diff --git a/.gitattributes b/.gitattributes index f3fa2bd01d02e51f1390daea371f28652855b3c1..fbe83a44aa559b2d87b91e0c688fa431d963fcdf 100644 --- a/.gitattributes +++ b/.gitattributes @@ -553,3 +553,29 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text .venv/lib/python3.12/site-packages/torch/_subclasses/__pycache__/fake_impls.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text .venv/lib/python3.12/site-packages/torch/_subclasses/__pycache__/fake_tensor.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text .venv/lib/python3.12/site-packages/torch/_vendor/quack/__pycache__/gemm_sm100.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/serializer.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/FileStoreTest filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/HashStoreTest filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/TCPStoreTest filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/protoc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/protoc-3.21.12.0 filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_aoti_abi_check filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_api filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_comms_id filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_cpp_rpc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_dist_autograd filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_global_state_manager filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_jit filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_lazy filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_privateuse1_profiler filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_profiler_collection filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_profiler_global_callback_stale_session filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/bin/test_shim filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/distributed_c10d.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/__init__.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/__init__.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_flat_param.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/fully_sharded_data_parallel.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/schedules.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/stage.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text +.venv/lib/python3.12/site-packages/torch/export/__pycache__/_trace.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_coreml/preprocess.py b/.venv/lib/python3.12/site-packages/torch/backends/_coreml/preprocess.py new file mode 100644 index 0000000000000000000000000000000000000000..57aa347a62e3cb3042f861df06e9820860a609d6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/_coreml/preprocess.py @@ -0,0 +1,154 @@ +# mypy: allow-untyped-defs +import hashlib +import json + +import coremltools as ct # type: ignore[import] +from coremltools.converters.mil.input_types import TensorType # type: ignore[import] +from coremltools.converters.mil.mil import types # type: ignore[import] +from coremltools.models.neural_network import quantization_utils # type: ignore[import] + +import torch + + +CT_METADATA_VERSION = "com.github.apple.coremltools.version" +CT_METADATA_SOURCE = "com.github.apple.coremltools.source" + + +class ScalarType: + Float = 0 + Double = 1 + Int = 2 + Long = 3 + Undefined = 4 + + +# Supported Tensor types in coremltools: +# https://github.com/apple/coremltools/blob/main/coremltools/converters/mil/frontend/torch/converter.py#L28 +torch_to_mil_types = { + ScalarType.Float: types.fp32, + ScalarType.Double: types.fp64, + ScalarType.Int: types.int32, + ScalarType.Long: types.int64, +} + + +class CoreMLComputeUnit: + CPU = "cpuOnly" + CPUAndGPU = "cpuAndGPU" + ALL = "all" + + +class CoreMLQuantizationMode: + LINEAR = "linear" + LINEAR_SYMMETRIC = "linear_symmetric" + NONE = "none" + + +def TensorSpec(shape, dtype=ScalarType.Float): + return (shape, dtype) + + +def CompileSpec( + inputs, + outputs, + backend=CoreMLComputeUnit.CPU, + allow_low_precision=True, + quantization_mode=CoreMLQuantizationMode.NONE, + mlmodel_export_path=None, + convert_to=None, +): + return ( + inputs, + outputs, + backend, + allow_low_precision, + quantization_mode, + mlmodel_export_path, + convert_to, + ) + + +def _check_enumerated_shape(shape): + for s in shape: + if not isinstance(s, (list, tuple)): + return False + return True + + +def _convert_to_mil_type(shape, dtype, name: str): + mil_shape = shape + if _check_enumerated_shape(shape): + mil_shape = ct.EnumeratedShapes(shape) + ml_type = TensorType(shape=mil_shape, dtype=torch_to_mil_types[dtype]) + ml_type.name = name + return ml_type + + +def preprocess(script_module: torch._C.ScriptObject, compile_spec: dict[str, tuple]): + spec = compile_spec["forward"] + ( + input_specs, + output_specs, + backend, + allow_low_precision, + quantization_mode, + mlmodel_export_path, + convert_to, + ) = spec + mil_inputs = [] + inputs = [] + for index, input in enumerate(input_specs): + shape, dtype = input + name = "input_" + str(index) + inputs.append([name, str(dtype), str(shape)]) + ml_type = _convert_to_mil_type(shape, dtype, name) + mil_inputs.append(ml_type) + model = torch.jit.RecursiveScriptModule._construct(script_module, lambda x: None) + mlmodel = ct.convert(model, inputs=mil_inputs, convert_to=convert_to) + + if quantization_mode != CoreMLQuantizationMode.NONE: + quant_model_spec = quantization_utils.quantize_weights( + mlmodel, nbits=8, quantization_mode=quantization_mode + ) + mlmodel = ct.models.MLModel(quant_model_spec) + + spec = mlmodel.get_spec() + if len(spec.description.output) != len(output_specs): # type: ignore[attr-defined] + raise AssertionError( + f"Number of outputs in spec ({len(spec.description.output)}) " # type: ignore[attr-defined] + f"does not match output_specs ({len(output_specs)})" + ) + outputs = [] + for index, output in enumerate(output_specs): + shape, dtype = output + name = spec.description.output[index].name # type: ignore[attr-defined] + outputs.append([name, str(dtype), str(shape)]) + mlmodel = ct.models.model.MLModel(spec) + print(mlmodel) + + if mlmodel_export_path is not None: + print(f"Saving CoreML .mlmodel file to {mlmodel_export_path}") + mlmodel.save(mlmodel_export_path) + + config = { + "spec_ver": str(spec.specificationVersion), # type: ignore[attr-defined] + "backend": backend, + "allow_low_precision": str(allow_low_precision), + } + metadata = { + "coremltool_ver": mlmodel.user_defined_metadata[CT_METADATA_VERSION], + "torch_ver": mlmodel.user_defined_metadata[CT_METADATA_SOURCE], + } + coreml_compile_spec = { + "inputs": inputs, + "outputs": outputs, + "config": config, + "metadata": metadata, + } + mlmodel = spec.SerializeToString() # type: ignore[attr-defined] + + return { + "model": mlmodel, + "hash": str(hashlib.sha256(mlmodel).hexdigest()), + "extra": json.dumps(coreml_compile_spec), + } diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..637a8161362ab9bb0a892c9026dc7992b8ec7ba8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/prepare.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/prepare.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bd163ce8098f0c3f646fce477aedd3dda75daaff Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/prepare.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/serializer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/serializer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eb9cf6de406621ffdcfc50d97f0310831426816f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/__pycache__/serializer.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83cbb3cc33e05f17bd096eed4f8e97c3e30e7c90a4b37258f32c07400f9201a4 +size 125075 diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/prepare.py b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/prepare.py new file mode 100644 index 0000000000000000000000000000000000000000..c83eac663e92fe2cb4b4c615107757cf6615e494 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/prepare.py @@ -0,0 +1,208 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs + +from typing import Optional + +import torch +from torch.backends._nnapi.serializer import _NnapiSerializer + + +ANEURALNETWORKS_PREFER_LOW_POWER = 0 +ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER = 1 +ANEURALNETWORKS_PREFER_SUSTAINED_SPEED = 2 + + +class NnapiModule(torch.nn.Module): + """Torch Module that wraps an NNAPI Compilation. + + This module handles preparing the weights, initializing the + NNAPI TorchBind object, and adjusting the memory formats + of all inputs and outputs. + """ + + # _nnapi.Compilation is defined + comp: Optional[torch.classes._nnapi.Compilation] # type: ignore[name-defined] # noqa: UP045 + weights: list[torch.Tensor] + out_templates: list[torch.Tensor] + + def __init__( + self, + shape_compute_module: torch.nn.Module, + ser_model: torch.Tensor, + weights: list[torch.Tensor], + inp_mem_fmts: list[int], + out_mem_fmts: list[int], + compilation_preference: int, + relax_f32_to_f16: bool, + ): + super().__init__() + self.shape_compute_module = shape_compute_module + self.ser_model = ser_model + self.weights = weights + self.inp_mem_fmts = inp_mem_fmts + self.out_mem_fmts = out_mem_fmts + self.out_templates = [] + self.comp = None + self.compilation_preference = compilation_preference + self.relax_f32_to_f16 = relax_f32_to_f16 + + @torch.jit.export + def init(self, args: list[torch.Tensor]): + if self.comp is not None: + raise AssertionError("comp must be None before initialization") + self.out_templates = self.shape_compute_module.prepare(self.ser_model, args) # type: ignore[operator] + self.weights = [w.contiguous() for w in self.weights] + comp = torch.classes._nnapi.Compilation() + comp.init2( + self.ser_model, + self.weights, + self.compilation_preference, + self.relax_f32_to_f16, + ) + + self.comp = comp + + def forward(self, args: list[torch.Tensor]) -> list[torch.Tensor]: + if self.comp is None: + self.init(args) + comp = self.comp + if comp is None: + raise AssertionError("comp must not be None") + outs = [torch.empty_like(out) for out in self.out_templates] + + if len(args) != len(self.inp_mem_fmts): + raise AssertionError( + f"args length {len(args)} != inp_mem_fmts length {len(self.inp_mem_fmts)}" + ) + fixed_args = [] + for idx in range(len(args)): + fmt = self.inp_mem_fmts[idx] + # These constants match the values in DimOrder in serializer.py + # TODO: See if it's possible to use those directly. + if fmt == 0: + fixed_args.append(args[idx].contiguous()) + elif fmt == 1: + fixed_args.append(args[idx].permute(0, 2, 3, 1).contiguous()) + else: + raise ValueError("Invalid mem_fmt") + comp.run(fixed_args, outs) + if len(outs) != len(self.out_mem_fmts): + raise AssertionError( + f"outs length {len(outs)} != out_mem_fmts length {len(self.out_mem_fmts)}" + ) + for idx in range(len(self.out_templates)): + fmt = self.out_mem_fmts[idx] + # These constants match the values in DimOrder in serializer.py + # TODO: See if it's possible to use those directly. + if fmt in (0, 2): + pass + elif fmt == 1: + outs[idx] = outs[idx].permute(0, 3, 1, 2) + else: + raise ValueError("Invalid mem_fmt") + return outs + + +def convert_model_to_nnapi( + model, + inputs, + serializer=None, + return_shapes=None, + use_int16_for_qint16=False, + compilation_preference=ANEURALNETWORKS_PREFER_SUSTAINED_SPEED, + relax_f32_to_f16=False, +): + ( + shape_compute_module, + ser_model_tensor, + used_weights, + inp_mem_fmts, + out_mem_fmts, + retval_count, + ) = process_for_nnapi( + model, inputs, serializer, return_shapes, use_int16_for_qint16 + ) + + nnapi_model = NnapiModule( + shape_compute_module, + ser_model_tensor, + used_weights, + inp_mem_fmts, + out_mem_fmts, + compilation_preference, + relax_f32_to_f16, + ) + + class NnapiInterfaceWrapper(torch.nn.Module): + """NNAPI list-ifying and de-list-ifying wrapper. + + NNAPI always expects a list of inputs and provides a list of outputs. + This module allows us to accept inputs as separate arguments. + It returns results as either a single tensor or tuple, + matching the original module. + """ + + def __init__(self, mod): + super().__init__() + self.mod = mod + + wrapper_model_py = NnapiInterfaceWrapper(nnapi_model) + wrapper_model = torch.jit.script(wrapper_model_py) + # TODO: Maybe make these names match the original. + arg_list = ", ".join(f"arg_{idx}" for idx in range(len(inputs))) + if retval_count < 0: + ret_expr = "retvals[0]" + else: + ret_expr = "".join(f"retvals[{idx}], " for idx in range(retval_count)) + wrapper_model.define( + f"def forward(self, {arg_list}):\n" + f" retvals = self.mod([{arg_list}])\n" + f" return {ret_expr}\n" + ) + return wrapper_model + + +def process_for_nnapi( + model, inputs, serializer=None, return_shapes=None, use_int16_for_qint16=False +): + model = torch.jit.freeze(model) + + if isinstance(inputs, torch.Tensor): + inputs = [inputs] + + serializer = serializer or _NnapiSerializer( + config=None, use_int16_for_qint16=use_int16_for_qint16 + ) + ( + ser_model, + used_weights, + inp_mem_fmts, + out_mem_fmts, + shape_compute_lines, + retval_count, + ) = serializer.serialize_model(model, inputs, return_shapes) + ser_model_tensor = torch.tensor(ser_model, dtype=torch.int32) + + # We have to create a new class here every time this function is called + # because module.define adds a method to the *class*, not the instance. + class ShapeComputeModule(torch.nn.Module): + """Code-gen-ed module for tensor shape computation. + + module.prepare will mutate ser_model according to the computed operand + shapes, based on the shapes of args. Returns a list of output templates. + """ + + shape_compute_module = torch.jit.script(ShapeComputeModule()) + real_shape_compute_lines = [ + "def prepare(self, ser_model: torch.Tensor, args: List[torch.Tensor]) -> List[torch.Tensor]:\n", + ] + [f" {line}\n" for line in shape_compute_lines] + shape_compute_module.define("".join(real_shape_compute_lines)) + + return ( + shape_compute_module, + ser_model_tensor, + used_weights, + inp_mem_fmts, + out_mem_fmts, + retval_count, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/serializer.py b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/serializer.py new file mode 100644 index 0000000000000000000000000000000000000000..b7c88b1341708704fa16cb97333eebbb0d0823f5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/_nnapi/serializer.py @@ -0,0 +1,2646 @@ +# mypy: allow-untyped-defs +import array +import enum +import functools +import logging +import operator +import struct +import sys +from typing import NamedTuple + +import torch + + +# TODO: Add type annotations +# TODO: Check tensor types for ops + + +LOG = logging.getLogger("nnapi_serialize") + + +class NNAPI_OperandCode: + FLOAT32 = 0 + INT32 = 1 + UINT32 = 2 + TENSOR_FLOAT32 = 3 + TENSOR_INT32 = 4 + TENSOR_QUANT8_ASYMM = 5 + BOOL = 6 + TENSOR_QUANT16_SYMM = 7 + TENSOR_FLOAT16 = 8 + TENSOR_BOOL8 = 9 + FLOAT16 = 10 + TENSOR_QUANT8_SYMM_PER_CHANNEL = 11 + TENSOR_QUANT16_ASYMM = 12 + + +class NNAPI_OperationCode: + ADD = 0 + AVERAGE_POOL_2D = 1 + CONCATENATION = 2 + CONV_2D = 3 + DEPTHWISE_CONV_2D = 4 + DEPTH_TO_SPACE = 5 + DEQUANTIZE = 6 + EMBEDDING_LOOKUP = 7 + FLOOR = 8 + FULLY_CONNECTED = 9 + HASHTABLE_LOOKUP = 10 + L2_NORMALIZATION = 11 + L2_POOL_2D = 12 + LOCAL_RESPONSE_NORMALIZATION = 13 + LOGISTIC = 14 + LSH_PROJECTION = 15 + LSTM = 16 + MAX_POOL_2D = 17 + MUL = 18 + RELU = 19 + RELU1 = 20 + RELU6 = 21 + RESHAPE = 22 + RESIZE_BILINEAR = 23 + RNN = 24 + SOFTMAX = 25 + SPACE_TO_DEPTH = 26 + SVDF = 27 + TANH = 28 + BATCH_TO_SPACE_ND = 29 + DIV = 30 + MEAN = 31 + PAD = 32 + SPACE_TO_BATCH_ND = 33 + SQUEEZE = 34 + STRIDED_SLICE = 35 + SUB = 36 + TRANSPOSE = 37 + ABS = 38 + ARGMAX = 39 + ARGMIN = 40 + AXIS_ALIGNED_BBOX_TRANSFORM = 41 + BIDIRECTIONAL_SEQUENCE_LSTM = 42 + BIDIRECTIONAL_SEQUENCE_RNN = 43 + BOX_WITH_NMS_LIMIT = 44 + CAST = 45 + CHANNEL_SHUFFLE = 46 + DETECTION_POSTPROCESSING = 47 + EQUAL = 48 + EXP = 49 + EXPAND_DIMS = 50 + GATHER = 51 + GENERATE_PROPOSALS = 52 + GREATER = 53 + GREATER_EQUAL = 54 + GROUPED_CONV_2D = 55 + HEATMAP_MAX_KEYPOINT = 56 + INSTANCE_NORMALIZATION = 57 + LESS = 58 + LESS_EQUAL = 59 + LOG = 60 + LOGICAL_AND = 61 + LOGICAL_NOT = 62 + LOGICAL_OR = 63 + LOG_SOFTMAX = 64 + MAXIMUM = 65 + MINIMUM = 66 + NEG = 67 + NOT_EQUAL = 68 + PAD_V2 = 69 + POW = 70 + PRELU = 71 + QUANTIZE = 72 + QUANTIZED_16BIT_LSTM = 73 + RANDOM_MULTINOMIAL = 74 + REDUCE_ALL = 75 + REDUCE_ANY = 76 + REDUCE_MAX = 77 + REDUCE_MIN = 78 + REDUCE_PROD = 79 + REDUCE_SUM = 80 + ROI_ALIGN = 81 + ROI_POOLING = 82 + RSQRT = 83 + SELECT = 84 + SIN = 85 + SLICE = 86 + SPLIT = 87 + SQRT = 88 + TILE = 89 + TOPK_V2 = 90 + TRANSPOSE_CONV_2D = 91 + UNIDIRECTIONAL_SEQUENCE_LSTM = 92 + UNIDIRECTIONAL_SEQUENCE_RNN = 93 + RESIZE_NEAREST_NEIGHBOR = 94 + + +class NNAPI_FuseCode: + FUSED_NONE = 0 + FUSED_RELU = 1 + FUSED_RELU1 = 2 + FUSED_RELU6 = 3 + + +class OperandValueSourceType: + IMMEDIATE = 0 + NUMBERED_BUFFER = 2 + NUMBERED_MEMORY = 3 + + +# Scalar types that appear explicitly in models. +# These must be kept in sync with +# AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_QINTS. +# TODO: Expose these directly to Python to avoid maintaining this list. +class TorchScalarTypes(enum.Enum): + QUINT8 = 13 + + +def approx_equal(lhs, rhs, tolerance=1e-6): + return abs(lhs - rhs) <= tolerance * min(lhs, rhs) + + +def tensor_size(op_type, dims): + ITEM_SIZES = { + NNAPI_OperandCode.TENSOR_FLOAT32: 4, + NNAPI_OperandCode.TENSOR_INT32: 4, + NNAPI_OperandCode.TENSOR_QUANT8_ASYMM: 1, + NNAPI_OperandCode.TENSOR_QUANT16_SYMM: 2, + NNAPI_OperandCode.TENSOR_QUANT16_ASYMM: 2, + } + size = ITEM_SIZES[op_type] + for d in dims: + size *= d + return size + + +def change_element(tup, index, value): + ls = list(tup) + ls[index] = value + return tuple(ls) + + +class ConvPoolArgs2d(NamedTuple): + """Configuration arguments for a convolution.""" + + kernel_h: int + kernel_w: int + stride_h: int + stride_w: int + pad_t: int + pad_b: int + pad_l: int + pad_r: int + dilation_h: int + dilation_w: int + group: int + + +class DimOrder(enum.Enum): + PRESUMED_CONTIGUOUS = 0 + CHANNELS_LAST = 1 + SCALAR_OR_VECTOR = 2 + UNKNOWN_CONSTANT = 999 + + +class Operand(NamedTuple): + """Representation of an NNAPI operand.""" + + # NNAPI operand type. One of NNAPI_OperandCode. + # TODO: Make this an enum. + op_type: int + + # This is always the PyTorch shape, which is NCHW for feature maps. + # The actual NNAPI operand might have a transposed shape. + # we use 0 for load time dynamic shapes & -1 for runtime dynamic shapes + shape: tuple[int, ...] + + # Specifies how the shape of the operand that we define in NNAPI + # relates to the shape we track above. + # - PRESUMED_CONTIGUOUS: physical NNAPI operand will exactly match + # the shape of the PyTorch tensor. + # - CHANNELS_LAST: The PyTorch tensor is expected to be NCHW, and + # the NNAPI operand will be represented explicitly as NHWC. + dim_order: DimOrder + + # Quantization params + scale: float + zero_point: int + + def use_nchw(self): + if self.dim_order is DimOrder.PRESUMED_CONTIGUOUS: + return True + if self.dim_order is DimOrder.CHANNELS_LAST: + return False + raise Exception("Unknown dim order") # noqa: TRY002 + + +def broadcast_shapes(shape1, shape2): + if len(shape1) <= 0: + raise AssertionError(f"shape1 must have length > 0, got {len(shape1)}") + if len(shape2) <= 0: + raise AssertionError(f"shape2 must have length > 0, got {len(shape2)}") + s1 = list(shape1) + s2 = list(shape2) + # TODO: Support non-equal-rank broadcast where semantics match. + # This can be tricky for NHWC tensors because dimension orders + # don't match between PT and NNAPI, even though semantics match. + if len(s1) > len(s2): + # s2 = [1] * (len(s1) - len(s2)) + s2 + raise Exception( # noqa: TRY002 + "Non-equal-rank broadcast is not supported yet." + ) + if len(s2) > len(s1): + # s3 = [1] * (len(s2) - len(s1)) + s1 + raise Exception( # noqa: TRY002 + "Non-equal-rank broadcast is not supported yet." + ) + ret = [] + for d1, d2 in zip(s1, s2): + if d1 == 1: + ret.append(d2) + elif d2 == 1: + ret.append(d1) + elif d1 == d2: + ret.append(d1) + else: + raise Exception( # noqa: TRY002 + f"Cannot broadcast shapes: {shape1} and {shape2}" + ) + return tuple(ret) + + +def get_conv_pool_shape(image_shape, args, out_ch, transpose): + batch, _in_c, in_h, in_w = image_shape + + # TODO: Handle dilation + if args.dilation_h != 1 or args.dilation_w != 1: + raise Exception("Dilation not supported yet.") # noqa: TRY002 + + if transpose: + out_h = (in_h - 1) * args.stride_h + args.kernel_h - args.pad_t - args.pad_b + out_w = (in_w - 1) * args.stride_w + args.kernel_w - args.pad_l - args.pad_l + else: + out_h = (in_h - args.kernel_h + args.pad_t + args.pad_b) // args.stride_h + 1 + out_w = (in_w - args.kernel_w + args.pad_l + args.pad_r) // args.stride_w + 1 + + # Handle variable-sized tensors. + if in_h == 0: + out_h = 0 + if in_w == 0: + out_w = 0 + + out_shape = (batch, out_ch, out_h, out_w) + return out_shape + + +def fix_shape(shape, dim_order): + # Return the actual shape that an operand should have in NNAPI, + # given a PyTorch shape and dimension order. This is where we + # convert from PyTorch's "always NCHW" shape to explicit NHWC. + if dim_order is DimOrder.PRESUMED_CONTIGUOUS: + return shape + if dim_order is DimOrder.CHANNELS_LAST: + return tuple([shape[0]] + list(shape[2:]) + [shape[1]]) + if dim_order is DimOrder.SCALAR_OR_VECTOR: + if not (len(shape) == 0 or len(shape) == 1): + raise AssertionError( + f"SCALAR_OR_VECTOR requires len(shape) == 0 or 1, got {len(shape)}" + ) + return shape + if dim_order is DimOrder.UNKNOWN_CONSTANT: + # XXX think this through + return shape + raise Exception(f"Bad dim_order: {dim_order!r}.") # noqa: TRY002 + + +def reverse_map_dim(dim_order, d): + # Return the original PyTorch dimension position for a given dimension. + # d should be the dimension that NNAPI will see. + # reverse_map_dim(PRESUMED_CONTIGUOUS, x) == x + # reverse_map_dim(CHANNELS_LAST, 3) == 1 + if dim_order in (DimOrder.PRESUMED_CONTIGUOUS, DimOrder.SCALAR_OR_VECTOR): + return d + if dim_order is DimOrder.CHANNELS_LAST: + return [0, 2, 3, 1][d] + raise AssertionError(f"expected DimOrder.CHANNELS_LAST, got {dim_order}") + + +def flex_name(op_id, dim): + # Return the local variable name for the computed flexible size + # for a given op and dimension. + return f"s_{op_id}_{dim}" + + +class _NnapiSerializer: + def __init__(self, config, use_int16_for_qint16=False): + self.operands = [] + self.values = [] + self.operations = [] + self.value_data = [] + self.operation_args = [] + self.inputs = [] + self.outputs = [] + self.flexible_shape_computation_lines = [] + + self.modules = {} + self.constants = {} + self.tensor_sequences = {} + self.jitval_operand_map = {} + self.cached_immediates = {} + self.used_weights = [] + self.weight_offset = 0 + self.use_int16_for_qint16 = use_int16_for_qint16 + + if config is None: + config = {} + + def get_next_operand_id(self): + return len(self.operands) + + # Add a tensor operand corresponding to a JIT Value. + # Returns the NNAPI operand ID. Can be looked up later with + # get_tensor_operand_by_jitval. + def add_tensor_operand(self, jitval, oper): + if not isinstance(oper, Operand): + raise AssertionError(f"expected Operand, got {type(oper)}") + if jitval in self.jitval_operand_map: + raise Exception(f"Duplicate tensor: {jitval!r}") # noqa: TRY002 + + operand_id = self.get_next_operand_id() + self.operands.append(oper) + self.jitval_operand_map[jitval] = operand_id + return operand_id + + # Add a tensor operand that does not correspond to a JIT Value. + # Useful for cases where multiple NNAPI operands are required + # to implement one JIT IR node. Returns the NNAPI operand ID. + def add_anonymous_tensor_operand(self, oper): + if not isinstance(oper, Operand): + raise AssertionError(f"expected Operand, got {type(oper)}") + operand_id = self.get_next_operand_id() + self.operands.append(oper) + return operand_id + + def torch_tensor_to_operand(self, tensor, dim_order): + dtype = str(tensor.dtype).replace("torch.", "") + scale = 0.0 + zero_point = 0 + if dtype == "float32": + op_type = NNAPI_OperandCode.TENSOR_FLOAT32 + elif dtype == "int32": + op_type = NNAPI_OperandCode.TENSOR_INT32 + elif dtype == "quint8": + op_type = NNAPI_OperandCode.TENSOR_QUANT8_ASYMM + scale = tensor.q_scale() + zero_point = tensor.q_zero_point() + elif dtype == "qint32": + op_type = NNAPI_OperandCode.TENSOR_INT32 + scale = tensor.q_scale() + zero_point = tensor.q_zero_point() + if zero_point != 0: + raise AssertionError(f"qint32 zero_point must be 0, got {zero_point}") + elif dtype == "int16": + if self.use_int16_for_qint16: + nnapi_dtype = getattr(tensor, "nnapi_dtype", None) + op_codes = ( + NNAPI_OperandCode.TENSOR_QUANT16_SYMM, + NNAPI_OperandCode.TENSOR_QUANT16_ASYMM, + ) + if nnapi_dtype in op_codes: + op_type = nnapi_dtype + scale = tensor.nnapi_scale + zero_point = tensor.nnapi_zero_point + else: + raise Exception( # noqa: TRY002 + f"`nnapi_type` needs to be one of {op_codes} for `int16`" + ) + else: + raise Exception( # noqa: TRY002 + "`int16` isn't supported. If you're trying to represent NNAPI" + " qint16 with Pytorch int16, set `use_int16_for_qint16 = True`" + ) + else: + raise Exception( # noqa: TRY002 + f"Can't handle input with dtype '{tensor.dtype}'" + ) + return Operand( + shape=tuple(tensor.shape), + # pyrefly: ignore [bad-argument-type] + op_type=op_type, + dim_order=dim_order, + scale=scale, + zero_point=zero_point, + ) + + def add_tensor_operand_for_input(self, arg_idx, jitval, tensor): + dim_order = ( + DimOrder.CHANNELS_LAST + if getattr(tensor, "nnapi_nhwc", False) + else DimOrder.PRESUMED_CONTIGUOUS + ) + toper = self.torch_tensor_to_operand(tensor, dim_order) + operand_id = self.add_tensor_operand(jitval, toper) + self.inputs.append(operand_id) + for dim, size in enumerate(tensor.shape): + if size == 0: + self.compute_operand_shape( + operand_id, dim, f"args[{arg_idx}].shape[{dim}]" + ) + return operand_id + + def add_tensor_operand_for_weight( + self, tensor, dim_order=DimOrder.UNKNOWN_CONSTANT + ): + toper = self.torch_tensor_to_operand(tensor, dim_order) + operand_id = len(self.operands) + self.operands.append(toper) + tsize = tensor_size(toper.op_type, toper.shape) + self.values.append((operand_id, OperandValueSourceType.NUMBERED_BUFFER)) + buf_num = len(self.used_weights) + offset = 0 + self.value_data.append(struct.pack("iii", buf_num, offset, tsize)) + # For NHWC NNAPI op, lay out data in the same dim order by permuting torch tensor + if dim_order == DimOrder.CHANNELS_LAST: + tensor = tensor.permute(0, 2, 3, 1) + self.used_weights.append(tensor) + return operand_id + + def add_immediate_operand(self, code, value, dims): + if not isinstance(dims, tuple): + raise AssertionError(f"dims must be a tuple, got {type(dims)}") + cache_key = (code, value) + if cache_key not in self.cached_immediates: + operand_id = len(self.operands) + self.operands.append(Operand(code, dims, DimOrder.SCALAR_OR_VECTOR, 0.0, 0)) + self.values.append((operand_id, OperandValueSourceType.IMMEDIATE)) + self.value_data.append(value) + self.cached_immediates[cache_key] = operand_id + return self.cached_immediates[cache_key] + + def add_immediate_int_scalar(self, value): + return self.add_immediate_operand( + NNAPI_OperandCode.INT32, struct.pack("i", value), () + ) + + def add_immediate_float_scalar(self, value): + return self.add_immediate_operand( + NNAPI_OperandCode.FLOAT32, struct.pack("f", value), () + ) + + def add_immediate_bool_scalar(self, value): + return self.add_immediate_operand( + NNAPI_OperandCode.BOOL, b"\x01" if value else b"\x00", () + ) + + def add_immediate_int_vector(self, value): + return self.add_immediate_operand( + NNAPI_OperandCode.TENSOR_INT32, + array.array("i", value).tobytes(), + (len(value),), + ) + + def has_operand_for_jitval(self, jitval): + return jitval in self.jitval_operand_map + + def get_tensor_operand_by_jitval(self, jitval): + operand_id = self.jitval_operand_map[jitval] + return (operand_id, self.operands[operand_id]) + + def get_tensor_operand_by_jitval_fixed_size(self, jitval): + op_id, oper = self.get_tensor_operand_by_jitval(jitval) + for s in oper.shape: + if s == 0: + # TODO: Improve this error message, possibly after converting + # many callsites to support flexible size. + raise Exception( # noqa: TRY002 + "Flexible size is not supported for this operand." + ) + if s < 0: + # runtime flex + LOG.warning("Operand %s has runtime flex shape", oper) + return op_id, oper + + def get_tensor_operand_or_constant( + self, jitval, dim_order=DimOrder.PRESUMED_CONTIGUOUS + ): + operand_id = self.jitval_operand_map.get(jitval) + if operand_id is None: + _, value = self.get_constant_value(jitval, "TensorType") + operand_id = self.add_tensor_operand_for_weight(value, dim_order) + return (operand_id, self.operands[operand_id]) + + def get_tensor_operand_for_weight(self, jitval): + _, value = self.get_constant_value(jitval, "TensorType") + operand_id = self.add_tensor_operand_for_weight(value) + return (operand_id, self.operands[operand_id]) + + def add_operation(self, opcode, inputs, outputs): + self.operations.append((opcode, len(inputs), len(outputs))) + self.operation_args.extend(inputs + outputs) + + def add_tensor_sequence(self, jitval, values): + if jitval in self.tensor_sequences: + raise AssertionError(f"jitval {jitval!r} already in tensor_sequences") + self.tensor_sequences[jitval] = values + + def add_constant_value(self, jitval, ctype, value): + if jitval in self.constants: + raise AssertionError(f"jitval {jitval!r} already in constants") + self.constants[jitval] = (ctype, value) + + def get_constant_value(self, jitval, typekind=None): + record = self.constants.get(jitval) + if record is None: + raise Exception( # noqa: TRY002 + f"Could not find constant value for '{jitval!r}'." + ) + ctype, _ = record + if typekind is not None and ctype.kind() != typekind: + raise Exception( # noqa: TRY002 + f"Expected constant value of type {typekind}, but got {ctype.kind()} for value '{jitval!r}'" + ) + return record + + def operand_to_template_torchscript(self, op_id, oper, shape=None): + """Return a TorchScript expression to build a template for a given operand.""" + if shape is None: + shape = oper.shape + else: + if len(shape) != len(oper.shape): + raise AssertionError( + f"shape length {len(shape)} != oper.shape length {len(oper.shape)}" + ) + + shape_parts = ["("] + for d, s in enumerate(shape): + if s > 0: + # Fixed shape dimension: just add the value. + shape_parts.append(str(s)) + elif s == 0: + # Load time flexible shape dimension: it should have been computed in a variable. + shape_parts.append(flex_name(op_id, d)) + elif s == -1: + # Runtime flexible shape + shape_parts.append("0") + else: + raise Exception( # noqa: TRY002 + "Unknown dim value, dimensions should be >= -1" + ) + shape_parts.append(",") + shape_parts.append(")") + shape_code = "".join(shape_parts) + if oper.op_type == NNAPI_OperandCode.TENSOR_FLOAT32: + return f"torch.zeros({shape_code}, dtype=torch.float32)" + elif oper.op_type == NNAPI_OperandCode.TENSOR_INT32: + return f"torch.zeros({shape_code}, dtype=torch.int32)" + elif oper.op_type == NNAPI_OperandCode.TENSOR_QUANT8_ASYMM: + return ( + f"torch.quantize_per_tensor(" + f"torch.zeros(1), scale={oper.scale}, zero_point={oper.zero_point}, dtype=torch.quint8)" + f".expand({shape_code}).contiguous()" + ) + elif oper.op_type in ( + NNAPI_OperandCode.TENSOR_QUANT16_ASYMM, + NNAPI_OperandCode.TENSOR_QUANT16_SYMM, + ): + if self.use_int16_for_qint16: + return f"torch.zeros({shape_code}, dtype=torch.int16)" + else: + raise Exception( # noqa: TRY002 + "`int16` isn't supported. If you're trying to represent NNAPI" + " qint16 with Pytorch int16, set `use_int16_for_qint16 = True`" + ) + + raise Exception( # noqa: TRY002 + f"Unsupported output operand type: {oper.op_type}" + ) + + def forward_operand_shape(self, out_op_id, out_dim, in_op_id, in_dim): + self.compute_operand_shape(out_op_id, out_dim, flex_name(in_op_id, in_dim)) + + def compute_operand_shape(self, op_id, dim, expr): + self.flexible_shape_computation_lines.append( + f"{flex_name(op_id, dim)} = {expr}" + ) + + def transpose_to_nhwc(self, in_id, oper): + if oper.shape[2:] != (1, 1): + raise Exception( # noqa: TRY002 + "Automatic transpose only supported for H,W == 1,1" + ) + + out_oper = oper._replace(dim_order=DimOrder.CHANNELS_LAST) + + inputs = [None] * 2 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_vector([0, 2, 3, 1]) + + outputs = [None] * 1 + outputs[0] = self.add_anonymous_tensor_operand(out_oper) + + self.add_operation(NNAPI_OperationCode.TRANSPOSE, inputs, outputs) + + return outputs[0], out_oper + + # Transpose inputs as necessary to allow broadcasting. + def transpose_for_broadcast(self, in0_id, in0_oper, in1_id, in1_oper): + if in0_oper.dim_order == in1_oper.dim_order: + return in0_id, in0_oper, in1_id, in1_oper + + # Assume NHWC is preferred if there is a mismatch. + orders = (in0_oper.dim_order, in1_oper.dim_order) + if orders == (DimOrder.PRESUMED_CONTIGUOUS, DimOrder.CHANNELS_LAST): + return self.transpose_to_nhwc(in0_id, in0_oper) + (in1_id, in1_oper) + if orders == (DimOrder.CHANNELS_LAST, DimOrder.PRESUMED_CONTIGUOUS): + return (in0_id, in0_oper) + self.transpose_to_nhwc(in1_id, in1_oper) + + raise Exception( # noqa: TRY002 + f"Automatic transpose not supported for dim_orders: {in0_oper.dim_order!r}, {in1_oper.dim_order!r}" + ) + + def get_size_arg(self, jitval): + ctype, value = self.get_constant_value(jitval) + if ctype.kind() == "ListType": + if ctype.getElementType().kind() != "IntType": + raise AssertionError( + f"expected ListType of IntType, got {ctype.getElementType().kind()}" + ) + return value + raise Exception( # noqa: TRY002 + f"Can't handle size arg of type '{ctype!r}' for '{jitval!r}'" + ) + + def get_conv_pool_args_2d_from_pack(self, kernel_size, packed_config): + pc = [i.item() for i in packed_config] + if pc[0] != 2: + raise AssertionError(f"expected pc[0] == 2, got {pc[0]}") + strides = [pc[1], pc[2]] + paddings = [pc[3], pc[4]] + dilations = [pc[5], pc[6]] + output_padding = [pc[7], pc[8]] + group_num = pc[9] + + if len(pc) != 11: + raise AssertionError(f"expected len(pc) == 11, got {len(pc)}") + if output_padding != [0, 0]: + raise AssertionError( + f"expected output_padding == [0, 0], got {output_padding}" + ) + + return self.get_conv_pool_args_2d_common( + kernel_size, strides, paddings, dilations, group_num + ) + + def get_conv_pool_args_2d_from_jit( + self, kernel_size, stride, padding, dilation=None, group=None + ): + strides = self.get_size_arg(stride) + paddings = self.get_size_arg(padding) + if dilation is None: + dilations = [1, 1] + else: + dilations = self.get_size_arg(dilation) + if group is not None: + _, group_num = self.get_constant_value(group, "IntType") + else: + group_num = None + return self.get_conv_pool_args_2d_common( + kernel_size, strides, paddings, dilations, group_num + ) + + def get_conv_pool_args_2d_common( + self, kernel_size, strides, paddings, dilations, group_num + ): + kernels = list(kernel_size) + + if len(kernels) != 2: + raise AssertionError(f"expected len(kernels) == 2, got {len(kernels)}") + if len(strides) != 2: + raise AssertionError(f"expected len(strides) == 2, got {len(strides)}") + if len(paddings) != 2: + raise AssertionError(f"expected len(paddings) == 2, got {len(paddings)}") + if len(dilations) != 2: + raise AssertionError(f"expected len(dilations) == 2, got {len(dilations)}") + + # NNAPI uses 4 values for padding. + ph, pw = paddings + real_paddings = [ph, ph, pw, pw] + + return ConvPoolArgs2d( + *(kernels + strides + real_paddings + dilations + [group_num]) + ) + + def serialize_model(self, model, inputs, return_shapes=None): + self.add_immediate_bool_scalar(False) + self.add_immediate_bool_scalar(True) + + inp_dim_orders = [] + out_dim_orders = [] + + self_jitval = next(model.graph.inputs()) + self.add_constant_value(self_jitval, self_jitval.type(), model) + + for arg_idx, (input_value, input_tensor) in enumerate( + zip(list(model.graph.inputs())[1:], inputs) + ): + op_id = self.add_tensor_operand_for_input( + arg_idx, input_value, input_tensor + ) + inp_dim_orders.append(self.operands[op_id].dim_order.value) + + for idx, node in enumerate(model.graph.nodes()): + LOG.debug("Processing node #%d: %r", idx, node) + self.add_node(node) + + retn = model.graph.return_node() + if retn.inputsSize() != 1: + raise AssertionError( + f"expected retn.inputsSize() == 1, got {retn.inputsSize()}" + ) + if retn.outputsSize() != 0: + raise AssertionError( + f"expected retn.outputsSize() == 0, got {retn.outputsSize()}" + ) + retn_input = retn.inputsAt(0) + template_return_lines = ["return ["] + if retn_input.type().kind() == "TensorType": + return_values = [retn_input] + retval_count = -1 + elif retn_input.type().kind() == "TupleType": + return_values = self.tensor_sequences[retn_input] + retval_count = len(return_values) + else: + raise Exception( # noqa: TRY002 + f"Unsupported return type: {retn_input.type()}" + ) + + if return_shapes is not None: + if len(return_shapes) != len(return_values): + raise AssertionError( + f"return_shapes length {len(return_shapes)} != return_values length {len(return_values)}" + ) + for i, v in enumerate(return_values): + op_id = self.jitval_operand_map[v] + self.outputs.append(op_id) + out_dim_orders.append(self.operands[op_id].dim_order.value) + shape = return_shapes[i] if return_shapes else None + template_return_lines.append( + self.operand_to_template_torchscript(op_id, self.operands[op_id], shape) + + "," + ) + template_return_lines.append("]") + + model = [] + + version = 1 + header = struct.pack( + "iiiiii", + version, + len(self.operands), + len(self.values), + len(self.operations), + len(self.inputs), + len(self.outputs), + ) + model.append(header) + + serialized_values, serialized_value_data = self.serialize_values() + + model.extend( + struct.pack("iifi", t, len(d), s, z) for (t, d, _m, s, z) in self.operands + ) + model.extend(serialized_values) + model.extend(struct.pack("iii", *x) for x in self.operations) + + # Compact the model so we can get its length so far. + model = [b"".join(model)] + model_offset = len(model[0]) + # Model offset is the index into the model (in 32-bit words, not bytes) + # of the next dimension we're about to serialize. If it's 0, + # generate code to mutate it before passing to NNAPI. + if model_offset % 4 != 0: + raise AssertionError( + f"model_offset must be divisible by 4, got {model_offset}" + ) + model_offset = int(model_offset / 4) + + for op_id, (_, dims, dim_order, _, _) in enumerate(self.operands): + shape = fix_shape(dims, dim_order) + for d, s in enumerate(shape): + if s == 0: + pt_d = reverse_map_dim(dim_order, d) + self.flexible_shape_computation_lines.append( + f"ser_model[{model_offset}] = {flex_name(op_id, pt_d)}" + ) + model_offset += 1 + + # convert runtime flex shape from -1 to 0 + shape = tuple(d if d != -1 else 0 for d in shape) + model.append(self.serialize_ints(shape)) + + model.extend(serialized_value_data) + model.append(self.serialize_ints(self.operation_args)) + model.append(self.serialize_ints(self.inputs)) + model.append(self.serialize_ints(self.outputs)) + + self.flexible_shape_computation_lines.extend(template_return_lines) + + return ( + array.array("i", b"".join(model)), + self.used_weights, + inp_dim_orders, + out_dim_orders, + self.flexible_shape_computation_lines, + retval_count, + ) + + def serialize_values(self): + serialized_values = [] + serialized_value_data = [] + if len(self.values) != len(self.value_data): + raise AssertionError( + f"values length {len(self.values)} != value_data length {len(self.value_data)}" + ) + for (op_index, source_type), data in zip(self.values, self.value_data): + source_length = len(data) + + # Pad with 0 bytes out to a multiple of 4 for alignment. + physical_length = ((source_length - 1) | 0x3) + 1 + padded_data = data + (b"\0" * (physical_length - source_length)) + + serialized_values.append( + struct.pack("iii", op_index, source_type, source_length) + ) + serialized_value_data.append(padded_data) + + return serialized_values, serialized_value_data + + @staticmethod + def serialize_ints(ints): + return array.array("i", ints).tobytes() + + ADDER_MAP = { + "prim::GetAttr": lambda self, node: self.add_getattr(node), + "prim::Constant": lambda self, node: self.add_constant_node(node), + "prim::ListConstruct": lambda self, node: self.add_list_construct(node), + "prim::TupleConstruct": lambda self, node: self.add_tuple_construct(node), + "aten::unsqueeze": lambda self, node: self.add_unsqueeze(node), + "aten::to": lambda self, node: self.add_to(node), + "aten::detach": lambda self, node: self._identity(node), + "aten::reshape": lambda self, node: self.add_reshape(node), + "aten::flatten": lambda self, node: self.add_flatten(node), + "aten::slice": lambda self, node: self.add_slice(node), + "aten::size": lambda self, node: self.add_size(node), + "aten::cat": lambda self, node: self.add_cat(node), + "aten::mean": lambda self, node: self.add_mean(node), + "aten::quantize_per_tensor": lambda self, node: self.add_quantize(node), + "aten::dequantize": lambda self, node: self.add_dequantize(node), + "aten::add": lambda self, node: self.add_add_sub_op( + node, NNAPI_OperationCode.ADD, NNAPI_FuseCode.FUSED_NONE + ), + "aten::sub": lambda self, node: self.add_add_sub_op( + node, NNAPI_OperationCode.SUB, NNAPI_FuseCode.FUSED_NONE + ), + "aten::mul": lambda self, node: self.add_pointwise_simple_binary_broadcast_op( + node, NNAPI_OperationCode.MUL, NNAPI_FuseCode.FUSED_NONE + ), + "aten::div": lambda self, node: self.add_pointwise_simple_binary_broadcast_op( + node, NNAPI_OperationCode.DIV, NNAPI_FuseCode.FUSED_NONE + ), + "aten::relu": lambda self, node: self.add_pointwise_simple_unary_op( + node, NNAPI_OperationCode.RELU + ), + "aten::sigmoid": lambda self, node: self.add_pointwise_simple_unary_op( + node, NNAPI_OperationCode.LOGISTIC + ), + "aten::softmax": lambda self, node: self.add_softmax(node), + "aten::hardtanh": lambda self, node: self.add_hardtanh(node), + "aten::avg_pool2d": lambda self, node: self.add_avg_pool2d(node), + "aten::max_pool2d": lambda self, node: self.add_pool2d_node( + node, NNAPI_OperationCode.MAX_POOL_2D + ), + "aten::adaptive_avg_pool2d": lambda self, node: self.add_adaptive_avg_pool2d( + node + ), + "aten::upsample_nearest2d": lambda self, node: self.add_upsample_nearest2d( + node + ), + "aten::prelu": lambda self, node: self.add_prelu_op(node), + "aten::addmm": lambda self, node: self.add_addmm(node), + "aten::linear": lambda self, node: self.add_linear(node), + "aten::_convolution": lambda self, node: self.add_conv_underscore(node), + "aten::conv2d": lambda self, node: self.add_conv2d(node), + "aten::log_softmax": lambda self, node: self.add_log_softmax(node), + "quantized::linear": lambda self, node: self.add_qlinear(node), + "quantized::conv2d": lambda self, node: self.add_qconv2d( + node, NNAPI_FuseCode.FUSED_NONE + ), + "quantized::conv2d_relu": lambda self, node: self.add_qconv2d( + node, NNAPI_FuseCode.FUSED_RELU + ), + "quantized::conv_transpose2d": lambda self, node: self.add_qconv2d( + node, NNAPI_FuseCode.FUSED_NONE, transpose=True + ), + "quantized::add": lambda self, node: self.add_qadd( + node, NNAPI_OperationCode.ADD, NNAPI_FuseCode.FUSED_NONE + ), + "quantized::add_relu": lambda self, node: self.add_qadd( + node, NNAPI_OperationCode.ADD, NNAPI_FuseCode.FUSED_RELU + ), + "quantized::mul": lambda self, node: self.add_qadd( + node, NNAPI_OperationCode.MUL, NNAPI_FuseCode.FUSED_NONE + ), + } + + def add_node(self, node): + adder = self.ADDER_MAP.get(node.kind()) + if not adder: + raise Exception( # noqa: TRY002 + f"Unsupported node kind ({node.kind()!r}) in node {node!r}" + ) + adder(self, node) + + def _identity(self, node): + in_id, _in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + jitval = node.outputsAt(0) + self.jitval_operand_map[jitval] = in_id + + def add_getattr(self, node): + if node.inputsSize() != 1: + raise AssertionError( + f"expected node.inputsSize() == 1, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + obj_ctype, obj = self.get_constant_value(node.inputsAt(0)) + if not str(obj_ctype).startswith("__torch__."): + raise AssertionError( + f"expected obj_ctype to start with '__torch__.', got {obj_ctype}" + ) + name = node.s("name") + value = getattr(obj, name) + output = node.outputsAt(0) + ctype = output.type() + self.add_constant_value(output, ctype, value) + + def add_constant_node(self, node): + if node.inputsSize() != 0: + raise AssertionError( + f"expected node.inputsSize() == 0, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + output = node.outputsAt(0) + ctype = output.type() + value = output.toIValue() + self.add_constant_value(output, ctype, value) + + def add_list_construct(self, node): + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + output = node.outputsAt(0) + ctype = output.type() + const_vals: list | None = [] + tensors: list | None = [] + for inp in node.inputs(): + if const_vals is not None and inp in self.constants: + _, val = self.get_constant_value(inp) + const_vals.append(val) + else: + const_vals = None + if tensors is not None and inp.type().kind() == "TensorType": + tensors.append(inp) + else: + tensors = None + + if const_vals is not None: + # NOTE: Now that TorchScript supports list constants, + # this code path might not be used anymore. + self.add_constant_value(output, ctype, const_vals) + if tensors is not None: + self.add_tensor_sequence(output, tensors) + if const_vals is None and tensors is None: + raise Exception( # noqa: TRY002 + f"Unable to handle ListConstruct node. Neither all constants nor all tensors. {node!r}" + ) + + def add_tuple_construct(self, node): + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + output = node.outputsAt(0) + values = list(node.inputs()) + self.add_tensor_sequence(output, values) + + def add_unsqueeze(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + + _, dim = self.get_constant_value(node.inputsAt(1), "IntType") + if in_oper.dim_order != DimOrder.PRESUMED_CONTIGUOUS: + raise AssertionError( + f"expected dim_order PRESUMED_CONTIGUOUS, got {in_oper.dim_order}" + ) + + real_dim = dim if dim >= 0 else dim + len(in_oper.shape) + 1 + out_shape_list = list(in_oper.shape) + out_shape_list.insert(real_dim, 1) + out_shape = tuple(out_shape_list) + out_oper = in_oper._replace(shape=out_shape) + + inputs = [None] * 2 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_scalar(dim) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.EXPAND_DIMS, inputs, outputs) + + def add_to(self, node): + # Handle to("cpu") / to("gpu") case + self._identity(node) + + def add_reshape(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + + shape_ctype, shape = self.get_constant_value(node.inputsAt(1)) + if shape_ctype.kind() != "ListType": + raise AssertionError( + f"expected shape_ctype ListType, got {shape_ctype.kind()}" + ) + if shape_ctype.getElementType().kind() != "IntType": + raise AssertionError( + f"expected shape element type IntType, got {shape_ctype.getElementType().kind()}" + ) + is_trivial_reshape = len(shape) == 2 and shape[1] == -1 + + if in_oper.dim_order != DimOrder.PRESUMED_CONTIGUOUS and not is_trivial_reshape: + raise Exception( # noqa: TRY002 + "Currently, reshape is only supported on NHWC tensors if the target size is [X, -1]." + ) + + # Bit of a hack here. Use a real tensor to infer the output shape. + out_shape = torch.zeros(1).expand(in_oper.shape).reshape(shape).shape + out_oper = in_oper._replace( + shape=out_shape, dim_order=DimOrder.PRESUMED_CONTIGUOUS + ) + + inputs = [None] * 2 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_vector(shape) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.RESHAPE, inputs, outputs) + + def add_flatten(self, node): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + + _start_ctype, start_dim = self.get_constant_value(node.inputsAt(1), "IntType") + _end_ctype, end_dim = self.get_constant_value(node.inputsAt(2), "IntType") + + # channels last with channels == 1 or (height & width both 1) + is_trivial_flatten = len(in_oper.shape) == 4 and ( + in_oper.shape[1] == 1 or (in_oper.shape[2] == 1 and in_oper.shape[3] == 1) + ) + if in_oper.dim_order != DimOrder.PRESUMED_CONTIGUOUS and not is_trivial_flatten: + raise Exception( # noqa: TRY002 + "Currently, flatten is not supported on NHWC tensors unless C=1 or H=W=1" + ) + + if start_dim < 0: + start_dim += len(in_oper.shape) + if end_dim < 0: + end_dim += len(in_oper.shape) + + out_shape = ( + in_oper.shape[:start_dim] + + (functools.reduce(operator.mul, in_oper.shape[start_dim : end_dim + 1]),) + + in_oper.shape[end_dim + 1 :] + ) + + if any(dim == 0 for dim in in_oper.shape[start_dim : end_dim + 1]): + raise Exception( # noqa: TRY002 + "Flattening flexible dims is not supported yet" + ) + non_flattened_dims = in_oper.shape[:start_dim] + in_oper.shape[end_dim + 1 :] + if non_flattened_dims.count(0) > 1: + raise Exception("Only 1 dim can be flexible") # noqa: TRY002 + + out_oper = in_oper._replace( + shape=out_shape, dim_order=DimOrder.PRESUMED_CONTIGUOUS + ) + out_id = self.add_tensor_operand(node.outputsAt(0), out_oper) + + for idx, dim in enumerate(out_shape): + if dim == 0: + self.forward_operand_shape(out_id, idx, in_id, in_oper.shape.index(0)) + + inputs_1 = tuple(dim if dim != 0 else -1 for dim in out_shape) + inputs = [None] * 2 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_vector(inputs_1) + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.RESHAPE, inputs, outputs) + + def add_slice(self, node): + if node.inputsSize() != 5: + raise AssertionError( + f"expected node.inputsSize() == 5, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + _, dim_value = self.get_constant_value(node.inputsAt(1)) + _, start_value = self.get_constant_value(node.inputsAt(2)) + _, stop_value = self.get_constant_value(node.inputsAt(3)) + _, step_value = self.get_constant_value(node.inputsAt(4)) + + if start_value is None: + start_value = 0 + if stop_value is None: + stop_value = sys.maxsize + + if start_value < 0: + start_value += in_oper.shape[dim_value] + elif start_value == sys.maxsize: + start_value = 0 + + if start_value == 0 and stop_value == sys.maxsize: + self._identity(node) + return + + if in_oper.shape[dim_value] == 0: + raise Exception("Unable to slice with flexible shape") # noqa: TRY002 + + if stop_value < 0: + stop_value += in_oper.shape[dim_value] + elif stop_value == sys.maxsize: + stop_value = in_oper.shape[dim_value] + + if start_value >= stop_value: + raise Exception( # noqa: TRY002 + "Slice start value should be less than stop value" + ) + + out_len = (stop_value - start_value) // step_value + out_shape = tuple( + out_len if i == dim_value else dim for i, dim in enumerate(in_oper.shape) + ) + out_id = self.add_tensor_operand( + node.outputsAt(0), in_oper._replace(shape=out_shape) + ) + + # flex inputs + end_mask = 0 + for idx, dim in enumerate(out_shape): + if dim == 0: + self.forward_operand_shape(out_id, idx, in_id, idx) + end_mask |= 1 << idx + + inputs = [None] * 7 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_vector( + [start_value if i == dim_value else 0 for i in range(len(in_oper.shape))] + ) + inputs[2] = self.add_immediate_int_vector( + [ + stop_value if i == dim_value else dim + for i, dim in enumerate(in_oper.shape) + ] + ) + inputs[3] = self.add_immediate_int_vector( + [step_value if i == dim_value else 1 for i in range(len(in_oper.shape))] + ) + inputs[4] = self.add_immediate_int_scalar(0) # begin mask + inputs[5] = self.add_immediate_int_scalar(end_mask) + inputs[6] = self.add_immediate_int_scalar(0) # shrink axis mask + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.STRIDED_SLICE, inputs, outputs) + + def add_size(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + _, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + _, value = self.constants[node.inputsAt(1)] + res = in_oper.shape[value] + output = node.outputsAt(0) + self.add_constant_value(output, output.type(), res) + + def add_cat(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + tensors = self.tensor_sequences[node.inputsAt(0)] + _, dim = self.get_constant_value(node.inputsAt(1), "IntType") + + if len(tensors) <= 0: + raise AssertionError(f"expected len(tensors) > 0, got {len(tensors)}") + in_ids = [] + out_oper = None + out_dim_size = 0 + for inp in tensors: + in_id, in_oper = self.get_tensor_operand_by_jitval(inp) + if out_oper is None: + out_shape = change_element(in_oper.shape, dim, -1) + out_oper = in_oper._replace(shape=out_shape) + if in_oper.op_type != out_oper.op_type: + raise AssertionError( + f"in_oper.op_type {in_oper.op_type} != out_oper.op_type {out_oper.op_type}" + ) + if in_oper.dim_order != out_oper.dim_order: + raise AssertionError( + f"in_oper.dim_order {in_oper.dim_order} != out_oper.dim_order {out_oper.dim_order}" + ) + if change_element(in_oper.shape, dim, -1) != change_element( + out_oper.shape, dim, -1 + ): + raise AssertionError( + f"shape mismatch: {change_element(in_oper.shape, dim, -1)} != {change_element(out_oper.shape, dim, -1)}" + ) + # TODO: Possibly check scale and zero point. + in_ids.append(in_id) + # TODO: Possibly support variable-sized inputs. + out_dim_size += in_oper.shape[dim] + + if out_oper is None: + raise AssertionError("out_oper must not be None") + out_oper = out_oper._replace( + shape=change_element(out_oper.shape, dim, out_dim_size) + ) + + if in_oper.dim_order == DimOrder.CHANNELS_LAST: # type: ignore[possibly-undefined] + if len(out_oper.shape) != 4: + raise AssertionError( + f"expected len(out_oper.shape) == 4 for CHANNELS_LAST, got {len(out_oper.shape)}" + ) + nnapi_dim = [0, 3, 1, 2][dim] + else: + nnapi_dim = dim + + out_id = self.add_tensor_operand(node.outputsAt(0), out_oper) + for idx, d in enumerate(out_oper.shape): + if d == 0: + if idx == dim: + shape = " + ".join(flex_name(ip_id, dim) for ip_id in in_ids) + self.compute_operand_shape(out_id, idx, shape) + else: + self.forward_operand_shape(out_id, idx, in_ids[0], idx) + + inputs = in_ids + [self.add_immediate_int_scalar(nnapi_dim)] + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.CONCATENATION, inputs, outputs) + + def add_mean(self, node): + if node.inputsSize() != 4: + raise AssertionError( + f"expected node.inputsSize() == 4, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + dim_ctype, dim = self.get_constant_value(node.inputsAt(1)) + if dim_ctype.kind() != "ListType": + raise AssertionError(f"expected dim_ctype ListType, got {dim_ctype.kind()}") + if dim_ctype.getElementType().kind() != "IntType": + raise AssertionError( + f"expected dim element type IntType, got {dim_ctype.getElementType().kind()}" + ) + _, keep_dim = self.get_constant_value(node.inputsAt(2), "BoolType") + # Expect None for dtype + self.get_constant_value(node.inputsAt(3), "NoneType") + + if in_oper.dim_order == DimOrder.CHANNELS_LAST: + if len(in_oper.shape) != 4: + raise AssertionError( + f"expected len(in_oper.shape) == 4 for CHANNELS_LAST, got {len(in_oper.shape)}" + ) + nnapi_dim = [[0, 3, 1, 2][d] for d in dim] + else: + nnapi_dim = dim + + collapsed_dims = set() + for d in dim: + if d < 0: + d += len(in_oper.shape) + collapsed_dims.add(d) + + if in_oper.dim_order == DimOrder.CHANNELS_LAST and not keep_dim: + if not collapsed_dims.issuperset({2, 3}): + raise AssertionError( + f"expected collapsed_dims to include {{2, 3}}, got {collapsed_dims}" + ) + out_dim_order = DimOrder.PRESUMED_CONTIGUOUS + else: + out_dim_order = in_oper.dim_order + + out_shape = [] + for i, s in enumerate(in_oper.shape): + if i not in collapsed_dims: + out_shape.append(s) + elif keep_dim: + out_shape.append(1) + + out_oper = in_oper._replace(shape=out_shape, dim_order=out_dim_order) + + inputs = [None] * 3 + inputs[0] = in_id + inputs[1] = self.add_immediate_int_vector(nnapi_dim) + inputs[2] = self.add_immediate_int_scalar(keep_dim) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.MEAN, inputs, outputs) + + def add_quantize(self, node): + if node.inputsSize() != 4: + raise AssertionError( + f"expected node.inputsSize() == 4, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + if in_oper.dim_order != DimOrder.CHANNELS_LAST: + raise Exception( # noqa: TRY002 + "Most hardware backends prefer NHWC quantized tensors. " + "Try setting `t.nnapi_nhwc = True` on your tensor inputs. " + ) + _, scale = self.get_constant_value(node.inputsAt(1), "FloatType") + _, zero_point = self.get_constant_value(node.inputsAt(2), "IntType") + _, scalar_type = self.get_constant_value(node.inputsAt(3), "IntType") + if scalar_type != TorchScalarTypes.QUINT8.value: + raise Exception( # noqa: TRY002 + "PyTorch NNAPI export only supports quantized tensors " + "with the quint8 dtype." + ) + op_type = NNAPI_OperandCode.TENSOR_QUANT8_ASYMM + + out_oper = in_oper._replace( + op_type=op_type, + scale=scale, + zero_point=zero_point, + ) + + inputs = [None] * 1 + inputs[0] = in_id + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.QUANTIZE, inputs, outputs) + + def add_dequantize(self, node): + if node.inputsSize() != 1: + raise AssertionError( + f"expected node.inputsSize() == 1, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + out_oper = in_oper._replace( + op_type=NNAPI_OperandCode.TENSOR_FLOAT32, + scale=0.0, + zero_point=0, + ) + + inputs = [None] * 1 + inputs[0] = in_id + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.DEQUANTIZE, inputs, outputs) + + def add_pointwise_simple_unary_op(self, node, opcode): + if node.inputsSize() != 1: + raise AssertionError( + f"expected node.inputsSize() == 1, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + + out_oper = in_oper + if opcode == NNAPI_OperationCode.LOGISTIC: + # NNAPI docs: For ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, the scale + # must be 1.f / 256 and the zeroPoint must be 0. + # https://fburl.com/h52stoog + if in_oper.op_type == NNAPI_OperandCode.TENSOR_QUANT8_ASYMM: + out_oper = in_oper._replace(zero_point=0, scale=1.0 / 256) + + out_id = self.add_tensor_operand(node.outputsAt(0), out_oper) + + for idx, dim in enumerate(in_oper.shape): + if dim == 0: + self.forward_operand_shape(out_id, idx, in_id, idx) + + inputs = [None] * 1 + inputs[0] = in_id + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(opcode, inputs, outputs) + + def _do_add_binary(self, node, opcode, fuse_code, *, qparams=None): + """Helper for pointwise binary broadcast ops with superfluous extra args.""" + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + if node.inputsAt(0).type().kind() != "TensorType": + raise AssertionError( + f"expected inputsAt(0) TensorType, got {node.inputsAt(0).type().kind()}" + ) + if node.inputsAt(1).type().kind() != "TensorType": + raise AssertionError( + f"expected inputsAt(1) TensorType, got {node.inputsAt(1).type().kind()}" + ) + + if self.has_operand_for_jitval(node.inputsAt(0)): + in0_id, in0_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + in1_id, in1_oper = self.get_tensor_operand_or_constant( + node.inputsAt(1), in0_oper.dim_order + ) + elif self.has_operand_for_jitval(node.inputsAt(1)): + in1_id, in1_oper = self.get_tensor_operand_by_jitval(node.inputsAt(1)) + in0_id, in0_oper = self.get_tensor_operand_or_constant( + node.inputsAt(0), in1_oper.dim_order + ) + else: + raise Exception( # noqa: TRY002 + f"Can't do a NNAPI binary op: {opcode} on two constants" + ) + + if in0_oper.op_type != in1_oper.op_type: + raise AssertionError( + f"in0_oper.op_type {in0_oper.op_type} != in1_oper.op_type {in1_oper.op_type}" + ) + in0_id, in0_oper, in1_id, in1_oper = self.transpose_for_broadcast( + in0_id, in0_oper, in1_id, in1_oper + ) + # NOTE: PyTorch and NNAPI have the same broadcast semantics. + out_shape = broadcast_shapes(in0_oper.shape, in1_oper.shape) + out_oper = in0_oper._replace(shape=out_shape) + if qparams is not None: + scale, zp = qparams + out_oper = out_oper._replace(scale=scale, zero_point=zp) + + out_id = self.add_tensor_operand(node.outputsAt(0), out_oper) + for idx, (d0, d1) in enumerate(zip(in0_oper.shape, in1_oper.shape)): + if d0 == 1 and d1 == 0: + self.forward_operand_shape(out_id, idx, in1_id, idx) + elif d0 == 0 and d1 == 1: + self.forward_operand_shape(out_id, idx, in0_id, idx) + elif d0 == 0 and d1 == 0: + self.flexible_shape_computation_lines.append( + f"assert {flex_name(in0_id, idx)} == {flex_name(in1_id, idx)}" + ) + self.forward_operand_shape(out_id, idx, in0_id, idx) + + inputs = [None] * 3 + inputs[0] = in0_id + inputs[1] = in1_id + inputs[2] = self.add_immediate_int_scalar(fuse_code) + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(opcode, inputs, outputs) + + def add_pointwise_simple_binary_broadcast_op(self, node, opcode, fuse_code): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + self._do_add_binary(node, opcode, fuse_code) + + def add_add_sub_op(self, node, opcode, fuse_code): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + + _, alpha = self.get_constant_value(node.inputsAt(2), "IntType") + if alpha != 1: + raise Exception( # noqa: TRY002 + "NNAPI does not support add/sub with alpha." + ) + + self._do_add_binary(node, opcode, fuse_code) + + def add_qadd(self, node, opcode, fuse_code): + if node.inputsSize() != 4: + raise AssertionError( + f"expected node.inputsSize() == 4, got {node.inputsSize()}" + ) + + _, scale = self.get_constant_value(node.inputsAt(2), "FloatType") + _, zero_point = self.get_constant_value(node.inputsAt(3), "IntType") + + self._do_add_binary(node, opcode, fuse_code, qparams=(scale, zero_point)) + + def add_softmax(self, node): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + in_id, in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + + _, softmax_dim = self.get_constant_value(node.inputsAt(1), "IntType") + + out_id = self.add_tensor_operand(node.outputsAt(0), in_oper) + for dim, size in enumerate(in_oper.shape): + if size == 0: + self.forward_operand_shape(out_id, dim, in_id, dim) + + inputs = [None] * 3 + inputs[0] = in_id + inputs[1] = self.add_immediate_float_scalar( + 1.0 + ) # positive scaling factor of exponent, beta + inputs[2] = self.add_immediate_int_scalar(softmax_dim) + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.SOFTMAX, inputs, outputs) + + def add_hardtanh(self, node): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval_fixed_size(node.inputsAt(0)) + _, min_val = self.get_constant_value(node.inputsAt(1), "FloatType") + _, max_val = self.get_constant_value(node.inputsAt(2), "FloatType") + + op_map = { + (-1, 1): NNAPI_OperationCode.RELU1, + (0, 6): NNAPI_OperationCode.RELU6, # noqa: E201 + } + + opcode = op_map.get((min_val, max_val)) + if opcode is None: + raise Exception( # noqa: TRY002 + "NNAPI only supports hardtanh with args (-1, 1) or (0, 6)." + ) + + inputs = [None] * 1 + inputs[0] = in_id + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), in_oper) + + self.add_operation(opcode, inputs, outputs) + + def add_prelu_op(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + if node.inputsAt(0).type().kind() != "TensorType": + raise AssertionError( + f"expected inputsAt(0) TensorType, got {node.inputsAt(0).type().kind()}" + ) + if node.inputsAt(1).type().kind() != "TensorType": + raise AssertionError( + f"expected inputsAt(1) TensorType, got {node.inputsAt(1).type().kind()}" + ) + + in_id, in_oper = self.get_tensor_operand_by_jitval(node.inputsAt(0)) + w_id, w_oper = self.get_tensor_operand_for_weight(node.inputsAt(1)) + if len(w_oper.shape) != 1: + raise AssertionError( + f"expected len(w_oper.shape) == 1, got {len(w_oper.shape)}" + ) + if w_oper.shape[0] <= 0: + raise AssertionError(f"expected w_oper.shape[0] > 0, got {w_oper.shape[0]}") + if w_oper.shape[0] > 1: + if in_oper.use_nchw(): + # TODO: Support this by adding trailing 1 dims. + raise Exception( # noqa: TRY002 + "Per-channel PReLU only supports channels_last right now." + ) + + out_id = self.add_tensor_operand(node.outputsAt(0), in_oper) + for dim, size in enumerate(in_oper.shape): + if size > 0: + pass + elif dim <= 1: + raise Exception( # noqa: TRY002 + "PReLU requires fixed size for dim 0 and dim 1." + ) + else: + self.forward_operand_shape(out_id, dim, in_id, dim) + + inputs = [None] * 2 + inputs[0] = in_id + inputs[1] = w_id + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.PRELU, inputs, outputs) + + def add_pool2d_node(self, node, opcode): + if node.inputsSize() != 6: + raise AssertionError( + f"expected node.inputsSize() == 6, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + image, kernel, stride, padding, dilation, _ceil_mode = node.inputs() + + stride = stride or kernel + + # TODO: Validate ceil_mode semantics. + + args = self.get_conv_pool_args_2d_from_jit( + self.get_size_arg(kernel), stride, padding, dilation + ) + if args.dilation_h != 1 or args.dilation_w != 1: + raise Exception("NNAPI does not support dilated pooling.") # noqa: TRY002 + + image_id, image_oper = self.get_tensor_operand_by_jitval_fixed_size(image) + if len(image_oper.shape) != 4: + raise AssertionError( + f"expected len(image_oper.shape) == 4, got {len(image_oper.shape)}" + ) + + out_shape = get_conv_pool_shape( + image_oper.shape, args, image_oper.shape[1], False + ) + use_nchw = image_oper.use_nchw() + + inputs = [None] * 11 + inputs[0] = image_id + inputs[1] = self.add_immediate_int_scalar(args.pad_l) + inputs[2] = self.add_immediate_int_scalar(args.pad_r) + inputs[3] = self.add_immediate_int_scalar(args.pad_t) + inputs[4] = self.add_immediate_int_scalar(args.pad_b) + inputs[5] = self.add_immediate_int_scalar(args.stride_w) + inputs[6] = self.add_immediate_int_scalar(args.stride_h) + inputs[7] = self.add_immediate_int_scalar(args.kernel_w) + inputs[8] = self.add_immediate_int_scalar(args.kernel_h) + inputs[9] = self.add_immediate_int_scalar(NNAPI_FuseCode.FUSED_NONE) + inputs[10] = self.add_immediate_bool_scalar(use_nchw) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand( + node.outputsAt(0), image_oper._replace(shape=out_shape) + ) + + self.add_operation(opcode, inputs, outputs) + + def add_avg_pool2d(self, node): + if node.inputsSize() != 7: + raise AssertionError( + f"expected node.inputsSize() == 7, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + ( + image, + kernel, + stride, + padding, + _ceil_mode, + count_include_pad, + divisor_override, + ) = node.inputs() + + _, count_include_pad_value = self.get_constant_value(count_include_pad) + _, divisor_override_value = self.get_constant_value(divisor_override) + if not count_include_pad_value or divisor_override_value: + raise Exception( # noqa: TRY002 + "NNAPI doesn't support count_include_pad=False or divisor_override" + ) + + args = self.get_conv_pool_args_2d_from_jit( + self.get_size_arg(kernel), stride, padding + ) + + image_id, image_oper = self.get_tensor_operand_by_jitval(image) + if len(image_oper.shape) != 4: + raise AssertionError( + f"expected len(image_oper.shape) == 4, got {len(image_oper.shape)}" + ) + + out_shape = get_conv_pool_shape( + image_oper.shape, args, image_oper.shape[1], False + ) + use_nchw = image_oper.use_nchw() + + inputs = [None] * 11 + inputs[0] = image_id + inputs[1] = self.add_immediate_int_scalar(args.pad_l) + inputs[2] = self.add_immediate_int_scalar(args.pad_r) + inputs[3] = self.add_immediate_int_scalar(args.pad_t) + inputs[4] = self.add_immediate_int_scalar(args.pad_b) + inputs[5] = self.add_immediate_int_scalar(args.stride_w) + inputs[6] = self.add_immediate_int_scalar(args.stride_h) + inputs[7] = self.add_immediate_int_scalar(args.kernel_w) + inputs[8] = self.add_immediate_int_scalar(args.kernel_h) + inputs[9] = self.add_immediate_int_scalar(NNAPI_FuseCode.FUSED_NONE) + inputs[10] = self.add_immediate_bool_scalar(use_nchw) + + outputs = [None] * 1 + out_id = self.add_tensor_operand( + node.outputsAt(0), image_oper._replace(shape=out_shape) + ) + self._handle_conv_pool_flexible_input(out_id, image, args, False) + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.AVERAGE_POOL_2D, inputs, outputs) + + def add_adaptive_avg_pool2d(self, node): + if node.inputsSize() != 2: + raise AssertionError( + f"expected node.inputsSize() == 2, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + image_id, image_oper = self.get_tensor_operand_by_jitval_fixed_size( + node.inputsAt(0) + ) + if len(image_oper.shape) != 4: + raise AssertionError( + f"expected len(image_oper.shape) == 4, got {len(image_oper.shape)}" + ) + + size_ctype, size_arg = self.get_constant_value(node.inputsAt(1)) + if size_ctype.kind() != "ListType": + raise AssertionError( + f"expected size_ctype ListType, got {size_ctype.kind()}" + ) + if size_ctype.getElementType().kind() != "IntType": + raise AssertionError( + f"expected size element type IntType, got {size_ctype.getElementType().kind()}" + ) + if size_arg != [1, 1]: + raise Exception( # noqa: TRY002 + "NNAPI only supports adaptive_avg_pool2d with output size (1, 1)." + ) + + out_shape = image_oper.shape[0:2] + tuple(size_arg) + use_nchw = image_oper.use_nchw() + + inputs = [None] * 11 + inputs[0] = image_id + inputs[1] = self.add_immediate_int_scalar(0) + inputs[2] = self.add_immediate_int_scalar(0) + inputs[3] = self.add_immediate_int_scalar(0) + inputs[4] = self.add_immediate_int_scalar(0) + inputs[5] = self.add_immediate_int_scalar(1) + inputs[6] = self.add_immediate_int_scalar(1) + inputs[7] = self.add_immediate_int_scalar(image_oper.shape[3]) + inputs[8] = self.add_immediate_int_scalar(image_oper.shape[2]) + inputs[9] = self.add_immediate_int_scalar(NNAPI_FuseCode.FUSED_NONE) + inputs[10] = self.add_immediate_bool_scalar(use_nchw) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand( + node.outputsAt(0), image_oper._replace(shape=out_shape) + ) + + self.add_operation(NNAPI_OperationCode.AVERAGE_POOL_2D, inputs, outputs) + + def add_upsample_nearest2d(self, node): + if not (node.inputsSize() == 3 or node.inputsSize() == 4): + raise AssertionError( + f"expected node.inputsSize() == 3 or 4, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + if node.inputsSize() == 3: + image, size_jit, scale_jit = node.inputs() + else: + image, size_jit, scale_h_jit, scale_w_jit = node.inputs() + size_ctype, size_arg = self.get_constant_value(size_jit) + + if node.inputsSize() == 3: + scale_ctype, scale_arg = self.get_constant_value(scale_jit) # type: ignore[possibly-undefined] + else: + scale_h_ctype, scale_h_arg = self.get_constant_value(scale_h_jit) # type: ignore[possibly-undefined] + scale_w_ctype, _scale_w_arg = self.get_constant_value(scale_w_jit) # type: ignore[possibly-undefined] + + # The only way for the 4-argument overload of upsample_nearest2d to + # have been added to the graph without error is if the scale_h and + # scale_w arguments are None + if scale_h_ctype.kind() != "NoneType": + raise AssertionError( + f"expected scale_h_ctype NoneType, got {scale_h_ctype.kind()}" + ) + if scale_w_ctype.kind() != "NoneType": + raise AssertionError( + f"expected scale_w_ctype NoneType, got {scale_w_ctype.kind()}" + ) + + scale_ctype = scale_h_ctype + scale_arg = scale_h_arg + + image_id, image_oper = self.get_tensor_operand_by_jitval(image) + if len(image_oper.shape) != 4: + raise AssertionError( + f"expected len(image_oper.shape) == 4, got {len(image_oper.shape)}" + ) + + if size_ctype.kind() != "NoneType" and scale_ctype.kind() != "NoneType": + raise Exception("Size and scale cannot both be non-None.") # noqa: TRY002 + elif size_ctype.kind() != "NoneType": + if size_ctype.kind() != "ListType": + raise AssertionError( + f"expected size_ctype ListType, got {size_ctype.kind()}" + ) + if size_ctype.getElementType().kind() != "IntType": + raise AssertionError( + f"expected size element type IntType, got {size_ctype.getElementType().kind()}" + ) + if scale_ctype.kind() != "NoneType": + raise AssertionError( + f"expected scale_ctype NoneType, got {scale_ctype.kind()}" + ) + if scale_arg is not None: + raise AssertionError(f"expected scale_arg None, got {scale_arg}") + if not isinstance(size_arg, list): + raise AssertionError( + f"expected size_arg to be list, got {type(size_arg)}" + ) + if not size_arg: + raise AssertionError("expected size_arg to be non-empty") + if not all(isinstance(val, int) for val in size_arg): + raise AssertionError("expected all size_arg values to be int") + if len(size_arg) == 1: + size_arg = size_arg * 2 + if len(size_arg) != 2: + raise AssertionError( + f"expected len(size_arg) == 2, got {len(size_arg)}" + ) + out_h = size_arg[0] + out_w = size_arg[1] + arg_h = self.add_immediate_int_scalar(out_h) + arg_w = self.add_immediate_int_scalar(out_w) + elif scale_ctype.kind() != "NoneType": + if scale_ctype.kind() != "ListType": + raise AssertionError( + f"expected scale_ctype ListType, got {scale_ctype.kind()}" + ) + if scale_ctype.getElementType().kind() != "FloatType": + raise AssertionError( + f"expected scale element type FloatType, got {scale_ctype.getElementType().kind()}" + ) + if size_ctype.kind() != "NoneType": + raise AssertionError( + f"expected size_ctype NoneType, got {size_ctype.kind()}" + ) + if size_arg is not None: + raise AssertionError(f"expected size_arg None, got {size_arg}") + if not isinstance(scale_arg, list): + raise AssertionError( + f"expected scale_arg to be list, got {type(scale_arg)}" + ) + if not scale_arg: + raise AssertionError("expected scale_arg to be non-empty") + if not all(isinstance(val, float) for val in scale_arg): + raise AssertionError("expected all scale_arg values to be float") + if len(scale_arg) == 1: + scale_arg = scale_arg * 2 + if len(scale_arg) != 2: + raise AssertionError( + f"expected len(scale_arg) == 2, got {len(scale_arg)}" + ) + out_h = int(scale_arg[0] * image_oper.shape[2]) + out_w = int(scale_arg[1] * image_oper.shape[3]) + arg_h = self.add_immediate_float_scalar(scale_arg[0]) + arg_w = self.add_immediate_float_scalar(scale_arg[1]) + else: + raise Exception("Size and scale cannot both be None.") # noqa: TRY002 + + out_shape = (image_oper.shape[0], image_oper.shape[1], out_h, out_w) + use_nchw = image_oper.use_nchw() + out_id = self.add_tensor_operand( + node.outputsAt(0), image_oper._replace(shape=out_shape) + ) + + if image_oper.shape[0] == 0 or image_oper.shape[1] == 0: + raise Exception("Flexible batch or channels not supported") # noqa: TRY002 + + # Handle variable input size + for dim in (2, 3): # h, w indices + if image_oper.shape[dim] == 0: + if size_ctype.kind() != "NoneType": + # pyrefly: ignore [unsupported-operation] + self.compute_operand_shape(out_id, dim, size_arg[dim - 2]) + elif scale_ctype.kind() != "NoneType": + self.compute_operand_shape( + out_id, + dim, + # pyrefly: ignore [unsupported-operation] + f"int({scale_arg[dim - 2]} * {flex_name(image_id, dim)})", + ) + else: + raise Exception( # noqa: TRY002 + "Size and scale cannot both be None." + ) + + inputs = [None] * 4 + inputs[0] = image_id + inputs[1] = arg_w + inputs[2] = arg_h + inputs[3] = self.add_immediate_bool_scalar(use_nchw) + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.RESIZE_NEAREST_NEIGHBOR, inputs, outputs) + + def add_addmm(self, node): + if node.inputsSize() != 5: + raise AssertionError( + f"expected node.inputsSize() == 5, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + jit_bias, jit_input, jit_weight, jit_beta, jit_alpha = node.inputs() + + for jitval in (jit_beta, jit_alpha): + scale_ctype, scale_value = self.get_constant_value(jitval) + if scale_ctype.kind() not in ("IntType", "FloatType"): + raise AssertionError( + f"expected scale_ctype IntType or FloatType, got {scale_ctype.kind()}" + ) + if scale_value != 1: + raise Exception( # noqa: TRY002 + "NNAPI Fully-Connected does not support alpha and beta." + ) + + self.add_addmm_or_linear(node, True, jit_input, jit_weight, jit_bias) + + def add_linear(self, node): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + jit_input, jit_weight, jit_bias = node.inputs() + + self.add_addmm_or_linear(node, False, jit_input, jit_weight, jit_bias) + + def add_addmm_or_linear( + self, node, transpose_weight, jit_input, jit_weight, jit_bias + ): + input_id, input_oper = self.get_tensor_operand_by_jitval(jit_input) + bias_id, bias_oper = self.get_tensor_operand_for_weight(jit_bias) + + if len(input_oper.shape) != 2: + raise AssertionError( + f"expected len(input_oper.shape) == 2, got {len(input_oper.shape)}" + ) + if len(bias_oper.shape) != 1: + raise AssertionError( + f"expected len(bias_oper.shape) == 1, got {len(bias_oper.shape)}" + ) + + # TODO: Transform at load time to share weights with CPU model. + _, weight_tensor = self.get_constant_value(jit_weight, "TensorType") + if len(weight_tensor.shape) != 2: + raise AssertionError( + f"expected len(weight_tensor.shape) == 2, got {len(weight_tensor.shape)}" + ) + if transpose_weight: + nnapi_weight_tensor = weight_tensor.t().contiguous() + else: + nnapi_weight_tensor = weight_tensor.contiguous() + weight_id = self.add_tensor_operand_for_weight(nnapi_weight_tensor) + weight_oper = self.operands[weight_id] + + out_shape = (input_oper.shape[0], weight_oper.shape[0]) + out_id = self.add_tensor_operand( + node.outputsAt(0), input_oper._replace(shape=out_shape) + ) + + if input_oper.shape[0] == 0: + self.forward_operand_shape(out_id, 0, input_id, 0) + + inputs = [None] * 4 + inputs[0] = input_id + inputs[1] = weight_id + inputs[2] = bias_id + inputs[3] = self.add_immediate_int_scalar(NNAPI_FuseCode.FUSED_NONE) + + outputs = [None] * 1 + outputs[0] = out_id + + self.add_operation(NNAPI_OperationCode.FULLY_CONNECTED, inputs, outputs) + + def add_qlinear(self, node): + if node.inputsSize() != 4: + raise AssertionError( + f"expected node.inputsSize() == 4, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + ( + jit_input, + jit_packed_weight, + jit_scale, + jit_zero_point, + ) = node.inputs() + + input_id, input_oper = self.get_tensor_operand_by_jitval_fixed_size(jit_input) + # TODO: Support automatic reshape + if len(input_oper.shape) != 2: + raise AssertionError( + f"expected len(input_oper.shape) == 2, got {len(input_oper.shape)}" + ) + + _, out_scale = self.get_constant_value(jit_scale, "FloatType") + _, out_zero_point = self.get_constant_value(jit_zero_point, "IntType") + weight_ctype, packed_weight = self.get_constant_value(jit_packed_weight) + if weight_ctype.name() != "LinearPackedParamsBase": + raise AssertionError( + f"expected weight_ctype LinearPackedParamsBase, got {weight_ctype.name()}" + ) + raw_weight, raw_bias = packed_weight.__getstate__()[0] + if raw_bias is None: + raise AssertionError("raw_bias must not be None") + + if len(raw_weight.shape) != 2: + raise AssertionError( + f"expected len(raw_weight.shape) == 2, got {len(raw_weight.shape)}" + ) + if len(raw_bias.shape) != 1: + raise AssertionError( + f"expected len(raw_bias.shape) == 1, got {len(raw_bias.shape)}" + ) + if raw_bias.shape[0] != raw_weight.shape[0]: + raise AssertionError( + f"raw_bias.shape[0] {raw_bias.shape[0]} != raw_weight.shape[0] {raw_weight.shape[0]}" + ) + if raw_weight.shape[1] != input_oper.shape[1]: + raise AssertionError( + f"raw_weight.shape[1] {raw_weight.shape[1]} != input_oper.shape[1] {input_oper.shape[1]}" + ) + + if raw_weight.qscheme() != torch.per_tensor_affine: + raise AssertionError( + f"expected raw_weight.qscheme() per_tensor_affine, got {raw_weight.qscheme()}" + ) + if raw_weight.dtype == torch.quint8: + unsigned_weight = raw_weight + else: + if raw_weight.dtype != torch.qint8: + raise AssertionError( + f"expected raw_weight.dtype qint8, got {raw_weight.dtype}" + ) + unsigned_weight = torch._make_per_tensor_quantized_tensor( + (raw_weight.int_repr().int() + 128).to(torch.uint8), + scale=raw_weight.q_scale(), + zero_point=raw_weight.q_zero_point() + 128, + ) + weight_scale = unsigned_weight.q_scale() + bias_scale = input_oper.scale * weight_scale + int_bias = torch.quantize_per_tensor(raw_bias, bias_scale, 0, torch.qint32) + bias_id = self.add_tensor_operand_for_weight(int_bias) + + multiplier = input_oper.scale * weight_scale / out_scale + if multiplier <= 0: + raise AssertionError(f"expected multiplier > 0, got {multiplier}") + if multiplier >= 1: + raise Exception( # noqa: TRY002 + "Quantized convolution multiplier is greater than 1. " + "This is supported by NNAPI, but not by most hardware backends. " + "Try training a model without quantization-aware training. " + ) + + # TODO: Transform at load time to share weights with CPU model. + nnapi_weight_tensor = unsigned_weight.contiguous() + weight_id = self.add_tensor_operand_for_weight(nnapi_weight_tensor) + weight_oper = self.operands[weight_id] + + out_shape = (input_oper.shape[0], weight_oper.shape[0]) + out_oper = input_oper._replace( + shape=out_shape, + scale=out_scale, + zero_point=out_zero_point, + ) + + inputs = [None] * 4 + inputs[0] = input_id + inputs[1] = weight_id + inputs[2] = bias_id + inputs[3] = self.add_immediate_int_scalar(NNAPI_FuseCode.FUSED_NONE) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand(node.outputsAt(0), out_oper) + + self.add_operation(NNAPI_OperationCode.FULLY_CONNECTED, inputs, outputs) + + def get_optional_bias(self, jit_bias, weight_tensor, transpose=False): + ctype, _value = self.get_constant_value(jit_bias) + if ctype.kind() == "NoneType": + bias_idx = 1 if transpose else 0 + nnapi_bias_tensor = torch.zeros( + weight_tensor.size()[bias_idx], dtype=weight_tensor.dtype + ) + bias_id = self.add_tensor_operand_for_weight(nnapi_bias_tensor) + bias_oper = self.operands[bias_id] + return bias_id, bias_oper + else: + return self.get_tensor_operand_for_weight(jit_bias) + + def add_conv2d(self, node): + if node.inputsSize() != 7: + raise AssertionError( + f"expected node.inputsSize() == 7, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + ( + jit_image, + jit_weight, + jit_bias, + jit_stride, + jit_pad, + jit_dilation, + jit_groups, + ) = node.inputs() + + _, weight_tensor = self.get_constant_value(jit_weight, "TensorType") + bias_id, _bias_oper = self.get_optional_bias(jit_bias, weight_tensor) + args = self.get_conv_pool_args_2d_from_jit( + weight_tensor.shape[2:4], jit_stride, jit_pad, jit_dilation, jit_groups + ) + + return self.add_conv2d_common( + node.outputsAt(0), + 0.0, + 0, + jit_image, + weight_tensor, + bias_id, + args, + False, # transpose + NNAPI_FuseCode.FUSED_NONE, + ) + + def add_conv_underscore(self, node): + if node.inputsSize() != 13: + raise AssertionError( + f"expected node.inputsSize() == 13, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + ( + jit_image, + jit_weight, + jit_bias, + jit_stride, + jit_pad, + jit_dilation, + jit_transpose, + _, + jit_groups, + _, + _, + _, + _, + ) = node.inputs() + + _, weight_tensor = self.get_constant_value(jit_weight, "TensorType") + _, transpose = self.get_constant_value(jit_transpose) + bias_id, _bias_oper = self.get_optional_bias(jit_bias, weight_tensor, transpose) + args = self.get_conv_pool_args_2d_from_jit( + weight_tensor.shape[2:4], jit_stride, jit_pad, jit_dilation, jit_groups + ) + + return self.add_conv2d_common( + node.outputsAt(0), + 0.0, + 0, + jit_image, + weight_tensor, + bias_id, + args, + transpose, + NNAPI_FuseCode.FUSED_NONE, + ) + + def add_log_softmax(self, node): + if node.inputsSize() != 3: + raise AssertionError( + f"expected node.inputsSize() == 3, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + jit_input, jit_dim, _jit_half_to_float = node.inputs() + input_id, input_oper = self.get_tensor_operand_by_jitval_fixed_size(jit_input) + _, dim = self.get_constant_value(jit_dim, "IntType") + + out_shape = input_oper.shape + + inputs = [None] * 3 + inputs[0] = input_id + # specifying 1 as the scaling factor for the exponent, beta + inputs[1] = self.add_immediate_float_scalar(1) + inputs[2] = self.add_immediate_int_scalar(dim) + + outputs = [None] * 1 + outputs[0] = self.add_tensor_operand( + node.outputsAt(0), input_oper._replace(shape=out_shape) + ) + self.add_operation(NNAPI_OperationCode.LOG_SOFTMAX, inputs, outputs) + + def add_qconv2d(self, node, fuse_code, transpose=False): + if node.inputsSize() != 4: + raise AssertionError( + f"expected node.inputsSize() == 4, got {node.inputsSize()}" + ) + if node.outputsSize() != 1: + raise AssertionError( + f"expected node.outputsSize() == 1, got {node.outputsSize()}" + ) + + ( + jit_image, + jit_packed_weight, + jit_scale, + jit_zero_point, + ) = node.inputs() + + _, out_scale = self.get_constant_value(jit_scale, "FloatType") + _, out_zero_point = self.get_constant_value(jit_zero_point, "IntType") + weight_ctype, packed_weight = self.get_constant_value(jit_packed_weight) + if weight_ctype.name() != "Conv2dPackedParamsBase": + raise AssertionError( + f"expected weight_ctype Conv2dPackedParamsBase, got {weight_ctype.name()}" + ) + ( + pack_version, + tensors, + opt_tensors, + ) = packed_weight.__getstate__()[0] + if pack_version != "2": + raise AssertionError(f"expected pack_version '2', got {pack_version!r}") + packed_config, raw_weight = tensors + (raw_bias,) = opt_tensors + if raw_bias is None: + raise AssertionError("raw_bias must not be None") + args = self.get_conv_pool_args_2d_from_pack( + raw_weight.shape[2:4], packed_config + ) + + if raw_weight.qscheme() != torch.per_tensor_affine: + raise AssertionError( + f"expected raw_weight.qscheme() per_tensor_affine, got {raw_weight.qscheme()}" + ) + if raw_weight.dtype == torch.quint8: + unsigned_weight = raw_weight + else: + if raw_weight.dtype != torch.qint8: + raise AssertionError( + f"expected raw_weight.dtype qint8, got {raw_weight.dtype}" + ) + unsigned_weight = torch._make_per_tensor_quantized_tensor( + (raw_weight.int_repr().int() + 128).to(torch.uint8), + scale=raw_weight.q_scale(), + zero_point=raw_weight.q_zero_point() + 128, + ) + weight_scale = unsigned_weight.q_scale() + _, image_oper = self.get_tensor_operand_by_jitval(jit_image) + bias_scale = image_oper.scale * weight_scale + int_bias = torch.quantize_per_tensor(raw_bias, bias_scale, 0, torch.qint32) + bias_id = self.add_tensor_operand_for_weight(int_bias) + + multiplier = image_oper.scale * weight_scale / out_scale + if multiplier <= 0: + raise AssertionError(f"expected multiplier > 0, got {multiplier}") + if multiplier >= 1: + raise Exception( # noqa: TRY002 + "Quantized convolution multiplier is greater than 1. " + "This is supported by NNAPI, but not by most hardware backends. " + "Try training a model without quantization-aware training. " + ) + + return self.add_conv2d_common( + node.outputsAt(0), + out_scale, + out_zero_point, + jit_image, + unsigned_weight, + bias_id, + args, + transpose, + fuse_code, + ) + + def add_conv2d_common( + self, + jit_out, + out_scale, + out_zero_point, + jit_image, + weight_tensor, + bias_id, + args, + transpose, + fuse_code, + ): + image_id, image_oper = self.get_tensor_operand_by_jitval(jit_image) + in_c = image_oper.shape[1] + + if args.group == 1: + # Full convolution + depthwise = False + if transpose: + weight_permutation = (1, 2, 3, 0) + else: + weight_permutation = (0, 2, 3, 1) + elif args.group == in_c: + # Depthwise convolution + depthwise = True + weight_permutation = (1, 2, 3, 0) + else: + raise Exception("Group convolution not supported yet.") # noqa: TRY002 + + # TODO: Transform at load time to share weights with CPU model. + nnapi_weight_tensor = weight_tensor.permute(*weight_permutation).contiguous() + weight_id = self.add_tensor_operand_for_weight(nnapi_weight_tensor) + weight_oper = self.operands[weight_id] + + bias_oper = self.operands[bias_id] + + if image_oper.op_type == NNAPI_OperandCode.TENSOR_FLOAT32: + if weight_oper.op_type != NNAPI_OperandCode.TENSOR_FLOAT32: + raise AssertionError( + f"expected weight_oper TENSOR_FLOAT32, got {weight_oper.op_type}" + ) + if bias_oper.op_type != NNAPI_OperandCode.TENSOR_FLOAT32: + raise AssertionError( + f"expected bias_oper TENSOR_FLOAT32, got {bias_oper.op_type}" + ) + elif image_oper.op_type == NNAPI_OperandCode.TENSOR_QUANT8_ASYMM: + if weight_oper.op_type != NNAPI_OperandCode.TENSOR_QUANT8_ASYMM: + raise AssertionError( + f"expected weight_oper TENSOR_QUANT8_ASYMM, got {weight_oper.op_type}" + ) + if bias_oper.op_type != NNAPI_OperandCode.TENSOR_INT32: + raise AssertionError( + f"expected bias_oper TENSOR_INT32, got {bias_oper.op_type}" + ) + if not approx_equal(image_oper.scale * weight_oper.scale, bias_oper.scale): + raise AssertionError( + f"scale mismatch: image*weight scale {image_oper.scale * weight_oper.scale} != bias scale {bias_oper.scale}" + ) + if bias_oper.zero_point != 0: + raise AssertionError( + f"expected bias_oper.zero_point == 0, got {bias_oper.zero_point}" + ) + else: + raise Exception( # noqa: TRY002 + f"Unsupported input type for conv2d: {image_oper.op_type}" + ) + + if len(image_oper.shape) != 4: + raise AssertionError( + f"expected len(image_oper.shape) == 4, got {len(image_oper.shape)}" + ) + if len(weight_oper.shape) != 4: + raise AssertionError( + f"expected len(weight_oper.shape) == 4, got {len(weight_oper.shape)}" + ) + if len(bias_oper.shape) != 1: + raise AssertionError( + f"expected len(bias_oper.shape) == 1, got {len(bias_oper.shape)}" + ) + + if depthwise: + # Depthwise convolution + one, _kern_h, _kern_w, out_c = weight_oper.shape + if one != 1: + raise AssertionError(f"expected weight_oper.shape[0] == 1, got {one}") + if out_c % in_c != 0: + raise AssertionError(f"out_c {out_c} must be divisible by in_c {in_c}") + channel_multiplier = out_c // in_c + if channel_multiplier != 1: + raise AssertionError( + f"channel_multiplier must be 1, got {channel_multiplier}" + ) + if out_c != in_c: + raise AssertionError(f"out_c {out_c} != in_c {in_c}") + else: + # Full convolution + out_c, _kern_h, _kern_w, kern_d = weight_oper.shape + if kern_d != in_c: + raise AssertionError(f"kern_d {kern_d} != in_c {in_c}") + + if out_c != bias_oper.shape[0]: + raise AssertionError( + f"out_c {out_c} != bias_oper.shape[0] {bias_oper.shape[0]}" + ) + + use_nchw = image_oper.use_nchw() + + if depthwise: + num_args = 12 + opcode = NNAPI_OperationCode.DEPTHWISE_CONV_2D + else: + num_args = 11 + if transpose: + opcode = NNAPI_OperationCode.TRANSPOSE_CONV_2D + else: + opcode = NNAPI_OperationCode.CONV_2D + + inputs = [None] * num_args + inputs[0] = image_id + inputs[1] = weight_id + inputs[2] = bias_id + inputs[3] = self.add_immediate_int_scalar(args.pad_l) + inputs[4] = self.add_immediate_int_scalar(args.pad_r) + inputs[5] = self.add_immediate_int_scalar(args.pad_t) + inputs[6] = self.add_immediate_int_scalar(args.pad_b) + inputs[7] = self.add_immediate_int_scalar(args.stride_w) + inputs[8] = self.add_immediate_int_scalar(args.stride_h) + if depthwise: + inputs[9] = self.add_immediate_int_scalar(1) + inputs[10] = self.add_immediate_int_scalar(fuse_code) + inputs[11] = self.add_immediate_bool_scalar(use_nchw) + else: + inputs[9] = self.add_immediate_int_scalar(fuse_code) + inputs[10] = self.add_immediate_bool_scalar(use_nchw) + + outputs = [None] * 1 + out_shape = get_conv_pool_shape(image_oper.shape, args, out_c, transpose) + out_oper = image_oper._replace( + shape=out_shape, + scale=out_scale, + zero_point=out_zero_point, + ) + out_id = self.add_tensor_operand(jit_out, out_oper) + self._handle_conv_pool_flexible_input(out_id, jit_image, args, transpose) + + outputs[0] = out_id + self.add_operation(opcode, inputs, outputs) + + def _handle_conv_pool_flexible_input(self, out_id, jit_image, args, transpose): + image_id, image_oper = self.get_tensor_operand_by_jitval(jit_image) + batch, in_ch, in_h, in_w = image_oper.shape + + if batch == 0: + self.forward_operand_shape(out_id, 0, image_id, 0) + if in_ch == 0: + raise Exception("Input channels can't be flexible") # noqa: TRY002 + # H & W + if transpose: + if in_h == 0: + self.compute_operand_shape( + out_id, + 2, + f"({flex_name(image_id, 2)} - 1) * {args.stride_h} + {args.kernel_h} - {args.pad_t} - {args.pad_b}", + ) + if in_w == 0: + self.compute_operand_shape( + out_id, + 3, + f"({flex_name(image_id, 3)} - 1) * {args.stride_w} + {args.kernel_w} - {args.pad_l} - {args.pad_r}", + ) + else: + if in_h == 0: + self.compute_operand_shape( + out_id, + 2, + f"({flex_name(image_id, 2)} - {args.kernel_h} + {args.pad_t} + {args.pad_b}) // {args.stride_h} + 1", + ) + if in_w == 0: + self.compute_operand_shape( + out_id, + 3, + f"({flex_name(image_id, 3)} - {args.kernel_w} + {args.pad_l} + {args.pad_r}) // {args.stride_w} + 1", + ) + + +def serialize_model( + module, inputs, *, config=None, return_shapes=None, use_int16_for_qint16=False +): + """Convert to NNAPI and serialize torchscript module. + + Parameters: + module: Torchscript module to convert + inputs: Tensors used to specify input details for NNAPI + config (optional): Optional config to attach to module + return_shapes (optional): Specify shape of outputs if + your module uses runtime flexible shapes to set output + buffer size for NNAPI + use_int16_for_qint16 (optional): Use Pytorch int16 to represent NNAPI qint16 values + """ + return _NnapiSerializer(config, use_int16_for_qint16).serialize_model( + module, inputs, return_shapes + ) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cpu/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/cpu/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..82dc52cd4904c1cda023c876c586550a5a33ff7a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/cpu/__init__.py @@ -0,0 +1,21 @@ +import torch + + +__all__ = [ + "get_cpu_capability", +] + + +def get_cpu_capability() -> str: + r"""Return cpu capability as a string value. + + Possible values: + - "DEFAULT" + - "VSX" + - "Z VECTOR" + - "NO AVX" + - "AVX2" + - "AVX512" + - "SVE256" + """ + return torch._C._get_cpu_capability() diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cpu/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/cpu/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..37eb6c4f51c2ba1a5f533cfffe93d27baaca8ed5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/cpu/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cuda/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/cuda/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..6b0f7f7fe94857c45574a8132ed418d54645803d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/cuda/__init__.py @@ -0,0 +1,715 @@ +# mypy: allow-untyped-defs +import contextlib +from typing import Any +from typing_extensions import deprecated + +import torch + + +__all__ = [ + "is_built", + "cuFFTPlanCacheAttrContextProp", + "cuFFTPlanCache", + "cuFFTPlanCacheManager", + "cuBLASModule", + "preferred_linalg_library", + "preferred_blas_library", + "cublas_workspace_size", + "cublaslt_workspace_size", + "blas_workspace_size", + "preferred_rocm_fa_library", + "is_ck_sdpa_available", + "cufft_plan_cache", + "matmul", + "SDPAParams", + "enable_cudnn_sdp", + "cudnn_sdp_enabled", + "enable_flash_sdp", + "flash_sdp_enabled", + "enable_mem_efficient_sdp", + "mem_efficient_sdp_enabled", + "math_sdp_enabled", + "enable_math_sdp", + "allow_fp16_bf16_reduction_math_sdp", + "fp16_bf16_reduction_math_sdp_allowed", + "is_flash_attention_available", + "can_use_flash_attention", + "can_use_efficient_attention", + "can_use_cudnn_attention", + "sdp_kernel", +] + + +def is_built(): + r""" + Return whether PyTorch is built with CUDA support. + + Note that this doesn't necessarily mean CUDA is available; just that if this PyTorch + binary were run on a machine with working CUDA drivers and devices, we would be able to use it. + """ + return torch._C._has_cuda + + +class cuFFTPlanCacheAttrContextProp: + # Like regular ContextProp, but uses the `.device_index` attribute from the + # calling object as the first argument to the getter and setter. + def __init__(self, getter, setter): + self.getter = getter + self.setter = setter + + def __get__(self, obj, objtype): + return self.getter(obj.device_index) + + def __set__(self, obj, val): + if isinstance(self.setter, str): + raise RuntimeError(self.setter) + self.setter(obj.device_index, val) + + +class cuFFTPlanCache: + r""" + Represent a specific plan cache for a specific `device_index`. + + The attributes `size` and `max_size`, and method `clear`, can fetch and/ or + change properties of the C++ cuFFT plan cache. + """ + + def __init__(self, device_index): + self.device_index = device_index + + size = cuFFTPlanCacheAttrContextProp( + torch._cufft_get_plan_cache_size, + ".size is a read-only property showing the number of plans currently in the " + "cache. To change the cache capacity, set cufft_plan_cache.max_size.", + ) + + max_size = cuFFTPlanCacheAttrContextProp( + torch._cufft_get_plan_cache_max_size, torch._cufft_set_plan_cache_max_size + ) + + def clear(self): + return torch._cufft_clear_plan_cache(self.device_index) + + +class cuFFTPlanCacheManager: + r""" + Represent all cuFFT plan caches, return the cuFFTPlanCache for a given device when indexed. + + Finally, this object, when used directly as a `cuFFTPlanCache` object (e.g., + setting the `.max_size`) attribute, the current device's cuFFT plan cache is + used. + """ + + __initialized = False + + def __init__(self): + self.caches = [] + self.__initialized = True + + def __getitem__(self, device): + index = torch.cuda._utils._get_device_index(device) + if index < 0 or index >= torch.cuda.device_count(): + raise RuntimeError( + f"cufft_plan_cache: expected 0 <= device index < {torch.cuda.device_count()}, but got " + f"device with index {index}" + ) + if len(self.caches) == 0: + self.caches.extend( + cuFFTPlanCache(index) for index in range(torch.cuda.device_count()) + ) + return self.caches[index] + + def __getattr__(self, name): + return getattr(self[torch.cuda.current_device()], name) + + def __setattr__(self, name, value): + if self.__initialized: + return setattr(self[torch.cuda.current_device()], name, value) + else: + return super().__setattr__(name, value) + + +class cuBLASModule: + @staticmethod + def _parse_reduction_setting(value: Any, attr_name: str) -> tuple[bool, bool]: + def _ensure_bool(obj: Any, which: str) -> bool: + if isinstance(obj, bool): + return obj + raise TypeError( + f"{attr_name} expects a bool for {which}, but got {type(obj)!r}" + ) + + if isinstance(value, bool): + return value, True + if isinstance(value, (list, tuple)): + if not value: + raise TypeError(f"{attr_name} expects at least one boolean argument") + if len(value) > 2: + raise TypeError(f"{attr_name} expects at most two boolean arguments") + allow_reduced_precision = _ensure_bool(value[0], "allow_reduced_precision") + if len(value) == 1: + return allow_reduced_precision, True + allow_splitk = _ensure_bool(value[1], "allow_splitk") + return allow_reduced_precision, allow_splitk + raise TypeError( + f"{attr_name} expects a bool or a tuple/list of bools, but got {type(value)!r}" + ) + + def __getattr__(self, name): + if name == "allow_tf32": + return torch._C._get_cublas_allow_tf32() + elif name == "prefer_cublaslt_grouped_gemm": + return torch._C._get_cublaslt_prefer_grouped_gemm() + elif name == "allow_fp16_reduced_precision_reduction": + allow_reduced_precision, _ = ( + torch._C._get_cublas_allow_fp16_reduced_precision_reduction() + ) + return allow_reduced_precision + elif name == "allow_fp16_reduced_precision_reduction_split_k": + _, allow_splitk = ( + torch._C._get_cublas_allow_fp16_reduced_precision_reduction() + ) + return allow_splitk + elif name == "allow_bf16_reduced_precision_reduction": + allow_reduced_precision, _ = ( + torch._C._get_cublas_allow_bf16_reduced_precision_reduction() + ) + return allow_reduced_precision + elif name == "allow_bf16_reduced_precision_reduction_split_k": + _, allow_splitk = ( + torch._C._get_cublas_allow_bf16_reduced_precision_reduction() + ) + return allow_splitk + elif name == "allow_fp16_accumulation": + return torch._C._get_cublas_allow_fp16_accumulation() + elif name == "fp32_precision": + return torch._C._get_fp32_precision_getter("cuda", "matmul") + raise AttributeError("Unknown attribute " + name) + + def __setattr__(self, name, value): + if name == "allow_tf32": + return torch._C._set_cublas_allow_tf32(value) + elif name == "prefer_cublaslt_grouped_gemm": + return torch._C._set_cublaslt_prefer_grouped_gemm(value) + elif name == "allow_fp16_reduced_precision_reduction": + allow_reduced_precision, allow_splitk = self._parse_reduction_setting( + value, "allow_fp16_reduced_precision_reduction" + ) + return torch._C._set_cublas_allow_fp16_reduced_precision_reduction( + allow_reduced_precision, + allow_splitk, + ) + elif name == "allow_bf16_reduced_precision_reduction": + allow_reduced_precision, allow_splitk = self._parse_reduction_setting( + value, "allow_bf16_reduced_precision_reduction" + ) + return torch._C._set_cublas_allow_bf16_reduced_precision_reduction( + allow_reduced_precision, + allow_splitk, + ) + elif name == "allow_fp16_accumulation": + return torch._C._set_cublas_allow_fp16_accumulation(value) + elif name == "fp32_precision": + return torch._C._set_fp32_precision_setter("cuda", "matmul", value) + raise AttributeError("Unknown attribute " + name) + + +_LinalgBackends = { + "default": torch._C._LinalgBackend.Default, + "cusolver": torch._C._LinalgBackend.Cusolver, + "magma": torch._C._LinalgBackend.Magma, +} +_LinalgBackends_str = ", ".join(_LinalgBackends.keys()) + + +def preferred_linalg_library( + backend: None | str | torch._C._LinalgBackend = None, +) -> torch._C._LinalgBackend: + r""" + Override the heuristic PyTorch uses to choose between cuSOLVER and MAGMA for CUDA linear algebra operations. + + .. warning:: This flag is experimental and subject to change. + + When PyTorch runs a CUDA linear algebra operation it often uses the cuSOLVER or MAGMA libraries, + and if both are available it decides which to use with a heuristic. + This flag (a :class:`str`) allows overriding those heuristics. + + * If `"cusolver"` is set then cuSOLVER will be used wherever possible. + * If `"magma"` is set then MAGMA will be used wherever possible. + * If `"default"` (the default) is set then heuristics will be used to pick between + cuSOLVER and MAGMA if both are available. + * When no input is given, this function returns the currently preferred library. + * User may use the environment variable TORCH_LINALG_PREFER_CUSOLVER=1 to set the preferred library to cuSOLVER + globally. + This flag only sets the initial value of the preferred library and the preferred library + may still be overridden by this function call later in your script. + + Note: When a library is preferred other libraries may still be used if the preferred library + doesn't implement the operation(s) called. + This flag may achieve better performance if PyTorch's heuristic library selection is incorrect + for your application's inputs. + + Currently supported linalg operators: + + * :func:`torch.linalg.inv` + * :func:`torch.linalg.inv_ex` + * :func:`torch.linalg.cholesky` + * :func:`torch.linalg.cholesky_ex` + * :func:`torch.cholesky_solve` + * :func:`torch.cholesky_inverse` + * :func:`torch.linalg.lu_factor` + * :func:`torch.linalg.lu` + * :func:`torch.linalg.lu_solve` + * :func:`torch.linalg.qr` + * :func:`torch.linalg.eigh` + * :func:`torch.linalg.eigvals` + * :func:`torch.linalg.svd` + * :func:`torch.linalg.svdvals` + """ + if backend is None: + pass + elif isinstance(backend, str): + if backend not in _LinalgBackends: + raise RuntimeError( + f"Unknown input value. Choose from: {_LinalgBackends_str}." + ) + torch._C._set_linalg_preferred_backend(_LinalgBackends[backend]) + elif isinstance(backend, torch._C._LinalgBackend): + torch._C._set_linalg_preferred_backend(backend) + else: + raise RuntimeError("Unknown input value type.") + + return torch._C._get_linalg_preferred_backend() + + +_BlasBackends = { + "default": torch._C._BlasBackend.Default, + "cublas": torch._C._BlasBackend.Cublas, + "hipblas": torch._C._BlasBackend.Cublas, # alias + "cublaslt": torch._C._BlasBackend.Cublaslt, + "hipblaslt": torch._C._BlasBackend.Cublaslt, # alias + "ck": torch._C._BlasBackend.Ck, +} +_BlasBackends_str = ", ".join(_BlasBackends.keys()) + + +def preferred_blas_library( + backend: None | str | torch._C._BlasBackend = None, +) -> torch._C._BlasBackend: + r""" + Override the library PyTorch uses for BLAS operations. Choose between cuBLAS, cuBLASLt, and CK [ROCm-only]. + + .. warning:: This flag is experimental and subject to change. + + When PyTorch runs a CUDA BLAS operation it defaults to cuBLAS even if both cuBLAS and cuBLASLt are available. + For PyTorch built for ROCm, hipBLAS, hipBLASLt, and CK may offer different performance. + This flag (a :class:`str`) allows overriding which BLAS library to use. + + * If `"cublas"` is set then cuBLAS will be used wherever possible. + * If `"cublaslt"` is set then cuBLASLt will be used wherever possible. + * If `"ck"` is set then CK will be used wherever possible. + * If `"default"` (the default) is set then heuristics will be used to pick between the other options. + * When no input is given, this function returns the currently preferred library. + * User may use the environment variable TORCH_BLAS_PREFER_CUBLASLT=1 to set the preferred library to cuBLASLt + globally. + This flag only sets the initial value of the preferred library and the preferred library + may still be overridden by this function call later in your script. + + Note: When a library is preferred other libraries may still be used if the preferred library + doesn't implement the operation(s) called. + This flag may achieve better performance if PyTorch's library selection is incorrect + for your application's inputs. + + """ + if backend is None: + pass + elif isinstance(backend, str): + if backend not in _BlasBackends: + raise RuntimeError( + f"Unknown input value. Choose from: {_BlasBackends_str}." + ) + torch._C._set_blas_preferred_backend(_BlasBackends[backend]) + elif isinstance(backend, torch._C._BlasBackend): + torch._C._set_blas_preferred_backend(backend) + else: + raise RuntimeError("Unknown input value type.") + + return torch._C._get_blas_preferred_backend() + + +def cublas_workspace_size(size: None | int = None) -> int: + r"""Query or set the cuBLAS workspace size in bytes. + + When called with no arguments, returns the current workspace size. + When called with a size argument, sets the workspace size and returns the new value. + Setting the workspace size will take precedence over the CUBLAS_WORKSPACE_CONFIG environment variable. + Changes take effect lazily: only handles used after the change get new workspaces. + + Args: + size (int, optional): workspace size in bytes. Must be non-negative. + + Returns: + int: the current (or newly set) workspace size in bytes. + """ + if size is not None: + torch._C._cuda_setCublasWorkspaceSize(size) + return torch._C._cuda_getCublasWorkspaceSize() + + +def cublaslt_workspace_size(size: None | int = None) -> int: + r"""Query or set the cuBLASLt workspace size in bytes. + + When called with no arguments, returns the current workspace size. + When called with a size argument, sets the workspace size and returns the new value. + Setting the workspace size will take precedence over the CUBLASLT_WORKSPACE_SIZE environment variable. + Changes take effect lazily: only handles used after the change get new workspaces. + + Args: + size (int, optional): workspace size in bytes. Must be non-negative. + + Returns: + int: the current (or newly set) workspace size in bytes. + """ + if size is not None: + torch._C._cuda_setCublasLtWorkspaceSize(size) + return torch._C._cuda_getCublasLtWorkspaceSize() + + +def blas_workspace_size( + size: None | int = None, + backend: None | str | torch._C._BlasBackend = None, +) -> int: + r"""Query or set the BLAS workspace size for a given backend. + + Convenience wrapper that dispatches to :func:`cublas_workspace_size` or + :func:`cublaslt_workspace_size` depending on the backend. + + When *backend* is ``None`` the current :func:`preferred_blas_library` is + used. ``Default`` is resolved to the platform's default backend (cuBLAS + on NVIDIA, potentially hipBLASLt on supported ROCm architectures). + + .. note:: + + When ``TORCH_CUBLASLT_UNIFIED_WORKSPACE`` is enabled (the default on + open-source CUDA builds), the cuBLASLt workspace is capped at the + cuBLAS workspace size and physically reuses the same allocation. + Setting a large cuBLASLt workspace via this function will therefore + *not* increase memory beyond the cuBLAS workspace size. + + .. note:: + + Setting the workspace size for the cublas backend will take precedence + over the CUBLAS_WORKSPACE_CONFIG environment variable, and setting the + workspace size for the cublaslt backend will take precedence over the + CUBLASLT_WORKSPACE_SIZE environment variable. + + Args: + size (int, optional): workspace size in bytes. Must be non-negative. + When omitted the current size is returned without modification. + backend (str | torch._C._BlasBackend, optional): which backend's + workspace to query/set. Accepts the same strings as + :func:`preferred_blas_library` (e.g. ``"cublas"``, ``"cublaslt"``). + + Returns: + int: the current (or newly set) workspace size in bytes. + + Raises: + RuntimeError: if the resolved backend is CK (no workspace concept). + """ + if backend is None: + resolved = preferred_blas_library() + elif isinstance(backend, str): + if backend not in _BlasBackends: + raise RuntimeError( + f"Unknown backend string. Choose from: {_BlasBackends_str}." + ) + resolved = _BlasBackends[backend] + elif isinstance(backend, torch._C._BlasBackend): + resolved = backend + else: + raise RuntimeError("Unknown backend type.") + + if resolved == torch._C._BlasBackend.Default: + resolved = torch._C._get_blas_default_backend() + + if resolved == torch._C._BlasBackend.Ck: + raise RuntimeError("CK backend does not use a workspace.") + + if resolved == torch._C._BlasBackend.Cublaslt: + return cublaslt_workspace_size(size) + return cublas_workspace_size(size) + + +_ROCmFABackends = { + "default": torch._C._ROCmFABackend.Default, + "aotriton": torch._C._ROCmFABackend.AOTriton, + "ck": torch._C._ROCmFABackend.Ck, +} +_ROCmFABackends_str = ", ".join(_ROCmFABackends.keys()) + + +from torch._C import _SDPAParams as SDPAParams, _SDPBackend as SDPBackend + + +def preferred_rocm_fa_library( + backend: None | str | torch._C._ROCmFABackend = None, +) -> torch._C._ROCmFABackend: + r""" + [ROCm-only] + Override the backend PyTorch uses in ROCm environments for Flash Attention. Choose between AOTriton and CK + + .. warning:: This flag is experimental and subject to change. + + When Flash Attention is enabled and desired, PyTorch defaults to using AOTriton as the backend. + This flag (a :class:`str`) allows users to override this backend to use composable_kernel + + * If `"default"` is set then the default backend will be used wherever possible. Currently AOTriton. + * If `"aotriton"` is set then AOTriton will be used wherever possible. + * If `"ck"` is set then CK will be used wherever possible. + * When no input is given, this function returns the currently preferred library. + * User may use the environment variable TORCH_ROCM_FA_PREFER_CK=1 to set the preferred library to CK + globally. + + Note: When a library is preferred other libraries may still be used if the preferred library + doesn't implement the operation(s) called. + This flag may achieve better performance if PyTorch's library selection is incorrect + for your application's inputs. + """ + if backend is None: + pass + elif isinstance(backend, str): + if backend not in _ROCmFABackends: + raise RuntimeError( + f"Unknown input value. Choose from: {_ROCmFABackends_str}." + ) + torch._C._set_rocm_fa_preferred_backend(_ROCmFABackends[backend]) + elif isinstance(backend, torch._C._ROCmFABackend): + torch._C._set_rocm_fa_preferred_backend(backend) + else: + raise ValueError(f"Unknown input value. Choose from: {_ROCmFABackends_str}.") + + return torch._C._get_rocm_fa_preferred_backend() + + +# Set the __module__ attribute +SDPAParams.__module__ = "torch.backends.cuda" +SDPAParams.__name__ = "SDPAParams" + + +def is_ck_sdpa_available() -> bool: + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether composable_kernel may be used as the backend for + scaled-dot-product-attention. + """ + # pyrefly: ignore [missing-attribute] + return torch._C._is_ck_sdpa_available() + + +def flash_sdp_enabled(): + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether flash scaled dot product attention is enabled or not. + """ + return torch._C._get_flash_sdp_enabled() + + +def enable_flash_sdp(enabled: bool): + r""" + .. warning:: This flag is beta and subject to change. + + Enables or disables flash scaled dot product attention. + """ + torch._C._set_sdp_use_flash(enabled) + + +def mem_efficient_sdp_enabled(): + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether memory efficient scaled dot product attention is enabled or not. + """ + return torch._C._get_mem_efficient_sdp_enabled() + + +def enable_mem_efficient_sdp(enabled: bool): + r""" + .. warning:: This flag is beta and subject to change. + + Enables or disables memory efficient scaled dot product attention. + """ + torch._C._set_sdp_use_mem_efficient(enabled) + + +def math_sdp_enabled(): + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether math scaled dot product attention is enabled or not. + """ + return torch._C._get_math_sdp_enabled() + + +def enable_math_sdp(enabled: bool): + r""" + .. warning:: This flag is beta and subject to change. + + Enables or disables math scaled dot product attention. + """ + torch._C._set_sdp_use_math(enabled) + + +def allow_fp16_bf16_reduction_math_sdp(enabled: bool): + r""" + .. warning:: This flag is beta and subject to change. + + Enables or disables fp16/bf16 reduction in math scaled dot product attention. + """ + torch._C._set_math_sdp_allow_fp16_bf16_reduction(enabled) + + +def fp16_bf16_reduction_math_sdp_allowed(): + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether fp16/bf16 reduction in math scaled dot product attention is enabled or not. + """ + return torch._C._get_math_sdp_allow_fp16_bf16_reduction() + + +def is_flash_attention_available() -> bool: + r"""Check if PyTorch was built with FlashAttention for scaled_dot_product_attention. + + Returns: + True if FlashAttention is built and available; otherwise, False. + + Note: + This function is dependent on a CUDA-enabled build of PyTorch. It will return False + in non-CUDA environments. + """ + return torch._C._is_flash_attention_available() + + +def can_use_flash_attention(params: SDPAParams, debug: bool = False) -> bool: + r"""Check if FlashAttention can be utilized in scaled_dot_product_attention. + + Args: + params: An instance of SDPAParams containing the tensors for query, + key, value, an optional attention mask, dropout rate, and + a flag indicating if the attention is causal. + debug: Whether to logging.warn debug information as to why FlashAttention could not be run. + Defaults to False. + + Returns: + True if FlashAttention can be used with the given parameters; otherwise, False. + + Note: + This function is dependent on a CUDA-enabled build of PyTorch. It will return False + in non-CUDA environments. + """ + return torch._C._can_use_flash_attention(params, debug) + + +def can_use_efficient_attention(params: SDPAParams, debug: bool = False) -> bool: + r"""Check if efficient_attention can be utilized in scaled_dot_product_attention. + + Args: + params: An instance of SDPAParams containing the tensors for query, + key, value, an optional attention mask, dropout rate, and + a flag indicating if the attention is causal. + debug: Whether to logging.warn with information as to why efficient_attention could not be run. + Defaults to False. + + Returns: + True if efficient_attention can be used with the given parameters; otherwise, False. + + Note: + This function is dependent on a CUDA-enabled build of PyTorch. It will return False + in non-CUDA environments. + """ + return torch._C._can_use_mem_efficient_attention(params, debug) + + +def can_use_cudnn_attention(params: SDPAParams, debug: bool = False) -> bool: + r"""Check if cudnn_attention can be utilized in scaled_dot_product_attention. + + Args: + params: An instance of SDPAParams containing the tensors for query, + key, value, an optional attention mask, dropout rate, and + a flag indicating if the attention is causal. + debug: Whether to logging.warn with information as to why cuDNN attention could not be run. + Defaults to False. + + Returns: + True if cuDNN can be used with the given parameters; otherwise, False. + + Note: + This function is dependent on a CUDA-enabled build of PyTorch. It will return False + in non-CUDA environments. + """ + return torch._C._can_use_cudnn_attention(params, debug) + + +def cudnn_sdp_enabled(): + r""" + .. warning:: This flag is beta and subject to change. + + Returns whether cuDNN scaled dot product attention is enabled or not. + """ + return torch._C._get_cudnn_sdp_enabled() + + +def enable_cudnn_sdp(enabled: bool): + r""" + .. warning:: This flag is beta and subject to change. + + Enables or disables cuDNN scaled dot product attention. + """ + torch._C._set_sdp_use_cudnn(enabled) + + +@contextlib.contextmanager +@deprecated( + ( + "`torch.backends.cuda.sdp_kernel()` is deprecated. " + "In the future, this context manager will be removed. " + "Please see `torch.nn.attention.sdpa_kernel()` for the new context manager, " + "with updated signature." + ), + category=FutureWarning, +) +def sdp_kernel( + enable_flash: bool = True, + enable_math: bool = True, + enable_mem_efficient: bool = True, + enable_cudnn: bool = True, +): + r""" + .. warning:: This flag is beta and subject to change. + + This context manager can be used to temporarily enable or disable any of the three backends for scaled dot product attention. + Upon exiting the context manager, the previous state of the flags will be restored. + """ + from torch.nn.attention import sdpa_kernel + + backend_list = [] + if enable_flash: + backend_list.append(SDPBackend.FLASH_ATTENTION) + if enable_mem_efficient: + backend_list.append(SDPBackend.EFFICIENT_ATTENTION) + if enable_math: + backend_list.append(SDPBackend.MATH) + if enable_cudnn: + backend_list.append(SDPBackend.CUDNN_ATTENTION) + + with sdpa_kernel(backend_list) as context: + try: + yield context + finally: + pass + + +cufft_plan_cache = cuFFTPlanCacheManager() +matmul = cuBLASModule() diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cuda/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/cuda/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6a16b4cc73f9c6e8c3f174fe5437a087995c27e7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/cuda/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..c98153dcfa6974746d091651064a4eed6e75f869 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__init__.py @@ -0,0 +1,261 @@ +# mypy: allow-untyped-defs +import os +import sys +import warnings +from contextlib import contextmanager +from typing import Optional + +import torch +from torch.backends import ( + __allow_nonbracketed_mutation, + _FP32Precision, + _get_fp32_precision_getter, + _set_fp32_precision_setter, + ContextProp, + PropModule, +) + +from . import rnn + + +try: + from torch._C import _cudnn +except ImportError: + _cudnn = None # type: ignore[assignment] + +# Write: +# +# torch.backends.cudnn.enabled = False +# +# to globally disable CuDNN/MIOpen + +__cudnn_version: int | None = None + +if _cudnn is not None: + + def _init(): + global __cudnn_version + if __cudnn_version is None: + # pyrefly: ignore [missing-attribute] + __cudnn_version = _cudnn.getVersionInt() + # pyrefly: ignore [missing-attribute] + runtime_version = _cudnn.getRuntimeVersion() + # pyrefly: ignore [missing-attribute] + compile_version = _cudnn.getCompileVersion() + runtime_major, runtime_minor, _ = runtime_version + compile_major, compile_minor, _ = compile_version + # Different major versions are always incompatible + # Starting with cuDNN 7, minor versions are backwards-compatible + # Not sure about MIOpen (ROCm), so always do a strict check + if runtime_major != compile_major: + cudnn_compatible = False + # pyrefly: ignore [missing-attribute] + elif runtime_major < 7 or not _cudnn.is_cuda: + cudnn_compatible = runtime_minor == compile_minor + else: + cudnn_compatible = runtime_minor >= compile_minor + if not cudnn_compatible: + if os.environ.get("PYTORCH_SKIP_CUDNN_COMPATIBILITY_CHECK", "0") == "1": + return True + base_error_msg = ( + f"cuDNN version incompatibility: " + f"PyTorch was compiled against {compile_version} " + f"but found runtime version {runtime_version}. " + f"PyTorch already comes bundled with cuDNN. " + f"One option to resolving this error is to ensure PyTorch " + f"can find the bundled cuDNN. " + ) + + if "LD_LIBRARY_PATH" in os.environ: + ld_library_path = os.environ.get("LD_LIBRARY_PATH", "") + if any( + substring in ld_library_path for substring in ["cuda", "cudnn"] + ): + raise RuntimeError( + f"{base_error_msg}" + f"Looks like your LD_LIBRARY_PATH contains incompatible version of cudnn. " + f"Please either remove it from the path or install cudnn {compile_version}" + ) + else: + raise RuntimeError( + f"{base_error_msg}" + f"one possibility is that there is a " + f"conflicting cuDNN in LD_LIBRARY_PATH." + ) + else: + raise RuntimeError(base_error_msg) + # Check if cuDNN version is compatible with available CUDA devices + if torch.cuda.is_available() and not torch.version.hip: + min_cc = min( + [ + torch.cuda.get_device_capability(i) + for i in range(torch.cuda.device_count()) + ] + ) + if __cudnn_version >= 91100 and min_cc < (7, 5): + raise RuntimeError( + f"cuDNN version {__cudnn_version} is not compatible with devices with SM < 7.5. " + f"Please install a version of PyTorch with a compatible cuDNN version. " + f"https://github.com/pytorch/pytorch/blob/main/RELEASE.md#release-compatibility-matrix" + ) + + return True + +else: + + def _init(): + return False + + +def version(): + """Return the version of cuDNN.""" + if not _init(): + return None + return __cudnn_version + + +CUDNN_TENSOR_DTYPES = { + torch.half, + torch.float, + torch.double, +} + + +def is_available(): + r"""Return a bool indicating if CUDNN is currently available.""" + return torch._C._has_cudnn + + +def is_acceptable(tensor): + if not torch._C._get_cudnn_enabled(): + return False + if tensor.device.type != "cuda" or tensor.dtype not in CUDNN_TENSOR_DTYPES: + return False + if not is_available(): + warnings.warn( + "PyTorch was compiled without cuDNN/MIOpen support. To use cuDNN/MIOpen, rebuild " + "PyTorch making sure the library is visible to the build system.", + stacklevel=2, + ) + return False + if not _init(): + warnings.warn( + "cuDNN/MIOpen library not found. Check your {libpath}".format( + libpath={"darwin": "DYLD_LIBRARY_PATH", "win32": "PATH"}.get( + sys.platform, "LD_LIBRARY_PATH" + ) + ), + stacklevel=2, + ) + return False + return True + + +def set_flags( + _enabled=None, + _benchmark=None, + _benchmark_limit=None, + _deterministic=None, + _allow_tf32=None, + _fp32_precision="none", + _depthwise_kernel=None, +): + orig_flags = ( + torch._C._get_cudnn_enabled(), + torch._C._get_cudnn_benchmark(), + None if not is_available() else torch._C._cuda_get_cudnn_benchmark_limit(), + torch._C._get_cudnn_deterministic(), + torch._C._get_cudnn_allow_tf32(), + torch._C._get_fp32_precision_getter("cuda", "all"), + torch._C._get_cudnn_depthwise_kernel(), + ) + if _enabled is not None: + torch._C._set_cudnn_enabled(_enabled) + if _benchmark is not None: + torch._C._set_cudnn_benchmark(_benchmark) + if _benchmark_limit is not None and is_available(): + torch._C._cuda_set_cudnn_benchmark_limit(_benchmark_limit) + if _deterministic is not None: + torch._C._set_cudnn_deterministic(_deterministic) + if _allow_tf32 is not None: + torch._C._set_cudnn_allow_tf32(_allow_tf32) + if _fp32_precision is not None: + torch._C._set_fp32_precision_setter("cuda", "all", _fp32_precision) + if _depthwise_kernel is not None: + torch._C._set_cudnn_depthwise_kernel(_depthwise_kernel) + return orig_flags + + +@contextmanager +def flags( + enabled=False, + benchmark=False, + benchmark_limit=10, + deterministic=False, + allow_tf32=True, + fp32_precision="none", + depthwise_kernel="auto", +): + with __allow_nonbracketed_mutation(): + orig_flags = set_flags( + enabled, + benchmark, + benchmark_limit, + deterministic, + allow_tf32, + fp32_precision, + depthwise_kernel, + ) + try: + yield + finally: + # recover the previous values + with __allow_nonbracketed_mutation(): + set_flags(*orig_flags) + + +# The magic here is to allow us to intercept code like this: +# +# torch.backends..enabled = True + + +class CudnnModule(PropModule): + enabled = ContextProp(torch._C._get_cudnn_enabled, torch._C._set_cudnn_enabled) + deterministic = ContextProp( + torch._C._get_cudnn_deterministic, torch._C._set_cudnn_deterministic + ) + benchmark = ContextProp( + torch._C._get_cudnn_benchmark, torch._C._set_cudnn_benchmark + ) + benchmark_limit = None + if is_available(): + benchmark_limit = ContextProp( + torch._C._cuda_get_cudnn_benchmark_limit, + torch._C._cuda_set_cudnn_benchmark_limit, + ) + allow_tf32 = ContextProp( + torch._C._get_cudnn_allow_tf32, torch._C._set_cudnn_allow_tf32 + ) + conv = _FP32Precision("cuda", "conv") + fp32_precision = ContextProp( + _get_fp32_precision_getter("cuda", "all"), + _set_fp32_precision_setter("cuda", "all"), + ) + depthwise_kernel = ContextProp( + torch._C._get_cudnn_depthwise_kernel, + torch._C._set_cudnn_depthwise_kernel, + ) + + +# This is the sys.modules replacement trick, see +# https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273 +sys.modules[__name__] = CudnnModule(sys.modules[__name__], __name__) + +# Add type annotation for the replaced module +enabled: bool +deterministic: bool +benchmark: bool +allow_tf32: bool +fp32_precision: str +benchmark_limit: int +depthwise_kernel: str diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..133da173b8d45b80b69a90b95152e0d61a437981 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/rnn.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/rnn.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d6a88de62e47cf7508c4dfc7165ace7daac5cdab Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/__pycache__/rnn.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cudnn/rnn.py b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/rnn.py new file mode 100644 index 0000000000000000000000000000000000000000..177b664aa3b0a610e4a6ae51da7d631e91bf278c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/cudnn/rnn.py @@ -0,0 +1,128 @@ +# mypy: allow-untyped-defs +import sys + +import torch._C +import torch.cuda +from torch.backends import ( + _get_fp32_precision_getter, + _set_fp32_precision_setter, + PropModule, +) + + +try: + from torch._C import _cudnn +except ImportError: + # Uses of all the functions below should be guarded by torch.backends.cudnn.is_available(), + # so it's safe to not emit any checks here. + _cudnn = None # type: ignore[assignment] + + +def get_cudnn_mode(mode): + if mode == "RNN_RELU": + # pyrefly: ignore [missing-attribute] + return int(_cudnn.RNNMode.rnn_relu) + elif mode == "RNN_TANH": + # pyrefly: ignore [missing-attribute] + return int(_cudnn.RNNMode.rnn_tanh) + elif mode == "LSTM": + # pyrefly: ignore [missing-attribute] + return int(_cudnn.RNNMode.lstm) + elif mode == "GRU": + # pyrefly: ignore [missing-attribute] + return int(_cudnn.RNNMode.gru) + else: + raise ValueError(f"Unknown mode: {mode}") + + +# NB: We don't actually need this class anymore (in fact, we could serialize the +# dropout state for even better reproducibility), but it is kept for backwards +# compatibility for old models. +class Unserializable: + def __init__(self, inner): + self.inner = inner + + def get(self): + return self.inner + + def __getstate__(self): + return "" + + def __setstate__(self, state): + self.inner = None + + +# we would like to use ContextProp from backends here but the +# frozen flags appears to be overzealous +class ContextProp: + def __init__(self, getter, setter): + self.getter = getter + self.setter = setter + + def __get__(self, obj, objtype): + return self.getter() + + def __set__(self, obj, val): + self.setter(val) + + +def init_dropout_state(dropout, train, dropout_seed, dropout_state): + dropout_desc_name = "desc_" + str(torch.cuda.current_device()) + dropout_p = dropout if train else 0 + if (dropout_desc_name not in dropout_state) or ( + dropout_state[dropout_desc_name].get() is None + ): + if dropout_p == 0: + dropout_state[dropout_desc_name] = Unserializable(None) + else: + dropout_state[dropout_desc_name] = Unserializable( + torch._cudnn_init_dropout_state( # type: ignore[call-arg] + dropout_p, + train, + dropout_seed, + # pyrefly: ignore [unexpected-keyword] + self_ty=torch.uint8, + device=torch.device("cuda"), + ) + ) + dropout_ts = dropout_state[dropout_desc_name].get() + return dropout_ts + + +class CudnnRNNModule(PropModule): + def __init__(self, m, name): + super().__init__(m, name) + self.m.Unserializable = Unserializable + self.m.get_cudnn_mode = get_cudnn_mode + self.m.init_dropout_state = init_dropout_state + + @staticmethod + def init_dropout_state(dropout, train, dropout_seed, dropout_state): + dropout_desc_name = "desc_" + str(torch.cuda.current_device()) + dropout_p = dropout if train else 0 + if (dropout_desc_name not in dropout_state) or ( + dropout_state[dropout_desc_name].get() is None + ): + if dropout_p == 0: + dropout_state[dropout_desc_name] = Unserializable(None) + else: + dropout_state[dropout_desc_name] = Unserializable( + torch._cudnn_init_dropout_state( # type: ignore[call-arg] + dropout_p, + train, + dropout_seed, + # pyrefly: ignore [unexpected-keyword] + self_ty=torch.uint8, + device=torch.device("cuda"), + ) + ) + dropout_ts = dropout_state[dropout_desc_name].get() + return dropout_ts + + fp32_precision = ContextProp( + _get_fp32_precision_getter("cuda", "rnn"), + _set_fp32_precision_setter("cuda", "rnn"), + ) + + +sys.modules[__name__] = CudnnRNNModule(sys.modules[__name__], __name__) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..6b0006d91637d2e35ac87a61b2393ad938d142bc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__init__.py @@ -0,0 +1,57 @@ +from typing import Optional + +import torch + + +__all__ = [ + "version", + "is_available", + "get_max_alg_id", +] + +try: + from torch._C import _cusparselt +except ImportError: + _cusparselt = None # type: ignore[assignment] + +__cusparselt_version: int | None = None +__MAX_ALG_ID: int | None = None + +if _cusparselt is not None: + + def _init() -> bool: + global __cusparselt_version + global __MAX_ALG_ID + if __cusparselt_version is None: + # pyrefly: ignore [missing-attribute] + __cusparselt_version = _cusparselt.getVersionInt() + if __cusparselt_version == 400: + __MAX_ALG_ID = 4 + elif __cusparselt_version == 502: + __MAX_ALG_ID = 5 + elif __cusparselt_version == 602: + __MAX_ALG_ID = 37 + return True + +else: + + def _init() -> bool: + return False + + +def version() -> int | None: + """Return the version of cuSPARSELt""" + if not _init(): + return None + return __cusparselt_version + + +def is_available() -> bool: + r"""Return a bool indicating if cuSPARSELt is currently available.""" + return torch._C._has_cusparselt + + +def get_max_alg_id() -> int | None: + if not _init(): + return None + return __MAX_ALG_ID diff --git a/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dcc9f3f1273df885243ca4063720ad95d66eaeed Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/cusparselt/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..1a681b77ef58ce1f390232b82c4a9843d5559ca3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__init__.py @@ -0,0 +1,7 @@ +# mypy: allow-untyped-defs +import torch + + +def is_available(): + r"""Return whether PyTorch is built with KleidiAI support.""" + return torch._C._has_kleidiai diff --git a/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9cf9ffc363aad55c1e7b69e5280396cade7d90e9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/kleidiai/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mha/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/mha/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e1dd2ebd688805bdf3359cb56b64d0854cf258c4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/mha/__init__.py @@ -0,0 +1,25 @@ +# Config options to enable/disable C++ kernel for nn.functional.MHA +# and nn.TransformerEncoder +import torch + + +_is_fastpath_enabled: bool = True + + +def get_fastpath_enabled() -> bool: + """Returns whether fast path for TransformerEncoder and MultiHeadAttention + is enabled, or ``True`` if jit is scripting. + + .. note:: + The fastpath might not be run even if ``get_fastpath_enabled`` returns + ``True`` unless all conditions on inputs are met. + """ + if not torch.jit.is_scripting(): + return _is_fastpath_enabled + return True + + +def set_fastpath_enabled(value: bool) -> None: + """Sets whether fast path is enabled""" + global _is_fastpath_enabled + _is_fastpath_enabled = value diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mha/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/mha/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..49f474dab2e57e924f3916292652fba20dae8106 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/mha/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/miopen/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/miopen/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..1b270b658e31a91dfb37380abec383009dfc5bfa --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/miopen/__init__.py @@ -0,0 +1,50 @@ +# mypy: allow-untyped-defs +import sys +from contextlib import contextmanager + +import torch +from torch.backends import __allow_nonbracketed_mutation, ContextProp, PropModule + + +def set_flags( + _immediate=None, +): + orig_flags = (torch._C._get_miopen_immediate(),) + if _immediate is not None: + torch._C._set_miopen_immediate(_immediate) + return orig_flags + + +@contextmanager +def flags( + immediate=False, +): + with __allow_nonbracketed_mutation(): + orig_flags = set_flags( + immediate, + ) + try: + yield + finally: + # recover the previous values + with __allow_nonbracketed_mutation(): + set_flags(*orig_flags) + + +# The magic here is to allow us to intercept code like this: +# +# torch.backends..immediate = True + + +class MiopenModule(PropModule): + immediate = ContextProp( + torch._C._get_miopen_immediate, torch._C._set_miopen_immediate + ) + + +# This is the sys.modules replacement trick, see +# https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273 +sys.modules[__name__] = MiopenModule(sys.modules[__name__], __name__) + +# Add type annotation for the replaced module +immediate: bool diff --git a/.venv/lib/python3.12/site-packages/torch/backends/miopen/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/miopen/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6f5f9d6931ff8c65a6f4f2beb94c72b8db8f3a76 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/miopen/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mkl/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/mkl/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..52900f0861decc8d7fe8d1221f73553c9b141781 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/mkl/__init__.py @@ -0,0 +1,59 @@ +# mypy: allow-untyped-defs +import torch + + +def is_available(): + r"""Return whether PyTorch is built with MKL support.""" + return torch._C.has_mkl + + +VERBOSE_OFF = 0 +VERBOSE_ON = 1 + + +class verbose: + """ + On-demand oneMKL verbosing functionality. + + To make it easier to debug performance issues, oneMKL can dump verbose + messages containing execution information like duration while executing + the kernel. The verbosing functionality can be invoked via an environment + variable named `MKL_VERBOSE`. However, this methodology dumps messages in + all steps. Those are a large amount of verbose messages. Moreover, for + investigating the performance issues, generally taking verbose messages + for one single iteration is enough. This on-demand verbosing functionality + makes it possible to control scope for verbose message dumping. In the + following example, verbose messages will be dumped out for the second + inference only. + + .. highlight:: python + .. code-block:: python + + import torch + + model(data) + with torch.backends.mkl.verbose(torch.backends.mkl.VERBOSE_ON): + model(data) + + Args: + level: Verbose level + - ``VERBOSE_OFF``: Disable verbosing + - ``VERBOSE_ON``: Enable verbosing + """ + + def __init__(self, enable): + self.enable = enable + + def __enter__(self): + if self.enable == VERBOSE_OFF: + return + st = torch._C._verbose.mkl_set_verbose(self.enable) + if not st: + raise AssertionError( + "Failed to set MKL into verbose mode. Please consider to disable this verbose scope." + ) + return self + + def __exit__(self, exc_type, exc_val, exc_tb): + torch._C._verbose.mkl_set_verbose(VERBOSE_OFF) + return False diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mkl/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/mkl/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..24c3ef5963c645b3898015fcbeaa3e56a8f2b3e9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/mkl/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..6c7326496a50727ad782e51891d245fc996908ba --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__init__.py @@ -0,0 +1,139 @@ +# mypy: allow-untyped-defs +import sys +from contextlib import contextmanager +from typing import TYPE_CHECKING + +import torch +from torch.backends import ( + __allow_nonbracketed_mutation, + _FP32Precision, + _get_fp32_precision_getter, + _set_fp32_precision_setter, + ContextProp, + PropModule, +) + + +def is_available(): + r"""Return whether PyTorch is built with MKL-DNN support.""" + return torch._C._has_mkldnn + + +def is_acl_available(): + r"""Return whether PyTorch is built with MKL-DNN + ACL support.""" + # pyrefly: ignore [missing-attribute] + return torch._C._has_mkldnn_acl + + +VERBOSE_OFF = 0 +VERBOSE_ON = 1 +VERBOSE_ON_CREATION = 2 + + +class verbose: + """ + On-demand oneDNN (former MKL-DNN) verbosing functionality. + + To make it easier to debug performance issues, oneDNN can dump verbose + messages containing information like kernel size, input data size and + execution duration while executing the kernel. The verbosing functionality + can be invoked via an environment variable named `DNNL_VERBOSE`. However, + this methodology dumps messages in all steps. Those are a large amount of + verbose messages. Moreover, for investigating the performance issues, + generally taking verbose messages for one single iteration is enough. + This on-demand verbosing functionality makes it possible to control scope + for verbose message dumping. In the following example, verbose messages + will be dumped out for the second inference only. + + .. highlight:: python + .. code-block:: python + + import torch + + model(data) + with torch.backends.mkldnn.verbose(torch.backends.mkldnn.VERBOSE_ON): + model(data) + + Args: + level: Verbose level + - ``VERBOSE_OFF``: Disable verbosing + - ``VERBOSE_ON``: Enable verbosing + - ``VERBOSE_ON_CREATION``: Enable verbosing, including oneDNN kernel creation + """ + + def __init__(self, level): + self.level = level + + def __enter__(self): + if self.level == VERBOSE_OFF: + return + st = torch._C._verbose.mkldnn_set_verbose(self.level) + if not st: + raise AssertionError( + "Failed to set MKLDNN into verbose mode. Please consider to disable this verbose scope." + ) + return self + + def __exit__(self, exc_type, exc_val, exc_tb): + torch._C._verbose.mkldnn_set_verbose(VERBOSE_OFF) + return False + + +def set_flags( + _enabled=None, _deterministic=None, _allow_tf32=None, _fp32_precision="none" +): + orig_flags = ( + torch._C._get_mkldnn_enabled(), + torch._C._get_mkldnn_deterministic(), + torch._C._get_onednn_allow_tf32(), + torch._C._get_fp32_precision_getter("mkldnn", "all"), + ) + if _enabled is not None: + torch._C._set_mkldnn_enabled(_enabled) + if _deterministic is not None: + torch._C._set_mkldnn_deterministic(_deterministic) + if _allow_tf32 is not None: + torch._C._set_onednn_allow_tf32(_allow_tf32) + if _fp32_precision is not None: + torch._C._set_fp32_precision_setter("mkldnn", "all", _fp32_precision) + return orig_flags + + +@contextmanager +def flags(enabled=False, deterministic=False, allow_tf32=True, fp32_precision="none"): + with __allow_nonbracketed_mutation(): + orig_flags = set_flags(enabled, deterministic, allow_tf32, fp32_precision) + try: + yield + finally: + with __allow_nonbracketed_mutation(): + set_flags(*orig_flags) + + +class MkldnnModule(PropModule): + def is_available(self): + return is_available() + + enabled = ContextProp(torch._C._get_mkldnn_enabled, torch._C._set_mkldnn_enabled) + deterministic = ContextProp( + torch._C._get_mkldnn_deterministic, torch._C._set_mkldnn_deterministic + ) + allow_tf32 = ContextProp( + torch._C._get_onednn_allow_tf32, torch._C._set_onednn_allow_tf32 + ) + matmul = _FP32Precision("mkldnn", "matmul") + conv = _FP32Precision("mkldnn", "conv") + rnn = _FP32Precision("mkldnn", "rnn") + fp32_precision = ContextProp( + _get_fp32_precision_getter("mkldnn", "all"), + _set_fp32_precision_setter("generic", "all"), + ) + + +if TYPE_CHECKING: + enabled: ContextProp + deterministic: ContextProp + allow_tf32: ContextProp + fp32_precision: str + +sys.modules[__name__] = MkldnnModule(sys.modules[__name__], __name__) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2807eee9825731836a0c1f8d793f7b072c2e1e2a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/mkldnn/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mps/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/mps/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e4a2f91374f4f306a7346618bc58b91c248cfbb8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/mps/__init__.py @@ -0,0 +1,60 @@ +from functools import lru_cache as _lru_cache +from typing import Optional, TYPE_CHECKING + +import torch +from torch.library import Library as _Library + + +__all__ = [ + "get_core_count", + "get_name", + "is_built", + "is_available", + "is_macos13_or_newer", + "is_macos_or_newer", +] + + +def is_built() -> bool: + r"""Return whether PyTorch is built with MPS support. + + Note that this doesn't necessarily mean MPS is available; just that + if this PyTorch binary were run on a machine with working MPS drivers + and devices, we would be able to use it. + """ + return torch._C._has_mps + + +@_lru_cache +def is_available() -> bool: + r"""Return a bool indicating if MPS is currently available.""" + return torch._C._mps_is_available() + + +@_lru_cache +def is_macos_or_newer(major: int, minor: int) -> bool: + r"""Return a bool indicating whether MPS is running on given MacOS or newer.""" + return torch._C._mps_is_on_macos_or_newer(major, minor) + + +@_lru_cache +def is_macos13_or_newer(minor: int = 0) -> bool: + r"""Return a bool indicating whether MPS is running on MacOS 13 or newer.""" + return torch._C._mps_is_on_macos_or_newer(13, minor) + + +@_lru_cache +def get_name() -> str: + r"""Return Metal device name""" + return torch._C._mps_get_name() + + +@_lru_cache +def get_core_count() -> int: + r"""Return GPU core count. + + According to the documentation, one core is comprised of 16 Execution Units. + One execution Unit has 8 ALUs. + And one ALU can run 24 threads, i.e. one core is capable of executing 3072 threads concurrently. + """ + return torch._C._mps_get_core_count() diff --git a/.venv/lib/python3.12/site-packages/torch/backends/mps/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/mps/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c11271abdb2dc0a035ff42446a792b7d1d5f4002 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/mps/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..8d8a72f3cda9b0da16702c0d7c6fe92ae8f3f153 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__init__.py @@ -0,0 +1,32 @@ +# mypy: allow-untyped-defs +from contextlib import contextmanager + +import torch +from torch.backends import __allow_nonbracketed_mutation, ContextProp, PropModule + + +__all__ = ["is_available", "flags", "set_flags"] + + +def is_available(): + r"""Return whether PyTorch is built with NNPACK support.""" + return torch._nnpack_available() + + +def set_flags(_enabled): + r"""Set if nnpack is enabled globally""" + orig_flags = (torch._C._get_nnpack_enabled(),) + torch._C._set_nnpack_enabled(_enabled) + return orig_flags + + +@contextmanager +def flags(enabled=False): + r"""Context manager for setting if nnpack is enabled globally""" + with __allow_nonbracketed_mutation(): + orig_flags = set_flags(enabled) + try: + yield + finally: + with __allow_nonbracketed_mutation(): + set_flags(orig_flags[0]) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8408552cf82bd65d8423afe918ac553440701e57 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/nnpack/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/openmp/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/openmp/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..aff8d46cd4ac2d9ff49942542d99ac2afbb85896 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/openmp/__init__.py @@ -0,0 +1,7 @@ +# mypy: allow-untyped-defs +import torch + + +def is_available(): + r"""Return whether PyTorch is built with OpenMP support.""" + return torch._C.has_openmp diff --git a/.venv/lib/python3.12/site-packages/torch/backends/openmp/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/openmp/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..43048f97d42054389168fb5694c8896f6ab020e4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/openmp/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..264be78aa9a1c24a4624a87782e2b2c5afd29c05 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__init__.py @@ -0,0 +1,117 @@ +# mypy: allow-untyped-defs +import sys +import warnings +from contextlib import contextmanager +from functools import lru_cache as _lru_cache +from typing import Any + +from torch.backends import __allow_nonbracketed_mutation, ContextProp, PropModule + + +try: + import opt_einsum as _opt_einsum # type: ignore[import] +except ImportError: + _opt_einsum = None + + +@_lru_cache +def is_available() -> bool: + r"""Return a bool indicating if opt_einsum is currently available. + + You must install opt-einsum in order for torch to automatically optimize einsum. To + make opt-einsum available, you can install it along with torch: ``pip install torch[opt-einsum]`` + or by itself: ``pip install opt-einsum``. If the package is installed, torch will import + it automatically and use it accordingly. Use this function to check whether opt-einsum + was installed and properly imported by torch. + """ + return _opt_einsum is not None + + +def get_opt_einsum() -> Any: + r"""Return the opt_einsum package if opt_einsum is currently available, else None.""" + return _opt_einsum + + +def _set_enabled(_enabled: bool) -> None: + if not is_available() and _enabled: + raise ValueError( + f"opt_einsum is not available, so setting `enabled` to {_enabled} will not reap " + "the benefits of calculating an optimal path for einsum. torch.einsum will " + "fall back to contracting from left to right. To enable this optimal path " + "calculation, please install opt-einsum." + ) + global enabled + enabled = _enabled + + +def _get_enabled() -> bool: + return enabled + + +def _set_strategy(_strategy: str) -> None: + if not is_available(): + raise ValueError( + f"opt_einsum is not available, so setting `strategy` to {_strategy} will not be meaningful. " + "torch.einsum will bypass path calculation and simply contract from left to right. " + "Please install opt_einsum or unset `strategy`." + ) + if not enabled: + raise ValueError( + f"opt_einsum is not enabled, so setting a `strategy` to {_strategy} will not be meaningful. " + "torch.einsum will bypass path calculation and simply contract from left to right. " + "Please set `enabled` to `True` as well or unset `strategy`." + ) + if _strategy not in ["auto", "greedy", "optimal"]: + raise ValueError( + f"`strategy` must be one of the following: [auto, greedy, optimal] but is {_strategy}" + ) + global strategy + strategy = _strategy + + +def _get_strategy() -> str: + # pyrefly: ignore [bad-return] + return strategy + + +def set_flags(_enabled=None, _strategy=None): + orig_flags = (enabled, None if not is_available() else strategy) + if _enabled is not None: + _set_enabled(_enabled) + if _strategy is not None: + _set_strategy(_strategy) + return orig_flags + + +@contextmanager +def flags(enabled=None, strategy=None): + with __allow_nonbracketed_mutation(): + orig_flags = set_flags(enabled, strategy) + try: + yield + finally: + # recover the previous values + with __allow_nonbracketed_mutation(): + set_flags(*orig_flags) + + +# The magic here is to allow us to intercept code like this: +# +# torch.backends.opt_einsum.enabled = True + + +class OptEinsumModule(PropModule): + global enabled + enabled = ContextProp(_get_enabled, _set_enabled) + global strategy + strategy = None + if is_available(): + strategy = ContextProp(_get_strategy, _set_strategy) + + +# This is the sys.modules replacement trick, see +# https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273 +sys.modules[__name__] = OptEinsumModule(sys.modules[__name__], __name__) + +enabled = bool(is_available()) +strategy = "auto" if is_available() else None diff --git a/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b0a81e9b1e7da9426f12eb63ff5bdb7b31caf671 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/opt_einsum/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/python_native/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/python_native/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..1e7848e26686c2d4c0a593e04d29d638ddd5c8be --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/python_native/__init__.py @@ -0,0 +1,395 @@ +""" +User-facing API for controlling DSL operation overrides. + +The torch.backends.python_native module provides control over DSL (Domain Specific Language) +operation overrides defined in torch._native. This allows users to selectively enable or disable +high-performance implementations from various DSLs like Triton and CuteDSL. + +The module supports both coarse-grained control (entire DSLs) and fine-grained control +(individual operations or dispatch keys). All control operations support context managers +for temporary state changes. + +Example usage:: + + import torch.backends.python_native as pn + + # DSL-level control + pn.triton.enabled = False # Disable all triton ops + pn.cutedsl.enabled = True # Enable all cutedsl ops + + # Individual operation control + pn.disable_operations("scaled_mm") # Disable specific op across all DSLs + pn.enable_operations("scaled_mm") # Re-enable specific op + + # Context manager support + with pn.triton.disabled(): + result = some_computation() # Triton ops disabled here + + # Query capabilities + print(pn.available_dsls) # ['triton', 'cutedsl'] + print(pn.get_dsl_operations("triton")) # Operations for triton +""" + +import functools +import sys +import types +from contextlib import contextmanager + +from torch.backends import ContextProp, flags_frozen, PropModule + + +@contextmanager +def _preserve_filter_state(): + """Context manager to save and restore registry filter state.""" + filter_state = _get_filter_state() + + # Save original state + original_state = ( + set(filter_state._dsl_names), + set(filter_state._op_symbols), + set(filter_state._dispatch_keys), + ) + + try: + yield filter_state + finally: + # Restore original state + filter_state._dsl_names.clear() + filter_state._op_symbols.clear() + filter_state._dispatch_keys.clear() + + filter_state._dsl_names.update(original_state[0]) + filter_state._op_symbols.update(original_state[1]) + filter_state._dispatch_keys.update(original_state[2]) + + +def _get_dsl_registry(): + """Lazy import to avoid circular imports.""" + from torch._native.dsl_registry import dsl_registry + + return dsl_registry + + +def _get_registry_functions(): + """Lazy import of registry functions.""" + from torch._native.registry import ( + _filter_state, + _graphs, + deregister_op_overrides, + reenable_op_overrides, + ) + + return deregister_op_overrides, reenable_op_overrides, _graphs, _filter_state + + +def _get_filter_state(): + """Direct access to filter state.""" + return _get_registry_functions()[3] + + +def _get_dsl_module(dsl_name: str): + """Get the registered DSL module for direct control. + + Uses the DSL registry to dynamically look up DSL modules instead of + hard-coding the mapping. This makes the function automatically extensible + for new DSLs without code changes. + + Args: + dsl_name (str): Name of the DSL to retrieve. + + Returns: + DSLModuleProtocol: The registered DSL module. + + Raises: + ValueError: If the DSL is not registered. + """ + registry = _get_dsl_registry() + + # Use the public API to get the DSL module + dsl_module = registry.get_dsl_module(dsl_name) + if dsl_module is not None: + return dsl_module + else: + raise ValueError( + f"Unknown DSL: {dsl_name}. Available DSLs: {registry.list_all_dsls()}" + ) + + +class DSLController: + """Controller for a specific DSL.""" + + def __init__(self, dsl_name: str): + self._dsl_name = dsl_name + + @property + def name(self) -> str: + return self._dsl_name + + @property + def available(self) -> bool: + """Check if DSL runtime is available.""" + registry = _get_dsl_registry() + return registry.is_dsl_available(self._dsl_name) + + @property + def version(self): + """Get DSL version.""" + registry = _get_dsl_registry() + return registry.get_dsl_version(self._dsl_name) + + @property + def backend(self) -> str: + """Chosen lowering backend for DSLs that have one (e.g. helion).""" + chosen = getattr(_get_dsl_module(self._dsl_name), "_chosen_backend", None) + if chosen is None: + raise AttributeError(f"{self._dsl_name} DSL has no configurable backend") + return chosen() + + @property + def enabled(self) -> bool: + """Check if DSL is currently enabled.""" + filter_state = _get_filter_state() + return self._dsl_name not in filter_state._dsl_names + + @enabled.setter + def enabled(self, value: bool): + """Enable or disable the DSL.""" + if flags_frozen(): + raise RuntimeError( + f"not allowed to set {self._dsl_name} DSL flags " + "after disable_global_flags; please use flags() context manager instead" + ) + if value: + self.enable() + else: + self.disable() + + def disable(self): + """Disable all operations for this DSL.""" + dsl_module = _get_dsl_module(self._dsl_name) + dsl_module.deregister_op_overrides() + + def enable(self): + """Re-enable all operations for this DSL.""" + reenable_op_overrides = _get_registry_functions()[1] + reenable_op_overrides(enable_dsl_names=self._dsl_name) + + @contextmanager + def disabled(self): + """Context manager to temporarily disable DSL.""" + original_state = self.enabled + try: + self.disable() + yield + finally: + if original_state: + self.enable() + + def __repr__(self): + status = "available" if self.available else "unavailable" + enabled_status = "enabled" if self.enabled else "disabled" + return f"DSLController({self._dsl_name}, {status}, {enabled_status})" + + +class PythonNativeModule(PropModule): + """Main module for python_native DSL control.""" + + def __init__(self, original_module): + super().__init__(original_module, original_module.__name__) + + @property + def available_dsls(self) -> list[str]: + """Get list of available DSLs.""" + registry = _get_dsl_registry() + result = registry.list_available_dsls() + return list(result) if not isinstance(result, list) else result + + @property + def all_dsls(self) -> list[str]: + """Get list of all registered DSLs.""" + registry = _get_dsl_registry() + result = registry.list_all_dsls() + return list(result) if not isinstance(result, list) else result + + def get_dsl_operations(self, dsl_name: str) -> list[str]: + """Get list of operations registered by a specific DSL. + + Args: + dsl_name (str): Name of the DSL to query (e.g., 'triton', 'cutedsl'). + + Returns: + list[str]: Sorted list of operation names registered by the DSL. + + Example:: + + ops = torch.backends.python_native.get_dsl_operations("triton") + print(ops) # ['triton_to_mxfp8_dim0', ...] + """ + from torch._native.registry import get_dsl_operations + + return get_dsl_operations(dsl_name) + + def disable_operations(self, *op_symbols: str): + """Disable specific operations across all DSLs. + + Args: + *op_symbols (str): Names of operations to disable. + + Example:: + + # Disable scaled matrix multiply across all DSLs + torch.backends.python_native.disable_operations("scaled_mm") + + # Disable multiple operations + torch.backends.python_native.disable_operations( + "scaled_mm", "flash_attention" + ) + """ + deregister_op_overrides = _get_registry_functions()[0] + deregister_op_overrides(disable_op_symbols=list(op_symbols)) + + def enable_operations(self, *op_symbols: str): + """Re-enable specific operations across all DSLs. + + Args: + *op_symbols (str): Names of operations to re-enable. + + Example:: + + # Re-enable previously disabled operations + torch.backends.python_native.enable_operations( + "scaled_mm", "flash_attention" + ) + """ + reenable_op_overrides = _get_registry_functions()[1] + reenable_op_overrides(enable_op_symbols=list(op_symbols)) + + def disable_dispatch_keys(self, *dispatch_keys: str): + """Disable operations at specific dispatch keys. + + Args: + *dispatch_keys (str): Dispatch keys to disable (e.g., 'CUDA', 'CPU'). + + Example:: + + # Disable all native operations on CUDA + torch.backends.python_native.disable_dispatch_keys("CUDA") + """ + deregister_op_overrides = _get_registry_functions()[0] + deregister_op_overrides(disable_dispatch_keys=list(dispatch_keys)) + + def enable_dispatch_keys(self, *dispatch_keys: str): + """Re-enable operations at specific dispatch keys. + + Args: + *dispatch_keys (str): Dispatch keys to re-enable (e.g., 'CUDA', 'CPU'). + + Example:: + + # Re-enable native operations on CUDA + torch.backends.python_native.enable_dispatch_keys("CUDA") + """ + reenable_op_overrides = _get_registry_functions()[1] + reenable_op_overrides(enable_dispatch_keys=list(dispatch_keys)) + + @contextmanager + def operations_disabled(self, *op_symbols: str): + """Context manager to temporarily disable operations. + + Args: + *op_symbols (str): Names of operations to temporarily disable. + + Example:: + + with torch.backends.python_native.operations_disabled("scaled_mm"): + # scaled_mm is disabled across all DSLs + result = model(input) + # scaled_mm is automatically re-enabled here + """ + filter_state = _get_filter_state() + previously_disabled_ops = { + op for op in op_symbols if op in filter_state._op_symbols + } + + self.disable_operations(*op_symbols) + try: + yield + finally: + # Only re-enable operations that weren't already disabled + ops_to_reenable = [ + op for op in op_symbols if op not in previously_disabled_ops + ] + if ops_to_reenable: + self.enable_operations(*ops_to_reenable) + + @functools.lru_cache(maxsize=16) # noqa: B019 + def _get_dsl_controller(self, name: str) -> "DSLController": + """Get or create a DSL controller (cached).""" + return DSLController(name) + + def _get_registry_functions(self): + """Expose registry functions for testing.""" + return _get_registry_functions() + + def is_operation_disabled(self, op_symbol: str) -> bool: + """Check if an operation is currently disabled.""" + filter_state = _get_filter_state() + return op_symbol in filter_state._op_symbols + + def is_dsl_disabled(self, dsl_name: str) -> bool: + """Check if a DSL is currently disabled.""" + filter_state = _get_filter_state() + return dsl_name in filter_state._dsl_names + + def __getattr__(self, name: str): + """Dynamic attribute access for DSL controllers.""" + # Skip dunder attributes to avoid triggering DSL registry lookups + # during torch initialization. inspect.getmodule() calls + # hasattr(module, '__file__') which would otherwise cause a circular + # import through _get_dsl_registry() while torch is still loading. + if name.startswith("__") and name.endswith("__"): + raise AttributeError(f"module '{self.__name__}' has no attribute '{name}'") + + if name in self.all_dsls: + return self._get_dsl_controller(name) + + # Expose private functions for testing + if name == "_get_dsl_module": + return _get_dsl_module + if name == "_get_registry_functions": + return self._get_registry_functions + if name == "_get_filter_state": + return _get_filter_state + + raise AttributeError(f"module '{self.__name__}' has no attribute '{name}'") + + def __dir__(self): + """Return available attributes including DSL names.""" + attrs = set(super().__dir__()) + attrs.update( + { + "available_dsls", + "all_dsls", + "get_dsl_operations", + "disable_operations", + "enable_operations", + "disable_dispatch_keys", + "enable_dispatch_keys", + "operations_disabled", + "is_operation_disabled", + "is_dsl_disabled", + } + ) + + # Add DSL names + try: + attrs.update(self.all_dsls) + except Exception: + # If registry not available yet, skip DSL names + pass + + return sorted(attrs) + + +# Replace the current module with our enhanced version +sys.modules[__name__] = PythonNativeModule(sys.modules[__name__]) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/python_native/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/python_native/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a31e2b5a7f323164981f6b48fe29bf8c825352e5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/python_native/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/quantized/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/quantized/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..caabfdf243783f2161a201c6a6ec9bd6eca83b18 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/quantized/__init__.py @@ -0,0 +1,65 @@ +# mypy: allow-untyped-defs +import sys +import types + +import torch + + +# This function should correspond to the enums present in c10/core/QEngine.h +def _get_qengine_id(qengine: str) -> int: + if qengine == "none" or qengine == "" or qengine is None: + ret = 0 + elif qengine == "fbgemm": + ret = 1 + elif qengine == "qnnpack": + ret = 2 + elif qengine == "onednn": + ret = 3 + elif qengine == "x86": + ret = 4 + else: + ret = -1 + raise RuntimeError(f"{qengine} is not a valid value for quantized engine") + return ret + + +# This function should correspond to the enums present in c10/core/QEngine.h +def _get_qengine_str(qengine: int) -> str: + all_engines = {0: "none", 1: "fbgemm", 2: "qnnpack", 3: "onednn", 4: "x86"} + return all_engines.get(qengine, "*undefined") + + +class _QEngineProp: + def __get__(self, obj, objtype) -> str: + return _get_qengine_str(torch._C._get_qengine()) + + def __set__(self, obj, val: str) -> None: + torch._C._set_qengine(_get_qengine_id(val)) + + +class _SupportedQEnginesProp: + def __get__(self, obj, objtype) -> list[str]: + qengines = torch._C._supported_qengines() + return [_get_qengine_str(qe) for qe in qengines] + + def __set__(self, obj, val) -> None: + raise RuntimeError("Assignment not supported") + + +class QuantizedEngine(types.ModuleType): + def __init__(self, m, name): + super().__init__(name) + self.m = m + + def __getattr__(self, attr): + return self.m.__getattribute__(attr) + + engine = _QEngineProp() + supported_engines = _SupportedQEnginesProp() + + +# This is the sys.modules replacement trick, see +# https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273 +sys.modules[__name__] = QuantizedEngine(sys.modules[__name__], __name__) +engine: str +supported_engines: list[str] diff --git a/.venv/lib/python3.12/site-packages/torch/backends/quantized/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/quantized/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4d07173ab52d054cc8119852704af27231479034 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/quantized/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xeon/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/xeon/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..508bff44aeb652aa6f93bdbd0fd3a66666aac89c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/run_cpu.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/run_cpu.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c3e35f182ae49a9b770d6cd54f93b1ce87fa6440 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/xeon/__pycache__/run_cpu.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xeon/run_cpu.py b/.venv/lib/python3.12/site-packages/torch/backends/xeon/run_cpu.py new file mode 100644 index 0000000000000000000000000000000000000000..f96a6c416696b480126e2b398ec6305f420d45af --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/xeon/run_cpu.py @@ -0,0 +1,1028 @@ +# mypy: allow-untyped-defs +""" +This is a script for launching PyTorch inference on Intel(R) Xeon(R) Scalable Processors with optimal configurations. + +Single instance inference, multi-instance inference are enabled. + +Note: term "instance" here doesn't refer to a cloud instance. This script is executed as a single process. It invokes +multiple "instances" which are formed from multiple threads for each. "instance" is kind of group of threads in this +context. + +Illustrated as below: + +:: + + +-----------------------------+----------------------+-------+ + | process | thread | core | + +=============================+======================+=======+ + | torch.backends.xeon.run_cpu | instance 0: thread 0 | 0 | + | | thread 1 | 1 | + | +----------------------+-------+ + | | instance 1: thread 0 | 2 | + | | thread 1 | 3 | + | +----------------------+-------+ + | | ... | ... | + | +----------------------+-------+ + | | instance N: thread 0 | M | + | | thread 1 | M+1 | + +-----------------------------+----------------------+-------+ + +To get the peak performance on Intel(R) Xeon(R) Scalable Processors, the script optimizes the configuration of thread and memory +management. For thread management, the script configures thread affinity and the preload of Intel OMP library. +For memory management, it configures NUMA binding and preload optimized memory allocation library (e.g. tcmalloc, jemalloc). + +Environment variables that will be set by this script: + ++------------------+-------------------------------------------------------------------------------------------------+ +| Environ Variable | Value | ++==================+=================================================================================================+ +| LD_PRELOAD | Depending on knobs you set, /libiomp5.so, /libjemalloc.so, /libtcmalloc.so might | +| | be appended to LD_PRELOAD. | ++------------------+-------------------------------------------------------------------------------------------------+ +| KMP_AFFINITY | If libiomp5.so is preloaded, KMP_AFFINITY could be set to "granularity=fine,compact,1,0". | ++------------------+-------------------------------------------------------------------------------------------------+ +| KMP_BLOCKTIME | If libiomp5.so is preloaded, KMP_BLOCKTIME is set to "1". | ++------------------+-------------------------------------------------------------------------------------------------+ +| OMP_NUM_THREADS | value of ncores_per_instance | ++------------------+-------------------------------------------------------------------------------------------------+ +| MALLOC_CONF | If libjemalloc.so is preloaded, MALLOC_CONF will be set to | +| | "oversize_threshold:1,background_thread:true,metadata_thp:auto". | ++------------------+-------------------------------------------------------------------------------------------------+ + +*Note*: This script respects environment variables set preliminarily. I.e. If you set the environment variables +mentioned above before running the script, the script will not overwrite the values in the script. + +How to use this module: +~~~~~~~~~~~~~~~~~~~~~~~ + +Single instance inference +------------------------- + +1. Run single-instance inference on a single node with all CPU nodes. + +:: + + python -m torch.backends.xeon.run_cpu --throughput-mode script.py args + +2. Run single-instance inference on a single CPU node. + +:: + + python -m torch.backends.xeon.run_cpu --node-id 1 script.py args + +Multi-instance inference +------------------------ + +1. Multi-instance + By default this tool runs one process per node. If you want to set the instance numbers and core per instance, + --ninstances and --ncores-per-instance should be set. + +:: + + python -m torch.backends.xeon.run_cpu -- python_script args + + eg: on an Intel(R) Xeon(R) Scalable Processor with 14 instance, 4 cores per instance + +:: + + python -m torch.backends.xeon.run_cpu --ninstances 14 --ncores-per-instance 4 python_script args + +2. Run single-instance inference among multiple instances. + By default, runs all ninstances. If you want to independently run a single instance among ninstances, specify rank. + + eg: run 0th instance on an Intel(R) Xeon(R) Scalable Processor with 2 instance (i.e., numactl -C 0-27) + +:: + + python -m torch.backends.xeon.run_cpu --ninstances 2 --rank 0 python_script args + + eg: run 1st instance on an Intel(R) Xeon(R) Scalable Processor with 2 instance (i.e., numactl -C 28-55) + +:: + + python -m torch.backends.xeon.run_cpu --ninstances 2 --rank 1 python_script args + + eg: run 0th instance on an Intel(R) Xeon(R) Scalable Processor with 2 instance, 2 cores per instance, + first four cores (i.e., numactl -C 0-1) + +:: + + python -m torch.backends.xeon.run_cpu --core-list "0, 1, 2, 3" --ninstances 2 --ncores-per-instance 2 + --rank 0 python_script args + +3. To look up what optional arguments this module offers: + +:: + + python -m torch.backends.xeon.run_cpu --help + +Memory allocator +---------------- + +"--enable-tcmalloc" and "--enable-jemalloc" can be used to enable different memory allocator. + +""" + +import glob +import logging +import os +import platform +import re +import subprocess +import sys +from argparse import ArgumentParser, RawTextHelpFormatter, REMAINDER +from os.path import expanduser + +from torch.distributed.elastic.multiprocessing import ( + DefaultLogsSpecs as _DefaultLogsSpecs, + start_processes, + Std, +) + + +format_str = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" +logging.basicConfig(level=logging.INFO, format=format_str) +logger = logging.getLogger(__name__) + + +class _CPUinfo: + """Get CPU information, such as cores list and NUMA information.""" + + def __init__(self, test_input=""): + self.cpuinfo = [] + if platform.system() in ["Windows", "Darwin"]: + raise RuntimeError(f"{platform.system()} is not supported!!!") + elif platform.system() == "Linux": + # Sample output of: `lscpu --parse=CPU,Core,Socket,Node` + # + # # The following is the parsable format, which can be fed to other + # # programs. Each different item in every column has a unique ID + # # starting from zero. + # # CPU,Core,Socket,Node + # 0,0,0,0 + # 1,1,0,0 + # ... + if test_input == "": + lscpu_cmd = ["lscpu", "--parse=CPU,Core,Socket,Node"] + lscpu_info = subprocess.check_output( + lscpu_cmd, universal_newlines=True + ).split("\n") + else: + lscpu_info = test_input.split("\n") + + # Get information about cpu, core, socket and node + for line in lscpu_info: + pattern = r"^([\d]+,[\d]+,[\d]+,[\d]?)" + regex_out = re.search(pattern, line) + if regex_out: + self.cpuinfo.append(regex_out.group(1).strip().split(",")) + + # physical cores := core column in lscpu output + # logical cores := cPU column in lscpu output + self.node_nums = int(max(line[3] for line in self.cpuinfo)) + 1 + self.node_physical_cores: list[list[int]] = [] # node_id is index + self.node_logical_cores: list[list[int]] = [] # node_id is index + self.physical_core_node_map = {} # physical core to numa node id + self.logical_core_node_map = {} # logical core to numa node id + + for node_id in range(self.node_nums): + cur_node_physical_core = [] + cur_node_logical_core = [] + for cpuinfo in self.cpuinfo: + nid = cpuinfo[3] if cpuinfo[3] != "" else "0" + if node_id == int(nid): + if int(cpuinfo[1]) not in cur_node_physical_core: + cur_node_physical_core.append(int(cpuinfo[1])) + self.physical_core_node_map[int(cpuinfo[1])] = int(node_id) + cur_node_logical_core.append(int(cpuinfo[0])) + self.logical_core_node_map[int(cpuinfo[0])] = int(node_id) + self.node_physical_cores.append(cur_node_physical_core) + self.node_logical_cores.append(cur_node_logical_core) + + def _physical_core_nums(self): + overall_core_num = 0 + for phycisal_core_list in self.node_physical_cores: + overall_core_num += len(phycisal_core_list) + return overall_core_num + + def _logical_core_nums(self): + overall_core_num = 0 + for logical_core_list in self.node_logical_cores: + overall_core_num += len(logical_core_list) + return overall_core_num + + def get_node_physical_cores(self, node_id): + if node_id < 0 or node_id > self.node_nums - 1: + raise ValueError( + f"Invalid node id: {node_id}. Valid node ids: {list(range(len(self.node_physical_cores)))}" + ) + return self.node_physical_cores[node_id] + + def get_node_logical_cores(self, node_id): + if node_id < 0 or node_id > self.node_nums - 1: + raise ValueError( + f"Invalid node id: {node_id}. Valid node ids: {list(range(len(self.node_physical_cores)))}" + ) + return self.node_logical_cores[node_id] + + def get_all_physical_cores(self): + all_cores = [] + for cores in self.node_physical_cores: + all_cores.extend(cores) + return all_cores + + def get_all_logical_cores(self): + all_cores = [] + for cores in self.node_logical_cores: + all_cores.extend(cores) + return all_cores + + def numa_aware_check(self, core_list): + """ + Check whether all cores in core_list are in the same NUMA node. + + Cross NUMA will reduce performance. + We strongly advise to not use cores on different nodes. + """ + cores_numa_map = self.logical_core_node_map + numa_ids = [] + for core in core_list: + numa_id = cores_numa_map[core] + if numa_id not in numa_ids: + numa_ids.append(numa_id) + if len(numa_ids) > 1: + logger.warning( + "Numa Aware: cores:%s on different NUMA nodes:%s. To avoid \ +this behavior, please use --ncores-per-instance knob to make sure number of cores is divisible by --ncores-per-\ +instance. Alternatively, please use --skip-cross-node-cores knob.", + core_list, + numa_ids, + ) + if len(numa_ids) == 0: + raise RuntimeError( + "invalid number of NUMA nodes; please make sure numa_ids >= 1" + ) + return numa_ids + + +class _Launcher: + r"""Class for launcher.""" + + msg_lib_notfound = ( + f"Unable to find the {{0}} library file lib{{1}}.so in $CONDA_PREFIX/lib or $VIRTUAL_ENV/lib \ +or /.local/lib/ or /usr/local/lib/ or /usr/local/lib64/ or /usr/lib or /usr/lib64 or \ +{expanduser('~')}/.local/lib/ so the LD_PRELOAD environment variable will not be set." + ) + + def __init__(self) -> None: + self.cpuinfo = _CPUinfo() + + def add_lib_preload(self, lib_type): + """Enable TCMalloc/JeMalloc/intel OpenMP.""" + library_paths = [] + if "CONDA_PREFIX" in os.environ: + library_paths.append(f"{os.environ['CONDA_PREFIX']}/lib") + if "VIRTUAL_ENV" in os.environ: + library_paths.append(f"{os.environ['VIRTUAL_ENV']}/lib") + + library_paths += [ + f"{expanduser('~')}/.local/lib", + "/usr/local/lib", + "/usr/local/lib64", + "/usr/lib", + "/usr/lib64", + ] + + lib_find = False + lib_set = False + for item in os.getenv("LD_PRELOAD", "").split(":"): + if item.endswith(f"lib{lib_type}.so"): + lib_set = True + break + if not lib_set: + for lib_path in library_paths: + # pyrefly: ignore [unbound-name] + library_file = os.path.join(lib_path, f"lib{lib_type}.so") + matches = glob.glob(library_file) + if len(matches) > 0: + # pyrefly: ignore [unbound-name] + ld_preloads = [f"{matches[0]}", os.getenv("LD_PRELOAD", "")] + # pyrefly: ignore [unbound-name] + os.environ["LD_PRELOAD"] = os.pathsep.join( + # pyrefly: ignore [unbound-name] + [p.strip(os.pathsep) for p in ld_preloads if p] + ) + lib_find = True + break + return lib_set or lib_find + + def is_numactl_available(self): + numactl_available = False + try: + cmd = ["numactl", "-C", "0", "-m", "0", "hostname"] + r = subprocess.run( + cmd, + env=os.environ, + stdout=subprocess.DEVNULL, + stderr=subprocess.DEVNULL, + check=False, + ) + if r.returncode == 0: + numactl_available = True + except Exception: + pass + return numactl_available + + def set_memory_allocator( + self, enable_tcmalloc=True, enable_jemalloc=False, use_default_allocator=False + ): + """ + Enable TCMalloc/JeMalloc with LD_PRELOAD and set configuration for JeMalloc. + + By default, PTMalloc will be used for PyTorch, but TCMalloc and JeMalloc can get better + memory reuse and reduce page fault to improve performance. + """ + if enable_tcmalloc and enable_jemalloc: + raise RuntimeError( + "Unable to enable TCMalloc and JEMalloc at the same time." + ) + + if enable_tcmalloc: + find_tc = self.add_lib_preload(lib_type="tcmalloc") + if not find_tc: + msg = f'{self.msg_lib_notfound} you can use "conda install -c conda-forge gperftools" to install {{0}}' + logger.warning(msg.format("TCmalloc", "tcmalloc")) + else: + logger.info("Use TCMalloc memory allocator") + + elif enable_jemalloc: + find_je = self.add_lib_preload(lib_type="jemalloc") + if not find_je: + msg = f'{self.msg_lib_notfound} you can use "conda install -c conda-forge jemalloc" to install {{0}}' + logger.warning(msg.format("Jemalloc", "jemalloc")) + else: + logger.info("Use JeMalloc memory allocator") + self.set_env( + "MALLOC_CONF", + "oversize_threshold:1,background_thread:true,metadata_thp:auto", + ) + + elif use_default_allocator: + pass + + else: + find_tc = self.add_lib_preload(lib_type="tcmalloc") + if find_tc: + logger.info("Use TCMalloc memory allocator") + return + find_je = self.add_lib_preload(lib_type="jemalloc") + if find_je: + logger.info("Use JeMalloc memory allocator") + return + logger.warning( + """Neither TCMalloc nor JeMalloc is found in $CONDA_PREFIX/lib or $VIRTUAL_ENV/lib + or /.local/lib/ or /usr/local/lib/ or /usr/local/lib64/ or /usr/lib or /usr/lib64 or + %s/.local/lib/ so the LD_PRELOAD environment variable will not be set. + This may drop the performance""", + expanduser("~"), + ) + + def log_env_var(self, env_var_name=""): + if env_var_name in os.environ: + logger.info("%s=%s", env_var_name, os.environ[env_var_name]) + + def set_env(self, env_name, env_value): + if not env_value: + logger.warning("%s is None", env_name) + if env_name not in os.environ: + os.environ[env_name] = env_value + elif os.environ[env_name] != env_value: + logger.warning( + "Overriding value with the one set in environment variable: %s. \ +Value applied: %s. Value ignored: %s", + env_name, + os.environ[env_name], + env_value, + ) + self.log_env_var(env_name) + + # set_kmp_affinity is used to control whether to set KMP_AFFINITY or not. + # In scenario that use all cores on all nodes, including logical cores, setting KMP_AFFINITY disables logical cores. + # In this case, KMP_AFFINITY should not be set. + def set_multi_thread_and_allocator( + self, + disable_iomp=False, + set_kmp_affinity=True, + enable_tcmalloc=True, + enable_jemalloc=False, + use_default_allocator=False, + ): + """ + Set multi-thread configuration and enable Intel openMP and TCMalloc/JeMalloc. + + By default, GNU openMP and PTMalloc are used in PyTorch. but Intel openMP and TCMalloc/JeMalloc are better alternatives + to get performance benefit. + """ + self.set_memory_allocator( + enable_tcmalloc, enable_jemalloc, use_default_allocator + ) + if not disable_iomp: + find_iomp = self.add_lib_preload(lib_type="iomp5") + if not find_iomp: + msg = f'{self.msg_lib_notfound} you can use "conda install mkl" to install {{0}}' + logger.warning(msg.format("iomp", "iomp5")) + else: + logger.info("Using Intel OpenMP") + if set_kmp_affinity: + self.set_env("KMP_AFFINITY", "granularity=fine,compact,1,0") + self.set_env("KMP_BLOCKTIME", "1") + self.log_env_var("LD_PRELOAD") + + r""" + Launcher for single instance and multi-instance + """ + + def launch(self, args): + cores = [] + set_kmp_affinity = True + enable_taskset = False + # ncores_per_instance param check and expand for multi-instance + if not isinstance(args.ncores_per_instance, list): + args.ncores_per_instance = [args.ncores_per_instance] + args.ncores_per_instance = [int(x) for x in args.ncores_per_instance] + if args.ncores_per_instance != [-1]: + if args.ninstances > 0: + if len(args.ncores_per_instance) == 1: + args.ncores_per_instance = ( + args.ncores_per_instance * args.ninstances + ) + if len(args.ncores_per_instance) != args.ninstances: + raise ValueError( + 'Mismatch for "--ninstances" and "--ncores-per-instance" settings.' + ) + elif len(args.ncores_per_instance) > 1: + args.ninstances = len(args.ncores_per_instance) + if args.core_list: # user specify what cores will be used by params + cores = [int(x) for x in args.core_list.split(",")] + if args.ncores_per_instance == [-1]: + raise RuntimeError( + 'please specify the "--ncores-per-instance" if you have pass the --core-list params' + ) + if args.ninstances > 0 and sum(args.ncores_per_instance) < len(cores): + logger.warning( + "only first %s cores will be used, \ +but you specify %s cores in core_list", + sum(args.ncores_per_instance), + len(cores), + ) + else: + if args.use_logical_core: + if args.node_id != -1: + cores = self.cpuinfo.get_node_logical_cores(args.node_id) + else: + cores = self.cpuinfo.get_all_logical_cores() + # When using all cores on all nodes, including logical cores, + # setting KMP_AFFINITY disables logical cores. Thus, KMP_AFFINITY should not be set. + set_kmp_affinity = False + else: + if args.node_id != -1: + cores = self.cpuinfo.get_node_physical_cores(args.node_id) + else: + cores = self.cpuinfo.get_all_physical_cores() + if ( + not args.multi_instance + and args.ninstances == -1 + and args.ncores_per_instance == [-1] + ): + args.ninstances = 1 + args.ncores_per_instance = [len(cores)] + elif ( + args.multi_instance + and args.ninstances == -1 + and args.ncores_per_instance == [-1] + ): + args.throughput_mode = True + elif args.ncores_per_instance == [-1] and args.ninstances != -1: + if args.ninstances > len(cores): + raise RuntimeError( + f"there are {len(cores)} total cores but you specify {args.ninstances} ninstances; \ +please make sure ninstances <= total_cores)" + ) + if args.skip_cross_node_cores: + logger.warning( + "with --skip-cross-node-cores set, the first %d sub-numa node(s) will be used \ +to launch the instance(s), each node for 1 instance.", + args.ninstances, + ) + if args.ninstances > self.cpuinfo.node_nums: + raise ValueError( + "--ninstances should not be larger than numa node number \ +when --skip-cross-node-cores is set" + ) + cores.clear() + args.ncores_per_instance.clear() + for i in range(args.ninstances): + per_node_cores = self.cpuinfo.get_node_physical_cores(i) + cores.extend(per_node_cores) + args.ncores_per_instance.append(len(per_node_cores)) + else: + logger.warning( + "all the CPU cores on the machine will be distributed \ +as evenly as possible to each instance. It is possible that some cores in an instance may cross the numa node." + ) + args.ncores_per_instance = [ + len(cores) // args.ninstances + ] * args.ninstances + elif args.ncores_per_instance != [-1]: + if args.skip_cross_node_cores: + utilized_node_cores = ( + self.cpuinfo.node_logical_cores + if args.use_logical_core + else self.cpuinfo.node_physical_cores + ) + ncore_per_node = [len(c) for c in utilized_node_cores] + if len(args.ncores_per_instance) > len(ncore_per_node): + # too many ncores_per_instance to skip cross-node cores + raise ValueError( + "there are %s nodes, but %s ncores-per-instance elements and \ +--skip-cross-node-cores is specified.", + len(ncore_per_node), + len(args.ncores_per_instance), + ) + leftover_cores = list() + for i, core_num in enumerate(args.ncores_per_instance): + leftover_num = core_num - ncore_per_node[i] + if leftover_num > 0: + # too many ncores_per_instance for a node + raise ValueError( + "there are %s core(s) in node %s, but specified %s cores for \ +this node and --skip-cross-node-cores is specified.", + ncore_per_node[i], + i, + core_num, + ) + elif leftover_num < 0: + # Exclude excessive cores from largest core IDs + leftover_cores.extend(utilized_node_cores[i][leftover_num:]) + # used node num < total node num, mark the rest node cores as unused + for j in range(i + 1, len(ncore_per_node)): + leftover_cores.extend(utilized_node_cores[j][:]) + for core_rm in leftover_cores: + if core_rm in cores: + cores.remove(core_rm) + args.ninstances = len(args.ncores_per_instance) + else: + if sum(args.ncores_per_instance) > len(cores): + raise RuntimeError( + "Please make sure the sum up of ncores_per_instance <= total_cores" + ) + if args.latency_mode: + logger.warning( + "--latency-mode is exclusive to --ninstances, --ncores-per-instance, --node-id and \ +--use-logical-core. They won't take effect even they are set explicitly." + ) + core_num_per_instance = 4 + cores = list() + args.ninstances = 0 + if args.skip_cross_node_cores: + # Per-numa core allocation, the remainder cores are not used + for node_id, node_core_list in enumerate( + self.cpuinfo.node_physical_cores + ): + node_core_num = len(node_core_list) + for i in range( + core_num_per_instance, + node_core_num + 1, + core_num_per_instance, + ): + args.ninstances += 1 + cores.extend(node_core_list[i - core_num_per_instance : i]) + if i < node_core_num: + logger.warning( + "With --skip-cross-node-cores set, not all the cores on node %d is used \ +because the core number on this node is not dividable by %d.", + node_id, + core_num_per_instance, + ) + args.ncores_per_instance = [core_num_per_instance] * args.ninstances + else: + # Compact allocation of all physical cores, regardless of numa + cores = self.cpuinfo.get_all_physical_cores() + args.ninstances = len(cores) // core_num_per_instance + args.ncores_per_instance = [core_num_per_instance] * args.ninstances + remainder = len(cores) % core_num_per_instance + if remainder > 0: + args.ninstances += 1 + args.ncores_per_instance.append(remainder) + if args.throughput_mode: + logger.warning( + "--throughput-mode is exclusive to --ninstances, --ncores-per-instance, --node-id and \ +--use-logical-core. They won't take effect even they are set explicitly." + ) + args.ninstances = self.cpuinfo.node_nums + cores = self.cpuinfo.get_all_physical_cores() + args.ncores_per_instance.clear() + for node_core_list in self.cpuinfo.node_physical_cores: + args.ncores_per_instance.append(len(node_core_list)) + + if args.ninstances > 1 and args.rank != -1: + logger.info( + "assigning %s cores for instance %s", + args.ncores_per_instance[args.rank], + args.rank, + ) + + if not args.disable_numactl: + numactl_available = self.is_numactl_available() + if not numactl_available: + if not args.disable_taskset: + logger.warning( + "Core binding with numactl is not available. Disabling numactl and using taskset instead. \ + This may affect performance in multi-socket system; please use numactl if memory binding is needed." + ) + args.disable_numactl = True + enable_taskset = True + else: + logger.warning( + "Core binding with numactl is not available, and --disable_taskset is set. \ + Please unset --disable_taskset to use taskset instead of numactl." + ) + sys.exit(-1) + + if not args.disable_taskset: + enable_taskset = True + + self.set_multi_thread_and_allocator( + args.disable_iomp, + set_kmp_affinity, + args.enable_tcmalloc, + args.enable_jemalloc, + args.use_default_allocator, + ) + entrypoint = "" + launch_args = {} + launch_envs: dict[int, dict] = {} + launch_tee = {} + # check whether is launched from torchrun with --nproc-per-node + local_size = int(os.environ.get("LOCAL_WORLD_SIZE", 1)) + local_rank = int(os.environ.get("LOCAL_RANK", 0)) + cores = sorted(cores) + for i in range(args.ninstances): + cmd = [] + cur_process_cores = "" + if not args.disable_numactl or enable_taskset: + if not args.disable_numactl: + cmd = ["numactl"] + elif enable_taskset: + cmd = ["taskset"] + core_index_start = 0 + if ( + args.rank == -1 + ): # sequentially assign ncores_per_instance to ninstances + for j in range(i): + core_index_start += args.ncores_per_instance[j] + core_list = cores[ + core_index_start : core_index_start + + args.ncores_per_instance[i] + ] + else: # assign ncores_per_instance from rank + for j in range(args.rank): + core_index_start += args.ncores_per_instance[j] + core_list = cores[ + core_index_start : core_index_start + + args.ncores_per_instance[args.rank] + ] + + core_ranges: list[dict] = [] + if local_size > 1: + total_num_cores = len(core_list) + cores_per_rank = total_num_cores // local_size + if cores_per_rank < 1: + raise AssertionError( + f"At least one core needs to be assigned to each rank, " + f"got {total_num_cores} cores for {local_size} ranks" + ) + core_list = core_list[ + cores_per_rank * local_rank : cores_per_rank * (local_rank + 1) + ] + for core in core_list: + if len(core_ranges) == 0: + range_elem = {"start": core, "end": core} + core_ranges.append(range_elem) + else: + if core - core_ranges[-1]["end"] == 1: + core_ranges[-1]["end"] = core + else: + range_elem = {"start": core, "end": core} + core_ranges.append(range_elem) + for r in core_ranges: + cur_process_cores = f"{cur_process_cores}{r['start']}-{r['end']}," + cur_process_cores = cur_process_cores[:-1] + if not args.disable_numactl: + numa_params = f"-C {cur_process_cores} " + numa_ids = ",".join( + [ + str(numa_id) + for numa_id in self.cpuinfo.numa_aware_check(core_list) + ] + ) + numa_params += f"-m {numa_ids}" + cmd.extend(numa_params.split()) + elif enable_taskset: + taskset_params = f"-c {cur_process_cores} " + cmd.extend(taskset_params.split()) + with_python = not args.no_python + if with_python: + cmd.append(sys.executable) + cmd.append("-u") + if args.module: + cmd.append("-m") + cmd.append(args.program) + cmd.extend(args.program_args) + cmd_s = " ".join(cmd) + logger.info(cmd_s) + if entrypoint == "": + entrypoint = cmd[0] + del cmd[0] + launch_args[i] = tuple(cmd) + # Use user defined OMP_NUM_THREADS if set explicitly in prior + # Else set it as ncores-per-instance of the current instance + predef_omp_thr_num = os.environ.get("OMP_NUM_THREADS") + launch_envs[i] = ( + {"OMP_NUM_THREADS": str(args.ncores_per_instance[i])} + if predef_omp_thr_num is None + else {} + ) + launch_tee[i] = Std.ALL + + if args.rank != -1: # launches single instance, rank, only + break + + ctx = start_processes( + name=args.log_file_prefix, + entrypoint=entrypoint, + args=launch_args, + envs=launch_envs, + logs_specs=_DefaultLogsSpecs(log_dir=args.log_path, tee=launch_tee), + ) + ctx.wait() + + +def _add_memory_allocator_params(parser): + group = parser.add_argument_group("Memory Allocator Parameters") + # allocator control + group.add_argument( + "--enable-tcmalloc", + "--enable_tcmalloc", + action="store_true", + default=False, + help="Enable tcmalloc allocator", + ) + group.add_argument( + "--enable-jemalloc", + "--enable_jemalloc", + action="store_true", + default=False, + help="Enable jemalloc allocator", + ) + group.add_argument( + "--use-default-allocator", + "--use_default_allocator", + action="store_true", + default=False, + help="Use default memory allocator", + ) + + +def _add_multi_instance_params(parser): + group = parser.add_argument_group("Multi-instance Parameters") + # multi-instance control + group.add_argument( + "--ncores-per-instance", + "--ncores_per_instance", + metavar="\b", + default=-1, + type=int, + nargs="+", + help="Cores per instance", + ) + group.add_argument( + "--ninstances", + metavar="\b", + default=-1, + type=int, + help="For multi-instance, you should give the cores number you used for per instance.", + ) + group.add_argument( + "--skip-cross-node-cores", + "--skip_cross_node_cores", + action="store_true", + default=False, + help="If specified --ncores-per-instance, skips cross-node cores.", + ) + group.add_argument( + "--rank", + metavar="\b", + default="-1", + type=int, + help="Specify instance index to assign ncores_per_instance for rank; \ +otherwise ncores_per_instance will be assigned sequentially to ninstances. Please refer to \ +https://github.com/intel/intel-extension-for-pytorch/blob/master/docs/tutorials/performance_tuning/launch_script.md", + ) + group.add_argument( + "--latency-mode", + "--latency_mode", + action="store_true", + default=False, + help="By default 4 core per instance and use all physical cores", + ) + group.add_argument( + "--throughput-mode", + "--throughput_mode", + action="store_true", + default=False, + help="By default one instance per node and use all physical cores", + ) + group.add_argument( + "--node-id", + "--node_id", + metavar="\b", + default=-1, + type=int, + help="node id for multi-instance, by default all nodes will be used", + ) + group.add_argument( + "--use-logical-core", + "--use_logical_core", + action="store_true", + default=False, + help="Whether only use physical cores", + ) + group.add_argument( + "--disable-numactl", + "--disable_numactl", + action="store_true", + default=False, + help="Disable numactl", + ) + group.add_argument( + "--disable-taskset", + "--disable_taskset", + action="store_true", + default=False, + help="Disable taskset", + ) + group.add_argument( + "--core-list", + "--core_list", + metavar="\b", + default=None, + type=str, + help='Specify the core list as "core_id, core_id, ....", otherwise, all the cores will be used.', + ) + group.add_argument( + "--log-path", + "--log_path", + metavar="\b", + default="", + type=str, + help="The log file directory. Default path is " + ", which means disable logging to files.", + ) + group.add_argument( + "--log-file-prefix", + "--log_file_prefix", + metavar="\b", + default="run", + type=str, + help="log file prefix", + ) + + +def _add_kmp_iomp_params(parser): + group = parser.add_argument_group("IOMP Parameters") + group.add_argument( + "--disable-iomp", + "--disable_iomp", + action="store_true", + default=False, + help="By default, we use Intel OpenMP and libiomp5.so will be add to LD_PRELOAD", + ) + + +def create_args(parser=None): + """ + Parse the command line options. + + @retval ArgumentParser + """ + # pyrefly: ignore [missing-attribute] + parser.add_argument( + "--multi-instance", + "--multi_instance", + action="store_true", + default=False, + help="Enable multi-instance, by default one instance per node", + ) + + # pyrefly: ignore [missing-attribute] + parser.add_argument( + "-m", + "--module", + default=False, + action="store_true", + help="Changes each process to interpret the launch script " + "as a python module, executing with the same behavior as" + '"python -m".', + ) + + # pyrefly: ignore [missing-attribute] + parser.add_argument( + "--no-python", + "--no_python", + default=False, + action="store_true", + help='Do not prepend the --program script with "python" - just exec ' + "it directly. Useful when the script is not a Python script.", + ) + + _add_memory_allocator_params(parser) + _add_kmp_iomp_params(parser) + + _add_multi_instance_params(parser) + # positional + # pyrefly: ignore [missing-attribute] + parser.add_argument( + "program", + type=str, + help="The full path to the program/script to be launched. " + "followed by all the arguments for the script", + ) + + # rest from the training program + # pyrefly: ignore [missing-attribute] + parser.add_argument("program_args", nargs=REMAINDER) + + +def main(args): + env_before = set(os.environ.keys()) + if platform.system() in ["Windows", "Darwin"]: + raise RuntimeError(f"{platform.system()} is not supported!!!") + + if args.log_path: + os.makedirs(args.log_path, exist_ok=True) + else: + args.log_path = os.devnull + + if args.latency_mode and args.throughput_mode: + raise RuntimeError( + "Either args.latency_mode or args.throughput_mode should be set" + ) + + if not args.no_python and not args.program.endswith(".py"): + raise RuntimeError( + 'For non Python script, you should use "--no-python" parameter.' + ) + + # Verify LD_PRELOAD + if "LD_PRELOAD" in os.environ: + lst_valid = [] + tmp_ldpreload = os.environ["LD_PRELOAD"] + for item in tmp_ldpreload.split(":"): + matches = glob.glob(item) + if len(matches) > 0: + lst_valid.append(item) + else: + logger.warning("%s doesn't exist. Removing it from LD_PRELOAD.", item) + if len(lst_valid) > 0: + os.environ["LD_PRELOAD"] = ":".join(lst_valid) + else: + os.environ["LD_PRELOAD"] = "" + + launcher = _Launcher() + launcher.launch(args) + for x in sorted(set(os.environ.keys()) - env_before): + logger.debug("%s=%s", x, os.environ[x]) + + +if __name__ == "__main__": + parser = ArgumentParser( + description="This is a script for launching PyTorch inference on Intel(R) Xeon(R) Scalable " + "Processors with optimal configurations. Single instance inference, " + "multi-instance inference are enable. To get the peak performance on Intel(R) " + "Xeon(R) Scalable Processors, the script optimizes the configuration " + "of thread and memory management. For thread management, the script configures thread " + "affinity and the preload of Intel OMP library. For memory management, it configures " + "NUMA binding and preload optimized memory allocation library (e.g. tcmalloc, jemalloc) " + "\n################################# Basic usage ############################# \n" + "\n 1. single instance\n" + "\n >>> python -m torch.backends.xeon.run_cpu python_script args \n" + "\n2. multi-instance \n" + "\n >>> python -m torch.backends.xeon.run_cpu --ninstances xxx " + "--ncores-per-instance xx python_script args\n" + "\n############################################################################# \n", + formatter_class=RawTextHelpFormatter, + ) + create_args(parser) + args = parser.parse_args() + main(args) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__init__.py b/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..31e69876927d01878a9d1cb836d72fd14adf95e9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__init__.py @@ -0,0 +1,29 @@ +# mypy: allow-untyped-defs +import sys +import types + +import torch + + +class _XNNPACKEnabled: + def __get__(self, obj, objtype): + return torch._C._is_xnnpack_enabled() + + def __set__(self, obj, val): + raise RuntimeError("Assignment not supported") + + +class XNNPACKEngine(types.ModuleType): + def __init__(self, m, name): + super().__init__(name) + self.m = m + + def __getattr__(self, attr): + return self.m.__getattribute__(attr) + + enabled = _XNNPACKEnabled() + + +# This is the sys.modules replacement trick, see +# https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273 +sys.modules[__name__] = XNNPACKEngine(sys.modules[__name__], __name__) diff --git a/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..856fdf9834db6cbc2273f23bef1ce8adbc85cd0a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/backends/xnnpack/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/FileStoreTest b/.venv/lib/python3.12/site-packages/torch/bin/FileStoreTest new file mode 100644 index 0000000000000000000000000000000000000000..8bf742af34e4d8812f0978e0bc475af014b3a4fd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/FileStoreTest @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ae4c542a0d5d348da76f4c30101d7fdcb97751119ca5c69d288e85ed3cb7497 +size 623896 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/HashStoreTest b/.venv/lib/python3.12/site-packages/torch/bin/HashStoreTest new file mode 100644 index 0000000000000000000000000000000000000000..4ff4da135eb75ead30b69987b8742ad58bba94d0 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/HashStoreTest @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a31476781197a2bd434f659316b8ca878324366236a12281a13e808588e3fc62 +size 629792 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/TCPStoreTest b/.venv/lib/python3.12/site-packages/torch/bin/TCPStoreTest new file mode 100644 index 0000000000000000000000000000000000000000..caddf608e7940a11f4505447a61f0698d1cbdbe7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/TCPStoreTest @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1d18dd894ae438f20946dfdb3c10f30652fc8552857194f0a5a30b5b07cd5de +size 687304 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/protoc b/.venv/lib/python3.12/site-packages/torch/bin/protoc new file mode 100644 index 0000000000000000000000000000000000000000..a650ef8b8b5d0798dea9a0be92910416bc90ff4d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/protoc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:866a9e66a42b2764b5f9d60bcb5e9fbd370a9bbc9fc0a1621cd54d245ce6e703 +size 5463672 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/protoc-3.21.12.0 b/.venv/lib/python3.12/site-packages/torch/bin/protoc-3.21.12.0 new file mode 100644 index 0000000000000000000000000000000000000000..a650ef8b8b5d0798dea9a0be92910416bc90ff4d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/protoc-3.21.12.0 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:866a9e66a42b2764b5f9d60bcb5e9fbd370a9bbc9fc0a1621cd54d245ce6e703 +size 5463672 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/script_module_v4.ptl b/.venv/lib/python3.12/site-packages/torch/bin/script_module_v4.ptl new file mode 100644 index 0000000000000000000000000000000000000000..7ce36eed3e4698f187d6c24f57f5c7b48e850b52 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/script_module_v4.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_aoti_abi_check b/.venv/lib/python3.12/site-packages/torch/bin/test_aoti_abi_check new file mode 100644 index 0000000000000000000000000000000000000000..3ea2c7968cdd5bcd97a459f106c7ef6537307d77 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_aoti_abi_check @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8686fe9ef1b6110dd301390433d884ce5886cd07029995943edc0aba5239694 +size 1363824 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_api b/.venv/lib/python3.12/site-packages/torch/bin/test_api new file mode 100644 index 0000000000000000000000000000000000000000..301f4ecd0e6f5a71c34c67ca9b0177876c62d9bc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_api @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41500ab0d205e7019d85cd0d3e780f8175084aedc85237b07c409bb279f0f621 +size 15106616 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_comms_id b/.venv/lib/python3.12/site-packages/torch/bin/test_comms_id new file mode 100644 index 0000000000000000000000000000000000000000..54deff2088d2585eb79e9f3e3b96e9383a4a4897 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_comms_id @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b400139d350d8430b1fe9b259c4e1dc4e07efdde4dc4b083690818597fe6069c +size 677384 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_cpp_rpc b/.venv/lib/python3.12/site-packages/torch/bin/test_cpp_rpc new file mode 100644 index 0000000000000000000000000000000000000000..05534ac63651f8677e0c42cf266911a370a399cf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_cpp_rpc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be6a892790b5eeefcf8d59a64a9bb7ddf3d0d9c9cbc36b478da962f553b15f91 +size 790168 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_dist_autograd b/.venv/lib/python3.12/site-packages/torch/bin/test_dist_autograd new file mode 100644 index 0000000000000000000000000000000000000000..0ce5c722f8183761e76240cd960498adc65f5ba4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_dist_autograd @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6afc31e0f9150ea959676c9e4494c8b0186aacd50f801499e4204295b668116 +size 651136 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_global_state_manager b/.venv/lib/python3.12/site-packages/torch/bin/test_global_state_manager new file mode 100644 index 0000000000000000000000000000000000000000..55cc85147372f4863c413d62ba5703b5a6ff1bc2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_global_state_manager @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c662ccc85d2602968bf421741cdae78ef8a656a7f4df4d867930aec50a1760c +size 631912 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_interpreter_async.pt b/.venv/lib/python3.12/site-packages/torch/bin/test_interpreter_async.pt new file mode 100644 index 0000000000000000000000000000000000000000..65918f5774cd9995407c4cde1891a8c42c080f29 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_interpreter_async.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:549d04b98a683b79f1c5831a5ad922b563311a4216b5a2c28226182790368b21 +size 2840 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_jit b/.venv/lib/python3.12/site-packages/torch/bin/test_jit new file mode 100644 index 0000000000000000000000000000000000000000..59df9fd00c9196d0a67457008c09e848b57888b6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_jit @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4da4c7cbad0d9c87545fa4325a6e72b25d08aa4332f3db2da737dbf06af186dd +size 13435808 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_lazy b/.venv/lib/python3.12/site-packages/torch/bin/test_lazy new file mode 100644 index 0000000000000000000000000000000000000000..6f1924d05e3d518371f313b055500450a2e53aec --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_lazy @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e4b8d47f0d319dba52588062f75e5db78ad4e9e78f01fb3c9db7585b2c4c1c4 +size 4960104 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_privateuse1_profiler b/.venv/lib/python3.12/site-packages/torch/bin/test_privateuse1_profiler new file mode 100644 index 0000000000000000000000000000000000000000..da44da35aa777874f4a78eb47890be822232ab62 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_privateuse1_profiler @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a7ee74bd78ee709c3bfd042578be1d1d969bde406d451cf77499f848af1655d +size 616880 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_collection b/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_collection new file mode 100644 index 0000000000000000000000000000000000000000..e778dc6c18c80d79201ff00a1cd6c946307f47ea --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_collection @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a5030f4b620ebd93cb172dc60f969effb1d7246f4b6fab6be9eca6924582479 +size 1068192 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_global_callback_stale_session b/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_global_callback_stale_session new file mode 100644 index 0000000000000000000000000000000000000000..2a3e3101e659ef61331905ce71e95eb4cdaede2b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_profiler_global_callback_stale_session @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f9bb30dc5439ca24016811d0fdbc17511d2b37cf61bee2e9e337535bcb26c42 +size 600432 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/test_shim b/.venv/lib/python3.12/site-packages/torch/bin/test_shim new file mode 100644 index 0000000000000000000000000000000000000000..a2388978df44eb20711cde1020cf02b77d2053d2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/bin/test_shim @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d99a3f78fb4b253b21efbd6818337a5ef1cac47768e54369f44d4c19b102fbf4 +size 630280 diff --git a/.venv/lib/python3.12/site-packages/torch/bin/torch_shm_manager b/.venv/lib/python3.12/site-packages/torch/bin/torch_shm_manager new file mode 100644 index 0000000000000000000000000000000000000000..24b1bb9a52e46dd819f335ad44e416489d654305 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/torch_shm_manager differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..ddee6be4c35afb9352e3c51df3b023c0b43da789 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..8860b3f1eeb35a14f5ec37669fa503c8d01fe0dd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_float_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..cb36f9aeba8bc66d8dfdd07395ce02e9569a439b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..ce612850374e367e90d48e12b3aa9c187e64d2b6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_float_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..443074fe7130cdb2270c289ff48e138e55061ff0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..d895b209afb3986eb70a2bb8b30a19e627a72481 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_inplace_int_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..ac8b1b918de7c5d30ee03f839d078d080485a8bb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..ba723415ecf23f2651745402b6c8e4e8d4619f1c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_int_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..323aa42dde4ec21f327ab731777e2098685153a6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..61b96da1f6fb0aad7f236c6dfd94b2b071dd289f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_float_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..6d06dea6b5896d84b9d7a86e00067935d87144c6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..3085b13a71be1d0c7875120d52521f13a49c0e0b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_reciprocal_int_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..4fd551d073aebfc52180a7efd63520704e5c05ef Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..2f8cfb95856c0ae9c76895ed2e31cef645af4246 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_scalar_scalar_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..9680713a83e280711e6430876f630d857eae37f3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..20524c64f7c2d8119ed88fba29eb87a6b1cfc69f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_inplace_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..0381636677b52b82d8de1592e04adabd4efc961c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..fc5e8f3bd2a029acf06f69c6b332e6065a5bc4b7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_out_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl new file mode 100644 index 0000000000000000000000000000000000000000..21792d35b8924f52d20bc6f00dc9ff90b5597675 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl differ diff --git a/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl.ff b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl.ff new file mode 100644 index 0000000000000000000000000000000000000000..bdababe90d9161966ac4c9b96492b728f6735fa8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/bin/upgrader_models/test_versioned_div_tensor_v2.ptl.ff differ diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/__init__.py b/.venv/lib/python3.12/site-packages/torch/compiler/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..5d75290acef78076f724739c5c8dacd99cce88c6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/compiler/__init__.py @@ -0,0 +1,1014 @@ +# mypy: allow-untyped-defs +import contextlib +import io +from collections.abc import Callable +from typing import Any, TYPE_CHECKING, TypeVar +from typing_extensions import ParamSpec + +import torch +from torch._higher_order_ops.invoke_subgraph import NestedCompileRegionOptions + +# ``torch.compiler.precompile``: make_fx AOT capture -> self-contained Python source +# plus an acceleration cache. Re-exported from the private impl module, whose +# ``_PrecompileApi.__module__`` is forced to "torch.compiler" so this is the single +# public location. Distinct from ``torch._dynamo.config.caching_precompile`` (a +# ``torch.compile`` guard-serialization caching mode), despite the shared word. +# ``PrecompileError`` is also re-exported here as ``torch.compiler.PrecompileError`` so the +# conventional ``except torch.compiler.PrecompileError`` works; its ``__module__`` is already +# forced to "torch.compiler" in the impl module, matching this public location. +from torch._precompile import ( + precompile as precompile, + PrecompileError as PrecompileError, +) + +from . import config +from ._cache import CacheInfo + + +if TYPE_CHECKING: + from torch._dynamo.eval_frame import StanceStr + + +__all__ = [ + "compile", + "config", + "assume_constant_result", + "reset", + "nonstrict_trace", + "allow_in_graph", + "substitute_in_graph", + "list_backends", + "disable", + "set_default_backend", + "get_default_backend", + "set_stance", + "set_enable_guard_collectives", + "cudagraph_mark_step_begin", + "cudagraph_mark_warmup_incomplete", + "load_compiled_function", + "precompile", + "PrecompileError", + "wrap_numpy", + "is_compiling", + "is_dynamo_compiling", + "is_exporting", + "save_cache_artifacts", + "load_cache_artifacts", + "keep_portable_guards_unsafe", + "skip_guard_on_inbuilt_nn_modules_unsafe", + "skip_guard_on_all_nn_modules_unsafe", + "keep_tensor_guards_unsafe", + "skip_guard_on_globals_unsafe", + "skip_all_guards_unsafe", + "nested_compile_region", +] + + +_P = ParamSpec("_P") +_R = TypeVar("_R") +FuncType = Callable[..., Any] +F = TypeVar("F", bound=FuncType) + + +def compile(*args, **kwargs): + """ + See :func:`torch.compile` for details on the arguments for this function. + """ + # pyrefly: ignore [not-iterable] + return torch.compile(*args, **kwargs) + + +def reset() -> None: + """ + Reset the in-process compiler state. + + This function clears Dynamo's in-memory compilation caches and related + process-local state used by :func:`torch.compile`. It does not delete + filesystem caches, such as Inductor's disk cache. + """ + import torch._dynamo + + torch._dynamo.reset() + + +def allow_in_graph(fn): + """ + Tells the compiler frontend (Dynamo) to skip symbolic introspection of the function + and instead directly write it to the graph when encountered. + + If you are using :func:`torch.compile` (with backend="inductor" (the default)), or + :func:`torch.export.export`, and trying to black-box a Python function throughout + all tracing, do not use this API. + Instead, please create a custom operator (see `PyTorch Custom Operators Landing Page + `_) + + .. warning:: + + If you're a typical torch.compile user (e.g. you're applying torch.compile to + a model to make it run faster), you probably don't want to use this function. + :func:`allow_in_graph` is a footgun because it skips the compiler frontend + (Dynamo) that is responsible for doing safety checks (graph breaks, handling + closures, etc). Incorrect usage will lead to difficult-to-debug silent + incorrectness issues. + + Given a Python function with no allow_in_graph decorator, regular execution + of torch.compile traces through the function. :func:`allow_in_graph` changes + it so that the frontend does not trace inside the function, but the compiler + backend still traces through it. Compare this to custom operators, which + treats a function as a black box throughout the torch.compile stack. The following + table compares these mechanisms. + + +------------------------+-----------------------+--------------------------------+ + | Mechanism | Frontend (Dynamo) | Backend (AOTAutograd+Inductor) | + +========================+=======================+================================+ + | no decorator | trace inside | trace inside | + +------------------------+-----------------------+--------------------------------+ + | allow_in_graph | opaque callable | trace inside | + +------------------------+-----------------------+--------------------------------+ + | custom op | opaque callable | opaque callable | + +------------------------+-----------------------+--------------------------------+ + + One common use case for :func:`allow_in_graph()` is as an escape hatch for the compiler + frontend: if you know the function works w.r.t. to the downstream components of the + compilation stack (AOTAutograd and Inductor) but there is a Dynamo bug that prevents it from + symbolically introspecting the function properly (or if your code is in C/C++ and + therefore cannot be introspected with Dynamo), then one can decorate said function + with :func:`allow_in_graph` to bypass Dynamo. + + We require that ``fn`` adhere to the following restrictions. Failure to adhere + results in undefined behavior: + + - The inputs to ``fn`` must be Proxy-able types in the FX graph. Valid types include: + Tensor/int/bool/float/None/List[Tensor?]/List[int?]/List[float?] + Tuple[Tensor?, ...]/Tuple[int?, ...]/Tuple[float?, ...]/torch.dtype/torch.device + - The outputs to ``fn`` must be Proxy-able types in the FX graph (see previous bullet) + - all Tensors used inside of ``fn`` must be passed directly as inputs to ``fn`` + (as opposed to being captured variables). + + See also :func:`nonstrict_trace()`, which has slightly fewer restrictions on the inputs. + + Args: + fn: A callable representing the function to be included in the graph. + If ``fn`` is a list or tuple of callables it recursively applies + :func:`allow_in_graph()` to each function and returns a new list or + tuple containing the modified functions. + + Example:: + + torch.compiler.allow_in_graph(my_custom_function) + + + @torch.compile(...) + def fn(x): + x = torch.add(x, 1) + x = my_custom_function(x) + x = torch.add(x, 1) + return x + + + fn(...) + + Will capture a single graph containing ``my_custom_function()``. + + """ + import torch._dynamo + + return torch._dynamo.allow_in_graph(fn) + + +def nonstrict_trace(traceable_fn: Callable[_P, _R]) -> Callable[_P, _R]: + """ + Decorator to mark a function as nonstrict-traceable for dynamo. + + A nonstrict-traced function appears as an opaque call in the dynamo graph. + Dynamo does not trace into the function body (hence the "nonstrict"), but + aot_autograd will trace into it. + + This is similar to ``allow_in_graph`` but with enhanced support for: + - User-defined classes as inputs (must be registered with pytree) + - ``nn.Module`` as input arguments (parameters and buffers are tracked for autograd) + - Global/captured tensors treated as constants (assumed not updated during execution) + + Note: + - With ``backend="eager"``, the original Python function runs directly. + With ``backend="aot_eager"``, the graph traced by aot_autograd runs. + With ``backend="inductor"``, the traced graph is compiled with inductor. + + - Training is supported: you can call ``.backward()`` on outputs and gradients + will flow through the nonstrict-traced function. + + Dangerous patterns (may cause silent incorrectness): + - Side effects between nonstric_trace'd fn and compiled region: The function should + not depend on variables mutated by other code inside the compiled function, and code + after the call should not depend on mutations made by it. + + - Implicit inputs (closures/globals): Tensors captured from enclosing scopes + are treated as constants. Gradients will NOT flow back to them. Pass tensors + as explicit arguments if gradients are needed. + + Restrictions: + - Both inputs and outputs must use pytree-compatible types. User-defined classes + must be registered via :func:`torch.utils._pytree.register_pytree_node`, + :func:`torch.utils._pytree.register_dataclass`, or + :func:`torch.utils._pytree.register_constant`. Tensors, Python primitives (int, float, bool, str), + symbolic types (SymInt, SymFloat, SymBool), and built-in containers (list, + tuple, dict) are already handled by default. + - Primitive values and container structure are specialized per call site: + each call site expects the same primitives and structure on every execution. + + Example:: + + >>> import torch + >>> @torch.compiler.nonstrict_trace + ... def traced_forward(model, x): + ... # It's OK to have dynamo graph break within nonstrict_trace region + ... torch._dynamo.graph_break() + ... return model(x) + x + ... + >>> class MyModule(torch.nn.Module): + ... def __init__(self): + ... super().__init__() + ... self.inner = torch.nn.Linear(10, 10) + ... + ... def forward(self, x): + ... return traced_forward(self.inner, x) + ... + >>> # Compile and run + >>> model = MyModule() + >>> opt_model = torch.compile(model, backend="aot_eager", fullgraph=True) + >>> out = opt_model(torch.randn(10, 10)) + >>> out.sum().backward() # Gradients flow through traced_forward + + """ + import torch._dynamo + + return torch._dynamo.nonstrict_trace(traceable_fn) + + +def substitute_in_graph( + original_fn: Callable[_P, _R], + *, + can_constant_fold_through: bool = False, + skip_signature_check: bool = False, +) -> Callable[[Callable[_P, _R]], Callable[_P, _R]]: + """ + Register a polyfill handler for a function, usually a C function from the C extension, to be + used in place of the original function when inlining the original function in the graph. + + .. note:: + + The polyfill handler is only used when inlining the original function. It is not used when + the original function is called directly. In the eager mode, the decorated function calls + the performant C function rather than the polyfill handler. + + The polyfill handler is a function that will be called in place of the original function when + inlining the original function. The polyfill handler should have the same signature and the same + behavior as the original function. + + Args: + original_fn (callable): The original function, usually a C function, to register a polyfill + handler for. + can_constant_fold_through (bool, optional): Whether the polyfill handler can be constant + folded through. That is, if the polyfill handler is a pure function and its arguments + are constant, the result of the polyfill handler can be constant folded during the + compilation. Defaults to ``False``. + skip_signature_check (bool, optional): Whether to skip the signature check between the + original function and the polyfill handler. Defaults to ``False``. + + Returns: + A decorator that registers the polyfill handler for the original function. + + Example:: + + >>> import operator + >>> operator.indexOf([1, 2, 3, 4, 5], 3) + 2 + >>> torch.compile(operator.indexOf, fullgraph=True)([1, 2, 3, 4, 5], 3) + ... # xdoctest: +SKIP("Long tracebacks") + Traceback (most recent call last): + ... + torch._dynamo.exc.Unsupported: ... + + >>> @torch.compiler.substitute_in_graph(operator.indexOf) + ... def indexOf(a, b, /): + ... for i, item in enumerate(a): + ... if item is b or item == b: + ... return i + ... raise ValueError("sequence.index(x): x not in sequence") + >>> + >>> torch.compile(operator.indexOf, fullgraph=True)([1, 2, 3, 4, 5], 3) + 2 + """ + import torch._dynamo + + return torch._dynamo.substitute_in_graph( + original_fn, + can_constant_fold_through=can_constant_fold_through, + skip_signature_check=skip_signature_check, + ) + + +def list_backends(exclude_tags=("debug", "experimental")) -> list[str]: + """ + Return valid strings that can be passed to `torch.compile(..., backend="name")`. + + Args: + exclude_tags(optional): A tuple of strings representing tags to exclude. + """ + import torch._dynamo + + return torch._dynamo.list_backends(exclude_tags) + + +def assume_constant_result(fn): + """ + This function is used to mark a function `fn` as having a constant result. + This allows the compiler to optimize away your function. + Returns The same function `fn` + + Args: + fn: The function to be marked as having a constant result. + + .. warning:: + `assume_constant_result` can, if invalid, cause safety and soundness issues, :func:`torch.compile` + will not attempt to validate whether the constant assumption is true or not + + """ + import torch._dynamo + + return torch._dynamo.assume_constant_result(fn) + + +def disable(fn=None, recursive=True, *, reason=None): + """ + This function provides a decorator to disable compilation on a function. + It also provides the option of recursively disabling called functions. + + Args: + fn (optional): The function to disable + recursive (optional): A boolean value indicating whether the disabling should be recursive. + reason (optional): A string value indicating the reason for disabling the function. + """ + import torch._dynamo + + return torch._dynamo.disable(fn, recursive, reason=reason) + + +def set_default_backend(backend: str | Callable[..., Any] | None) -> None: + """Set the default backend for ``torch.compile`` when no ``backend`` argument is specified. + + Passing ``None`` resets the default back to ``"inductor"``. + + Args: + backend: A backend name (string), a callable backend, or ``None``. + + Example:: + + >>> torch.compiler.set_default_backend("eager") + >>> torch.compiler.get_default_backend() + 'eager' + >>> torch.compiler.set_default_backend(None) # reset + >>> torch.compiler.get_default_backend() + 'inductor' + """ + from torch._dynamo.backends.registry import set_default_backend + + set_default_backend(backend) + + +def get_default_backend() -> str | Callable[..., Any]: + """Return the current default backend for ``torch.compile``. + + Returns: + The current default backend (string or callable). Initially ``"inductor"``. + """ + from torch._dynamo.backends.registry import get_default_backend + + return get_default_backend() + + +def set_stance( + stance: "StanceStr" = "default", + *, + skip_guard_eval_unsafe: bool = False, + force_backend: str | Callable[..., Any] | None = None, +): + """ + Set the current stance of the compiler. + Can be used as a function, context manager, or decorator. + Do not use this function inside a `torch.compile` region - an error will be raised otherwise. + + .. code-block:: python + + @torch.compile + def foo(x): ... + + + @torch.compiler.set_stance("force_eager") + def bar(): + # will not be compiled + foo(...) + + + bar() + + with torch.compiler.set_stance("force_eager"): + # will also not be compiled + foo(...) + + torch.compiler.set_stance("force_eager") + # will also not be compiled + foo(...) + torch.compiler.set_stance("default") + + # will be compiled + foo(...) + + Args: + stance: The stance to set the compiler to. Valid values are: + + - "default": The default stance, used for normal compilation. + - "force_eager": Ignore all `torch.compile` directives. + - "eager_on_recompile": Run code eagerly when a recompile is necessary. + If there is cached compiled code valid for the input, it will still be used. + - "fail_on_recompile": Raise an error when recompiling a function. + - "eager_then_compile": Run the first invocation in eager mode, then compile on + subsequent calls. This is beneficial for dynamic shapes as it allows inferring + dynamism from the first two invocations instead of wasting a static compile on + the first invocation. + - "aot_eager_then_compile": Run the first invocation with AOT eager to get memory + benefits from activation checkpointing, then compile on subsequent calls. Like + eager_then_compile, this improves handling of dynamic shapes by avoiding an + initial static compile. + + + skip_guard_eval_unsafe: A flag to run only differentiating guards. + CAUTION - This flag is unsafe and should only be used if your setup + meets the following conditions. + + torch.compile uses a guard system to support recompilations and + choose which compiled artifact to run at runtime. These guards, + though efficient, add some overhead, which may impact performance in + scenarios where you need to optimize for minimal guard processing + time. This API enables you to disable guard evaluation, assuming + that you have warmed up the compiled model with a sufficient variety + of inputs. This assumption means that, after the warmup phase, no + further recompilations will be necessary. If this assumption fails, + there is a risk of silently producing incorrect results (hence the + term "unsafe" in the API name). + + force_backend: If `stance` is "default", this argument can be used to force `torch.compile` + to use a specific backend. Otherwise, an error is raised. + """ + import torch._dynamo + + return torch._dynamo.set_stance( + stance, + skip_guard_eval_unsafe=skip_guard_eval_unsafe, + force_backend=force_backend, + ) + + +# forbid in graph +set_stance._dynamo_forbidden = True # type: ignore[attr-defined] + + +def set_enable_guard_collectives(enabled: bool): + """ + Enables use of collectives *during* guard evaluation to synchronize behavior + across ranks. This is expensive: we have to issue a collective every time + we enter a compiled code region, even if no rank actually would need to + compile. This can help prevent NCCL hangs by ensuring that we never have a + situation where one rank starts recompiling while other ranks don't compile; + it is especially useful in conjunction with enable_compiler_collectives + where such a situation would immediately cause a hang (as it is necessary + for all ranks to compile at the same time to run compiler collectives). Like + compiler collectives, you can only run this on SPMD programs; you will hang + otherwise. Note that a guard collective is only issued if there is any + compiled code to guard on; if this is the first time we encounter a frame or + the frame is skipped, we don't issue collectives. + + Returns the previous setting of enabled. + """ + from torch._C._dynamo.eval_frame import set_guard_complete_hook + from torch._dynamo.eval_frame import guard_collectives_hook + + if enabled: + return set_guard_complete_hook(guard_collectives_hook) is not None # type: ignore[arg-type] + else: + return set_guard_complete_hook(None) is not None + + +set_enable_guard_collectives._dynamo_forbidden = True # type: ignore[attr-defined] + + +def cudagraph_mark_step_begin(): + """ + Indicates that a new iteration of inference or training is about to begin. + + CUDA Graphs will free tensors of a prior iteration. A new iteration is started on each invocation of + torch.compile, so long as there is not a pending backward that has not been called. + + If that heuristic is wrong, such as in the following example, manually mark it with this api. + + .. code-block:: python + + @torch.compile(mode="reduce-overhead") + def rand_foo(): + return torch.rand([4], device="cuda") + + + for _ in range(5): + torch.compiler.cudagraph_mark_step_begin() + rand_foo() + rand_foo() + + For more details, see `torch.compiler_cudagraph_trees `__ # noqa: B950 + """ + from torch._inductor import cudagraph_trees + + cudagraph_trees.mark_step_begin() + + +def cudagraph_mark_warmup_incomplete(): + """Request another warmup for the active CUDA Graph Trees function. + + Call this synchronously from an autotuner or other code running during CUDA + Graph Trees warmup when the current function needs another warmup iteration. + The function will run eagerly again on its next invocation instead of being + recorded. This is a no-op outside CUDA Graph Trees warmup, including during + recording and replay or when CUDA Graph Trees are disabled. + """ + from torch._inductor import cudagraph_trees + + cudagraph_trees.mark_warmup_incomplete() + + +def wrap_numpy(fn): + r"""Decorator that turns a function from ``np.ndarray``\ s to ``np.ndarray``\ s into a function + from ``torch.Tensor``\ s to ``torch.Tensor``\ s. + + It is designed to be used with :func:`torch.compile` with ``fullgraph=True``. It allows you to + compile a NumPy function as if it were a PyTorch function. This allows you to run NumPy code + on CUDA or compute its gradients. + + .. note:: + + This decorator does not work without :func:`torch.compile`. + + Example:: + + >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA) + >>> # Compile a NumPy function as a Tensor -> Tensor function + >>> @torch.compile(fullgraph=True) + >>> @torch.compiler.wrap_numpy + >>> def fn(a: np.ndarray): + >>> return np.sum(a * a) + >>> # Execute the NumPy function using Tensors on CUDA and compute the gradients + >>> x = torch.arange(6, dtype=torch.float32, device="cuda", requires_grad=True) + >>> out = fn(x) + >>> out.backward() + >>> print(x.grad) + tensor([ 0., 2., 4., 6., 8., 10.], device='cuda:0') + """ + from torch._dynamo.external_utils import wrap_numpy as wrap + + return wrap(fn) + + +_is_compiling_flag: bool = False +_is_exporting_flag: bool = False +_is_non_strict_tracing_flag: bool = False + + +def is_compiling() -> bool: + """ + Indicates whether a graph is executed/traced as part of torch.compile() or torch.export(). + + Note that there are 2 other related flags that should deprecated eventually: + * torch._dynamo.external_utils.is_compiling() + * torch._utils.is_compiling() + + Example:: + + >>> def forward(self, x): + >>> if not torch.compiler.is_compiling(): + >>> pass # ...logic that is not needed in a compiled/traced graph... + >>> + >>> # ...rest of the function... + """ + if torch.jit.is_scripting(): + return False + else: + return _is_compiling_flag + + +def _is_non_strict_tracing() -> bool: + """ + Indicates whether we are inside a non-strict make_fx-based tracing session. + """ + return _is_non_strict_tracing_flag + + +@contextlib.contextmanager +def _non_strict_tracing_context(): + """Context manager that sets the non-strict tracing flag.""" + global _is_non_strict_tracing_flag + old = _is_non_strict_tracing_flag + try: + _is_non_strict_tracing_flag = True + yield + finally: + _is_non_strict_tracing_flag = old + + +@contextlib.contextmanager +def _compile_session_context(): + """Context manager that sets _is_compiling_flag for the duration of a + torch.compile session. + """ + global _is_compiling_flag + old = _is_compiling_flag + try: + _is_compiling_flag = True + yield + finally: + _is_compiling_flag = old + + +@contextlib.contextmanager +def _patch_autograd_grad(): + """Patch autograd.grad for non-strict make_fx tracing. + + This patch installs autograd hooks so traced backward nodes preserve + stack trace, seq_nr, and autograd_backward metadata before delegating to + the real torch.autograd.grad. + """ + import functools + + import torch.autograd + from torch._dynamo.utils import warn_once + from torch._functorch._aot_autograd.logging_utils import ( + setup_stacktrace_preservation_hooks_from_tensors, + ) + + warn_once( + "torch.compiler._patch_autograd_grad() is deprecated; " + "use torch.compiler._patch_engine_backward() instead." + ) + # TODO: Remove this helper once Titan no longer depends on it. + + _orig_grad = torch.autograd.grad + + @functools.wraps(_orig_grad) + def _patched_grad(outputs, inputs, *args, **kwargs): + if not _is_non_strict_tracing(): + raise AssertionError( + "_patch_autograd_grad() must be used under " + "_non_strict_tracing_context()" + ) + + setup_stacktrace_preservation_hooks_from_tensors(outputs) + return _orig_grad(outputs, inputs, *args, **kwargs) + + torch.autograd.grad = _patched_grad + try: + yield + finally: + torch.autograd.grad = _orig_grad + + +@contextlib.contextmanager +def _patch_engine_backward(): + """Patch _engine_run_backward for non-strict make_fx tracing. + + This patch installs autograd hooks so traced backward nodes preserve + stack trace, seq_nr, and autograd_backward metadata before delegating to + the real autograd engine entrypoint used by backward(). + """ + import functools + + import torch.autograd + import torch.autograd.graph + from torch._functorch._aot_autograd.logging_utils import ( + setup_stacktrace_preservation_hooks_from_tensors, + ) + + _orig_engine_run_backward = torch.autograd.graph._engine_run_backward + + @functools.wraps(_orig_engine_run_backward) + def _patched_engine_backward(outputs, *args, **kwargs): + if not _is_non_strict_tracing(): + raise AssertionError( + "_patch_engine_backward() must be used under " + "_non_strict_tracing_context()" + ) + + setup_stacktrace_preservation_hooks_from_tensors(outputs) + return _orig_engine_run_backward(outputs, *args, **kwargs) + + torch.autograd.graph._engine_run_backward = _patched_engine_backward + torch.autograd._engine_run_backward = _patched_engine_backward + try: + yield + finally: + torch.autograd.graph._engine_run_backward = _orig_engine_run_backward + torch.autograd._engine_run_backward = _orig_engine_run_backward + + +def is_dynamo_compiling() -> bool: + """ + Indicates whether a graph is traced via TorchDynamo. + + It's stricter than is_compiling() flag, as it would only be set to True when + TorchDynamo is used. + + Example:: + + >>> def forward(self, x): + >>> if not torch.compiler.is_dynamo_compiling(): + >>> pass # ...logic that is not needed in a TorchDynamo-traced graph... + >>> + >>> # ...rest of the function... + """ + return False + + +def is_exporting() -> bool: + """ + Indicates whether we're under exporting. + + It's stricter than is_compiling() flag, as it would only be set to True when + torch.export is used. + + Example:: + + >>> def forward(self, x): + >>> if not torch.compiler.is_exporting(): + >>> pass # ...logic that is not needed in export... + >>> + >>> # ...rest of the function... + """ + return _is_exporting_flag + + +def save_cache_artifacts() -> tuple[bytes, CacheInfo] | None: + """ + Serializes all the cache artifacts that were created during the compilation + + Example: + + - Execute torch.compile + - Call torch.compiler.save_cache_artifacts() + """ + from ._cache import CacheArtifactManager + + if torch._dynamo.config.caching_precompile: + from torch._dynamo.precompile_context import PrecompileContext + + PrecompileContext.save_to_dynamo_cache() + + return CacheArtifactManager.serialize() + + +def load_cache_artifacts(serialized_artifacts: bytes) -> CacheInfo | None: + """ + Hot loads cache artifacts that were previously serialized via + save_cache_artifacts + + Example: + + # From a previous invocation + artifacts = torch.compiler.save_cache_artifacts() + + torch.compiler.load_cache_artifacts(artifacts[0]) + """ + from ._cache import CacheArtifactManager, CacheInfo + + artifacts = CacheArtifactManager.deserialize(serialized_artifacts) + if artifacts is not None: + return CacheArtifactManager.populate_caches(artifacts) + return None + + +def keep_portable_guards_unsafe(guard_entries): + """ + A common function to only keep guards that can be used in both Python and non-Python environments. + This includes: + - Tensor metadata and dynamic shape information. + - Global contexts state (e.g. autocast, no_grad, etc.) + + This is unsafe to use by default. + To use this API, use guard_filter_fn argument while calling torch.compile + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.keep_global_context_and_tensor_guards_unsafe}, + >> ) + """ + return [ + ( + g.guard_type in ("GLOBAL_STATE", "SHAPE_ENV") + or (g.guard_type == "TENSOR_MATCH" and not g.is_global) + ) + for g in guard_entries + ] + + +def skip_guard_on_inbuilt_nn_modules_unsafe(guard_entries): + """ + A common function to skip guards on the inbuilt nn modules like + torch.nn.Linear. This is unsafe to use by default. But for majority of + torch.compile users, the model code does not modify the inbuilt nn module + attributes. They can benefit from reduction in guard latency overhead using + this API. + + To use this API, use guard_filter_fn argument while calling torch.compile + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.skip_guard_on_all_nn_modules_unsafe}, + >> ) + """ + return [ + not entry.orig_guard.source.is_unspecialized_builtin_nn_module() + for entry in guard_entries + ] + + +def skip_guard_on_all_nn_modules_unsafe(guard_entries): + """ + A common function to skip guards on all nn modules, both user defined as + well inbuilt nn modules (like torch.nn.Linear). This is unsafe to use by + default. But for majority of torch.compile users, the model code does not + modify the nn module attributes. They can benefit from reduction in guard + latency overhead using this API. + + To use this API, use guard_filter_fn argument while calling torch.compile + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.skip_guard_on_all_nn_modules_unsafe}, + >> ) + """ + + return [ + not entry.orig_guard.source.is_unspecialized_nn_module() + for entry in guard_entries + ] + + +def keep_tensor_guards_unsafe(guard_entries, keep_parameters=False): + """ + A common function to keep tensor guards on all tensors. This is unsafe to + use by default. But if you don't expect any changes in the model code, you + can just keep the tensor guards. + + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.keep_tensor_guards}, + >> ) + """ + + keep_flags = [] + for entry in guard_entries: + if entry.guard_type == "TENSOR_MATCH": + if not isinstance(entry.value, torch.nn.Parameter): + keep_flags.append(True) + elif keep_parameters: + keep_flags.append(True) + else: + keep_flags.append(False) + else: + keep_flags.append(False) + return keep_flags + + +def skip_guard_on_globals_unsafe(guard_entries): + """ + A common function to skip guards on all globals. This is unsafe to use by + default. But if you don't expect any changes in the globals, you can just + keep the tensor guards. + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.skip_guard_on_globals}, + >> ) + """ + + return [not entry.is_global for entry in guard_entries] + + +def skip_all_guards_unsafe(guard_entries): + """ + A function for skipping all guards on a compiled function. + + WARNING: This function will drop all the safety guarantees from Dynamo + compiled function. Use this with caution. + + To use this API, use guard_filter_fn argument while calling torch.compile + + >> opt_mod = torch.compile( + >> mod, + >> options={"guard_filter_fn": torch.compiler.skip_all_guards_unsafe}, + >> ) + """ + return [False for entry in guard_entries] + + +def nested_compile_region( + fn=None, + *, + options: NestedCompileRegionOptions | None = None, + max_reuse_entries: int = 8, + reuse_hash_fn=None, +): + """ + Tells **``torch.compile``** that the marked set of operations forms a nested + compile region (which is often repeated in the full model) whose code can be + compiled once and safely reused. ``nested_compile_region`` can also be used + as a decorator. + + During **``torch.compile``** tracing, the compiler applies *hierarchical + compilation* with ``nested_compile_region``: it emits optimized code for the + marked region the first time it is encountered and re-emits (or "stamps + out") the previously compiled code on every subsequent invocation. This can + substantially reduce overall compile time for deeply-stacked, + structurally-identical components such as the transformer layers of a + large-language-model (LLM). + + Outside a ``torch.compile`` context—i.e., in standard eager execution—the + call is a no-op, so existing workflows remain unaffected. + + Note that ``nested_compile_region`` **does not** promise that a region will + be compiled exactly once. If the compiler detects that new input conditions + (shape, dtype, device, stride, globals etc.) make the cached version invalid + to reuse, it will transparently re-compile the region. Using it is + therefore *safe*: correctness is always preserved, and you pay the extra + compilation cost only when required. + + Args: + fn: The function to wrap + options: Optional backend to use for compiling the subgraph. + Warning: this is an experimental feature under development and + not ready for use yet. + max_reuse_entries: Maximum number of reuse cache entries per function + before raising an error. If this limit is hit, guards keep failing + across invocations and hierarchical compilation is not effective. + reuse_hash_fn: Optional callable that takes the same ``*args, **kwargs`` + as the wrapped function and returns an integer hash key. When + provided, Dynamo traces this function to obtain a constant integer + and uses it as the cache key for subgraph reuse, bypassing the + automatic fingerprint/guard machinery. Two calls that produce the + same hash key reuse the same cached subgraph. The hash function + must be fully traceable (no graph breaks) and must return a + constant integer. + """ + + if options is not None: + from torch._dynamo import config as dynamo_config + + if not dynamo_config.enable_invoke_subgraph_regional_compile: + raise RuntimeError( + "nested_compile_region config is an experimental feature for testing only." + ) + + from torch._higher_order_ops.invoke_subgraph import ( + mark_compile_region as _mark_compile_region, + ) + + return _mark_compile_region( + fn, + options=options, + max_reuse_entries=max_reuse_entries, + reuse_hash_fn=reuse_hash_fn, + ) + + +def load_compiled_function( + file: io.IOBase, + *, + f_globals: dict[str, object] | None = None, + external_data: dict[str, Any] | None = None, +) -> Callable[..., Any]: + """ + Load an aot-compiled function from a file. + + .. warning:: + + This API is currently experimental and subject to change. + + Args: + file: A file-like object containing the serialized compiled function. + f_globals: Optional global scope enclosing the compiled function. + external_data: Optional data to be loaded into the runtime environment + of the compiled function. This should contain the same + data as AOTCompileResult.external_data returned from save_compiled_function() call. + + Returns: + A torch-compiled function with compilation preloaded from disk. + """ + from torch._dynamo.aot_compile import AOTCompiledFunction + + data = file.read() + return AOTCompiledFunction.deserialize(data, f_globals, external_data) diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2c90808f752cec2c75e480f3275693c59163aedc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/_cache.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/_cache.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..12f987168e89490b5318dcb17cac2032e61b3f5d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/_cache.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/config.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/config.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..68824a5438796246be776cc5a6f8bc816a7ad221 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/compiler/__pycache__/config.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/_cache.py b/.venv/lib/python3.12/site-packages/torch/compiler/_cache.py new file mode 100644 index 0000000000000000000000000000000000000000..9f993e43f7ac53077e689afd2b7c9a98112fa3bb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/compiler/_cache.py @@ -0,0 +1,342 @@ +import copy +import dataclasses +import logging +from abc import ABC, abstractmethod +from collections import defaultdict +from collections.abc import Generator +from contextlib import contextmanager +from itertools import chain +from typing import Any + +from torch.utils._appending_byte_serializer import ( + AppendingByteSerializer, + BytesReader, + BytesWriter, +) +from torch.utils._ordered_set import OrderedSet + + +log = logging.getLogger(__name__) + + +@dataclasses.dataclass(frozen=True) +class CacheArtifact(ABC): + """ + Data for each cache artifact that will be serialized and deserialized + """ + + key: str + content: bytes = dataclasses.field(repr=False) # Do not display potential binary + + @staticmethod + def serialize(writer: BytesWriter, cls: "CacheArtifact") -> None: + writer.write_str(cls.key) + writer.write_bytes(cls.content) + + @staticmethod + def deserialize(artifact_type: str, reader: BytesReader) -> "CacheArtifact": + key = reader.read_str() + content = reader.read_bytes() + return CacheArtifactFactory.create(artifact_type, key, content) + + @staticmethod + def encode(content: Any) -> bytes: + if not isinstance(content, bytes): + raise AssertionError(f"Expected bytes, got {type(content)}") + return content + + @abstractmethod + def populate_cache(self) -> None: + pass + + @staticmethod + def type() -> str: + """ + Returns the type of the artifact. Must be unique across all CacheArtifact classes. + + CacheArtifactFactory.register will add property method to CacheInfo based on this (def {type}_artifacts) + that returns all artifacts for specific cache. + """ + raise RuntimeError("CacheArtifact is an abstract class, please use a subclass") + + +class CacheArtifactFactory: + """ + Factory for creating CacheArtifact objects based on their type + """ + + _artifact_types: dict[str, type[CacheArtifact]] = {} + + @classmethod + def register(cls, artifact_cls: type[CacheArtifact]) -> type[CacheArtifact]: + artifact_type_key = artifact_cls.type() + if artifact_cls.type() in cls._artifact_types: + raise AssertionError( + f"Artifact of type={artifact_type_key} already registered in mega-cache artifact factory" + ) + cls._artifact_types[artifact_type_key] = artifact_cls + setattr( + CacheInfo, + f"{artifact_type_key}_artifacts", + property(lambda self: self.artifacts[artifact_type_key]), + ) + return artifact_cls + + @classmethod + def _get_artifact_type(cls, artifact_type_key: str) -> type[CacheArtifact]: + if artifact_type_key not in cls._artifact_types: + raise AssertionError( + f"Artifact of type={artifact_type_key} not registered in mega-cache artifact factory" + ) + return cls._artifact_types[artifact_type_key] + + @classmethod + def create(cls, artifact_type_key: str, key: str, content: bytes) -> CacheArtifact: + artifact_cls = cls._get_artifact_type(artifact_type_key) + # pyrefly: ignore [bad-instantiation] + return artifact_cls(key, content) + + @classmethod + def encode_create( + cls, artifact_type_key: str, key: str, content: Any + ) -> CacheArtifact: + artifact_cls = cls._get_artifact_type(artifact_type_key) + # pyrefly: ignore [bad-instantiation] + return artifact_cls(key, artifact_cls.encode(content)) + + +@dataclasses.dataclass +class CacheInfo: + """ + Return value of serialization and deserialization for the purpose of + instrumentation + """ + + artifacts: defaultdict[str, list[str]] = dataclasses.field( + default_factory=lambda: defaultdict(list) + ) + + # Methods set by CacheArtifactFactory.register based on CacheArtifact.type() + @property + def inductor_artifacts(self) -> list[str]: # type: ignore[empty-body] + ... + + @property + def autotune_artifacts(self) -> list[str]: # type: ignore[empty-body] + ... + + @property + def aot_autograd_artifacts(self) -> list[str]: # type: ignore[empty-body] + ... + + @property + def pgo_artifacts(self) -> list[str]: # type: ignore[empty-body] + ... + + @property + def precompile_artifacts(self) -> list[str]: # type: ignore[empty-body] + ... + + def add(self, artifact: CacheArtifact) -> None: + self.artifacts[artifact.type()].append(artifact.key) + + def clear(self) -> None: + self.artifacts.clear() + + def empty(self) -> bool: + return not self.artifacts + + +def _serialize_single_cache( + writer: BytesWriter, cls: "tuple[str, list[CacheArtifact]]" +) -> None: + writer.write_str(cls[0]) + writer.write_uint64(len(cls[1])) + for artifact in cls[1]: + CacheArtifact.serialize(writer, artifact) + + +def _deserialize_single_cache( + reader: BytesReader, +) -> "tuple[str, list[CacheArtifact]]": + artifacts = [] + artifact_type_key = reader.read_str() + num_artifacts = reader.read_uint64() + for _ in range(num_artifacts): + artifacts.append(CacheArtifact.deserialize(artifact_type_key, reader)) + + return artifact_type_key, artifacts + + +CacheArtifactsResult = dict[str, list[CacheArtifact]] + + +@dataclasses.dataclass(frozen=True) +class CacheArtifactRecorder: + """ + Shared entry point for cache implementations to record artifacts. + """ + + artifact_type: str + key: str + + def record(self, content: Any) -> None: + CacheArtifactManager.record_artifact(self.artifact_type, self.key, content) + + def record_if_present(self, content: Any | None) -> None: + if content is not None: + self.record(content) + + +class CacheArtifactManager: + """ + Lightweight manager class for collecting and processing cache artifacts for + hot loading + + Intended Lifecycle: + - Execute code via torch.compile, this will call + CacheArtifactRecorder.record on each cache artifact + - Call CacheArtifactManager.serialize to convert all the cache artifacts + to portable format + - Call CacheArtifactManager.deserialize to hot load the cache artifacts on + a potentially different process + + NOTE: There's no FB/FC guarantees, results of cache artifacts will not be + used unless code version matches. + """ + + # Protected by the compile_lock + _new_cache_artifacts: CacheArtifactsResult = defaultdict(list) + # Keep a separate seen artifacts list to avoid unnecessary duplicates + # This list will not be cleared between serialize() calls + _seen_artifacts: OrderedSet[CacheArtifact] = OrderedSet() + # When serialize() is called, artifacts are transferred from _cache_artifacts to + # internal data structure of the _serializer + # This allows us to only pay the cost of serialization if serialize() is called + _serializer: AppendingByteSerializer[tuple[str, list[CacheArtifact]]] = ( + AppendingByteSerializer(serialize_fn=_serialize_single_cache) + ) + _cache_info: CacheInfo = CacheInfo() + + @classmethod + def clear(cls) -> None: + cls._new_cache_artifacts.clear() + cls._seen_artifacts.clear() + cls._serializer.clear() + cls._cache_info.clear() + + @classmethod + @contextmanager + def with_fresh_cache(cls) -> Generator[None, None, None]: + original_new_cache_artifacts = cls._new_cache_artifacts + original_seen_artifacts = cls._seen_artifacts + original_serializer = cls._serializer + original_cache_info = cls._cache_info + + cls._new_cache_artifacts = defaultdict(list) + cls._seen_artifacts = OrderedSet() + cls._serializer = AppendingByteSerializer(serialize_fn=_serialize_single_cache) + cls._cache_info = cls._cache_info.__class__() + try: + yield + finally: + cls._new_cache_artifacts = original_new_cache_artifacts + cls._seen_artifacts = original_seen_artifacts + cls._serializer = original_serializer + cls._cache_info = original_cache_info + + @classmethod + def record_artifact( + cls, + artifact_type: str, + key: str, + content: Any, + ) -> None: + """ + Called from each caching operation to record the artifact in this + "mega" list + """ + artifact = CacheArtifactFactory.encode_create(artifact_type, key, content) + if artifact in cls._seen_artifacts: + return + log.debug("Recording %s", artifact) + cls._new_cache_artifacts[artifact_type].append(artifact) + cls._seen_artifacts.add(artifact) + + @classmethod + def need_serialize(cls) -> bool: + """ + Have we seen new artifacts since last serialize call? + """ + return len(cls._new_cache_artifacts) != 0 + + @classmethod + def serialize(cls) -> tuple[bytes, CacheInfo] | None: + """ + Converts the "mega" list into portable format + """ + for artifact in chain(*cls._new_cache_artifacts.values()): + log.debug("saving: %s", artifact) + cls._cache_info.add(artifact) + + if cls._cache_info.empty(): + # If there are no artifacts, don't just return bytes with + # version. + return None + + try: + # We deep copy cls._cache_info since later compilations + # can keep adding to cache_info + info = copy.deepcopy(cls._cache_info) + cls._serializer.extend(cls._new_cache_artifacts.items()) + artifact_bytes = cls._serializer.to_bytes() + cls._new_cache_artifacts.clear() + return artifact_bytes, info + except Exception: + log.warning("Failed to pickle cache artifacts", exc_info=True) + return None + + @staticmethod + def deserialize(serialized_artifacts: bytes) -> CacheArtifactsResult | None: + """ + Converts the portable format back into CacheArtifacts + """ + try: + CacheArtifactManager._ensure_cache_artifacts_registered() + artifacts = dict( + AppendingByteSerializer.to_list( + serialized_artifacts, + deserialize_fn=_deserialize_single_cache, + ) + ) + except Exception: + log.warning("Failed to un-pickle cache artifacts", exc_info=True) + return None + + return artifacts + + @staticmethod + def populate_caches(artifacts: CacheArtifactsResult) -> CacheInfo: + info = CacheInfo() + for artifact in chain(*artifacts.values()): + log.debug("writing: %s", artifact) + info.add(artifact) + artifact.populate_cache() + + return info + + @classmethod + def _ensure_cache_artifacts_registered(cls) -> None: + """When deserializing caches in fresh process, we need to ensure that all + cache artifacts are registered in the cache registry. This is done by + simply importing all the cache artifacts already wrapped with register call. + """ + from torch._dynamo.package import PrecompileCacheArtifact # noqa: F401 + from torch._dynamo.pgo import PGOCacheArtifact # noqa: F401 + from torch._functorch._aot_autograd.autograd_cache import ( # noqa: F401 + AOTAutogradCacheArtifact, + ) + from torch._inductor.codecache import InductorCacheArtifact # noqa: F401 + from torch._inductor.runtime.autotune_cache import ( # noqa: F401 + AutotuneCacheArtifact, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/compiler/config.py b/.venv/lib/python3.12/site-packages/torch/compiler/config.py new file mode 100644 index 0000000000000000000000000000000000000000..bdec1ad7d17bc3b221eb12647d487c67eb075cde --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/compiler/config.py @@ -0,0 +1,344 @@ +""" +This is the top-level configuration module for the compiler, containing +cross-cutting configuration options that affect all parts of the compiler +stack. + +You may also be interested in the per-component configuration modules, which +contain configuration options that affect only a specific part of the compiler: + +* :mod:`torch._dynamo.config` +* :mod:`torch._inductor.config` +* :mod:`torch._functorch.config` +* :mod:`torch.fx.experimental.config` +""" + +import sys + +from torch.utils._config_module import Config, install_config_module + + +__all__ = [ + "job_id", + "dynamic_shapes", + "assume_static_by_default", + "automatic_dynamic_shapes", + "recompile_limit", + "accumulated_recompile_limit", + "verbose", + "capture_scalar_outputs", + "capture_dynamic_output_shape_ops", + "log_file_name", + "fail_on_recompile_limit_hit", + "allow_unspec_int_on_nn_module", + "skip_tensor_guards_with_matching_dict_tags", + "enable_cpp_symbolic_shape_guards", + "wrap_top_frame", + "reorderable_logging_functions", + "force_disable_caches", + "compile_on_one_rank", +] + + +# NB: Docblocks go UNDER variable definitions! Use spacing to make the +# grouping clear. + +# FB-internal note: you do NOT have to specify this explicitly specify this if +# you run on MAST, we will automatically default this to +# mast:MAST_JOB_NAME:MAST_JOB_VERSION. +job_id: str | None = Config( + env_name_default=["TORCH_COMPILE_JOB_ID", "TORCH_COMPILE_STICKY_PGO_KEY"], + default=None, +) +""" +Semantically, this should be an identifier that uniquely identifies, e.g., a +training job. You might have multiple attempts of the same job, e.g., if it was +preempted or needed to be restarted, but each attempt should be running +substantially the same workload with the same distributed topology. You can +set this by environment variable with :envvar:`TORCH_COMPILE_JOB_ID`. + +Operationally, this controls the effect of profile-guided optimization related +persistent state. PGO state can affect how we perform compilation across +multiple invocations of PyTorch, e.g., the first time you run your program we +may compile twice as we discover what inputs are dynamic, and then PGO will +save this state so subsequent invocations only need to compile once, because +they remember it is dynamic. This profile information, however, is sensitive +to what workload you are running, so we require you to tell us that two jobs +are *related* (i.e., are the same workload) before we are willing to reuse +this information. Notably, PGO does nothing (even if explicitly enabled) +unless a valid ``job_id`` is available. In some situations, PyTorch can be +configured to automatically compute a ``job_id`` based on the environment it +is running in. + +Profiles are always collected on a per rank basis, so different ranks may have +different profiles. If you know your workload is truly SPMD, you can run with +:data:`torch._dynamo.config.enable_compiler_collectives` to ensure nodes get +consistent profiles across all ranks. +""" + +pgo_extra_read_key: str | None = Config( + env_name_default="TORCH_COMPILE_STICKY_PGO_READ", default=None +) +pgo_extra_write_key: str | None = Config( + env_name_default="TORCH_COMPILE_STICKY_PGO_WRITE", default=None +) +""" +Additional read/write keys for PGO. +Write key: Besides writing to the default local/remote PGO state, this also writes to the specified key. +Read key: Besides reading from the default state, this also reads from the specified key (if written to before) +and merges it with the default state. +""" + + +cache_key_tag: str = Config(env_name_default="TORCH_COMPILE_CACHE_KEY_TAG", default="") +""" +Tag to be included in the cache key generation for all torch compile caching. +A common use case for such a tag is to break caches. +""" + +force_disable_caches: bool = Config( + justknob="pytorch/remote_cache:force_disable_caches", + env_name_force=[ + "TORCHINDUCTOR_FORCE_DISABLE_CACHES", + "TORCH_COMPILE_FORCE_DISABLE_CACHES", + ], + default=False, +) +""" +Force disables all caching -- This will take precedence over and override any other caching flag +""" + +compile_on_one_rank: bool = Config( + default=False, + env_name_default=[ + "TORCH_COMPILE_ON_ONE_RANK", + "TORCH_DISTRIBUTED_COMPILE_ON_ONE_RANK", + ], +) +""" +When enabled, device- and rank-specific values (devices, process groups) are computed at +runtime via custom ops rather than baked in at compile time, so a graph can be compiled on +one rank and run on all ranks. Read across the stack: make_fx, inductor, and distributed. +""" + +dynamic_sources: str = Config( + env_name_default="TORCH_COMPILE_DYNAMIC_SOURCES", default="" +) +r""" +Comma delimited list of sources that should be marked as dynamic. Primarily useful for large +models with graph breaks where you need intermediate tensors and ints to be marked dynamic. + +Each entry is either an exact source name or a Python regex (regex applies only to the +source name, not to the dim) matched with ``re.match``. An optional ``:N`` suffix restricts +the entry to dim ``N`` of the matched tensor; without the suffix all dims are marked dynamic. +``N`` must be a literal non-negative integer. To target multiple dims of the same source, +list it multiple times. Examples:: + + L['x'] # all dims of x dynamic (current behavior) + L['x']:0 # only dim 0 of x dynamic + L['x']:0, L['x']:2 # dims 0 and 2 of x dynamic + +Regex examples (``re.match`` anchors at the start, not the end):: + + L\['x.*'\] # all sources whose name starts with L['x (e.g. L['x'], L['x1']) + L\['x.*'\]:0 # dim 0 dynamic for every source matching L['x.* + .* # mark every source dynamic (all dims) + .*:0 # mark dim 0 dynamic for every tensor source + +The ``:N`` suffix is ignored for non-tensor (int/scalar) sources, which have no dim. + +Entries in this list are dominant over all other flags dynamic=False, force_nn_module_property_static_shapes +and force_parameter_static_shapes. +""" + +dynamic_values: str = Config( + env_name_default="TORCH_COMPILE_DYNAMIC_VALUES", default="" +) +""" +Comma delimited list of integer values that should cause any matching int source or +tensor dim to be marked dynamic. + +This is useful for JIT workflows with graph breaks where you don't know up-front which +sources need to be dynamic, but you can warm up the model with a distinctive "sentinel" +value (e.g. an odd batch size like 1111) and have any int or tensor dim that equals one +of those values be marked dynamic automatically. This is needed because with graph +breaks we do not propagate dynamic properties through eager code, so an int or tensor +derived in eager between two compiled regions cannot inherit the dynamic-ness of its +source — matching on the sentinel value lets us recover it on the next compiled region. + +Example:: + + TORCH_COMPILE_DYNAMIC_VALUES="1111,2222" + +Pick sentinel values unlikely to occur as legitimate shapes/ints in your +workload (e.g. 1111, 2222) — matching is purely by value, so any int source +or tensor dim equal to a listed value anywhere in the program will be +marked dynamic. + +unlike ``dynamic_sources``, ``dynamic_values`` does NOT override +``force_parameter_static_shapes`` or ``force_nn_module_property_static_shapes``. +""" + +unbacked_sources: str = Config( + env_name_default="TORCH_COMPILE_UNBACKED_SOURCES", default="" +) +""" +Comma delimited list of sources that should be marked as unbacked. Primarily useful for large +models with graph breaks where you need intermediate tensors marked unbacked. + +Supports the same ``:N`` per-dim suffix syntax as ``dynamic_sources``; see that config's +docstring for details. + +Entries in this list are dominant over all other flags dynamic=False, force_nn_module_property_static_shapes +and force_parameter_static_shapes. +""" + +# force a python GC before recording cudagraphs +force_cudagraph_gc: bool = Config(env_name_default="TORCH_CUDAGRAPH_GC", default=False) +""" +If True (the backward-compatible behavior) then gc.collect() before recording +any cudagraph. +""" + + +# Cross-cutting configuration options that affect the entire compilation pipeline + +dynamic_shapes: bool = Config(alias="torch._dynamo.config.dynamic_shapes") +""" +Controls whether the compilation pipeline supports dynamic tensor shapes. +When enabled, the compiler can handle tensors with varying dimensions across +different invocations. This is a cross-cutting setting that affects shape +inference, guard generation, and code generation across the entire compilation +stack. +""" + +assume_static_by_default: bool = Config( + alias="torch._dynamo.config.assume_static_by_default" +) +""" +When enabled, all tensor dimensions are assumed to be static unless explicitly +marked as dynamic or detected as changing. This compilation-wide behavior affects +how the entire stack handles shape specialization and can improve performance +for static workloads. +""" + +automatic_dynamic_shapes: bool = Config( + alias="torch._dynamo.config.automatic_dynamic_shapes" +) +""" +Enables automatic detection and handling of dynamic shapes. When a tensor's +shape changes between compilations, the system automatically marks those +dimensions as dynamic rather than requiring manual specification. This +cross-cutting optimization improves the user experience by reducing recompilations. +""" + +recompile_limit: int = Config(alias="torch._dynamo.config.recompile_limit") +""" +Maximum number of recompilations allowed for a single function before falling +back to eager execution. This compilation performance control prevents excessive +recompilation overhead that can degrade overall performance. +""" + +accumulated_recompile_limit: int = Config( + alias="torch._dynamo.config.accumulated_recompile_limit" +) +""" +Global limit on total recompilations across all compiled functions to prevent +runaway recompilation scenarios. This safeguard protects against compilation +performance issues that could affect the entire program. +""" + +verbose: bool = Config(alias="torch._dynamo.config.verbose") +""" +Enables verbose debugging output for Dynamo. When enabled, provides detailed +information about Dynamo's compilation decisions, optimizations, and potential +issues. +""" + + +# TorchDynamo-specific configuration options + +capture_scalar_outputs: bool = Config( + alias="torch._dynamo.config.capture_scalar_outputs" +) +""" +Controls whether TorchDynamo captures operations that return scalar values (like .item()) +into the FX graph. When disabled, these operations cause graph breaks. This is a +TorchDynamo-specific tracing behavior that affects how the tracer handles +scalar-returning operations. +""" + +capture_dynamic_output_shape_ops: bool = Config( + alias="torch._dynamo.config.capture_dynamic_output_shape_ops" +) +""" +Controls whether TorchDynamo captures operations with dynamic output shapes (like +nonzero, unique) into the FX graph. When disabled, these operations cause graph breaks. +This is a TorchDynamo-specific setting for handling operations with unpredictable +output shapes during tracing. +""" + +log_file_name: str | None = Config(alias="torch._dynamo.config.log_file_name") +""" +Specifies a file path for TorchDynamo-specific logging output. When set, internal +TorchDynamo debug information is written to this file rather than stdout. This is +useful for debugging TorchDynamo's internal tracing behavior. +""" + +fail_on_recompile_limit_hit: bool = Config( + alias="torch._dynamo.config.fail_on_recompile_limit_hit" +) +""" +Raises a hard error when recompile limits are exceeded instead of falling back +to eager execution. This is useful for detecting excessive recompilation in +performance-critical deployments where you want to ensure compilation overhead +is kept under control. +""" + +allow_unspec_int_on_nn_module: bool = Config( + alias="torch._dynamo.config.allow_unspec_int_on_nn_module" +) +""" +Allows integer attributes of nn.Module instances to be unspecialized through +the dynamic shape mechanism. By default, TorchDynamo specializes on all integer +module attributes, but this can cause excessive recompilation when integers +like step counters change frequently. +""" + +skip_tensor_guards_with_matching_dict_tags: bool = Config( + alias="torch._dynamo.config.skip_tensor_guards_with_matching_dict_tags" +) +""" +Optimizes guard generation by treating tensors as immutable when they are +dictionary values with consistent dictionary tags across invocations. This +reduces guard overhead for tensors stored in persistent data structures. +""" + +enable_cpp_symbolic_shape_guards: bool = Config( + alias="torch._dynamo.config.enable_cpp_symbolic_shape_guards" +) +""" +Uses C++ implementation for symbolic shape guard evaluation to improve performance. +The C++ guard manager can significantly speed up guard checking for symbolic shapes +in shape-polymorphic compilations. +""" + +wrap_top_frame: bool = Config(alias="torch._dynamo.config.wrap_top_frame") +""" +Wraps the top-level decorated function/module in a frame wrapper to ensure +nn.Module hooks are compiled within the same frame as the main function. This +improves compilation coverage for models that rely on hooks. +""" + +reorderable_logging_functions: set = Config( + alias="torch._dynamo.config.reorderable_logging_functions" +) +""" +A set of logging functions that can be reordered to execute after the compiled +portion of the graph, allowing larger graphs to be captured. Functions in this +set will have their execution deferred to avoid graph breaks, though this may +affect the timing of log output. In particular, mutated values will not be logged +at the right time, leading to incorrect logging. +""" + + +install_config_module(sys.modules[__name__]) diff --git a/.venv/lib/python3.12/site-packages/torch/contrib/__init__.py b/.venv/lib/python3.12/site-packages/torch/contrib/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..df0f068226a5262448b34e73c62573c8a16c5e11 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/_tensorboard_vis.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/_tensorboard_vis.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2eb9a18d8693f9c41dfa48fe771cc78cedf3ddd3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/contrib/__pycache__/_tensorboard_vis.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/contrib/_tensorboard_vis.py b/.venv/lib/python3.12/site-packages/torch/contrib/_tensorboard_vis.py new file mode 100644 index 0000000000000000000000000000000000000000..fd54513d76e390f80d0079e88e399658832d0eb9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/contrib/_tensorboard_vis.py @@ -0,0 +1,152 @@ +# mypy: allow-untyped-defs +import time +from collections import defaultdict +from functools import partial + +import torch + + +# Unfortunately it doesn't seem as if there was any way to get TensorBoard to do +# anything without having TF installed, and so this file has a hard dependency on it +# as well. It really is a debugging tool, so it doesn't matter. +try: + from tensorflow.core.framework import graph_pb2 + from tensorflow.core.util import event_pb2 + from tensorflow.python.summary.writer.writer import FileWriter +except ImportError: + raise ImportError( + "TensorBoard visualization of GraphExecutors requires having " + "TensorFlow installed" + ) from None + + +def dump_tensorboard_summary(graph_executor, logdir): + with FileWriter(logdir) as w: + pb_graph = visualize(graph_executor) + evt = event_pb2.Event( + wall_time=time.time(), graph_def=pb_graph.SerializeToString() + ) + w.add_event(evt) + + +def visualize(graph, name_prefix="", pb_graph=None, executors_it=None): + """Visualizes an independent graph, or a graph executor.""" + value_map = {} + pb_graph = pb_graph or graph_pb2.GraphDef() + + if isinstance(graph, torch._C.GraphExecutorState): + visualize_graph_executor( + graph, name_prefix, pb_graph, partial(visualize, pb_graph=pb_graph) + ) + return pb_graph + + # Set up an input node + pb_graph.node.add(op="input", name=name_prefix + "input") + for i, value in enumerate(graph.param_node().outputs()): + value_map[value.unique()] = name_prefix + "input:" + str(i) + + visualize_rec(graph, value_map, name_prefix, pb_graph, executors_it) + + # Gather all outputs + return_node = pb_graph.node.add(op="output", name=name_prefix + "output") + for value in graph.return_node().inputs(): + return_node.input.append(value_map[value.unique()]) + + return pb_graph + + +def visualize_graph_executor(state, name_prefix, pb_graph, inline_graph): + """Append the state of a given GraphExecutor to the graph protobuf. + + Args: + state (GraphExecutor or GraphExecutorState): GraphExecutor to display. + name_prefix (str): Name prefix of the containing subgraph. + pb_graph (GraphDef): graph to append to. + inline_graph (Callable): a function that handles setting up a value_map, + so that some graphs in here can be inlined. This is necessary, because + this will simply be `visualize` for the top-level GraphExecutor, + or `inline_graph` for all nested ones. + + The signature should look like (Graph, name_prefix) -> (). + It will be called exactly once. + + The strategy is to embed all different configurations as independent subgraphs, + while inlining the original graph as the one that actually produces the values. + """ + if state.autograd_fallback_graph is not None: + visualize( + graph=state.autograd_fallback_graph, + name_prefix=name_prefix + "autograd_fallback/", + pb_graph=pb_graph, + executors_it=iter(state.autograd_fallback.executors()), + ) + + for i, (arg_spec, plan) in enumerate(state.execution_plans.items()): + subgraph_name = name_prefix + f"plan{i}/" + + # Create a disconnected node that will keep information regarding the input + # types of this trace. This is unfortunately a bit too verbose to be included + # in the subgraph name. + input_kinds = pb_graph.node.add(op="INPUT_KIND", name=subgraph_name) + input_kinds.attr["inputs"].s = repr(arg_spec).encode("ascii") + + visualize(plan.graph, subgraph_name, pb_graph, iter(plan.code.executors())) + + # Show gradient as an independent subgraph of this plan + if plan.grad_executor is not None: + grad_subgraph_name = subgraph_name + "grad/" + visualize(plan.grad_executor, grad_subgraph_name, pb_graph) + + return inline_graph(state.graph, name_prefix + "original/") + + +def visualize_rec(graph, value_map, name_prefix, pb_graph, executors_it=None): + """Recursive part of visualize (basically skips setting up the input and output nodes).""" + + def inline_graph(subgraph, name, node): + rec_value_map = { + inp.unique(): value_map[val.unique()] + for inp, val in zip(subgraph.inputs(), node.inputs()) + } + visualize_rec( + graph=subgraph, value_map=rec_value_map, name_prefix=name, pb_graph=pb_graph + ) + for out, val in zip(subgraph.outputs(), node.outputs()): + value_map[val.unique()] = rec_value_map[out.unique()] + + op_id_counter: defaultdict[str, int] = defaultdict(int) + + def name_for(node): + kind = node.kind()[node.kind().index("::") + 2 :] + op_id_counter[kind] += 1 + return kind, name_prefix + kind + "_" + str(op_id_counter[kind]) + + def add_fusion_group(node): + op, name = name_for(node) + inline_graph(node.g("Subgraph"), name + "/", node) + + def add_graph_executor(node): + op, name = name_for(node) + if executors_it is None: + add_node(node) + else: + ge = next(executors_it) + visualize_graph_executor( + ge, name + "/", pb_graph, partial(inline_graph, node=node) + ) + + def add_node(node): + if node.kind() == "prim::FusionGroup": + return add_fusion_group(node) + elif node.kind() == "prim::GraphExecutor": + return add_graph_executor(node) + op, name = name_for(node) + pb_node = pb_graph.node.add(op=op, name=name) + for value in node.inputs(): + pb_node.input.append(value_map[value.unique()]) + # TODO: handle attrs + for i, value in enumerate(node.outputs()): + value_map[value.unique()] = name + ":" + str(i) + + for node in graph.nodes(): + add_node(node) diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/__init__.py b/.venv/lib/python3.12/site-packages/torch/cpu/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f98f06d2a7bde6f4b477e1488584dae969095e43 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cpu/__init__.py @@ -0,0 +1,252 @@ +# mypy: allow-untyped-defs +r""" +This package implements abstractions found in ``torch.cuda`` +to facilitate writing device-agnostic code. +""" + +from collections.abc import Mapping +from contextlib import AbstractContextManager +from functools import lru_cache +from types import MappingProxyType +from typing import Any + +import torch + +from .. import device as _device +from . import amp + + +__all__ = [ + "is_available", + "is_initialized", + "synchronize", + "current_device", + "current_stream", + "stream", + "set_device", + "device_count", + "Stream", + "StreamContext", + "Event", + "get_capabilities", +] + + +@lru_cache(None) +def get_capabilities() -> Mapping[str, Any]: + """ + Returns an immutable mapping of CPU capabilities detected at runtime. + + This function queries the CPU for supported instruction sets and features + using cpuinfo. The result is cached after the first call for efficiency. + + The returned mapping contains architecture-specific capabilities: + + For x86/x86_64: + - SSE family: sse, sse2, sse3, ssse3, sse4_1, sse4_2, sse4a + - AVX family: avx, avx2, avx_vnni + - AVX-512 family: avx512_f, avx512_cd, avx512_dq, avx512_bw, avx512_vl, + avx512_ifma, avx512_vbmi, avx512_vbmi2, avx512_bitalg, avx512_vpopcntdq, + avx512_vnni, avx512_bf16, avx512_fp16, avx512_vp2intersect, + avx512_4vnniw, avx512_4fmaps + - AVX10 family: avx10_1, avx10_2 + - AVX-VNNI-INT: avx_vnni_int8, avx_vnni_int16, avx_ne_convert + - AMX: amx_bf16, amx_tile, amx_int8, amx_fp16 + - FMA: fma3, fma4 + - Other: f16c, bmi, bmi2, popcnt, lzcnt, aes, sha, clflush, clflushopt, clwb + + For ARM64: + - SIMD: neon, fp16_arith, bf16, i8mm, dot + - SVE: sve, sve2, sve_bf16, sve_max_length (when supported) + - SME: sme, sme2, sme_max_length (when supported) + - Other: atomics, fhm, rdm, crc32, aes, sha1, sha2, pmull + + Common to all architectures: + - architecture: string identifying the CPU architecture + + Returns: + MappingProxyType: An immutable mapping where keys are capability names + (e.g., 'avx2', 'sve') and values are booleans indicating + support, or integers for properties like vector lengths. + + Example: + >>> caps = torch.cpu.get_capabilities() + >>> if caps.get("avx2", False): + ... print("AVX2 is supported") + >>> print(f"Architecture: {caps['architecture']}") + """ + return MappingProxyType(torch._C._cpu._get_cpu_capability()) + + +def _is_avx2_supported() -> bool: + r"""Returns a bool indicating if CPU supports AVX2.""" + return get_capabilities().get("avx2", False) + + +def _is_avx512_supported() -> bool: + r"""Returns a bool indicating if CPU supports AVX512.""" + return get_capabilities().get("avx512_f", False) + + +def _is_avx512_bf16_supported() -> bool: + r"""Returns a bool indicating if CPU supports AVX512_BF16.""" + return get_capabilities().get("avx512_bf16", False) + + +def _is_vnni_supported() -> bool: + r"""Returns a bool indicating if CPU supports VNNI.""" + # Note: Currently, it only checks avx512_vnni, will add the support of avx2_vnni later. + return get_capabilities().get("avx512_vnni", False) + + +def _is_amx_tile_supported() -> bool: + r"""Returns a bool indicating if CPU supports AMX_TILE.""" + return get_capabilities().get("amx_tile", False) + + +def _is_amx_fp16_supported() -> bool: + r"""Returns a bool indicating if CPU supports AMX FP16.""" + return get_capabilities().get("amx_fp16", False) + + +def _init_amx() -> bool: + r"""Initializes AMX instructions.""" + return torch._C._cpu._init_amx() + + +def is_available() -> bool: + r"""Returns a bool indicating if CPU is currently available. + + N.B. This function only exists to facilitate device-agnostic code + + """ + return True + + +def synchronize(device: torch.types.Device = None) -> None: + r"""Waits for all kernels in all streams on the CPU device to complete. + + Args: + device (torch.device or int, optional): ignored, there's only one CPU device. + + N.B. This function only exists to facilitate device-agnostic code. + """ + + +class Stream: + """ + N.B. This class only exists to facilitate device-agnostic code + """ + + def __init__(self, priority: int = -1) -> None: + pass + + def wait_stream(self, stream) -> None: + pass + + def record_event(self) -> None: + pass + + def wait_event(self, event) -> None: + pass + + +class Event: + def query(self) -> bool: + return True + + def record(self, stream=None) -> None: + pass + + def synchronize(self) -> None: + pass + + def wait(self, stream=None) -> None: + pass + + +_default_cpu_stream = Stream() +_current_stream = _default_cpu_stream + + +def current_stream(device: torch.types.Device = None) -> Stream: + r"""Returns the currently selected :class:`Stream` for a given device. + + Args: + device (torch.device or int, optional): Ignored. + + N.B. This function only exists to facilitate device-agnostic code + + """ + return _current_stream + + +class StreamContext(AbstractContextManager): + r"""Context-manager that selects a given stream. + + N.B. This class only exists to facilitate device-agnostic code + + """ + + cur_stream: Stream | None + + def __init__(self, stream): + self.stream = stream + self.prev_stream = _default_cpu_stream + + def __enter__(self): + cur_stream = self.stream + if cur_stream is None: + return + + global _current_stream + self.prev_stream = _current_stream + _current_stream = cur_stream + + def __exit__(self, type: Any, value: Any, traceback: Any) -> None: + cur_stream = self.stream + if cur_stream is None: + return + + global _current_stream + _current_stream = self.prev_stream + + +def stream(stream: Stream) -> AbstractContextManager: + r"""Wrapper around the Context-manager StreamContext that + selects a given stream. + + N.B. This function only exists to facilitate device-agnostic code + """ + return StreamContext(stream) + + +def device_count() -> int: + r"""Returns number of CPU devices (not cores). Always 1. + + N.B. This function only exists to facilitate device-agnostic code + """ + return 1 + + +def set_device(device: torch.types.Device) -> None: + r"""Sets the current device, in CPU we do nothing. + + N.B. This function only exists to facilitate device-agnostic code + """ + + +def current_device() -> str: + r"""Returns current device for cpu. Always 'cpu'. + + N.B. This function only exists to facilitate device-agnostic code + """ + return "cpu" + + +def is_initialized() -> bool: + r"""Returns True if the CPU is initialized. Always True. + + N.B. This function only exists to facilitate device-agnostic code + """ + return True diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cpu/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..971284c55853ac7342b0cb4324dbb543c493483f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cpu/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/__init__.py b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..dae673c7b2313480a940a9cc19517dba21d20d3a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__init__.py @@ -0,0 +1,3 @@ +# pyrefly: ignore [deprecated] +from .autocast_mode import autocast +from .grad_scaler import GradScaler diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b40e712388ac3e68aefede3954a526a7507dfff4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/autocast_mode.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/autocast_mode.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..699b54b29f998308a7f76339a120855034bcef5e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/autocast_mode.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/grad_scaler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/grad_scaler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d363e3b10a147607c5c2871cece9c1d9cca1b51b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cpu/amp/__pycache__/grad_scaler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/autocast_mode.py b/.venv/lib/python3.12/site-packages/torch/cpu/amp/autocast_mode.py new file mode 100644 index 0000000000000000000000000000000000000000..f0f81060d4a01fc6857138c49ec8276bee59b90d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cpu/amp/autocast_mode.py @@ -0,0 +1,71 @@ +# mypy: allow-untyped-defs +import sys +from typing import Any +from typing_extensions import deprecated + +import torch + + +__all__ = ["autocast"] + + +@deprecated( + "`torch.cpu.amp.autocast(args...)` is deprecated. " + "Please use `torch.amp.autocast('cpu', args...)` instead.", + category=FutureWarning, +) +class autocast(torch.amp.autocast_mode.autocast): + r""" + See :class:`torch.autocast`. + ``torch.cpu.amp.autocast(args...)`` is deprecated. Please use ``torch.amp.autocast("cpu", args...)`` instead. + """ + + # TODO: remove this conditional once we stop supporting Python < 3.13 + # Prior to Python 3.13, inspect.signature could not retrieve the correct + # signature information for classes decorated with @deprecated (unless + # the __new__ static method was explicitly defined); + # + # However, this issue has been fixed in Python 3.13 and later versions. + if sys.version_info < (3, 13): + + def __new__( + cls, + enabled: bool = True, + dtype: torch.dtype = torch.bfloat16, + cache_enabled: bool = True, + ): + return super().__new__(cls) + + def __init_subclass__(cls): + pass + + def __init__( + self, + enabled: bool = True, + dtype: torch.dtype = torch.bfloat16, + cache_enabled: bool = True, + ): + if torch._jit_internal.is_scripting(): + self._enabled = enabled + self.device = "cpu" + self.fast_dtype = dtype + return + super().__init__( + "cpu", enabled=enabled, dtype=dtype, cache_enabled=cache_enabled + ) + + def __enter__(self): + if torch._jit_internal.is_scripting(): + return self + return super().__enter__() + + # TODO: discuss a unified TorchScript-friendly API for autocast + def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any): # type: ignore[override] + if torch._jit_internal.is_scripting(): + return + return super().__exit__(exc_type, exc_val, exc_tb) + + def __call__(self, func): + if torch._jit_internal.is_scripting(): + return func + return super().__call__(func) diff --git a/.venv/lib/python3.12/site-packages/torch/cpu/amp/grad_scaler.py b/.venv/lib/python3.12/site-packages/torch/cpu/amp/grad_scaler.py new file mode 100644 index 0000000000000000000000000000000000000000..aefaa1c323f5ff9089fc69c7a7aabbb380cc7233 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cpu/amp/grad_scaler.py @@ -0,0 +1,35 @@ +from typing_extensions import deprecated + +import torch + + +__all__ = ["GradScaler"] + + +class GradScaler(torch.amp.GradScaler): + r""" + See :class:`torch.amp.GradScaler`. + ``torch.cpu.amp.GradScaler(args...)`` is deprecated. Please use ``torch.amp.GradScaler("cpu", args...)`` instead. + """ + + @deprecated( + "`torch.cpu.amp.GradScaler(args...)` is deprecated. " + "Please use `torch.amp.GradScaler('cpu', args...)` instead.", + category=FutureWarning, + ) + def __init__( + self, + init_scale: float = 2.0**16, + growth_factor: float = 2.0, + backoff_factor: float = 0.5, + growth_interval: int = 2000, + enabled: bool = True, + ) -> None: + super().__init__( + "cpu", + init_scale=init_scale, + growth_factor=growth_factor, + backoff_factor=backoff_factor, + growth_interval=growth_interval, + enabled=enabled, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/csrc/inductor/aoti_runtime/model.h b/.venv/lib/python3.12/site-packages/torch/csrc/inductor/aoti_runtime/model.h new file mode 100644 index 0000000000000000000000000000000000000000..41edb7c1e8e42f2346216f23bffafa6f5e7ea649 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/csrc/inductor/aoti_runtime/model.h @@ -0,0 +1,70 @@ +#pragma once + +// WARNING: Be careful when adding new includes here. This header will be used +// in model.so, and should not refer to any aten/c10 headers except the stable +// C ABI defined in torch/csrc/inductor/aoti_torch/c/shim.h. The same rule +// applies to other files under torch/csrc/inductor/aoti_runtime/. +#include + +struct AOTInductorArrayRefTensor; + +namespace torch::aot_inductor { + +class AOTInductorModel : public AOTInductorModelBase { + public: + AOTInductorModel( + std::shared_ptr constants_map, + std::shared_ptr> constants_array, + const std::string& device_str, + std::optional cubin_dir); + + std::unordered_map const_run_impl( + DeviceStreamType stream, + AOTIProxyExecutorHandle proxy_executor, + bool initialization = false); + + void _const_run_impl( + std::vector& output_handles, + DeviceStreamType stream, + AOTIProxyExecutorHandle proxy_executor); + + void run_impl( + AtenTensorHandle* + input_handles, // array of input AtenTensorHandle; handles + // are stolen; the array itself is borrowed + AtenTensorHandle* + output_handles, // array for writing output AtenTensorHandle; handles + // will be stolen by the caller; the array itself is + // borrowed + DeviceStreamType stream, + AOTIProxyExecutorHandle proxy_executor); + + template + Outputs run_impl_minimal_arrayref_interface( + const Inputs& inputs, + DeviceStreamType stream, + AOTIProxyExecutorHandle proxy_executor); + + void run_impl_minimal_arrayref_interface_v2_raw( + const AOTInductorArrayRefTensor* c_inputs, + AOTInductorArrayRefTensor* c_outputs, + DeviceStreamType stream, + AOTIProxyExecutorHandle proxy_executor); + + static std::unique_ptr Create( + std::shared_ptr constants_map, + std::shared_ptr> constants_array, + const std::string& device_str, + std::optional cubin_dir) { + return std::make_unique( + std::move(constants_map), + std::move(constants_array), + device_str, + std::move(cubin_dir)); + } + + private: + std::unique_ptr kernels_; +}; + +} // namespace torch::aot_inductor diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py b/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..ad77c7b60dfb1274364c003495152bc7fc3f1b01 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py @@ -0,0 +1,2191 @@ +# mypy: allow-untyped-defs +r""" +This package adds support for CUDA tensor types. + +It implements the same function as CPU tensors, but they utilize +GPUs for computation. + +It is lazily initialized, so you can always import it, and use +:func:`is_available()` to determine if your system supports CUDA. + +:ref:`cuda-semantics` has more details about working with CUDA. +""" + +import glob +import importlib +import importlib.util +import os +import platform +import threading +import traceback +import warnings +from collections.abc import Callable +from functools import lru_cache +from typing import Any, cast, NewType, Optional, TYPE_CHECKING + +import torch +import torch._C +from torch._utils import _dummy_type, _LazySeedTracker, classproperty +from torch.types import Device + +from . import _device_limits, gds +from ._utils import _get_device_index +from .graphs import ( + CUDAGraph, + export_dot, + export_graph_data, + graph, + graph_pool_handle, + is_current_stream_capturing, + make_graphed_callables, +) +from .green_contexts import GreenContext +from .streams import Event, ExternalStream, Stream + + +try: + from torch._C import _cudart # type: ignore[attr-defined] +except ImportError: + _cudart = None + +_initialized = False +_tls = threading.local() +_initialization_lock = threading.Lock() +_queued_calls: list[ + tuple[Callable[[], None], list[str]] +] = [] # don't invoke these until initialization occurs +_is_in_bad_fork = getattr(torch._C, "_cuda_isInBadFork", lambda: False) + +_HAS_PYNVML = False +_HAS_AMDSMI = False +_PYNVML_ERR = None +_AMDSMI_ERR = None +try: + from torch import version as _version + + try: + if not _version.hip: + import pynvml # type: ignore[import] + else: + import ctypes + from pathlib import Path + + # In ROCm (at least up through 6.3.2) there're 2 copies of libamd_smi.so: + # - One at lib/libamd_smi.so + # - One at share/amd_smi/amdsmi/libamd_smi.so + # + # The amdsmi python module hardcodes loading the second one in share- + # https://github.com/ROCm/amdsmi/blob/1d305dc9708e87080f64f668402887794cd46584/py-interface/amdsmi_wrapper.py#L174 + # + # See also https://github.com/ROCm/amdsmi/issues/72. + # + # This creates an ODR violation if the copy of libamd_smi.so from lib + # is also loaded (via `ld` linking, `LD_LIBRARY_PATH` or `rpath`). + # + # In order to avoid the violation we hook CDLL and try using the + # already loaded version of amdsmi, or any version in the processes + # rpath/LD_LIBRARY_PATH first, so that we only load a single copy + # of the .so. + class _amdsmi_cdll_hook: + def __init__(self) -> None: + self.original_CDLL = ctypes.CDLL # type: ignore[misc,assignment] + paths = ["libamd_smi.so"] + if rocm_home := os.getenv("ROCM_HOME", os.getenv("ROCM_PATH")): + paths = [os.path.join(rocm_home, "lib/libamd_smi.so")] + paths + # TheRock ROCm wheels install the amdsmi python package in + # site-packages/amdsmi and ship the native library under the + # _rocm_sdk_core wheel (site-packages/_rocm_sdk_core/lib) + # rather than on the loader path or under $ROCM_HOME. The + # file there is a versioned soname (e.g. libamd_smi.so.26), + # so amdsmi's bare CDLL("libamd_smi.so") can't find it. Locate + # the package and append the concrete versioned files as a + # last-resort fallback so the hook can redirect to them. + if platform.system() == "Linux": + paths = paths + self._rocm_sdk_core_amdsmi_paths() + self.paths: list[str] = paths + + @staticmethod + def _rocm_sdk_core_amdsmi_paths() -> list[str]: + try: + spec = importlib.util.find_spec("_rocm_sdk_core") + except (ImportError, ValueError): + return [] + if spec is None or not spec.submodule_search_locations: + return [] + found: list[str] = [] + for location in spec.submodule_search_locations: + found += glob.glob( + os.path.join(location, "lib", "libamd_smi.so*") + ) + return sorted(found) + + def hooked_CDLL( + self, name: str | Path | None, *args: Any, **kwargs: Any + ) -> ctypes.CDLL: + if name and Path(name).name.startswith("libamd_smi.so"): + for path in self.paths: + try: + return self.original_CDLL(path, *args, **kwargs) + except OSError: + pass + return self.original_CDLL(name, *args, **kwargs) # type: ignore[arg-type] + + def __enter__(self) -> None: + ctypes.CDLL = self.hooked_CDLL # type: ignore[misc,assignment] + + def __exit__(self, type: Any, value: Any, traceback: Any) -> None: + ctypes.CDLL = self.original_CDLL # type: ignore[misc] + + try: + with _amdsmi_cdll_hook(): + import amdsmi # type: ignore[import] + _HAS_AMDSMI = True + except ModuleNotFoundError as err: + _AMDSMI_ERR = err + raise + except (KeyError, OSError) as err: + # The amdsmi python package is installed but its native library + # (libamd_smi.so) could not be discovered/loaded -- e.g. TheRock + # ROCm wheels lay amdsmi out so that its own find_smi_library() + # misses the versioned libamd_smi.so.* and raises KeyError. Treat + # this like a missing optional dependency (degrade to + # _HAS_AMDSMI=False) instead of aborting `import torch`. + _AMDSMI_ERR = err + raise ModuleNotFoundError( + "amdsmi is installed but libamd_smi.so could not be loaded" + ) from err + + _HAS_PYNVML = True + except ModuleNotFoundError: + pass + finally: + del _version +except ImportError as err: + _PYNVML_ERR = err # sometimes a lib is installed but the import fails for some other reason, so we log the error for later + +_lazy_seed_tracker = _LazySeedTracker() + +# Define dummy _CudaDeviceProperties type if PyTorch was compiled without CUDA +if hasattr(torch._C, "_CudaDeviceProperties"): + _CudaDeviceProperties = torch._C._CudaDeviceProperties +else: + _CudaDeviceProperties = _dummy_type("_CudaDeviceProperties") # type: ignore[assignment, misc] + +if hasattr(torch._C, "_cuda_exchangeDevice"): + _exchange_device = torch._C._cuda_exchangeDevice +else: + + def _exchange_device(device: int) -> int: + if device < 0: + return -1 + raise RuntimeError("PyTorch was compiled without CUDA support") + + +if hasattr(torch._C, "_cuda_maybeExchangeDevice"): + _maybe_exchange_device = torch._C._cuda_maybeExchangeDevice +else: + + def _maybe_exchange_device(device: int) -> int: + if device < 0: + return -1 + raise RuntimeError("PyTorch was compiled without CUDA support") + + +has_half: bool = True +has_magma: bool = torch._C._has_magma + +default_generators: tuple[torch._C.Generator] = () # type: ignore[assignment] + + +def _is_compiled() -> bool: + r"""Return true if compile with CUDA support.""" + return hasattr(torch._C, "_cuda_getDeviceCount") + + +def _nvml_based_avail() -> bool: + return os.getenv("PYTORCH_NVML_BASED_CUDA_CHECK") == "1" + + +def is_available() -> bool: + r""" + Return a bool indicating if CUDA is currently available. + + .. note:: This function will NOT poison fork if the environment variable + ``PYTORCH_NVML_BASED_CUDA_CHECK=1`` is set. For more details, see + :ref:`multiprocessing-poison-fork-note`. + """ + if not _is_compiled(): + return False + if _nvml_based_avail(): + # The user has set an env variable to request this availability check that attempts to avoid fork poisoning by + # using NVML at the cost of a weaker CUDA availability assessment. Note that if NVML discovery/initialization + # fails, this assessment falls back to the default CUDA Runtime API assessment (`cudaGetDeviceCount`) + return device_count() > 0 + else: + # The default availability inspection never throws and returns 0 if the driver is missing or can't + # be initialized. This uses the CUDA Runtime API `cudaGetDeviceCount` which in turn initializes the CUDA Driver + # API via `cuInit` + return torch._C._cuda_getDeviceCount() > 0 + + +def is_bf16_supported(including_emulation: bool = True): + r"""Return a bool indicating if the current CUDA/ROCm device supports dtype bfloat16.""" + # Check for ROCm, if true return true, no ROCM_VERSION check required, + # since it is supported on AMD GPU archs. + if torch.version.hip: + return True + + # If CUDA is not available, then it does not support bf16 either + if not is_available(): + return False + + device = torch.cuda.current_device() + + if torch.cuda.get_device_properties(device).major >= 8: + return True + + if not including_emulation: + return False + + # Finally try to create a bfloat16 device. + return _check_bf16_tensor_supported(device) + + +@lru_cache(maxsize=16) +def _check_bf16_tensor_supported(device: Device): + try: + torch.tensor([1.0], dtype=torch.bfloat16, device=device) + return True + except Exception: + return False + + +def is_tf32_supported() -> bool: + r"""Return a bool indicating if the current CUDA/ROCm device supports dtype tf32.""" + if torch.version.hip: + prop_name = torch.cuda.get_device_properties().gcnArchName + archs = ("gfx94", "gfx95") + for arch in archs: + if arch in prop_name: + return True + return False + + # Otherwise, tf32 is supported on CUDA platforms that natively (i.e. no emulation) + # support bfloat16. + return is_bf16_supported(including_emulation=False) + + +def _sleep(cycles): + torch._C._cuda_sleep(cycles) + + +def _extract_arch_version(arch_string: str) -> int: + """Extracts the architecture string from a CUDA version""" + base = arch_string.split("_", maxsplit=2)[1] + base = base.removesuffix("a").removesuffix("f") + return int(base) + + +class _CompatInterval: + """ + Defines a range of compute capabilities starting at a given + version and going up to the end of that major version. This + also allows excluding specific versions from the range. + """ + + def __init__(self, start, exclude: set[int] | None = None): + self.major, self.minor = start // 10, start % 10 + self.exclude = set() if exclude is None else exclude + + def __contains__(self, x): + if x in self.exclude: + return False + x_major, x_minor = x // 10, x % 10 + return x_major == self.major and x_minor >= self.minor + + def __str__(self): + result = f">={self.major}.{self.minor},<{self.major + 1}.0" + if len(self.exclude) > 0: + exceptions = ", ".join(f"{x // 10}.{x % 10}" for x in self.exclude) + result += f" except {{{exceptions}}}" + return result + + +class _CompatSet: + """ + A set of compute capabilities. It exists primarily to support custom + printing logic and is otherwise equivalent to a plain python set(). + """ + + def __init__(self, values: set[int]): + self.values = values + + def __contains__(self, x): + return x in self.values + + def __str__(self): + return "{" + ", ".join(f"{v // 10}.{v % 10}" for v in self.values) + "}" + + +# (code SM)->(device SM required to execute the code) +# +# Developer Notes: +# - This dict should be kept up to date with keys corresponding +# to SM versions that PyTorch can be built for. An out of date +# mapping will lead to false warnings. +# - The keys in dict correspond to known sm versions but the values +# are merely rules based on sm compatibility guarantees for NVIDIA +# devices while accounting for incompatibility of iGPU and dGPU. +DEVICE_REQUIREMENT: dict[int, _CompatSet | _CompatInterval] = { + 50: _CompatInterval(start=50, exclude={53}), + 52: _CompatInterval(start=52, exclude={53}), + 53: _CompatSet({53}), + 60: _CompatInterval(start=60, exclude={62}), + 61: _CompatInterval(start=61, exclude={62}), + 62: _CompatSet({62}), + 70: _CompatInterval(start=70, exclude={72}), + 72: _CompatSet({72}), + 75: _CompatInterval(start=75), + 80: _CompatInterval(start=80, exclude={87}), + 86: _CompatInterval(start=86, exclude={87}), + 87: _CompatSet({87}), + 89: _CompatInterval(start=89), + 90: _CompatInterval(start=90), + 100: _CompatInterval(start=100, exclude={101}), + 101: _CompatSet({101, 110}), # 101 was renamed to 110 + 103: _CompatInterval(start=103), + 110: _CompatSet({101, 110}), # 101 was renamed to 110 + 120: _CompatInterval(start=120), + 121: _CompatInterval(start=121), +} + +# CUDA 13.2 allows SBSA binaries to run on Jetson devices. +# This dict can be combined with DEVICE_REQUIREMENT once +# the minimum supported CUDA version is 13.2. +DEVICE_REQUIREMENT_POST_JETSON_SBSA_UNIFICATION: dict[ + int, _CompatSet | _CompatInterval +] = DEVICE_REQUIREMENT | { + 70: _CompatInterval(start=70), + 80: _CompatInterval(start=80), + 86: _CompatInterval(start=86), +} + +# TORCH_CUDA_ARCH_LIST for PyTorch releases, keyed by host arch. +# Kept in sync with .ci/manywheel/build_cuda.sh by the validator in +# .github/scripts/generate_binary_build_matrix.py. +PYTORCH_RELEASES_CODE_CC: dict[str, dict[str, set[int]]] = { + "12.6": { + "x86_64": {50, 60, 70, 75, 80, 86, 90}, + "aarch64": {80, 90}, + }, + "13.0": { + "x86_64": {75, 80, 86, 90, 100, 120}, + "aarch64": {80, 90, 100, 110, 120}, + }, + "13.2": { + "x86_64": {75, 80, 86, 90, 100, 120}, + "aarch64": {80, 90, 100, 110, 120}, + }, + "13.4": { + "x86_64": {75, 80, 86, 90, 100, 120}, + "aarch64": {80, 90, 100, 110, 120}, + }, +} + + +def _device_requirement(code_cc): + if torch.version.cuda is None: + return None + requirement = ( + DEVICE_REQUIREMENT_POST_JETSON_SBSA_UNIFICATION + if tuple(int(x) for x in torch.version.cuda.split(".")) >= (13, 2) + else DEVICE_REQUIREMENT + ) + return requirement.get(code_cc, None) + + +def _host_arch_key() -> str: + machine = platform.machine().lower() + return "aarch64" if machine == "aarch64" else "x86_64" + + +def _code_compatible_with_device(device_cc: int, code_cc: int): + compatible_devices = _device_requirement(code_cc) + if compatible_devices is None: + warnings.warn( + f"PyTorch was compiled with an unknown compute capability {code_cc // 10}.{code_cc % 10}. " + + " Please create an issue on Github if this is a valid compute capability.", + stacklevel=2, + ) + return device_cc in _CompatInterval(start=code_cc) + return device_cc in compatible_devices + + +def _warn_unsupported_code(device_index: int, device_cc: int, code_ccs: list[int]): + name = get_device_name(device_index) + + arch = _host_arch_key() + compatible_releases: list[str] = [] + for cuda, by_arch in PYTORCH_RELEASES_CODE_CC.items(): + build_ccs = by_arch.get(arch, set()) + if any(_code_compatible_with_device(device_cc, cc) for cc in build_ccs): + compatible_releases.append(cuda) + + lines = [ + f"Found GPU{device_index} {name} which is of compute capability (CC) {device_cc // 10}.{device_cc % 10}.", + "The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports:", + ] + [ + f"- {cc // 10}.{cc % 10} which supports hardware CC {_device_requirement(cc)}" + for cc in code_ccs + ] + + if len(compatible_releases) > 0: + version = torch.__version__ + base_version = version.split("+")[0] + is_nightly = "dev" in base_version + index_root = ( + "https://download.pytorch.org/whl/nightly" + if is_nightly + else "https://download.pytorch.org/whl" + ) + lines.append( + f"Your installed torch=={version} does not include kernels for this GPU. " + "Reinstall the same version against a CUDA build that does, e.g.:" + ) + for cuda in compatible_releases: + cu_tag = "cu" + cuda.replace(".", "") + lines.append( + f" For CUDA {cuda} use pip install torch=={base_version} --index-url {index_root}/{cu_tag}" + ) + else: + lines.append( + f"No published PyTorch CUDA builds for release {torch.__version__} support this GPU. " + "Visit https://pytorch.org/get-started/locally/ to find a compatible release." + ) + + warnings.warn("\n".join(lines), stacklevel=2) + + +def _check_capability(): + if torch.version.cuda is None: # on ROCm we don't want this check + return + + arch_list = get_arch_list() + if len(arch_list) == 0: + return + + code_ccs = [_extract_arch_version(cc) for cc in arch_list] + for d in range(device_count()): + major, minor = get_device_capability(d) + device_cc = 10 * major + minor + if not any( + _code_compatible_with_device(device_cc, code_cc) for code_cc in code_ccs + ): + _warn_unsupported_code(d, device_cc, code_ccs) + + +def _check_cubins(): + incompatible_device_warn = """ +{} with CUDA capability sm_{} is not compatible with the current PyTorch installation. +The current PyTorch install supports CUDA capabilities {}. +If you want to use the {} GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/ +""" + if torch.version.cuda is None: # on ROCm we don't want this check + return + arch_list = get_arch_list() + if len(arch_list) == 0: + return + supported_sm = [_extract_arch_version(arch) for arch in arch_list if "sm_" in arch] + for idx in range(device_count()): + cap_major, cap_minor = get_device_capability(idx) + # NVIDIA GPU compute architectures are backward compatible within major version + supported = any(sm // 10 == cap_major for sm in supported_sm) + if not supported: + device_name = get_device_name(idx) + capability = cap_major * 10 + cap_minor + warnings.warn( + incompatible_device_warn.format( + device_name, capability, " ".join(arch_list), device_name + ), + stacklevel=2, + ) + + +def is_initialized(): + r"""Return whether PyTorch's CUDA state has been initialized.""" + return _initialized and not _is_in_bad_fork() + + +def _lazy_call(callable, **kwargs): + with _initialization_lock: + if is_initialized(): + callable() + else: + # TODO(torch_deploy): this accesses linecache, which attempts to read the + # file system to get traceback info. Patch linecache or do something + # else here if this ends up being important. + global _lazy_seed_tracker + if kwargs.get("seed_all", False): + _lazy_seed_tracker.queue_seed_all(callable, traceback.format_stack()) + elif kwargs.get("seed", False): + _lazy_seed_tracker.queue_seed(callable, traceback.format_stack()) + else: + # Don't store the actual traceback to avoid memory cycle + _queued_calls.append((callable, traceback.format_stack())) + + +_lazy_call(_check_capability) +_lazy_call(_check_cubins) + + +class DeferredCudaCallError(Exception): + pass + + +AcceleratorError = torch._C.AcceleratorError +OutOfMemoryError = torch._C.OutOfMemoryError + + +def init(): + r"""Initialize PyTorch's CUDA state. + + You may need to call this explicitly if you are interacting with + PyTorch via its C API, as Python bindings for CUDA functionality + will not be available until this initialization takes place. + Ordinary users should not need this, as all of PyTorch's CUDA methods + automatically initialize CUDA state on-demand. + + Does nothing if the CUDA state is already initialized. + """ + _lazy_init() + + +def _lazy_init(): + global _initialized, _queued_calls + if is_initialized() or hasattr(_tls, "is_initializing"): + return + with _initialization_lock: + # We be double-checked locking, boys! This is OK because + # the above test was GIL protected anyway. The inner test + # is for when a thread blocked on some other thread which was + # doing the initialization; when they get the lock, they will + # find there is nothing left to do. + if is_initialized(): + return + # It is important to prevent other threads from entering _lazy_init + # immediately, while we are still guaranteed to have the GIL, because some + # of the C calls we make below will release the GIL + if _is_in_bad_fork(): + raise RuntimeError( + "Cannot re-initialize CUDA in forked subprocess. To use CUDA with " + "multiprocessing, you must use the 'spawn' start method" + ) + if not hasattr(torch._C, "_cuda_getDeviceCount"): + raise AssertionError("Torch not compiled with CUDA enabled") + if _cudart is None: + raise AssertionError( + "libcudart functions unavailable. It looks like you have a broken build?" + ) + # This function throws if there's a driver initialization error, no GPUs + # are found or any other error occurs + torch._C._cuda_init() + # Some of the queued calls may reentrantly call _lazy_init(); + # we need to just return without initializing in that case. + # However, we must not let any *other* threads in! + _tls.is_initializing = True + + _queued_calls.extend(calls for calls in _lazy_seed_tracker.get_calls() if calls) + + try: + for queued_call, orig_traceback in _queued_calls: + try: + queued_call() + except Exception as e: + msg = ( + f"CUDA call failed lazily at initialization with error: {str(e)}\n\n" + f"CUDA call was originally invoked at:\n\n{''.join(orig_traceback)}" + ) + raise DeferredCudaCallError(msg) from e + finally: + delattr(_tls, "is_initializing") + _initialized = True + + +def cudart(): + r"""Retrieves the CUDA runtime API module. + + + This function initializes the CUDA runtime environment if it is not already + initialized and returns the CUDA runtime API module (_cudart). The CUDA + runtime API module provides access to various CUDA runtime functions. + + Args: + ``None`` + + Returns: + module: The CUDA runtime API module (_cudart). + + Raises: + RuntimeError: If CUDA cannot be re-initialized in a forked subprocess. + AssertionError: If PyTorch is not compiled with CUDA support or if libcudart functions are unavailable. + + Example of CUDA operations with profiling: + >>> import torch + >>> from torch.cuda import cudart, check_error + >>> import os + >>> + >>> os.environ["CUDA_PROFILE"] = "1" + >>> + >>> def perform_cuda_operations_with_streams(): + >>> stream = torch.cuda.Stream() + >>> with torch.cuda.stream(stream): + >>> x = torch.randn(100, 100, device='cuda') + >>> y = torch.randn(100, 100, device='cuda') + >>> z = torch.mul(x, y) + >>> return z + >>> + >>> torch.cuda.synchronize() + >>> print("====== Start nsys profiling ======") + >>> check_error(cudart().cudaProfilerStart()) + >>> with torch.autograd.profiler.emit_nvtx(): + >>> result = perform_cuda_operations_with_streams() + >>> print("CUDA operations completed.") + >>> check_error(torch.cuda.cudart().cudaProfilerStop()) + >>> print("====== End nsys profiling ======") + + To run this example and save the profiling information, execute: + >>> $ nvprof --profile-from-start off --csv --print-summary -o trace_name.prof -f -- python cudart_test.py + + This command profiles the CUDA operations in the provided script and saves + the profiling information to a file named `trace_name.prof`. + The `--profile-from-start off` option ensures that profiling starts only + after the `cudaProfilerStart` call in the script. + The `--csv` and `--print-summary` options format the profiling output as a + CSV file and print a summary, respectively. + The `-o` option specifies the output file name, and the `-f` option forces the + overwrite of the output file if it already exists. + """ + _lazy_init() + return _cudart + + +class cudaStatus: + SUCCESS: int = 0 + ERROR_NOT_READY: int = 34 + + +class CudaError(RuntimeError): + def __init__(self, code: int) -> None: + # pyrefly: ignore [missing-attribute] + msg = _cudart.cudaGetErrorString(_cudart.cudaError(code)) + super().__init__(f"{msg} ({code})") + + +def check_error(res: int) -> None: + r"""Raise an error if the result of a CUDA runtime API call is not success.""" + # pyrefly: ignore [missing-attribute] + if res != _cudart.cudaError.success: + raise CudaError(res) + + +class _DeviceGuard: + def __init__(self, index: int): + self.idx = index + self.prev_idx = -1 + + def __enter__(self): + self.prev_idx = torch.cuda._exchange_device(self.idx) + + def __exit__(self, type: Any, value: Any, traceback: Any): + self.idx = torch.cuda._maybe_exchange_device(self.prev_idx) + return False + + +class device: + r"""Context-manager that changes the selected device. + + Args: + device (torch.device or int): device index to select. It's a no-op if + this argument is a negative integer or ``None``. + """ + + def __init__(self, device: Any): + self.idx = _get_device_index(device, optional=True) + self.prev_idx = -1 + + def __enter__(self): + self.prev_idx = torch.cuda._exchange_device(self.idx) + + def __exit__(self, type: Any, value: Any, traceback: Any): + self.idx = torch.cuda._maybe_exchange_device(self.prev_idx) + return False + + +class device_of(device): + r"""Context-manager that changes the current device to that of given object. + + You can use both tensors and storages as arguments. If a given object is + not allocated on a GPU, this is a no-op. + + Args: + obj (Tensor or Storage): object allocated on the selected device. + """ + + def __init__(self, obj): + idx = obj.get_device() if obj.is_cuda else -1 + super().__init__(idx) + + +def set_device(device: Device) -> None: + r"""Set the current device. + + Usage of this function is discouraged in favor of :any:`device`. In most + cases it's better to use ``CUDA_VISIBLE_DEVICES`` environmental variable. + + Args: + device (torch.device or int): selected device. This function is a no-op + if this argument is negative. + """ + device = _get_device_index(device) + if device >= 0: + torch._C._cuda_setDevice(device) + + +def get_device_name(device: Device = None) -> str: + r"""Get the name of a device. + + Args: + device (torch.device or int or str, optional): device for which to return the + name. This function is a no-op if this argument is a negative + integer. It uses the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Returns: + str: the name of the device + """ + return get_device_properties(device).name + + +def get_device_capability(device: Device = None) -> tuple[int, int]: + r"""Get the cuda capability of a device. + + Args: + device (torch.device or int or str, optional): device for which to return the + device capability. This function is a no-op if this argument is + a negative integer. It uses the current device, given by + :func:`~torch.cuda.current_device`, if :attr:`device` is ``None`` + (default). + + Returns: + tuple(int, int): the major and minor cuda capability of the device + """ + prop = get_device_properties(device) + return prop.major, prop.minor + + +# pyrefly: ignore [not-a-type] +def get_device_properties(device: Device = None) -> _CudaDeviceProperties: + r"""Get the properties of a device. + + Args: + device (torch.device or int or str, optional): device for which to return the + properties of the device. It uses the current device, given by + :func:`~torch.cuda.current_device`, if :attr:`device` is ``None`` + (default). + + Returns: + _CudaDeviceProperties: the properties of the device + """ + _lazy_init() # will define _get_device_properties + device = _get_device_index(device, optional=True) + if device < 0 or device >= device_count(): + raise AssertionError("Invalid device id") + return _get_device_properties(device) # type: ignore[name-defined] + + +def can_device_access_peer(device: Device, peer_device: Device) -> bool: + r"""Check if peer access between two devices is possible.""" + _lazy_init() + device = _get_device_index(device, optional=True) + peer_device = _get_device_index(peer_device) + if device < 0 or device >= device_count(): + raise AssertionError("Invalid device id") + if peer_device < 0 or peer_device >= device_count(): + raise AssertionError("Invalid peer device id") + return torch._C._cuda_canDeviceAccessPeer(device, peer_device) + + +class StreamContext: + r"""Context-manager that selects a given stream. + + All CUDA kernels queued within its context will be enqueued on a selected + stream. + + Args: + Stream (Stream): selected stream. This manager is a no-op if it's + ``None``. + .. note:: Streams are per-device. + """ + + cur_stream: Optional["torch.cuda.Stream"] + + def __init__(self, stream: Optional["torch.cuda.Stream"]): + self.stream = stream + self.idx = _get_device_index(None, True) + if not torch.jit.is_scripting(): + if self.idx is None: + # pyrefly: ignore [bad-assignment] + self.idx = -1 + + self.src_prev_stream = ( + None if not torch.jit.is_scripting() else torch.cuda.default_stream(None) + ) + self.dst_prev_stream = ( + None if not torch.jit.is_scripting() else torch.cuda.default_stream(None) + ) + + def __enter__(self): + # Local cur_stream variable for type refinement + cur_stream = self.stream + # Return if stream is None or CUDA device not available + if cur_stream is None or self.idx == -1: + return + self.src_prev_stream = torch.cuda.current_stream(None) + + # If the stream is not on the current device, then + # set the current stream on the device + if self.src_prev_stream.device != cur_stream.device: + with device(cur_stream.device): + self.dst_prev_stream = torch.cuda.current_stream(cur_stream.device) + torch.cuda.set_stream(cur_stream) + + def __exit__(self, type: Any, value: Any, traceback: Any): + # Local cur_stream variable for type refinement + cur_stream = self.stream + # If stream is None or no CUDA device available, return + if cur_stream is None or self.idx == -1: + return + + # Reset the stream on the original device + # and destination device + if self.src_prev_stream.device != cur_stream.device: # type: ignore[union-attr] + torch.cuda.set_stream(self.dst_prev_stream) # type: ignore[arg-type] + torch.cuda.set_stream(self.src_prev_stream) # type: ignore[arg-type] + + +def stream(stream: Optional["torch.cuda.Stream"]) -> StreamContext: + r"""Wrap around the Context-manager StreamContext that selects a given stream. + + Arguments: + stream (Stream): selected stream. This manager is a no-op if it's + ``None``. + .. note:: + In eager mode stream is of type Stream class while in JIT it is + an object of the custom class ``torch.classes.cuda.Stream``. + """ + return StreamContext(stream) + + +def _set_stream_by_id(stream_id, device_index, device_type): + r"""set stream specified by the stream id, device index and + device type + + Args: stream_id (int): stream id in stream pool + device_index (int): device index in topo + device_type (int): enum device type + """ + torch._C._cuda_setStream( + stream_id=stream_id, + device_index=device_index, + device_type=device_type, + ) + + +def set_stream(stream: Stream): + r"""Set the current stream. This is a wrapper API to set the stream. + Usage of this function is discouraged in favor of the ``stream`` + context manager. + + Args: + stream (Stream): selected stream. This function is a no-op + if this argument is ``None``. + """ + if stream is None: + return + _set_stream_by_id( + stream_id=stream.stream_id, + device_index=stream.device_index, + device_type=stream.device_type, + ) + + +def _parse_visible_devices() -> list[int] | list[str]: + r"""Parse CUDA_VISIBLE_DEVICES environment variable.""" + var = os.getenv("CUDA_VISIBLE_DEVICES") + + if torch.version.hip: + hip_devices = os.getenv("HIP_VISIBLE_DEVICES") + rocr_devices = os.getenv("ROCR_VISIBLE_DEVICES") + + # You must take care if both HIP and ROCR env vars are set as they have + # different meanings. Both env vars accept either a list of ints or a + # list of UUIDs. The ROCR env var is processed first which then reduces + # the number of GPUs that HIP can select from. + if rocr_devices is not None: + rocr_count = len(rocr_devices.split(",")) + if hip_devices is not None: + # sanity check if both env vars are set + if len(hip_devices.split(",")) > rocr_count: + raise RuntimeError( + "HIP_VISIBLE_DEVICES contains more devices than ROCR_VISIBLE_DEVICES" + ) + # HIP_VISIBLE_DEVICES is preferred over ROCR_VISIBLE_DEVICES + var = hip_devices + else: + return list(range(rocr_count)) + elif hip_devices is not None: + var = hip_devices + + if var is None: + return list(range(64)) + + def _strtoul(s: str) -> int: + """Return -1 or positive integer sequence string starts with.""" + if not s: + return -1 + for idx, c in enumerate(s): + if not (c.isdigit() or (idx == 0 and c in "+-")): + break + if idx + 1 == len(s): + idx += 1 + return int(s[:idx]) if idx > 0 else -1 + + def parse_list_with_prefix(lst: str, prefix: str) -> list[str]: + rcs: list[str] = [] + for elem in lst.split(","): + # Repeated id results in empty set + if elem in rcs: + return cast(list[str], []) + # Anything other but prefix is ignored + if not elem.startswith(prefix): + break + rcs.append(elem) + return rcs + + if var.startswith("GPU-"): + return parse_list_with_prefix(var, "GPU-") + if var.startswith("MIG-"): + return parse_list_with_prefix(var, "MIG-") + # CUDA_VISIBLE_DEVICES uses something like strtoul + # which makes `1gpu2,2ampere` is equivalent to `1,2` + rc: list[int] = [] + for elem in var.split(","): + x = _strtoul(elem.strip()) + # Repeated ordinal results in empty set + if x in rc: + return cast(list[int], []) + # Negative value aborts the sequence + if x < 0: + break + rc.append(x) + return rc + + +def _raw_device_count_amdsmi() -> int: + if not _HAS_AMDSMI: + return -1 + try: + amdsmi.amdsmi_init() + except amdsmi.AmdSmiException as e: + warnings.warn( + f"Can't initialize amdsmi - Error code: {e.err_code}", stacklevel=2 + ) + return -1 + socket_handles = amdsmi.amdsmi_get_processor_handles() + return len(socket_handles) + + +def _raw_device_count_nvml() -> int: + r"""Return number of devices as reported by NVML or negative value if NVML discovery/initialization failed.""" + from ctypes import byref, c_int, CDLL + + nvml_h = CDLL("libnvidia-ml.so.1") + rc = nvml_h.nvmlInit() + if rc != 0: + warnings.warn("Can't initialize NVML", stacklevel=2) + return -1 + dev_count = c_int(-1) + rc = nvml_h.nvmlDeviceGetCount_v2(byref(dev_count)) + if rc != 0: + warnings.warn("Can't get nvml device count", stacklevel=2) + return -1 + del nvml_h + return dev_count.value + + +def _raw_device_uuid_amdsmi() -> list[str] | None: + from ctypes import byref, c_int, c_void_p, CDLL, create_string_buffer + + if not _HAS_AMDSMI: + return None + try: + amdsmi.amdsmi_init() + except amdsmi.AmdSmiException: + warnings.warn("Can't initialize amdsmi", stacklevel=2) + return None + try: + socket_handles = amdsmi.amdsmi_get_processor_handles() + dev_count = len(socket_handles) + except amdsmi.AmdSmiException: + warnings.warn("Can't get amdsmi device count", stacklevel=2) + return None + uuids: list[str] = [] + for idx in range(dev_count): + try: + handler = amdsmi.amdsmi_get_processor_handles()[idx] + except amdsmi.AmdSmiException: + warnings.warn("Cannot get amd device handler", stacklevel=2) + return None + try: + uuid = amdsmi.amdsmi_get_gpu_asic_info(handler)["asic_serial"][ + 2: + ] # Removes 0x prefix from serial + except amdsmi.AmdSmiException: + warnings.warn("Cannot get uuid for amd device", stacklevel=2) + return None + uuids.append( + str(uuid).lower() + ) # Lower-case to match expected HIP_VISIBLE_DEVICES uuid input + return uuids + + +def _raw_device_uuid_nvml() -> list[str] | None: + r"""Return list of device UUID as reported by NVML or None if NVM discovery/initialization failed.""" + from ctypes import byref, c_int, c_void_p, CDLL, create_string_buffer + + nvml_h = CDLL("libnvidia-ml.so.1") + rc = nvml_h.nvmlInit() + if rc != 0: + warnings.warn("Can't initialize NVML", stacklevel=2) + return None + dev_count = c_int(-1) + rc = nvml_h.nvmlDeviceGetCount_v2(byref(dev_count)) + if rc != 0: + warnings.warn("Can't get nvml device count", stacklevel=2) + return None + uuids: list[str] = [] + for idx in range(dev_count.value): + dev_id = c_void_p() + rc = nvml_h.nvmlDeviceGetHandleByIndex_v2(idx, byref(dev_id)) + if rc != 0: + warnings.warn("Can't get device handle", stacklevel=2) + return None + buf_len = 96 + buf = create_string_buffer(buf_len) + rc = nvml_h.nvmlDeviceGetUUID(dev_id, buf, buf_len) + if rc != 0: + warnings.warn("Can't get device UUID", stacklevel=2) + return None + uuids.append(buf.raw.decode("ascii").strip("\0")) + del nvml_h + return uuids + + +def _transform_uuid_to_ordinals(candidates: list[str], uuids: list[str]) -> list[int]: + r"""Given the set of partial uuids and list of known uuids builds a set of ordinals excluding ambiguous partials IDs.""" + + def uuid_to_ordinal(candidate: str, uuids: list[str]) -> int: + best_match = -1 + for idx, uuid in enumerate(uuids): + if not uuid.startswith(candidate): + continue + # Ambiguous candidate + if best_match != -1: + return -1 + best_match = idx + return best_match + + rc: list[int] = [] + for candidate in candidates: + if torch.version.hip: + candidate = candidate.replace( + "GPU-", "", 1 + ) # Remove GPU-prefix to match amdsmi asic serial + idx = uuid_to_ordinal(candidate, uuids) + # First invalid ordinal stops parsing + if idx < 0: + break + # Duplicates result in empty set + if idx in rc: + return cast(list[int], []) + rc.append(idx) + return rc + + +def _device_count_amdsmi() -> int: + visible_devices = _parse_visible_devices() + if not visible_devices: + return 0 + try: + if type(visible_devices[0]) is str: + uuids = _raw_device_uuid_amdsmi() + if uuids is None: + return -1 + # Create string version of visible devices to avoid mypy warnings + visible_device_str = cast(list[str], visible_devices) + visible_devices = _transform_uuid_to_ordinals(visible_device_str, uuids) + else: + raw_cnt = _raw_device_count_amdsmi() + if raw_cnt <= 0: + return raw_cnt + # Trim the list up to a maximum available device + # pyrefly: ignore [bad-argument-type] + for idx, val in enumerate(visible_devices): + # pyrefly: ignore [redundant-cast] + if cast(int, val) >= raw_cnt: + return idx + except OSError: + return -1 + except AttributeError: + return -1 + return len(visible_devices) + + +def _device_count_nvml() -> int: + r"""Return number of devices as reported by NVML taking CUDA_VISIBLE_DEVICES into account. + + Negative value is returned if NVML discovery or initialization has failed. + """ + visible_devices = _parse_visible_devices() + if not visible_devices: + return 0 + try: + if type(visible_devices[0]) is str: + # Skip MIG parsing + if visible_devices[0].startswith("MIG-"): + return -1 + uuids = _raw_device_uuid_nvml() + if uuids is None: + return -1 + visible_devices = _transform_uuid_to_ordinals( + cast(list[str], visible_devices), uuids + ) + else: + raw_cnt = _raw_device_count_nvml() + if raw_cnt <= 0: + return raw_cnt + # Trim the list up to a maximum available device + # pyrefly: ignore [bad-argument-type] + for idx, val in enumerate(visible_devices): + # pyrefly: ignore [redundant-cast] + if cast(int, val) >= raw_cnt: + return idx + except OSError: + return -1 + except AttributeError: + return -1 + return len(visible_devices) + + +def _get_nvml_device_index(device: Device) -> int: + r"""Return the NVML index of the device, taking CUDA_VISIBLE_DEVICES into account.""" + idx = _get_device_index(device, optional=True) + visible_devices = _parse_visible_devices() + if type(visible_devices[0]) is str: + uuids = _raw_device_uuid_nvml() + if uuids is None: + raise RuntimeError("Can't get device UUIDs") + visible_devices = _transform_uuid_to_ordinals( + cast(list[str], visible_devices), uuids + ) + visible_devices = cast(list[int], visible_devices) + if idx < 0 or idx >= len(visible_devices): + raise RuntimeError( + f"device {idx} is not visible (CUDA_VISIBLE_DEVICES={visible_devices})" + ) + return visible_devices[idx] + + +_cached_device_count: int | None = None + + +def device_count() -> int: + r""" + Return the number of GPUs available. + + .. note:: This API will NOT poison fork if NVML discovery succeeds. + See :ref:`multiprocessing-poison-fork-note` for more details. + """ + global _cached_device_count + if not _is_compiled(): + return 0 + if _cached_device_count is not None: + return _cached_device_count + if _initialized or hasattr(_tls, "is_initializing"): + r = torch._C._cuda_getDeviceCount() + else: + # bypass _device_count_nvml() if rocm (not supported) + if torch.version.hip: + nvml_count = _device_count_amdsmi() + else: + nvml_count = _device_count_nvml() + r = torch._C._cuda_getDeviceCount() if nvml_count < 0 else nvml_count + # NB: Do not cache the device count prior to CUDA initialization, because + # the number of devices can change due to changes to CUDA_VISIBLE_DEVICES + # setting prior to CUDA initialization. + if _initialized: + _cached_device_count = r + return r + + +def get_arch_list() -> list[str]: + r"""Return list CUDA architectures this library was compiled for.""" + if not _is_compiled(): + return [] + arch_flags = torch._C._cuda_getArchFlags() + if arch_flags is None: + return [] + return arch_flags.split() + + +def get_gencode_flags() -> str: + r"""Return NVCC gencode flags this library was compiled with.""" + arch_list = get_arch_list() + if len(arch_list) == 0: + return "" + arch_list_ = [arch.split("_") for arch in arch_list] + return " ".join( + [ + f"-gencode compute=compute_{arch},code={kind}_{arch}" + for (kind, arch) in arch_list_ + ] + ) + + +def current_device() -> int: + r"""Return the index of a currently selected device.""" + _lazy_init() + return torch._C._cuda_getDevice() + + +def synchronize(device: Device = None) -> None: + r"""Wait for all kernels in all streams on a CUDA device to complete. + + Args: + device (torch.device or int, optional): device for which to synchronize. + It uses the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + """ + _lazy_init() + with torch.cuda.device(device): + return torch._C._cuda_synchronize() + + +def ipc_collect(): + r"""Force collects GPU memory after it has been released by CUDA IPC. + + .. note:: + Checks if any sent CUDA tensors could be cleaned from the memory. Force + closes shared memory file used for reference counting if there is no + active counters. Useful when the producer process stopped actively sending + tensors and want to release unused memory. + """ + _lazy_init() + return torch._C._cuda_ipc_collect() + + +def current_stream(device: Device = None) -> Stream: + r"""Return the currently selected :class:`Stream` for a given device. + + Args: + device (torch.device or int, optional): selected device. Returns + the currently selected :class:`Stream` for the current device, given + by :func:`~torch.cuda.current_device`, if :attr:`device` is ``None`` + (default). + """ + _lazy_init() + streamdata = torch._C._cuda_getCurrentStream( + _get_device_index(device, optional=True) + ) + return Stream( + stream_id=streamdata[0], device_index=streamdata[1], device_type=streamdata[2] + ) + + +def default_stream(device: Device = None) -> Stream: + r"""Return the default :class:`Stream` for a given device. + + Args: + device (torch.device or int, optional): selected device. Returns + the default :class:`Stream` for the current device, given by + :func:`~torch.cuda.current_device`, if :attr:`device` is ``None`` + (default). + """ + _lazy_init() + streamdata = torch._C._cuda_getDefaultStream( + _get_device_index(device, optional=True) + ) + return Stream( + stream_id=streamdata[0], device_index=streamdata[1], device_type=streamdata[2] + ) + + +def get_stream_from_external(data_ptr: int, device: Device = None) -> Stream: + r"""Return a :class:`Stream` from an externally allocated CUDA stream. + + This function is used to wrap streams allocated in other libraries in order + to facilitate data exchange and multi-library interactions. + + .. note:: This function doesn't manage the stream life-cycle, it is the user + responsibility to keep the referenced stream alive while this returned + stream is being used. + + Args: + data_ptr(int): Integer representation of the `cudaStream_t` value that + is allocated externally. + device(torch.device or int, optional): the device where the stream + was originally allocated. If device is specified incorrectly, + subsequent launches using this stream may fail. + """ + _lazy_init() + streamdata = torch._C._cuda_getStreamFromExternal( + data_ptr, _get_device_index(device, optional=True) + ) + return Stream( + stream_id=streamdata[0], device_index=streamdata[1], device_type=streamdata[2] + ) + + +def current_blas_handle(): + r"""Return cublasHandle_t pointer to current cuBLAS handle""" + _lazy_init() + return torch._C._cuda_getCurrentBlasHandle() + + +def current_solver_handle(): + r"""Return cusolverDnHandle_t pointer to current cuSOLVER handle""" + _lazy_init() + return torch._C._cuda_getCurrentSolverHandle() + + +def set_sync_debug_mode(debug_mode: int | str) -> None: + r"""Set the debug mode for cuda synchronizing operations. + + Args: + debug_mode(str or int): if "default" or 0, don't error or warn on synchronizing operations, + if "warn" or 1, warn on synchronizing operations, if "error" or 2, error out synchronizing operations. + + Warning: + This is an experimental feature, and not all synchronizing operations will trigger warning or error. In + particular, operations in torch.distributed and torch.sparse namespaces are not covered yet. + """ + _lazy_init() + if isinstance(debug_mode, str): + if debug_mode == "default": + debug_mode = 0 + elif debug_mode == "warn": + debug_mode = 1 + elif debug_mode == "error": + debug_mode = 2 + else: + raise RuntimeError( + "invalid value of debug_mode, expected one of `default`, `warn`, `error`" + ) + + torch._C._cuda_set_sync_debug_mode(debug_mode) + + +def get_sync_debug_mode() -> int: + r"""Return current value of debug mode for cuda synchronizing operations.""" + _lazy_init() + return torch._C._cuda_get_sync_debug_mode() + + +def _get_pynvml_handler(device: Device = None): + if not _HAS_PYNVML: + raise ModuleNotFoundError( + "nvidia-ml-py does not seem to be installed or it can't be imported." + # pyrefly: ignore [invalid-inheritance] + ) from _PYNVML_ERR + # pyrefly: ignore [import-error, missing-import, missing-module-attribute] + from pynvml import NVMLError_DriverNotLoaded + + try: + pynvml.nvmlInit() + except NVMLError_DriverNotLoaded as e: + raise RuntimeError("cuda driver can't be loaded, is cuda enabled?") from e + + device = _get_nvml_device_index(device) + handle = pynvml.nvmlDeviceGetHandleByIndex(device) + return handle + + +def _get_amdsmi_handler(device: Device = None): + if not _HAS_AMDSMI: + raise ModuleNotFoundError( + "amdsmi does not seem to be installed or it can't be imported." + ) from _AMDSMI_ERR + try: + amdsmi.amdsmi_init() + except amdsmi.AmdSmiException as e: + raise RuntimeError( + "amdsmi driver can't be loaded, requires >=ROCm6.0 installation" + ) from e + device = _get_amdsmi_device_index(device) + handle = amdsmi.amdsmi_get_processor_handles()[device] + return handle + + +_cached_hip_to_amdsmi: dict[int, int] | None = None + + +def _get_amdsmi_device_index_from_hip_index(device: int) -> int: + r"""Return amdsmi index from HIP device index. They are not always the same. + + Assume amdsmi_init() already completes successfully.""" + global _cached_hip_to_amdsmi + if _cached_hip_to_amdsmi is None: + amdsmi_handles = amdsmi.amdsmi_get_processor_handles() + + def gen(): + for amdsmi_idx, handle in enumerate(amdsmi_handles): + info = amdsmi.amdsmi_get_gpu_enumeration_info(handle) + if "hip_id" in info: + yield info["hip_id"], amdsmi_idx + + _cached_hip_to_amdsmi = dict(gen()) + if not _cached_hip_to_amdsmi and len(amdsmi_handles) > 1: + warnings.warn( + "Cannot translate HIP ID to AMD SMI ID due to" + " lack of translation information prior to ROCm 6.4." + " Functions that rely on amdsmi" + " (e.g. temperature()) may operate on wrong devices." + ) + if device not in _cached_hip_to_amdsmi: + warnings.warn( + f"Cannot translate HIP ID {device} to AMD SMI ID due to" + " undetected HIP ID from amdsmi." + " amdsmi_get_gpu_enumeration_info() only report these HIP IDs" + f" {list(_cached_hip_to_amdsmi.keys())}." + " Functions that rely on amdsmi" + " (e.g. temperature()) may operate on wrong devices." + ) + return _cached_hip_to_amdsmi.get(device, device) + + +def _get_amdsmi_device_index(device: Device) -> int: + r"""Return the amdsmi index of the device, taking visible_devices into account.""" + idx = _get_device_index(device, optional=True) + visible_devices = _parse_visible_devices() + visible_device_is_str = type(visible_devices[0]) is str + if visible_device_is_str: + uuids = _raw_device_uuid_amdsmi() + if uuids is None: + raise RuntimeError("Can't get device UUIDs") + visible_devices_str = cast( + list[str], visible_devices + ) # Create str variable for mypy + visible_devices = _transform_uuid_to_ordinals(visible_devices_str, uuids) + idx_map = dict(enumerate(cast(list[int], visible_devices))) + if idx not in idx_map: + raise RuntimeError( + f"device {idx} is not visible (HIP_VISIBLE_DEVICES={visible_devices})" + ) + if visible_device_is_str: + return idx_map[idx] + else: + return _get_amdsmi_device_index_from_hip_index(idx_map[idx]) + + +def _get_amdsmi_device_memory_used(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + # amdsmi_get_gpu_vram_usage returns mem usage in megabytes + mem_mega_bytes = amdsmi.amdsmi_get_gpu_vram_usage(handle)["vram_used"] + mem_bytes = mem_mega_bytes * 1024 * 1024 + return mem_bytes + + +def _get_amdsmi_memory_usage(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + return amdsmi.amdsmi_get_gpu_activity(handle)["umc_activity"] + + +def _get_amdsmi_utilization(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + return amdsmi.amdsmi_get_gpu_activity(handle)["gfx_activity"] + + +def _get_amdsmi_temperature(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + return amdsmi.amdsmi_get_temp_metric( + handle, + amdsmi.AmdSmiTemperatureType.JUNCTION, + amdsmi.AmdSmiTemperatureMetric.CURRENT, + ) + + +def _get_amdsmi_power_draw(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + socket_power = amdsmi.amdsmi_get_power_info(handle)["average_socket_power"] + if socket_power != "N/A": + return socket_power + else: + socket_power = amdsmi.amdsmi_get_power_info(handle)["current_socket_power"] + if socket_power != "N/A": + return socket_power + else: + return 0 + + +def _get_amdsmi_clock_rate(device: Device = None) -> int: + handle = _get_amdsmi_handler(device) + clock_info = amdsmi.amdsmi_get_clock_info(handle, amdsmi.AmdSmiClkType.GFX) + if "cur_clk" in clock_info: # ROCm 6.2 deprecation + clock_rate = clock_info["cur_clk"] + else: + clock_rate = clock_info["clk"] + if clock_rate != "N/A": + return clock_rate + else: + return 0 + + +def device_memory_used(device: Device = None) -> int: + r"""Return used global (device) memory in bytes as given by `nvidia-smi` or `amd-smi`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + """ + if not torch.version.hip: + handle = _get_pynvml_handler() + device = _get_nvml_device_index(device) + handle = pynvml.nvmlDeviceGetHandleByIndex(device) + return pynvml.nvmlDeviceGetMemoryInfo(handle).used + else: + return _get_amdsmi_device_memory_used(device) + + +def memory_usage(device: Device = None) -> int: + r"""Return the percent of time over the past sample period during which global (device) + memory was being read or written as given by `nvidia-smi`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Warning: Each sample period may be between 1 second and 1/6 second, + depending on the product being queried. + """ + if not torch.version.hip: + handle = _get_pynvml_handler() + device = _get_nvml_device_index(device) + handle = pynvml.nvmlDeviceGetHandleByIndex(device) + return pynvml.nvmlDeviceGetUtilizationRates(handle).memory + else: + return _get_amdsmi_memory_usage(device) + + +def utilization(device: Device = None) -> int: + r"""Return the percent of time over the past sample period during which one or + more kernels was executing on the GPU as given by `nvidia-smi`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Warning: Each sample period may be between 1 second and 1/6 second, + depending on the product being queried. + """ + if not torch.version.hip: + handle = _get_pynvml_handler(device) + device = _get_nvml_device_index(device) + handle = pynvml.nvmlDeviceGetHandleByIndex(device) + return pynvml.nvmlDeviceGetUtilizationRates(handle).gpu + else: + return _get_amdsmi_utilization(device) + + +def temperature(device: Device = None) -> int: + r"""Return the average temperature of the GPU sensor in Degrees C (Centigrades). + + The average temperature is computed based on past sample period as given by `nvidia-smi`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Warning: Each sample period may be between 1 second and 1/6 second, + depending on the product being queried. + """ + if not torch.version.hip: + handle = _get_pynvml_handler(device) + # 0 refers to the temperature sensor for the GPU die. + return pynvml.nvmlDeviceGetTemperature(handle, 0) + else: + return _get_amdsmi_temperature(device) + + +def power_draw(device: Device = None) -> int: + r"""Return the average power draw of the GPU sensor in mW (MilliWatts) + over the past sample period as given by `nvidia-smi` for Fermi or newer fully supported devices. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Warning: Each sample period may be between 1 second and 1/6 second, + depending on the product being queried. + """ + if not torch.version.hip: + handle = _get_pynvml_handler(device) + return pynvml.nvmlDeviceGetPowerUsage(handle) + else: + return _get_amdsmi_power_draw(device) + + +def clock_rate(device: Device = None) -> int: + r"""Return the clock speed of the GPU SM in MHz (megahertz) over the past sample period as given by `nvidia-smi`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Warning: Each sample period may be between 1 second and 1/6 second, + depending on the product being queried. + """ + if not torch.version.hip: + handle = _get_pynvml_handler(device) + return pynvml.nvmlDeviceGetClockInfo(handle, 1) + else: + return _get_amdsmi_clock_rate(device) + + +def _get_device(device: int | str | torch.device) -> torch.device: + r"""Return the torch.device type object from the passed in device. + + Args: + device (torch.device or int): selected device. + """ + if isinstance(device, str): + device = torch.device(device) + elif isinstance(device, int): + device = torch.device("cuda", device) + return device + + +def _get_generator(device: torch.device) -> torch._C.Generator: + r"""Return the CUDA Generator object for the given device. + + Args: + device (torch.device): selected device. + """ + idx = device.index + if idx is None: + idx = current_device() + return torch.cuda.default_generators[idx] + + +def _set_rng_state_offset( + offset: int, device: int | str | torch.device = "cuda" +) -> None: + r"""Set the random number generator state offset of the specified GPU. + + Args: + offset (int): The desired offset + device (torch.device or int, optional): The device to set the RNG state. + Default: ``'cuda'`` (i.e., ``torch.device('cuda')``, the current CUDA device). + """ + final_device = _get_device(device) + + def cb(): + default_generator = _get_generator(final_device) + default_generator.set_offset(offset) + + _lazy_call(cb) + + +def _get_rng_state_offset(device: int | str | torch.device = "cuda") -> int: + r"""Return the random number generator state offset of the specified GPU. + + Args: + device (torch.device or int, optional): The device to return the RNG state offset of. + Default: ``'cuda'`` (i.e., ``torch.device('cuda')``, the current CUDA device). + + .. warning:: + This function eagerly initializes CUDA. + """ + _lazy_init() + final_device = _get_device(device) + default_generator = _get_generator(final_device) + return default_generator.get_offset() + + +# pyrefly: ignore [deprecated] +from .memory import * # noqa: F403 +from .memory import _use_uvm +from .random import * # noqa: F403 + + +################################################################################ +# Define Storage and Tensor classes +################################################################################ + + +@staticmethod # type: ignore[misc] +def _lazy_new(cls, *args, **kwargs): + _lazy_init() + # We may need to call lazy init again if we are a forked child + # del _CudaBase.__new__ + return super(_CudaBase, cls).__new__(cls, *args, **kwargs) + + +class _CudaBase: + is_cuda = True + is_sparse = False + + def type(self, *args, **kwargs): + # We could use a Protocol here to tell mypy that self has `get_device` method + # but it is only available in the typing module on Python >= 3.8 + # or on typing_extensions module on Python >= 3.6 + with device(self.get_device()): # type: ignore[attr-defined] + return super().type(*args, **kwargs) # type: ignore[misc] + + __new__ = _lazy_new + + +from torch.storage import _LegacyStorage, _warn_typed_storage_removal + + +class _CudaLegacyStorage(_LegacyStorage): + @classmethod + def from_buffer(cls, *args, **kwargs): + _warn_typed_storage_removal() + raise RuntimeError("from_buffer: Not available for CUDA storage") + + @classmethod + def _new_with_weak_ptr(cls, *args, **kwargs): + raise RuntimeError("_new_with_weak_ptr: Not available for CUDA storage") + + @classmethod + def _new_shared_filename(cls, manager, obj, size, *, device=None, dtype=None): + raise RuntimeError("_new_shared_filename: Not available for CUDA storage") + + +class ByteStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.uint8 + + +class DoubleStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.double + + +class FloatStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.float + + +class HalfStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.half + + +class LongStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.long + + +class IntStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.int + + +class ShortStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.short + + +class CharStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.int8 + + +class BoolStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.bool + + +class BFloat16Storage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.bfloat16 + + +class ComplexDoubleStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.cdouble + + +class ComplexFloatStorage(_CudaLegacyStorage): + @classproperty + def dtype(self): + _warn_typed_storage_removal() + return self._dtype + + @classproperty + def _dtype(self): + return torch.cfloat + + +del _LegacyStorage +del _CudaLegacyStorage + +torch._storage_classes.add(DoubleStorage) +torch._storage_classes.add(FloatStorage) +torch._storage_classes.add(LongStorage) +torch._storage_classes.add(IntStorage) +torch._storage_classes.add(ShortStorage) +torch._storage_classes.add(CharStorage) +torch._storage_classes.add(ByteStorage) +torch._storage_classes.add(HalfStorage) +torch._storage_classes.add(BoolStorage) +torch._storage_classes.add(BFloat16Storage) +torch._storage_classes.add(ComplexDoubleStorage) +torch._storage_classes.add(ComplexFloatStorage) + + +class _WrappedTritonKernel: + """Just a simple wrapper to store some metadata for testing purposes.""" + + def __init__(self, kernel): + self.kernel = kernel + self.kernel_invoked = False + + def __call__(self, *args, **kwargs): + res = self.kernel(*args, **kwargs) + self.kernel_invoked = True + return res + + +def _register_triton_kernels(): + @_WrappedTritonKernel + def kernel_impl(*args, **kwargs): + from torch.sparse._triton_ops import bsr_dense_mm + + # pyrefly: ignore [not-callable] + return bsr_dense_mm(*args, skip_checks=True, **kwargs) + + @_WrappedTritonKernel + def addmm_kernel_impl(*args, **kwargs): + from torch.sparse._triton_ops import bsr_dense_addmm + + return bsr_dense_addmm(*args, skip_checks=True, **kwargs) + + has_triton = importlib.util.find_spec("triton") is not None + if has_triton: + torch._TritonLibrary.registerOp( + "_triton_bsr_dense_mm_out", + "_triton_bsr_dense_mm_out(Tensor bsr, Tensor dense, *, Tensor(a!) out) -> Tensor(a!)", + kernel_impl, + "SparseCsrCUDA", + ) + + torch._TritonLibrary.registerOp( + "_triton_bsr_dense_addmm_out", + ( + "_triton_bsr_dense_addmm_out(Tensor input, Tensor bsr, Tensor dense," + " *, Scalar beta, Scalar alpha, Tensor(a!) out) -> Tensor(a!)" + ), + addmm_kernel_impl, + "SparseCsrCUDA", + ) + + +_lazy_call(_register_triton_kernels) + + +def _compile_kernel( + kernel_source: str, + kernel_name: str, + compute_capability: str | None = None, + cuda_include_dirs: list | None = None, + nvcc_options: list | None = None, +): + """ + Compiles a CUDA kernel using NVRTC and returns a callable function. + + This function is a wrapper for NVRTC that enables runtime compilation of CUDA kernels. + Note that this returns a raw CUDA kernel that operates on raw memory pointers. + To use this kernel as a proper PyTorch operator, you should wrap it following the guide at: + pytorch.org/tutorials/advanced/python_custom_ops.html + + Args: + kernel_source (str): The CUDA kernel source code as a string + kernel_name (str): The name of the kernel function to compile + compute_capability (str, optional): The compute capability to target (e.g., "86"). + If None, will detect from current device. + cuda_include_dirs (list, optional): List of directories containing CUDA headers + nvcc_options (list, optional): Additional options to pass to NVRTC + + Returns: + callable: A Python function that can be called with PyTorch tensor arguments to execute the kernel + + Example: + >>> # xdoctest: +SKIP + >>> kernel_code = ''' + extern "C" + __global__ void add_tensors(const float* a, const float* b, float* c, int n) { + int i = threadIdx.x + blockIdx.x * blockDim.x; + if (i < n) + c[i] = a[i] + b[i]; + } + ''' + >>> add_kernel = torch.cuda.compile_kernel(kernel_code, "add_tensors") + >>> a = torch.randn(1024, device="cuda") + >>> b = torch.randn(1024, device="cuda") + >>> c = torch.empty_like(a) + >>> add_kernel(grid=(4, 1, 1), block=(256, 1, 1), args=[a, b, c, a.numel()]) + """ + from torch.cuda._utils import _cuda_load_module, _nvrtc_compile + + # Compile the kernel to PTX + ptx, mangled_name = _nvrtc_compile( + kernel_source, + kernel_name, + compute_capability, + cuda_include_dirs, + nvcc_options, + ) + + # Load the module and get the kernel + result = _cuda_load_module(ptx, [mangled_name]) + + if isinstance(result, dict): + return result[mangled_name] + else: + # This branch shouldn't be executed if kernel_names is provided, + # but MyPy needs this to understand type narrowing + return getattr(result, mangled_name) + + +from . import amp, graph_annotations, jiterator, nvtx, profiler, sparse, tunable + + +_POOL_HANDLE = NewType("_POOL_HANDLE", tuple[int, int]) + + +__all__ = [ + # Typed storage and tensors + "BFloat16Storage", + # pyrefly: ignore [bad-dunder-all] + "BFloat16Tensor", + "BoolStorage", + # pyrefly: ignore [bad-dunder-all] + "BoolTensor", + "ByteStorage", + # pyrefly: ignore [bad-dunder-all] + "ByteTensor", + "CharStorage", + # pyrefly: ignore [bad-dunder-all] + "CharTensor", + "ComplexDoubleStorage", + "ComplexFloatStorage", + "DoubleStorage", + # pyrefly: ignore [bad-dunder-all] + "DoubleTensor", + "FloatStorage", + # pyrefly: ignore [bad-dunder-all] + "FloatTensor", + "HalfStorage", + # pyrefly: ignore [bad-dunder-all] + "HalfTensor", + "IntStorage", + # pyrefly: ignore [bad-dunder-all] + "IntTensor", + "LongStorage", + # pyrefly: ignore [bad-dunder-all] + "LongTensor", + "ShortStorage", + # pyrefly: ignore [bad-dunder-all] + "ShortTensor", + "CUDAGraph", + "CudaError", + "DeferredCudaCallError", + "Event", + "ExternalStream", + "Stream", + "StreamContext", + "GreenContext", + "amp", + "caching_allocator_alloc", + "caching_allocator_delete", + "caching_allocator_disabled", + "caching_allocator_enable", + "can_device_access_peer", + "check_error", + "cudaStatus", + "cudart", + "current_blas_handle", + "current_solver_handle", + "current_device", + "current_stream", + "default_generators", + "default_stream", + "device", + "device_count", + "device_memory_used", + "device_of", + "empty_cache", + "export_dot", + "export_graph_data", + "get_allocator_backend", + "CUDAPluggableAllocator", + "change_current_allocator", + "get_arch_list", + "get_device_capability", + "get_device_name", + "get_device_properties", + "get_gencode_flags", + "get_per_process_memory_fraction", + "get_rng_state", + "get_rng_state_all", + "get_stream_from_external", + "get_sync_debug_mode", + "graph", + "graph_annotations", + "graph_pool_handle", + "graphs", + "has_half", + "has_magma", + "host_memory_stats", + "host_memory_stats_as_nested_dict", + "init", + "initial_seed", + "ipc_collect", + "is_available", + "is_bf16_supported", + "is_current_stream_capturing", + "is_initialized", + "is_tf32_supported", + "jiterator", + "list_gpu_processes", + "make_graphed_callables", + "manual_seed", + "manual_seed_all", + "max_memory_allocated", + "max_memory_cached", + "max_memory_reserved", + "mem_get_info", + "memory", + "memory_allocated", + "memory_cached", + "memory_reserved", + "memory_snapshot", + "memory_stats", + "memory_stats_as_nested_dict", + "memory_summary", + "memory_usage", + "MemPool", + "use_mem_pool", + "temperature", + "power_draw", + "clock_rate", + "nccl", + "nvtx", + "profiler", + "random", + "reset_accumulated_host_memory_stats", + "reset_accumulated_memory_stats", + "reset_max_memory_allocated", + "reset_max_memory_cached", + "reset_peak_host_memory_stats", + "reset_peak_memory_stats", + "seed", + "seed_all", + "set_device", + "set_per_process_memory_fraction", + "set_rng_state", + "set_rng_state_all", + "set_stream", + "set_sync_debug_mode", + "sparse", + "stream", + "streams", + "synchronize", + "tunable", + "utilization", +] diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..aef2d7a935693ebb719ef5d673f660c3e5925ff5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_annotate_cuda_graph_trace.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_annotate_cuda_graph_trace.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9eb82ff4229cb00d96176b9cee96423bc01fd239 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_annotate_cuda_graph_trace.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_device_limits.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_device_limits.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..441f08b2478137e055914c00612133dd5dbac2ea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_device_limits.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_gpu_trace.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_gpu_trace.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b7d1536dd5d1ef3cc737b7c545dd76c2af0bb36b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_gpu_trace.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_graph_annotations.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_graph_annotations.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fc60eeeedbae6cdad34deef28199ab1f83cf3da5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_graph_annotations.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_memory_viz.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_memory_viz.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9de176aaeae9d493f5e254db974892768886b506 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_memory_viz.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_pin_memory_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_pin_memory_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7019f236785bd4cfb34e691db5c6e7ac2714bb27 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_pin_memory_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_sanitizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_sanitizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..65a15af45be073cc13aa777402b95a3bbcded812 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_sanitizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..489d1dece4c74ad363929215ea801baf15847281 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/comm.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/comm.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..45b9a38000cc0bd46352a2d21f593e3c66b33047 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/comm.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/gds.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/gds.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b3bd9ab40a8d5dc393ffa186cec524f48738e5f5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/gds.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graph_annotations.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graph_annotations.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4e046261155807511873391afb5f527c1232ff06 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graph_annotations.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graphs.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graphs.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e367363676746b19ad0fbae8b3425fc401bd59bc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/graphs.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/green_contexts.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/green_contexts.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8c006f2cbed3fc4f8ac541f3d421ffae6c8b04ec Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/green_contexts.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/jiterator.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/jiterator.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8989e8b65ddc7f246517bfe2dcbe21931780009b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/jiterator.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/memory.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/memory.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..680ddc24df5b5bb870179edd3e3c0a352525ca9b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/memory.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nccl.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nccl.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9f02765b194490d57dc4fa420d8fa1bfe5273ca4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nccl.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nvtx.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nvtx.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ceef401fea77e28ce4d38b698e4e081f8aa3e946 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/nvtx.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/profiler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/profiler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..943b6c0fa77fb0f698f17321201e06813d1109d6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/profiler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/random.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/random.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3969fa4d8288f360ef5f1151fdb8a25a74cf393f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/random.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/sparse.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/sparse.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cfcdfee4fe973a5962246092f0cf8af20cebc7d4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/sparse.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/streams.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/streams.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..94c46fde3d57206c94df92d07473c0d3806ac0b8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/streams.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/tunable.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/tunable.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..176540aa99fb874b231098a092ffca5b0a212ae3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/__pycache__/tunable.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_annotate_cuda_graph_trace.py b/.venv/lib/python3.12/site-packages/torch/cuda/_annotate_cuda_graph_trace.py new file mode 100644 index 0000000000000000000000000000000000000000..fc5f3a84b69387097cf357b2633beb8e263d8e9e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_annotate_cuda_graph_trace.py @@ -0,0 +1,365 @@ +#!/usr/bin/env python3 +"""Post-process a profiler trace to add CUDA graph kernel annotations. + +Reads a profiler trace (gzipped or plain JSON) and a kernel annotations +pickle, matches kernel events by their graph node id, and writes an +annotated trace with the annotation fields added to each kernel event's +args (displayed alongside grid/block size in trace viewers). + +The annotations pickle is auto-discovered from the trace file's parent +directory (one level up, matching the rank from the trace filename). + +Usage: + python -m torch.cuda._annotate_cuda_graph_trace [-a ] [-o ] + +Examples: + # Auto-discover annotations pickle from trace location + python -m torch.cuda._annotate_cuda_graph_trace \\ + traces/step_000000000014/000000.*.pt.trace.json.gz + + # Explicit annotations pickle + python -m torch.cuda._annotate_cuda_graph_trace trace.json.gz -a annotations.pkl +""" + +import argparse +import gzip +import json +import pickle +import re +import sys +from collections import defaultdict +from pathlib import Path +from typing import Any + + +_WORK_CATEGORIES = {"kernel", "gpu_memcpy", "gpu_memset"} + + +def _move_overlapping_to_stream( + trace: dict, default_stream: int = 7, overlap_stream: int = 8 +) -> int: + """Move graphed kernels that overlap with their predecessor to a separate stream. + + Perfetto cannot display overlapping (non-nested) events on the same + stream -- they get hidden. This pass detects graphed kernel events on + *default_stream* whose start timestamp falls before the previous + kernel's end, and moves them to *overlap_stream* so they're visible. + + Returns the number of events moved. + """ + graphed_on_default = [ + e + for e in trace["traceEvents"] + if e.get("cat") == "kernel" + and e.get("tid") == default_stream + and e.get("args", {}).get("graph node id", 0) != 0 + ] + graphed_on_default.sort(key=lambda e: e["ts"]) + + moved = 0 + prev_end = 0.0 + for event in graphed_on_default: + ts = event["ts"] + dur = event.get("dur", 0) + if ts < prev_end: + event["tid"] = overlap_stream + event.get("args", {})["stream"] = overlap_stream + moved += 1 + else: + prev_end = ts + dur + + return moved + + +def _fix_overlapping_timestamps(trace: dict, max_adjust_us: float = 1.0) -> int: + """Clamp graphed kernel/memcpy timestamps so they don't overlap on the same stream. + + CUPTI can produce slightly overlapping timestamps for consecutive graphed + events, causing Perfetto to hide events that sit entirely "under" their + neighbours. This pass sorts graphed work events per stream and ensures + each event starts at or after the previous event's end. + + Overlaps larger than *max_adjust_us* are flagged as warnings and left + unchanged, since they likely indicate a real issue rather than CUPTI + timestamp jitter. + + Returns the number of events adjusted. + """ + per_stream: dict[int, list[dict]] = defaultdict(list) + for event in trace["traceEvents"]: + if ( + event.get("cat") in _WORK_CATEGORIES + and event.get("args", {}).get("graph node id", 0) != 0 + ): + per_stream[event.get("tid")].append(event) + + adjusted = 0 + for tid, events in per_stream.items(): + events.sort(key=lambda e: e["ts"]) + prev_end = 0.0 + for event in events: + ts = event["ts"] + dur = event.get("dur", 0) + if ts < prev_end: + overlap = prev_end - ts + if overlap > max_adjust_us: + print( + f"WARNING: large overlap {overlap:.3f}us on stream {tid} " + f"for {event.get('name', '?')[:60]}, skipping adjustment" + ) + else: + event["ts"] = prev_end + adjusted += 1 + prev_end = event["ts"] + dur + + return adjusted + + +def annotate_trace( + trace: dict, + annotations: dict[int, list[Any]], + default_stream: int = 7, +) -> int: + """Add annotation fields to kernel events matching the annotations dict. + + Each annotation entry is a list (from nested ``mark_kernels`` scopes). + Fields from all annotations are merged into the event args; if multiple + annotations define the same key, later entries in the list win. + + For graphed events (graph_node_id != 0), reassigns ``tid`` and + ``args["stream"]`` to the stream recorded in annotations, or to + *default_stream* if there is no annotation. Also moves the + corresponding ``ac2g`` flow-finish events to the new tid so that + CPU-to-GPU correlation arrows are preserved. + + Removes ``gpu_user_annotation`` events and orphaned ``ac2g`` events + from streams that have no kernel or memcpy events after reassignment, + since CUPTI replicates these onto every stream during graph replay. + + Returns the number of events annotated. + """ + # Build an index of ac2g 'f' events keyed by (tid, ts) so we can + # move them together with the kernel events they correspond to. + ac2g_f_index: dict[tuple, list] = {} + for event in trace["traceEvents"]: + if event.get("cat") == "ac2g" and event.get("ph") == "f": + key = (event.get("tid"), event.get("ts")) + ac2g_f_index.setdefault(key, []).append(event) + + annotated = 0 + for event in trace.get("traceEvents", []): + args = event.get("args", {}) + graph_node_id = args.get("graph node id") + if graph_node_id is None or graph_node_id == 0: + continue + stream_id = None + if graph_node_id in annotations: + for ann in annotations[graph_node_id]: + if isinstance(ann, dict): + for key, value in ann.items(): + # Legacy pickles wrapped bare strings as {"str": ...}; + # the canonical key is "name". + args["name" if key == "str" else key] = str(value) + if "stream" in ann: + stream_id = int(ann["stream"]) + else: + # Legacy pickles stored bare strings unwrapped. + args["name"] = str(ann) + annotated += 1 + + # Reassign stream: use annotated stream if available, else default + if stream_id is None: + stream_id = default_stream + old_key = (event.get("tid"), event.get("ts")) + event["tid"] = stream_id + args["stream"] = stream_id + + # Move the matching ac2g 'f' event(s) to the same new tid + for ac2g_event in ac2g_f_index.get(old_key, ()): + ac2g_event["tid"] = stream_id + + # Remove gpu_user_annotation events and ac2g flow-finish events from + # streams that have no real kernel/memcpy/memset work -- these are + # noise replicated by CUPTI onto every stream during graph replay. + tids_with_work = set() + for event in trace["traceEvents"]: + if event.get("cat") in _WORK_CATEGORIES: + tids_with_work.add(event.get("tid")) + + def _is_noise(event: dict) -> bool: + cat = event.get("cat") + if cat == "gpu_user_annotation": + return event.get("tid") not in tids_with_work + if cat == "ac2g" and event.get("ph") == "f": + return event.get("tid") not in tids_with_work + return False + + original_count = len(trace["traceEvents"]) + trace["traceEvents"] = [ + event for event in trace["traceEvents"] if not _is_noise(event) + ] + removed = original_count - len(trace["traceEvents"]) + if removed: + print(f"Removed {removed} noise events from empty streams") + + # Clean up metadata: remove thread_name / thread_sort_index entries + # for noise streams that have no real (non-metadata) events, and add + # thread_name entries for our new annotation streams. + all_tids_in_trace = { + e.get("tid") for e in trace["traceEvents"] if e.get("ph") != "M" + } + # Find the GPU process pid from existing thread_name metadata + gpu_pid = 0 + for event in trace["traceEvents"]: + if ( + event.get("ph") == "M" + and event.get("name") == "thread_name" + and str(event.get("args", {}).get("name", "")).startswith("stream ") + ): + gpu_pid = event.get("pid", 0) + break + + # Remove metadata entries for tids with no non-metadata events + trace["traceEvents"] = [ + event + for event in trace["traceEvents"] + if event.get("ph") != "M" or event.get("tid") in all_tids_in_trace + ] + + # Add thread_name metadata for new annotation tids that lack one + existing_thread_names = { + e.get("tid") + for e in trace["traceEvents"] + if e.get("ph") == "M" and e.get("name") == "thread_name" + } + for tid in sorted(tids_with_work - existing_thread_names): + trace["traceEvents"].append( + { + "ph": "M", + "pid": gpu_pid, + "tid": tid, + "name": "thread_name", + "args": {"name": f"stream {tid}"}, + } + ) + + return annotated + + +def load_trace(path: Path) -> dict: + if path.suffix == ".gz" or path.name.endswith(".json.gz"): + with gzip.open(path, "rt") as f: + return json.load(f) + else: + with open(path) as f: + return json.load(f) + + +def save_trace(trace: dict, path: Path) -> None: + if path.suffix == ".gz" or path.name.endswith(".json.gz"): + with gzip.open(path, "wt") as f: + json.dump(trace, f) + else: + with open(path, "w") as f: + json.dump(trace, f) + + +def _find_annotations_pkl(trace_file: Path) -> Path | None: + """Auto-discover the annotations pickle from the trace file location. + + Trace files live in e.g. ``traces/step_000000000014/000000..pt.trace.json.gz`` + where the leading digits are the rank. The pickle lives one level up: + ``traces/kernel_annotations_rank0_*.pkl``. + """ + match = re.match(r"^(\d+)", trace_file.name) + if not match: + return None + rank = int(match.group(1)) + + traces_dir = trace_file.parent.parent + candidates = sorted(traces_dir.glob(f"kernel_annotations_rank{rank}_*.pkl")) + if candidates: + return candidates[0] + return None + + +def main() -> None: + parser = argparse.ArgumentParser( + description="Annotate a profiler trace with CUDA graph kernel annotations." + ) + parser.add_argument( + "trace_file", type=Path, help="Input trace file (.json or .json.gz)" + ) + parser.add_argument( + "-a", + "--annotations", + type=Path, + default=None, + help="Kernel annotations pickle file. Auto-discovered if omitted.", + ) + parser.add_argument( + "-o", + "--output", + type=Path, + default=None, + help="Output file path. Defaults to .annotated.", + ) + parser.add_argument( + "--default-stream", + type=int, + default=7, + help="Stream ID to assign to unannotated graphed events (default: 7).", + ) + args = parser.parse_args() + + annotations_pkl = args.annotations + if annotations_pkl is None: + annotations_pkl = _find_annotations_pkl(args.trace_file) + if annotations_pkl is None: + print( + f"Could not auto-discover annotations pickle for {args.trace_file}. " + f"Use -a to specify it explicitly.", + file=sys.stderr, + ) + sys.exit(1) + print(f"Auto-discovered annotations: {annotations_pkl}") + + with open(annotations_pkl, "rb") as f: + annotations = pickle.load(f) + print(f"Loaded {len(annotations)} kernel annotations") + + trace = load_trace(args.trace_file) + total_events = len(trace.get("traceEvents", [])) + print(f"Loaded trace with {total_events} events") + + count = annotate_trace(trace, annotations, default_stream=args.default_stream) + print(f"Annotated {count} kernel events") + + overlap_moved = _move_overlapping_to_stream( + trace, default_stream=args.default_stream + ) + if overlap_moved: + print(f"Moved {overlap_moved} overlapping events to stream 8") + + ts_fixed = _fix_overlapping_timestamps(trace) + if ts_fixed: + print(f"Fixed {ts_fixed} overlapping graphed event timestamps") + + output = args.output + if output is None: + name = args.trace_file.name + if name.endswith(".json.gz"): + output = args.trace_file.with_name( + name.replace(".json.gz", ".annotated.json.gz") + ) + elif name.endswith(".json"): + output = args.trace_file.with_suffix(".annotated.json") + else: + output = args.trace_file.with_suffix(args.trace_file.suffix + ".annotated") + + save_trace(trace, output) + print(f"Saved annotated trace to {output}") + + +if __name__ == "__main__": + main() diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_device_limits.py b/.venv/lib/python3.12/site-packages/torch/cuda/_device_limits.py new file mode 100644 index 0000000000000000000000000000000000000000..4c4218661f40df45b8429b1f5a7b77b25a858213 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_device_limits.py @@ -0,0 +1,140 @@ +import torch +from torch._C import dtype + + +__all__ = ["GPULimits"] + + +class GPULimits: + r"""Utility class that provides the theoretical limits of Nvidia GPU devices. The + limits don't take into account thermal throttling (assume that the GPU run at its + peak rated frequency). This is because user hardware configuration may influence + power behavior. + """ + + def __init__(self, target_device: torch.device): + # The device properties object is obtained by calling 'cudaGetDeviceProperties' CUDA + # runtime function. We need the total memory bus width and the memory clock rate to + # calculate the memory bandwidth. + self.device_properties = torch.cuda.get_device_properties(target_device) + + # The compute capability is needed to determine the number of FLOPs per cycle per SM + self.compute_capability = int( + f"{self.device_properties.major}{self.device_properties.minor}" + ) + + # FLOPs per cycle information derived from Table 2 in: + # https://resources.nvidia.com/en-us-hopper-architecture/nvidia-h100-tensor-c + + # Returns the number of FMA instructions retired per cycle per SM for a given + # data type, when tensor cores are NOT used + def get_fma_per_cycle_per_sm_cuda_cores(self, data_type: dtype) -> int: + hardcoded_device_values = { + # Ampere Architecture + "fp16_80": 256, + "fp32_80": 64, + "fp64_80": 32, + # Hopper Architecture + "fp16_90": 64, + "fp32_90": 128, + "fp64_90": 64, + # Blackwell Architecture + "fp16_100": 256, + "fp32_100": 128, + "fp64_100": 64, + } + dict_key = "" + if data_type is torch.float16: + dict_key = f"fp16_{self.compute_capability}" + elif data_type is torch.float32: + dict_key = f"fp32_{self.compute_capability}" + elif data_type is torch.float64: + dict_key = f"fp64_{self.compute_capability}" + else: + dict_key = "unknown" + + if dict_key not in hardcoded_device_values: + raise RuntimeError( + f"No data for sm_{self.compute_capability} and {data_type}." + ) + + return hardcoded_device_values[dict_key] + + # Returns the number of FMA instructions retired per cycle per SM for a given + # data type, when tensor cores ARE used + def get_fma_per_cycle_per_sm_tensor_cores(self, data_type: dtype) -> int: + hardcoded_device_values = { + # Ampere Architecture + "int8_80": 2048, + "fp16_80": 1024, + "fp32_80": 512, + "fp64_80": 64, + # Hopper Architecture + "int8_90": 4096, + "fp8_90": 4096, + "fp16_90": 2048, + "fp32_90": 1024, + "fp64_90": 128, + # Blackwell Architecture + "int8_100": 8192, + "fp8_100": 8192, + "fp16_100": 4096, + "fp32_100": 2048, + } + dict_key = "" + if data_type is torch.float16: + dict_key = f"fp16_{self.compute_capability}" + elif data_type is torch.bfloat16: + # FP16 and BF16 are equivalent in terms of FLOPs per cycle per SM + dict_key = f"fp16_{self.compute_capability}" + elif data_type is torch.float32: + dict_key = f"fp32_{self.compute_capability}" + elif data_type is torch.int8: + dict_key = f"int8_{self.compute_capability}" + elif data_type is torch.float64: + dict_key = f"fp64_{self.compute_capability}" + else: + dict_key = "unknown" + + if dict_key not in hardcoded_device_values: + raise RuntimeError( + f"No data for sm_{self.compute_capability} and {data_type}." + ) + + return hardcoded_device_values[dict_key] + + def get_tflops_per_second( + self, data_type: dtype, use_tensor_cores: bool = True + ) -> float: + num_sms = self.device_properties.multi_processor_count + clock_rate = self.device_properties.clock_rate # KHz + + fma_per_cycle = 0 + if use_tensor_cores: + fma_per_cycle = self.get_fma_per_cycle_per_sm_tensor_cores(data_type) + else: + fma_per_cycle = self.get_fma_per_cycle_per_sm_cuda_cores(data_type) + + # 1 FMA counts as 2 floating point operations + # Clock rate is in KHz + tflops_per_second = num_sms * fma_per_cycle * 2 * clock_rate / 1e9 + return tflops_per_second + + def get_memory_bandwidth_Bps(self) -> int: + # DRAM devices are Double-Data which means they provide an output at both fronts of + # a clock beat + bus_bytes_per_cycle = int(2 * self.device_properties.memory_bus_width / 8) + mem_clock_rate_Hz = self.device_properties.memory_clock_rate * 1000 + bytes_per_second = bus_bytes_per_cycle * mem_clock_rate_Hz + return bytes_per_second + + def get_shared_memory_bandwidth_Bps(self) -> int: + # Each warp can LD or ST 32 x 4 bytes per cycle. To calculate the + # device's throughput we need to multiply with frequency and number of SMs. + num_sms = self.device_properties.multi_processor_count + bytes_per_cycle_per_sm = 128 + bytes_per_cycle_per_device = num_sms * bytes_per_cycle_per_sm + bytes_per_second = ( + bytes_per_cycle_per_device * self.device_properties.clock_rate * 1000 + ) + return bytes_per_second diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_gpu_trace.py b/.venv/lib/python3.12/site-packages/torch/cuda/_gpu_trace.py new file mode 100644 index 0000000000000000000000000000000000000000..d3b8f7e4626f9bec478a7c74292f78894d62ee6f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_gpu_trace.py @@ -0,0 +1,73 @@ +from collections.abc import Callable + +from torch._utils import CallbackRegistry + + +EventCreationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA event creation" +) +EventDeletionCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA event deletion" +) +EventRecordCallbacks: "CallbackRegistry[int, int]" = CallbackRegistry( + "CUDA event record" +) +EventWaitCallbacks: "CallbackRegistry[int, int]" = CallbackRegistry("CUDA event wait") +MemoryAllocationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA memory allocation" +) +MemoryDeallocationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA memory deallocation" +) +StreamCreationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA stream creation" +) +DeviceSynchronizationCallbacks: "CallbackRegistry[[]]" = CallbackRegistry( + "CUDA device synchronization" +) +StreamSynchronizationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA stream synchronization" +) +EventSynchronizationCallbacks: "CallbackRegistry[int]" = CallbackRegistry( + "CUDA event synchronization" +) + + +def register_callback_for_event_creation(cb: Callable[[int], None]) -> None: + EventCreationCallbacks.add_callback(cb) + + +def register_callback_for_event_deletion(cb: Callable[[int], None]) -> None: + EventDeletionCallbacks.add_callback(cb) + + +def register_callback_for_event_record(cb: Callable[[int, int], None]) -> None: + EventRecordCallbacks.add_callback(cb) + + +def register_callback_for_event_wait(cb: Callable[[int, int], None]) -> None: + EventWaitCallbacks.add_callback(cb) + + +def register_callback_for_memory_allocation(cb: Callable[[int], None]) -> None: + MemoryAllocationCallbacks.add_callback(cb) + + +def register_callback_for_memory_deallocation(cb: Callable[[int], None]) -> None: + MemoryDeallocationCallbacks.add_callback(cb) + + +def register_callback_for_stream_creation(cb: Callable[[int], None]) -> None: + StreamCreationCallbacks.add_callback(cb) + + +def register_callback_for_device_synchronization(cb: Callable[[], None]) -> None: + DeviceSynchronizationCallbacks.add_callback(cb) + + +def register_callback_for_stream_synchronization(cb: Callable[[int], None]) -> None: + StreamSynchronizationCallbacks.add_callback(cb) + + +def register_callback_for_event_synchronization(cb: Callable[[int], None]) -> None: + EventSynchronizationCallbacks.add_callback(cb) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_graph_annotations.py b/.venv/lib/python3.12/site-packages/torch/cuda/_graph_annotations.py new file mode 100644 index 0000000000000000000000000000000000000000..a383012265db8fdd6aa7cb6fe4770cc9d54586e6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_graph_annotations.py @@ -0,0 +1,1015 @@ +"""Annotate CUDA graph kernel nodes during capture. + +During CUDA graph capture, ``mark_kernels`` records the current capture +frontier and the direct dependents already attached to that frontier. +On scope exit it walks only the newly added dependent edges to find the +nodes created within the scope. Each kernel, memcpy, or memset node found +is annotated by its ``toolsId`` so it can later be matched to profiler +trace events. + +``mark_kernels`` now snapshots capture state from whatever stream is +current on scope entry, so that stream must already be participating in +the capture. ``mark_stream`` handles this by starting ``mark_kernels`` +before switching to the target stream. + +The annotations can be pickled and later merged into a Chrome profiler +trace using ``torch.cuda._annotate_cuda_graph_trace``. + +Requires ``cuda.bindings`` package and a CUDA driver that supports +``cudaGraphNodeGetToolsId`` (CUDA >= 13.1 or appropriate cuda-compat). +When unavailable, ``mark_kernels`` silently becomes a no-op. + +Annotation recording is enabled per capture via ``torch.cuda.graph``'s +``enable_annotations`` argument, which also resolves and remaps the recorded +annotations to the exec graph automatically on context exit. + +Usage during capture:: + + from torch.cuda._graph_annotations import mark_kernels, get_kernel_annotations + + with torch.cuda.graph(graph, enable_annotations=True): + with mark_kernels("phase_A"): + y = workload_a(x) + with mark_kernels("phase_B"): + z = workload_b(y) + + annotations = get_kernel_annotations() + +When you need to drive this outside the context manager's automatic path, +``resolve_and_remap(graph)`` is shorthand for ``resolve_pending_annotations()`` +followed by ``remap_to_exec_graph(graph)``; call those directly for finer +control (e.g. resolving once before remapping several graphs). +""" + +from __future__ import annotations + +import importlib.metadata +import threading +import warnings +from collections import defaultdict +from contextlib import contextmanager +from logging import getLogger +from typing import Any, NamedTuple, TYPE_CHECKING, TypeAlias + + +if TYPE_CHECKING: + from collections.abc import Iterable, Mapping + +import torch +from torch.cuda._utils import ( + _check_cuda_bindings, + _check_cuda_bindings_driver, + _HAS_CUDA_BINDINGS, +) + + +try: + from cuda.bindings import ( # pyrefly: ignore[missing-import] + driver as _cuda_driver, + runtime as _cuda_runtime, + ) +except ImportError: + _cuda_driver = None # type: ignore[assignment] + _cuda_runtime = None # type: ignore[assignment] + + +logger = getLogger(__name__) + + +_CaptureState: TypeAlias = tuple[Any, list[Any]] +_ExistingDirectDependents: TypeAlias = dict[int, set[int]] + + +# Tri-state: None = not probed, True = available, False = unavailable. +# Deferred to first use to avoid premature CUDA initialization. +_tools_id_available: bool | None = None + +# Whether annotation recording is active. Scoped to a ``torch.cuda.graph`` +# capture: set by the context manager on entry (from its ``enable_annotations`` +# argument) and cleared on exit. When False, mark_kernels/mark_stream and the +# capture-id stamp are no-ops. Module-level (not thread-local) because capture +# can span threads (e.g. autograd). +_annotations_enabled: bool = False + + +# Id of the top-level capture graph of the active capture. A conditional node's body is +# captured into a separate cudaGraph_t (see CUDAGraph::begin_capture_to_conditional_node), +# and its node ids live in that graph's id space, which remap_to_exec_graph never rekeys -- +# so mark_kernels compares against this to detect a scope inside a body. +_capture_root_graph_id: int | None = None + + +def _set_annotations_enabled(enabled: bool) -> None: + """Set whether annotation recording is active. Used by ``torch.cuda.graph`` + to scope annotations to a capture; not a public API.""" + global _annotations_enabled, _capture_root_graph_id + _annotations_enabled = enabled + if not enabled: + _capture_root_graph_id = None + + +def _graph_id(graph: Any) -> int: + """The driver's unique id for a ``cudaGraph_t``. + + Identifies a graph across its whole lifetime, unlike the handle itself: that is a + pointer, so a graph created after another is destroyed can reuse the value and + compare equal to it.""" + return _check_cuda_bindings( + _cuda_runtime.cudaGraphGetId(graph) # pyrefly: ignore[missing-attribute] + ) + + +def maybe_stamp_capture_root(stream: Any) -> None: + """Record the top-level capture graph of the capture just begun on ``stream``. + + Called by ``torch.cuda.graph`` right after ``capture_begin``, while ``stream`` is + still capturing into the top-level graph. Not a public API.""" + global _capture_root_graph_id + _capture_root_graph_id = None + if not _annotations_enabled or _is_tools_id_unavailable(): + return + stream_handle = _cuda_runtime.cudaStream_t( # pyrefly: ignore[missing-attribute] + init_value=stream.cuda_stream + ) + state = _get_capture_state(stream_handle) + if state is not None: + _capture_root_graph_id = _graph_id(state[0]) + + +def _probe_tools_id() -> bool: + """Probe whether cudaGraphNodeGetToolsId is supported by the driver. + + Calls with a null node: cudaErrorInvalidValue means the API exists + in the driver (good), cudaErrorCallRequiresNewerDriver means it + does not (bad). + """ + if not hasattr(_cuda_runtime, "cudaGraphNodeGetToolsId"): + # API is missing from cuda-bindings - likely version too old + cuda_bindings_version = importlib.metadata.version("cuda-bindings") + + logger.warning( + "cudaGraphNodeGetToolsId API not found in cuda-bindings. " + "Current version: %s, required: >= 13.1.0. " + "CUDA graph kernel annotations will be disabled. " + "To enable annotations, upgrade cuda-bindings: " + "pip install --upgrade cuda-bindings", + cuda_bindings_version, + ) + return False + err, *_ = _cuda_runtime.cudaGraphNodeGetToolsId( + 0 + ) # pyrefly: ignore[missing-attribute] + if ( + err + == _cuda_runtime.cudaError_t.cudaErrorCallRequiresNewerDriver # pyrefly: ignore[missing-attribute] + ): + logger.info( + "cudaGraphNodeGetToolsId requires a newer driver " + "(missing cuda-compat?); " + "CUDA graph kernel annotations will be disabled" + ) + return False + return True + + +def _is_tools_id_unavailable() -> bool: + """Return True if cudaGraphNodeGetToolsId is not usable.""" + global _tools_id_available + if not _HAS_CUDA_BINDINGS: + return True + if _tools_id_available is not None: + return not _tools_id_available + _tools_id_available = _probe_tools_id() + return not _tools_id_available + + +def is_available() -> bool: + r"""is_available() -> bool + + Return whether CUDA graph annotation recording is supported. + + Requires a CUDA device, the ``cuda-bindings`` package, and a driver + that supports ``cudaGraphNodeGetToolsId`` (CUDA >= 13.1 or an + equivalent cuda-compat package). When this returns ``False``, + :func:`mark_kernels` is a silent no-op and no annotations are + recorded. + + The first call may probe the CUDA driver; the result is cached. + """ + return torch.cuda.is_available() and not _is_tools_id_unavailable() + + +def _get_capture_state(stream: Any) -> _CaptureState | None: + """Return ``(graph, frontier)`` for an active capture, else ``None``.""" + status, _id, graph, _deps, _edge_data, _num_deps = _check_cuda_bindings( + _cuda_runtime.cudaStreamGetCaptureInfo( # pyrefly: ignore[missing-attribute] + stream + ) + ) + if ( + status + != _cuda_runtime.cudaStreamCaptureStatus.cudaStreamCaptureStatusActive # pyrefly: ignore[missing-attribute] + ): + return None + return graph, list(_deps[:_num_deps]) + + +def _get_root_nodes(graph: Any) -> list[Any]: + """Return the current root nodes for the graph.""" + _, num_roots = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetRootNodes( # pyrefly: ignore[missing-attribute] + graph + ) + ) + if num_roots == 0: + return [] + roots, num_roots = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetRootNodes( # pyrefly: ignore[missing-attribute] + graph, pNumRootNodes=num_roots + ) + ) + return list(roots[:num_roots]) + + +def _get_dependent_nodes(node: Any) -> list[Any]: + """Return the direct dependents of a graph node.""" + _, _, num_dependents = _check_cuda_bindings( + _cuda_runtime.cudaGraphNodeGetDependentNodes( # pyrefly: ignore[missing-attribute] + node + ) + ) + if num_dependents == 0: + return [] + dependents, _edge_data, num_dependents = _check_cuda_bindings( + _cuda_runtime.cudaGraphNodeGetDependentNodes( # pyrefly: ignore[missing-attribute] + node, pNumDependentNodes=num_dependents + ) + ) + return list(dependents[:num_dependents]) + + +def _get_node_type(node: Any) -> Any: + """Return graph node type without tripping runtime bugs on newer node kinds. + + The runtime ``cudaGraphNodeGetType`` can return ``cudaErrorUnknown`` (999) + for valid nodes whose driver type is ``CU_GRAPH_NODE_TYPE_BATCH_MEM_OP``. + Query via the driver API instead. + """ + return _check_cuda_bindings_driver( + _cuda_driver.cuGraphNodeGetType(node) # pyrefly: ignore[missing-attribute] + ) + + +def _collect_descendants( + start_nodes: list[Any], + *, + existing_direct_dependents: _ExistingDirectDependents | None = None, + include_start_nodes: bool = False, +) -> dict[int, Any]: + """Walk dependent edges starting at ``start_nodes``. + + ``existing_direct_dependents`` maps each node in ``start_nodes`` to + the direct dependent node keys that were already present at scope + entry. Those edges are skipped so the traversal only follows nodes + added after scope entry. + """ + existing_direct_dependents = existing_direct_dependents or {} + seen = {int(node) for node in start_nodes} + descendants: dict[int, Any] = {} + stack = list(start_nodes) + + if include_start_nodes: + for node in start_nodes: + descendants[int(node)] = node + + while stack: + node = stack.pop() + old_dependents = existing_direct_dependents.get(int(node), set()) + for dependent in _get_dependent_nodes(node): + dependent_key = int(dependent) + if dependent_key in old_dependents or dependent_key in seen: + continue + seen.add(dependent_key) + descendants[dependent_key] = dependent + stack.append(dependent) + + return descendants + + +# toolsId -> list of annotation objects. +_kernel_annotations: defaultdict[int, list[Any]] = defaultdict(list) + +# Node types we annotate (kernels, memcpys, memsets, batch mem ops, event +# record/wait nodes, and host nodes), as driver node-type enums. Event and host +# nodes are stream-ordered but do not occupy the device; annotating them lets the +# profiler place their spans on the intended stream lane. Initialized lazily to +# avoid touching cuda.bindings at import time. +_ANNOTATABLE_TYPES: set[Any] | None = None + + +def _get_annotatable_types() -> set[Any]: + global _ANNOTATABLE_TYPES + if _ANNOTATABLE_TYPES is None: + node_types = _cuda_driver.CUgraphNodeType # pyrefly: ignore[missing-attribute] + _ANNOTATABLE_TYPES = { + node_types.CU_GRAPH_NODE_TYPE_KERNEL, + node_types.CU_GRAPH_NODE_TYPE_MEMCPY, + node_types.CU_GRAPH_NODE_TYPE_MEMSET, + node_types.CU_GRAPH_NODE_TYPE_BATCH_MEM_OP, + node_types.CU_GRAPH_NODE_TYPE_EVENT_RECORD, + node_types.CU_GRAPH_NODE_TYPE_WAIT_EVENT, + node_types.CU_GRAPH_NODE_TYPE_HOST, + } + return _ANNOTATABLE_TYPES + + +# Node types whose work lives in a separate cudaGraph_t (child graphs, conditional +# bodies). The dependent-edge walk does not descend into such a node, and the nodes +# inside are numbered in the body graph's id space, which remap_to_exec_graph never +# rekeys -- so annotations there would be silently lost. See mark_kernels. Initialized +# lazily, like _ANNOTATABLE_TYPES above: _cuda_driver is None when cuda.bindings is +# absent, so reading the enum at import time would break `import torch`. +_NESTED_GRAPH_TYPES: set[Any] | None = None + + +def _get_nested_graph_types() -> set[Any]: + global _NESTED_GRAPH_TYPES + if _NESTED_GRAPH_TYPES is None: + node_types = _cuda_driver.CUgraphNodeType # pyrefly: ignore[missing-attribute] + _NESTED_GRAPH_TYPES = { + node_types.CU_GRAPH_NODE_TYPE_GRAPH, + node_types.CU_GRAPH_NODE_TYPE_CONDITIONAL, + } + return _NESTED_GRAPH_TYPES + + +# Pending scopes: (annotation, toolsIds discovered for the scope). +# +# Notably, this is NOT the current dynamic scope; instead, we fill this in +# after we exit each mark_kernels region, and this holds the FULL extra +# CUDA graph node attribution, steadily growing as we execute. +# +# Scopes are recorded as we exit a `mark_kernels`. Later, when we resolve +# all of the annotations, we do merges on the annotations where +# first-recorded-wins (it works: you end up preferring inner scopes over +# outer scopes, what you'd expect). For example: +# +# with mark_kernels({"name": "outer", "color": "red"}): +# with mark_kernels({"name": "inner"}): +# y = x + 1 # toolsId: 101 +# z = y * 2 # toolsId: 102 +# +# where the add captures kernel node with toolsId 101 and the mul 102. After +# executing ALL of this code, the _pending_scopes would be (inner exits first, +# so it is recorded first): +# +# ({"name": "inner"}, [101]) +# ({"name": "outer", "color": "red"}, [101, 102]) +# +# toolsId 101 appears in both lists, and so the first-recorded-wins rule then +# means we merge all the dicts to {"name": "inner", "color": "red"}. +_pending_scopes: list[tuple[Any, list[int]]] = [] + + +# TLS slot carrying the current annotation region: the annotations of every +# open backward-annotating ``mark_kernels`` scope collapsed into one dict +# (inner scopes win common keys). It lives in ThreadLocalPythonObjects rather +# than a ContextVar or threading.local because that is the one Python-writable +# TLS that ``at::ThreadLocalState`` snapshots into autograd engine worker and +# device threads, so a backward bracket can re-establish the region for nodes +# created while it executes (higher-order grad, checkpoint recomputation) even +# across engine thread hops. Region dicts are never mutated after creation, so +# holding a reference is a snapshot. +_REGION_TLS_KEY = "cuda_graph_annotation_region" + + +def _current_region() -> dict[str, Any] | None: + if torch._C._is_key_in_tls(_REGION_TLS_KEY): + return torch._C._get_obj_in_tls(_REGION_TLS_KEY) + return None + + +def _enter_region(collapsed: dict[str, Any]) -> dict[str, Any] | None: + """Install ``collapsed`` as the current region; return the previous one.""" + prev = _current_region() + torch._C._stash_obj_in_tls(_REGION_TLS_KEY, collapsed) + return prev + + +def _exit_region(prev: dict[str, Any] | None) -> None: + if prev is None: + torch._C._remove_obj_from_tls(_REGION_TLS_KEY) + else: + torch._C._stash_obj_in_tls(_REGION_TLS_KEY, prev) + + +# node.metadata marker recording that a node's backward hook pair is already +# installed, so redundant firings of ``_freeze_region_hook`` (one per live +# push, e.g. nested scopes) attach only once. An identity sentinel so no +# other metadata user can collide with it. +_HOOKED_KEY: Any = object() + +# Bumped by clear_kernel_annotations. Node hooks record only while the +# generation they were created under is current, so clearing revokes +# recording from every scope opened before the clear (the hooks stay on +# the graph but become inert) without mutating live autograd graphs. +_annotation_generation: int = 0 + + +class _KernelScope(NamedTuple): + """Capture-frontier snapshot taken at scope entry, consumed at scope exit.""" + + graph: Any + frontier: list[Any] + entry_root_keys: set[int] | None + entry_direct_dependents: _ExistingDirectDependents + + +def _begin_kernel_scope() -> _KernelScope | None: + """Snapshot the current stream's capture frontier; ``None`` if not capturing.""" + stream = _cuda_runtime.cudaStream_t( # pyrefly: ignore[missing-attribute] + init_value=torch.cuda.current_stream().cuda_stream + ) + capture_state = _get_capture_state(stream) + if capture_state is None: + return None + graph, frontier = capture_state + + entry_root_keys: set[int] | None = None + entry_direct_dependents = { + int(node): {int(dep) for dep in _get_dependent_nodes(node)} for node in frontier + } + if not frontier: + entry_root_keys = {int(node) for node in _get_root_nodes(graph)} + return _KernelScope(graph, frontier, entry_root_keys, entry_direct_dependents) + + +def _end_kernel_scope(scope: _KernelScope) -> list[int]: + """Walk nodes captured since ``scope`` was begun; return their toolsIds.""" + if scope.frontier: + scope_nodes = _collect_descendants( + scope.frontier, + existing_direct_dependents=scope.entry_direct_dependents, + ) + else: + new_roots = [ + node + for node in _get_root_nodes(scope.graph) + if int(node) not in (scope.entry_root_keys or set()) + ] + scope_nodes = _collect_descendants(new_roots, include_start_nodes=True) + + annotatable = _get_annotatable_types() + nested = _get_nested_graph_types() + nested_seen: set[str] = set() + tools_ids: list[int] = [] + for node in scope_nodes.values(): + node_type = _get_node_type(node) + if node_type in nested: + nested_seen.add(node_type.name) + if node_type not in annotatable: + continue + tools_ids.append( + _check_cuda_bindings( + _cuda_runtime.cudaGraphNodeGetToolsId( # pyrefly: ignore[missing-attribute] + node + ) + ) + ) + + if nested_seen: + # The annotations recorded above are still correct -- the exec graph preserves + # top-level node ids -- they just do not cover the nested body. + warnings.warn( + f"mark_kernels: this scope contains {sorted(nested_seen)} nodes; the work " + "inside them is in a separate cudaGraph_t that this walk does not descend " + "into, so it is left unannotated", + stacklevel=4, + ) + return tools_ids + + +class _BracketState(threading.local): + """Per-thread stack of open backward brackets for one hook pair. + + Thread-local so concurrent executions of a retained node (separate + graph tasks on different engine threads) cannot interleave each + other's scope snapshots; within one thread pre/post pair up LIFO. + + If the node throws, its posthook never runs (the engine's + call_function has no try/finally around fn) and the prehook's entry + is left on the stack. That is harmless. Say a retained node runs + three times on one thread and the second run throws: + + run 1: prehook push [e1]; posthook pops e1 -> [] + run 2: prehook push [e2]; node throws, no pop -> [e2] + run 3: prehook push [e2, e3]; posthook pops e3 -> [e2] + + Pops always take the top, so the leaked e2 sits below all later + entries and every subsequent run still pops its own. The only loss + is e2 itself: that run's kernels go unrecorded, which is moot since + its backward failed. (The region/creation-hook TLS the prehook set + is not leaked -- the engine task's ThreadLocalStateGuard restores + those on unwind.) + """ + + def __init__(self) -> None: + # Entries are (scope, prev_region, merged_region) or _SKIPPED. + self.stack: list[Any] = [] + + +# Bracket-stack marker for "prehook did nothing": distinct from an entry with +# scope=None, which means the region and creation hook were installed but no +# capture was active (the posthook must still pop them). +_SKIPPED: Any = object() + + +def _freeze_region_hook(node: Any, generation: int) -> None: + """Node creation hook: freeze the current region into ``node``'s bracket. + + Reads the region TLS slot and, if a region is open, installs the node's + single backward hook pair closing over that snapshot. Every backward- + annotating ``mark_kernels`` scope pushes one of these (and so does an + executing hooked node, to cover nodes created during its backward), so + a node created under nested scopes sees several identical firings; the + ``node.metadata`` marker makes only the first attach. + + AccumulateGrad is always excluded: a leaf's node is created once and + cached (possibly during warmup), so which scope would own it is an + accident of first use, and its work (the ``.grad`` accumulation) + belongs to the leaf rather than to any forward region. If accumulation + ever needs annotation it should get a dedicated leaf-accumulation + region, not first-use assignment. + """ + if isinstance(node, torch._C._functions.AccumulateGrad): + return + if _HOOKED_KEY in node.metadata: + return + frozen = _current_region() + if frozen is None: + return + node.metadata[_HOOKED_KEY] = True + _attach_backward_hooks(node, frozen, generation) + + +def _attach_backward_hooks(node: Any, frozen: dict[str, Any], generation: int) -> None: + """Register one pre/post hook pair bracketing ``node``'s backward execution. + + ``frozen`` is the region that was current when the node was created: + the annotations of every scope then open, already collapsed inner-wins. + The pair brackets the node's execution the same way ``mark_kernels`` + brackets forward work: snapshot the capture frontier before, collect + the new nodes after. It runs on the autograd engine thread executing + the node, which during a whole-graph capture participates in the + capture; when backward runs outside a capture the prehook sees no + active capture and the pair is a no-op. + + Executing the node is semantically re-entering its frozen region, so + the prehook merges ``frozen`` over the live region (a ``mark_kernels`` + enclosing the whole ``backward()`` call, if any) and installs the + result while the node runs. That gives nodes created during its + backward (``create_graph=True``, checkpoint recomputation) the forward + region's ownership -- a creation hook is pushed alongside so they get + hooked even when this backward runs eagerly and no scope's hook is + live -- and it puts a ``mark_kernels`` opened inside the execution + dynamically inner, so it outranks ``frozen``, which outranks the + enclosing scope. The posthook records the merged region once; scope + completion order then yields exactly that precedence at resolve. + + The pops are exception-safe without a finally: each engine task runs + under an ``at::ThreadLocalStateGuard`` that restores both the + creation-hook TLS and the region slot even when the node throws. + + ``generation`` is the annotation generation the owning scope was opened + under; the bracket only records (and only propagates to created nodes) + while it is current, so ``clear_kernel_annotations`` makes stale hooks + inert. + """ + state = _BracketState() + + def creation_hook(child: Any) -> None: + _freeze_region_hook(child, generation) + + def prehook(_grad_outputs: Any) -> None: + if generation != _annotation_generation or _is_tools_id_unavailable(): + state.stack.append(_SKIPPED) + return + # Re-establish ownership even when this backward is not itself + # captured: kernels of nodes created here may be captured by a + # later (e.g. second-order) annotated capture. + merged = {**(_current_region() or {}), **frozen} + prev = _enter_region(merged) + torch._C._autograd._push_node_creation_hook(creation_hook) + scope = _begin_kernel_scope() if _annotations_enabled else None + state.stack.append((scope, prev, merged)) + + def posthook(_grad_inputs: Any, _grad_outputs: Any) -> None: + if not state.stack: + return + entry = state.stack.pop() + if entry is _SKIPPED: + return + scope, prev, merged = entry + torch._C._autograd._pop_node_creation_hook() + _exit_region(prev) + if scope is None: + return + tools_ids = _end_kernel_scope(scope) + if tools_ids: + tagged = {**merged, "autograd_phase": "backward"} + _pending_scopes.append((tagged, tools_ids)) + + node.register_prehook(prehook) + node.register_hook(posthook) + + +@contextmanager # type: ignore[arg-type] +def mark_kernels(annotation: str | dict[str, Any], *, backward: bool = True): + r"""mark_kernels(annotation, *, backward=True) + + Context manager that annotates GPU work captured within its scope. + + Must be used inside an active :class:`torch.cuda.graph` capture with + ``enable_annotations=True``. Every kernel, memcpy, and memset node the + capture adds within the scope is tagged with :attr:`annotation`. Outside + a capture, with annotations disabled, or when :func:`is_available` is + ``False``, the context manager is a no-op. + + When scopes overlap on the same node (e.g. nested scopes), their + annotation dicts are merged key-by-key with the inner scope winning + common keys. + + By default backward work is annotated too: autograd nodes created by + forward operations inside the scope get hooks (via + :class:`torch.autograd.graph.node_creation_hook`) that bracket their + backward execution, so when the backward pass is itself captured -- + in the same capture as the forward or in a later one -- its kernels + are tagged with the same annotation, plus an ``"autograd_phase": + "backward"`` key marking them as backward work (``"autograd_phase"`` + is therefore reserved: backward annotation overwrites it). When + backward runs outside a capture the hooks are a no-op. Ownership + extends to higher-order gradients: nodes created while a hooked node + executes (``create_graph=True``, checkpoint recomputation) inherit its + annotations, so a later grad-of-grad capture is attributed too. + ``AccumulateGrad`` nodes are never annotated: a leaf's node is created + once and cached, so scope ownership would be an accident of first use. + Pass ``backward=False`` to annotate only the forward work, e.g. when a + wrapper implements its own backward attribution. The keyword's + presence also serves as the feature probe for that native backward + support: ``"backward" in inspect.signature(mark_kernels).parameters``. + + Implementation: on entry, records the current stream's capture frontier + and its existing direct dependents; on scope exit, walks only the + dependent nodes added since entry (falling back to newly created graph + roots when the scope is the first captured work). + + Args: + annotation (str or dict): Metadata to attach to each captured node. + A string ``s`` is recorded as ``{"name": s}``. Dict values must + be picklable. The key ``"name"`` names the region in trace + tooling; ``"stream"`` is reserved for stream-lane assignment. + backward (bool): Whether to also annotate the backward kernels of + autograd nodes created inside the scope. Default: ``True``. + + .. note:: + The nodes to annotate must be reachable from the capture frontier of + the stream that is current on scope entry. Work on a different + already-capturing stream must be synchronized with the current + stream first. + + .. note:: + Child-graph and conditional nodes have bodies in a separate + ``cudaGraph_t`` that this walk does not descend into, so their work is + left unannotated and a warning is issued. Descending is possible + (``cudaGraphNodeGetParams`` exposes the body graphs), but would not be + enough on its own: a body's nodes are numbered in that graph's id space + and are renumbered again when the exec graph inlines them, and nothing + exposes that renumbering, so :func:`remap_to_exec_graph` could not key + the annotations to what a profiler reports. For the same reason a scope + *inside* a conditional body (``torch.cond`` / ``torch.while_loop``) + records nothing at all. + + .. warning:: + This API is in prototype and may change in future releases. + + Example:: + + >>> # xdoctest: +SKIP("requires cuda-bindings and driver >= 13.1") + >>> g = torch.cuda.CUDAGraph() + >>> x = torch.randn(8, device="cuda") + >>> with torch.cuda.graph(g, enable_annotations=True): + ... with torch.cuda.graph_annotations.mark_kernels("phase_A"): + ... y = x + 1 + """ + if not _annotations_enabled or _is_tools_id_unavailable(): + yield + return + + if isinstance(annotation, str): + annotation = {"name": annotation} + + scope = _begin_kernel_scope() + if scope is None: + yield + return + + if ( + _capture_root_graph_id is not None + and _graph_id(scope.graph) != _capture_root_graph_id + ): + # Inside a conditional node's body: torch.cond / torch.while_loop capture into a + # separate cudaGraph_t. Its node ids are in that graph's id space and are + # renumbered again in the exec graph, so anything recorded here would be a key + # that matches nothing in a trace. Record nothing rather than dead keys. + warnings.warn( + "mark_kernels: this scope is inside a CUDA graph conditional-node body " + "(torch.cond / torch.while_loop), which is captured into a separate " + "cudaGraph_t whose node ids are never remapped to the exec graph; " + "nothing is annotated for it", + stacklevel=3, + ) + yield + return + + generation = _annotation_generation + + def creation_hook(node: Any) -> None: + _freeze_region_hook(node, generation) + + # Enter the region even when backward=False: the region must reflect the + # full dynamic scope so that nodes hooked by a nested backward=True scope + # (or by an executing hooked node) freeze this annotation too. + prev = _enter_region({**(_current_region() or {}), **annotation}) + try: + if backward: + with torch.autograd.graph.node_creation_hook(creation_hook): + yield + else: + yield + finally: + _exit_region(prev) + + tools_ids = _end_kernel_scope(scope) + if tools_ids: + _pending_scopes.append((annotation, tools_ids)) + + +def maybe_stamp_capture_graph_id(torch_cuda_graph: torch.cuda.CUDAGraph) -> None: + """Record the captured graph's id on the graph object for later remap. + + Called from ``capture_end`` in the window after capture ends and before the + graph is finalized, where the template ``cudaGraph_t`` is live for both + ``keep_graph`` modes (and reachable via ``raw_cuda_graph``). ``remap_to_exec_graph`` + later matches this graph's annotations to its exec id via this stamp, without + relying on call ordering. No-op if annotations are disabled or + ``cudaGraphNodeGetToolsId`` is unavailable. Harmless when no ``mark_kernels`` + regions run: the annotation map stays empty so resolve and remap are no-ops. + """ + if not _annotations_enabled or _is_tools_id_unavailable(): + return + # Past the _is_tools_id_unavailable() guard cuda-bindings is present and the + # driver supports the toolsId API (same version gate as cudaGraphGetId), so + # any error here is unexpected: error-check and let it raise. cudaGraphGetId + # accepts the raw cudaGraph_t handle (int) directly. + torch_cuda_graph._capture_graph_id = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetId( # pyrefly: ignore[missing-attribute] + torch_cuda_graph.raw_cuda_graph() + ) + ) + # Fresh capture: annotations are keyed by this capture id until remapped. + torch_cuda_graph._remapped_exec_id = None + + +def resolve_pending_annotations() -> None: + """Resolve pending scope toolsIds into kernel annotations.""" + if not _pending_scopes: + return + + try: + per_tools_id: defaultdict[int, list[Any]] = defaultdict(list) + for annotation, tools_ids in _pending_scopes: + for tools_id in tools_ids: + per_tools_id[tools_id].append(annotation) + + for tools_id, ann_list in per_tools_id.items(): + if len(ann_list) == 1: + _kernel_annotations[tools_id].append(ann_list[0]) + continue + + merged: dict[str, Any] = {} + for annotation in ann_list: + if isinstance(annotation, dict): + for key, value in annotation.items(): + merged.setdefault(key, value) + else: + merged.setdefault("name", annotation) + _kernel_annotations[tools_id].append(merged) + except Exception: + logger.exception("resolve_pending_annotations failed") + finally: + _pending_scopes.clear() + + +def remap_to_exec_graph(torch_cuda_graph: torch.cuda.CUDAGraph) -> None: + """Remap annotation keys from capture graph ID to exec graph ID. + + During capture, toolsId encodes the capture graph's ID in the upper + 32 bits. After instantiation, the profiler uses the exec graph's ID. + This function rewrites the keys so annotations match the trace. + + The graph's capture id is read from the ``_capture_graph_id`` stamped on it + by ``maybe_stamp_capture_graph_id`` in the capture_end window, so only the annotations + belonging to this graph are rekeyed. This is order-independent and correct + when several graphs are captured in sequence: call once per graph. Graphs + captured with annotations disabled have no capture id and are skipped. + + The exec graph id is only defined once the graph is instantiated. With + ``keep_graph=True`` instantiation is deferred past the ``torch.cuda.graph()`` + context, so the remap is driven from the graph's ``instantiate()``/ + ``replay()`` instead of context exit. Each ``instantiate()`` (even on an + unmodified template) produces a fresh exec graph id, so this rekeys from + wherever the annotations are currently keyed -- the capture id before the + first remap, the previous exec id after a re-instantiate -- to the current + exec id. ``_remapped_exec_id`` tracks that current key; when it already + matches the live exec id (e.g. replay after instantiate) this is a no-op. + """ + capture_graph_id = torch_cuda_graph._capture_graph_id + if not _kernel_annotations or capture_graph_id is None: + return + + exec_graph_id = _check_cuda_bindings( + _cuda_runtime.cudaGraphExecGetId( # pyrefly: ignore[missing-attribute] + torch_cuda_graph.raw_cuda_graph_exec() + ) + ) + + current_key_id = ( + capture_graph_id + if torch_cuda_graph._remapped_exec_id is None + else torch_cuda_graph._remapped_exec_id + ) + if current_key_id == exec_graph_id: + return + + remapped = _rekey_annotations(_kernel_annotations, current_key_id, exec_graph_id) + _kernel_annotations.clear() + _kernel_annotations.update(remapped) + torch_cuda_graph._remapped_exec_id = exec_graph_id + + +def _rekey_annotations( + annotations: dict[int, list[Any]], + capture_graph_id: int, + exec_graph_id: int, +) -> dict[int, list[Any]]: + """Rekey one graph's annotations from its capture id to its exec id. + + A toolsId packs the graph id in the upper 32 bits and the node id in the + lower 32. Only entries whose upper bits match ``capture_graph_id`` are + rewritten to ``exec_graph_id``; entries from other graphs (already remapped + to their own exec ids, or pending their own remap) are kept as-is. When two + capture-side ids collide on the same rekeyed id their lists are merged. + """ + remapped: dict[int, list[Any]] = {} + for tools_id, ann_list in annotations.items(): + if tools_id >> 32 != capture_graph_id: + remapped[tools_id] = ann_list + continue + node_id = tools_id & 0xFFFFFFFF + new_tools_id = (exec_graph_id << 32) | node_id + if new_tools_id in remapped: + remapped[new_tools_id].extend(ann_list) + else: + remapped[new_tools_id] = list(ann_list) + return remapped + + +def resolve_and_remap(torch_cuda_graph: torch.cuda.CUDAGraph) -> None: + """Resolve any pending scopes and remap one graph in a single call. + + Shorthand for ``resolve_pending_annotations()`` followed by + ``remap_to_exec_graph(graph)``; the pair normally run after a capture. + """ + resolve_pending_annotations() + remap_to_exec_graph(torch_cuda_graph) + + +def get_kernel_annotations() -> Mapping[int, list[Any]]: + r"""get_kernel_annotations() -> Mapping[int, list] + + Return the live registry of recorded kernel annotations. + + Keys are opaque integers matching the ``graph node id`` field that + CUPTI-based profilers attach to kernel events; values are the lists of + annotation dicts recorded for that node. The registry accumulates + across captures and is global to the process. + + The returned mapping is a **live view**: it is updated in place when a + graph is instantiated (annotation keys are rekeyed to the executable + graph's ids), so a reference obtained early stays current. Keys are + valid for joining against a profiler trace once the corresponding + graphs have been instantiated. Treat the mapping as read-only; snapshot + it with ``dict(...)`` if isolation is needed. + + .. warning:: + This API is in prototype and may change in future releases. + + Example:: + + >>> # xdoctest: +SKIP("requires cuda-bindings and driver >= 13.1") + >>> annotations = torch.cuda.graph_annotations.get_kernel_annotations() + >>> with open("annotations.pkl", "wb") as f: + ... pickle.dump(dict(annotations), f) + """ + return _kernel_annotations + + +def clear_kernel_annotations() -> None: + r"""clear_kernel_annotations() -> None + + Clear all recorded kernel annotations. + + The annotation registry is process-global and accumulates across + captures; long-running workloads that capture many graphs should clear + it once recorded annotations have been consumed (e.g. after saving + them alongside a profiler trace). + + Clearing forgets everything recorded so far and revokes recording from + every scope opened before the clear: backward hooks that + :func:`mark_kernels` attached to existing autograd nodes stay on the + graph but become inert. Scopes opened after the clear record normally. + + .. warning:: + This API is in prototype and may change in future releases. + """ + global _annotation_generation + _kernel_annotations.clear() + _pending_scopes.clear() + _annotation_generation += 1 + + +def remove_kernel_annotations(exec_graph_ids: Iterable[int]) -> None: + """Drop kernel-annotation entries whose exec graph id (tools_id >> 32) is in + exec_graph_ids, so the map does not grow across the run. Run by the annotation + resolver's graph-destroy handler.""" + ids = set(exec_graph_ids) + if not ids: + return + for key in [k for k in _kernel_annotations if k >> 32 in ids]: + del _kernel_annotations[key] + + +# Counter-based stream ID registry. IDs start at 60 (above the highest +# observed non-graphed CUDA stream ID) so every assigned lane is visually +# distinct in Perfetto and doesn't collide with real streams. +_stream_id_counter: int = 60 +_stream_id_map: dict[int, int] = {} + + +def _get_stream_id(stream: torch.cuda.Stream) -> int: + """Return a small, stable stream ID for the given CUDA stream.""" + global _stream_id_counter + key = stream.cuda_stream + if key not in _stream_id_map: + _stream_id_map[key] = _stream_id_counter + _stream_id_counter += 1 + return _stream_id_map[key] + + +def get_stream_for_pg(pg_key: str) -> int: + """Return a unique stream ID for the given process group key.""" + global _stream_id_counter + if pg_key not in _stream_id_map: + _stream_id_map[pg_key] = _stream_id_counter # type: ignore[assignment] + _stream_id_counter += 1 + return _stream_id_map[pg_key] # type: ignore[return-value] + + +@contextmanager # type: ignore[arg-type] +def mark_stream(stream: torch.cuda.Stream, annotation: str | dict[str, Any]): + """Switch to stream, inject its ID into annotation, and mark kernels. + + If *stream* is already the current stream, no stream switch or stream ID + injection happens — the kernels stay on whatever stream is active (which + keeps the trace faithful when e.g. FSDP uses the current stream for + copy-in instead of a separate one). When switching to a different stream, + this snapshots the current capturing branch before the target stream + runs marked work. If the target stream is already capturing, the marked + work must still be synchronized with the current stream so it is + reachable from that snapped frontier. + """ + if not _annotations_enabled: + with torch.cuda.stream(stream): + yield + return + if stream.cuda_stream == torch.cuda.current_stream().cuda_stream: + with mark_kernels(annotation): + yield + else: + if isinstance(annotation, str): + annotation = {"name": annotation} + if isinstance(annotation, dict): + annotation["stream"] = _get_stream_id(stream) + with mark_kernels(annotation): + with torch.cuda.stream(stream): + yield diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_memory_viz.py b/.venv/lib/python3.12/site-packages/torch/cuda/_memory_viz.py new file mode 100644 index 0000000000000000000000000000000000000000..f58f92090150d68183af79f6e817012faa26f8b9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_memory_viz.py @@ -0,0 +1,815 @@ +# mypy: allow-untyped-defs +import base64 +import io +import json +import operator +import os +import pickle +import subprocess +import sys +import warnings +from functools import lru_cache +from itertools import groupby +from typing import Any + + +cache = lru_cache(None) + +__all__ = ["format_flamegraph", "segments", "memory", "compare"] + + +def _frame_fmt(f, full_filename=False): + i = f["line"] + fname = f["filename"] + if not full_filename: + fname = fname.split("/")[-1] + func = f["name"] + return f"{fname}:{i}:{func}" + + +@cache +def _frame_filter(name, filename): + omit_functions = [ + "unwind::unwind", + "CapturedTraceback::gather", + "gather_with_cpp", + "_start", + "__libc_start_main", + "PyEval_", + "PyObject_", + "PyFunction_", + ] + omit_filenames = [ + "core/boxing", + "/Register", + "/Redispatch", + "pythonrun.c", + "Modules/main.c", + "Objects/call.c", + "Objects/methodobject.c", + "pycore_ceval.h", + "ceval.c", + "cpython/abstract.h", + ] + for of in omit_functions: + if of in name: + return False + for of in omit_filenames: + if of in filename: + return False + return True + + +def _frames_fmt(frames, full_filename=False, reverse=False): + if reverse: + frames = reversed(frames) + return [ + _frame_fmt(f, full_filename) + for f in frames + if _frame_filter(f["name"], f["filename"]) + ] + + +def _block_extra_legacy(b): + if "history" in b: + frames = b["history"][0].get("frames", []) + real_size = b["history"][0]["real_size"] + else: + real_size = b.get("requested_size", b["size"]) + frames = [] + return frames, real_size + + +def _block_extra(b): + if "frames" not in b: + # old snapshot format made it more complicated to get frames/allocated size + return _block_extra_legacy(b) + return b["frames"], b["requested_size"] + + +def format_flamegraph(flamegraph_lines, flamegraph_script=None): + if flamegraph_script is None: + cache_dir = os.path.expanduser("~/.cache/") + os.makedirs(cache_dir, exist_ok=True) + flamegraph_script = f"{cache_dir}/flamegraph.pl" + if not os.path.exists(flamegraph_script): + import tempfile + import urllib.request + + print(f"Downloading flamegraph.pl to: {flamegraph_script}") + # Download to a temp file next to the target, then atomically move it + # into place so a concurrent reader never sees a partial file. The temp + # file lives in the same directory (same filesystem) so os.replace is + # atomic, and is created with delete=False + manual cleanup: moving a + # delete=True NamedTemporaryFile out from under its context manager makes + # __exit__ fail to unlink the now-missing file. + fd, tmp_name = tempfile.mkstemp( + suffix=".pl", dir=os.path.dirname(flamegraph_script) or None + ) + os.close(fd) + try: + urllib.request.urlretrieve( + "https://raw.githubusercontent.com/brendangregg/FlameGraph/master/flamegraph.pl", + tmp_name, + ) + os.chmod(tmp_name, 0o755) + os.replace(tmp_name, flamegraph_script) + finally: + if os.path.exists(tmp_name): + os.remove(tmp_name) + args = [flamegraph_script, "--countname", "bytes"] + with subprocess.Popen( + args, stdin=subprocess.PIPE, stdout=subprocess.PIPE, encoding="utf-8" + ) as p: + if p.stdin is None: + raise AssertionError("p.stdin is None") + if p.stdout is None: + raise AssertionError("p.stdout is None") + p.stdin.write(flamegraph_lines) + p.stdin.close() + result = p.stdout.read() + p.stdout.close() + p.wait() + if p.wait() != 0: + raise AssertionError(f"flamegraph process exited with code {p.wait()}") + return result + + +def _write_blocks(f, prefix, blocks): + def frames_fragment(frames): + if not frames: + return "" + return ";".join(_frames_fmt(frames, reverse=True)) + + for b in blocks: + if "history" not in b: + frames, accounted_for_size = _block_extra(b) + f.write( + f"{prefix};{b['state']};{frames_fragment(frames)} {accounted_for_size}\n" + ) + else: + accounted_for_size = 0 + for h in b["history"]: + sz = h["real_size"] + accounted_for_size += sz + if "frames" in h: + frames = h["frames"] + f.write(f"{prefix};{b['state']};{frames_fragment(frames)} {sz}\n") + else: + f.write(f"{prefix};{b['state']}; {sz}\n") + gaps = b["size"] - accounted_for_size + if gaps: + f.write(f"{prefix};{b['state']}; {gaps}\n") + + +def segments(snapshot, format_flamegraph=format_flamegraph): + f = io.StringIO() + for seg in snapshot["segments"]: + prefix = f"stream_{seg['stream']};seg_{seg['address']}" + _write_blocks(f, prefix, seg["blocks"]) + return format_flamegraph(f.getvalue()) + + +def memory(snapshot, format_flamegraph=format_flamegraph): + f = io.StringIO() + for seg in snapshot["segments"]: + prefix = f"stream_{seg['stream']}" + _write_blocks(f, prefix, seg["blocks"]) + return format_flamegraph(f.getvalue()) + + +def compare(before, after, format_flamegraph=format_flamegraph): + def _seg_key(seg): + return (seg["address"], seg["total_size"]) + + def _seg_info(seg): + return f"stream_{seg['stream']};seg_{seg['address']}" + + f = io.StringIO() + + before_segs = {_seg_key(seg) for seg in before} + after_segs = {_seg_key(seg) for seg in after} + + print(f"only_before = {[a for a, _ in (before_segs - after_segs)]}") + print(f"only_after = {[a for a, _ in (after_segs - before_segs)]}") + + for seg in before: + if _seg_key(seg) not in after_segs: + _write_blocks(f, f"only_before;{_seg_info(seg)}", seg["blocks"]) + + for seg in after: + if _seg_key(seg) not in before_segs: + _write_blocks(f, f"only_after;{_seg_info(seg)}", seg["blocks"]) + + return format_flamegraph(f.getvalue()) + + +def _format_size(num): + # https://stackoverflow.com/questions/1094841/get-human-readable-version-of-file-size + for unit in ["", "Ki", "Mi", "Gi", "Ti", "Pi", "Ei", "Zi"]: + if abs(num) < 1024.0: + return f"{num:3.1f}{unit}B" + num /= 1024.0 + return f"{num:.1f}YiB" + + +class Bytes: + def __init__(self, value): + self.value = value + + def __add__(self, rhs): + return Bytes(self.value + rhs) + + def __repr__(self): + return _format_size(self.value) + + +def calc_active(seg): + return sum(b["size"] for b in seg["blocks"] if b["state"] == "active_allocated") + + +def _report_free(free_external, free_internal): + total = free_external + free_internal + suffix = "" + if total != 0: + pct = (free_internal / total) * 100 + suffix = f" ({pct:.1f}% internal)" + return f"{Bytes(total)}{suffix}" + + +PAGE_SIZE = 1024 * 1024 * 20 +legend = f"""\ + +Legend: + [a ] - a segment in the allocator + ^-- a page {Bytes(PAGE_SIZE)} of memory in the segment + a-z: pages filled with a single block's content + ' ': page is completely free + *: page if completely full with multiple blocks + 0-9: page is partially full with tensors of multiple blocks (9 == 90% full) + (X% internal) - of the free memory, X% is free because we rounded the size of the allocation. +""" + + +def segsum(data): + r"""Visually reports how the allocator has filled its segments. + + This printout can help debug fragmentation issues since free fragments + will appear as gaps in this printout. The amount of free space is reported + for each segment. + We distinguish between internal free memory which occurs because the + allocator rounds the allocation size, and external free memory, which are + the gaps between allocations in a segment. + Args: + data: snapshot dictionary created from _snapshot() + """ + out = io.StringIO() + out.write(f"Summary of segments >= {Bytes(PAGE_SIZE)} in size\n") + total_reserved = 0 + total_allocated = 0 + free_external = 0 + free_internal = 0 + for seg in sorted( + data["segments"], key=lambda x: (x["total_size"], calc_active(x)) + ): + total_reserved += seg["total_size"] + + seg_free_external = 0 + seg_free_internal = 0 + seg_allocated = 0 + all_ranges = [] + boffset = 0 + for b in seg["blocks"]: + active = b["state"] == "active_allocated" + if active: + _, allocated_size = _block_extra(b) + all_ranges.append((boffset, allocated_size, True)) + seg_allocated += allocated_size + seg_free_internal += b["size"] - allocated_size + else: + seg_free_external += b["size"] + + boffset += b["size"] + + total_allocated += seg_allocated + free_external += seg_free_external + free_internal += seg_free_internal + + nseg = (seg["total_size"] - 1) // PAGE_SIZE + 1 + occupied = [" " for _ in range(nseg)] + frac = [0.0 for _ in range(nseg)] + active_size = 0 + for i, (start_, size, active) in enumerate(all_ranges): + active_size += size + finish_ = start_ + size + start = start_ // PAGE_SIZE + finish = (finish_ - 1) // PAGE_SIZE + 1 + m = chr(ord("a" if active else "A") + (i % 26)) + for j in range(start, finish): + s = max(start_, j * PAGE_SIZE) + e = min(finish_, (j + 1) * PAGE_SIZE) + frac[j] += (e - s) / PAGE_SIZE + if occupied[j] != " ": + occupied[j] = "0123456789*"[int(frac[j] * 10)] + else: + occupied[j] = m + stream = "" if seg["stream"] == 0 else f", stream_{seg['stream']}" + body = "".join(occupied) + if seg_free_external + seg_free_internal + seg_allocated != seg["total_size"]: + raise AssertionError( + f"Segment size mismatch: {seg_free_external} + {seg_free_internal} + {seg_allocated} != {seg['total_size']}" + ) + stream = f" stream_{seg['stream']}" if seg["stream"] != 0 else "" + if seg["total_size"] >= PAGE_SIZE: + out.write( + f"[{body}] {Bytes(seg['total_size'])} allocated, " + f"{_report_free(seg_free_external, seg_free_internal)} free{stream}\n" + ) + out.write(f"segments: {len(data['segments'])}\n") + out.write(f"total_reserved: {Bytes(total_reserved)}\n") + out.write(f"total_allocated: {Bytes(total_allocated)}\n") + out.write(f"total_free: {_report_free(free_external, free_internal)}\n") + out.write(legend) + if free_internal + free_external + total_allocated != total_reserved: + raise AssertionError( + f"Memory accounting error: {free_internal} + {free_external} + {total_allocated} != {total_reserved}" + ) + return out.getvalue() + + +def trace(data): + out = io.StringIO() + + def format(entries): + segment_intervals: list = [] + segment_addr_to_name = {} + allocation_addr_to_name = {} + + free_names: list = [] + next_name = 0 + + def _name(): + nonlocal next_name + if free_names: + return free_names.pop() + r, m = next_name // 26, next_name % 26 + next_name += 1 + return f"{chr(ord('a') + m)}{'' if r == 0 else r}" + + def find_segment(addr): + for name, saddr, size in segment_intervals: + if addr >= saddr and addr < saddr + size: + return name, saddr + for i, seg in enumerate(data["segments"]): + saddr = seg["address"] + size = seg["allocated_size"] + if addr >= saddr and addr < saddr + size: + return f"seg_{i}", saddr + return None, None + + count = 0 + out.write(f"{len(entries)} entries\n") + + total_reserved = 0 + for seg in data["segments"]: + total_reserved += seg["total_size"] + + for count, e in enumerate(entries): + if e["action"] == "alloc": + addr, size = e["addr"], e["size"] + n = _name() + seg_name, seg_addr = find_segment(addr) + if seg_name is None: + seg_name = "MEM" + offset = addr + else: + offset = addr - seg_addr + out.write(f"{n} = {seg_name}[{offset}:{Bytes(size)}]\n") + allocation_addr_to_name[addr] = (n, size, count) + count += size + elif e["action"] == "free_requested": + addr, size = e["addr"], e["size"] + name, _, _ = allocation_addr_to_name.get(addr, (addr, None, None)) + out.write(f"del {name} # {Bytes(size)}\n") + elif e["action"] == "free_completed": + addr, size = e["addr"], e["size"] + count -= size + name, _, _ = allocation_addr_to_name.get(addr, (addr, None, None)) + out.write(f"# free completed for {name} {Bytes(size)}\n") + if name in allocation_addr_to_name: + free_names.append(name) + del allocation_addr_to_name[name] + elif e["action"] == "segment_alloc": + addr, size = e["addr"], e["size"] + name = _name() + out.write(f"{name} = cudaMalloc({addr}, {Bytes(size)})\n") + segment_intervals.append((name, addr, size)) + segment_addr_to_name[addr] = name + elif e["action"] == "segment_free": + addr, size = e["addr"], e["size"] + name = segment_addr_to_name.get(addr, addr) + out.write(f"cudaFree({name}) # {Bytes(size)}\n") + if name in segment_addr_to_name: + free_names.append(name) + del segment_addr_to_name[name] + elif e["action"] == "oom": + size = e["size"] + free = e["device_free"] + out.write( + f"raise OutOfMemoryError # {Bytes(size)} requested, {Bytes(free)} free in CUDA\n" + ) + elif e["action"] == "annotate": + addr = e["addr"] + name, _, _ = allocation_addr_to_name.get(addr, (addr, None, None)) + out.write(f"# annotate {name}: {e['user_metadata']}\n") + else: + out.write(f"{e}\n") + out.write(f"TOTAL MEM: {Bytes(count)}") + + for i, d in enumerate(data["device_traces"]): + if d: + out.write(f"Device {i} ----------------\n") + format(d) + return out.getvalue() + + +_memory_viz_template = r""" + + + + + + + +""" + + +def _format_viz(data, viz_kind, device): + if device is not None: + warnings.warn( + "device argument is deprecated, plots now contain all device", + FutureWarning, + stacklevel=3, + ) + buffer = pickle.dumps(data) + buffer += b"\x00" * (3 - len(buffer) % 3) + # Encode the buffer with base64 + encoded_buffer = base64.b64encode(buffer).decode("utf-8") + + json_format = json.dumps([{"name": "snapshot.pickle", "base64": encoded_buffer}]) + return _memory_viz_template.replace("$VIZ_KIND", repr(viz_kind)).replace( + "$SNAPSHOT", json_format + ) + + +def filter_alloc_free_pairs(data): + for dev_id in range(len(data["device_traces"])): + # set of indexes of trace events for alloc-free pairs + filterSet = set() + # map from addr to index of alloc event + allocMap = {} + # set of addrs from free_requested events + freeRequested = set() + for idx, event in enumerate(data["device_traces"][dev_id]): + if event["action"] == "alloc": + allocMap[event["addr"]] = idx + elif event["action"] == "free_requested": + freeRequested.add(event["addr"]) + if allocMap.get(event["addr"]) is not None: + filterSet.add(idx) + filterSet.add(allocMap[event["addr"]]) + allocMap.pop(event["addr"]) + elif event["action"] == "free_completed": + if event["addr"] in freeRequested: + freeRequested.remove(event["addr"]) + filterSet.add(idx) + else: + print(f"free_completed without free_requested: {event}") + + # Remove events whose index is in filterSet + if filterSet: + # Create a new list excluding events with indices in filterSet + data["device_traces"][dev_id] = [ + event + for idx, event in enumerate(data["device_traces"][dev_id]) + if idx not in filterSet + ] + + return data + + +def trace_plot(data, device=None, plot_segments=False, filter_freed=False): + """Generate a visualization over time of the memory usage recorded by the trace as an html file. + + Args: + data: Memory snapshot as generated from torch.cuda.memory._snapshot() + device (torch.device, optional): Generate the trace for this device, needed if multiple devices have allocations. + plot_segments (bool, optional): Plots memory returned from cudaMalloc, rather than individual allocations. + Defaults to False. + filter_freed (bool, optional): Filter out alloc-free paired events to only plot allocations that are not freed yet. + Defaults to False to plot all trace events. + + Returns: + str: HTML of visualization + """ + if filter_freed: + data = filter_alloc_free_pairs(data) + + return _format_viz( + data, + "Active Memory Timeline" + if not plot_segments + else "Active Cached Memory Timeline", + device, + ) + + +def _profile_to_snapshot(profile): + import torch + from torch._C._profiler import _EventType + from torch.profiler._memory_profiler import Action, TensorKey + + memory_profile = profile._memory_profile() + + allocation_stacks = {} + for event in memory_profile._op_tree.sorted_nodes: + if event.tag == _EventType.Allocation: + parent = event.parent + python_parents = [] + while parent: + if parent.tag in (_EventType.PyCall, _EventType.PyCCall): + python_parents.append(parent) + parent = parent.parent + key = TensorKey.from_allocation(event.extra_fields) + + # Corner case: If allocation doesn't have an ID (can't prove it was used as a Tensor) + # key will be None. I should add some way to identify these, I just haven't yet. + if key and event.extra_fields.alloc_size > 0: + allocation_stacks[key] = python_parents + + device_count = torch.cuda.device_count() + snapshot: dict[str, list[Any]] = { + "device_traces": [[] for _ in range(device_count + 1)], + "segments": [ + { + "device": device, + "address": None, + "total_size": 0, + "stream": 0, + "blocks": [], + } + for device in range(device_count + 1) + ], + } + + def to_device(device): + if device.type == "cuda": + return device.index + else: + return device_count + + def allocate(size, tensor_key, version, during_trace=True): + device = to_device(tensor_key.device) + addr = tensor_key.storage.ptr + + seg = snapshot["segments"][device] # type: ignore[index] + if seg["address"] is None or seg["address"] > addr: + seg["address"] = addr + seg["total_size"] = max( + seg["total_size"], addr + size + ) # record max addr for now, we will make it the size later + category = memory_profile._categories.get(tensor_key, version) + category = category.name.lower() if category is not None else "unknown" + stack = allocation_stacks.get(tensor_key, ()) + stack = [{"filename": "none", "line": 0, "name": p.name} for p in stack] + r = { + "action": "alloc", + "addr": addr, + "size": size, + "stream": 0, + "frames": stack, + "category": category, + } + if during_trace: + snapshot["device_traces"][device].append(r) + return r + + def free(alloc, device): + for e in ("free_requested", "free_completed"): + snapshot["device_traces"][device].append( + { + "action": e, + "addr": alloc["addr"], + "size": alloc["size"], + "stream": 0, + "frames": alloc["frames"], + } + ) + + kv_to_elem = {} + + # create the device trace + for _time, action, (tensor_key, version), size in memory_profile.timeline: + if not isinstance(tensor_key, TensorKey): + continue + if action == Action.CREATE: + kv_to_elem[(tensor_key, version)] = allocate(size, tensor_key, version) + elif action == Action.DESTROY: + free(kv_to_elem.pop((tensor_key, version)), to_device(tensor_key.device)) + elif action == Action.INCREMENT_VERSION: + free(kv_to_elem.pop((tensor_key, version)), to_device(tensor_key.device)) + kv_to_elem[(tensor_key, version + 1)] = allocate( + size, tensor_key, version + 1 + ) + elif action == Action.PREEXISTING: + kv_to_elem[(tensor_key, version)] = allocate( + size, tensor_key, version, during_trace=False + ) + + # create the final snapshot state + blocks_at_end = [ + (to_device(tensor_key.device), event["addr"], event["size"], event["frames"]) + for (tensor_key, version), event in kv_to_elem.items() + ] + for device, blocks in groupby(sorted(blocks_at_end), key=operator.itemgetter(0)): + seg = snapshot["segments"][device] # type: ignore[index] + last_addr = seg["address"] + for _, addr, size, frames in blocks: + if last_addr < addr: + seg["blocks"].append({"size": addr - last_addr, "state": "inactive"}) + seg["blocks"].append( + { + "size": size, + "state": "active_allocated", + "requested_size": size, + "frames": frames, + } + ) + last_addr = addr + size + if last_addr < seg["total_size"]: + seg["blocks"].append( + {"size": seg["total_size"] - last_addr, "state": "inactive"} + ) + + snapshot["segments"] = [seg for seg in snapshot["segments"] if seg["blocks"]] # type: ignore[attr-defined] + for seg in snapshot["segments"]: # type: ignore[attr-defined, name-defined, no-redef] + seg["total_size"] -= seg["address"] + if not seg["blocks"]: + seg["blocks"].append({"size": seg["total_size"], "state": "inactive"}) + + return snapshot + + +def profile_plot(profile, device=None): + """Generate a visualization over time of the memory usage recorded by kineto memory profiling as an html file. + + Args: + profile: profile as generated by `torch.profiler.profile(profile_memory=True)` + device (torch.device, optional): Generate the trace for this device, needed if multiple devices have allocations. + + Returns: + str: HTML of visualization + """ + snapshot = _profile_to_snapshot(profile) + return _format_viz(snapshot, "Active Memory Timeline", device) + + +def segment_plot(data: Any, device=None): + return _format_viz(data, "Allocator State History", device) + + +if __name__ == "__main__": + import os.path + + thedir = os.path.realpath(os.path.dirname(__file__)) + if thedir in sys.path: + # otherwise we find cuda/random.py as random... + sys.path.remove(thedir) + import argparse + + fn_name = "torch.cuda.memory._snapshot()" + pickled = f"pickled memory statistics from {fn_name}" + parser = argparse.ArgumentParser( + description=f"Visualize memory dumps produced by {fn_name}" + ) + + subparsers = parser.add_subparsers(dest="action") + + def _output(p): + p.add_argument( + "-o", + "--output", + default="output.svg", + help="flamegraph svg (default: output.svg)", + ) + + description = "Prints overall allocation statistics and a visualization of how the allocators segments are currently filled." + stats_a = subparsers.add_parser("stats", description=description) + stats_a.add_argument("input", help=pickled) + + description = "Prints buffer of the most recent allocation events embedded in the snapshot in a Pythonic style." + trace_a = subparsers.add_parser("trace", description=description) + trace_a.add_argument("input", help=pickled) + + description = "Generate a flamegraph that visualizes what memory is stored in each allocator segment (aka block)" + segments_a = subparsers.add_parser("segments", description=description) + segments_a.add_argument("input", help=pickled) + _output(segments_a) + + description = ( + "Generate a flamegraph the program locations contributing to CUDA memory usage." + ) + memory_a = subparsers.add_parser("memory", description=description) + memory_a.add_argument("input", help=pickled) + _output(memory_a) + + description = ( + "Generate a flamegraph that shows segments (aka blocks) that have been added " + "or removed between two different memory snapshots." + ) + compare_a = subparsers.add_parser("compare", description=description) + compare_a.add_argument("before", help=pickled) + compare_a.add_argument("after", help=pickled) + _output(compare_a) + + plots = ( + ( + "trace_plot", + "Generate a visualization over time of the memory usage recorded by the trace as an html file.", + ), + ( + "segment_plot", + "Visualize how allocations are packed into allocator segments at each point in a trace as an html file.", + ), + ) + for cmd, description in plots: + trace_plot_a = subparsers.add_parser(cmd, description=description) + trace_plot_a.add_argument("input", help=pickled) + help = "visualize trace from this device (default: chooses the only device with trace info or errors)" + trace_plot_a.add_argument("-d", "--device", type=int, default=None, help=help) + help = "path to save the visualization(default: output.html)" + trace_plot_a.add_argument("-o", "--output", default="output.html", help=help) + if cmd == "trace_plot": + help = "visualize change to segments rather than individual allocations" + trace_plot_a.add_argument( + "-s", "--segments", action="store_true", help=help + ) + + help = ( + "filter out allocation-free pairs to only visualize the allocations that are not freed yet;" + "useful to reduce the number of events for large traces for debugging OOM" + ) + trace_plot_a.add_argument( + "-f", "--filter_freed", action="store_true", help=help + ) + + args = parser.parse_args() + + def _read(name): + if name == "-": + data = pickle.load(sys.stdin.buffer) + else: + with open(name, "rb") as f: + data = pickle.load(f) + if isinstance(data, list): # segments only... + data = {"segments": data, "traces": []} + return data + + def _write(name, data): + with open(name, "w") as f: + f.write(data) + + if args.action == "segments": + data = _read(args.input) + _write(args.output, segments(data)) + elif args.action == "memory": + data = _read(args.input) + _write(args.output, memory(data)) + elif args.action == "stats": + data = _read(args.input) + print(segsum(data)) + elif args.action == "trace": + data = _read(args.input) + print(trace(data)) + elif args.action == "compare": + before = _read(args.before) + after = _read(args.after) + _write(args.output, compare(before, after)) + elif args.action == "trace_plot": + data = _read(args.input) + _write( + args.output, + trace_plot( + data, + device=args.device, + plot_segments=args.segments, + filter_freed=args.filter_freed, + ), + ) + elif args.action == "segment_plot": + data = _read(args.input) + _write(args.output, segment_plot(data, device=args.device)) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_pin_memory_utils.py b/.venv/lib/python3.12/site-packages/torch/cuda/_pin_memory_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..2254c3ff109fd1cd3be12fa683a284b527af88ee --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_pin_memory_utils.py @@ -0,0 +1,25 @@ +import torch + + +def pin_memory(data_ptr: int, size: int) -> None: + cudart = torch.cuda.cudart() + succ = int( + cudart.cudaHostRegister( + data_ptr, + size, + 1, # lines up with 'cudaHostRegisterPortable' + ) + ) + + if succ != 0: + raise RuntimeError( + f"Registering memory failed with cudaError: {succ}." + " It's possible that this is an asynchronous error raised from a previous cuda operation." + " Consider launching with CUDA_LAUNCH_BLOCKING=1 to debug." + ) + + +def unpin_memory(data_ptr: int) -> None: + succ = int(torch.cuda.cudart().cudaHostUnregister(data_ptr)) + if succ != 0: + raise AssertionError(f"Unpinning shared memory failed with error-code: {succ}") diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_sanitizer.py b/.venv/lib/python3.12/site-packages/torch/cuda/_sanitizer.py new file mode 100644 index 0000000000000000000000000000000000000000..b9b5e773136d4689cbe8de83ce9fdd1e2abeada2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_sanitizer.py @@ -0,0 +1,670 @@ +# mypy: allow-untyped-defs +r""" +This module introduces CUDA Sanitizer, a tool for detecting synchronization errors between kernels ran on different streams. + +It stores information on accesses to tensors to determine if they are synchronized +or not. When enabled in a python program and a possible data race is detected, a +detailed warning will be printed and the program will exit. + +It can be enabled either by importing this module and calling +:func:`enable_cuda_sanitizer()` or by exporting the ``TORCH_CUDA_SANITIZER`` +environment variable. +""" + +import enum +import functools +import inspect +import io +import logging +import re +import sys +import textwrap +import traceback +from collections.abc import Iterator +from dataclasses import dataclass, field +from typing import Any, TypeVar + +import torch +import torch.cuda._gpu_trace as gpu_trace +from torch.utils import _pytree as pytree +from torch.utils._python_dispatch import TorchDispatchMode + + +aten = torch.ops.aten + +DEFAULT_STREAM_ID = 0 + +TK = TypeVar("TK") +TVa = TypeVar("TVa") +TVb = TypeVar("TVb") + +DataPtr = int +StreamId = int +EventId = int +SeqNum = int + +logger = logging.getLogger(__name__) + +# Note that this is only factories that take Tensor as input as they are +# the ones we care about. +FACTORY_FUNCTION_REGEX = re.compile("(new_.*|.*_like)") + + +class AccessType(enum.Enum): + READ = enum.auto() + WRITE = enum.auto() + + def __str__(self): + return "reading from" if self is AccessType.READ else "writing to" + + +@dataclass +class Access: + r"""Stores information about a single access to a tensor by a kernel. + + Args: + type: either AccessType.READ or AccessType.Write. + seq_num: the sequential number of the kernel performing the access. + stream: the stream id of the stream executing the kernel. + operator: the schema of the launched kernel, which lists the + arguments and return type. + aliases: the arguments in the schema this access corresponds to. + is_output: Whether the tensor was an output of the kernel. + stack_trace: the stack summary object captured during access. + """ + + type: AccessType + seq_num: SeqNum + stream: StreamId + operator: str + aliases: list[str] + is_output: bool + stack_trace: traceback.StackSummary + + +class SynchronizationError(Exception): + """Base class for errors detected by CUDA Sanitizer.""" + + +class UnsynchronizedAccessError(SynchronizationError): + """Stores information about two unsynchronized accesses to one data pointer.""" + + def __init__( + self, + data_ptr: DataPtr, + allocation_stack_trace: traceback.StackSummary | None, + current_access: Access, + previous_access: Access, + ): + self.data_ptr = data_ptr + self.allocation_stack_trace = allocation_stack_trace + self.current_access = current_access + self.previous_access = previous_access + + def __str__(self): + def format_access(access: Access): + message.write(f"{access.operator}\n{access.type}") + if access.aliases: + message.write(" argument(s) " + ", ".join(access.aliases)) + if access.is_output: + message.write(", and to") + if access.is_output: + message.write(" the output") + message.write( + f"\nWith stack trace:\n{''.join(access.stack_trace.format())}\n" + ) + + with io.StringIO() as message: + message.write( + textwrap.dedent( + f"""\ + ============================ + CSAN detected a possible data race on tensor with data pointer {self.data_ptr} + Access by stream {self.current_access.stream} during kernel: + """ + ) + ) + format_access(self.current_access) + + message.write( + f"Previous access by stream {self.previous_access.stream} during kernel:\n" + ) + format_access(self.previous_access) + + if self.allocation_stack_trace: + message.write( + "Tensor was allocated with stack trace:\n" + f"{''.join(self.allocation_stack_trace.format())}" + ) + else: + message.write("Trace for tensor allocation not found.") + return message.getvalue() + + +class CUDASanitizerErrors(Exception): + """Wrapper class for errors reported by CUDA Sanitizer.""" + + def __init__(self, errors: list[SynchronizationError]): + self.errors = errors + + def __str__(self): + return f"detected {len(self.errors)} errors" + + +@dataclass +class TensorInfo: + r"""Stores information about a single tensor and recent accesses to it. + + Args: + allocation_stack_trace: the stack summary object captured during tensor + allocation. Can be ``None`` if the allocation wasn't caught by CSAN. + reads: list of read accesses to the tensor that were performed since + the last write. + write: the last write access to the tensor. + """ + + allocation_stack_trace: traceback.StackSummary | None + reads: list[Access] = field(default_factory=list) + write: Access | None = None + + +class _TensorsAccessed: + def __init__(self) -> None: + self.accesses: dict[DataPtr, TensorInfo] = {} + + def ensure_tensor_exists(self, data_ptr: DataPtr) -> None: + if data_ptr not in self.accesses: + logger.info( + "Found tensor with pointer: %s, but no matching tensor " + "allocation in the trace. Backfilling the trace now. " + "Perhaps the sanitizer was enabled after some torch operations?", + data_ptr, + ) + self.create_tensor(data_ptr, None) + + def ensure_tensor_does_not_exist(self, data_ptr: DataPtr) -> None: + if data_ptr in self.accesses: + logger.info( + "Found duplicate tensor allocation in the trace for tensor with " + "pointer: %s. Assuming the trace for tensor deallocation " + "wasn't caught and backfilling it now. " + "Perhaps the sanitizer was enabled after some torch operations?", + data_ptr, + ) + self.delete_tensor(data_ptr) + + def create_tensor( + self, data_ptr: DataPtr, stack_trace: traceback.StackSummary | None + ) -> None: + self.accesses[data_ptr] = TensorInfo(stack_trace) + + def delete_tensor(self, data_ptr: DataPtr) -> None: + del self.accesses[data_ptr] + + def were_there_reads_since_last_write(self, data_ptr: DataPtr) -> bool: + return bool(self.accesses[data_ptr].reads) + + def get_allocation_stack_trace( + self, data_ptr: DataPtr + ) -> traceback.StackSummary | None: + return self.accesses[data_ptr].allocation_stack_trace + + def get_write(self, data_ptr: DataPtr) -> Access | None: + return self.accesses[data_ptr].write + + def get_reads(self, data_ptr: DataPtr) -> list[Access]: + return self.accesses[data_ptr].reads + + def add_read(self, data_ptr: DataPtr, access: Access) -> None: + self.accesses[data_ptr].reads.append(access) + + def set_write(self, data_ptr: DataPtr, access: Access) -> None: + self.accesses[data_ptr].write = access + self.accesses[data_ptr].reads = [] + + +class StreamSynchronizations: + def __init__(self) -> None: + self.current_sync_states: dict[StreamId, dict[StreamId, SeqNum]] = {} + self.recorded_sync_states: dict[EventId, dict[StreamId, SeqNum]] = {} + self.host_sync_state: dict[StreamId, SeqNum] = {} + self.create_stream(DEFAULT_STREAM_ID) + + def _ensure_stream_exists(self, stream: StreamId) -> None: + if stream not in self.current_sync_states: + logger.info( + "Found Stream with id: %s, but no matching stream " + "creation in the trace. Backfilling the trace now. " + "Perhaps the sanitizer was enabled after some torch operations?", + stream, + ) + self.create_stream(stream) + + def _ensure_event_exists(self, event: EventId) -> None: + if event not in self.recorded_sync_states: + logger.info( + "Found Event with id: %s, but no matching event " + "creation in the trace. Backfilling the trace now. " + "Perhaps the sanitizer was enabled after some torch operations?", + event, + ) + self.create_event(event) + + def _ensure_event_does_not_exist(self, event: EventId) -> None: + if event in self.recorded_sync_states: + logger.info( + "Found duplicate event creation in the trace for event with " + "id: %s. Assuming the trace for event deletion wasn't caught " + "and backfilling it now. " + "Perhaps the sanitizer was enabled after some torch operations?", + event, + ) + self.delete_event(event) + + def create_stream(self, stream: StreamId) -> None: + if stream in self.current_sync_states: + logger.info( + "Found duplicate Stream creation in the trace for Stream with " + "id: %s. PyTorch Streams are only created once, so this " + "trace entry is ignored.", + stream, + ) + else: + self.host_sync_state[stream] = 0 + self.current_sync_states[stream] = self.host_sync_state.copy() + + def create_event(self, event: EventId) -> None: + self._ensure_event_does_not_exist(event) + self.recorded_sync_states[event] = {} + + def delete_event(self, event: EventId) -> None: + self._ensure_event_exists(event) + del self.recorded_sync_states[event] + + def update_seq_num(self, stream: StreamId, seq_num: SeqNum) -> None: + self._ensure_stream_exists(stream) + self.current_sync_states[stream][stream] = seq_num + + def record_state(self, event: EventId, stream: StreamId) -> None: + self._ensure_event_exists(event) + self._ensure_stream_exists(stream) + self.recorded_sync_states[event] = self.current_sync_states[stream].copy() + + def _state_wait_for_other( + self, state: dict[StreamId, SeqNum], other: dict[StreamId, SeqNum] + ) -> None: + for stream, seq_num in other.items(): + state[stream] = max(state.get(stream, -1), seq_num) + + def stream_wait_for_event(self, stream: StreamId, event: EventId) -> None: + self._ensure_stream_exists(stream) + self._ensure_event_exists(event) + self._state_wait_for_other( + self.current_sync_states[stream], self.recorded_sync_states[event] + ) + + def all_streams_wait_for_event(self, event: EventId) -> None: + self._ensure_event_exists(event) + for stream in self.current_sync_states: + self.stream_wait_for_event(stream, event) + + self._state_wait_for_other( + self.host_sync_state, self.recorded_sync_states[event] + ) + + def all_streams_wait_for_stream(self, stream: StreamId) -> None: + self._ensure_stream_exists(stream) + for state in self.current_sync_states.values(): + self._state_wait_for_other(state, self.current_sync_states[stream]) + + self._state_wait_for_other( + self.host_sync_state, self.current_sync_states[stream] + ) + + def sync_all_streams(self) -> None: + for stream, state in self.current_sync_states.items(): + self.host_sync_state[stream] = state[stream] + + for state in self.current_sync_states.values(): + self._state_wait_for_other(state, self.host_sync_state) + + def is_ordered_after( + self, current_stream: StreamId, seq_num: SeqNum, other_stream: StreamId + ) -> bool: + self._ensure_stream_exists(current_stream) + self._ensure_stream_exists(other_stream) + return seq_num <= self.current_sync_states[current_stream].get(other_stream, -1) + + +class EventHandler: + """Analyzes CSAN trace for synchronization errors. + + Stores information on each stream's synchronizations with other streams as well + as tensor accesses to determine whether a given kernel launch might cause a + data race. + """ + + def __init__(self) -> None: + self.tensors_accessed = _TensorsAccessed() + self.syncs = StreamSynchronizations() + self.seq_num: SeqNum = 0 + + def _handle_kernel_launch( + self, + stream: StreamId, + read_only: set[DataPtr], + read_write: set[DataPtr], + outputs: set[DataPtr], + operator: str, + tensor_aliases: dict[int, list[str]], + ) -> list[SynchronizationError]: + def check_conflict( + data_ptr: DataPtr, current_access: Access, previous_access: Access | None + ) -> None: + if previous_access is None: + return + if not self.syncs.is_ordered_after( + current_access.stream, previous_access.seq_num, previous_access.stream + ): + error_list.append( + UnsynchronizedAccessError( + data_ptr, + self.tensors_accessed.get_allocation_stack_trace(data_ptr), + current_access, + previous_access, + ) + ) + + error_list: list[SynchronizationError] = [] + self.seq_num += 1 + self.syncs.update_seq_num(stream, self.seq_num) + stack_trace = traceback.StackSummary.extract( + traceback.walk_stack(inspect.currentframe()), lookup_lines=False + ) + # The stack trace generated in this way is in the inverse order, so it must be + # reversed. + stack_trace.reverse() + + for data_ptr in read_only: + self.tensors_accessed.ensure_tensor_exists(data_ptr) + current_access = Access( + AccessType.READ, + self.seq_num, + stream, + operator, + tensor_aliases[data_ptr], + data_ptr in outputs, + stack_trace, + ) + check_conflict( + data_ptr, current_access, self.tensors_accessed.get_write(data_ptr) + ) + self.tensors_accessed.add_read(data_ptr, current_access) + + for data_ptr in read_write: + self.tensors_accessed.ensure_tensor_exists(data_ptr) + current_access = Access( + AccessType.WRITE, + self.seq_num, + stream, + operator, + tensor_aliases[data_ptr], + data_ptr in outputs, + stack_trace, + ) + if self.tensors_accessed.were_there_reads_since_last_write(data_ptr): + for previous_access in self.tensors_accessed.get_reads(data_ptr): + check_conflict(data_ptr, current_access, previous_access) + else: + check_conflict( + data_ptr, current_access, self.tensors_accessed.get_write(data_ptr) + ) + self.tensors_accessed.set_write(data_ptr, current_access) + + return error_list + + def _handle_event_creation(self, event: EventId) -> None: + self.syncs.create_event(event) + + def _handle_event_deletion(self, event: EventId) -> None: + self.syncs.delete_event(event) + + def _handle_event_record(self, event: EventId, stream: StreamId) -> None: + self.syncs.record_state(event, stream) + + def _handle_event_wait(self, event: EventId, stream: StreamId) -> None: + self.syncs.stream_wait_for_event(stream, event) + + def _handle_memory_allocation(self, data_ptr: DataPtr) -> None: + self.tensors_accessed.ensure_tensor_does_not_exist(data_ptr) + stack_trace = traceback.StackSummary.extract( + traceback.walk_stack(inspect.currentframe()), lookup_lines=False + ) + # The stack trace generated in this way is in the inverse order, so it must be + # reversed. + stack_trace.reverse() + self.tensors_accessed.create_tensor( + data_ptr, + stack_trace, + ) + + def _handle_memory_deallocation(self, data_ptr: DataPtr) -> None: + self.tensors_accessed.ensure_tensor_exists(data_ptr) + self.tensors_accessed.delete_tensor(data_ptr) + + def _handle_stream_creation(self, stream: StreamId) -> None: + self.syncs.create_stream(stream) + + def _handle_device_synchronization(self) -> None: + self.syncs.sync_all_streams() + + def _handle_stream_synchronization(self, stream: StreamId) -> None: + self.syncs.all_streams_wait_for_stream(stream) + + def _handle_event_synchronization(self, event: EventId) -> None: + self.syncs.all_streams_wait_for_event(event) + + +def zip_by_key(a: dict[TK, TVa], b: dict[TK, TVb]) -> Iterator[tuple[TK, TVa, TVb]]: + for arg, value in a.items(): + if arg in b: + yield arg, value, b[arg] + + +def zip_arguments( + schema: torch.FunctionSchema, args: tuple[Any, ...], kwargs: dict[str, Any] +) -> Iterator[tuple[torch.Argument, Any]]: + schema_args = schema.arguments[: len(args)] + schema_kwargs = {arg.name: arg for arg in schema.arguments[len(args) :]} + + yield from zip(schema_args, args) + + for _, argument, value in zip_by_key(schema_kwargs, kwargs): + yield (argument, value) + + +class ArgumentHandler: + def __init__(self) -> None: + self.dataptrs_read: set[DataPtr] = set() + self.dataptrs_written: set[DataPtr] = set() + self.tensor_aliases: dict[DataPtr, list[str]] = {} + self.outputs: set[DataPtr] = set() + + def _handle_argument( + self, + value: Any, + is_write: bool, + metadata_only: bool, + name: str | None = None, + is_output: bool = False, + ) -> None: + if isinstance(value, torch.Tensor) and value.is_cuda: + # data_ptr() is preferred, but distinguish Tensors with null data_ptr() + # otherwise two empty Tensors could incorrectly match as a conflict + data_ptr = value.data_ptr() if value.data_ptr() else id(value) + if is_write: + self.dataptrs_written.add(data_ptr) + elif not metadata_only: + self.dataptrs_read.add(data_ptr) + + self.tensor_aliases.setdefault(data_ptr, []) + if name is not None: + self.tensor_aliases[data_ptr].append(name) + if is_output: + self.outputs.add(data_ptr) + + def parse_inputs( + self, + schema: torch.FunctionSchema, + args: tuple[Any, ...], + kwargs: dict[str, Any], + *, + is_factory: bool, + ) -> None: + for argument, value in zip_arguments(schema, args, kwargs): + is_write = argument.alias_info is not None and argument.alias_info.is_write + # A change is metadata only if it is a view or a factory function that + # reads only metadata + metadata_only = is_factory or ( + argument.alias_info is not None and not argument.alias_info.is_write + ) + pytree.tree_map_( + functools.partial( + self._handle_argument, + is_write=is_write, + name=argument.name, + metadata_only=metadata_only, + ), + value, + ) + + def parse_outputs( + self, schema: torch.FunctionSchema, outputs: Any, *, is_factory: bool + ) -> None: + for res, value in zip(schema.returns, (outputs,)): + metadata_only = is_factory or ( + res.alias_info is not None and not res.alias_info.is_write + ) + pytree.tree_map_( + functools.partial( + self._handle_argument, + is_write=not metadata_only, + is_output=True, + metadata_only=metadata_only, + ), + value, + ) + + +class CUDASanitizerDispatchMode(TorchDispatchMode): + def __init__(self) -> None: + self.event_handler = EventHandler() + torch._C._activate_gpu_trace() + gpu_trace.register_callback_for_event_creation( + self.event_handler._handle_event_creation + ) + gpu_trace.register_callback_for_event_deletion( + self.event_handler._handle_event_deletion + ) + gpu_trace.register_callback_for_event_record( + self.event_handler._handle_event_record + ) + gpu_trace.register_callback_for_event_wait( + self.event_handler._handle_event_wait + ) + gpu_trace.register_callback_for_memory_allocation( + self.event_handler._handle_memory_allocation + ) + gpu_trace.register_callback_for_memory_deallocation( + self.event_handler._handle_memory_deallocation + ) + gpu_trace.register_callback_for_stream_creation( + self.event_handler._handle_stream_creation + ) + gpu_trace.register_callback_for_device_synchronization( + self.event_handler._handle_device_synchronization + ) + gpu_trace.register_callback_for_stream_synchronization( + self.event_handler._handle_stream_synchronization + ) + gpu_trace.register_callback_for_event_synchronization( + self.event_handler._handle_event_synchronization + ) + + def __torch_dispatch__(self, func, types, args=(), kwargs=None): + if kwargs is None: + kwargs = {} + + # record_stream is not a kernel dispatch, skip it + if func is aten.record_stream.default: + return func(*args, **kwargs) + + is_factory = bool(FACTORY_FUNCTION_REGEX.match(func._schema.name)) + + argument_handler = ArgumentHandler() + argument_handler.parse_inputs(func._schema, args, kwargs, is_factory=is_factory) + + outputs = func(*args, **kwargs) + + argument_handler.parse_outputs(func._schema, outputs, is_factory=is_factory) + errors = self.event_handler._handle_kernel_launch( + torch.cuda.current_stream().cuda_stream, + argument_handler.dataptrs_read - argument_handler.dataptrs_written, + argument_handler.dataptrs_written, + argument_handler.outputs, + func._schema, + argument_handler.tensor_aliases, + ) + if errors: + for error in errors: + print(error, file=sys.stderr) + raise CUDASanitizerErrors(errors) + + return outputs + + +class CUDASanitizer: + """Manages the lifetime of a CUDASanitizer dispatch mode object. + + The CUDASanitizer class wraps the entering/exiting functions of the dispatch mode + context manager in the enable function/destructor, respectively. This is to + explicitly set the lifetime of the dispatch mode object to that of the application. + This approach was deemed more elegant than using the atexit module. + """ + + def __init__(self) -> None: + self.dispatch = CUDASanitizerDispatchMode() + self.enabled = False + + def enable(self): + self.dispatch.__enter__() + self.enabled = True + + def disable(self): + self.dispatch.__exit__(None, None, None) + self.enabled = False + + def __del__(self): + # Since this object lifetime is linked to the `torch.cuda._sanitizer` python + # module, it often gets deleted as part of the overall `torch` module cleanup + # At that time, depending on CPython version, the torch.* module might be in + # different states of being already cleaned up. + # Similarly other imports might already have been cleaned up so `sys` might + # be already gone as well. + # Skip exiting the mode if it outlived the runtime. + if (sys is not None) and (not sys.is_finalizing()) and self.enabled: + self.disable() + + +def enable_cuda_sanitizer(): + """Enable CUDA Sanitizer. + + The sanitizer will begin to analyze low-level CUDA calls invoked by torch functions + for synchronization errors. All data races found will be printed to the standard + error output along with stack traces of suspected causes. For best results, the + sanitizer should be enabled at the very beginning of the program. + """ + cuda_sanitizer.enable() + + +cuda_sanitizer = CUDASanitizer() diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/_utils.py b/.venv/lib/python3.12/site-packages/torch/cuda/_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..f9f0e979f996601b05f750fafa33c4c768665e07 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/_utils.py @@ -0,0 +1,634 @@ +import ctypes +import sys +from typing import Any + +import torch + + +try: + from cuda.bindings import ( # pyrefly: ignore[missing-import] + driver as _cuda_bindings_driver, + runtime as _cuda_bindings_runtime, + ) + + _HAS_CUDA_BINDINGS = True +except ImportError: + _cuda_bindings_driver = None # type: ignore[assignment] + _cuda_bindings_runtime = None # type: ignore[assignment] + _HAS_CUDA_BINDINGS = False + +# The _get_device_index has been moved to torch.utils._get_device_index +from torch._utils import _get_device_index as _torch_get_device_index + + +def _get_hip_runtime_library() -> ctypes.CDLL: + # If ROCm python packages are available, query the OS-independent absolute + # path to the library provided by those packages, including any version suffix. + # See https://github.com/ROCm/TheRock/blob/main/docs/packaging/python_packaging.md#dynamic-library-resolution + try: + # pyrefly: ignore [import-error, missing-import] + import rocm_sdk + + lib = ctypes.CDLL(str(rocm_sdk.find_libraries("amdhip64")[0])) + except (ImportError, IndexError): + if sys.platform == "win32": + lib = ctypes.CDLL(f"amdhip64_{torch.version.hip[0]}.dll") + else: # Unix-based systems + lib = ctypes.CDLL("libamdhip64.so") + + lib.cuGetErrorString = lib.hipGetErrorString # type: ignore[attr-defined] + lib.cuModuleLoadData = lib.hipModuleLoadData # type: ignore[attr-defined] + lib.cuModuleGetFunction = lib.hipModuleGetFunction # type: ignore[attr-defined] + lib.cuLaunchKernel = lib.hipModuleLaunchKernel # type: ignore[attr-defined] + lib.cuFuncSetAttribute = lib.hipFuncSetAttribute # type: ignore[attr-defined] + return lib + + +def _get_cuda_library() -> ctypes.CDLL: + if sys.platform == "win32": + return ctypes.CDLL("nvcuda.dll") + else: # Unix-based systems + return ctypes.CDLL("libcuda.so.1") + + +# Load GPU driver runtime +def _get_gpu_runtime_library() -> ctypes.CDLL: + if torch.version.hip: + return _get_hip_runtime_library() + else: + return _get_cuda_library() + + +# Helper: check CUDA errors +def _check_cuda(result: int) -> None: + if result == 0: + return + err_str = ctypes.c_char_p() + libcuda = _get_gpu_runtime_library() # Get reference to CUDA library + libcuda.cuGetErrorString(result, ctypes.byref(err_str)) + error_message = ( + err_str.value.decode() if err_str.value is not None else "Unknown CUDA error" + ) + raise RuntimeError(f"CUDA error: {error_message}") + + +def _check_cuda_bindings(result: Any) -> Any: + """Check a cuda.bindings (cuda-python) call result for errors. + + All cuda.bindings driver/runtime calls return ``(error, *outputs)``. This + helper unpacks the tuple, raises on non-success, and returns the + outputs (``None`` for zero outputs, scalar for one, tuple otherwise). + """ + if not _HAS_CUDA_BINDINGS: + raise RuntimeError("cuda.bindings is not available") + # We assume that `result`` is a return value from cuda bindings API calls. + # This means it is always a tuple, of length >= 1 and the first element + # represents the error with a `value` attribute. + # See also cuda-python error handling here: + # https://github.com/NVIDIA/cuda-python/blob/88363f8f17636ecd28772b2b8ad41a33895b41bb/ + # cuda_bindings/cuda/bindings/_example_helpers/helper_cuda.py#L11-L31 + err, *out = result + if err.value: + # pyrefly: ignore [missing-attribute] + if isinstance(err, _cuda_bindings_driver.CUresult): + # pyrefly: ignore [missing-attribute] + err2, name = _cuda_bindings_driver.cuGetErrorString(err) + # pyrefly: ignore [missing-attribute] + success = _cuda_bindings_driver.CUresult.CUDA_SUCCESS + # cuGetErrorString can fail and leave name NULL. cudaGetErrorString + # returns a fallback string for unknown runtime errors. + err_str = name if err2 == success else "" + # pyrefly: ignore [missing-attribute] + elif isinstance(err, _cuda_bindings_runtime.cudaError_t): + # pyrefly: ignore [missing-attribute] + err_str = _cuda_bindings_runtime.cudaGetErrorString(err)[1] + else: + err_str = "unknown error type" + if isinstance(err_str, bytes): + err_str = err_str.decode() + raise RuntimeError(f"CUDA error: {err} ({err_str})") + + if len(out) == 0: + return None + if len(out) == 1: + return out[0] + return out + + +def _check_cuda_bindings_driver(result: Any) -> Any: + """Check a cuda.bindings (cuda-python) driver call result for errors. + + Like ``_check_cuda_bindings`` but for the CUDA driver API, whose calls + return ``(CUresult, *outputs)`` and report errors via ``cuGetErrorString``. + """ + if not _HAS_CUDA_BINDINGS: + raise RuntimeError("cuda.bindings is not available") + err, *out = result + if ( + err + != _cuda_bindings_driver.CUresult.CUDA_SUCCESS # pyrefly: ignore[missing-attribute] + ): + _, err_str = ( + _cuda_bindings_driver.cuGetErrorString( # pyrefly: ignore[missing-attribute] + err + ) + ) + if isinstance(err_str, bytes): + err_str = err_str.decode() + raise RuntimeError(f"CUDA driver error: {err} ({err_str})") + if len(out) == 0: + return None + if len(out) == 1: + return out[0] + return out + + +def _get_hiprtc_library() -> ctypes.CDLL: + try: + # pyrefly: ignore [import-error, missing-import] + import rocm_sdk + + lib = ctypes.CDLL(str(rocm_sdk.find_libraries("hiprtc")[0])) + except (ImportError, IndexError): + if sys.platform == "win32": + version_str = "".join( + ["0", torch.version.hip[0], "0", torch.version.hip[2]] + ) + lib = ctypes.CDLL(f"hiprtc{version_str}.dll") + else: + lib = ctypes.CDLL("libhiprtc.so") + + # Provide aliases for HIP RTC functions to match NVRTC API + lib.nvrtcGetErrorString = lib.hiprtcGetErrorString # type: ignore[attr-defined] + lib.nvrtcCreateProgram = lib.hiprtcCreateProgram # type: ignore[attr-defined] + lib.nvrtcDestroyProgram = lib.hiprtcDestroyProgram # type: ignore[attr-defined] + lib.nvrtcCompileProgram = lib.hiprtcCompileProgram # type: ignore[attr-defined] + lib.nvrtcGetCUBINSize = lib.hiprtcGetCodeSize # type: ignore[attr-defined] + lib.nvrtcGetCUBIN = lib.hiprtcGetCode # type: ignore[attr-defined] + lib.nvrtcGetProgramLogSize = lib.hiprtcGetProgramLogSize # type: ignore[attr-defined] + lib.nvrtcGetProgramLog = lib.hiprtcGetProgramLog # type: ignore[attr-defined] + lib.nvrtcAddNameExpression = lib.hiprtcAddNameExpression # type: ignore[attr-defined] + lib.nvrtcGetLoweredName = lib.hiprtcGetLoweredName # type: ignore[attr-defined] + return lib + + +def _get_nvrtc_library() -> ctypes.CDLL: + major_version = int(torch.version.cuda.split(".")[0]) # type: ignore[union-attr] + if sys.platform == "win32": + nvrtc_libs = [ + f"nvrtc64_{major_version}0_0.dll", + ] + else: + nvrtc_libs = [ + f"libnvrtc.so.{major_version}", + "libnvrtc.so", # Fallback to unversioned + ] + for lib_name in nvrtc_libs: + try: + return ctypes.CDLL(lib_name) + except OSError: + continue + raise OSError("Could not find any NVRTC library") + + +def _get_gpu_rtc_library() -> ctypes.CDLL: + # Since PyTorch already loads the GPU RTC library, we can use the system library + # which should be compatible with PyTorch's version + if torch.version.hip: + return _get_hiprtc_library() + else: + return _get_nvrtc_library() + + +def _get_gpu_rtc_compatible_flags() -> list[str]: + """ + Get HIPCC/NVCC flags that are compatible with NVRTC compilation. + + Returns: + List of HIPCC/NVCC flags that can be safely used with NVRTC. + """ + from torch.utils.cpp_extension import COMMON_HIPCC_FLAGS, COMMON_NVCC_FLAGS + + nvrtc_unsupported_flags = { + "--expt-relaxed-constexpr", + } + + # Filter out unsupported flags + compatible_flags = [ + flag for flag in COMMON_NVCC_FLAGS if flag not in nvrtc_unsupported_flags + ] + + if torch.version.hip: + compatible_flags.extend(COMMON_HIPCC_FLAGS) + + return compatible_flags + + +def _nvrtc_compile( + kernel_source: str, + kernel_name: str, + compute_capability: str | None = None, + cuda_include_dirs: list | None = None, + nvcc_options: list | None = None, + auto_pch: bool = False, +) -> tuple[bytes, str]: + """ + Compiles a CUDA kernel using NVRTC and returns the PTX code. + + Args: + kernel_source (str): The CUDA kernel source code as a string + kernel_name (str): The name of the kernel function to compile + compute_capability (str, None): The compute capability to target (e.g., "86"). + If None, will detect from current device. + cuda_include_dirs (list, None): List of directories containing CUDA headers + nvcc_options (list, None): Additional options to pass to NVRTC + auto_pch (bool): Enable automatic precompiled headers (CUDA 12.8+) + + Returns: + Tuple[bytes, str]: The compiled PTX code and mangled kernel name + """ + # Ensure CUDA is initialized + import torch.cuda + + # Load NVRTC library + libnvrtc = _get_gpu_rtc_library() + + # NVRTC constants + NVRTC_SUCCESS = 0 + + # Helper: check NVRTC errors + def check_nvrtc(result: int) -> None: + if result != NVRTC_SUCCESS: + err_str = ctypes.c_char_p() + libnvrtc.nvrtcGetErrorString(result, ctypes.byref(err_str)) + error_message = ( + err_str.value.decode() + if err_str.value is not None + else "Unknown CUDA error" + ) + raise RuntimeError(f"CUDA error: {error_message}") + + # Convert source to bytes + source_bytes = kernel_source.encode("utf-8") + + # Get compute capability if not provided + if compute_capability is None: + props = torch.cuda.get_device_properties(torch.cuda.current_device()) + if torch.version.hip: + compute_capability = f"{props.gcnArchName}" + else: + compute_capability = f"{props.major}{props.minor}" + + # Prepare compilation options + options = [] + if torch.version.hip: + options.append(f"--offload-arch={compute_capability}".encode()) + else: + options.append(f"--gpu-architecture=sm_{compute_capability}".encode()) + + # Auto-detect and add CUDA include paths + from torch.utils.cpp_extension import include_paths + + cuda_include_paths = include_paths("cuda") + for cuda_path in cuda_include_paths: + options.append(f"-I{cuda_path}".encode()) + + # Add custom include directories + if cuda_include_dirs: + for directory in cuda_include_dirs: + options.append(f"-I{directory}".encode()) + + # Enable automatic precompiled headers (CUDA 12.8+) + if auto_pch: + if str(torch.version.cuda) < "12.8": + raise AssertionError(f"PCH requires CUDA 12.8+, got {torch.version.cuda}") + if nvcc_options is None: + nvcc_options = [] + nvcc_options.append("--pch") + + # Add custom NVCC options + if nvcc_options: + for option in nvcc_options: + options.append(option.encode("utf-8")) + + nvrtc_compatible_flags = _get_gpu_rtc_compatible_flags() + options.extend([flag.encode("utf-8") for flag in nvrtc_compatible_flags]) + + # Convert options to C array + num_options = len(options) + options_array = (ctypes.c_char_p * num_options)(*options) + + # Create program + prog = ctypes.c_void_p() + check_nvrtc( + libnvrtc.nvrtcCreateProgram( + ctypes.byref(prog), + source_bytes, + f"{kernel_name}.cu".encode(), + 0, + None, + None, + ) + ) + + # Add kernel name, which can be a template expression + c_kernel_name = kernel_name.encode("utf-8") + check_nvrtc(libnvrtc.nvrtcAddNameExpression(prog, c_kernel_name)) + + # Compile program + res = libnvrtc.nvrtcCompileProgram(prog, num_options, options_array) + + # Handle compilation errors + if res != NVRTC_SUCCESS: + # Get log + log_size = ctypes.c_size_t() + libnvrtc.nvrtcGetProgramLogSize(prog, ctypes.byref(log_size)) + log = ctypes.create_string_buffer(log_size.value) + libnvrtc.nvrtcGetProgramLog(prog, log) + raise RuntimeError(f"Kernel compilation failed:\n{log.value.decode()}") + + # Get binary + binary_size = ctypes.c_size_t() + check_nvrtc(libnvrtc.nvrtcGetCUBINSize(prog, ctypes.byref(binary_size))) + binary = ctypes.create_string_buffer(binary_size.value) + check_nvrtc(libnvrtc.nvrtcGetCUBIN(prog, binary)) + + # Get mangled name + c_mangled_name = ctypes.c_char_p() + check_nvrtc( + libnvrtc.nvrtcGetLoweredName(prog, c_kernel_name, ctypes.byref(c_mangled_name)) + ) + if c_mangled_name.value is not None: + mangled_name = c_mangled_name.value.decode() # make a copy + else: + mangled_name = "" + + libnvrtc.nvrtcDestroyProgram(ctypes.byref(prog)) + + # For some reason, ".value" causes the string to be truncated, + # likely due to the presence of '\0' in the string. So we use .raw instead. + return binary.raw, mangled_name + + +class _CudaModule: + def __init__(self, module: ctypes.c_void_p) -> None: + self._module = module + self._kernels: dict[str, _CudaKernel] = {} + + def __getattr__(self, name: str) -> "_CudaKernel": + if name in self._kernels: + return self._kernels[name] + + # Import the CUDA library inside the method + # pyrefly: ignore [missing-module-attribute] + from torch.cuda._utils import _get_gpu_runtime_library + + libcuda = _get_gpu_runtime_library() + + func = ctypes.c_void_p() + try: + _check_cuda( + libcuda.cuModuleGetFunction( + ctypes.byref(func), self._module, name.encode("utf-8") + ) + ) + kernel = _CudaKernel(func, self._module) + self._kernels[name] = kernel + return kernel + + except RuntimeError as err: + raise AttributeError(f"No kernel named '{name}' in this module") from err + + +class _CudaKernel: + """ + Represents a compiled CUDA kernel that can be called with PyTorch tensors. + """ + + def __init__(self, func: ctypes.c_void_p, module: ctypes.c_void_p) -> None: + self.func = func + self.module = module + self._max_shared_mem_bytes = 0 + + def __call__( + self, + grid: tuple[int, int, int] = (1, 1, 1), + block: tuple[int, int, int] = (1, 1, 1), + args: list | None = None, + shared_mem: int = 0, + stream: Any | None = None, + ) -> None: + """ + Call the compiled CUDA kernel + + Args: + grid (tuple): Grid dimensions (grid_x, grid_y, grid_z) + block (tuple): Block dimensions (block_x, block_y, block_z) + args (list): List of arguments to pass to the kernel. + PyTorch tensor arguments will be automatically converted to pointers. + shared_mem (int): Shared memory size in bytes + stream (torch.cuda.Stream): CUDA stream to use. If None, uses current stream. + """ + import torch + + libcuda = torch.cuda._utils._get_gpu_runtime_library() + + if not args: + args = [] + + # Process arguments and convert tensors to pointers + processed_args: list[ctypes.c_void_p] = [] + c_args = [] + + for arg in args: + if isinstance(arg, torch.Tensor): + if not arg.is_cuda and not (arg.is_cpu and arg.is_pinned()): + raise ValueError( + "All tensor arguments must be CUDA tensors or pinned CPU tensors" + ) + # Get pointer to tensor data + ptr = ctypes.c_void_p(arg.data_ptr()) + processed_args.append(ptr) + c_args.append(ctypes.byref(ptr)) + elif isinstance(arg, int): + # Convert integers to C int + c_int = ctypes.c_int(arg) + # Store the C int for reference keeping, not in processed_args + c_args.append(ctypes.byref(c_int)) + elif isinstance(arg, float): + # Python floats are doubles - use double by default + c_double = ctypes.c_double(arg) + # Store the C double for reference keeping, not in processed_args + c_args.append(ctypes.byref(c_double)) + else: + raise TypeError(f"Unsupported argument type: {type(arg)}") + + # Convert to array of void pointers + c_args_array = (ctypes.c_void_p * len(c_args))() + for i, arg in enumerate(c_args): + c_args_array[i] = ctypes.cast(arg, ctypes.c_void_p) + + # Get the stream + if stream is None: + # Defer import to avoid circular imports + import torch.cuda + + stream = torch.cuda.current_stream() + + # Check if kernel requires large shared memory but hasn't been configured + if shared_mem >= 48 * 1024 and ( + self._max_shared_mem_bytes == 0 or shared_mem > self._max_shared_mem_bytes + ): + configured_msg = ( + "not configured" + if self._max_shared_mem_bytes == 0 + else f"only {self._max_shared_mem_bytes} bytes configured" + ) + raise RuntimeError( + f"Kernel requires {shared_mem} bytes of shared memory (>= 48KB), " + f"but {configured_msg}. " + "Call kernel.set_shared_memory_config(shared_mem) after compilation " + "and before launching the kernel." + ) + + _check_cuda( + libcuda.cuLaunchKernel( + self.func, + grid[0], + grid[1], + grid[2], + block[0], + block[1], + block[2], + shared_mem, + stream._as_parameter_, + c_args_array, + None, + ) + ) + + def set_shared_memory_config(self, shared_mem_bytes: int) -> None: + if shared_mem_bytes < 48 * 1024: + # No configuration needed for <= 48KB, just update the value + self._max_shared_mem_bytes = shared_mem_bytes + return + + libcuda = _get_gpu_runtime_library() + + # Get device properties to validate against limits + device_props = torch.cuda.get_device_properties() + # HIP doesn't have shared_memory_per_block_optin in device properties, so we hard-code it here + if torch.version.hip: + # navi, CDNA1-CDNA3 allows a max of 64KB shared memory, + # CDNA4 (gfx950) 160KB, and CDNA5 (gfx1250) 320KB. + gcn_arch = device_props.gcnArchName.split(":", 1)[0] + if gcn_arch == "gfx950": + max_shared_mem = 160 * 1024 + elif gcn_arch == "gfx1250": + max_shared_mem = 320 * 1024 + else: + max_shared_mem = 65536 + else: + max_shared_mem = getattr( + device_props, "shared_memory_per_block_optin", 49152 + ) + + if shared_mem_bytes > max_shared_mem: + raise RuntimeError( + f"Requested shared memory ({shared_mem_bytes} bytes) exceeds " + f"device limit ({max_shared_mem} bytes). " + "Consider reducing block size or shared memory usage." + ) + + # Set the function attribute once + # https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html + cudaFuncAttributeMaxDynamicSharedMemorySize = 8 + _check_cuda( + libcuda.cuFuncSetAttribute( + self.func, + cudaFuncAttributeMaxDynamicSharedMemorySize, + shared_mem_bytes, + ) + ) + + self._max_shared_mem_bytes = shared_mem_bytes + + +def _cuda_load_module( + ptx: str | bytes, kernel_names: list[str] | None = None +) -> _CudaModule | dict[str, "_CudaKernel"]: + """ + Loads a CUDA module from PTX code and returns a module object that can access kernels. + + Args: + ptx (bytes or str): The PTX code to load + kernel_names (list, optional): List of kernel names to extract from the module. + If None, will return a module object with __getattr__. + + Returns: + object: If kernel_names is None, returns a module object with __getattr__ to access kernels. + If kernel_names is provided, returns a dict mapping kernel names to _CudaKernel objects. + """ + # Ensure CUDA is initialized + import torch.cuda + + # Load CUDA driver library + libcuda = _get_gpu_runtime_library() + + # Convert PTX to bytes if it's a string + if isinstance(ptx, str): + ptx = ptx.encode("utf-8") + + # Load PTX module + module = ctypes.c_void_p() + # Get the current stream without directly importing torch.cuda at module level + stream = torch.cuda.current_stream() + with stream: + _check_cuda(libcuda.cuModuleLoadData(ctypes.byref(module), ptx)) + + if not kernel_names: + return _CudaModule(module) + + # Return specific kernels + kernels = {} + for name in kernel_names: + func = ctypes.c_void_p() + _check_cuda( + libcuda.cuModuleGetFunction( + ctypes.byref(func), module, name.encode("utf-8") + ) + ) + kernels[name] = _CudaKernel(func, module) + return kernels + + +def _get_device_index( + device: Any, optional: bool = False, allow_cpu: bool = False +) -> int: + r"""Get the device index from :attr:`device`, which can be a torch.device object, a Python integer, or ``None``. + + If :attr:`device` is a torch.device object, returns the device index if it + is a CUDA device. Note that for a CUDA device without a specified index, + i.e., ``torch.device('cuda')``, this will return the current default CUDA + device if :attr:`optional` is ``True``. If :attr:`allow_cpu` is ``True``, + CPU devices will be accepted and ``-1`` will be returned in this case. + + If :attr:`device` is a Python integer, it is returned as is. + + If :attr:`device` is ``None``, this will return the current default CUDA + device if :attr:`optional` is ``True``. + """ + if isinstance(device, int): + return device + if isinstance(device, str): + device = torch.device(device) + if isinstance(device, torch.device): + if allow_cpu: + if device.type not in ["cuda", "cpu"]: + raise ValueError(f"Expected a cuda or cpu device, but got: {device}") + elif device.type != "cuda": + raise ValueError(f"Expected a cuda device, but got: {device}") + if not torch.jit.is_scripting(): + if isinstance(device, torch.cuda.device): + return device.idx + return _torch_get_device_index(device, optional, allow_cpu) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/__init__.py b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..88ef0b5acac5e5bdeb034169052bcf5aa7456e33 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__init__.py @@ -0,0 +1,13 @@ +# pyrefly: ignore [deprecated] +from .autocast_mode import autocast, custom_bwd, custom_fwd +from .common import amp_definitely_not_available +from .grad_scaler import GradScaler + + +__all__ = [ + "amp_definitely_not_available", + "autocast", + "custom_bwd", + "custom_fwd", + "GradScaler", +] diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d010dc2ee61bc059f184f4007eaa7d29f2337aaa Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/autocast_mode.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/autocast_mode.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b9ab7e221e6d16b0ea46fbec7411665e633be77b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/autocast_mode.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/common.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/common.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..284a506d28377cd1a8f63284ef3d6a5ee153ef19 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/common.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/grad_scaler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/grad_scaler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ea38359e738f2ed4a04698f857cb90d2e3126382 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/cuda/amp/__pycache__/grad_scaler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/autocast_mode.py b/.venv/lib/python3.12/site-packages/torch/cuda/amp/autocast_mode.py new file mode 100644 index 0000000000000000000000000000000000000000..e6b63c708d3f2ddfc162a4431e114a2bcf47e9eb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/amp/autocast_mode.py @@ -0,0 +1,110 @@ +# mypy: allow-untyped-defs +import functools +import sys +from typing import Any +from typing_extensions import deprecated + +import torch + + +__all__ = ["autocast", "custom_fwd", "custom_bwd"] + + +@deprecated( + "`torch.cuda.amp.autocast(args...)` is deprecated. " + "Please use `torch.amp.autocast('cuda', args...)` instead.", + category=FutureWarning, +) +class autocast(torch.amp.autocast_mode.autocast): + r"""See :class:`torch.autocast`. + + ``torch.cuda.amp.autocast(args...)`` is deprecated. Please use ``torch.amp.autocast("cuda", args...)`` instead. + """ + + # TODO: remove this conditional once we stop supporting Python < 3.13 + # Prior to Python 3.13, inspect.signature could not retrieve the correct + # signature information for classes decorated with @deprecated (unless + # the __new__ static method was explicitly defined); + # + # However, this issue has been fixed in Python 3.13 and later versions. + if sys.version_info < (3, 13): + + def __new__( + cls, + enabled: bool = True, + dtype: torch.dtype = torch.float16, + cache_enabled: bool = True, + ): + return super().__new__(cls) + + def __init_subclass__(cls): + pass + + def __init__( + self, + enabled: bool = True, + dtype: torch.dtype = torch.float16, + cache_enabled: bool = True, + ): + if torch._jit_internal.is_scripting(): + self._enabled = enabled + self.device = "cuda" + self.fast_dtype = dtype + return + super().__init__( + "cuda", enabled=enabled, dtype=dtype, cache_enabled=cache_enabled + ) + + def __enter__(self): + if torch._jit_internal.is_scripting(): + return self + return super().__enter__() + + # TODO: discuss a unified TorchScript-friendly API for autocast + def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any): # type: ignore[override] + if torch._jit_internal.is_scripting(): + return + return super().__exit__(exc_type, exc_val, exc_tb) + + def __call__(self, func): + if torch._jit_internal.is_scripting(): + return func + return super().__call__(func) + + +# Preserved only for BC reasons +@deprecated( + "`torch.cuda.amp.autocast_mode._cast(value, dtype)` is deprecated. " + "Please use `torch.amp.autocast_mode._cast(value, 'cuda', dtype)` instead.", + category=FutureWarning, +) +def _cast(value, dtype): + return torch.amp.autocast_mode._cast(value, "cuda", dtype) + + +@deprecated( + "`torch.cuda.amp.custom_fwd(args...)` is deprecated. " + "Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead.", + category=FutureWarning, +) +def custom_fwd(fwd=None, *, cast_inputs=None): + """ + ``torch.cuda.amp.custom_fwd(args...)`` is deprecated. Please use + ``torch.amp.custom_fwd(args..., device_type='cuda')`` instead. + """ + return functools.partial(torch.amp.custom_fwd, device_type="cuda")( + fwd=fwd, cast_inputs=cast_inputs + ) + + +@deprecated( + "`torch.cuda.amp.custom_bwd(args...)` is deprecated. " + "Please use `torch.amp.custom_bwd(args..., device_type='cuda')` instead.", + category=FutureWarning, +) +def custom_bwd(bwd): + """ + ``torch.cuda.amp.custom_bwd(args...)`` is deprecated. Please use + ``torch.amp.custom_bwd(args..., device_type='cuda')`` instead. + """ + return functools.partial(torch.amp.custom_bwd, device_type="cuda")(bwd) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/common.py b/.venv/lib/python3.12/site-packages/torch/cuda/amp/common.py new file mode 100644 index 0000000000000000000000000000000000000000..915a9b4f4a9ca6c147abefd7c8ab1891ee5a8179 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/amp/common.py @@ -0,0 +1,11 @@ +# mypy: allow-untyped-defs +from importlib.util import find_spec + +import torch + + +__all__ = ["amp_definitely_not_available"] + + +def amp_definitely_not_available(): + return not (torch.cuda.is_available() or find_spec("torch_xla")) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/amp/grad_scaler.py b/.venv/lib/python3.12/site-packages/torch/cuda/amp/grad_scaler.py new file mode 100644 index 0000000000000000000000000000000000000000..62e2020073c8ed99f7295edd1aaea4c54d815f63 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/amp/grad_scaler.py @@ -0,0 +1,38 @@ +from typing_extensions import deprecated + +import torch + +# We need to keep this unused import for BC reasons +from torch.amp.grad_scaler import OptState # noqa: F401 + + +__all__ = ["GradScaler"] + + +class GradScaler(torch.amp.GradScaler): + r""" + See :class:`torch.amp.GradScaler`. + ``torch.cuda.amp.GradScaler(args...)`` is deprecated. Please use ``torch.amp.GradScaler("cuda", args...)`` instead. + """ + + @deprecated( + "`torch.cuda.amp.GradScaler(args...)` is deprecated. " + "Please use `torch.amp.GradScaler('cuda', args...)` instead.", + category=FutureWarning, + ) + def __init__( + self, + init_scale: float = 2.0**16, + growth_factor: float = 2.0, + backoff_factor: float = 0.5, + growth_interval: int = 2000, + enabled: bool = True, + ) -> None: + super().__init__( + "cuda", + init_scale=init_scale, + growth_factor=growth_factor, + backoff_factor=backoff_factor, + growth_interval=growth_interval, + enabled=enabled, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/comm.py b/.venv/lib/python3.12/site-packages/torch/cuda/comm.py new file mode 100644 index 0000000000000000000000000000000000000000..2915de5d090fd18c82540beedb9971a0b7b6cc3e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/comm.py @@ -0,0 +1,19 @@ +# The functions here have been moved to torch.nn.parallel.comm +from torch.nn.parallel.comm import ( + broadcast, + broadcast_coalesced, + gather, + reduce_add, + reduce_add_coalesced, + scatter, +) + + +__all__ = [ + "broadcast", + "broadcast_coalesced", + "reduce_add", + "reduce_add_coalesced", + "scatter", + "gather", +] diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/gds.py b/.venv/lib/python3.12/site-packages/torch/cuda/gds.py new file mode 100644 index 0000000000000000000000000000000000000000..d9ece65ad429e4cb745de9e55c7db5ef02c0271a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/gds.py @@ -0,0 +1,187 @@ +import os +import sys +from collections.abc import Callable + +import torch +from torch.types import Storage + + +__all__: list[str] = [ + "is_available", + "gds_register_buffer", + "gds_deregister_buffer", + "GdsFile", +] + + +def _dummy_fn(name: str) -> Callable: + def fn(*args, **kwargs): # type: ignore[no-untyped-def] + raise RuntimeError(f"torch._C.{name} is not supported on this platform") + + return fn + + +def is_available() -> bool: + r"""Return ``True`` if GDS (GPUDirect Storage) support is built in. + + This requires PyTorch to be built with cuFile (CUDA) or hipFile (ROCm). + """ + return torch._C._has_gds + + +# Without GDS built in, install dummy stubs so the API raises a clear error. +if not is_available(): + if hasattr(torch._C, "_gds_deregister_buffer"): + raise AssertionError( + "_gds_deregister_buffer exists but _gds_register_buffer does not" + ) + if hasattr(torch._C, "_gds_register_handle"): + raise AssertionError( + "_gds_register_handle exists but _gds_register_buffer does not" + ) + if hasattr(torch._C, "_gds_deregister_handle"): + raise AssertionError( + "_gds_deregister_handle exists but _gds_register_buffer does not" + ) + if hasattr(torch._C, "_gds_load_storage"): + raise AssertionError( + "_gds_load_storage exists but _gds_register_buffer does not" + ) + if hasattr(torch._C, "_gds_save_storage"): + raise AssertionError( + "_gds_save_storage exists but _gds_register_buffer does not" + ) + # Define functions + torch._C.__dict__["_gds_register_buffer"] = _dummy_fn("_gds_register_buffer") + torch._C.__dict__["_gds_deregister_buffer"] = _dummy_fn("_gds_deregister_buffer") + torch._C.__dict__["_gds_register_handle"] = _dummy_fn("_gds_register_handle") + torch._C.__dict__["_gds_deregister_handle"] = _dummy_fn("_gds_deregister_handle") + torch._C.__dict__["_gds_load_storage"] = _dummy_fn("_gds_load_storage") + torch._C.__dict__["_gds_save_storage"] = _dummy_fn("_gds_save_storage") + + +def gds_register_buffer(s: Storage) -> None: + """Registers a storage on a CUDA device as a cufile buffer. + + Example:: + + >>> # xdoctest: +SKIP("gds filesystem requirements") + >>> src = torch.randn(1024, device="cuda") + >>> s = src.untyped_storage() + >>> gds_register_buffer(s) + + Args: + s (Storage): Buffer to register. + """ + torch._C._gds_register_buffer(s) + + +def gds_deregister_buffer(s: Storage) -> None: + """Deregisters a previously registered storage on a CUDA device as a cufile buffer. + + Example:: + + >>> # xdoctest: +SKIP("gds filesystem requirements") + >>> src = torch.randn(1024, device="cuda") + >>> s = src.untyped_storage() + >>> gds_register_buffer(s) + >>> gds_deregister_buffer(s) + + Args: + s (Storage): Buffer to register. + """ + torch._C._gds_deregister_buffer(s) + + +class GdsFile: + r"""Wrapper around cuFile. + + cuFile is a file-like interface to the GPUDirect Storage (GDS) API. + + See the `cufile docs `_ + for more details. + + Args: + filename (str): Name of the file to open. + flags (int): Flags to pass to ``os.open`` when opening the file. ``os.O_DIRECT`` will + be added automatically. + + Example:: + + >>> # xdoctest: +SKIP("gds filesystem requirements") + >>> src1 = torch.randn(1024, device="cuda") + >>> src2 = torch.randn(2, 1024, device="cuda") + >>> file = torch.cuda.gds.GdsFile(f, os.O_CREAT | os.O_RDWR) + >>> file.save_storage(src1.untyped_storage(), offset=0) + >>> file.save_storage(src2.untyped_storage(), offset=src1.nbytes) + >>> dest1 = torch.empty(1024, device="cuda") + >>> dest2 = torch.empty(2, 1024, device="cuda") + >>> file.load_storage(dest1.untyped_storage(), offset=0) + >>> file.load_storage(dest2.untyped_storage(), offset=src1.nbytes) + >>> torch.equal(src1, dest1) + True + >>> torch.equal(src2, dest2) + True + + """ + + def __init__(self, filename: str, flags: int): + if sys.platform == "win32": + raise RuntimeError("GdsFile is not supported on this platform.") + self.filename = filename + self.flags = flags + self.fd = os.open(filename, flags | os.O_DIRECT) # type: ignore[attr-defined] + self.handle: int | None = None + self.register_handle() + + def __del__(self) -> None: + if self.handle is not None: + self.deregister_handle() + os.close(self.fd) + + def register_handle(self) -> None: + """Registers file descriptor to cuFile Driver. + + This is a wrapper around ``cuFileHandleRegister``. + """ + if self.handle is not None: + raise AssertionError("Cannot register a handle that is already registered.") + self.handle = torch._C._gds_register_handle(self.fd) + + def deregister_handle(self) -> None: + """Deregisters file descriptor from cuFile Driver. + + This is a wrapper around ``cuFileHandleDeregister``. + """ + if self.handle is None: + raise AssertionError("Cannot deregister a handle that is not registered.") + torch._C._gds_deregister_handle(self.handle) + self.handle = None + + def load_storage(self, storage: Storage, offset: int = 0) -> None: + """Loads data from the file into the storage. + + This is a wrapper around ``cuFileRead``. ``storage.nbytes()`` of data + will be loaded from the file at ``offset`` into the storage. + + Args: + storage (Storage): Storage to load data into. + offset (int, optional): Offset into the file to start loading from. (Default: 0) + """ + if self.handle is None: + raise AssertionError("Cannot load data from a file that is not registered.") + torch._C._gds_load_storage(self.handle, storage, offset) + + def save_storage(self, storage: Storage, offset: int = 0) -> None: + """Saves data from the storage into the file. + + This is a wrapper around ``cuFileWrite``. All bytes of the storage + will be written to the file at ``offset``. + + Args: + storage (Storage): Storage to save data from. + offset (int, optional): Offset into the file to start saving to. (Default: 0) + """ + if self.handle is None: + raise AssertionError("Cannot save data to a file that is not registered.") + torch._C._gds_save_storage(self.handle, storage, offset) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/graph_annotations.py b/.venv/lib/python3.12/site-packages/torch/cuda/graph_annotations.py new file mode 100644 index 0000000000000000000000000000000000000000..132b7ee2c278d7ac9adfd449a0b5abc78de1fb3e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/graph_annotations.py @@ -0,0 +1,57 @@ +r"""Annotate kernels captured in CUDA graphs for profiler traces. + +During CUDA graph capture, :func:`mark_kernels` tags the GPU work nodes +(kernels, memcpys, memsets) captured within its scope with user metadata. +The recorded annotations are keyed by the same ``graph node id`` values +that CUPTI-based profilers attach to kernel events during graph replay, so +the annotations can later be joined against a profiler trace to identify +which kernels belong to which region of the captured workload. + +Annotation recording is enabled per capture via the ``enable_annotations`` +argument of :class:`torch.cuda.graph`:: + + import torch + from torch.cuda.graph_annotations import mark_kernels, get_kernel_annotations + + g = torch.cuda.CUDAGraph() + with torch.cuda.graph(g, enable_annotations=True): + with mark_kernels("phase_A"): + y = workload_a(x) + with mark_kernels({"name": "phase_B", "dtype": "bf16"}): + z = workload_b(y) + + annotations = get_kernel_annotations() + +The annotation mapping is typically pickled and joined against a profiler +trace offline (matching each kernel event's ``graph node id`` field). + +Requires the ``cuda-bindings`` package and a CUDA driver that supports +``cudaGraphNodeGetToolsId`` (CUDA >= 13.1, or an equivalent cuda-compat +package). When unavailable, recording silently degrades to a no-op; use +:func:`is_available` to check support programmatically. + +.. warning:: + This API is in prototype and may change in future releases. +""" + +from torch.cuda._graph_annotations import ( + clear_kernel_annotations, + get_kernel_annotations, + is_available, + mark_kernels, +) + + +__all__ = [ + "clear_kernel_annotations", + "get_kernel_annotations", + "is_available", + "mark_kernels", +] + +# The implementation lives in the private module (which predates this one and +# has external users); re-export and claim the names so they present as +# torch.cuda.graph_annotations APIs (test_correct_module_names requires it). +for _name in __all__: + globals()[_name].__module__ = "torch.cuda.graph_annotations" +del _name diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/graphs.py b/.venv/lib/python3.12/site-packages/torch/cuda/graphs.py new file mode 100644 index 0000000000000000000000000000000000000000..18f3d555cac2fe32270011ab65735d1cddaa1a40 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/graphs.py @@ -0,0 +1,1598 @@ +# pylint: disable=useless-parent-delegation +from __future__ import annotations + +import ctypes +import gc +import typing +import warnings +import weakref +from collections import OrderedDict +from collections.abc import Callable +from typing import overload, TYPE_CHECKING, TypeAlias, TypeGuard, Union +from typing_extensions import ParamSpec, Self, TypeVar + +import torch +from torch import Tensor +from torch.cuda._utils import _check_cuda_bindings + + +try: + from cuda.bindings import ( # pyrefly: ignore[missing-import] + driver as _cuda_driver, + runtime as _cuda_runtime, + ) +except ImportError: + _cuda_driver = None # type: ignore[assignment] + _cuda_runtime = None # type: ignore[assignment] + + +if TYPE_CHECKING: + # importing _POOL_HANDLE at runtime toplevel causes an import cycle + from torch.cuda import _POOL_HANDLE, MemPool + from torch.utils._cuda_debug import _CUDAGraphInputLivenessTracker + from torch.utils.hooks import RemovableHandle + + _GraphPool: TypeAlias = _POOL_HANDLE | MemPool +else: + _GraphPool: TypeAlias = typing.Any + +from .._utils import _dummy_type + + +__all__ = [ + "is_current_stream_capturing", + "graph_pool_handle", + "CUDAGraph", + "graph", + "make_graphed_callables", + "export_dot", + "export_graph_data", + "register_graph_capture_start_hook", + "register_graph_capture_end_hook", + "register_graph_instantiate_hook", + "register_graph_replay_start_hook", + "register_graph_replay_end_hook", + "register_graph_destroy_hook", +] + + +_R = TypeVar("_R") +_P = ParamSpec("_P") + + +def _is_mem_pool(pool: object) -> TypeGuard[MemPool]: + mempool_cls = getattr(torch.cuda, "MemPool", None) + return mempool_cls is not None and isinstance(pool, mempool_cls) + + +def _get_pool_id(pool: _GraphPool | None) -> _POOL_HANDLE | None: + if pool is None: + return None + if _is_mem_pool(pool): + return typing.cast("_POOL_HANDLE", pool.id) + return typing.cast("_POOL_HANDLE", pool) + + +if not hasattr(torch._C, "_CudaStreamBase"): + # Define dummy base classes + torch._C.__dict__["_CUDAGraph"] = _dummy_type("_CUDAGraph") + torch._C.__dict__["_graph_pool_handle"] = _dummy_type("_graph_pool_handle") + torch._C.__dict__["_cuda_isCurrentStreamCapturing"] = _dummy_type( + "_cuda_isCurrentStreamCapturing" + ) + +from torch._C import _cuda_isCurrentStreamCapturing, _CUDAGraph, _graph_pool_handle + + +def is_current_stream_capturing() -> bool: + r"""Return True if CUDA graph capture is underway on the current CUDA stream, False otherwise. + + If a CUDA context does not exist on the current device, returns False without initializing the context. + """ + return _cuda_isCurrentStreamCapturing() + + +# Python shim helps Sphinx process docstrings more reliably. +def graph_pool_handle() -> _POOL_HANDLE: + r"""Return an opaque token representing the id of a graph memory pool. + + See :ref:`Graph memory management`. + + .. warning:: + This API is in beta and may change in future releases. + """ + return torch.cuda._POOL_HANDLE(_graph_pool_handle()) + + +def _require_cuda_bindings() -> None: + """Raise a uniform error if the cuda-bindings package is unavailable. + + ``_cuda_runtime`` and ``_cuda_driver`` are imported together, so checking one + suffices. + """ + if _cuda_runtime is None: + raise RuntimeError( + "This CUDAGraph API requires the cuda-bindings package; " + "install it with `pip install cuda-bindings`." + ) + + +class _RetainedCallbacks: + r"""Holds destroy callbacks and retained objects for a single capture cycle + of a :class:`CUDAGraph`. + + A :class:`CUDAGraph` owns one of these and arms a :func:`weakref.finalize` + bound to :meth:`fire`. The holder deliberately does NOT reference the graph, + so the finalizer cannot keep the graph alive; :meth:`fire` can still sync the + replay streams because they are mirrored onto the holder, not read off the + graph. Callbacks and objects live in ``OrderedDict``s keyed by + ``RemovableHandle`` id so registrations can be individually removed. + """ + + def __init__(self) -> None: + self.callbacks: OrderedDict[int, Callable[[], None]] = OrderedDict() + self.objects: OrderedDict[int, object] = OrderedDict() + # Streams the graph was replayed on; drained before firing. A set (not a + # single stream) because concurrent replays on multiple streams leave + # work in flight on all of them and we cannot know which finishes last. + # Stream hashes/compares by value, so same-stream replays dedup to one + # entry -- the common case. Only populated when a caller requested sync. + self.replay_streams: set[torch.cuda.Stream] = set() + self.sync_before_fire: bool = False + self._fired: bool = False + + def fire(self) -> None: + # Fired at most once (guards the reset()-then-destroy double trigger). + if self._fired: + return + self._fired = True + if self.sync_before_fire: + # Drain in-flight replays before releasing anything the graph may + # reference. Guarded so a sync failure does not skip the callbacks. + for stream in self.replay_streams: + try: + stream.synchronize() + except Exception: + pass + for cb in list(self.callbacks.values()): + try: + cb() + except Exception: + pass # match finalizer semantics: swallow, don't abort the rest + self.callbacks.clear() + self.objects.clear() + + +# Global CUDA-graph lifecycle hooks (the module-level counterpart of the per-graph +# _capture_start_hooks / _capture_end_hooks / _post_instantiate_hooks / _replay_*_hooks). A +# consumer (e.g. a profiler observer) registers a hook that fires for every graph, so the graph +# code stays free of consumer knowledge and a consumer sees captures of graphs it did not build +# (an inductor or NCCL capture); registering a hook is the opt-in. Keyed by RemovableHandle id. +# The fan-out (_run_global_hooks) swallows per-hook errors so one consumer cannot break a graph +# lifecycle step for another -- unlike the per-graph hooks, whose errors propagate to the owner. +_global_capture_start_hooks: OrderedDict[int, Callable[[CUDAGraph], None]] = ( + OrderedDict() +) +_global_capture_end_hooks: OrderedDict[int, Callable[[CUDAGraph], None]] = OrderedDict() +_global_instantiate_hooks: OrderedDict[int, Callable[[CUDAGraph], None]] = OrderedDict() +_global_replay_start_hooks: OrderedDict[int, Callable[[CUDAGraph], None]] = ( + OrderedDict() +) +_global_replay_end_hooks: OrderedDict[int, Callable[[CUDAGraph], None]] = OrderedDict() + + +def _register_global_hook( + registry: OrderedDict[int, Callable[[CUDAGraph], None]], + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(registry) + registry[handle.id] = fn + return handle + + +def _run_global_hooks( + registry: OrderedDict[int, Callable[[CUDAGraph], None]], + torch_cuda_graph: CUDAGraph, +) -> None: + for fn in list(registry.values()): + try: + fn(torch_cuda_graph) + except Exception: + pass + + +def register_graph_capture_start_hook( + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + """Register a hook run with each CUDA graph as its capture begins. Returns a + RemovableHandle; call ``.remove()`` to unregister. + + .. warning:: + The hook runs with capture already live on the current stream, so it must not issue + CUDA work: anything it launches is captured into the graph, and under the default + ``"global"`` capture error mode an unsafe call (e.g. an allocation) raises. Querying + capture state is fine. Do preparation that needs CUDA before the capture instead. + """ + return _register_global_hook(_global_capture_start_hooks, fn) + + +def register_graph_capture_end_hook( + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + """Register a hook run with each CUDA graph when its capture ends, while the captured + ``cudaGraph_t`` is still live (see :meth:`CUDAGraph.register_capture_end_hook`). Returns a + RemovableHandle; call ``.remove()`` to unregister.""" + return _register_global_hook(_global_capture_end_hooks, fn) + + +def register_graph_instantiate_hook( + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + """Register a hook run with each CUDA graph right after it is instantiated. Returns a + RemovableHandle; call ``.remove()`` to unregister.""" + return _register_global_hook(_global_instantiate_hooks, fn) + + +def register_graph_replay_start_hook( + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + """Register a hook run with each CUDA graph at the start of every replay, just before it + is launched. Returns a RemovableHandle; call ``.remove()`` to unregister. + + .. note:: + Replay is the hot path and this fires for EVERY graph on EVERY replay -- keep it + cheap. With nothing registered the cost is a single dict emptiness check. + """ + return _register_global_hook(_global_replay_start_hooks, fn) + + +def register_graph_replay_end_hook( + fn: Callable[[CUDAGraph], None], +) -> RemovableHandle: + """Register a hook run with each CUDA graph at the end of every replay, once the replay is + *enqueued* (the launch is asynchronous, so the GPU work has not completed). Fires even if + the launch raised, so a start hook is always balanced by an end. Returns a RemovableHandle; + call ``.remove()`` to unregister. See the hot-path note on + :func:`register_graph_replay_start_hook`.""" + return _register_global_hook(_global_replay_end_hooks, fn) + + +# Graph-destroy hooks, each handed the destroyed graph's exec ids (tools_id >> 32) so a +# consumer can purge its per-graph state. A CUDAGraph arms a single destroy callback (only +# while a hook is registered, see _graph_destroy_hooks_active) that fans out here. +_global_destroy_hooks: OrderedDict[int, Callable[[set[int]], None]] = OrderedDict() + + +def register_graph_destroy_hook(fn: Callable[[set[int]], None]) -> RemovableHandle: + """Register ``fn(exec_ids)`` to run when a CUDA graph is destroyed. Returns a handle whose + ``remove()`` unregisters it.""" + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(_global_destroy_hooks) + _global_destroy_hooks[handle.id] = fn + return handle + + +def _graph_destroy_hooks_active() -> bool: + """True when any graph-destroy hook is registered -- the gate a CUDAGraph checks before + arming its destroy callback.""" + return bool(_global_destroy_hooks) + + +def _run_graph_destroy_hooks(exec_graph_ids: set[int]) -> None: + """Invoke every registered hook with the destroyed exec graph ids, swallowing per-hook + errors so one failure does not abort the rest (matching the destroy-callback fire + semantics). The single entry point a graph's destroy callback calls.""" + for fn in list(_global_destroy_hooks.values()): + try: + fn(exec_graph_ids) + except Exception: + pass + + +# Python shim helps Sphinx process docstrings more reliably. +class CUDAGraph(_CUDAGraph): + r"""Wrapper around a CUDA graph. + + Arguments: + keep_graph (bool, optional): If ``keep_graph=False``, the + cudaGraphExec_t will be instantiated on GPU at the end of + ``capture_end`` and the underlying cudaGraph_t will be + destroyed. Users who want to query or otherwise modify the + underlying cudaGraph_t before instantiation can set + ``keep_graph=True`` and access it via ``raw_cuda_graph`` after + ``capture_end``. Note that the cudaGraphExec_t will not be + instantiated at the end of ``capture_end`` in this + case. Instead, it will be instantiated via an explicit call + to ``instantiate`` or automatically on the first call to + ``replay`` if ``instantiate`` was not already called. Calling + ``instantiate`` manually before ``replay`` is recommended to + prevent increased latency on the first call to ``replay``. It + is allowed to modify the raw cudaGraph_t after first calling + ``instantiate``, but the user must call ``instantiate`` again + manually to make sure the instantiated graph has these + changes. Pytorch has no means of tracking these changes. + + .. warning:: + This API is in beta and may change in future releases. + + """ + + _tracker: _CUDAGraphInputLivenessTracker | None + # Read-only property exposed from the C++ _CUDAGraph base via pybind; + # annotated (not assigned) so the type checker sees it without shadowing it. + _has_graph_exec: bool + # Stays None unless maybe_stamp_capture_graph_id stamps it during capture_end + # (requires annotations enabled and cudaGraphNodeGetToolsId available). + _capture_graph_id: int | None + # Exec graph id the recorded annotations are currently keyed to, or None + # before the first remap. Lets a re-instantiate (which produces a fresh exec + # id) rekey annotations from the previous exec id to the new one. + _remapped_exec_id: int | None + # Exec graph ids a consumer has recorded per-graph state under (one per + # instantiate). Handed to the graph-destroy hooks on destruction so consumers + # can purge that state and their maps do not grow across the run. + _recorded_exec_ids: set[int] + _keep_graph: bool + # User hooks fired by capture_begin / capture_end / instantiate (see register_*_hook). + _capture_start_hooks: dict[int, Callable[[CUDAGraph], None]] + _capture_end_hooks: dict[int, Callable[[CUDAGraph], None]] + _post_instantiate_hooks: dict[int, Callable[[CUDAGraph], None]] + # Transient get_graph_data() cache shared across a single instantiate()'s + # post-instantiate hooks, so multiple consumers (dependency recording, an + # export_graph_data dump) query the topology once instead of once each. + # Only populated while _caching_graph_data is set; dropped right after the + # hooks run so the topology is not retained for the graph's lifetime. + _caching_graph_data: bool + _instantiate_graph_data: dict | None + # Replay lifecycle hooks, fired around each replay (hot path -- only iterated + # when non-empty). See register_replay_start_hook / register_replay_end_hook. + _replay_start_hooks: dict[int, Callable[[CUDAGraph], None]] + _replay_end_hooks: dict[int, Callable[[CUDAGraph], None]] + # Destroy callbacks / retained objects for the current capture cycle, plus + # the weakref.finalize armed on this graph that fires the holder on + # collection. reset() fires the current holder and re-arms a fresh pair. + _retained: _RetainedCallbacks + _retained_finalizer: weakref.finalize | None + + def __new__(cls, keep_graph: bool = False) -> Self: + instance = super().__new__(cls, keep_graph) + instance._tracker = None + instance._capture_graph_id = None + instance._remapped_exec_id = None + instance._recorded_exec_ids = set() + instance._keep_graph = keep_graph + # OrderedDict (not dict): RemovableHandle weak-references the mapping. + instance._capture_start_hooks = OrderedDict() + instance._capture_end_hooks = OrderedDict() + instance._post_instantiate_hooks = OrderedDict() + instance._caching_graph_data = False + instance._instantiate_graph_data = None + instance._replay_start_hooks = OrderedDict() + instance._replay_end_hooks = OrderedDict() + instance._retained_finalizer = None + instance._arm_retained() + return instance + + def _arm_retained(self) -> None: + # Install a fresh destroy-callback holder and finalizer. The finalizer is + # bound to the holder, NOT to `self`: a finalizer arg referencing the + # graph would keep it alive forever and never fire. The holder does not + # point back at the graph, so fire() can still sync the last replay + # stream off it without pinning the graph. + self._retained = _RetainedCallbacks() + self._retained_finalizer = weakref.finalize(self, self._retained.fire) + # When a consumer (e.g. the CUPTI monitor) has registered graph-destroy + # hooks, arm the per-graph state purge for this capture cycle. The callback + # captures the exec-id SET OBJECT (empty now, filled as this graph + # records/instantiates) and the module fan-out function, never self and never + # a hook: a closure reachable to the graph would pin it past collection so it + # never fires. reset() re-arms a fresh holder, so this re-registers per cycle; + # a graph that records nothing just fires on an empty set (a no-op). + if _graph_destroy_hooks_active(): + exec_ids = self._recorded_exec_ids + self.register_destroy_callback(lambda: _run_graph_destroy_hooks(exec_ids)) + + def register_capture_start_hook( + self, hook: Callable[[CUDAGraph], None] + ) -> RemovableHandle: + r"""Register ``hook(graph)`` to run when capture begins on this graph, right + after capture is under way on the current stream. Hooks fire in registration + order. Returns a handle whose ``remove()`` deregisters the hook. + + .. warning:: + The hook runs inside the capture: any CUDA work it issues is captured into + the graph, and under the default ``"global"`` capture error mode an unsafe + call raises. See :func:`torch.cuda.graphs.register_graph_capture_start_hook`. + """ + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(self._capture_start_hooks) + self._capture_start_hooks[handle.id] = hook + return handle + + def register_capture_end_hook( + self, hook: Callable[[CUDAGraph], None] + ) -> RemovableHandle: + r"""Register ``hook(graph)`` to run when capture ends, after capture + completes but before the graph is finalized. The captured ``cudaGraph_t`` + is live (via :meth:`raw_cuda_graph`) for both ``keep_graph`` modes. Hooks + fire in registration order. Returns a handle whose ``remove()`` + deregisters the hook. + """ + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(self._capture_end_hooks) + self._capture_end_hooks[handle.id] = hook + return handle + + def register_post_instantiate_hook( + self, hook: Callable[[CUDAGraph], None] + ) -> RemovableHandle: + r"""Register ``hook(graph)`` to run after each instantiation (including + re-instantiation, which produces a fresh exec graph). The instantiated + graph is available via :meth:`raw_cuda_graph_exec`. Hooks fire in + registration order. Returns a handle whose ``remove()`` deregisters the + hook. + """ + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(self._post_instantiate_hooks) + self._post_instantiate_hooks[handle.id] = hook + return handle + + def register_replay_start_hook( + self, hook: Callable[[CUDAGraph], None] + ) -> RemovableHandle: + r"""Register ``hook(graph)`` to run at the start of every :meth:`replay`, + just before the graph is launched (after any on-demand instantiation, so + :meth:`raw_cuda_graph_exec` is valid). Hooks fire in registration order. + Returns a handle whose ``remove()`` deregisters the hook. + + .. note:: + Replay is the hot path and a registered hook runs on every replay -- + keep it cheap. With no hook registered the cost is a single dict + emptiness check. + """ + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(self._replay_start_hooks) + self._replay_start_hooks[handle.id] = hook + return handle + + def register_replay_end_hook( + self, hook: Callable[[CUDAGraph], None] + ) -> RemovableHandle: + r"""Register ``hook(graph)`` to run at the end of every :meth:`replay`, + just after the graph is launched. The launch is asynchronous, so the hook + runs once the replay is *enqueued*, not once the GPU work completes. Hooks + fire in registration order. Returns a handle whose ``remove()`` + deregisters the hook. See the hot-path note on + :meth:`register_replay_start_hook`. + + End hooks fire even if the launch raises -- so a start hook is always + balanced by an end -- and the launch error then propagates. (Start hooks + that raise abort the replay before launch, and no end hook fires.) + """ + from torch.utils.hooks import RemovableHandle + + handle = RemovableHandle(self._replay_end_hooks) + self._replay_end_hooks[handle.id] = hook + return handle + + def register_destroy_callback( + self, + cb: Callable[[], None], + *, + synchronize_before_release: bool = False, + ) -> RemovableHandle: + r"""Register ``cb()`` to run when this graph is destroyed (finalized) or + explicitly :meth:`reset`, just before its CUDA resources are freed. + Callbacks fire once per capture cycle, in registration order; exceptions + are swallowed so one failure does not abort the rest. ``cb`` must NOT + reference this graph: the finalizer that fires it is held by a global + registry, so a callback reachable to the graph keeps the graph alive + until interpreter exit (it is never collected, hence never fired). + Returns a handle whose ``remove()`` deregisters the callback. + + Teardown does not synchronize CUDA, and ``cudaGraphExecDestroy`` frees an + in-flight graph only asynchronously, so a callback that frees device + memory the graph reads/writes is a use-after-free if a replay is still + in flight. Pass ``synchronize_before_release=True`` to synchronize every + stream this graph was replayed on before firing. Otherwise callbacks + must not free anything the graph references. + """ + from torch.utils.hooks import RemovableHandle + + if synchronize_before_release: + self._retained.sync_before_fire = True + handle = RemovableHandle(self._retained.callbacks) + self._retained.callbacks[handle.id] = cb + return handle + + def retain_object( + self, + obj: object, + *, + synchronize_before_release: bool = False, + ) -> RemovableHandle: + r"""Keep ``obj`` alive for this graph's current capture cycle and release + it when the graph is destroyed (finalized) or explicitly :meth:`reset`. + No callback runs; normal refcounting drops ``obj`` when the retained + reference is released. Returns a handle whose ``remove()`` drops the + retained reference early. As with :meth:`register_destroy_callback`, + ``obj`` must NOT reference this graph, or the graph is kept alive until + interpreter exit and ``obj`` is never released. + + ``synchronize_before_release`` has the same meaning and caveats as in + :meth:`register_destroy_callback`: set it if releasing ``obj`` frees device + memory the graph reads/writes (e.g. ``obj`` is the last reference to a + tensor the graph uses) and replays may still be in flight. + """ + from torch.utils.hooks import RemovableHandle + + if synchronize_before_release: + self._retained.sync_before_fire = True + handle = RemovableHandle(self._retained.objects) + self._retained.objects[handle.id] = obj + return handle + + def _maybe_remap_annotations(self) -> None: + # Remap recorded kernel annotations to the current exec graph id. No-op + # unless a capture id was stamped (annotations enabled). Called from + # instantiate() -- which replay() routes through for keep_graph=True -- + # so every fresh exec graph (each instantiate() produces a new exec id) + # rekeys the annotations; remap_to_exec_graph self-skips when the exec id + # is unchanged. + if self._capture_graph_id is None: + return + from torch.cuda._graph_annotations import remap_to_exec_graph + + remap_to_exec_graph(self) + # Record the exec id we remapped to so the destroy callback (armed in + # capture_end_post) hands it to the graph-destroy hooks for per-graph state + # cleanup. A re-instantiate mints a fresh exec id, so the set accumulates each. + if self._remapped_exec_id is not None: + self._recorded_exec_ids.add(self._remapped_exec_id) + + def _release_python_resources(self) -> None: + # Single source of truth for GC-critical Python resources released by + # both reset() and __del__. Destroy callbacks are NOT fired here: on the + # destruction path the armed weakref.finalize fires them (after __del__, + # before the C++ resources are freed), so user code stays out of + # __del__; reset() fires + re-arms the holder itself (see reset()). + tracker, self._tracker = self._tracker, None + if tracker is not None: + tracker.stop() + + def __del__(self) -> None: + try: + self._release_python_resources() + except Exception: + pass # don't raise under GC + + def capture_begin( + self, + pool: _GraphPool | None = None, + capture_error_mode: str = "global", + check_input_liveness: bool = False, + ) -> None: + r"""Begin capturing CUDA work on the current stream. + + Typically, you shouldn't call ``capture_begin`` yourself. + Use :class:`~torch.cuda.graph` or :func:`~torch.cuda.make_graphed_callables`, + which call ``capture_begin`` internally. + + Arguments: + pool (optional): Token (returned by :func:`~torch.cuda.graph_pool_handle` or + :meth:`other_Graph_instance.pool()`) or + :class:`~torch.cuda.MemPool` that hints this graph may share memory + with the indicated pool. See :ref:`Graph memory management`. + capture_error_mode (str, optional): specifies the cudaStreamCaptureMode for the graph capture stream. + Can be "global", "thread_local" or "relaxed". During cuda graph capture, some actions, such as cudaMalloc, + may be unsafe. "global" will error on actions in other threads, "thread_local" will only error for + actions in the current thread, and "relaxed" will not error on these actions. Do NOT change this setting + unless you're familiar with `cudaStreamCaptureMode `_ + check_input_liveness (bool, optional): + If ``True``, tracks external tensor inputs during graph capture and + raises an error if any are deallocated before replay. This helps debug "use after free" errors + where input tensors are garbage collected between capture and replay. Default: ``False``. + + .. note:: + Custom CUDA kernels added outside PyTorch (e.g., via cuLaunchKernel or DLPack) are not + tracked by this mechanism. + """ + if self._tracker is not None: + self._tracker.stop() + self._tracker = None + super().capture_begin( + pool=_get_pool_id(pool), capture_error_mode=capture_error_mode + ) + if check_input_liveness: + from torch.utils._cuda_debug import _CUDAGraphInputLivenessTracker + + self._tracker = _CUDAGraphInputLivenessTracker() + self._tracker.start() + # Capture is live from here, so a hook must not issue CUDA work (see + # register_capture_start_hook). Global hooks run before the per-graph ones, + # matching instantiate(). + if _global_capture_start_hooks: + _run_global_hooks(_global_capture_start_hooks, self) + for hook in list(self._capture_start_hooks.values()): + hook(self) + + def capture_end_pre(self) -> None: + r"""End capture but do not finalize: leaves the captured ``cudaGraph_t`` + live (for both ``keep_graph`` modes) so it can be inspected before + :meth:`capture_end_post` instantiates and/or destroys it.""" + super().capture_end_pre() + + def capture_end_post(self) -> None: + r"""Finalize a capture started by :meth:`capture_end_pre`: destroy the + template when ``keep_graph=False`` (the graph must already be + instantiated; :meth:`capture_end` and the context manager do so).""" + super().capture_end_post() + if self._tracker is not None: + self._tracker.stop() + + def capture_end(self) -> None: + r"""End CUDA graph capture on the current stream. + + After ``capture_end``, ``replay`` may be called on this instance. + + Typically, you shouldn't call ``capture_end`` yourself. + Use :class:`~torch.cuda.graph` or :func:`~torch.cuda.make_graphed_callables`, + which call ``capture_end`` internally. + """ + self.capture_end_pre() + # Run the in-window work (stamp, then user capture-end hooks) while the + # template is live (both keep_graph modes). Errors here are unexpected + # (stamp) or user bugs (hooks) and propagate -- we deliberately don't + # wrap them in a finally that calls capture_end_post(), since a failing + # finalize would mask the real error. + from torch.cuda._graph_annotations import maybe_stamp_capture_graph_id + + maybe_stamp_capture_graph_id(self) + if _global_capture_end_hooks: + _run_global_hooks(_global_capture_end_hooks, self) + for hook in list(self._capture_end_hooks.values()): + hook(self) + if not self._keep_graph: + # Route keep_graph=False instantiation through Python so the remap + # and post-instantiate hooks fire from instantiate(). + self.instantiate() + self.capture_end_post() + + def instantiate(self) -> None: + r"""Instantiate the CUDA graph. Will be called by + ``capture_end`` if ``keep_graph=False``, or by ``replay`` if + ``keep_graph=True`` and ``instantiate`` has not already been + explicitly called. Does not destroy the cudaGraph_t returned + by ``raw_cuda_graph``. + """ + super().instantiate() + self._maybe_remap_annotations() + # Fire registered instantiate hooks (e.g. a profiler observer inspecting the freshly + # instantiated graph). Registering a hook is the opt-in; no consumer -> a no-op. + # get_graph_data() is cached for the duration of the hooks so several consumers share + # one query; the cache is dropped afterwards so nothing is retained for the graph. + self._caching_graph_data = True + try: + if _global_instantiate_hooks: + _run_global_hooks(_global_instantiate_hooks, self) + for hook in list(self._post_instantiate_hooks.values()): + hook(self) + finally: + self._caching_graph_data = False + self._instantiate_graph_data = None + + def replay(self) -> None: + r"""Replay the CUDA work captured by this graph.""" + if self._tracker is not None: + self._tracker.check_alive(self.pools()) + if self._retained.sync_before_fire: + # Record the replay stream on the holder so fire() can drain it + # before releasing graph-referenced resources, without the holder + # referencing the graph. Only recorded when a caller opted into sync. + self._retained.replay_streams.add(torch.cuda.current_stream()) + # With keep_graph=True the exec graph is instantiated on demand here on + # the first replay; the C++ replay() requires it to already exist. The + # annotation remap rides on instantiate(), so it is handled by that call. + if not self._has_graph_exec: + self.instantiate() + if _global_replay_start_hooks: + _run_global_hooks(_global_replay_start_hooks, self) + if self._replay_start_hooks: + for hook in list(self._replay_start_hooks.values()): + hook(self) + try: + super().replay() + finally: + if _global_replay_end_hooks: + _run_global_hooks(_global_replay_end_hooks, self) + if self._replay_end_hooks: + for hook in list(self._replay_end_hooks.values()): + hook(self) + + def reset(self) -> None: + r"""Delete the graph currently held by this instance.""" + self._release_python_resources() + # also-fire-on-reset: reset() destroys this capture's CUDA resources and + # the graph may be re-captured, so fire the current destroy callbacks and + # re-arm a fresh holder + finalizer for the next capture cycle (the old + # finalizer is bound to the now-spent holder). detach() cancels it and + # drops its reference to the spent holder promptly. + self._retained.fire() + if self._retained_finalizer is not None: + self._retained_finalizer.detach() + # Start a fresh id set BEFORE re-arming: _arm_retained captures it into the + # next cycle's destroy callback. The just-fired holder dropped the old one. + self._recorded_exec_ids = set() + self._arm_retained() + # Reset-only state: scrubbed here because the object is reused after + # reset(); on death these ints die with the object. + self._capture_graph_id = None + self._remapped_exec_id = None + super().reset() + + def pool(self) -> _POOL_HANDLE: + r"""Return an opaque token representing the id of this graph's memory pool. + + This id can optionally be passed to another graph's ``capture_begin``, + which hints the other graph may share the same memory pool. + """ + return torch.cuda._POOL_HANDLE(super().pool()) + + def pools(self) -> list[_POOL_HANDLE]: + r"""Return opaque tokens for all memory pools retained by this graph.""" + return [torch.cuda._POOL_HANDLE(pool) for pool in super().pools()] + + def _retain_pool(self, pool: _GraphPool) -> None: + pool_id = _get_pool_id(pool) + if pool_id is None: + raise RuntimeError("CUDAGraph._retain_pool expected a memory pool") + return super()._retain_pool(pool_id) + + def enable_debug_mode(self) -> None: + r"""Retain the captured graph (equivalent to ``keep_graph=True``) so it + can be inspected, e.g. via :meth:`debug_dump`. Kept for backward + compatibility.""" + super().enable_debug_mode() + # Keep the Python-side flag in sync with the C++ keep_graph_ it now sets. + self._keep_graph = True + + def debug_dump(self, debug_path: str, *, verbose: bool = True) -> None: + r"""Dump the captured graph to ``debug_path`` in Graphviz DOT format. + + The graph's template must be live: ``keep_graph=True`` (or + :meth:`enable_debug_mode`), or called from a capture-end hook. Requires + the ``cuda.bindings`` package. + + Arguments: + debug_path (required): Path to dump the graph to. + verbose: If ``True`` (default), use the most verbose DOT output. + """ + _dump_graph_dot(self, debug_path, verbose=verbose) + + def raw_cuda_graph(self) -> int: + r"""Returns the underlying cudaGraph_t. The template must be live: this + requires ``keep_graph=True`` (it persists after ``capture_end``), or + access from within a capture-end hook (before the template is destroyed + for ``keep_graph=False``). + + See the following for APIs for how to manipulate this object: `Graph Management `_ and `cuda-python Graph Management bindings `_ + """ + return super().raw_cuda_graph() + + def raw_cuda_graph_exec(self) -> int: + r"""Returns the underlying cudaGraphExec_t. ``instantiate`` must have been called if ``keep_graph`` is True, or ``capture_end`` must have been called if ``keep_graph`` is False. If you call ``instantiate()`` after ``raw_cuda_graph_exec()``, the previously returned cudaGraphExec_t will be destroyed. It is your responsibility not to use this object after destruction. + + See the following for APIs for how to manipulate this object: `Graph Execution `_ and `cuda-python Graph Execution bindings `_ + """ + return super().raw_cuda_graph_exec() + + def get_graph_data(self) -> dict: + r"""Return a dictionary describing the graph's topology and node metadata. + + ``keep_graph`` must be True. The graph must have been instantiated + (via :meth:`instantiate`) before calling this method. + Requires the ``cuda.bindings`` package. + + Returns a dictionary with structure:: + + { + "exec_graph_id": int, + "nodes": [ + { + "index": int, + "node_type": str, + "tools_id": int, + "graph_id": int, + "node_id": int, + "kernel_name": str or None, + "event_ptr": int, + "host_fn_addr": int, + "host_fn_name": str or None, + "dependencies": [int, ...], + "dependents": [int, ...], + }, + ..., + ], + } + + ``event_ptr`` is the ``cudaEvent_t`` handle (as an int) an event-record + or event-wait node records / waits on -- these nodes produce no timed + CUPTI record, so matching a wait to the record that signals it is the + only way to reason about the cross-stream sync it encodes. It is ``0`` + for other node types. + + ``host_fn_addr`` / ``host_fn_name`` are populated for host nodes (a CPU + callback run as a graph node): the callback address and a best-effort + demangled symbol name for it (``None`` when it resolves to no exported + symbol). They are ``0`` / ``None`` for other node types. + + Each node's ``graph_id`` is remapped to the exec graph id so that + ``tools_id`` values match those reported by CUPTI-based profilers. + ``dependencies`` and ``dependents`` are lists of node indices within + the ``nodes`` list. + + This structure is useful for inspecting a profiler trace and + establishing whether a particular dependency observed in the profile + is a true dependency (encoded in the graph) or a fake dependency + caused by mapping of independent streams to the same hardware + channel. + + Child-graph and conditional nodes are reported as nodes in their own + right, but this walk does not descend into their bodies (those are + separate ``cudaGraph_t`` objects), so the work inside them is absent + from ``nodes`` and a warning is issued. The ids that *are* reported + stay valid: the exec graph preserves top-level node ids. + """ + # Serve the shared cache when instantiate() has it live (see _caching_graph_data). + if self._instantiate_graph_data is not None: + return self._instantiate_graph_data + from torch.cuda._graph_annotations import _is_tools_id_unavailable + + _require_cuda_bindings() + # Narrow for the type checker (cuda bindings are present past the check). + if _cuda_runtime is None or _cuda_driver is None: + raise AssertionError( + "expected _cuda_runtime and _cuda_driver to be not None" + ) + + if _is_tools_id_unavailable(): + raise RuntimeError( + "get_graph_data requires cudaGraphNodeGetToolsId which needs " + "cuda.bindings >= 13.1 and CUDA driver >= 13.1 " + "(or cuda-compat >= 13.1 in LD_LIBRARY_PATH)" + ) + + node_type_names = { + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeKernel: "kernel", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeMemcpy: "memcpy", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeMemset: "memset", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeHost: "host", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeGraph: "child_graph", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeEmpty: "empty", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeWaitEvent: "wait_event", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeEventRecord: "event_record", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeMemAlloc: "mem_alloc", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeMemFree: "mem_free", + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeConditional: "conditional", + } + # Node types whose work lives in a separate cudaGraph_t (see the warning below). + nested_graph_types = { + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeGraph, + _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeConditional, + } + nested_node_types: set[str] = set() + + raw = self.raw_cuda_graph() + + _, num = _check_cuda_bindings(_cuda_runtime.cudaGraphGetNodes(raw, numNodes=0)) + nodes, num = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetNodes(raw, numNodes=num) + ) + + handle_to_idx: dict[int, int] = {} + node_infos: list[dict] = [] + + for i in range(num): + node = nodes[i] + handle_to_idx[int(node)] = i + + ntype = _check_cuda_bindings(_cuda_runtime.cudaGraphNodeGetType(node)) + if ntype in nested_graph_types: + nested_node_types.add(node_type_names.get(ntype, ntype)) + tools_id = _check_cuda_bindings(_cuda_runtime.cudaGraphNodeGetToolsId(node)) + graph_id = tools_id >> 32 + node_id = tools_id & 0xFFFFFFFF + + kernel_name = None + if ntype == _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeKernel: + cu_node = _cuda_driver.CUgraphNode(init_value=int(node)) + err, params = _cuda_driver.cuGraphKernelNodeGetParams(cu_node) + if err == _cuda_driver.CUresult.CUDA_SUCCESS and int(params.func): + cu_func = _cuda_driver.CUfunction(init_value=int(params.func)) + err, name = _cuda_driver.cuFuncGetName(cu_func) + if err == _cuda_driver.CUresult.CUDA_SUCCESS: + kernel_name = name.decode() if isinstance(name, bytes) else name + + # Event record/wait nodes carry a cudaEvent_t but emit no timed CUPTI record; + # capture the handle so a wait node can be matched to the record that signals it. + # Not best-effort like the kernel-name lookup above (a name can legitimately be + # unavailable): the node type is already established here, so these calls are + # expected to succeed. Swallowing a failure would leave event_ptr 0 and make the + # record/wait match quietly wrong rather than loud. + event_ptr = 0 + if ntype == _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeEventRecord: + event_ptr = int( + _check_cuda_bindings( + _cuda_runtime.cudaGraphEventRecordNodeGetEvent(node) + ) + ) + elif ntype == _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeWaitEvent: + event_ptr = int( + _check_cuda_bindings( + _cuda_runtime.cudaGraphEventWaitNodeGetEvent(node) + ) + ) + + # Host nodes carry no name in the CUPTI record; recover the callback address + # from the node params and resolve a best-effort symbol name (see + # _resolve_host_fn_name). Both are None/0 for other node types. + host_fn_addr = 0 + host_fn_name = None + if ntype == _cuda_runtime.cudaGraphNodeType.cudaGraphNodeTypeHost: + err, params = _cuda_runtime.cudaGraphHostNodeGetParams(node) + if err == _cuda_runtime.cudaError_t.cudaSuccess: + host_fn_addr = int(params.fn) + host_fn_name = _resolve_host_fn_name(host_fn_addr) + + node_infos.append( + { + "index": i, + "node_type": node_type_names.get(ntype, str(ntype)), + "tools_id": tools_id, + "graph_id": graph_id, + "node_id": node_id, + "kernel_name": kernel_name, + "event_ptr": event_ptr, + "host_fn_addr": host_fn_addr, + "host_fn_name": host_fn_name, + "dependencies": [], + "dependents": [], + } + ) + + _, _, _, num_edges = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetEdges(raw, numEdges=0) + ) + if num_edges > 0: + from_nodes, to_nodes, _edge_data, num_edges = _check_cuda_bindings( + _cuda_runtime.cudaGraphGetEdges(raw, numEdges=num_edges) + ) + for i in range(num_edges): + src = handle_to_idx.get(int(from_nodes[i])) + dst = handle_to_idx.get(int(to_nodes[i])) + if src is not None and dst is not None: + node_infos[src]["dependents"].append(dst) + node_infos[dst]["dependencies"].append(src) + + exec_handle = _cuda_runtime.cudaGraphExec_t( + init_value=self.raw_cuda_graph_exec() + ) + exec_graph_id = _check_cuda_bindings( + _cuda_runtime.cudaGraphExecGetId(exec_handle) + ) + for info in node_infos: + info["tools_id"] = (exec_graph_id << 32) | info["node_id"] + info["graph_id"] = exec_graph_id + + if nested_node_types: + # What is returned stays correct: the exec graph preserves top-level node + # ids and numbers the nested ones after them. Those nested nodes do execute + # and do show up in a profiler trace (under this exec graph id, with the + # appended node ids), they are just missing from the topology below. + warnings.warn( + f"get_graph_data: this graph contains {sorted(nested_node_types)} " + "nodes; the work inside them is in a separate cudaGraph_t that this " + "walk does not descend into, so it is absent from the returned nodes", + stacklevel=2, + ) + + data = { + "exec_graph_id": exec_graph_id, + "nodes": node_infos, + } + if self._caching_graph_data: + self._instantiate_graph_data = data + return data + + +class _DlInfo(ctypes.Structure): + _fields_ = [ + ("dli_fname", ctypes.c_char_p), + ("dli_fbase", ctypes.c_void_p), + ("dli_sname", ctypes.c_char_p), + ("dli_saddr", ctypes.c_void_p), + ] + + +def _resolve_host_fn_name(addr: int) -> str | None: + """Best-effort symbol name for a host-node callback address via dladdr + demangle. + Returns None when the address resolves to no exported symbol (e.g. an anonymous + trampoline). Host nodes carry no name in the CUPTI record, so this is the only source.""" + if not addr: + return None + info = _DlInfo() + if not ctypes.CDLL(None).dladdr(ctypes.c_void_p(addr), ctypes.byref(info)): + return None + if not info.dli_sname: + return None + return torch._C._demangle(info.dli_sname.decode()) + + +def _dump_graph_dot(cuda_graph: CUDAGraph, path: str, *, verbose: bool = True) -> None: + """Write ``cuda_graph``'s template to ``path`` in Graphviz DOT format.""" + _require_cuda_bindings() + flags = ( + _cuda_runtime.cudaGraphDebugDotFlags.cudaGraphDebugDotFlagsVerbose # pyrefly: ignore[missing-attribute] + if verbose + else 0 + ) + _check_cuda_bindings( + _cuda_runtime.cudaGraphDebugDotPrint( # pyrefly: ignore[missing-attribute] + cuda_graph.raw_cuda_graph(), path.encode(), flags + ) + ) + + +def export_dot(path: str, *, verbose: bool = True) -> Callable[[CUDAGraph], None]: + r"""Return a capture-end hook that dumps the captured graph to ``path`` in + Graphviz DOT format. Register it with + :meth:`CUDAGraph.register_capture_end_hook`; works for both ``keep_graph`` + modes since it runs while the template is still live.""" + + def _hook(cuda_graph: CUDAGraph) -> None: + _dump_graph_dot(cuda_graph, path, verbose=verbose) + + return _hook + + +def export_graph_data(path: str) -> Callable[[CUDAGraph], None]: + r"""Return a post-instantiate hook that pickles :meth:`CUDAGraph.get_graph_data` + to ``path``. Register it with :meth:`CUDAGraph.register_post_instantiate_hook`: + ``get_graph_data`` needs the graph instantiated (it remaps node ids to the + exec graph id), and at post-instantiate time the template is still live, so + this works for both ``keep_graph`` modes.""" + + def _hook(cuda_graph: CUDAGraph) -> None: + import pickle + + with open(path, "wb") as f: + pickle.dump(cuda_graph.get_graph_data(), f) + + return _hook + + +class graph: + r"""Context-manager that captures CUDA work into a :class:`torch.cuda.CUDAGraph` object for later replay. + + See :ref:`CUDA Graphs ` for a general introduction, + detailed use, and constraints. + + Arguments: + cuda_graph (torch.cuda.CUDAGraph): Graph object used for capture. + pool (optional): Opaque token (returned by a call to :func:`~torch.cuda.graph_pool_handle()` or + :meth:`other_Graph_instance.pool()`) or + :class:`~torch.cuda.MemPool` hinting this graph's capture may share memory + from the specified pool. See :ref:`Graph memory management`. + stream (torch.cuda.Stream, optional): If supplied, will be set as the current stream in the context. + If not supplied, ``graph`` sets its own internal side stream as the current stream in the context. + capture_error_mode (str, optional): specifies the cudaStreamCaptureMode for the graph capture stream. + Can be "global", "thread_local" or "relaxed". During cuda graph capture, some actions, such as cudaMalloc, + may be unsafe. "global" will error on actions in other threads, "thread_local" will only error for + actions in the current thread, and "relaxed" will not error on actions. Do NOT change this setting + unless you're familiar with `cudaStreamCaptureMode `_ + enable_annotations (bool, optional): If ``True``, enables kernel annotation + recording on entry and automatically calls + :func:`~torch.cuda._graph_annotations.resolve_pending_annotations` before + the capture ends. Annotations are **not** cleared on exit so that multiple + graphs in the same workload can accumulate annotations. + Requires ``cuda.bindings`` package and cuda-compat >= 13.1 or CUDA driver >= 13.1. + check_input_liveness (bool, optional): If ``True``, tracks external tensor inputs during graph capture and + raises an error if any are deallocated before replay. This helps debug "use after free" errors + where input tensors are garbage collected between capture and replay. Default: ``False``. + + .. note:: + Custom CUDA kernels added outside PyTorch (e.g., via cuLaunchKernel or DLPack) are not + tracked by this mechanism. + + .. note:: + For effective memory sharing, if you pass a ``pool`` used by a previous capture and the previous capture + used an explicit ``stream`` argument, you should pass the same ``stream`` argument to this capture. + + .. warning:: + This API is in beta and may change in future releases. + + .. _cudaStreamCaptureMode: + https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html#group__CUDART__STREAM_1g9d0535d93a214cbf126835257b16ba85 + """ + + default_capture_stream: torch.cuda.Stream | None = None + + def __init__( + self, + cuda_graph: CUDAGraph, + pool: _GraphPool | None = None, + stream: torch.cuda.Stream | None = None, + capture_error_mode: str = "global", + enable_annotations: bool = False, + check_input_liveness: bool = False, + ): + # Lazy-init of default_capture_stream helps avoid circular-import errors. + # Not thread safe, but graphs already have the general (explicitly documented) + # restriction that only one capture may be underway at a time in the process. + if stream is None and self.__class__.default_capture_stream is None: + self.__class__.default_capture_stream = torch.cuda.Stream() + + pool_id = _get_pool_id(pool) + self.pool: tuple[()] | tuple[_POOL_HANDLE] = ( + () if pool_id is None else (pool_id,) + ) + self.capture_stream = ( + stream if stream is not None else self.__class__.default_capture_stream + ) + if self.capture_stream is None: + raise AssertionError("capture_stream must not be None") + self.stream_ctx = torch.cuda.stream(self.capture_stream) + self.cuda_graph = cuda_graph + self.capture_error_mode = capture_error_mode + self._enable_annotations = enable_annotations + self.check_input_liveness = check_input_liveness + + def __enter__(self) -> None: + # Free as much memory as we can for the graph + torch.cuda.synchronize() + + if torch.compiler.config.force_cudagraph_gc: + # Originally we unconditionally garbage collected here. On one hand + # that's nice because we have a chance to collect more memory, but + # on the other hand it is REALLY expensive, especially for doing + # multiple cudagraph captures in a row. In theory it will only help + # when a dead python cycle is holding onto CUDA memory. + gc.collect() + + torch.cuda.empty_cache() + # pyrefly: ignore [missing-attribute] + torch._C._host_emptyCache() + + # Scope annotation recording to this capture: stamp/mark_kernels gate on + # this flag, and __exit__ always clears it. + from torch.cuda._graph_annotations import _set_annotations_enabled + + _set_annotations_enabled(self._enable_annotations) + + # Stackoverflow seems comfortable with this pattern + # https://stackoverflow.com/questions/26635684/calling-enter-and-exit-manually#39172487 + self.stream_ctx.__enter__() + + self.cuda_graph.capture_begin( + # type: ignore[misc] + *self.pool, + # pyrefly: ignore [bad-keyword-argument] + capture_error_mode=self.capture_error_mode, + # pyrefly: ignore [bad-keyword-argument] + check_input_liveness=self.check_input_liveness, + ) + # The capture stream is now capturing into the top-level graph; remember it so + # mark_kernels can tell a conditional-node body apart from this graph. + from torch.cuda._graph_annotations import maybe_stamp_capture_root + + maybe_stamp_capture_root(torch.cuda.current_stream()) + + def __exit__(self, *args: object) -> None: + from torch.cuda._graph_annotations import ( + _set_annotations_enabled, + resolve_pending_annotations, + ) + + try: + if self._enable_annotations: + resolve_pending_annotations() + + # capture_end stamps the capture id and, for keep_graph=False, + # instantiates (which remaps annotations to the exec id). For + # keep_graph=True the remap is owned by the later instantiate()/replay(). + self.cuda_graph.capture_end() + self.stream_ctx.__exit__(*args) + finally: + # Annotation recording is capture-scoped; clear it unconditionally. + _set_annotations_enabled(False) + # returning None should propagate exceptions from either capture_end or stream_ctx.__exit__() + + +_ModuleOrCallable: TypeAlias = Union["torch.nn.Module", Callable[..., object]] + + +@overload +def make_graphed_callables( + callables: _ModuleOrCallable, + sample_args: tuple[Tensor, ...], + num_warmup_iters: int = 3, + allow_unused_input: bool = False, + pool: _GraphPool | None = None, + capture_error_mode: str = "global", + enable_annotations: bool = False, +) -> _ModuleOrCallable: ... + + +@overload +def make_graphed_callables( + callables: tuple[_ModuleOrCallable, ...], + sample_args: tuple[tuple[Tensor, ...], ...], + num_warmup_iters: int = 3, + allow_unused_input: bool = False, + pool: _GraphPool | None = None, + capture_error_mode: str = "global", + enable_annotations: bool = False, +) -> tuple[_ModuleOrCallable, ...]: ... + + +def make_graphed_callables( + callables: _ModuleOrCallable | tuple[_ModuleOrCallable, ...], + sample_args: tuple[Tensor, ...] | tuple[tuple[Tensor, ...], ...], + num_warmup_iters: int = 3, + allow_unused_input: bool = False, + pool: _GraphPool | None = None, + capture_error_mode: str = "global", + enable_annotations: bool = False, +) -> _ModuleOrCallable | tuple[_ModuleOrCallable, ...]: + r"""Accept callables (functions or :class:`nn.Module`\ s) and returns graphed versions. + + Each graphed callable's forward pass runs its source callable's + forward CUDA work as a CUDA graph inside a single autograd node. + + The graphed callable's forward pass also appends + a backward node to the autograd graph. During backward, this node runs the + callable's backward work as a CUDA graph. + + Therefore, each graphed callable should be a drop-in replacement for its source callable + in an autograd-enabled training loop. + + See :ref:`Partial-network capture` for detailed use and constraints. + + If you pass a tuple of several callables, their captures will use the same memory pool. + See :ref:`Graph memory management` for when this is appropriate. + + Arguments: + callables (torch.nn.Module or Python function, or tuple of these): Callable or callables to graph. + See :ref:`Graph memory management` for when passing a tuple of callables + is appropriate. If you pass a tuple of callables, their order in the tuple must be the same order + they'll run in the live workload. + sample_args (tuple of Tensors, or tuple of tuples of Tensors): Samples args for each callable. + If a single callable was passed, ``sample_args`` must be a single tuple of argument Tensors. + If a tuple of callables was passed, ``sample_args`` must be tuple of tuples of argument Tensors. + num_warmup_iters (int): The number of warmup iterations. Currently, ``DataDistributedParallel`` needs + 11 iterations for warm up. Default: ``3``. + allow_unused_input (bool): If False, specifying inputs that were not used when computing outputs + (and therefore their grad is always zero) is an error. Defaults to False. + pool (optional): Token (returned by :func:`~torch.cuda.graph_pool_handle` or + :meth:`other_Graph_instance.pool()`) or + :class:`~torch.cuda.MemPool` that hints this graph may share memory + with the indicated pool. See :ref:`Graph memory management`. + enable_annotations (bool, optional): If ``True``, the forward and backward + captures record kernel annotations from + :func:`torch.cuda.graph_annotations.mark_kernels` scopes inside the + callables (backward kernels are tagged via the scopes' autograd node + hooks). See :mod:`torch.cuda.graph_annotations`. Default: ``False``. + + .. note:: + The ``requires_grad`` state of each Tensor in ``sample_args`` must match the state + that's expected for the corresponding real input in the training loop. + + .. warning:: + This API is in beta and may change in future releases. + + .. warning:: + ``sample_args`` for each callable must contain only Tensors. Other types are not allowed. + + .. warning:: + Returned callables do not support higher order differentiation (e.g., double backward). + + .. warning:: + In any :class:`~torch.nn.Module` passed to :func:`~make_graphed_callables`, only parameters + may be trainable. Buffers must have ``requires_grad=False``. + + .. warning:: + After you pass a :class:`torch.nn.Module` through :func:`~make_graphed_callables`, + you may not add or remove any of that Module's parameters or buffers. + + .. warning:: + :class:`torch.nn.Module`\s passed to :func:`~torch.cuda.make_graphed_callables` must not have module hooks + registered on them at the time they are passed. However, registering hooks on modules *after* passing them + through :func:`~torch.cuda.make_graphed_callables` is allowed. + + .. warning:: + When running a graphed callable, you must pass its arguments in the same order and format + they appeared in that callable's ``sample_args``. + + .. warning:: + The automatic mixed precision is supported in :func:`~torch.cuda.make_graphed_callables` only with disabled + caching. The context manager `torch.cuda.amp.autocast()` must have `cache_enabled=False`. + """ + if torch.is_autocast_enabled() and torch.is_autocast_cache_enabled(): + raise RuntimeError( + "make_graphed_callables does not support the autocast caching. Please set `cache_enabled=False`." + ) + + just_one_callable = False + + _sample_args: tuple[tuple[Tensor, ...], ...] + if not isinstance(callables, tuple): + just_one_callable = True + callables = (callables,) + _sample_args = (typing.cast(tuple[Tensor, ...], sample_args),) + else: + _sample_args = typing.cast(tuple[tuple[Tensor, ...], ...], sample_args) + + flatten_sample_args = [] + + for c, args in zip(callables, _sample_args): + if isinstance(c, torch.nn.Module): + if not ( + len(c._backward_hooks) == 0 + and len(c._forward_hooks) == 0 + and len(c._forward_pre_hooks) == 0 + ): + raise AssertionError( + "Modules must not have hooks registered at the time they are passed. However, registering hooks " + + "on modules after passing them through make_graphed_callables is allowed." + ) + if not all(b.requires_grad is False for b in c.buffers()): + raise AssertionError( + "In any :class:`~torch.nn.Module` passed to " + + ":func:`~make_graphed_callables`, only parameters may be trainable. All buffers must have " + + "``requires_grad=False``." + ) + flatten_arg = torch.utils._pytree.arg_tree_leaves(*args) + flatten_sample_args.append(tuple(flatten_arg)) + if not all(isinstance(arg, torch.Tensor) for arg in flatten_arg): + raise AssertionError( + "In the beta API, sample_args " + + "for each callable must contain only Tensors. Other types are not allowed." + ) + + # If a callable is an nn.Module, its graph's full input surface is the args the user explicitly + # passes to forward (ie, its sample_args) AND the module's parameter attributes. + per_callable_len_user_args = [len(args) for args in flatten_sample_args] + per_callable_module_params = [ + tuple(c.parameters()) if isinstance(c, torch.nn.Module) else () + for c in callables + ] + per_callable_static_input_surfaces = [ + flatten_sample_args[i] + per_callable_module_params[i] + for i in range(len(callables)) + ] + + fwd_graphs = [torch.cuda.CUDAGraph() for _ in range(len(callables))] + bwd_graphs = [torch.cuda.CUDAGraph() for _ in range(len(callables))] + + mempool = graph_pool_handle() if pool is None else _get_pool_id(pool) + + # Warmup + # Hopefully prevents cudnn benchmarking and other lazy-initialization cuda work + # from ending up in any captures. + torch.cuda.synchronize() + # hipBLASLt handles are per-(device, stream) on ROCm and lazily created. + # Need to use the same stream for warmup and capture to avoid capture errors. + stream = torch.cuda.Stream() + with torch.cuda.stream(stream): + for func, args, static_input_surface in zip( + callables, _sample_args, per_callable_static_input_surfaces + ): + grad_inputs, outputs, outputs_grad = None, None, None + for _ in range(num_warmup_iters): + outputs = torch.utils._pytree.tree_leaves(func(*args)) + outputs_grad = tuple(o for o in outputs if o.requires_grad) + if len(outputs_grad) > 0: + grad_inputs = torch.autograd.grad( + outputs=outputs_grad, + inputs=tuple( + i for i in static_input_surface if i.requires_grad + ), + grad_outputs=tuple( + torch.empty_like(o) for o in outputs if o.requires_grad + ), + only_inputs=True, + allow_unused=allow_unused_input, + ) + for v in [outputs, outputs_grad, grad_inputs]: + del v + + torch.cuda.synchronize() + + # All captures here share a mempool. To avoid replays corrupting each other's memory, + # the safest approach is to capture all passes in the same order they'll run: + # fwd 1, fwd 2, ... fwd N, then bwd N, bwd N-1, ... bwd 1. + + # Capture forward graphs + per_callable_static_outputs = [] + per_callable_output_unflatten_spec = [] + for func, args, fwd_graph in zip(callables, _sample_args, fwd_graphs): + with torch.cuda.graph( + fwd_graph, + stream=stream, + pool=mempool, + capture_error_mode=capture_error_mode, + enable_annotations=enable_annotations, + ): + func_outputs = func(*args) + + flatten_outputs, spec = torch.utils._pytree.tree_flatten(func_outputs) + per_callable_static_outputs.append(tuple(flatten_outputs)) + per_callable_output_unflatten_spec.append(spec) + + # Capture backward graphs in reverse order + per_callable_static_grad_outputs = [] + per_callable_static_grad_inputs = [] + for static_input_surface, static_outputs, bwd_graph in zip( + reversed(per_callable_static_input_surfaces), + reversed(per_callable_static_outputs), + reversed(bwd_graphs), + ): + # For now, assumes all static_outputs require grad + # assert all(o.requires_grad for o in static_outputs), "Outputs of graphed callables must require grad." + static_grad_outputs = tuple( + torch.empty_like(o) if o.requires_grad else None for o in static_outputs + ) + + outputs_grad = tuple(o for o in static_outputs if o.requires_grad) + grad_inputs = None + if len(outputs_grad) > 0: + with torch.cuda.graph( + bwd_graph, + stream=stream, + pool=mempool, + capture_error_mode=capture_error_mode, + enable_annotations=enable_annotations, + ): + grad_inputs = torch.autograd.grad( + outputs=outputs_grad, + inputs=tuple(i for i in static_input_surface if i.requires_grad), + grad_outputs=tuple(o for o in static_grad_outputs if o is not None), + only_inputs=True, + allow_unused=allow_unused_input, + ) + + # Constructs a tuple suitable for returning from Graphed.backward: + # Pads out the actually-needed grads with Nones in gradient slots for inputs that don't require grad. + # I couldn't think of a slick one-liner for this pattern. + static_grad_inputs = [] + grad_idx = 0 + for arg in static_input_surface: + if arg.requires_grad and grad_inputs is not None: + static_grad_inputs.append(grad_inputs[grad_idx]) + grad_idx += 1 + else: + static_grad_inputs.append(None) # type: ignore[arg-type] + static_grad_inputs = tuple(static_grad_inputs) # type: ignore[assignment] + + per_callable_static_grad_outputs.append(static_grad_outputs) + per_callable_static_grad_inputs.append(static_grad_inputs) + + # Reverses the most recent two lists + per_callable_static_grad_outputs.reverse() + per_callable_static_grad_inputs.reverse() + # Now for every per_callable list, per_callable_*[i] holds the stuff for the ith callable. + + def make_graphed_autograd_function( + fwd_graph: CUDAGraph, + bwd_graph: CUDAGraph, + module_params: tuple[torch.nn.Parameter, ...], + len_user_args: int, + output_unflatten_spec: torch.utils._pytree.TreeSpec, + static_input_surface: tuple[Tensor, ...], + static_outputs: tuple[Tensor, ...], + static_grad_outputs: tuple[Tensor | None, ...], + static_grad_inputs: tuple[Tensor, ...], + ) -> Callable[..., object]: + class Graphed(torch.autograd.Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx: object, *inputs: Tensor) -> tuple[Tensor, ...]: + # At this stage, only the user args may (potentially) be new tensors. + for i in range(len_user_args): + if static_input_surface[i].data_ptr() != inputs[i].data_ptr(): + static_input_surface[i].copy_(inputs[i]) + fwd_graph.replay() + if not isinstance(static_outputs, tuple): + raise AssertionError( + f"static_outputs must be tuple, got {type(static_outputs)}" + ) + return tuple(o.detach() for o in static_outputs) + + @staticmethod + @torch.autograd.function.once_differentiable + # pyrefly: ignore [bad-override] + def backward(ctx: object, *grads: Tensor) -> tuple[Tensor, ...]: + if len(grads) != len(static_grad_outputs): + raise AssertionError( + f"len(grads)={len(grads)} != len(static_grad_outputs)={len(static_grad_outputs)}" + ) + for g, grad in zip(static_grad_outputs, grads): + if g is not None: + # don't copy if autograd gods have been kind and the + # incoming grad is already in the right place + if g.data_ptr() != grad.data_ptr(): + g.copy_(grad) + bwd_graph.replay() + + # Input args that didn't require grad expect a None gradient. + if not isinstance(static_grad_inputs, tuple): + raise AssertionError( + f"static_grad_inputs must be tuple, got {type(static_grad_inputs)}" + ) + return tuple( + # pyrefly: ignore [bad-argument-type] + b.detach() if b is not None else b + for b in static_grad_inputs + ) + + def functionalized(*user_args: object) -> object: + # Runs the autograd function with inputs == all inputs to the graph that might require grad + # (explicit user args + module parameters) + # Assumes module params didn't change since capture. + flatten_user_args = torch.utils._pytree.arg_tree_leaves(*user_args) + out = Graphed.apply(*(tuple(flatten_user_args) + module_params)) + return torch.utils._pytree.tree_unflatten(out, output_unflatten_spec) + + return functionalized + + # Put together the final graphed callables + ret: list[_ModuleOrCallable] = [] + for i, func in enumerate(callables): + graphed = make_graphed_autograd_function( + fwd_graphs[i], + bwd_graphs[i], + per_callable_module_params[i], + per_callable_len_user_args[i], + per_callable_output_unflatten_spec[i], + per_callable_static_input_surfaces[i], + per_callable_static_outputs[i], + per_callable_static_grad_outputs[i], + per_callable_static_grad_inputs[i], + ) + + if isinstance(func, torch.nn.Module): + + def make_graphed_forward( + func: torch.nn.Module, + graph_training_state: bool, + graphed: Callable[_P, _R], + orig_fwd: Callable[_P, _R], + ) -> Callable[_P, _R]: + def new_fwd(*user_args: _P.args, **user_kwargs: _P.kwargs) -> _R: + # If the module's training-or-eval state matches what we graphed, + # run the graph, otherwise run the original forward method + if func.training == graph_training_state: + return graphed(*user_args, **user_kwargs) + else: + return orig_fwd(*user_args, **user_kwargs) + + return new_fwd + + func.forward = make_graphed_forward( + func, func.training, graphed, func.forward + ) + ret.append(func) + else: + ret.append(graphed) + + if just_one_callable: + return ret[0] + + return tuple(ret) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/green_contexts.py b/.venv/lib/python3.12/site-packages/torch/cuda/green_contexts.py new file mode 100644 index 0000000000000000000000000000000000000000..b8e2a1b1b2daae4178e13b217ec85c9fbc610900 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/green_contexts.py @@ -0,0 +1,387 @@ +from __future__ import annotations + +import functools +import sys +import warnings +from typing import Any +from typing_extensions import deprecated + +import torch +from torch.cuda._utils import ( + _check_cuda_bindings, + _cuda_bindings_driver as _drv, + _cuda_bindings_runtime as _rt, + _HAS_CUDA_BINDINGS, +) + + +__all__ = [ + "GreenContext", +] + +_STREAMS_PER_GREEN_CONTEXT_POOL = 32 + +_WORKQUEUE_SCOPE_VALUES = { + "device_ctx": 0, + "balanced": 1, +} + +_CONTEXT_STACK_DEPRECATION = ( + "`GreenContext.set_context` and `GreenContext.pop_context` are deprecated. " + "Please create a stream with `GreenContext.Stream()` and use " + "`torch.cuda.stream(stream)` instead." +) + + +# note: this can safely be cached in a process/thread because +# the driver version cannot change during the lifetime of a process +@functools.cache +def _get_driver_version() -> int: + try: + # pyrefly: ignore [missing-attribute] + return _check_cuda_bindings(_drv.cuDriverGetVersion()) + except RuntimeError as e: + warnings.warn(f"Error while querying CUDA driver version: {e}") + return -1 + + +def _ensure_driver_version(version: int, message: str) -> None: + drv_version = _get_driver_version() + if drv_version < 0 or drv_version < version: + raise RuntimeError(message) + + +def _ensure_supported() -> None: + if torch.version.cuda is None or torch.version.hip is not None: + raise RuntimeError("Green Context is only supported on Nvidia CUDA") + if sys.platform == "win32": + raise RuntimeError("Green Context is not supported on Windows") + if not _HAS_CUDA_BINDINGS: + raise RuntimeError("GreenContext requires the cuda.bindings package") + _ensure_driver_version(12080, "Green Context requires user mode driver 12.8+") + + +def _ensure_workqueue_supported() -> None: + _ensure_driver_version( + 13010, "CUDA user mode driver too old to use workqueue configuration!" + ) + + +def _parse_workqueue_scope(workqueue_scope: str | None) -> int | None: + if workqueue_scope is None: + return None + if workqueue_scope not in _WORKQUEUE_SCOPE_VALUES: + raise ValueError( + "workqueue_scope must be 'device_ctx' or 'balanced', " + f"got '{workqueue_scope}'" + ) + return _WORKQUEUE_SCOPE_VALUES[workqueue_scope] + + +class GreenContext: + r"""Wrapper around a CUDA green context. + + .. warning:: + This API is in beta and may change in future releases. + + CUDA work should be placed on streams created from the green context: + + .. code-block:: python + + ctx = GreenContext(...) + stream = ctx.Stream() + with torch.cuda.stream(stream): + # torch operations here are using resources from `ctx` + pass + + Green-context streams are custom CUDA streams. Synchronization with other + streams is the user's responsibility and should be handled with CUDA events, + as with any other custom stream. + """ + + def __init__( + self, + *, + num_sms: int | None = None, + workqueue_scope: str | None = None, + workqueue_concurrency_limit: int | None = None, + device_id: int | None = None, + ) -> None: + r"""Create a CUDA green context. + + At least one of ``num_sms`` or ``workqueue_scope`` must be specified. + Both can be combined to partition SMs and configure workqueues in the + same green context. + + Arguments: + num_sms (int, optional): The number of SMs to use in the green + context. When ``None``, SMs are not partitioned. + workqueue_scope (str, optional): Workqueue sharing scope. One of + ``"device_ctx"`` (shared across all contexts, default driver + behavior) or ``"balanced"`` (non-overlapping workqueues with + other balanced green contexts). When ``None``, no workqueue + configuration is applied. + workqueue_concurrency_limit (int, optional): Maximum number of + concurrent stream-ordered workloads for the workqueue. Requires + ``workqueue_scope`` to be set. + device_id (int, optional): The device index of green context. + When ``None``, the current device is used. + """ + self._device_id = None + self._green_ctx = None + self._context = None + _ensure_supported() + + scope_value = _parse_workqueue_scope(workqueue_scope) + if scope_value is not None: + _ensure_workqueue_supported() + + if num_sms is None and scope_value is None: + raise RuntimeError( + "At least one of num_sms or workqueue_scope must be specified" + ) + if workqueue_concurrency_limit is not None and scope_value is None: + raise RuntimeError( + "workqueue_concurrency_limit requires workqueue_scope to be set" + ) + + if device_id is None: + device_id = torch.cuda.current_device() + + # pyrefly: ignore [missing-attribute] + current_ctx = _check_cuda_bindings(_drv.cuCtxGetCurrent()) + + if int(current_ctx) == 0: + warnings.warn( + "Attempted to create a green context but there was no primary " + "context! Creating a primary context...", + stacklevel=2, + ) + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_rt.cudaFree(0)) + + # pyrefly: ignore [missing-attribute] + device = _check_cuda_bindings(_drv.cuDeviceGet(device_id)) + resources = [] + + if num_sms is not None: + sm_resource = _check_cuda_bindings( + _drv.cuDeviceGetDevResource( # pyrefly: ignore [missing-attribute] + device, + _drv.CUdevResourceType.CU_DEV_RESOURCE_TYPE_SM, # pyrefly: ignore [missing-attribute] + ) + ) + if num_sms <= 0 or num_sms > sm_resource.sm.smCount: + raise RuntimeError( + "Invalid number of SMs requested for green context: " + f"{num_sms} (device has {sm_resource.sm.smCount} SMs)" + ) + split_result, nb_groups, _remaining = _check_cuda_bindings( + _drv.cuDevSmResourceSplitByCount( # pyrefly: ignore [missing-attribute] + 1, sm_resource, 0, num_sms + ) + ) + if nb_groups != 1: + raise RuntimeError("Failed to create single SM resource group") + resources.append(split_result[0]) + + if scope_value is not None: + wq_resource = _check_cuda_bindings( + _drv.cuDeviceGetDevResource( # pyrefly: ignore [missing-attribute] + device, + _drv.CUdevResourceType.CU_DEV_RESOURCE_TYPE_WORKQUEUE_CONFIG, # pyrefly: ignore [missing-attribute] + ) + ) + wq_resource.wqConfig.sharingScope = scope_value + if workqueue_concurrency_limit is not None: + wq_resource.wqConfig.wqConcurrencyLimit = workqueue_concurrency_limit + resources.append(wq_resource) + + desc = _check_cuda_bindings( + _drv.cuDevResourceGenerateDesc( # pyrefly: ignore [missing-attribute] + resources, len(resources) + ) + ) + green_ctx = _check_cuda_bindings( + _drv.cuGreenCtxCreate( # pyrefly: ignore [missing-attribute] + desc, + device, + _drv.CUgreenCtxCreate_flags.CU_GREEN_CTX_DEFAULT_STREAM, # pyrefly: ignore [missing-attribute] + ) + ) + try: + # pyrefly: ignore [missing-attribute] + context = _check_cuda_bindings(_drv.cuCtxFromGreenCtx(green_ctx)) + if int(context) == 0: + raise RuntimeError("Green ctx conversion to regular ctx failed!") + except Exception: + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_drv.cuGreenCtxDestroy(green_ctx)) + raise + + self._init_from_cuda_objects(device_id, green_ctx, context) + + def __del__(self) -> None: + green_ctx = getattr(self, "_green_ctx", None) + if green_ctx is None: + return + + # attempt to destroy streams related to this green context + # we ignore errors to avoid leaking exceptions during __del__ + end = min(_STREAMS_PER_GREEN_CONTEXT_POOL, self._curr_stream_idx + 1) + for idx in reversed(range(end)): + try: + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_drv.cuStreamDestroy(self._green_ctx_streams[idx])) + except RuntimeError as e: + warnings.warn( + f"Error while destroying green context stream at idx {idx} " + f"for green context {green_ctx}: {e}" + ) + self._green_ctx = None + try: + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_drv.cuGreenCtxDestroy(green_ctx)) + except RuntimeError as e: + warnings.warn(f"Error while destroying green context {green_ctx}: {e}") + + def _init_from_cuda_objects(self, device_id: int, green_ctx, context) -> None: + self._device_id = device_id + self._green_ctx = green_ctx + self._context = context + self._parent_stream: torch.cuda.Stream | None = None + self._green_ctx_streams: list[Any | None] = [ + None + ] * _STREAMS_PER_GREEN_CONTEXT_POOL + self._curr_stream_idx = -1 + + @staticmethod + def create( + *, + num_sms: int | None = None, + workqueue_scope: str | None = None, + workqueue_concurrency_limit: int | None = None, + device_id: int | None = None, + ) -> GreenContext: + r"""Create a CUDA green context. + + Kept for compatibility, see `GreenContext` constructor. + """ + return GreenContext( + num_sms=num_sms, + workqueue_scope=workqueue_scope, + workqueue_concurrency_limit=workqueue_concurrency_limit, + device_id=device_id, + ) + + @staticmethod + def max_workqueue_concurrency(device_id: int | None = None) -> int: + r"""Return the maximum workqueue concurrency limit for the device. + + This queries the device for the default number of concurrent + stream-ordered workloads supported by workqueue configuration + resources. + + Arguments: + device_id (int, optional): The device index to query. When + ``None``, the current device is used. + """ + _ensure_supported() + _ensure_workqueue_supported() + if device_id is None: + device_id = torch.cuda.current_device() + + # pyrefly: ignore [missing-attribute] + device = _check_cuda_bindings(_drv.cuDeviceGet(device_id)) + wq_resource = _check_cuda_bindings( + _drv.cuDeviceGetDevResource( # pyrefly: ignore [missing-attribute] + device, + _drv.CUdevResourceType.CU_DEV_RESOURCE_TYPE_WORKQUEUE_CONFIG, # pyrefly: ignore [missing-attribute] + ) + ) + return wq_resource.wqConfig.wqConcurrencyLimit + + def _ensure_alive(self) -> None: + if self._green_ctx is None or self._context is None: + raise RuntimeError("GreenContext has been destroyed") + + @deprecated(_CONTEXT_STACK_DEPRECATION, category=FutureWarning) + def set_context(self) -> None: + r"""Make the green context the current context. + + Deprecated. Create streams with :meth:`Stream` and use + :func:`torch.cuda.stream` instead. + """ + self._ensure_alive() + if self._parent_stream is not None: + raise RuntimeError("set_context called twice before pop_context") + current_stream = torch.cuda.current_stream() + self._parent_stream = current_stream + + event = torch.cuda.Event() + event.record(current_stream) + + # pyrefly: ignore [missing-attribute] + current_ctx = _check_cuda_bindings(_drv.cuCtxGetCurrent()) + if int(current_ctx) == 0: + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_drv.cuCtxSetCurrent(self._context)) + else: + # pyrefly: ignore [missing-attribute] + _check_cuda_bindings(_drv.cuCtxPushCurrent(self._context)) + + green_ctx_stream = torch.cuda.default_stream(self._device_id) + event.wait(green_ctx_stream) + torch.cuda.set_stream(green_ctx_stream) + + @deprecated(_CONTEXT_STACK_DEPRECATION, category=FutureWarning) + def pop_context(self) -> None: + r"""Assuming the green context is the current context, pop it from the + context stack and restore the previous context. + + Deprecated. Create streams with :meth:`Stream` and use + :func:`torch.cuda.stream` instead. + """ + try: + self._ensure_alive() + if self._parent_stream is None: + raise RuntimeError("pop_context called without matching set_context") + + event = torch.cuda.Event() + event.record(torch.cuda.current_stream()) + + # pyrefly: ignore [missing-attribute] + popped = _check_cuda_bindings(_drv.cuCtxPopCurrent()) + # pyrefly: ignore [bad-argument-type] + if int(popped) != int(self._context): + raise RuntimeError("expected popped context to be the current ctx") + + event.wait(self._parent_stream) + torch.cuda.set_stream(self._parent_stream) + finally: + self._parent_stream = None + + def Stream(self) -> torch.cuda.Stream: + r"""Return a CUDA stream associated with this green context. + + Use the returned stream with :func:`torch.cuda.stream` to run work on + the green context. Synchronization with other streams is not automatic; + use CUDA events as with any other custom stream. + """ + self._ensure_alive() + curr_idx = self._curr_stream_idx + 1 + idx = curr_idx % _STREAMS_PER_GREEN_CONTEXT_POOL + if curr_idx < _STREAMS_PER_GREEN_CONTEXT_POOL: + green_ctx_stream = _check_cuda_bindings( + _drv.cuGreenCtxStreamCreate( # pyrefly: ignore [missing-attribute] + self._green_ctx, + _drv.CUstream_flags.CU_STREAM_NON_BLOCKING, # pyrefly: ignore [missing-attribute] + 0, + ) + ) + self._green_ctx_streams[idx] = green_ctx_stream + else: + green_ctx_stream = self._green_ctx_streams[idx] + self._curr_stream_idx = curr_idx + # pyrefly: ignore [bad-argument-type] + return torch.cuda.ExternalStream(int(green_ctx_stream), self._device_id) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/jiterator.py b/.venv/lib/python3.12/site-packages/torch/cuda/jiterator.py new file mode 100644 index 0000000000000000000000000000000000000000..4b0c27e8b94cb2e36115b91f313651feb9dc9550 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/jiterator.py @@ -0,0 +1,192 @@ +# mypy: allow-untyped-defs +import re +from collections.abc import Callable + +import torch +from torch import Tensor + + +__all__: list[str] = [] + + +class _CodeParser: + def __init__(self, code_string: str): + optional_ws = r"\s*" + required_ws = r"\s+" + template_params = r"(?P\<.+\>)" + return_type = r"(?P\w+)" + function_name = r"(?P\w+)" + function_params = r"(?P\(.+\))" + function_body = r"(?P\{.+\})" + + pattern = ( + optional_ws + + "template" + + optional_ws + + template_params + + optional_ws + + return_type + + required_ws + + function_name + + optional_ws + + function_params + + optional_ws + + function_body + + optional_ws + ) + + result = re.match( + pattern, code_string, re.DOTALL + ) # DOTALL for matching multiline + + if result is None: + raise Exception( # noqa: TRY002 + f"Couldn't parse code, please check correctness:\n {code_string}" + ) + + self.template_params = result["template_params"] + self.return_type = result["return_type"] + self.function_name = result["function_name"] + self.function_params = result["function_params"] + self.function_body = result["function_body"] + + +class _JittedFunction: + def __init__( + self, code_string: str, return_by_ref: bool, num_outputs: int, **kwargs + ): + self.code_string = code_string + + if not (return_by_ref or num_outputs == 1): + raise AssertionError("Return by value only works for single output.") + self.return_by_ref = return_by_ref + self.num_outputs = num_outputs + + parsed_code = _CodeParser(code_string) + self.kernel_name = parsed_code.function_name + + self.kwargs_dict = kwargs + self.is_cuda_available = torch.cuda.is_available() + + def __call__(self, *tensors: Tensor, **kwargs): + # Jiterator follow torch.cuda's lazy initialization behavior + # Defer checking cuda's availability at the function invocation time + if not self.is_cuda_available: + raise AssertionError( + "Jiterator is only supported on CUDA and ROCm GPUs, none are available." + ) + + if len(tensors) > 8: + raise AssertionError( + f"jiterator only supports up to 8 tensor inputs, got {len(tensors)}" + ) + + expanded_kwargs = self.kwargs_dict.copy() + for key, value in kwargs.items(): + if key in self.kwargs_dict: + expanded_kwargs[key] = value + else: + raise KeyError(f"{key} is not declared in function definition") + + return torch._C._cuda_jiterator_compile_and_launch_kernel( + self.code_string, + self.kernel_name, + self.return_by_ref, + self.num_outputs, + tensors, + expanded_kwargs, + ) + + +def _create_jit_fn(code_string: str, **kwargs) -> Callable: + """ + Create a jiterator-generated cuda kernel for an elementwise op. + + The code string has to be a valid CUDA function that describes the computation for a single element. The code + string has to follow the c++ template pattern, as shown in the example below. This function will be inlined + into elementwise kernel template, and compiled on the fly. Compiled kernel will be cached in memory, as well as + local temp dir. + + Jiterator-generated kernels accept noncontiguous tensors, and support broadcasting and type promotion. + + Args: + code_string (str): CUDA code string to be compiled by jiterator. The entry functor must return by value. + kwargs (Dict, optional): Keyword arguments for generated function + + Example:: + + code_string = "template T my_kernel(T x, T y, T alpha) { return -x + alpha * y; }" + jitted_fn = create_jit_fn(code_string, alpha=1.0) + a = torch.rand(3, device="cuda") + b = torch.rand(3, device="cuda") + # invoke jitted function like a regular python function + result = jitted_fn(a, b, alpha=3.14) + + code_string also allows multiple function definitions, and the last function will be treated as the entry function. + + Example:: + + code_string = ( + "template T util_fn(T x, T y) { return ::sin(x) + ::cos(y); }" + ) + code_string += "template T my_kernel(T x, T y, T val) { return ::min(val, util_fn(x, y)); }" + jitted_fn = create_jit_fn(code_string, val=0.0) + a = torch.rand(3, device="cuda") + b = torch.rand(3, device="cuda") + # invoke jitted function like a regular python function + result = jitted_fn(a, b) # using default val=0.0 + + Jiterator can be used together with python registration to override an operator's cuda kernel. + Following example is overriding gelu's cuda kernel with relu. + + Example:: + + code_string = "template T my_gelu(T a) { return a > 0 ? a : 0; }" + my_gelu = create_jit_fn(code_string) + my_lib = torch.library.Library("aten", "IMPL") + my_lib.impl("aten::gelu", my_gelu, "CUDA") + # torch.nn.GELU and torch.nn.function.gelu are now overridden + a = torch.rand(3, device="cuda") + torch.allclose(torch.nn.functional.gelu(a), torch.nn.functional.relu(a)) + + .. warning:: + This API is in beta and may change in future releases. + + .. warning:: + This API only supports up to 8 inputs and 1 output + + .. warning:: + All input tensors must live in CUDA device + """ + return _JittedFunction(code_string, return_by_ref=False, num_outputs=1, **kwargs) + + +def _create_multi_output_jit_fn( + code_string: str, num_outputs: int, **kwargs +) -> Callable: + """ + Create a jiterator-generated cuda kernel for an elementwise op that supports returning one or more outputs. + + Args: + code_string (str): CUDA code string to be compiled by jiterator. The entry functor must return value by reference. + num_outputs(int): number of outputs return by the kernel + kwargs (Dict, optional): Keyword arguments for generated function + + Example:: + + code_string = "template void my_kernel(T x, T y, T alpha, T& out) { out = -x + alpha * y; }" + jitted_fn = create_jit_fn(code_string, alpha=1.0) + a = torch.rand(3, device="cuda") + b = torch.rand(3, device="cuda") + # invoke jitted function like a regular python function + result = jitted_fn(a, b, alpha=3.14) + + .. warning:: + This API is in beta and may change in future releases. + + .. warning:: + This API only supports up to 8 inputs and 8 outputs + """ + return _JittedFunction( + code_string, return_by_ref=True, num_outputs=num_outputs, **kwargs + ) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/memory.py b/.venv/lib/python3.12/site-packages/torch/cuda/memory.py new file mode 100644 index 0000000000000000000000000000000000000000..a551ae599bd41c7e1eab3952c232d2f598379b24 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/memory.py @@ -0,0 +1,1642 @@ +# mypy: allow-untyped-defs +r"""This package adds support for device memory management implemented in CUDA.""" + +import collections +import contextlib +import ctypes +import pickle +import sys +import threading +import warnings +from inspect import signature +from typing import Any, Literal, TYPE_CHECKING +from typing_extensions import deprecated + +import torch +from torch import _C +from torch._utils import _augment_memory_snapshot_stack_traces, _dummy_type + +from . import ( + _get_amdsmi_device_index, + _get_device_index, + _get_nvml_device_index, + _lazy_init, + is_initialized, +) +from ._memory_viz import memory as _memory, segments as _segments + + +if TYPE_CHECKING: + from torch.types import Device + + +__all__ = [ + "caching_allocator_alloc", + "caching_allocator_delete", + "caching_allocator_disabled", + "caching_allocator_enable", + "get_per_process_memory_fraction", + "set_per_process_memory_fraction", + "empty_cache", + "memory_stats", + "memory_stats_as_nested_dict", + "reset_accumulated_memory_stats", + "reset_peak_memory_stats", + "reset_max_memory_allocated", + "reset_max_memory_cached", + "host_memory_stats", + "host_memory_stats_as_nested_dict", + "reset_accumulated_host_memory_stats", + "reset_peak_host_memory_stats", + "memory_allocated", + "max_memory_allocated", + "memory_reserved", + "max_memory_reserved", + "memory_cached", + "max_memory_cached", + "memory_snapshot", + "memory_summary", + "list_gpu_processes", + "mem_get_info", + "get_allocator_backend", + "CUDAPluggableAllocator", + "change_current_allocator", + "MemPool", + "use_mem_pool", +] + + +if not hasattr(torch._C, "_cuda_CUDAAllocator"): + # Define dummy base classes + torch._C.__dict__["_cuda_CUDAAllocator"] = _dummy_type("_cuda_CUDAAllocator") + + +if not hasattr(torch._C, "_MemPool"): + # Define dummy base classes + torch._C.__dict__["_MemPool"] = _dummy_type("_MemPool") + torch._C.__dict__["_cuda_beginAllocateToPool"] = _dummy_type( + "_cuda_beginAllocateToPool" + ) + torch._C.__dict__["_cuda_beginAllocateCurrentThreadToPool"] = _dummy_type( + "_cuda_beginAllocateCurrentThreadToPool" + ) + torch._C.__dict__["_cuda_endAllocateToPool"] = _dummy_type( + "_cuda_endAllocateToPool" + ) + torch._C.__dict__["_cuda_releasePool"] = _dummy_type("_cuda_releasePool") + +from torch._C import ( # noqa: F401 + _cuda_beginAllocateCurrentThreadToPool, + _cuda_beginAllocateToPool, + _cuda_CUDAAllocator, + _cuda_endAllocateToPool, + _cuda_releasePool, + _MemPool, +) + + +def _host_allocator(): + _lazy_init() + return torch._C._cuda_cudaHostAllocator() + + +def caching_allocator_alloc(size, device: "Device" = None, stream=None): + r"""Perform a memory allocation using the CUDA memory allocator. + + Memory is allocated for a given device and a stream, this + function is intended to be used for interoperability with other + frameworks. Allocated memory is released through + :func:`~torch.cuda.caching_allocator_delete`. + + Args: + size (int): number of bytes to be allocated. + device (torch.device or int, optional): selected device. If it is + ``None`` the default CUDA device is used. + stream (torch.cuda.Stream or int, optional): selected stream. If is ``None`` then + the default stream for the selected device is used. + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + if device is None: + device = torch.cuda.current_device() + device = _get_device_index(device) + if stream is None: + stream = torch.cuda.current_stream(device) + if isinstance(stream, torch.cuda.streams.Stream): + stream = stream.cuda_stream + if not isinstance(stream, int): + raise TypeError( + "Invalid type for stream argument, must be " + "`torch.cuda.Stream` or `int` representing a pointer " + "to an existing stream" + ) + with torch.cuda.device(device): + return torch._C._cuda_cudaCachingAllocator_raw_alloc(size, stream) + + +def caching_allocator_delete(mem_ptr): + r"""Delete memory allocated using the CUDA memory allocator. + + Memory allocated with :func:`~torch.cuda.caching_allocator_alloc`. + is freed here. The associated device and stream are tracked inside + the allocator. + + Args: + mem_ptr (int): memory address to be freed by the allocator. + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + torch._C._cuda_cudaCachingAllocator_raw_delete(mem_ptr) + + +def caching_allocator_enable(value: bool = True) -> None: + r"""Enable or disable the CUDA memory allocator. On by default.""" + if is_initialized(): + torch._C._cuda_cudaCachingAllocator_enable(value) + + +@contextlib.contextmanager +def caching_allocator_disabled(): + r"""Context manager that temporarily disables the CUDA caching allocator.""" + # pyrefly: ignore [missing-attribute] + prev = torch._C._cuda_cudaCachingAllocator_is_enabled() + caching_allocator_enable(False) + try: + yield + finally: + caching_allocator_enable(prev) + + +def set_per_process_memory_fraction(fraction, device: "Device" = None) -> None: + r"""Set memory fraction for a process. + + The fraction is used to limit an caching allocator to allocated memory on a CUDA device. + The allowed value equals the total visible memory multiplied fraction. + If trying to allocate more than the allowed value in a process, will raise an out of + memory error in allocator. + + Args: + fraction(float): Range: 0~1. Allowed memory equals total_memory * fraction. + device (torch.device or int, optional): selected device. If it is + ``None`` the default CUDA device is used. + .. note:: + In general, the total available free memory is less than the total capacity. + """ + _lazy_init() + if device is None: + device = torch.cuda.current_device() + device = _get_device_index(device) + if not isinstance(fraction, float): + raise TypeError("Invalid type for fraction argument, must be `float`") + if fraction < 0 or fraction > 1: + raise ValueError(f"Invalid fraction value: {fraction}. Allowed range: 0~1") + + torch._C._cuda_setMemoryFraction(fraction, device) + + +def get_per_process_memory_fraction(device: "Device" = None) -> float: + r"""Get memory fraction for a process. + + Args: + device (torch.device or int, optional): selected device. If it is + ``None`` the default CUDA device is used. + Returns: + memory fraction, in range 0~1. Allowed memory equals total_memory * fraction. + """ + _lazy_init() + if device is None: + device = torch.cuda.current_device() + device = _get_device_index(device) + return torch._C._cuda_getMemoryFraction(device) + + +def empty_cache() -> None: + r"""Release all unoccupied cached memory currently held by the caching + allocator so that those can be used in other GPU application and visible in + `nvidia-smi`. + + .. note:: + :func:`~torch.cuda.empty_cache` doesn't increase the amount of GPU + memory available for PyTorch. However, it may help reduce fragmentation + of GPU memory in certain cases. See :ref:`cuda-memory-management` for + more details about GPU memory management. + """ + if is_initialized(): + torch._C._cuda_emptyCache() + + +def _recurse_add_to_result(result, prefix, obj, format_key): + if isinstance(obj, dict): + if prefix: + prefix += "." + for key, value in obj.items(): + _recurse_add_to_result(result, prefix + format_key(key), value, format_key) + else: + result.append((prefix, obj)) + + +def memory_stats(device: "Device" = None) -> dict[str, Any]: + r"""Return a dictionary of CUDA memory allocator statistics for a given device. + + The return value of this function is a dictionary of statistics, each of + which is a non-negative integer. + + Core statistics: + + - ``"allocated.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + number of allocation requests received by the memory allocator. + - ``"allocated_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + amount of allocated memory. + - ``"segment.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + number of reserved segments from ``cudaMalloc()``. + - ``"reserved_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + amount of reserved memory. + - ``"active.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + number of active memory blocks. + - ``"active_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + amount of active memory. + - ``"inactive_split.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + number of inactive, non-releasable memory blocks. + - ``"inactive_split_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + amount of inactive, non-releasable memory. + + For these core statistics, values are broken down as follows. + + Pool type: + + - ``all``: combined statistics across all memory pools. + - ``large_pool``: statistics for the large allocation pool + (as of June 2025, for size >= 1MB allocations). + - ``small_pool``: statistics for the small allocation pool + (as of June 2025, for size < 1MB allocations). + + Metric type: + + - ``current``: current value of this metric. + - ``peak``: maximum value of this metric. + - ``allocated``: historical total increase in this metric. + - ``freed``: historical total decrease in this metric. + + In addition to the core statistics, we also provide some simple event + counters: + + - ``"num_alloc_retries"``: number of failed ``cudaMalloc`` calls that + result in a cache flush and retry. + - ``"num_ooms"``: number of out-of-memory errors thrown. + - ``"num_sync_all_streams"``: number of ``synchronize_and_free_events`` calls. + - ``"num_device_alloc"``: number of CUDA allocation calls. This includes both + cuMemMap and cudaMalloc. + - ``"num_device_free"``: number of CUDA free calls. This includes both cuMemUnmap + and cudaFree. + - ``"num_oom_rejections"``: number of allocations preemptively rejected by the + throw_on_cudamalloc_oom + per_process_memory_fraction policy. + + The caching allocator can be configured via ENV to not split blocks larger than a + defined size (see Memory Management section of the Cuda Semantics documentation). + This helps avoid memory fragmentation but may have a performance + penalty. Additional outputs to assist with tuning and evaluating impact: + + - ``"max_split_size"``: blocks above this size will not be split. + - ``"oversize_allocations.{current,peak,allocated,freed}"``: + number of over-size allocation requests received by the memory allocator. + - ``"oversize_segments.{current,peak,allocated,freed}"``: + number of over-size reserved segments from ``cudaMalloc()``. + + The caching allocator can be configured via ENV to round memory allocations in order + to reduce fragmentation. Sometimes the overhead from rounding can be higher than + the fragmentation it helps reduce. The following stat can be used to check if + rounding adds too much overhead: + + - ``"requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``: + memory requested by client code, compare this with allocated_bytes to check if + allocation rounding adds too much overhead. + - ``"reserved_bytes_by_private_pools"``: nested dictionary keyed by + ``torch.cuda.MemPool.id`` tuples. Each value has the same + ``{all,large_pool,small_pool}.{current,peak,allocated,freed}`` structure + as ``reserved_bytes``, but scoped to a single private pool. In + :func:`~torch.cuda.memory_stats`, tuple keys are flattened by joining + their stringified elements with ``"_"``, so ``(0, 1)`` becomes ``"0_1"``. + + Args: + device (torch.device or int, optional): selected device. Returns + statistics for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + + .. note:: + With :ref:`backend:cudaMallocAsync`, some stats are not + meaningful, and are always reported as zero. + """ + result = [] + + def _format_key(key): + if isinstance(key, str): + return key + if isinstance(key, tuple): + return "_".join(str(part) for part in key) + return str(key) + + stats = memory_stats_as_nested_dict(device=device) + _recurse_add_to_result(result, "", stats, _format_key) + result.sort() + + return collections.OrderedDict(result) + + +def memory_stats_as_nested_dict(device: "Device" = None) -> dict[str, Any]: + r"""Return the result of :func:`~torch.cuda.memory_stats` as a nested dictionary.""" + if not is_initialized(): + return {} + device = _get_device_index(device, optional=True) + return torch._C._cuda_memoryStats(device) + + +def reset_accumulated_memory_stats(device: "Device" = None) -> None: + r"""Reset the "accumulated" (historical) stats tracked by the CUDA memory allocator. + + See :func:`~torch.cuda.memory_stats` for details. Accumulated stats correspond to + the `"allocated"` and `"freed"` keys in each individual stat dict, as well as + `"num_alloc_retries"` and `"num_ooms"`. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + device = _get_device_index(device, optional=True) + return torch._C._cuda_resetAccumulatedMemoryStats(device) + + +def reset_peak_memory_stats(device: "Device" = None) -> None: + r"""Reset the "peak" stats tracked by the CUDA memory allocator. + + See :func:`~torch.cuda.memory_stats` for details. Peak stats correspond to the + `"peak"` key in each individual stat dict. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + device = _get_device_index(device, optional=True) + return torch._C._cuda_resetPeakMemoryStats(device) + + +def host_memory_stats() -> dict[str, Any]: + r"""Return a dictionary of pinned (host) allocator statistics. + + Core statistics (host pinned allocator): + + - ``"allocations.{current,peak,allocated,freed}"``: + pinned blocks owned by the allocator (active + cached). Grows when a new + block is created via CUDA and shrinks when cached blocks are returned. + - ``"allocated_bytes.{current,peak,allocated,freed}"``: + bytes of pinned blocks owned by the allocator (active + cached), using + the rounded block size requested from CUDA. + - ``"active_requests.{current,peak,allocated,freed}"``: + blocks currently checked out to callers (increments on handout, decrements + when the block becomes reusable after stream deps finish). + - ``"active_bytes.{current,peak,allocated,freed}"``: + bytes corresponding to active blocks. + + Metric type: + + - ``current``: current value. + - ``peak``: maximum value. + - ``allocated``: historical total increase. + - ``freed``: historical total decrease. + + Event/timing counters: + + - ``"num_host_alloc"`` / ``"num_host_free"``: blocks created to grow the + pool / cached blocks returned to CUDA (matches allocations allocated/freed). + - ``"host_alloc_time.{total,max,min,count,avg}"``: time in CUDA alloc calls + when growing the pool (microseconds). + - ``"host_free_time.{total,max,min,count,avg}"``: time in CUDA free calls + when cached blocks are returned (microseconds). + + Block sizes are rounded up to the next power of two before calling CUDA, so + byte stats reflect the rounded size. Peak values are aggregated per bucket + and are a best-effort approximation of the true peak. + """ + result = [] + + stats = host_memory_stats_as_nested_dict() + _recurse_add_to_result(result, "", stats, str) + result.sort() + + return collections.OrderedDict(result) + + +def host_memory_stats_as_nested_dict() -> dict[str, Any]: + r"""Return the result of :func:`~torch.cuda.host_memory_stats` as a nested dictionary.""" + if not is_initialized(): + return {} + return torch._C._cuda_hostMemoryStats() + + +def reset_accumulated_host_memory_stats() -> None: + r"""Reset the "accumulated" (historical) stats tracked by the host memory allocator. + + See :func:`~torch.cuda.host_memory_stats` for details. Accumulated stats correspond to + the `"allocated"` and `"freed"` keys in each individual stat dict. + """ + return torch._C._cuda_resetAccumulatedHostMemoryStats() + + +def reset_peak_host_memory_stats() -> None: + r"""Reset the "peak" stats tracked by the host memory allocator. + + See :func:`~torch.cuda.host_memory_stats` for details. Peak stats correspond to the + `"peak"` key in each individual stat dict. + """ + return torch._C._cuda_resetPeakHostMemoryStats() + + +def reset_max_memory_allocated(device: "Device" = None) -> None: + r"""Reset the starting point in tracking maximum GPU memory occupied by tensors for a given device. + + See :func:`~torch.cuda.max_memory_allocated` for details. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. warning:: + This function now calls :func:`~torch.cuda.reset_peak_memory_stats`, which resets + /all/ peak memory stats. + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + warnings.warn( + "torch.cuda.reset_max_memory_allocated now calls torch.cuda.reset_peak_memory_stats, " + "which resets /all/ peak memory stats.", + FutureWarning, + stacklevel=2, + ) + return reset_peak_memory_stats(device=device) + + +def reset_max_memory_cached(device: "Device" = None) -> None: + r"""Reset the starting point in tracking maximum GPU memory managed by the caching allocator for a given device. + + See :func:`~torch.cuda.max_memory_cached` for details. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. warning:: + This function now calls :func:`~torch.cuda.reset_peak_memory_stats`, which resets + /all/ peak memory stats. + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + warnings.warn( + "torch.cuda.reset_max_memory_cached now calls torch.cuda.reset_peak_memory_stats, " + "which resets /all/ peak memory stats.", + FutureWarning, + stacklevel=2, + ) + return reset_peak_memory_stats(device=device) + + +def memory_allocated(device: "Device" = None) -> int: + r"""Return the current GPU memory occupied by tensors in bytes for a given device. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + This is likely less than the amount shown in `nvidia-smi` since some + unused memory can be held by the caching allocator and some context + needs to be created on GPU. See :ref:`cuda-memory-management` for more + details about GPU memory management. + """ + return memory_stats(device=device).get("allocated_bytes.all.current", 0) + + +def max_memory_allocated(device: "Device" = None) -> int: + r"""Return the maximum GPU memory occupied by tensors in bytes for a given device. + + By default, this returns the peak allocated memory since the beginning of + this program. :func:`~torch.cuda.reset_peak_memory_stats` can be used to + reset the starting point in tracking this metric. For example, these two + functions can measure the peak allocated memory usage of each iteration in a + training loop. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + return memory_stats(device=device).get("allocated_bytes.all.peak", 0) + + +def memory_reserved(device: "Device" = None) -> int: + r"""Return the current GPU memory managed by the caching allocator in bytes for a given device. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + return memory_stats(device=device).get("reserved_bytes.all.current", 0) + + +def max_memory_reserved(device: "Device" = None) -> int: + r"""Return the maximum GPU memory managed by the caching allocator in bytes for a given device. + + By default, this returns the peak cached memory since the beginning of this + program. :func:`~torch.cuda.reset_peak_memory_stats` can be used to reset + the starting point in tracking this metric. For example, these two functions + can measure the peak cached memory amount of each iteration in a training + loop. + + Args: + device (torch.device or int, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + + .. note:: + Under ``PYTORCH_CUDA_ALLOC_CONF=backend:cudaMallocAsync``, the peak is + computed by summing the high-water marks of the default mempool and the + device graph-memory pool (CUDA graph captures reserve backing in the + latter). Because those two high-water marks need not occur at the same + instant, the reported peak is a conservative *upper bound* on the true + simultaneous peak. The current value + (:func:`~torch.cuda.memory_reserved`) is exact. + """ + return memory_stats(device=device).get("reserved_bytes.all.peak", 0) + + +@deprecated( + "`torch.cuda.memory_cached` has been renamed to `torch.cuda.memory_reserved`", + category=FutureWarning, +) +def memory_cached(device: "Device" = None) -> int: + r"""Deprecated; see :func:`~torch.cuda.memory_reserved`.""" + return memory_reserved(device=device) + + +@deprecated( + "`torch.cuda.max_memory_cached` has been renamed to `torch.cuda.max_memory_reserved`", + category=FutureWarning, +) +def max_memory_cached(device: "Device" = None) -> int: + r"""Deprecated; see :func:`~torch.cuda.max_memory_reserved`.""" + return max_memory_reserved(device=device) + + +def memory_snapshot(mempool_id=None, include_traces=True): + r"""Return a snapshot of the CUDA memory allocator state across all devices. + + Interpreting the output of this function requires familiarity with the + memory allocator internals. + + Args: + mempool_id: Optional memory pool ID to get snapshot for a specific pool + include_traces: Whether to include trace entries in the snapshot. + If True (default), all trace entries are included. + If False, no trace entries are included (lightweight/fast snapshot). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + if mempool_id is None: + # pyrefly: ignore [bad-argument-type] + return torch._C._cuda_memorySnapshot((0, 0, include_traces))["segments"] + else: + return torch._C._cuda_memorySnapshot( + # pyrefly: ignore [bad-argument-type] + (mempool_id[0], mempool_id[1], include_traces) + )["segments"] + + +def memory_summary(device: "Device" = None, abbreviated: bool = False) -> str: + r"""Return a human-readable printout of the current memory allocator statistics for a given device. + + This can be useful to display periodically during training, or when + handling out-of-memory exceptions. + + Args: + device (torch.device or int, optional): selected device. Returns + printout for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + abbreviated (bool, optional): whether to return an abbreviated summary + (default: False). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + device = _get_device_index(device, optional=True) + stats = memory_stats(device=device) + + def _format_size(sz, pref_sz): + prefixes = ["B ", "KiB", "MiB", "GiB", "TiB", "PiB"] + prefix = prefixes[0] + for new_prefix in prefixes[1:]: + if pref_sz < 768 * 1024: + break + prefix = new_prefix + sz //= 1024 + pref_sz /= 1024 + return f"{sz:6d} {prefix}" + + def _format_count(cnt, pref_cnt): + prefixes = [" ", "K", "M"] + prefix = prefixes[0] + for new_prefix in prefixes[1:]: + if pref_cnt < 750 * 1000: + break + prefix = new_prefix + cnt //= 1000 + pref_cnt /= 1000 + return f"{cnt:7d} {prefix} " + + metrics_to_display = [ + ("allocated_bytes", "Allocated memory", _format_size), + ("active_bytes", "Active memory", _format_size), + ("requested_bytes", "Requested memory", _format_size), + ("reserved_bytes", "GPU reserved memory", _format_size), + ("inactive_split_bytes", "Non-releasable memory", _format_size), + ("allocation", "Allocations", _format_count), + ("active", "Active allocs", _format_count), + ("segment", "GPU reserved segments", _format_count), + ("inactive_split", "Non-releasable allocs", _format_count), + ] + + lines = [] + lines.append("=" * 75) + lines.append(" {_:16} PyTorch CUDA memory summary, device ID {device:<17d} ") + lines.append("-" * 75) + lines.append( + " {_:9} CUDA OOMs: {num_ooms:<12d} | {_:6} cudaMalloc retries: {num_alloc_retries:<8d} " + ) + lines.append("=" * 75) + lines.append( + " Metric | Cur Usage | Peak Usage | Tot Alloc | Tot Freed " + ) + + for metric_key, metric_name, formatter in metrics_to_display: + lines.append("-" * 75) + submetrics = [("all", metric_name)] + if not abbreviated: + submetrics.append(("large_pool", " from large pool")) + submetrics.append(("small_pool", " from small pool")) + + current_prefval, peak_prefval, allocated_prefval, freed_prefval = ( + None, + None, + None, + None, + ) + + for submetric_key, submetric_name in submetrics: + prefix = metric_key + "." + submetric_key + "." + + current = stats[prefix + "current"] + peak = stats[prefix + "peak"] + allocated = stats[prefix + "allocated"] + freed = stats[prefix + "freed"] + + if current_prefval is None: + current_prefval = current + peak_prefval = peak + allocated_prefval = allocated + freed_prefval = freed + + lines.append( + # pyrefly: ignore [bad-argument-type] + f" {submetric_name:<21} | {formatter(current, current_prefval)} | {formatter(peak, peak_prefval)} | " + f"{formatter(allocated, allocated_prefval)} | {formatter(freed, freed_prefval)} ", + ) + + metrics_to_display = [ + ("oversize_allocations", "Oversize allocations", _format_count), + ("oversize_segments", "Oversize GPU segments", _format_count), + ] + + for metric_key, metric_name, formatter in metrics_to_display: + lines.append("-" * 75) + + prefix = metric_key + "." + + current = stats[prefix + "current"] + peak = stats[prefix + "peak"] + allocated = stats[prefix + "allocated"] + freed = stats[prefix + "freed"] + + lines.append( + # pyrefly: ignore [bad-argument-type] + f" {metric_name:<21} | {formatter(current, current)} | {formatter(peak, peak)} | " + f"{formatter(allocated, allocated)} | {formatter(freed, freed)} ", + ) + + lines.append("=" * 75) + + fmt_dict = {"_": "", "device": device} + for k, v in stats.items(): + fmt_dict[k.replace(".", "-")] = v + return "|" + "|\n|".join(lines).format(**fmt_dict) + "|\n" + + +def list_gpu_processes(device: "Device" = None) -> str: + r"""Return a human-readable printout of the running processes and their GPU memory use for a given device. + + This can be useful to display periodically during training, or when + handling out-of-memory exceptions. + + Args: + device (torch.device or int, optional): selected device. Returns + printout for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + """ + if not torch.version.hip: + try: + import pynvml # type: ignore[import] + except ModuleNotFoundError: + return "pynvml module not found, please install nvidia-ml-py" + # pyrefly: ignore [import-error, missing-import, missing-module-attribute] + from pynvml import NVMLError_DriverNotLoaded + + try: + pynvml.nvmlInit() + except NVMLError_DriverNotLoaded: + return "cuda driver can't be loaded, is cuda enabled?" + + device = _get_nvml_device_index(device) + handle = pynvml.nvmlDeviceGetHandleByIndex(device) + procs = pynvml.nvmlDeviceGetComputeRunningProcesses(handle) + else: + try: + import amdsmi # type: ignore[import] + except ModuleNotFoundError: + return "amdsmi module not found, please install amdsmi" + try: + amdsmi.amdsmi_init() # type: ignore[attr-defined] + except amdsmi.AmdSmiException: # type: ignore[attr-defined] + return "amdsmi driver can't be loaded, is ROCm installed?" + + device = _get_amdsmi_device_index(device) + + try: + handle = amdsmi.amdsmi_get_processor_handles()[device] # type: ignore[attr-defined] + procs = amdsmi.amdsmi_get_gpu_process_list(handle) # type: ignore[attr-defined] + except amdsmi.AmdSmiException: # type: ignore[attr-defined] + return "amdsmi cannot list processes from other users" + + lines = [] + lines.append(f"GPU:{device}") + if len(procs) == 0: + lines.append("no processes are running") + for p in procs: + if not torch.version.hip: + mem = p.usedGpuMemory / (1024 * 1024) + pid = p.pid + else: + try: + proc_info = amdsmi.amdsmi_get_gpu_process_info(handle, p) # type: ignore[possibly-undefined] + except AttributeError: + # https://github.com/ROCm/amdsmi/commit/c551c3caedbd903ba828e7fdffa5b56d475a15e7 + # is a BC-breaking change that removes amdsmi_get_gpu_process_info API from amdsmi + proc_info = p + mem = proc_info["memory_usage"]["vram_mem"] / (1024 * 1024) + pid = proc_info["pid"] + lines.append(f"process {pid:>10d} uses {mem:>12.3f} MB GPU memory") + return "\n".join(lines) + + +def mem_get_info(device: "Device" = None) -> tuple[int, int]: + r"""Return the global free and total GPU memory for a given device using cudaMemGetInfo. + + Args: + device (torch.device or int or str, optional): selected device. Returns + statistic for the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default) or if the device index is not specified. + + .. note:: + See :ref:`cuda-memory-management` for more + details about GPU memory management. + """ + if device is None: + device = torch.cuda.current_device() + # optional=True allows `device = torch.device('cuda')` for which device.index is None + device = _get_device_index(device, optional=True) + return torch.cuda.cudart().cudaMemGetInfo(device) + + +def _record_memory_history_legacy( + enabled: bool, + record_context=True, + trace_alloc_max_entries=1, + trace_alloc_record_context=False, + device: "Device" = None, + record_context_cpp=False, + clear_history=False, + compile_context=False, + global_record_annotations=False, + skip_actions=None, + record_pinned_host_memory=False, + record_cuda=True, +): + _C._cuda_record_memory_history_legacy( # type: ignore[call-arg] + enabled, + record_context, + # pyrefly: ignore [bad-argument-type] + trace_alloc_max_entries, + trace_alloc_record_context, + record_context_cpp, + clear_history, + compile_context, + global_record_annotations, + # pyrefly: ignore [bad-argument-count] + skip_actions if skip_actions is not None else [], + record_pinned_host_memory, + record_cuda, + ) + + +def _record_memory_history( + enabled: Literal["state", "all"] | None = "all", *args, **kwargs +) -> None: + """Enable recording of stack traces associated with memory + allocations, so you can tell what allocated any piece of memory in + :func:`torch.cuda.memory._snapshot()`. + + In addition to keeping stack traces with each current allocation and free, + this will also enable recording of a history of all alloc/free events. + + Use :func:`torch.cuda.memory._snapshot()` to retrieve this information, + and the tools in `_memory_viz.py` to visualize snapshots. + + Buffer behavior + --------------- + + This will store up to `max_entries` instances of `TraceEntry` when enabled. + Python trace collection defaults to `sys.maxsize`, meaning long-running + or indefinitely running jobs should set a reasonable limit to avoid excessive + memory use. Expect each entry to be several KB. + + Longer running workflows or those with smaller `max_entries` values will only + store the last accumulated `max_entries` entries, meaning new entries overwrite + older entries. + + C++ implementation for reference to ring buffer implementation: + + .. code-block:: cpp + + if (record_history) { + if (alloc_trace->size() < alloc_trace_max_entries_) { + alloc_trace->emplace_back(te); + } else { + (*alloc_trace)[alloc_trace_next++] = te; + if (alloc_trace_next == alloc_trace_max_entries_) { + alloc_trace_next = 0; + } + } + } + + Latency impact + -------------- + + The Python trace collection is fast (2us per trace), so you may consider + enabling this on production jobs if you anticipate ever having to debug + memory issues. + + C++ trace collection is also fast (~50ns/frame), which for many typical programs + works out to ~2us per trace, but can vary depending on stack depth. + + Args: + enabled (Literal[None, "state", "all"], optional): + `None`, disable recording memory history. + `"state"`, keep information for currently allocated memory. + `"all"`, additionally keep a history of all alloc/free calls. + Defaults to "all". + context (Literal[None, "state", "alloc", "all"], optional): + `None`, Do not record any tracebacks. + `"state"`, Record tracebacks for currently allocated memory. + `"alloc"`, additionally keep tracebacks for alloc calls. + `"all"`, additionally keep tracebacks for free calls. + Defaults to "all". + stacks (Literal["python", "all"], optional): + `"python"`, include Python, TorchScript, and inductor frames in tracebacks + `"all"`, additionally include C++ frames + Defaults to "all". + max_entries (int, optional): Keep a maximum of `max_entries` + alloc/free events in the recorded history recorded. + clear_history (bool, optional): Clear history when enabling, defaults to False. + skip_actions (list[str], optional): List of action types to skip when recording + memory history. This can be used to reduce memory overhead by excluding + certain types of events from being recorded. Valid action types are: + + - `"alloc"`: Memory allocation events + - `"free_requested"`: Free requests (memory marked for freeing) + - `"free_completed"`: Completed free operations (memory actually freed) + - `"segment_alloc"`: Segment allocation from cudaMalloc + - `"segment_free"`: Segment freed back to CUDA via cudaFree + - `"oom"`: Out-of-memory exceptions + - `"snapshot"`: Memory snapshot generation events + + For example, to skip recording free_requested events: + `skip_actions=["free_requested"]` + + Defaults to None (record all actions). + record_pinned_host_memory (bool, optional): If True, also record memory history for + the CPU pinned memory (host) allocator. Host allocator traces will + appear in the ``host_segments`` and ``host_traces`` keys of the + snapshot returned by :func:`_snapshot`. Defaults to False. + record_cuda (bool, optional): If True, record memory history for the + CUDA device allocator. Set to False (together with + ``record_pinned_host_memory=True``) to record only host pinned memory. + Defaults to True. + + """ + if isinstance(enabled, bool): + return _record_memory_history_legacy(enabled, *args, **kwargs) + else: + return _record_memory_history_impl(enabled, *args, **kwargs) + + +def _record_memory_history_impl( + enabled: str | None = "all", + context: str | None = "all", + stacks: str = "all", + max_entries: int = sys.maxsize, + device: "Device" = None, + clear_history: bool = False, + compile_context: bool = False, + global_record_annotations: bool = False, + skip_actions: list[str] | None = None, + record_pinned_host_memory: bool = False, + record_cuda: bool = True, +): + _C._cuda_record_memory_history( # type: ignore[call-arg] + enabled, + context, + stacks, + max_entries, + clear_history, + compile_context, + global_record_annotations, + # pyrefly: ignore [bad-argument-count] + skip_actions if skip_actions is not None else [], + record_pinned_host_memory, + record_cuda, + ) + + +_record_memory_history.__signature__ = signature(_record_memory_history_impl) # type: ignore[attr-defined] + + +def _allocation_traceback(data_ptr: int) -> list[dict[str, Any]] | None: + r"""Return the allocation traceback for a currently-allocated CUDA pointer, + or ``None`` if the pointer is not found or recording was not enabled. + + ``data_ptr`` must be the base address of a recorded allocation, e.g. the + result of ``tensor.untyped_storage().data_ptr()``. Passing the result of + ``tensor.data_ptr()`` for a view with a nonzero ``storage_offset`` may + return ``None`` even if the underlying allocation is still live. + + Requires :func:`_record_memory_history` to have been called with + ``context != None`` beforehand. + """ + return torch._C._cuda_allocationTraceback(data_ptr) + + +def _snapshot(device: "Device" = None, augment_with_fx_traces=False): + """Save a snapshot of CUDA memory state at the time it was called. + + The state is represented as a dictionary with the following structure. + + .. code-block:: python + + class Snapshot(TypedDict): + segments: List[Segment] + device_traces: List[List[TraceEntry]] + + + class Segment(TypedDict): + # Segments are memory returned from a cudaMalloc call. + # The size of reserved memory is the sum of all Segments. + # Segments are cached and reused for future allocations. + # If the reuse is smaller than the segment, the segment + # is split into more than one Block. + # empty_cache() frees Segments that are entirely inactive. + address: int + total_size: int # cudaMalloc'd size of segment + stream: int + segment_type: Literal["small", "large"] # 'large' (>1MB) + segment_pool_id: Tuple[ + int, int + ] # id of the memory pool owning this segment + allocated_size: int # size of memory in use + active_size: int # size of memory in use or in active_awaiting_free state + blocks: List[Block] + + + class Block(TypedDict): + # A piece of memory returned from the allocator, or + # current cached but inactive. + size: int + requested_size: int # size requested during malloc, may be smaller than + # size due to rounding + address: int + state: Literal[ + "active_allocated", # used by a tensor + "active_awaiting_free", # waiting for another stream to finish using + # this, then it will become free + "inactive", + ] # free for reuse + frames: List[Frame] # stack trace from where the allocation occurred + + + class Frame(TypedDict): + filename: str + line: int + name: str + # Optional FX debug fields (present when augment_with_fx_traces=True + # and the frame corresponds to FX-generated code) + fx_node_op: str # FX node operation type (e.g., 'call_function', 'output') + fx_node_name: str # FX node name (e.g., 'linear', 'relu_1') + fx_original_trace: str # Original model source code stack trace + + + class TraceEntry(TypedDict): + # When `torch.cuda.memory._record_memory_history()` is enabled, + # the snapshot will contain TraceEntry objects that record each + # action the allocator took. + action: Literal[ + "alloc" # memory allocated + "free_requested", # the allocated received a call to free memory + "free_completed", # the memory that was requested to be freed is now + # able to be used in future allocation calls + "segment_alloc", # the caching allocator ask cudaMalloc for more memory + # and added it as a segment in its cache + "segment_free", # the caching allocator called cudaFree to return memory + # to cuda possibly trying free up memory to + # allocate more segments or because empty_caches was called + "oom", # the allocator threw an OOM exception. 'size' is + # the requested number of bytes that did not succeed + "snapshot", # the allocator generated a memory snapshot + # useful to correlate a previously taken + # snapshot with this trace + "annotate", # metadata was attached to a live allocation + # via _annotate_tensor. 'addr' is the allocation's base + # address and 'user_metadata' holds the annotation + ] + addr: int # not present for OOM + frames: List[Frame] + size: int + stream: int + device_free: int # only present for OOM, the amount of + # memory cuda still reports to be free + pool_id: Tuple[int, int] # id of the memory pool for this entry + + Args: + device: Device to capture snapshot for. If None, captures for current device. + augment_with_fx_traces: If True, augment stack trace frames with FX debug information + that maps generated FX code back to original model source code. + This adds fx_node_op, fx_node_name, fx_original_trace, and + fx_node_info fields to Frame objects. Default: False. + + Returns: + The Snapshot dictionary object + """ + s = _C._cuda_memorySnapshot(None) + if augment_with_fx_traces: + s = _augment_memory_snapshot_stack_traces(s) # type: ignore[assignment, arg-type] + return s + + +def _dump_snapshot(filename="dump_snapshot.pickle", augment_with_fx_traces=False): + """ + Save a pickled version of the `torch.memory._snapshot()` dictionary to a file. + + This file can be opened by the interactive snapshot viewer at pytorch.org/memory_viz + + Snapshot file sizes scale with `max_entries` and stack trace depth per entry, + with several KB per entry. These can easily be in the GB range for longer running + workflows with large `max_entries`. + + Args: + filename (str, optional): Name of the file to create. Defaults to "dump_snapshot.pickle". + augment_with_fx_traces (bool, optional): If True, augment the snapshot with FX debug information + before dumping. This maps generated FX code stack traces + back to original model source code. Defaults to False. + """ + s = _snapshot(augment_with_fx_traces=augment_with_fx_traces) + + with open(filename, "wb") as f: + pickle.dump(s, f) + + +def _memory_metadata_supported() -> bool: + """ + Return whether the active CUDA allocator backend records memory metadata. + + Only the native caching allocator supports :func:`_set_memory_metadata`. + With other backends (e.g. ``cudaMallocAsync`` or a pluggable allocator), + :func:`_set_memory_metadata` is a no-op and :func:`_get_memory_metadata` + always returns the empty string. Callers doing best-effort labeling can + check this before setting metadata to avoid the one-time warning emitted on + unsupported backends. + """ + # pyrefly: ignore [missing-attribute] + return torch._C._cuda_memoryMetadataSupported() + + +def _set_memory_metadata(metadata: str): + """ + Set custom metadata to be recorded on memory history trace entries. + + While memory history recording is enabled (see :func:`_record_memory_history`), + every allocator trace event generated by the calling thread carries the + current metadata string; it appears as the ``user_metadata`` key of trace + entries in the snapshot returned by :func:`_snapshot` and is displayed by + the pytorch.org/memory_viz visualizer. If memory history recording is not + enabled, this setting has no observable effect. The metadata does not + appear in the snapshot's segments/blocks view, in :func:`memory_stats`, or + on pinned host-memory allocations. + + The metadata is thread-local: allocations made by other threads (including + autograd's backward threads) are not affected. It applies to allocations + on all devices used by the calling thread. + + This is only supported by the native caching allocator; with other + backends (e.g. ``cudaMallocAsync`` or a pluggable allocator) it is + silently ignored. + + Args: + metadata (str): Custom metadata string to record on trace entries. + Pass an empty string to clear the metadata. + """ + # pyrefly: ignore [missing-attribute] + torch._C._cuda_setMemoryMetadata(metadata) + + +def _get_memory_metadata() -> str: + """ + Get the calling thread's current memory history metadata. + + See :func:`_set_memory_metadata`. Note that with allocator backends that + do not support metadata this always returns the empty string, which is + indistinguishable from no metadata being set. + + Returns: + str: The current metadata string, or empty string if no metadata is set. + """ + # pyrefly: ignore [missing-attribute] + return torch._C._cuda_getMemoryMetadata() + + +def _annotate_tensor(tensor: torch.Tensor, metadata: str): + """ + Attach metadata to the allocation backing ``tensor``, post facto. + + Unlike :func:`_set_memory_metadata`, which stamps metadata onto trace + events as they are generated, this records a dedicated ``annotate`` trace + event for an allocation that already exists. This is useful when the + information only becomes available after the allocation happened, e.g. + noting that a tensor was packed into the autograd graph. Metadata recorded + at allocation time is not modified; annotations accumulate as separate + trace entries keyed to the allocation's address, each carrying the + annotation string as its ``user_metadata`` and a stack trace of the + annotation site (subject to the ``context`` setting of + :func:`_record_memory_history`). The pytorch.org/memory_viz visualizer + shows annotations alongside the allocation's own metadata. + + The annotation is keyed to the base address of the tensor's untyped + storage, so it works for views and tensors with a nonzero + ``storage_offset``. + + If memory history recording is not enabled (see + :func:`_record_memory_history`), this has no observable effect. It is only + supported by the native caching allocator; with other backends (e.g. + ``cudaMallocAsync`` or a pluggable allocator) it is silently ignored. + + Args: + tensor (torch.Tensor): CUDA tensor whose backing allocation to + annotate. + metadata (str): Annotation string to record. + """ + if not tensor.is_cuda: + raise ValueError(f"expected a CUDA tensor, got device {tensor.device}") + # pyrefly: ignore [missing-attribute] + torch._C._cuda_annotateMemory(tensor.untyped_storage().data_ptr(), metadata) + + +def _save_segment_usage(filename="output.svg", snapshot=None): + if snapshot is None: + snapshot = _snapshot() + with open(filename, "w") as f: + f.write(_segments(snapshot)) + + +def _save_memory_usage(filename="output.svg", snapshot=None): + if snapshot is None: + snapshot = _snapshot() + with open(filename, "w") as f: + f.write(_memory(snapshot)) + + +@deprecated( + "torch.cuda._set_allocator_settings is deprecated. Use torch._C._accelerator_setAllocatorSettings instead.", + category=FutureWarning, +) +def _set_allocator_settings(env: str): + # pyrefly: ignore [missing-attribute] + return torch._C._accelerator_setAllocatorSettings(env) + + +def get_allocator_backend() -> str: + r"""Return a string describing the active allocator backend as set by + ``PYTORCH_ALLOC_CONF``. Currently available backends are + ``native`` (PyTorch's native caching allocator) and `cudaMallocAsync`` + (CUDA's built-in asynchronous allocator). + + .. note:: + See :ref:`cuda-memory-management` for details on choosing the allocator backend. + """ + return torch._C._cuda_getAllocatorBackend() + + +class _CUDAAllocator: + r"""Wrapper over internal CUDA memory allocators.""" + + def __init__(self, allocator: torch._C._cuda_CUDAAllocator): + self._allocator = allocator + + def allocator(self): + return self._allocator + + +class CUDAPluggableAllocator(_CUDAAllocator): + r"""CUDA memory allocator loaded from a so file.""" + + def __init__(self, path_to_so_file: str, alloc_fn_name: str, free_fn_name: str): + r"""Memory allocators are compiled in .so files and loaded dynamically using ctypes. + + To change the active allocator use the :func:`torch.memory.cuda.change_current_allocator` function. + + Args: + path_to_so_file(str): Path in the filesystem to the `.so` file containing + the allocator functions + alloc_fn_name(str): Name of the function to perform the memory allocation + in the so file. The signature must be: + void* alloc_fn_name(ssize_t size, int device, cudaStream_t stream); + free_fn_name(str): Name of the function to perform the memory release + in the so file. The signature must be: + void free_fn_name(void* ptr, size_t size, cudaStream_t stream); + + .. warning:: + This is currently supported only in unix OSs + + .. note:: + See :ref:`cuda-memory-management` for details on creating and using a custom allocator + """ + allocator = ctypes.CDLL(path_to_so_file) + alloc_fn = ctypes.cast(getattr(allocator, alloc_fn_name), ctypes.c_void_p).value + free_fn = ctypes.cast(getattr(allocator, free_fn_name), ctypes.c_void_p).value + if alloc_fn is None: + raise AssertionError(f"alloc_fn '{alloc_fn_name}' is None") + if free_fn is None: + raise AssertionError(f"free_fn '{free_fn_name}' is None") + self._allocator = torch._C._cuda_customAllocator(alloc_fn, free_fn) + + +def change_current_allocator(allocator: _CUDAAllocator) -> None: + r"""Change the currently used memory allocator to be the one provided. + + If the current allocator has already been used/initialized, this function will error. + + + Args: + allocator (torch.cuda.memory._CUDAAllocator): allocator to be set as the active one. + .. note:: + See :ref:`cuda-memory-management` for details on creating and using a custom allocator + """ + torch._C._cuda_changeCurrentAllocator(allocator.allocator()) + + +def _get_current_allocator() -> _CUDAAllocator: + r"""Return the allocator being currently used. + + .. note:: + See :ref:`cuda-memory-management` for details on creating and using a custom allocator + """ + return _CUDAAllocator(torch._C._cuda_getAllocator()) + + +class MemPool(_MemPool): + r"""MemPool represents a pool of memory in a caching allocator. Currently, + it's just the ID of the pool object maintained in the CUDACachingAllocator. + + Args: + allocator(torch._C._cuda_CUDAAllocator, optional): a + torch._C._cuda_CUDAAllocator object that can be used to + define how memory gets allocated in the pool. If :attr:`allocator` + is ``None`` (default), memory allocation follows the default/ + current configuration of the CUDACachingAllocator. + use_on_oom(bool): a bool that indicates if this pool can be used + as a last resort if a memory allocation outside of the pool fails due + to Out Of Memory. This is False by default. + no_split(bool): a bool that indicates if this pool should not split a segment. + This is False by default. + """ + + def __init__( + self, + allocator: _cuda_CUDAAllocator | None = None, + use_on_oom: bool = False, + no_split: bool = False, + ): + # pyrefly: ignore [bad-argument-count] + super().__init__(allocator, True, use_on_oom, no_split) + + @property + def id(self) -> tuple[int, int]: + r"""Returns the ID of this pool as a tuple of two ints.""" + return super().id + + def use_count(self) -> int: # pylint: disable=useless-parent-delegation + r"""Returns the reference count of this pool.""" + return super().use_count() + + def snapshot(self, include_traces=True): + r"""Return a snapshot of the CUDA memory allocator pool state across all + devices. + + Interpreting the output of this function requires familiarity with the + memory allocator internals. + + Args: + include_traces: Whether to include trace entries in the snapshot. + If True (default), all trace entries are included. + If False, no trace entries are included (lightweight/fast snapshot). + + .. note:: + See :ref:`cuda-memory-management` for more details about GPU memory + management. + """ + snapshot = torch.cuda.memory_snapshot(self.id, include_traces=include_traces) + return snapshot + + +@contextlib.contextmanager +def use_mem_pool(pool: MemPool, device: "Device" = None): + r"""A context manager that routes allocations to a given pool. + + Args: + pool(torch.cuda.MemPool): a MemPool object to be made active so that + allocations route to this pool. + device (torch.device or int, optional): selected device. Uses MemPool on + the current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + .. note:: + This context manager makes only current thread's allocations route to + the given pool. If a new thread is spawned inside the context manager + (e.g. by calling backward) the allocations in that thread will not + route to the given pool. + + .. note:: + When used during :class:`~torch.cuda.CUDAGraph` capture, the graph + retains the pool until the graph is reset or destroyed. + """ + device_index = ( + torch.cuda.current_device() if device is None else _get_device_index(device) + ) + _cuda_beginAllocateCurrentThreadToPool(device_index, pool.id) + try: + if torch.cuda.is_current_stream_capturing(): + graph = torch.cuda.CUDAGraph.get_currently_capturing_graph() + graph._retain_pool(pool) + yield + finally: + _cuda_endAllocateToPool(device_index, pool.id) + _cuda_releasePool(device_index, pool.id) + + +# Process-lifetime singleton: the UVM pluggable allocator + the ctypes closures +# behind it, lazily built (lock-guarded) by _make_uvm_allocator and reused. +_UVM_ALLOCATOR = None +_UVM_ALLOCATOR_LOCK = threading.Lock() + + +def _make_uvm_allocator(): + r"""Build the UVM ``CUDAPluggableAllocator`` and the ctypes closures behind it. + + Returns ``(c_alloc, c_free, allocator)``. UVM alloc/free are stateless (device + is passed per call), so one allocator serves all calls and devices. Raises + ``ImportError`` if ``cuda-python`` is unavailable. + """ + import logging + import traceback + + try: + from cuda.bindings import runtime as _rt # pyrefly: ignore[missing-import] + except ImportError: + raise ImportError( + "torch.cuda._use_uvm() requires the 'cuda-python' package " + "(cuda.bindings.runtime) for cudaMallocManaged, cudaMemAdvise, " + "and cudaFree." + ) from None + + log = logging.getLogger(__name__) + + _ALLOC_FN = ctypes.CFUNCTYPE( + ctypes.c_void_p, ctypes.c_size_t, ctypes.c_int, ctypes.c_void_p + ) + _FREE_FN = ctypes.CFUNCTYPE( + None, ctypes.c_void_p, ctypes.c_size_t, ctypes.c_int, ctypes.c_void_p + ) + + def _check(result, msg: str = ""): + err = result if not isinstance(result, tuple) else result[0] + if err != _rt.cudaError_t.cudaSuccess: + raise RuntimeError(f"CUDA error: {err}. {msg}") + + _advise_uses_struct = hasattr(_rt, "cudaMemLocation") + + def _mem_advise(ptr, size, advice, device_id, _runtime=_rt): + """Call cudaMemAdvise, handling struct-vs-int API difference. + + cuda-bindings 13.x requires a cudaMemLocation struct; + cuda-bindings 12.x expects a plain int device ordinal. + We try struct first and latch to int on TypeError. + """ + nonlocal _advise_uses_struct + if _advise_uses_struct: + try: + loc = _runtime.cudaMemLocation() + loc.type = _runtime.cudaMemLocationType.cudaMemLocationTypeDevice + loc.id = device_id + return _runtime.cudaMemAdvise(ptr, size, advice, loc) + except TypeError: + _advise_uses_struct = False + return _runtime.cudaMemAdvise(ptr, size, advice, device_id) + + _uvm_advise_supported_cache: dict[tuple[int, int], bool] = {} + + def _device_supports_uvm_advise(device_id, _runtime=_rt): + cache_key = (device_id, id(_runtime)) + if cache_key in _uvm_advise_supported_cache: + return _uvm_advise_supported_cache[cache_key] + attr = getattr( + _runtime.cudaDeviceAttr, "cudaDevAttrConcurrentManagedAccess", None + ) + if attr is None: + supported = False + else: + result = _runtime.cudaDeviceGetAttribute(attr, device_id) + err = result if not isinstance(result, tuple) else result[0] + if err != _runtime.cudaError_t.cudaSuccess: + supported = False + else: + supported = bool(result[1]) + _uvm_advise_supported_cache[cache_key] = supported + return supported + + def _uvm_alloc(size, device, stream, _runtime=_rt): + try: + err, ptr = _runtime.cudaMallocManaged(size, _runtime.cudaMemAttachGlobal) + _check(err, f"cudaMallocManaged({size})") + ptr = int(ptr) + if device >= 0 and _device_supports_uvm_advise(device, _runtime): + _check( + _mem_advise( + ptr, + size, + _runtime.cudaMemoryAdvise.cudaMemAdviseSetPreferredLocation, + device, + ), + "cudaMemAdvise(SetPreferredLocation)", + ) + _check( + _mem_advise( + ptr, + size, + _runtime.cudaMemoryAdvise.cudaMemAdviseSetAccessedBy, + device, + ), + "cudaMemAdvise(SetAccessedBy)", + ) + return ptr + except Exception: + log.error( + "[_use_uvm] FAILED to allocate %d bytes (%.2f GiB) via UVM." + " CUDACachingAllocator will raise an OOM error as a result." + " You can ignore free-memory numbers reported by PyTorch" + " as they are irrelevant for UVM.\nException:\n%s", + size, + size / (1024**3), + traceback.format_exc(), + ) + return 0 + + def _uvm_free(ptr, size, device, stream, _runtime=_rt): + """Best-effort free; guards against interpreter shutdown.""" + try: + if ptr: + _check(_runtime.cudaFree(ptr)) + except Exception: + if log is not None and traceback is not None: + try: + log.error( + "[_use_uvm] exception in free:\n%s", + traceback.format_exc(), + ) + except Exception: + pass + + c_alloc = _ALLOC_FN(_uvm_alloc) + c_free = _FREE_FN(_uvm_free) + alloc_ptr = ctypes.cast(c_alloc, ctypes.c_void_p).value + free_ptr = ctypes.cast(c_free, ctypes.c_void_p).value + # pyrefly: ignore[bad-argument-type] + allocator = torch._C._cuda_customAllocator(alloc_ptr, free_ptr) + return c_alloc, c_free, allocator + + +@contextlib.contextmanager +def _use_uvm(device: "Device" = None): + r"""A context manager that routes CUDA allocations through ``cudaMallocManaged`` (UVM). + + All tensors allocated inside this context use CUDA Unified Virtual Memory, + which allows oversubscribing GPU device memory by transparently paging to + system RAM on demand. Numerics are identical to regular device allocations; + only performance is affected due to page migration overhead. + + Args: + device (torch.device or int, optional): selected device. Uses the + current device, given by :func:`~torch.cuda.current_device`, + if :attr:`device` is ``None`` (default). + + Example:: + + >>> # xdoctest: +SKIP(reason="requires CUDA and cuda-python") + >>> with torch.cuda._use_uvm(): + ... x = torch.randn(1024, 1024, device="cuda") + ... y = x @ x.T # computed on GPU, pages in/out as needed + + .. note:: + Only the current thread's allocations are routed to managed memory. + Allocations in threads spawned inside this context (e.g. by backward) + will use the default allocator. Use + ``torch.autograd.set_multithreading_enabled(False)`` to force backward + onto the calling thread so that backward allocations also use UVM. + + .. note:: + Requires the ``cuda-python`` package (``cuda.bindings``). + """ + # A tensor's block can stay cached in the PrivatePool after _use_uvm() returns + # and be freed later by a global empty_cache(); per-call ctypes closures would + # be GC'd by then, dangling the free callback. Build once and reuse instead. + global _UVM_ALLOCATOR + if _UVM_ALLOCATOR is None: + with _UVM_ALLOCATOR_LOCK: + if _UVM_ALLOCATOR is None: + _UVM_ALLOCATOR = _make_uvm_allocator() + allocator = _UVM_ALLOCATOR[2] + + pool = MemPool(allocator=allocator) + with use_mem_pool(pool, device=device): + yield pool diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/nccl.py b/.venv/lib/python3.12/site-packages/torch/cuda/nccl.py new file mode 100644 index 0000000000000000000000000000000000000000..f850f8ad1f58d814fb236f4f2ce53e6541c2f1ae --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/nccl.py @@ -0,0 +1,151 @@ +# mypy: allow-untyped-defs +import collections +import warnings +from collections.abc import Sequence + +import torch.cuda + + +__all__ = ["all_reduce", "reduce", "broadcast", "all_gather", "reduce_scatter"] + +SUM = 0 # ncclRedOp_t + + +def is_available(tensors): + if not hasattr(torch._C, "_nccl_all_reduce"): + warnings.warn("PyTorch is not compiled with NCCL support", stacklevel=2) + return False + + devices = set() + for tensor in tensors: + if tensor.is_sparse: + return False + if not tensor.is_contiguous(): + return False + if not tensor.is_cuda: + return False + device = tensor.get_device() + if device in devices: + return False + devices.add(device) + + return True + + +def version(): + """ + Returns the version of the NCCL. + + + This function returns a tuple containing the major, minor, and patch version numbers of the NCCL. + The suffix is also included in the tuple if a version suffix exists. + Returns: + tuple: The version information of the NCCL. + """ + ver = torch._C._nccl_version() + major = ver >> 32 + minor = (ver >> 16) & 65535 + patch = ver & 65535 + suffix = torch._C._nccl_version_suffix().decode("utf-8") + if suffix == "": + return (major, minor, patch) + else: + return (major, minor, patch, suffix) + + +def unique_id(): + return torch._C._nccl_unique_id() + + +def init_rank(num_ranks, uid, rank): + return torch._C._nccl_init_rank(num_ranks, uid, rank) + + +def _check_sequence_type(inputs: torch.Tensor | Sequence[torch.Tensor]) -> None: + if not isinstance(inputs, collections.abc.Container) or isinstance( + inputs, torch.Tensor + ): + raise TypeError("Inputs should be a collection of tensors") + + +def all_reduce(inputs, outputs=None, op=SUM, streams=None, comms=None): + _check_sequence_type(inputs) + if outputs is None: + outputs = inputs + _check_sequence_type(outputs) + torch._C._nccl_all_reduce(inputs, outputs, op, streams, comms) + + +# `output` used to be `outputs`, taking in a list of tensors. So we have two +# arguments for BC reasons. +def reduce( + inputs: Sequence[torch.Tensor], + output: torch.Tensor | Sequence[torch.Tensor] | None = None, + root: int = 0, + op: int = SUM, + streams: Sequence[torch.cuda.Stream] | None = None, + comms=None, + *, + outputs: Sequence[torch.Tensor] | None = None, +) -> None: + _check_sequence_type(inputs) + _output: torch.Tensor + if outputs is not None: + if output is not None: + raise ValueError( + "'output' and 'outputs' can not be both specified. 'outputs' is deprecated in " + "favor of 'output', taking in a single output tensor. The signature of reduce is: " + "reduce(inputs, output=None, root=0, op=SUM, streams=None, comms=None)." + ) + else: + warnings.warn( + "`nccl.reduce` with an output tensor list is deprecated. " + "Please specify a single output tensor with argument 'output' instead.", + FutureWarning, + stacklevel=2, + ) + _output = outputs[root] + elif not isinstance(output, torch.Tensor) and isinstance( + output, collections.abc.Sequence + ): + # User called old API with positional arguments of list of output tensors. + warnings.warn( + "nccl.reduce with an output tensor list is deprecated. " + "Please specify a single output tensor.", + FutureWarning, + stacklevel=2, + ) + _output = output[root] + else: + _output = inputs[root] if output is None else output + torch._C._nccl_reduce(inputs, _output, root, op, streams, comms) + + +def broadcast( + inputs: Sequence[torch.Tensor], root: int = 0, streams=None, comms=None +) -> None: + _check_sequence_type(inputs) + torch._C._nccl_broadcast(inputs, root, streams, comms) + + +def all_gather( + inputs: Sequence[torch.Tensor], + outputs: Sequence[torch.Tensor], + streams=None, + comms=None, +) -> None: + _check_sequence_type(inputs) + _check_sequence_type(outputs) + torch._C._nccl_all_gather(inputs, outputs, streams, comms) + + +def reduce_scatter( + inputs: Sequence[torch.Tensor], + outputs: Sequence[torch.Tensor], + op: int = SUM, + streams=None, + comms=None, +) -> None: + _check_sequence_type(inputs) + _check_sequence_type(outputs) + torch._C._nccl_reduce_scatter(inputs, outputs, op, streams, comms) diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/nvtx.py b/.venv/lib/python3.12/site-packages/torch/cuda/nvtx.py new file mode 100644 index 0000000000000000000000000000000000000000..ddb5fed3e16c06240bf6cf5b28f76ee49253e134 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/nvtx.py @@ -0,0 +1,129 @@ +# mypy: allow-untyped-defs +r"""This package adds support for NVIDIA Tools Extension (NVTX) used in profiling.""" + +from contextlib import contextmanager + + +try: + from torch._C import _nvtx +except ImportError: + + class _NVTXStub: + @staticmethod + def _fail(*args, **kwargs): + raise RuntimeError( + "NVTX functions not installed. Are you sure you have a CUDA build?" + ) + + rangePushA = _fail + rangePop = _fail + markA = _fail + + _nvtx = _NVTXStub() # type: ignore[assignment] + +__all__ = ["range_push", "range_pop", "range_start", "range_end", "mark", "range"] + + +def range_push(msg): + """ + Push a range onto a stack of nested range span. Returns zero-based depth of the range that is started. + + Args: + msg (str): ASCII message to associate with range + """ + return _nvtx.rangePushA(msg) + + +def range_pop(): + """Pop a range off of a stack of nested range spans. Returns the zero-based depth of the range that is ended.""" + return _nvtx.rangePop() + + +def range_start(msg) -> int: + """ + Mark the start of a range with string message. It returns a unique handle + for this range to pass to the corresponding call to rangeEnd(). + + A key difference between this and range_push/range_pop is that the + range_start/range_end version supports range across threads (start on one + thread and end on another thread). + + Returns: A range handle (uint64_t) that can be passed to range_end(). + + Args: + msg (str): ASCII message to associate with the range. + """ + # pyrefly: ignore [missing-attribute] + return _nvtx.rangeStartA(msg) + + +def range_end(range_id) -> None: + """ + Mark the end of a range for a given range_id. + + Args: + range_id (int): a unique handle for the start range. + """ + # pyrefly: ignore [missing-attribute] + _nvtx.rangeEnd(range_id) + + +def _device_range_start(msg: str, stream: int = 0) -> object: + """ + Marks the start of a range with string message. + It returns an opaque heap-allocated handle for this range + to pass to the corresponding call to device_range_end(). + + A key difference between this and range_start is that the + range_start marks the range right away, while _device_range_start + marks the start of the range as soon as all the tasks on the + CUDA stream are completed. + + Returns: An opaque heap-allocated handle that should be passed to _device_range_end(). + + Args: + msg (str): ASCII message to associate with the range. + stream (int): CUDA stream id. + """ + # pyrefly: ignore [missing-attribute] + return _nvtx.deviceRangeStart(msg, stream) + + +def _device_range_end(range_handle: object, stream: int = 0) -> None: + """ + Mark the end of a range for a given range_handle as soon as all the tasks + on the CUDA stream are completed. + + Args: + range_handle: a unique handle for the start range. + stream (int): CUDA stream id. + """ + # pyrefly: ignore [missing-attribute] + _nvtx.deviceRangeEnd(range_handle, stream) + + +def mark(msg): + """ + Describe an instantaneous event that occurred at some point. + + Args: + msg (str): ASCII message to associate with the event. + """ + return _nvtx.markA(msg) + + +@contextmanager +def range(msg, *args, **kwargs): + """ + Context manager / decorator that pushes an NVTX range at the beginning + of its scope, and pops it at the end. If extra arguments are given, + they are passed as arguments to msg.format(). + + Args: + msg (str): message to associate with the range + """ + range_push(msg.format(*args, **kwargs)) + try: + yield + finally: + range_pop() diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/profiler.py b/.venv/lib/python3.12/site-packages/torch/cuda/profiler.py new file mode 100644 index 0000000000000000000000000000000000000000..01e728e6feb5e6f433dc85540022adb572046d97 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/profiler.py @@ -0,0 +1,56 @@ +# mypy: allow-untyped-defs +import contextlib + +from . import check_error, cudart + + +__all__ = ["start", "stop", "profile"] + +DEFAULT_FLAGS = [ + "gpustarttimestamp", + "gpuendtimestamp", + "gridsize3d", + "threadblocksize", + "streamid", + "enableonstart 0", + "conckerneltrace", +] + + +def start(): + r"""Starts cuda profiler data collection. + + .. warning:: + Raises CudaError in case of it is unable to start the profiler. + """ + check_error(cudart().cudaProfilerStart()) + + +def stop(): + r"""Stops cuda profiler data collection. + + .. warning:: + Raises CudaError in case of it is unable to stop the profiler. + """ + check_error(cudart().cudaProfilerStop()) + + +@contextlib.contextmanager +def profile(): + """ + Enable profiling. + + Context Manager to enabling profile collection by the active profiling tool from CUDA backend. + Example: + >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA) + >>> import torch + >>> model = torch.nn.Linear(20, 30).cuda() + >>> inputs = torch.randn(128, 20).cuda() + >>> with torch.cuda.profiler.profile() as prof: + ... model(inputs) + """ + try: + start() + yield + finally: + stop() diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/random.py b/.venv/lib/python3.12/site-packages/torch/cuda/random.py new file mode 100644 index 0000000000000000000000000000000000000000..e597884cb214dff9cb40538a9929af38b596f521 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/random.py @@ -0,0 +1,181 @@ +# mypy: allow-untyped-defs +from collections.abc import Iterable + +import torch +from torch import Tensor + +from . import _lazy_call, _lazy_init, current_device, device_count, is_initialized + + +__all__ = [ + "get_rng_state", + "get_rng_state_all", + "set_rng_state", + "set_rng_state_all", + "manual_seed", + "manual_seed_all", + "seed", + "seed_all", + "initial_seed", +] + + +def get_rng_state(device: int | str | torch.device = "cuda") -> Tensor: + r"""Return the random number generator state of the specified GPU as a ByteTensor. + + Args: + device (torch.device or int, optional): The device to return the RNG state of. + Default: ``'cuda'`` (i.e., ``torch.device('cuda')``, the current CUDA device). + + .. warning:: + This function eagerly initializes CUDA. + """ + _lazy_init() + if isinstance(device, str): + device = torch.device(device) + elif isinstance(device, int): + device = torch.device("cuda", device) + idx = device.index + if idx is None: + idx = current_device() + default_generator = torch.cuda.default_generators[idx] + return default_generator.get_state() + + +def get_rng_state_all() -> list[Tensor]: + r"""Return a list of ByteTensor representing the random number states of all devices.""" + results = [get_rng_state(i) for i in range(device_count())] + return results + + +def set_rng_state(new_state: Tensor, device: int | str | torch.device = "cuda") -> None: + r"""Set the random number generator state of the specified GPU. + + Args: + new_state (torch.ByteTensor): The desired state + device (torch.device or int, optional): The device to set the RNG state. + Default: ``'cuda'`` (i.e., ``torch.device('cuda')``, the current CUDA device). + """ + if not is_initialized(): + with torch._C._DisableFuncTorch(): + # Clone the state because the callback will be triggered + # later when CUDA is lazy initialized. + new_state = new_state.clone(memory_format=torch.contiguous_format) + if isinstance(device, str): + device = torch.device(device) + elif isinstance(device, int): + device = torch.device("cuda", device) + + def cb(): + idx = device.index + if idx is None: + idx = current_device() + default_generator = torch.cuda.default_generators[idx] + default_generator.set_state(new_state) + + _lazy_call(cb) + + +def set_rng_state_all(new_states: Iterable[Tensor]) -> None: + r"""Set the random number generator state of all devices. + + Args: + new_states (Iterable of torch.ByteTensor): The desired state for each device. + """ + for i, state in enumerate(new_states): + set_rng_state(state, i) + + +def manual_seed(seed: int) -> None: + r"""Set the seed for generating random numbers for the current GPU. + + It's safe to call this function if CUDA is not available; in that + case, it is silently ignored. + + Args: + seed (int): The desired seed. + + .. warning:: + If you are working with a multi-GPU model, this function is insufficient + to get determinism. To seed all GPUs, use :func:`manual_seed_all`. + """ + seed = int(seed) + + def cb(): + idx = current_device() + default_generator = torch.cuda.default_generators[idx] + default_generator.manual_seed(seed) + + _lazy_call(cb, seed=True) + + +def manual_seed_all(seed: int) -> None: + r"""Set the seed for generating random numbers on all GPUs. + + It's safe to call this function if CUDA is not available; in that + case, it is silently ignored. + + Args: + seed (int): The desired seed. + """ + seed = int(seed) + + def cb(): + for i in range(device_count()): + default_generator = torch.cuda.default_generators[i] + default_generator.manual_seed(seed) + + _lazy_call(cb, seed_all=True) + + +def seed() -> None: + r"""Set the seed for generating random numbers to a random number for the current GPU. + + It's safe to call this function if CUDA is not available; in that + case, it is silently ignored. + + .. warning:: + If you are working with a multi-GPU model, this function will only initialize + the seed on one GPU. To initialize all GPUs, use :func:`seed_all`. + """ + + def cb(): + idx = current_device() + default_generator = torch.cuda.default_generators[idx] + default_generator.seed() + + _lazy_call(cb) + + +def seed_all() -> None: + r"""Set the seed for generating random numbers to a random number on all GPUs. + + It's safe to call this function if CUDA is not available; in that + case, it is silently ignored. + """ + + def cb(): + random_seed = 0 + seeded = False + for i in range(device_count()): + default_generator = torch.cuda.default_generators[i] + if not seeded: + default_generator.seed() + random_seed = default_generator.initial_seed() + seeded = True + else: + default_generator.manual_seed(random_seed) + + _lazy_call(cb) + + +def initial_seed() -> int: + r"""Return the current random seed of the current GPU. + + .. warning:: + This function eagerly initializes CUDA. + """ + _lazy_init() + idx = current_device() + default_generator = torch.cuda.default_generators[idx] + return default_generator.initial_seed() diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/sparse.py b/.venv/lib/python3.12/site-packages/torch/cuda/sparse.py new file mode 100644 index 0000000000000000000000000000000000000000..f37a34118d2d8f73437dee54337a666df1b99a09 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/sparse.py @@ -0,0 +1 @@ +# The Tensor classes are added to this module by python_tensor.cpp diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/streams.py b/.venv/lib/python3.12/site-packages/torch/cuda/streams.py new file mode 100644 index 0000000000000000000000000000000000000000..77981e465257aa85ca141c8b9aa254768f792bcd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/streams.py @@ -0,0 +1,271 @@ +# mypy: allow-untyped-defs +# pylint: disable=useless-parent-delegation +from __future__ import annotations + +import ctypes + +import torch +from torch._utils import _dummy_type + + +if not hasattr(torch._C, "_CudaStreamBase"): + # Define dummy base classes + torch._C.__dict__["_CudaStreamBase"] = _dummy_type("_CudaStreamBase") + torch._C.__dict__["_CudaEventBase"] = _dummy_type("_CudaEventBase") + + +class Stream(torch._C._CudaStreamBase): + r"""Wrapper around a CUDA stream. + + A CUDA stream is a linear sequence of execution that belongs to a specific + device, independent from other streams. It supports with statement as a + context manager to ensure the operators within the with block are running + on the corresponding stream. See :ref:`cuda-semantics` for details. + + Args: + device(torch.device or int, optional): a device on which to allocate + the stream. If :attr:`device` is ``None`` (default) or a negative + integer, this will use the current device. + priority(int, optional): priority of the stream, which can be positive, 0, or negative. + A lower number indicates a higher priority. By default, the priority is set to 0. + If the value falls outside of the allowed priority range, it will automatically be + mapped to the nearest valid priority (lowest for large positive numbers or + highest for large negative numbers). + + """ + + def __new__(cls, device=None, priority=0, **kwargs): + # Check CUDA availability + if not torch.backends.cuda.is_built(): + raise RuntimeError("torch.cuda.Stream requires CUDA support") + # setting device manager is expensive, so we avoid it unless necessary + if device is None or ("stream_id" in kwargs and "device_index" in kwargs): + return super().__new__(cls, priority=priority, **kwargs) + else: + with torch.cuda.device(device): + return super().__new__(cls, priority=priority, **kwargs) + + def wait_event(self, event: Event | torch.Event) -> None: + r"""Make all future work submitted to the stream wait for an event. + + Args: + event (Event, torch.Event): an event to wait for. + + .. note:: This is a wrapper around ``cudaStreamWaitEvent()``: see + `CUDA Stream documentation`_ for more info. + + This function returns without waiting for :attr:`event`: only future + operations are affected. + + .. _CUDA Stream documentation: + https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html + """ + event.wait(self) + + def wait_stream(self, stream: Stream | torch.Stream) -> None: + r"""Synchronize with another stream. + + All future work submitted to this stream will wait until all kernels + submitted to a given stream at the time of call complete. + + Args: + stream (Stream, torch.Stream): a stream to synchronize. + + .. note:: This function returns without waiting for currently enqueued + kernels in :attr:`stream`: only future operations are affected. + """ + self.wait_event(stream.record_event()) + + def record_event(self, event: Event | torch.Event | None = None): + r"""Record an event. + + Args: + event (Event, torch.Event, optional): event to record. If not given, a new one + will be allocated. + + Returns: + Recorded event. + """ + if event is None: + event = Event() + event.record(self) + return event + + def query(self) -> bool: + r"""Check if all the work submitted has been completed. + + Returns: + A boolean indicating if all kernels in this stream are completed. + """ + return super().query() + + def synchronize(self) -> None: + r"""Wait for all the kernels in this stream to complete. + + .. note:: This is a wrapper around ``cudaStreamSynchronize()``: see + `CUDA Stream documentation`_ for more info. + """ + super().synchronize() + + @property + def _as_parameter_(self): + return ctypes.c_void_p(self.cuda_stream) + + def __eq__(self, o) -> bool: + if isinstance(o, Stream): + return super().__eq__(o) + return False + + def __hash__(self): + return hash((self.cuda_stream, self.device)) + + def __repr__(self): + return f"" + + def __cuda_stream__(self): + """Implements the CUDA Stream Protocol: + https://nvidia.github.io/cuda-python/cuda-core/latest/interoperability.html#cuda-stream-protocol + + Returns: + tuple: A 2-tuple of (version, handle) where version is the protocol version + and handle is the address of cudaStream_t (CUDA) or hipStream_t (ROCm) as a Python int. + """ + return (0, self.cuda_stream) + + +class ExternalStream(Stream): + r"""Wrapper around an externally allocated CUDA stream. + + This class is used to wrap streams allocated in other libraries in order + to facilitate data exchange and multi-library interactions. + + .. note:: This class doesn't manage the stream life-cycle, it is the user + responsibility to keep the referenced stream alive while this class is + being used. + + Args: + stream_ptr(int): Integer representation of the `cudaStream_t` value. + allocated externally. + device(torch.device or int, optional): the device where the stream + was originally allocated. If device is specified incorrectly, + subsequent launches using this stream may fail. + """ + + def __new__(cls, stream_ptr, device=None, **kwargs): + with torch.cuda.device(device): + return super().__new__(cls, stream_ptr=stream_ptr, **kwargs) + + +class Event(torch._C._CudaEventBase): + r"""Wrapper around a CUDA event. + + CUDA events are synchronization markers that can be used to monitor the + device's progress, to accurately measure timing, and to synchronize CUDA + streams. + + The underlying CUDA events are lazily initialized when the event is first + recorded or exported to another process. After creation, only streams on the + same device may record the event. However, streams on any device can wait on + the event. + + Args: + enable_timing (bool, optional): indicates if the event should measure time + (default: ``False``) + blocking (bool, optional): if ``True``, :meth:`wait` will be blocking (default: ``False``) + interprocess (bool): if ``True``, the event can be shared between processes + (default: ``False``) + external (bool, optional): indicates whether this event should create event record and event wait nodes, or create an internal cross-stream dependency, when captured in a cuda graph. See `cross-stream dependencies `_, `cudaEventRecordExternal `_, and `cudaEventWaitExternal `_ for more information about internal vs. external events. (default: ``False``) + + .. _CUDA Event Documentation: + https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__EVENT.html + """ + + def __new__( + cls, enable_timing=False, blocking=False, interprocess=False, external=False + ): + return super().__new__( + cls, + enable_timing=enable_timing, + blocking=blocking, + interprocess=interprocess, + external=external, + ) + + @classmethod + def from_ipc_handle(cls, device, handle): + r"""Reconstruct an event from an IPC handle on the given device.""" + return super().from_ipc_handle(device, handle) + + def record(self, stream: Stream | torch.Stream | None = None): + r"""Record the event in a given stream. + + Args: + stream (Stream, torch.Stream, optional): Uses ``torch.cuda.current_stream()`` if no stream is specified. + The stream's device must match the event's device. + """ + if stream is None: + stream = torch.cuda.current_stream() + # pyrefly: ignore [bad-argument-type] + super().record(stream) + + def wait(self, stream: Stream | torch.Stream | None = None) -> None: + r"""Make all future work submitted to the given stream wait for this event. + + Args: + stream (Stream, torch.Stream, optional): Uses ``torch.cuda.current_stream()`` if no stream is specified. + + .. note:: This is a wrapper around ``cudaStreamWaitEvent()``: see + `CUDA Event documentation`_ for more info. + """ + if stream is None: + stream = torch.cuda.current_stream() + # pyrefly: ignore [bad-argument-type] + super().wait(stream) + + def query(self): + r"""Check if all work currently captured by event has completed. + + Returns: + A boolean indicating if all work currently captured by event has + completed. + """ + return super().query() + + def elapsed_time(self, end_event: Event): + r"""Return the time elapsed. + + Time reported in milliseconds after the event was recorded and + before the end_event was recorded. + + Args: + end_event (Event): the end event. + """ + return super().elapsed_time(end_event) + + def synchronize(self) -> None: + r"""Wait for the event to complete. + + Waits until the completion of all work currently captured in this event. + This prevents the CPU thread from proceeding until the event completes. + + .. note:: This is a wrapper around ``cudaEventSynchronize()``: see + `CUDA Event documentation`_ for more info. + """ + super().synchronize() + + def ipc_handle(self): + r"""Return an IPC handle of this event. + + If not recorded yet, the event will use the current device. + """ + return super().ipc_handle() + + @property + def _as_parameter_(self): + return ctypes.c_void_p(self.cuda_event) + + def __repr__(self) -> str: + if self.cuda_event: + return f"" + else: + return "" diff --git a/.venv/lib/python3.12/site-packages/torch/cuda/tunable.py b/.venv/lib/python3.12/site-packages/torch/cuda/tunable.py new file mode 100644 index 0000000000000000000000000000000000000000..f7a267fbc1d278212ef173a139ae3bc9a90f26f2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/cuda/tunable.py @@ -0,0 +1,881 @@ +r""" +This module exposes a TunableOp interface. + +Some operations, such as GEMMs, could be implemented using more than one library +or more than one technique. For example, a GEMM could be implemented for CUDA or +ROCm using either the blas or blasLt libraries. Further, ROCm's rocblas and +hipblaslt libraries allow the user to query for all possible algorithms and then +choose one. How does one know which implementation is the fastest and should be +chosen? That's what TunableOp provides. + +Enabling TunableOp and Tuning Separately +======================================== + +The TunableOp feature is enabled separately from enabling the tuning phase +itself. Enabling TunableOp means that PyTorch will replace any standard +operators with their Tunable implementations. Any call to a TunableOp first +checks whether it has already been tuned for the given operator inputs. If so, +it will immediately call the tuned operation; no further tuning will take place +even when the tuning setting is enabled. Instead if no tuning result is found, +and tuning is enabled, the TunableOp will benchmark every registered +implementation of that operator for the given set of inputs and select the +fastest. + +File Input and Output +===================== + +The first time any TunableOp is invoked, the internal database of tuned +operations will be prepared by attempting to read the results from the given +file. The default filename is 'tunableop_results.csv'. To support tuning when +multiple GPUs are used across multiple processes, the GPU device ordinal is +automatically inserted into the filename to avoid multiple processes overwriting +the same file. + +If tuning is enabled and new tunings are discovered during the course of your +workload, it will also write out to this same filename with all tunings, both +the ones it read in at startup as well as the new ones found at runtime. This +can be used, for example, to build up a tunings file across many workloads by +reusing the same file. The output file is automatically created when the +application terminates. This behavior can be controlled by the C++ and Python +APIs but not the environment variables. + +Assuming you specified a filename, you'll end up with a CSV file with contents +like so:: + + Validator,PT_VERSION,2.2.0 + Validator,ROCM_VERSION,6.0.0.0-12969-1544e39 + Validator,HIPBLASLT_VERSION,0.6.0-a9c5cc7 + Validator,ROCBLAS_VERSION,4.0.0-72e57364-dirty + GemmTunableOp_float_NT,nt_25088_4096_64,Gemm_Hipblaslt_1219,1.262 + GemmTunableOp_float_NT,nt_4096_4096_64,Gemm_Rocblas_1216,0.033 + +Note the "Validator" lines. If you change a library version, or ROCm version, or +PyTorch version, TunableOp will detect this and reject the tunings file because +the prior tunings are likely affected by other software changes. + +The remaining lines are the tuned solutions for each TunableOp encountered +during your execution. Each line consists of 4 comma-separated fields: operator +name, operator parameters, solution name, and average execution time. The +execution time is an optional field. The CSV file can be edited, but with +caution. For example, the solution name (field 3) can be changed to "Default" +and it will fall back to the original PyTorch untuned implementation. Or, in the +case of ROCm's hipBLAS or hipBLASLt libraries, if you know the specific solution +index you can override the solution that TunableOp selected by replacing the +value. The operator name and parameters (fields 1 and 2) are internally named +and should not be modified. In the case of GemmTunableOp, field 1 indicates the +datatype and whether the inputs are transposed (T) or not (N) and field 2 +indicates the M, N, K input shapes. + +There is an option to enable verbose output but it is only recommended for +debugging purposes. This will produce a lot of diagnostic messages but may be +useful to see if TunableOp is being used at all. Otherwise, TunableOp is +completely silent, besides file output, unless there is a warning or error +during its use. The verbose option is only available by setting the environment +variable PYTORCH_TUNABLEOP_VERBOSE=1. + +A Note on Tuning Behavior, Warmup, and Cache Effects +==================================================== + +Tuning an operator consists of iterating through the list or registered +implementations and profiling each one. The profile is established by running a +single implementation in a loop multiple times and taking the average execution +time. There is also an optional warmup phase prior to tuning that can help with +reaching stable power states by the hardware. During tuning of a workload the +various hardware caches will more likely produce hits than when not tuning. +There are options for flushing the instruction cache and rotate the input tensors +which might help produce a more faithful profile of the tuned operator as if the +operator were run within a larger workload instead of in a tight, repetitive loop. + +By default, each possible solution for a given operator will be run for either +100 iterations or as many iterations that can be run within 30ms, whichever is +smaller, and its average execution will be calculated. The fastest solution +among all that were successfully profiled will be chosen. A profile might fail +if the given solution doesn't achieve the same accuracy as the default +implementation or if the solution returns an error code. + +CUDA cuBLASLt support uses the TunableOp result cache and profiling machinery +to time a configurable number of cuBLASLt heuristic candidates. + +Current Tunable Operators +========================= + +TunableGemm for ROCm +-------------------- + +Any call to at::cuda::blas::gemm() or ::bgemm() will be routed through TunableOp +when enabled. Calling gemm() for a given set of input arguments +(transa, transb, m, n, k) on ROCm will attempt to use the fastest available +implementation across both rocblas and hipblaslt. On CUDA, TunableGemm registers +cuBLASLt heuristic candidates for GEMM paths that already use cuBLASLt. + +cuBLASLt Heuristic Tuning for CUDA +---------------------------------- + +The number of cuBLASLt heuristic candidates is controlled by +set_cublaslt_requested_algo_count() or +PYTORCH_TUNABLEOP_CUBLASLT_REQUESTED_ALGO_COUNT, which defaults to 8. If this +count is 1, only the top cuBLASLt heuristic candidate is available. + +Offline Tuning +============== + +Motivation +---------- +There are several use cases for offline tuning. + +One use case involves a workload with a high-memory utilization, where regular tuning might lead to running out of memory. + +Another use case is for compute-intensive workloads. In such cases, it is more resource-efficient to collect +the GEMMs for the workload once and then tune repeatedly with different tuning parameters or libraries. + +Workflow +-------- +There are basically two steps: +1) Set the environment variables to collect the untuned GEMM and this will generate ``tunableop_untuned0.csv``: + +.. code-block:: bash + + export PYTORCH_TUNABLEOP_ENABLED=1 + export PYTORCH_TUNABLEOP_TUNING=0 + export PYTORCH_TUNABLEOP_RECORD_UNTUNED=1 + ... + +2) Run a Python script that reads the ``tunableop_untuned0.csv`` and generates the ``tunableop_results0.csv``, like this: + +.. code-block:: python + + import torch.cuda.tunable as tunable + import os + + os.putenv("PYTORCH_TUNABLEOP_ENABLED", "1") + os.putenv("PYTORCH_TUNABLEOP_TUNING", "1") + os.putenv("PYTORCH_TUNABLEOP_RECORD_UNTUNED", "0") + tunable.tune_gemm_in_file("tunableop_untuned0.csv") + + +It is also possible to take multiple untuned files and distribute the GEMMs for tuning to multiple GPUs +within a single node. In the first step, the GEMMs are first gathered and duplicate GEMMs are eliminated. +Next, the GEMMs are distributed to different GPUs for tuning. After all GEMMs are tuned, the results from +all the GPUs are then gathered into a single file whose base filename has ``_full0`` appended to it +(for example ``tunableop_results_full0.csv``). Finally, this new file, containing the gathered results, will be +duplicated N times, once for each GPU as convenience to the user will run the workload with the tuned +configuration on N GPUs. + +.. code-block:: python + + if __name__ == "__main__": + num_gpus = 8 # number of GPUs that will be used during the tuning process + tunable.mgpu_tune_gemm_in_file("tunableop_untuned?.csv", num_gpus) + +Note that the usage of the ``mgpu_tune_gemm_in_file`` API is different from its single GPU counterpart +(``tune_gemm_in_file``). The body of the Python script that calls the API must be wrapped in ``main()`` as shown +due to the use of concurrent futures module. The argument to ``mgpu_tune_gemm_in_file`` must contain a wild card +expression (``?`` or ``*``) to generate the list of untuned files containing the GEMMs to be processed. The ``num_gpus`` +must between 1 and the total number of GPUs available. + +Tuning Context +============== + +The behavior of TunableOp is currently manipulated through environment +variables, the C++ interface of at::cuda::tunable::getTuningContext(), or the +torch.cuda.tunable python interfaces. The environment variables take precedence +over any setting you manipulate using the C++ or Python APIs. + +Environment Variable Interface +------------------------------ +Environment variables are cached the first time they are read. You cannot use the +environment variable interface programmatically since the settings become fixed. +Use the C++ or Python APIs instead. + +""" + +import glob +import multiprocessing as mp +import os +import shutil +import warnings + +import torch + + +__all__ = [ + "enable", + "is_enabled", + "tuning_enable", + "tuning_is_enabled", + "record_untuned_enable", + "record_untuned_is_enabled", + "set_max_tuning_duration", + "get_max_tuning_duration", + "set_max_tuning_iterations", + "get_max_tuning_iterations", + "set_cublaslt_requested_algo_count", + "get_cublaslt_requested_algo_count", + "set_filename", + "get_filename", + "get_results", + "get_validators", + "read_file", + "tune_gemm_in_file", + "mgpu_tune_gemm_in_file", + "set_rotating_buffer_size", + "get_rotating_buffer_size", + "set_numerical_check_tolerances", +] + + +def enable(val: bool = True) -> None: + r"""This is the big on/off switch for all TunableOp implementations.""" + torch._C._cuda_tunableop_enable(val) # type: ignore[attr-defined] + + +def is_enabled() -> bool: + r"""Returns whether the TunableOp feature is enabled.""" + return torch._C._cuda_tunableop_is_enabled() # type: ignore[attr-defined] + + +def tuning_enable(val: bool = True) -> None: + r"""Enable tuning of TunableOp implementations. + + When enabled, if a tuned entry isn't found, run the tuning step and record + the entry. + """ + torch._C._cuda_tunableop_tuning_enable(val) # type: ignore[attr-defined] + + +def tuning_is_enabled() -> bool: + r"""Returns whether TunableOp implementations can be tuned.""" + return torch._C._cuda_tunableop_tuning_is_enabled() # type: ignore[attr-defined] + + +def record_untuned_enable(val: bool = True) -> None: + r"""Enable recording untuned TunableOp operations for offline tuning. + + When enabled, if a tuned entry isn't found, write it to the untuned file. + """ + torch._C._cuda_record_untuned_enable(val) # type: ignore[attr-defined] + + +def record_untuned_is_enabled() -> bool: + r"""Returns whether TunableOp operations are recorded for offline tuning.""" + return torch._C._cuda_record_untuned_is_enabled() # type: ignore[attr-defined] + + +def set_max_tuning_duration(duration: int) -> None: + r"""Set max time in milliseconds to spend tuning a given solution. + + If both max tuning duration and iterations are set, the smaller of the two + will be honored. At minimum 1 tuning iteration will always be run. + """ + torch._C._cuda_tunableop_set_max_tuning_duration(duration) # type: ignore[attr-defined] + + +def get_max_tuning_duration() -> int: + r"""Get max time to spend tuning a given solution.""" + return torch._C._cuda_tunableop_get_max_tuning_duration() # type: ignore[attr-defined] + + +def set_max_tuning_iterations(iterations: int) -> None: + r"""Set max number of iterations to spend tuning a given solution. + + If both max tuning duration and iterations are set, the smaller of the two + will be honored. At minimum 1 tuning iteration will always be run. + """ + torch._C._cuda_tunableop_set_max_tuning_iterations(iterations) # type: ignore[attr-defined] + + +def get_max_tuning_iterations() -> int: + r"""Get max iterations to spend tuning a given solution.""" + return torch._C._cuda_tunableop_get_max_tuning_iterations() # type: ignore[attr-defined] + + +def set_cublaslt_requested_algo_count(count: int) -> None: + r"""Set the number of cuBLASLt heuristic algorithms to request on CUDA. + + Values less than 1 are clamped to 1. + """ + torch._C._cuda_tunableop_set_cublaslt_requested_algo_count(count) # type: ignore[attr-defined] + + +def get_cublaslt_requested_algo_count() -> int: + r"""Get the number of cuBLASLt heuristic algorithms requested on CUDA.""" + get_count = ( + torch._C._cuda_tunableop_get_cublaslt_requested_algo_count # type: ignore[attr-defined] + ) + return get_count() + + +def set_filename(filename: str, insert_device_ordinal: bool = False) -> None: + r"""Set the filename to use for input/output of tuning results. + + If :attr:`insert_device_ordinal` is ``True`` then the current device ordinal + will be added to the given filename automatically. This can be used in a + 1-process-per-gpu scenario to ensure all processes write to a separate file. + """ + torch._C._cuda_tunableop_set_filename(filename, insert_device_ordinal) # type: ignore[attr-defined] + + +def get_filename() -> str: + r"""Get the results filename.""" + return torch._C._cuda_tunableop_get_filename() # type: ignore[attr-defined] + + +def get_results() -> tuple[str, str, str, float]: + r"""Return all TunableOp results.""" + return torch._C._cuda_tunableop_get_results() # type: ignore[attr-defined] + + +def get_validators() -> tuple[str, str]: + r"""Return the TunableOp validators.""" + return torch._C._cuda_tunableop_get_validators() # type: ignore[attr-defined] + + +def read_file(filename: str | None = None) -> bool: + r"""Read results from a TunableOp CSV file. + + If :attr:`filename` is not given, ``get_filename()`` is called. + """ + if filename is None: + filename = get_filename() + return torch._C._cuda_tunableop_read_file(filename) # type: ignore[attr-defined] + + +def set_rotating_buffer_size(buffer_size: int) -> None: + r"""Set rotating buffer size to this value in MB, if the buffer size is greater than zero. + + If less than zero, query L2 cache size. If equal to zero, means deactivate rotating buffer. + """ + return torch._C._cuda_tunableop_set_rotating_buffer_size(buffer_size) # type: ignore[attr-defined] + + +def get_rotating_buffer_size() -> int: + r"""Get the rotating buffer size in kilobytes.""" + return torch._C._cuda_tunableop_get_rotating_buffer_size() # type: ignore[attr-defined] + + +def set_numerical_check_tolerances( + enable: bool, atol: float = 1e-5, rtol: float = 1e-5 +) -> None: + r"""Set the atol and rtol values in numeric check""" + return torch._C._cuda_tunableop_set_numerical_check_tolerances(enable, atol, rtol) # type: ignore[attr-defined] + + +def tune_gemm_in_file(filename: str) -> None: + r"""tune GEMM in file.""" + + if not is_enabled(): + raise AssertionError("TunableOp is not enabled") + if not tuning_is_enabled(): + raise AssertionError("Tuning is not enabled") + + deviceid = torch.cuda.current_device() + + with open(filename) as file: + for line in file: + if line.startswith(("Gemm", "ScaledGemm")): + _process_single_offline_gemm(line, deviceid) + + +def _gather_unique_untuned_gemm_from_files(filename_pattern: str) -> set[str]: + r"""Process multiple untuned results file and return a set with duplicates removed.""" + unique_gemm_entries = set() # set will avoid duplicates + + for file_path in glob.glob(filename_pattern): + with open(file_path) as file: + for line in file: + if line.startswith(("Gemm", "ScaledGemm")): + unique_gemm_entries.add(line) + + return unique_gemm_entries + + +def _gather_tunableop_results() -> None: + r"""Gather results from multiple tunableop results file and create a single file.""" + gemm_lines = set() + validator_lines = [] + + # Need to allow for the possibility that results filename was + # set with the Python API instead of with environment variable. + # Also possible that results filename was not set at all. + # There are several test cases to check, but ultimately we + # need a glob-able expression + results_filename = get_filename() # Note empty string could be returned here + + if ( + results_filename is not None and results_filename != "" + ): # Case were the Python API was used to set the filename + dot_pos = results_filename.find(".") + if dot_pos != -1 and dot_pos > 0: + # Replace the character just to the left of the dot + filename_pattern = ( + results_filename[: dot_pos - 1] + "?" + results_filename[dot_pos:] + ) + else: + filename_pattern = "" # Needed to make linter happy + else: # Case where the environment variable was used to set the filename. + results_filename_env = os.getenv("PYTORCH_TUNABLEOP_FILENAME") + if results_filename_env is None or results_filename_env == "": + filename_pattern = "tunableop_results?.csv" + elif "%d" in results_filename_env: + filename_pattern = results_filename_env.replace("%d", "?") + else: + filename_pattern = results_filename_env.replace(".", "?.") + + if "?" not in filename_pattern: + raise AssertionError( + f"filename_pattern must contain '?', got {filename_pattern!r}" + ) + + FirstFile = False + matching_files = glob.glob(filename_pattern) + num_matching_files = len(matching_files) + for file_path in matching_files: + with open(file_path) as file: + for line in file: + if line.startswith("Validator"): + if not (FirstFile): + # Only read Validator from first file + validator_lines.append(line) + else: + gemm_lines.add(line) + + FirstFile = True + + output_file = filename_pattern.replace("?", "_full0") + + with open(output_file, "w") as out_file: + for line in validator_lines: + out_file.write(line) + for line in gemm_lines: + out_file.write(line) + + # Create num_matching_copies of the results file + for i in range(1, num_matching_files): + duplicate_file = output_file.replace("0", str(i)) + shutil.copy(output_file, duplicate_file) + + +def _create_matrices( + m: int, + n: int, + k: int, + lda: int, + ldb: int, + ldc: int, + transA: bool, + transB: bool, + dtypeA: torch.dtype, + deviceid: str, + dtypeB: torch.dtype | None = None, + randn: bool = True, + subMatrix: bool = False, +) -> tuple[torch.Tensor, torch.Tensor]: + r"""Helper function for _process_single_offline_gemm. + Creates matrices that are then consumed by one of the Torch GEMM APIs. + """ + # Fill parameters set for use with ScaledGEMM + fillA = 0.25 + fillB = 0.75 + + if dtypeB is None: + dtypeB = dtypeA + + if subMatrix: + # User reference for understanding leading dimension: + # https://github.com/Reference-LAPACK/lapack/blob/master/BLAS/SRC/dgemm.f + # TO DO: According to lines 108 - 133, there is no lower bound on rowsA, + # but there is a restriction on rowsB. Using this formula for now as it + # seems to work for all UTs. + rowsA = rowsB = max(ldc, k) + + if randn: + matA = torch.randn(rowsA, lda, dtype=dtypeA, device=deviceid) + matB = torch.randn(rowsB, ldb, dtype=dtypeA, device=deviceid) + else: + matA = torch.full((rowsA, lda), fillA, dtype=dtypeB, device=deviceid) + matB = torch.full((rowsB, ldb), fillB, dtype=dtypeB, device=deviceid) + + subA = matA[:k, :m].t() if transA else matA[:m, :k] + subB = matB[:n, :k].t() if transB else matB[:k, :n] + return subA, subB + else: + if randn: + matA = ( + torch.rand(k, m, dtype=dtypeA, device=deviceid).t() + if transA + else torch.rand(m, k, dtype=dtypeA, device=deviceid) + ) + matB = ( + torch.rand(n, k, dtype=dtypeB, device=deviceid).t() + if transB + else torch.rand(k, n, dtype=dtypeB, device=deviceid) + ) + else: + matA = ( + torch.full((k, m), fillA, dtype=dtypeA, device=deviceid).t() + if transA + else torch.full((m, k), fillA, dtype=dtypeA, device=deviceid) + ) + matB = ( + torch.full((n, k), fillB, dtype=dtypeB, device=deviceid).t() + if transB + else torch.full((k, n), fillB, dtype=dtypeB, device=deviceid) + ) + return matA, matB + + +def _create_batch_matrices( + m: int, + n: int, + k: int, + b: int, + lda: int, + ldb: int, + ldc: int, + transA: bool, + transB: bool, + dtype: torch.dtype, + deviceid: str, + subMatrix: bool = False, +) -> tuple[torch.Tensor, torch.Tensor]: + r"""Helper function for _process_single_offline_gemm. + Creates batch matrices that are then consumed by one of the Torch GEMM APIs. + Similar to _create_matrices but for 3D batch matrices. + """ + if subMatrix: + # User reference for understanding leading dimension: + # https://github.com/Reference-LAPACK/lapack/blob/master/BLAS/SRC/dgemm.f + # TO DO: According to lines 108 - 133, there is no lower bound on rowsA, + # but there is a restriction on rowsB. Using this formula for now as it + # seems to work for all UTs. + rowsA = rowsB = max(ldc, k) + + matA = torch.randn(b, rowsA, lda, dtype=dtype, device=deviceid) + matB = torch.randn(b, rowsB, ldb, dtype=dtype, device=deviceid) + + subA = matA[:b, :k, :m].transpose(1, 2) if transA else matA[:b, :m, :k] + subB = matB[:b, :n, :k].transpose(1, 2) if transB else matB[:b, :k, :n] + return subA, subB + else: + matA = ( + torch.rand(b, k, m, dtype=dtype, device=deviceid) + if transA + else torch.rand(b, m, k, dtype=dtype, device=deviceid) + ) + matB = ( + torch.rand(b, n, k, dtype=dtype, device=deviceid) + if transB + else torch.rand(b, k, n, dtype=dtype, device=deviceid) + ) + matA = matA.transpose(1, 2) if transA else matA + matB = matB.transpose(1, 2) if transB else matB + return matA, matB + + +def _process_single_offline_gemm(untuned_gemm_line: str, gpu_id: int) -> None: + r"""Process a single untuned GEMM.""" + + deviceid = "cuda:" + str(gpu_id) + torch.cuda.set_device(deviceid) + + dtype_dict = { + "Float": torch.float32, + "float": torch.float32, + "tf32": torch.float32, + "double": torch.float64, + "BFloat16": torch.bfloat16, + "Half": torch.half, + "c10::complex": torch.complex128, + "c10::complex": torch.complex64, + "Float8_e4m3fn": torch.float8_e4m3fn, + "Float8_e5m2": torch.float8_e5m2, + "Float8_e4m3fnuz": torch.float8_e4m3fnuz, + "Float8_e5m2fnuz": torch.float8_e5m2fnuz, + } + + untuned_gemm = untuned_gemm_line.strip().split(",")[:] + + underscore_count = untuned_gemm[0].count("_") + + # Initialize dtype to make linter happy + dtype = None + dtypeA = None + dtypeB = None + dtypeC = None + + # Extract BLAS parameters + if underscore_count == 2: + [op_sig, data_type, layout] = untuned_gemm[0].split("_") + transB = layout[0] == "T" + transA = layout[1] == "T" + dtype = dtype_dict.get(data_type) + if data_type == "tf32": + torch.backends.cuda.matmul.allow_tf32 = True + else: + torch.backends.cuda.matmul.allow_tf32 = False + + else: # ScaledGEMM + count = untuned_gemm[0].count("_") + if count not in [6, 7]: + raise AssertionError(f"count must be 6 or 7, got {count}") + untuned_gemm_temp = untuned_gemm[0].split("_") + # dtypeC = might not be FP8 type, keep track + # of the number of underscores + op_sig = untuned_gemm_temp[0] + data_typeA = untuned_gemm_temp[1] + "_" + untuned_gemm_temp[2] + data_typeB = untuned_gemm_temp[3] + "_" + untuned_gemm_temp[4] + if count == 7: + data_typeC = untuned_gemm_temp[5] + "_" + untuned_gemm_temp[6] + else: + data_typeC = untuned_gemm_temp[5] + transB = untuned_gemm_temp[count][0] == "T" + transA = untuned_gemm_temp[count][1] == "T" + dtypeA = dtype_dict.get(data_typeA) + dtypeB = dtype_dict.get(data_typeB) + dtypeC = dtype_dict.get(data_typeC) + + untuned_gemm_temp = untuned_gemm[1].split("_") + [n, m, k] = [int(g) for g in untuned_gemm_temp[1:4]] + if op_sig == "GemmStridedBatchedTunableOp": + if untuned_gemm_temp[6] != "ld": + raise AssertionError( + f"expected 'ld' at index 6, got {untuned_gemm_temp[6]!r}" + ) + [ldb, lda, ldc] = [int(g) for g in untuned_gemm_temp[7:10]] + else: + if untuned_gemm_temp[4] != "ld": + raise AssertionError( + f"expected 'ld' at index 4, got {untuned_gemm_temp[4]!r}" + ) + [ldb, lda, ldc] = [int(g) for g in untuned_gemm_temp[5:8]] + + # Detect subMatrix case + # A GEMM is "tight" (not a sub-matrix) only when each leading dimension + # equals its expected contiguous value for the given transpose layout. + # Checking mere membership in {n, m, k} is wrong: a padded leading + # dimension can coincidentally equal one of n/m/k (e.g. lda == n), which + # silently rewrites the requested shape and tunes the wrong GEMM. + # See https://github.com/ROCm/TheRock/issues/5553 + lda_tight = m if transA else k + ldb_tight = k if transB else n + ldc_tight = n + if lda == lda_tight and ldb == ldb_tight and ldc == ldc_tight: + subMatrix = False + else: + subMatrix = True + + if op_sig == "GemmTunableOp": + # Warnings for unsupported cases: + if m == 1 or n == 1 or k == 1: + if (not transA) and (not transB): + pass # case is supported + elif transA and n == 1: + pass # case is supported + else: + warnings.warn( + "Offline tuning is not supported for this GEMM. Use online tuning instead. " + + f"Skipped tuning for: {untuned_gemm[1]}", + stacklevel=2, + ) + return + + # Resolve linter issue + if dtype is None or not isinstance(dtype, torch.dtype): + raise TypeError(f"dtype must be a torch.dtype, but got {dtype}") + + matA, matB = _create_matrices( + m, n, k, lda, ldb, ldc, transA, transB, dtype, deviceid, subMatrix=subMatrix + ) + torch.mm(matA, matB) + + elif op_sig == "GemmStridedBatchedTunableOp": + # Warnings for unsupported cases: + if m == 1 or n == 1 or k == 1: + warnings.warn( + "Offline tuning is not supported for this GEMM. Use online tuning instead. " + + f"Skipped tuning for: {untuned_gemm[1]}", + stacklevel=2, + ) + return + + [b] = [int(g) for g in untuned_gemm_temp[5:6]] + + # Resolve linter issue + if dtype is None or not isinstance(dtype, torch.dtype): + raise TypeError(f"dtype must be a torch.dtype, but got {dtype}") + + matA, matB = _create_batch_matrices( + m, + n, + k, + b, + lda, + ldb, + ldc, + transA, + transB, + dtype, + deviceid, + subMatrix=subMatrix, + ) + torch.bmm(matA, matB) + elif op_sig == "ScaledGemmTunableOp": + # Only combination supported by PyTorch + if transB is not True: + raise AssertionError( + f"transB must be True for ScaledGemmTunableOp, got {transB}" + ) + if transA is not False: + raise AssertionError( + f"transA must be False for ScaledGemmTunableOp, got {transA}" + ) + + # Resolve linter issue + if dtypeA is None or not isinstance(dtypeA, torch.dtype): + raise TypeError(f"dtype must be a torch.dtype, but got {dtypeA}") + + matA, matB = _create_matrices( + m, + n, + k, + lda, + ldb, + ldc, + transA, + transB, + dtypeA, + deviceid, + dtypeB=dtypeB, + randn=False, + subMatrix=subMatrix, + ) + + if untuned_gemm_temp[8] != "rw": + raise AssertionError( + f"expected 'rw' at index 8, got {untuned_gemm_temp[8]!r}" + ) + if untuned_gemm_temp[9] == "1": + rowwise = True + else: + rowwise = False + if rowwise: + scaleA = ( + torch.ones((1, m), device=deviceid) + if transA + else torch.ones((m, 1), device=deviceid) + ) + scaleB = ( + torch.ones((1, n), device=deviceid) + if transB + else torch.ones((n, 1), device=deviceid) + ) + else: + scaleA = torch.tensor(0.8, device=deviceid) + scaleB = torch.tensor(0.9, device=deviceid) + + if untuned_gemm_temp[10] != "bias": + raise AssertionError( + f"expected 'bias' at index 10, got {untuned_gemm_temp[10]!r}" + ) + if untuned_gemm_temp[11] == "None": # no bias vector + torch._scaled_mm( + matA, matB, scale_a=scaleA, scale_b=scaleB, out_dtype=dtypeC + ) + else: # bias vector present + fillbias = 0.10 + bias_dtype = dtype_dict.get(untuned_gemm_temp[11]) + bias = ( + torch.full((n,), fillbias, dtype=bias_dtype, device=deviceid) + if transB + else torch.full((m,), fillbias, dtype=bias_dtype, device=deviceid) + ) + torch._scaled_mm( + matA, matB, scale_a=scaleA, scale_b=scaleB, out_dtype=dtypeC, bias=bias + ) + + elif op_sig == "GemmAndBiasTunableOp": + # y = x*A^T + b + if transA == transB: + raise AssertionError( + f"transA and transB must differ for GemmAndBiasTunableOp, got transA={transA}, transB={transB}" + ) + + # Resolve linter issue + if dtype is None or not isinstance(dtype, torch.dtype): + raise TypeError(f"dtype must be a torch.dtype, but got {dtype}") + + bias = torch.rand(n, dtype=dtype, device=deviceid) + + X, matA = _create_matrices( + m, n, k, lda, ldb, ldc, transA, transB, dtype, deviceid, subMatrix=subMatrix + ) + matA = matA.t() + torch.nn.functional.linear(X, matA, bias) + else: + warnings.warn(f"error: unknown op {op_sig}", stacklevel=2) + + +def _process_offline_gemms(untuned_gemm_lines: list[str], gpu_id: int) -> None: + r"""Process multiple untuned GEMMs on a single GPU.""" + _check_tuning_assertions() + for line in untuned_gemm_lines: + _process_single_offline_gemm(line, gpu_id) + + +def _check_tuning_assertions() -> None: + r"""Helper function for multi-GPU tuning case. Need to check that TunableOp feature + is enabled and that tuning is enabled. + """ + + if is_enabled() is False: + warnings.warn("TunableOp was disabled. Trying to enable now.", stacklevel=2) + enable(True) + if is_enabled() is not True: + raise AssertionError("is_enabled() must be True") + if tuning_is_enabled() is not True: + raise AssertionError("tuning_is_enabled() must be True") + if record_untuned_is_enabled() is not False: + raise AssertionError("record_untuned_is_enabled() must be False") + + +def mgpu_tune_gemm_in_file(filename_pattern: str, num_gpus: int) -> None: + r"""Process one or more files and distribute work over one or more GPUs.""" + unique_gemm_entries = _gather_unique_untuned_gemm_from_files(filename_pattern) + + total_gpus = torch.cuda.device_count() + + if not (1 <= num_gpus <= total_gpus): + raise AssertionError( + f"num_gpus must be between 1 and {total_gpus}, got {num_gpus}" + ) + + mp_context = mp.get_context("spawn") + + gemm_entries_by_gpu: list[list[str]] = [[] for _ in range(num_gpus)] + + # GEMM are assigned to GPUs in a round robin manner + for h, line in enumerate(unique_gemm_entries): + gemm_entries_by_gpu[h % num_gpus].append(line) + + processes = [] + for h, entries in enumerate(gemm_entries_by_gpu): + if not entries: + continue + # TunableOp initializes its output filename once per process, so keep + # each spawned process bound to a single GPU. + process = mp_context.Process(target=_process_offline_gemms, args=(entries, h)) + process.start() + processes.append((h, process)) + + failed_processes = [] + for h, process in processes: + process.join() + if process.exitcode != 0: + failed_processes.append((h, process.exitcode)) + + if failed_processes: + raise RuntimeError(f"offline tuning processes failed: {failed_processes}") + + torch.cuda.synchronize() + + _gather_tunableop_results() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..792054dc4c7db849a873ccc7aa566344ba180ca2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/__init__.py @@ -0,0 +1,209 @@ +# mypy: allow-untyped-defs +import contextlib +import logging +import sys +import traceback +import typing +from datetime import timedelta + +import torch + + +RankType = int | torch.SymInt + + +log = logging.getLogger(__name__) + + +def is_available() -> bool: + """ + Return ``True`` if the distributed package is available. + + Otherwise, + ``torch.distributed`` does not expose any other APIs. Currently, + ``torch.distributed`` is available on Linux, MacOS and Windows. Set + ``USE_DISTRIBUTED=1`` to enable it when building PyTorch from source. + Currently, the default value is ``USE_DISTRIBUTED=1`` for Linux and Windows, + ``USE_DISTRIBUTED=0`` for MacOS. + """ + return hasattr(torch._C, "_c10d_init") + + +_spmd_types_available: bool | None = None + + +def _is_spmd_types_available() -> bool: + """Check if the spmd_types package is installed.""" + import importlib.util + + global _spmd_types_available + if _spmd_types_available is None: + _spmd_types_available = importlib.util.find_spec("spmd_types") is not None + return _spmd_types_available + + +def _spmd_no_typecheck(): + """ + Return a spmd_types no_typecheck context, or a no-op if not installed. + """ + if _is_spmd_types_available(): + import spmd_types + + return spmd_types.no_typecheck() + + @contextlib.contextmanager + def no_typecheck(): + yield + + return no_typecheck() + + +if is_available() and not torch._C._c10d_init(): + raise RuntimeError("Failed to initialize torch.distributed") + +# Custom Runtime Errors thrown from the distributed package +DistError = torch._C._DistError +DistBackendError = torch._C._DistBackendError +DistNetworkError = torch._C._DistNetworkError +DistStoreError = torch._C._DistStoreError +QueueEmptyError = torch._C._DistQueueEmptyError + +if is_available(): + from torch._C._distributed_c10d import ( + _broadcast_coalesced, + _compute_bucket_assignment_by_size, + _DEFAULT_FIRST_BUCKET_BYTES, + _make_nccl_premul_sum, + _register_builtin_comm_hook, + _register_comm_hook, + _test_python_store, + _verify_params_across_processes, + Backend as _Backend, + BuiltinCommHookType, + DebugLevel, + FileStore, + get_debug_level, + GradBucket, + Logger, + PrefixStore, + ProcessGroup as ProcessGroup, + Reducer, + set_debug_level, + set_debug_level_from_env, + Store, + TCPStore, + Work as _Work, + ) + + def _make_distributed_pdb(): + """ + Supports using PDB from inside a multiprocessing child process. + + Usage: + _make_distributed_pdb().set_trace() + """ + + # Lazy import pdb only if we set breakpoints. + import pdb + + class _DistributedPdb(pdb.Pdb): + def interaction(self, *args, **kwargs): + _stdin = sys.stdin + try: + with open("/dev/stdin") as sys.stdin: + pdb.Pdb.interaction(self, *args, **kwargs) + finally: + sys.stdin = _stdin + + return _DistributedPdb() + + _breakpoint_cache: dict[int, typing.Any] = {} + + def breakpoint(rank: int = 0, skip: int = 0, timeout_s=3600): + """ + Set a breakpoint, but only on a single rank. All other ranks will wait for you to be + done with the breakpoint before continuing. + + Args: + rank (int): Which rank to break on. Default: ``0`` + skip (int): Skip the first ``skip`` calls to this breakpoint. Default: ``0``. + """ + if skip > 0: + key = hash(str(traceback.format_exc())) + counter = _breakpoint_cache.get(key, 0) + 1 + _breakpoint_cache[key] = counter + if counter <= skip: + log.warning("Skip the breakpoint, counter=%d", counter) + return + + # avoid having the default timeout (if short) interrupt your debug session + if timeout_s is not None: + for group in torch.distributed.distributed_c10d._pg_map: + torch.distributed.distributed_c10d.set_timeout( + timedelta(seconds=timeout_s), group + ) + + if get_rank() == rank: + pdb = _make_distributed_pdb() + pdb.message( + "\n!!! ATTENTION !!!\n\n" + f"Type 'up' to get to the frame that called dist.breakpoint(rank={rank})\n" + ) + pdb.set_trace() + # If Meta/Python keys are in the TLS, we want to make sure that we ignore them + # and hit the (default) CPU/CUDA implementation of barrier. + meta_in_tls = torch._C._meta_in_tls_dispatch_include() + guard = torch._C._DisableTorchDispatch() # type: ignore[attr-defined] + torch._C._set_meta_in_tls_dispatch_include(False) + try: + barrier() + finally: + torch._C._set_meta_in_tls_dispatch_include(meta_in_tls) + del guard + + if sys.platform != "win32": + from torch._C._distributed_c10d import HashStore + + from .device_mesh import DeviceMesh, init_device_mesh + + # Variables prefixed with underscore are not auto imported + # See the comment in `distributed_c10d.py` above `_backend` on why we expose + # this. + from .distributed_c10d import * # noqa: F403 + from .distributed_c10d import ( + _all_gather_base, + _coalescing_manager, + _CoalescingManager, + _create_process_group_wrapper, + _get_process_group_name, + _get_reconfigure_handle, + _new_window, + _rank_not_in_group, + _reconfigure, + _reduce_scatter_base, + _supports_reconfigure, + _supports_window, + _time_estimator, + get_node_local_rank, + ) + from .remote_device import _remote_device + from .rendezvous import ( + _create_store_from_options, + register_rendezvous_handler, + rendezvous, + ) + + set_debug_level_from_env() + +else: + # This stub is sufficient to get + # python test/test_public_bindings.py -k test_correct_module_names + # working even when USE_DISTRIBUTED=0. Feel free to add more + # stubs as necessary. + # We cannot define stubs directly because they confuse pyre + + class _Stub: + pass + + sys.modules["torch.distributed"].GroupName = _Stub # type: ignore[attr-defined] + sys.modules["torch.distributed"].ProcessGroup = _Stub # type: ignore[attr-defined] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3af13bcfb640a45d601af9a232281174901f3d25 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_checkpointable.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_checkpointable.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4a7e62a93dd56035bdb1ea04331624d4f5a6f639 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_checkpointable.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_composable_state.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_composable_state.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1b0ae7e445bc529d616b32374d7bbaa0670277dc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_composable_state.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2a59f486c6c80db231a8fcd0cdc3e6a371ad839f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives_impl.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives_impl.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cff4f5607041e9452de9d253c0dd58d303fd5085 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_functional_collectives_impl.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_mesh_layout.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_mesh_layout.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a092a75c3528da6c5d990a215cb8046f2bca219c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_mesh_layout.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_meta_registrations.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_meta_registrations.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..85cf673250d4a3029b834d0f8ed12ca94287b3d9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_meta_registrations.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_serialization.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_serialization.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..345abb75378974876f1a64a321c680962be76c04 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_serialization.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_state_dict_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_state_dict_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1c0e54913b0e7a30ee09e52690aca04e170302de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_state_dict_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_token_switch.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_token_switch.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..13aba08e158337c8b0103d9a0211437ebc82b97c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/_token_switch.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/argparse_util.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/argparse_util.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ca1f88c98985c69eeb313cd73b67371be4402675 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/argparse_util.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/c10d_logger.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/c10d_logger.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..656cd2d6de53dcbba951d6eec73e48c7c1b350ac Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/c10d_logger.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/collective_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/collective_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..05f6126f76da5cca813bb0d5e05737690adc1a82 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/collective_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/config.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/config.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5e6cb2b2c5ba1d2e2a718abe5a4b2f78762d6d80 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/config.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/constants.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/constants.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8179580f1ee4991e5d36fc2cb3984d3e68828a55 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/constants.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/device_mesh.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/device_mesh.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..415c6814e567fd42806e218141666fa2970ba804 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/device_mesh.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/distributed_c10d.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/distributed_c10d.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9480cdd7f674d86e5a96ecbea6eb18ada4dc6cb5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/distributed_c10d.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8b32ec83597fec2bcbc95daba4bd7c873f327fd8f2e1280e16065a2f878e825 +size 302967 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/launch.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/launch.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9529faa3ddec0c38d1484118d8cc7f578fa12370 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/launch.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/logging_handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/logging_handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7d60ae22c087b36a7371102f71f6381a853b927b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/logging_handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/remote_device.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/remote_device.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc206d87eb0d1dac70cbd87970bf0e05c831ce99 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/remote_device.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/rendezvous.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/rendezvous.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..18892a032bdcfb203a539b65615de6d930d018b7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/rendezvous.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/run.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/run.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f4b4a33f03760aeb12f26aea54a9c66ed234605d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/run.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a7b2b92242724158fdbca0a36e2fad073aecb46d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_checkpointable.py b/.venv/lib/python3.12/site-packages/torch/distributed/_checkpointable.py new file mode 100644 index 0000000000000000000000000000000000000000..0594c20337b3bf1c73fb40e2218e0c71580b75c5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_checkpointable.py @@ -0,0 +1,37 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from typing_extensions import Protocol, runtime_checkable + +import torch + + +@runtime_checkable +class _Checkpointable(Protocol): # noqa: PYI046 + """ + Interface for checkpointable objects. + Implemented as a protocol, implicit subtyping is supported so subclasses do not need to inherit this explicitly. + This is to allow arbitrary objects/tensor subclasses to hook into DCP seamlessly through implementing the interface. + """ + + def __create_write_items__(self, fqn: str, object: object) -> list[object]: + """ + Return a list of WriteItems based on object's contents. + """ + raise NotImplementedError( + "_Checkpointable._create_write_items is not implemented" + ) + + def __create_chunk_list__(self) -> list[object]: + """ + Return a list of `ChunkStorageMetadata` based on object's contents. + """ + raise NotImplementedError( + "_Checkpointable._create_chunk_list is not implemented" + ) + + def __get_tensor_shard__(self, index: int) -> torch.Tensor: + """ + Return a 'torch.Tensor' shard based on 'MetadataIndex'. + """ + raise NotImplementedError( + "_Checkpointable._get_tensor_shard is not implemented" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..3e38281810696814a7eae148eff19b58c10e072b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__init__.py @@ -0,0 +1,3 @@ +from .checkpoint_activation import checkpoint +from .contract import _get_registry, contract +from .replicate import replicate diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..54a9608d3f5371dde436e6aee850255230903f21 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/checkpoint_activation.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/checkpoint_activation.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..98a141cd60585a7109da94336859081ff4f7f8d0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/checkpoint_activation.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/contract.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/contract.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..be928f7d987659aa15ea10b3369f6fd9bcefe731 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/contract.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..72ad3194b28410eb2888caa8a6bc580370095f44 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate_with_fsdp.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate_with_fsdp.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9b3ce00fb8d5092548611f17622b6fbc421afcfe Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/__pycache__/replicate_with_fsdp.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/checkpoint_activation.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/checkpoint_activation.py new file mode 100644 index 0000000000000000000000000000000000000000..2922cf7c1ff6ed10b393377605b8a9f8ebf50327 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/checkpoint_activation.py @@ -0,0 +1,135 @@ +# mypy: allow-untyped-defs +from collections.abc import Generator +from contextlib import AbstractContextManager, contextmanager, nullcontext +from typing import Any + +import torch +import torch.nn as nn +from torch.utils.checkpoint import ( + _checkpoint_without_reentrant_generator_impl, + _DEFAULT_DETERMINISM_MODE, +) + +from .contract import _State, contract + + +@contextmanager +def _no_hook(module: nn.Module, user_ctx: AbstractContextManager | None = None): + r""" + Disable hooks installed by checkpoint to avoid unintentional recursion + during backward recomputation. + """ + + with user_ctx if user_ctx else nullcontext(): + orig_enable_hook = checkpoint.state(module).enable_hook + checkpoint.state(module).enable_hook = False + try: + yield + finally: + checkpoint.state(module).enable_hook = orig_enable_hook + + +class _CheckpointState(_State): + enable_hook: bool = False + _ac_generator: Generator[None, None, None] | None + + +@contract(_CheckpointState) +def checkpoint(module: nn.Module, **kwargs) -> nn.Module: + r""" + This is a composable activation checkpointing API. Unlike functional + activation checkpointing APIs, this one does not require changing model + source code. Unlike ``nn.Module`` wrapper activation checkpointing APIs, + this one does not modify model structure or fully-qualified names either. + Under the hood, it registers activation checkpointing logic as pre- and + post-forward hooks. Hence, this API can be easily applied to any model or + sub-modules in the model. + + Args: + module (nn.Module): the target model or sub-module to apply activation + checkpointing. + + Example:: + >>> # xdoctest: +SKIP + >>> import torch.nn as nn + >>> + >>> class MyModel(nn.Module): + >>> def __init__(self) -> None: + >>> super().__init__() + >>> self.l1 = nn.Linear(10, 10) + >>> self.l2 = nn.Linear(10, 10) + >>> + >>> def forward(self, x): + >>> return self.l2(self.l1(x)) + >>> + >>> model = MyModel() + >>> checkpoint(model.l1) # apply activation checkpointing only to l1 + >>> model(torch.zeros(2, 10)).sum().backward() + + """ + torch._C._log_api_usage_once("torch.distributed.checkpoint") + + use_reentrant = kwargs.pop("use_reentrant", False) + if use_reentrant: + raise NotImplementedError( + "use_reentrant=True is not supported in composable checkpoint. " + "Please use torch.utils.checkpoint.checkpoint instead." + ) + preserve_rng_state = kwargs.pop("preserve_rng_state", True) + user_context_fns = kwargs.pop("context_fn", None) + determinism_check = kwargs.pop("determinism_check", _DEFAULT_DETERMINISM_MODE) + debug = kwargs.pop("debug", False) + early_stop = kwargs.pop("early_stop", True) + + if kwargs: + raise ValueError( + "Unexpected keyword arguments: " + ",".join(arg for arg in kwargs) + ) + + def forward_pre_hook( + module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> None: + if checkpoint.state(module).enable_hook: + + def context_fns(): + if user_context_fns is not None: + ctx1, ctx2 = user_context_fns() + return ctx1, _no_hook(module, ctx2) + else: + return nullcontext(), _no_hook(module) + + gen = _checkpoint_without_reentrant_generator_impl( + module, + args, + kwargs, + preserve_rng_state=preserve_rng_state, + context_fn=context_fns, + determinism_check=determinism_check, + debug=debug, + early_stop=early_stop, + ) + checkpoint.state(module)._ac_generator = gen + next(gen) + + def forward_hook(module: nn.Module, inputs: tuple[Any, ...], output: Any) -> Any: + if checkpoint.state(module).enable_hook: + try: + gen = checkpoint.state(module)._ac_generator + if gen is None: + raise AssertionError + next(gen) + except StopIteration: + pass + else: + raise RuntimeError( + "Expected non-reentrant activation checkpoint generator to be exhausted, but it was not!" + ) + + # Ensure that we no longer hold on to the generator. always_call=True helps ensure we + # clear this even in the case of exception in fwd pass. + checkpoint.state(module)._ac_generator = None + + checkpoint.state(module).enable_hook = True + module.register_forward_pre_hook(forward_pre_hook, with_kwargs=True) + module.register_forward_hook(forward_hook, prepend=True, always_call=True) + return module diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/contract.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/contract.py new file mode 100644 index 0000000000000000000000000000000000000000..1bf3727aa63d2cc519668af899681ad4e2bf7ee3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/contract.py @@ -0,0 +1,250 @@ +# mypy: allow-untyped-defs +import uuid +from collections import OrderedDict +from collections.abc import Callable +from functools import wraps +from typing import Concatenate, Generic, Protocol +from typing_extensions import ParamSpec, TypeVar + +import torch +import torch.nn as nn +from torch.distributed._composable_state import _State +from torch.distributed.utils import _get_root_modules + + +_T = TypeVar("_T", covariant=True) +_P = ParamSpec("_P") + + +def generate_state_key(string="__composable_api_state_key"): + return f"{string}_{str(uuid.uuid4())}" + + +STATE_KEY = generate_state_key() +REGISTRY_KEY = generate_state_key() + + +# TODO: we can add additional info to RegistryItem to share across APIs. E.g., +# we can add args and kwargs here, and then we can detect whether fully_shard +# is combined with reentrant activation checkpointing and error out with a clear +# message. +class RegistryItem: + pass + + +_TState = TypeVar("_TState", bound="_State", covariant=True) +_M = TypeVar("_M", nn.Module, list[nn.Module]) + + +class _ContractFn(Protocol, Generic[_P, _T, _TState]): + def __call__(self, *args: _P.args, **kwargs: _P.kwargs) -> _T: ... + + def state(self, module: nn.Module) -> _TState: ... + + +def contract( + state_cls: type[_TState] = _State, # type: ignore[assignment] +) -> Callable[ + [Callable[Concatenate[_M, _P], _M]], + _ContractFn[Concatenate[_M, _P], _M, _TState], +]: + r""" + Decorate a function as a composable distributed API, where the first + argument of the function must be an :class:`nn.Module` instance or sequence + of :class:`nn.Module` instances. + + The decorator verifies that the decorated function does not modify + fully-qualified names (FQNs) for parameters, buffers, or modules. The + decorated function can return different module instances than the input + modules; the FQN invariant will be enforced following the input order. + + When a function ``func`` is decorated by ``@contract()``, a + ``.state(module: nn.Module)`` method will be installed to the decorated + function. Then you can retrieve and modify the state on a module by calling + ``func.state(module)``. + + Example:: + >>> # xdoctest: +SKIP + >>> import torch.nn as nn + >>> + >>> class MyModel(nn.Module): + >>> def __init__(self) -> None: + >>> super().__init__() + >>> self.l1 = nn.Linear(10, 10) + >>> self.l2 = nn.Linear(10, 10) + >>> + >>> def forward(self, x): + >>> return self.l2(self.l1(x)) + >>> + >>> @contract() + >>> def my_feature(module: nn.Module) -> nn.Module: + >>> my_feature.state(module).some_state = "any value" + >>> return module + >>> + >>> model = MyModel() + >>> my_feature(model.l1) + >>> assert my_feature.state(model.l1).some_state == "any value" + >>> my_feature(model.l2) + >>> model(torch.randn(2, 10)).sum().backward() + """ + + # wraps will make functions decorated with contract() pickleable - needed for integration with torch.package + @wraps(state_cls) # type: ignore[arg-type] + def inner( + func: Callable[Concatenate[_M, _P], _M], + ) -> _ContractFn[Concatenate[_M, _P], _M, _TState]: + @wraps(func) + def wrapper( + module: _M, + *args: _P.args, + **kwargs: _P.kwargs, + ) -> _M: + inp_module = module + modules: list[nn.Module] + if isinstance(module, nn.Module): + modules = [module] + else: + # If the user passes a sequence of modules, then we assume that + # we only need to insert the state object on the root modules + # (i.e. those without a parent) among the passed-in modules. + # pyrefly: ignore [no-matching-overload] + modules = _get_root_modules(list(module)) + state = state_cls() # shared across all modules + registry_item = RegistryItem() # shared across all modules + + # `func` is allowed to return different module instances than the + # input modules as long as FQNs are preserved following the input + # module order + all_orig_named_params: list[dict[str, nn.Parameter]] = [] + all_orig_named_buffers: list[dict[str, torch.Tensor]] = [] + all_orig_named_modules: list[dict[str, nn.Module]] = [] + + for module in modules: + default_all_state: dict[Callable, _State] = OrderedDict() + default_registry: dict[str, RegistryItem] = OrderedDict() + all_state: dict[Callable, _State] = module.__dict__.setdefault( # type: ignore[call-overload] + STATE_KEY, default_all_state + ) + if not isinstance(all_state, dict): + raise AssertionError( + f"Distributed composable API states corrupted: {all_state}" + ) + registry: dict[str, RegistryItem] = module.__dict__.setdefault( # type: ignore[call-overload] + REGISTRY_KEY, default_registry + ) + if not isinstance(registry, dict): + raise AssertionError( + f"Distributed composable API registry corrupted: {registry}" + ) + if func in all_state or func.__name__ in registry: + raise AssertionError( + "Each distinct composable distributed API can only be applied to a " + f"module once. {func.__name__} has already been applied to the " + f"following module:\n{module}" + ) + all_state.setdefault(func, state) + registry.setdefault(func.__name__, registry_item) + + all_orig_named_params.append(OrderedDict(module.named_parameters())) + all_orig_named_buffers.append(OrderedDict(module.named_buffers())) + all_orig_named_modules.append(OrderedDict(module.named_modules())) + + updated = func(inp_module, *args, **kwargs) + if updated is None: + updated = inp_module # type: ignore[assignment] + updated_modules: list[nn.Module] + if isinstance(updated, nn.Module): + updated_modules = [updated] + else: + updated_modules = _get_root_modules(list(inp_module)) # type: ignore[arg-type, call-overload] + + all_new_named_params: list[dict[str, nn.Parameter]] = [] + all_new_named_buffers: list[dict[str, torch.Tensor]] = [] + all_new_named_modules: list[dict[str, nn.Module]] = [] + for module in updated_modules: + all_new_named_params.append(OrderedDict(module.named_parameters())) + all_new_named_buffers.append(OrderedDict(module.named_buffers())) + all_new_named_modules.append(OrderedDict(module.named_modules())) + + num_orig_modules = len(all_orig_named_modules) + num_new_modules = len(all_new_named_modules) + if num_orig_modules != num_new_modules: + raise AssertionError( + f"{func.__name__} should return the same number of modules as input modules" + f"Inputs: {num_orig_modules} modules\n" + f"Outputs: {num_new_modules} modules" + ) + + def check_fqn(orig_fqns: list[str], new_fqns: list[str], check_key: str): + if orig_fqns == new_fqns: + return + + orig_fqn_set, new_fqn_set = set(orig_fqns), set(new_fqns) + orig_only = orig_fqn_set - new_fqn_set + new_only = new_fqn_set - orig_fqn_set + if len(orig_only) or len(new_only): + raise RuntimeError( + f"{check_key}" + "Composable distributed API implementations cannot modify FQNs.\n" + f"FQNs only in original: {orig_only}\n" + f"FQNs only in new: {new_only}" + ) + else: + raise RuntimeError( + f"{check_key}" + "Composable distributed API implementations cannot modify " + "the order of FQNs.\n" + f"Original FQNs: {orig_only}\n" + f"New FQNs: {new_only}" + ) + + for orig_named_params, new_named_params in zip( + all_orig_named_params, all_new_named_params + ): + check_fqn( + list(orig_named_params.keys()), + list(new_named_params.keys()), + "Checking parameters: ", + ) + for orig_named_buffers, new_named_buffers in zip( + all_orig_named_buffers, all_new_named_buffers + ): + check_fqn( + list(orig_named_buffers.keys()), + list(new_named_buffers.keys()), + "Checking buffers: ", + ) + for orig_named_modules, new_named_modules in zip( + all_orig_named_modules, all_new_named_modules + ): + check_fqn( + list(orig_named_modules.keys()), + list(new_named_modules.keys()), + "Checking modules: ", + ) + + # TODO: verify that installed distributed paradigms are compatible with + # each other. + + return updated + + def get_state(module: nn.Module) -> _State: + return module.__dict__.setdefault( # type: ignore[call-overload] + STATE_KEY, + {}, # TODO(@yhcharles): this is a temporary fix, need a better way + ).get(func) # type: ignore[call-overload] + + wrapper.state = get_state # type: ignore[attr-defined] + + return wrapper # type: ignore[return-value] + + return inner # type: ignore[return-value] + + +def _get_registry(module: nn.Module) -> dict[str, RegistryItem] | None: + r""" + Get an ``OrderedDict`` of composable APIs that have been applied to the + ``module``, indexed by the API name. If no API has been applied, then this + returns ``None``. + """ + return getattr(module, REGISTRY_KEY, None) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..108c765ba4766bf7d9110aa67e09ac02cab00410 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__init__.py @@ -0,0 +1,3 @@ +from torch.distributed.fsdp import CPUOffloadPolicy, MixedPrecisionPolicy, OffloadPolicy + +from .fully_shard import FSDPModule, fully_shard, register_fsdp_forward_method diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..12355e843d8bf62887c856cbbdd6046d98e492eb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/fully_shard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/fully_shard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..06636e8be3b172a33904ce32510841690875ba2f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/__pycache__/fully_shard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/fully_shard.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/fully_shard.py new file mode 100644 index 0000000000000000000000000000000000000000..9e36c7b430fc89dd58cc5742f299ac607eb4367b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/fsdp/fully_shard.py @@ -0,0 +1,8 @@ +# TODO: For backward compatibility, we are importing the public objects +# originally from this file. +from torch.distributed.fsdp import ( # noqa: F401 + FSDPModule, + fully_shard, + register_fsdp_forward_method, + UnshardHandle, +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate.py new file mode 100644 index 0000000000000000000000000000000000000000..fb4d5f574a6718bfe496662cd294bbc73164d91c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate.py @@ -0,0 +1,257 @@ +# mypy: allow-untyped-defs +import weakref +from collections.abc import Iterable +from typing import Any, NoReturn + +import torch +import torch.nn as nn +from torch.distributed._composable_state import _State +from torch.nn.parallel import DistributedDataParallel + +from .contract import _get_registry, contract + + +_ROOT_MODULE_PREFIX = "" + + +class _ReplicateState(_State): + _ddp_weakref: weakref.ref + + def __init__(self) -> None: + super().__init__() + self.module: nn.Module = nn.ParameterList() + self.has_initialized: bool = False + self._param_list: nn.ParameterList = nn.ParameterList() + # TODO(@fegin): this variable is originally created for testing, we + # should remove this if possible. + self._orig_module = self.module + self._param_names: list[str] = [] + self._no_sync: bool = False + self._init_args: tuple[Any, ...] | None = None + self._init_kwargs: dict[str, Any] = {} + self._comm_hook_args: list[Any] = [] + + def _collect_params( + self, + module: nn.Module, + ignored_modules: set[nn.Module], + ignored_params: set[nn.Parameter], + prefix: str = _ROOT_MODULE_PREFIX, + ) -> None: + # skip if managed by fully_sharded API + if _is_fully_sharded(module): + return + + # if a module is ignored, all descendants of the module are ignored. + if module in ignored_modules: + return + + recurse_prefix = ( + f"{prefix}." if prefix != _ROOT_MODULE_PREFIX else _ROOT_MODULE_PREFIX + ) + + for n, p in module.named_parameters(recurse=False): + if p not in ignored_params: + self._param_list.append(p) + self._param_names.append(f"{recurse_prefix}{n}") + + for name, child_module in module.named_children(): + self._collect_params( + child_module, + ignored_modules, + ignored_params, + prefix=f"{recurse_prefix}{name}", + ) + + def lazy_init(self) -> None: + @torch._disable_dynamo(recursive=True) + def _lazy_init(): + if self._init_args is None: + raise AssertionError + self.init(*self._init_args, **self._init_kwargs) + self.register_comm_hook() + self._init_args = () + self._init_kwargs = {} + + _lazy_init() + + def init( + self, + module: nn.Module, + ignored_modules: set[nn.Module], + **kwargs, + ) -> None: + if self.has_initialized: + return + + self.has_initialized = True + self.module = module + ignored_params = {p for m in ignored_modules for p in m.parameters()} + for submodule in module.modules(): + if _is_fully_sharded(submodule): + ignored_params.update(submodule.parameters()) + from torch.distributed.tensor.parallel.ddp import _localize_dtensor + + _localize_dtensor(module, ignored_params=ignored_params) + self._collect_params(module, ignored_modules, ignored_params) + + if "device_id" in kwargs: + # replicate() supports a small usability enhancement where + # user can pass in device_id as a Union[int, torch.device] even for + # CPU devices so users don't have to change code for CPU/GPU runs. + # We derive the right device_ids to feed into DDP to support this. + if kwargs["device_id"] is not None: + device_id = kwargs["device_id"] + # Convert to device_ids that DDP expects. + if isinstance(device_id, torch.device) and device_id.type == "cpu": + # CPU modules receive device_ids None + kwargs["device_ids"] = None + else: + # GPU modules expect device_ids=[cuda_device] + kwargs["device_ids"] = [device_id] + else: + kwargs["device_ids"] = None + kwargs.pop("device_id") + + self._ddp = DistributedDataParallel(self._param_list, **kwargs) + # Weakref to the DDP instance is currently only used for testing. + replicate.state(self.module)._ddp_weakref = weakref.ref(self._ddp) + + def register_comm_hook(self) -> None: + for comm_args, comm_kwargs in self._comm_hook_args: + self._ddp.register_comm_hook(*comm_args, **comm_kwargs) + self._comm_hook_args.clear() + + def record_init_args(self, *args, **kwargs) -> None: + self._init_args = args + self._init_kwargs = kwargs + + def forward_pre_hook( + self, module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> Any: + if self._init_args or self._init_kwargs: + self.lazy_init() + self._ddp.require_backward_grad_sync = not self._no_sync + DistributedDataParallel._active_ddp_module = self._ddp + return self._ddp._pre_forward(*args, **kwargs) + + def forward_post_hook( + self, + module: nn.Module, + input: tuple[torch.Tensor], + output: torch.Tensor, + ) -> torch.Tensor: + DistributedDataParallel._active_ddp_module = None + return self._ddp._post_forward(output) + + +def unimplemented_deepcopy(*args: Any, **kwargs: Any) -> NoReturn: + raise AssertionError( + "DDP does not support deepcopy. Please use state dict for serialization." + ) + + +# Follow the same pattern as FSDP/fully_shard +class DDP: + def __new__(cls, *args, **kwargs): + """ + Override ``__new__`` to remove the DDP class and directly construct + the original class for cases like indexing into a container module. + """ + # Use index 2 since 0 is the dynamically constructed `DDP<...>` class + # and index 1 is the `DDP` class itself + orig_cls = cls.__mro__[2] + return orig_cls.__new__(orig_cls, *args, **kwargs) + + def set_requires_gradient_sync(self, requires_gradient_sync: bool) -> None: + """ + Sets if the module should sync gradients. This can be used to implement + gradient accumulation without communication. + + Args: + requires_gradient_sync (bool): Whether to reduce gradients for the + module's parameters. + """ + replicate.state(self)._no_sync = not requires_gradient_sync # type: ignore[arg-type] + + def register_comm_hook(self, *args, **kwargs) -> None: + replicate.state(self)._comm_hook_args.append((args, kwargs)) # type: ignore[arg-type] + + +@contract(state_cls=_ReplicateState) +def replicate( + module: nn.Module, + ignored_modules: Iterable[torch.nn.Module] | None = None, + **kwargs, +) -> nn.Module: + r"""Replicates a module + + Args: + module (torch.nn.Module): module to replicate + + Example:: + >>> # xdoctest: +REQUIRES(module:torch._C._distributed_c10d) + >>> module = nn.Linear(3, 3) + >>> replicate(module) + """ + torch._C._log_api_usage_once("torch.distributed.replicate") + + # TODO(fegin): using kwargs is not a good idea if we would like to make + # replicate a formal API to replace DDP. + if "device_id" in kwargs: + if not isinstance(kwargs["device_id"], (int, torch.device)): + raise RuntimeError( + "Expected device_id to be int or torch.device, " + f"but got {type(kwargs['device_id'])}" + ) + + if _is_fully_sharded(module): + raise RuntimeError( + "Cannot apply `replicate()` on a Module already managed by `fully_shard`" + ) + + if ignored_modules is None: + ignored_modules = {} + else: + ignored_modules = set(ignored_modules) + + state = replicate.state(module) + module.register_forward_pre_hook(state.forward_pre_hook, with_kwargs=True) + device_mesh = kwargs.get("device_mesh") + if device_mesh is not None: + root_mesh = device_mesh._get_root_mesh() + # if a root mesh is not the same as device_mesh, + # meaning the device_mesh is sliced out from the root mesh. + if root_mesh != device_mesh: + # TODO: This is a temporary work around to enable DDP + TP. + # We should do the logic in DDP so that the 2D implementation is + # sound and the state_dict works out of the box. + # + # This won't conflict with what is done in DDP class as the module + # replicate is going to pass is NOT the original module. + from torch.distributed.tensor.parallel.ddp import ( + _localize_dtensor, + _reconstruct_dtensor, + ) + + module.register_forward_pre_hook(_reconstruct_dtensor) + module.register_forward_hook(_localize_dtensor) + + module.register_forward_hook(state.forward_post_hook) # type: ignore[arg-type] + + state.record_init_args(module, ignored_modules, **kwargs) + + # Place DDP leftmost for highest priority in the method resolution order + cls = module.__class__ + dct = {"__deepcopy__": unimplemented_deepcopy} + new_cls = type(f"DDP{cls.__name__}", (DDP, cls), dct) + module.__class__ = new_cls + return module + + +def _is_fully_sharded(module: nn.Module) -> bool: + r"""Check if module is marked with fully_shard.""" + registry = _get_registry(module) + if registry is None: + return False + return "fully_shard" in registry diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate_with_fsdp.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate_with_fsdp.py new file mode 100644 index 0000000000000000000000000000000000000000..e0c9125fca52dd11b49a686baf3379209bcbe30d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable/replicate_with_fsdp.py @@ -0,0 +1,204 @@ +# mypy: allow-untyped-defs +from __future__ import annotations + +import logging +from typing import overload, TYPE_CHECKING + +import torch +import torch.nn as nn +from torch.distributed._composable_state import _get_module_state +from torch.distributed.fsdp._fully_shard._fsdp_api import ( + MixedPrecisionPolicy, + OffloadPolicy, +) +from torch.distributed.fsdp._fully_shard._fsdp_common import DDPMeshInfo +from torch.distributed.fsdp._fully_shard._fsdp_init import ( + _apply_to_module, + _get_device_from_mesh, + _get_mesh_info, + _get_modules_and_states, + _init_default_mesh, + _init_param_group, + _validate_mesh as _validate_mesh_common, + _validate_module as _validate_module_common, +) +from torch.distributed.fsdp._fully_shard._fsdp_state import FSDPState, FSDPStateContext +from torch.distributed.fsdp._fully_shard._fully_shard import ( + _unimplemented_deepcopy, + FSDPModule, +) + +from .contract import _get_registry, contract + + +if TYPE_CHECKING: + from torch.distributed.fsdp._fully_shard._fsdp_api import DataParallelMeshDims + from torch.distributed.tensor import DeviceMesh + + +cls_to_replicate_cls: dict[type, type] = {} + +logger = logging.getLogger("torch.distributed._composable.replicate_with_fsdp") + + +class _ReplicateStateContext(FSDPStateContext["_ReplicateState"]): + """ + State shared across Replicate states. + + This is a typed subclass of FSDPStateContext parameterized with _ReplicateState, + providing correct type annotations (e.g., all_states: list[_ReplicateState]). + It also allows call sites to differentiate between Replicate and FSDP contexts + via isinstance checks if needed. + """ + + +def _get_module_replicate_state(module: nn.Module) -> _ReplicateState | None: + state = _get_module_state(module) + if isinstance(state, _ReplicateState): + return state + return None + + +class _ReplicateState(FSDPState): + _state_name: str = "Replicate" + + def __init__(self) -> None: + super().__init__() + self._state_ctx = _ReplicateStateContext() + + def _get_state_for_module(self, module: nn.Module) -> FSDPState | None: + return _get_module_replicate_state(module) + + def init( + self, + modules: tuple[nn.Module, ...], + device: torch.device, + mp_policy: MixedPrecisionPolicy, + auto_reshard_after_forward: bool = False, + ) -> None: + super().init(modules, device, mp_policy, auto_reshard_after_forward) + + +@overload +# pyrefly: ignore [inconsistent-overload] +def replicate( + module: nn.Module, + *, + mesh: DeviceMesh | None = ..., + mp_policy: MixedPrecisionPolicy = ..., + offload_policy: OffloadPolicy = ..., + ignored_params: set[nn.Parameter] | None = ..., + dp_mesh_dims: DataParallelMeshDims | None = ..., +) -> ReplicateModule: ... + + +@overload +# pyrefly: ignore [inconsistent-overload] +def replicate( + module: list[nn.Module], + *, + mesh: DeviceMesh | None = ..., + mp_policy: MixedPrecisionPolicy = ..., + offload_policy: OffloadPolicy = ..., + ignored_params: set[nn.Parameter] | None = ..., + dp_mesh_dims: DataParallelMeshDims | None = ..., +) -> list[ReplicateModule]: ... + + +@contract(state_cls=_ReplicateState) # type: ignore[misc] +def replicate( + module: nn.Module, + *, + mesh: DeviceMesh | None = None, + mp_policy: MixedPrecisionPolicy = MixedPrecisionPolicy(), + offload_policy: OffloadPolicy = OffloadPolicy(), + ignored_params: set[nn.Parameter] | None = None, + dp_mesh_dims: DataParallelMeshDims | None = None, +): + r"""Replicates a module + + Args: + module (torch.nn.Module): module to replicate + + Example:: + >>> # xdoctest: +REQUIRES(module:torch._C._distributed_c10d) + >>> module = nn.Linear(3, 3) + >>> replicate(module) + """ + torch._C._log_api_usage_once("torch.distributed._composable.replicate_with_fsdp") + _validate_module(module) + mesh = mesh or _init_default_mesh(mesh_dim_names=("replicate",)) + if dp_mesh_dims is not None: + _validate_mesh_common(mesh, dp_mesh_dims) + mesh_info = _get_mesh_info(mesh, dp_mesh_dims) + if not isinstance(mesh_info, DDPMeshInfo): + raise ValueError( + "replicate() with dp_mesh_dims requires replicate-only " + "dims (no shard dims). Use fully_shard() for sharding." + ) + else: + _validate_mesh(mesh) + mesh_info = DDPMeshInfo(mesh, replicate_mesh_dim=0) + device = _get_device_from_mesh(mesh) + # managed_modules (3rd return) and buffers (5th return) are unused: + # - managed_modules: FSDP uses this to set Dynamo-specific attributes + # (_is_fsdp_managed_module, _fsdp_use_orig_params), which replicate doesn't need + # - buffers: already moved to device by _get_modules_and_states; replicate + # doesn't need to track them separately + arg_module, modules, _, params, _ = _get_modules_and_states( + module, + device, + ignored_params, + is_composable_fn=is_composable_with_replicate, + get_state_fn=_get_module_replicate_state, + ) + state = replicate.state(modules[0]) # type: ignore[attr-defined] + state.init(modules, device, mp_policy) + + _init_param_group( + state, + params, + modules, + mesh_info, + None, # post_forward_mesh_info + device, + None, # shard_placement_fn + mp_policy, + offload_policy, + ) + + # Place Replicate leftmost for highest priority in the method resolution order + _apply_to_module( + modules, + cls_to_replicate_cls, + ReplicateModule, + "Replicate", + _unimplemented_deepcopy, + ) + return arg_module + + +class ReplicateModule(FSDPModule): + # Index in MRO where the original class is found. + # For Replicate: [Replicate, ReplicateModule, FSDPModule, Orig, ...] -> index 3 + _orig_cls_mro_index: int = 3 + + +def is_composable_with_replicate(module: nn.Module) -> bool: + registry = _get_registry(module) + if registry is None: + return True + return "fully_shard" not in registry + + +def _validate_module(module: nn.Module) -> None: + if not is_composable_with_replicate(module): + raise RuntimeError( + "Cannot apply `replicate()` on a Module already managed by `fully_shard`" + ) + _validate_module_common(module, "replicate") + + +def _validate_mesh(mesh: DeviceMesh) -> None: + if mesh.ndim != 1: + raise ValueError(f"replicate expects a 1D DeviceMesh but got {mesh}") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_composable_state.py b/.venv/lib/python3.12/site-packages/torch/distributed/_composable_state.py new file mode 100644 index 0000000000000000000000000000000000000000..a963f9af79db7b2383611ac939f77577e803a5b1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_composable_state.py @@ -0,0 +1,45 @@ +import weakref +from typing import cast + +import torch.nn as nn + + +class _State: + pass + + +_module_state_mapping: weakref.WeakKeyDictionary[ + nn.Module, weakref.ReferenceType[_State] +] = weakref.WeakKeyDictionary() + + +def _insert_module_state(module: nn.Module, state: _State) -> None: + global _module_state_mapping + if module in _module_state_mapping: + raise AssertionError(f"Inserting {module} more than once.") + _module_state_mapping[module] = weakref.ref(state) + + +def _get_module_state(module: nn.Module) -> _State | None: + """ + Return the ``_State`` in ``model``. + + Given a ``module``, this API finds out if the module is also a ``_State`` + instance or if the module is managed by a composable API. If the module + is also a ``_State``, ``module`` will be casted to ``_State` and returned. + If it is managed by a composable API, the corresponding ``_State`` will + be returned. + """ + global _module_state_mapping + if isinstance(module, _State): + return cast(_State, module) + else: + # https://github.com/pytorch/pytorch/issues/107054 + if module in _module_state_mapping: + state_ref = _module_state_mapping[module] + state = state_ref() + if state is None: + raise AssertionError("State has already been garbage collected") + return state + else: + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives.py b/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives.py new file mode 100644 index 0000000000000000000000000000000000000000..c30b34a70d4b8600a1ab93d1766584ca63544284 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives.py @@ -0,0 +1,2096 @@ +# mypy: allow-untyped-defs +import contextlib +import math +import sys +import warnings +from collections.abc import Callable, Sequence +from typing import Any, cast, TYPE_CHECKING + +import torch +import torch.compiler.config +import torch.distributed as dist +import torch.distributed.distributed_c10d as c10d +from torch._utils import _maybe_view_chunk_cat +from torch.distributed.device_mesh import DeviceMesh +from torch.fx.experimental.proxy_tensor import get_proxy_mode + +from . import _functional_collectives_impl as fun_col_impl + + +try: + from torch.utils._cxx_pytree import tree_map_only +except ImportError: + from torch.utils._pytree import tree_map_only # type: ignore[no-redef] + + +try: + from torch.compiler import is_dynamo_compiling as is_torchdynamo_compiling +except Exception: + warnings.warn( + "Unable to import torchdynamo util `is_torchdynamo_compiling`, so won't support torchdynamo correctly", + stacklevel=2, + ) + + def is_torchdynamo_compiling(): # type: ignore[misc] + return False + + +""" +New traceable, functional collectives. +RFC: https://github.com/pytorch/pytorch/issues/93173 + + compiler: trace these ops with plain-old-data schemas, then choose how to lower them. + eager: execute these 'functional' ops which in eager return AsyncCollectiveTensor subclasses, + automatically calling .wait() on underlying/hidden async 'work' obj only when fed to + a downstream op. + +Issues: +* Where should these ops live? Couldn't `import torch` if putting these ops in existing torch.distributed files +* Proper support for eager requires inplace ops. We should explore having it as an option for the API. +""" + +""" +Functional collectives are asynchronous only and we perform implicit stream synchronization +on behalf of the user. + +We use AsyncCollectiveTensor to wrap the result tensor of a collective and it lets us witness +first usage of the tensor and insert cross stream sync at the right place. + +The above are the easy bits, the hard one is how we match the Work object returned by +c10d and the tensor AsyncCollectiveTensor wraps. We alloc the tensor inside the collective +op implementation (see ``clone()`` call in ``_all_reduce``) and then it's handled by the +dispatcher which might call other implementations that are allowed to change the returned +tensor - even return a tensor with a different shape (see ``torch.vmap``). + +This means the caller of our ops receives a Tensor that is not guaranteed to be the same +allocated by our implementations and that makes pairing The AsyncTensor to the original +tensor a lot harder. This pairing is needed so we can lookup the Work object to use. + +Originally, we tried WeakKeyDictionary to map from Tensor to Work, but because Tensor's +identity is not stable across dispatch, the op caller would end up with a different Tensor +instance that would not match any in the dictionary. + +With Tensor identity out of the question, we decided use the tensor data pointer, which +should be stable across all the Tensor changes done during dispatch. + +We have a dictionary of tensor::data_ptr -> Work that we insert right after we call into c10d. + +We use this dictionary when AsyncCollectiveTensor is used to invoke Work::wait() + +Finally, we setup a finalizer against the tensor wrapper to observe it getting collected so we +can clean up stale entries in the dictionary. + +To eliminate the possibility of races we have a global version counter that is used by the finalizer. + +As a wise man said once: Don't cross the streams (https://www.youtube.com/watch?v=wyKQe_i9yyo) + +""" + +""" +Functional collectives can accept any of these types to describe the ranks participating in collectives. + +The different types will be desugared to a canonical format +""" +RANK_TYPES = ( + list[int] + | list[list[int]] + | dist.ProcessGroup + | DeviceMesh + | tuple["dist.tensor.DeviceMesh", int] + | c10d.GroupName +) + + +from torch._utils import _chunk_or_narrow_cat + + +""" +User facing APIs for functional collectives +------------------------------------------- + +These apis are called by user code and expected to work both in eager execution and compilation, +but there are significant differences to how the two modes are implemented underneath. + +Eager execution is 'optimized' using a tensor subclass that schedules the synchronization (via wait_tensor() op) +just before the tensor is first used. Compiled tracing currently relies on the compiler to perform this optimization, +and cannot yet correctly trace the AsyncTensor wrapper class. In the future, these paths may be unified +if sufficient subclass support is added in dynamo. + +Example: all_reduce is an entrypoint API, and other collectives follow a similar pattern. + +Here's how it works under torch.compile/dynamo: +all_reduce(...) + |--> _expand_group(...) - desugars processgroup into canonical/traceable format + |--> c10d_functional.all_reduce(...) - dynamo captures this op call, doesn't trace deeper + |--> _maybe_wrap_tensor(...) - wait_tensor() op is immediately called, no AsyncTensor subclass needed + +And under eager execution: +all_reduce(...) + |--> _expand_group(...) - same as above, but less critical for eager + |--> c10d_functional.all_reduce(...) - dispatches to real kernel OR records op in trace + |--> _maybe_wrap_tensor(...) - AsyncTensor wrapper applied to returned tensor, + which issues wait_tensor() at the time of first use +""" + + +def wait_tensor(tensor): + """ + Wait on a tensor returned by the collectives ops. + + Waiting follows device semantics, which means blocking on CPU and synchronizing streams on CUDA. + """ + return torch.ops._c10d_functional.wait_tensor(tensor) # type: ignore[attr-defined] + + +def broadcast(self: torch.Tensor, src: int, group: RANK_TYPES, tag: str = ""): + """ + Broadcasts the tensor to all processes in the given process group. + + Args: + src (int): Source rank + group (ProcessGroup or List[int]): The process group to work on. + tag (str, optional): A unique identifier for the collective. Default: empty string + """ + group = _resolve_group(group, tag) + tensor = torch.ops._c10d_functional.broadcast( + self, src, _group_or_group_name(group) + ) + return _maybe_wrap_tensor(tensor) + + +def all_reduce(self: torch.Tensor, reduceOp: str, group: RANK_TYPES, tag: str = ""): + """ + Reduces the tensor data across all machines in such a way that all get + the final result. + + The input tensor is left unmodified. + + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + tensor = torch.ops._c10d_functional.all_reduce( + self, reduceOp.lower(), _group_or_group_name(group) + ) + return _maybe_wrap_tensor(tensor) + + +def all_gather_single( + self: torch.Tensor, + gather_dim: int, + group: RANK_TYPES, + tag: str = "", +) -> torch.Tensor: + """ + Gather tensor data across from all machines and concatenate over ``gather_dim``. + + Note that it currently only supports gather_dim = 0. + + The input tensor is left unmodified. + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + group_size = c10d._get_group_size_by_name(group) + tensor = torch.ops._c10d_functional.all_gather_into_tensor( + self, group_size, _group_or_group_name(group) + ) + res = _maybe_wrap_tensor(tensor) + if gather_dim != 0: + # Check if _maybe_view_chunk_cat can use the view optimization. + # If not, it will use torch.cat which needs the data anyway, so + # wait early to avoid AsyncCollectiveTensor dispatch overhead. + if isinstance(res, AsyncCollectiveTensor): + shape = list(res.shape) + numel_between = math.prod(shape[1:gather_dim]) if gather_dim > 1 else 1 + can_use_view = shape[0] == group_size and numel_between == 1 + if not can_use_view: + res = res.wait() + res = _maybe_view_chunk_cat(res, group_size, gather_dim) + return res + + +def all_gather_single_autograd( + self: torch.Tensor, + gather_dim: int, + group: RANK_TYPES, + tag: str = "", +): + """ + Gather tensor data across from all machines and concatenate over ``gather_dim``. + + Note that it currently only supports gather_dim = 0. + + This function is the same as all_gather_single but will propagate the + backwards gradient across workers. + + See all_gather_single for more details on usage. + """ + # The base all_gather_single now has autograd support, so we can just call it + return all_gather_single(self, gather_dim, group, tag) + + +def reduce_scatter_single( + self: torch.Tensor, + reduceOp: str, + scatter_dim: int, + group: RANK_TYPES, + tag: str = "", +): + """ + Reduces the tensor data across all machines in such a way that all get + the final result, then scatter the results to corresponding ranks. + + + The input tensor is left unmodified. + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + group_size = c10d._get_group_size_by_name(group) + + if self.size(scatter_dim) % group_size != 0: + raise AssertionError( + f"input dimension 0 ({self.size(0)} must be a multiple of group_size {group_size})" + ) + if scatter_dim != 0: + self = _chunk_or_narrow_cat(self, group_size, narrow_dim=scatter_dim, cat_dim=0) + + tensor = torch.ops._c10d_functional.reduce_scatter_tensor( + self, + reduceOp.lower(), + group_size, + _group_or_group_name(group), + ) + res = _maybe_wrap_tensor(tensor) + return res + + +def reduce_scatter_single_autograd( + self: torch.Tensor, + reduceOp: str, + scatter_dim: int, + group: RANK_TYPES, + tag: str = "", +): + """ + Reduces the tensor data across all machines in such a way that all get + the final result, then scatter the results to corresponding ranks. + + This function is the same as reduce_scatter_single but will propagate the + backwards gradient across workers. + + Currently only the "sum" reduceOp is supported. + + See reduce_scatter_single for more details on usage. + """ + # The base reduce_scatter_single now has autograd support, so we can just call it + return reduce_scatter_single(self, reduceOp, scatter_dim, group, tag) + + +# The deprecated *_tensor aliases warn only outside compile: the +# typing_extensions @deprecated wrapper lives in a Dynamo-skipped module, and a +# raw warnings.warn is an untraceable builtin, either of which breaks +# fullgraph=True tracing of these still-supported entry points. Guarding on +# is_compiling() keeps the eager-time deprecation while letting Dynamo trace +# straight through to the _single implementation. +def all_gather_tensor( + self: torch.Tensor, + gather_dim: int, + group: RANK_TYPES, + tag: str = "", +) -> torch.Tensor: + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.all_gather_tensor` is deprecated. " + "Please use `torch.distributed._functional_collectives.all_gather_single` instead.", + FutureWarning, + stacklevel=2, + ) + return all_gather_single(self, gather_dim, group, tag) + + +def all_gather_tensor_autograd( + self: torch.Tensor, + gather_dim: int, + group: RANK_TYPES, + tag: str = "", +): + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.all_gather_tensor_autograd` is deprecated. " + "Please use `torch.distributed._functional_collectives.all_gather_single_autograd` instead.", + FutureWarning, + stacklevel=2, + ) + return all_gather_single_autograd(self, gather_dim, group, tag) + + +def reduce_scatter_tensor( + self: torch.Tensor, + reduceOp: str, + scatter_dim: int, + group: RANK_TYPES, + tag: str = "", +): + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.reduce_scatter_tensor` is deprecated. " + "Please use `torch.distributed._functional_collectives.reduce_scatter_single` instead.", + FutureWarning, + stacklevel=2, + ) + return reduce_scatter_single(self, reduceOp, scatter_dim, group, tag) + + +def reduce_scatter_tensor_autograd( + self: torch.Tensor, + reduceOp: str, + scatter_dim: int, + group: RANK_TYPES, + tag: str = "", +): + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.reduce_scatter_tensor_autograd` is deprecated. " + "Please use `torch.distributed._functional_collectives.reduce_scatter_single_autograd` instead.", + FutureWarning, + stacklevel=2, + ) + return reduce_scatter_single_autograd(self, reduceOp, scatter_dim, group, tag) + + +def all_reduce_coalesced( + self: list[torch.Tensor], reduceOp: str, group: RANK_TYPES, tag: str = "" +) -> list[torch.Tensor]: + """ + Reduces a list of tensors across all machines in such a way that all get + the final result. + + The all tensors in the input list are left unmodified. + + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + tensor_list = torch.ops._c10d_functional.all_reduce_coalesced( # type: ignore[attr-defined] + self, + reduceOp.lower(), + _group_or_group_name(group), + ) + return list(map(_maybe_wrap_tensor, tensor_list)) + + +def all_gather_single_coalesced( + self: list[torch.Tensor], group: RANK_TYPES, tag: str = "" +) -> list[torch.Tensor]: + """ + Gather a list of tensors across from all machines. + + Note that it currently only supports gather_dim = 0. + + The input tensor is left unmodified. + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + group_size = c10d._get_group_size_by_name(group) + tensor_list = torch.ops._c10d_functional.all_gather_into_tensor_coalesced( # type: ignore[attr-defined] + self, + group_size, + _group_or_group_name(group), + ) + return list(map(_maybe_wrap_tensor, tensor_list)) + + +def reduce_scatter_single_coalesced( + inputs: list[torch.Tensor], + reduceOp: str, + scatter_dim: list[int], + group: RANK_TYPES, + tag: str = "", +) -> list[torch.Tensor]: + """ + Reduces a list of tensors across all machines in such a way that all get + the final result, then scatter the results to corresponding ranks. + + The input tensors are left unmodified. + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + group = _resolve_group(group, tag) + group_size = c10d._get_group_size_by_name(group) + + if len(scatter_dim) != len(inputs): + raise AssertionError( + f"Length of scatter_dim ({len(scatter_dim)}) must equal length of inputs ({len(inputs)})" + ) + for idx, (dim, tensor) in enumerate(zip(scatter_dim, inputs)): + if tensor.size(dim) % group_size != 0: + raise AssertionError( + f"input dimension {dim} ({tensor.size(dim)} must be a multiple of group_size {group_size} for tensor at index {idx}" + ) + if dim != 0: + tensor_list = torch.chunk(tensor, group_size, dim=dim) + inputs[idx] = torch.cat(tensor_list) + + tensor_list = torch.ops._c10d_functional.reduce_scatter_tensor_coalesced( # type: ignore[attr-defined] + inputs, + reduceOp.lower(), + group_size, + _group_or_group_name(group), + ) + + return list(map(_maybe_wrap_tensor, tensor_list)) + + +# Guarded warning rather than the @deprecated wrapper so Dynamo can trace +# through to the _single_coalesced implementation; see the non-coalesced +# aliases above for the rationale. +def all_gather_into_tensor_coalesced( + self: list[torch.Tensor], group: RANK_TYPES, tag: str = "" +) -> list[torch.Tensor]: + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.all_gather_into_tensor_coalesced` is deprecated. " + "Please use `torch.distributed._functional_collectives.all_gather_single_coalesced` instead.", + FutureWarning, + stacklevel=2, + ) + return all_gather_single_coalesced(self, group, tag) + + +def reduce_scatter_tensor_coalesced( + inputs: list[torch.Tensor], + reduceOp: str, + scatter_dim: list[int], + group: RANK_TYPES, + tag: str = "", +) -> list[torch.Tensor]: + if not torch.compiler.is_compiling(): + warnings.warn( + "`torch.distributed._functional_collectives.reduce_scatter_tensor_coalesced` is deprecated. " + "Please use `torch.distributed._functional_collectives.reduce_scatter_single_coalesced` instead.", + FutureWarning, + stacklevel=2, + ) + return reduce_scatter_single_coalesced(inputs, reduceOp, scatter_dim, group, tag) + + +# This is a bit unsafe: it checks if the first argument in the schema reports as a non-mutable alias. +# Today, this maps 1:1 with "aten ops that are views". +def _is_view_op(tgt): + if not isinstance(tgt, torch._ops.OpOverload): + raise AssertionError(f"Expected torch._ops.OpOverload, got {type(tgt)}") + # Don't apply the view optimization to any `CompositeImplicitAutograd` ops. + # See issue: https://github.com/pytorch/pytorch/issues/133421 + if torch._C._dispatch_has_kernel_for_dispatch_key( + tgt.name(), torch.DispatchKey.CompositeImplicitAutograd + ): + return False + schema = tgt._schema + if len(schema.arguments) > 0: + first_arg = schema.arguments[0] + # check if op is a view + return first_arg.alias_info is not None and not first_arg.alias_info.is_write + + +def all_to_all_single( + self: torch.Tensor, + output_split_sizes: list[int] | None, + input_split_sizes: list[int] | None, + group: RANK_TYPES, + tag: str = "", +) -> torch.Tensor: + """ + Each process splits input tensor and then scatters the split list + to all processes in a group. Then concatenate the received tensors from all + the processes in the group and return single output tensor. + + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one dimension of the DeviceMesh + + :: N.B. If you pass a PG or a 1D list to perform a MPMD collective, the compiler won't be able to recover + that information and perform collective algebraic optimization. Use other forms of input for that. + """ + if output_split_sizes is not None: + if not all( + isinstance(size, (int, torch.SymInt)) for size in output_split_sizes + ): + raise AssertionError( + f"All output_split_sizes must be int or SymInt, got {output_split_sizes}" + ) + if input_split_sizes is not None: + if not all(isinstance(size, (int, torch.SymInt)) for size in input_split_sizes): + raise AssertionError( + f"All input_split_sizes must be int or SymInt, got {input_split_sizes}" + ) + group = _resolve_group(group, tag) + group_size = c10d._get_group_size_by_name(group) + if output_split_sizes is None or input_split_sizes is None: + if not (output_split_sizes is None and input_split_sizes is None): + raise AssertionError( + "output_split_sizes and input_split_sizes must either be " + "specified together or both set to None" + ) + output_split_sizes = [self.shape[0] // group_size] * group_size + input_split_sizes = output_split_sizes + tensor = torch.ops._c10d_functional.all_to_all_single( # type: ignore[attr-defined] + self, + output_split_sizes, + input_split_sizes, + _group_or_group_name(group), + ) + return _maybe_wrap_tensor(tensor) + + +def all_to_all_single_autograd( + self: torch.Tensor, + output_split_sizes: list[int] | None, + input_split_sizes: list[int] | None, + group: RANK_TYPES, + tag: str = "", +) -> torch.Tensor: + """ + Same as all_to_all_single but supports autograd. + """ + # The base all_to_all_single now has autograd support, so we can just call it + return all_to_all_single(self, output_split_sizes, input_split_sizes, group, tag) + + +# ============================================================================ +# Collective Autograd Functions / Custom Ops +# ============================================================================ + + +def wait_tensor_backward(ctx, grad_output: torch.Tensor): + """ + Backward for wait_tensor: identity (no-op). + Wait is just a synchronization primitive, so gradient flows through unchanged. + + Args: + ctx: Context object + grad_output: Gradient from downstream operations + + Returns: + Gradient unchanged (identity) + """ + return grad_output + + +def wait_tensor_setup_context(ctx, inputs, output): + """ + Setup context for wait_tensor backward. + Args: + ctx: Context object to save state for backward + inputs: Tuple of (tensor,) + output: Output from forward pass + """ + return + + +torch.library.register_autograd( + "_c10d_functional::wait_tensor", + wait_tensor_backward, + setup_context=wait_tensor_setup_context, +) + + +def all_reduce_backward(ctx, grad_output: torch.Tensor): + """ + Backward for all_reduce: all_reduce with same reduce_op. + Forward aggregates tensors, backward aggregates gradients. + + Args: + ctx: Context object + grad_output: Gradient from downstream operations + + Returns: + Tuple of (grad_input, grad_group_name, grad_reduce_op) + grad_group_name and grad_reduce_op are None (not differentiable) + """ + group_name = ctx.group_name + reduce_op = ctx.reduce_op + + # Only linear reductions have a well-defined all_reduce backward: for both + # 'sum' and 'avg' the gradient is an all_reduce of grad_output with the same + # op (grad wrt each rank's input is the same reduction of the per-rank + # grad_outputs). Nonlinear ops (min/max/product/premul_sum) do not. + if reduce_op not in ("sum", "avg"): + raise RuntimeError( + f"all_reduce backward only supports 'sum' and 'avg' reductions, got '{reduce_op}'" + ) + + # Backward does all_reduce with the same reduce_op + output = torch.ops._c10d_functional.all_reduce( + grad_output.contiguous(), reduce_op, group_name + ) + return wait_tensor(output), None, None + + +def all_reduce_setup_context(ctx, inputs, output): + """ + Setup context for all_reduce backward. + Args: + ctx: Context object to save state for backward + inputs: Tuple of (input, reduce_op, group_name) + output: Output from forward pass + """ + input, reduce_op, group_name = inputs + ctx.group_name = group_name + ctx.reduce_op = reduce_op.lower() + + +torch.library.register_autograd( + "_c10d_functional::all_reduce", + all_reduce_backward, + setup_context=all_reduce_setup_context, +) + + +def all_gather_into_tensor_backward(ctx, grad_output: torch.Tensor): + """ + Backward for all_gather_into_tensor: reduce_scatter with sum. + + Forward gathers tensors from all ranks, backward scatters gradients back + with sum reduction. + + Args: + ctx: Context object with group_name and group_size + grad_output: Gradient from downstream operations + + Returns: + Tuple of (grad_input, grad_group_size, grad_group_name) + grad_group_size and grad_group_name are None (not differentiable) + """ + group_name = ctx.group_name + group_size = ctx.group_size + + # Backward is reduce_scatter with sum + output = torch.ops._c10d_functional.reduce_scatter_tensor( + grad_output.contiguous(), + "sum", + group_size, + group_name, + ) + return wait_tensor(output), None, None + + +def all_gather_into_tensor_setup_context(ctx, inputs, output): + """ + Setup context for all_gather_into_tensor backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (input, group_size, group_name) + output: Output from forward pass + """ + input, group_size, group_name = inputs + ctx.group_name = group_name + ctx.group_size = group_size + + +torch.library.register_autograd( + "_c10d_functional::all_gather_into_tensor", + all_gather_into_tensor_backward, + setup_context=all_gather_into_tensor_setup_context, +) + + +def reduce_scatter_tensor_backward(ctx, grad_output: torch.Tensor): + """ + Backward for reduce_scatter_tensor: all_gather. + + Forward reduces and scatters tensors to ranks, backward gathers gradients + from all ranks. + + Args: + ctx: Context object with group_name, group_size, and reduce_op + grad_output: Gradient from downstream operations + + Returns: + Tuple of (grad_input, grad_reduce_op, grad_group_size, grad_group_name) + grad_reduce_op, grad_group_size, grad_group_name are None (not differentiable) + """ + group_name = ctx.group_name + group_size = ctx.group_size + reduce_op = ctx.reduce_op + + # Lazy validation: check reduce_op only when backward is called + if reduce_op != "sum": + raise RuntimeError( + f"reduce_scatter_tensor backward only supports 'sum' reduction, got '{reduce_op}'" + ) + + # Backward is all_gather + output = torch.ops._c10d_functional.all_gather_into_tensor( + grad_output.contiguous(), + group_size, + group_name, + ) + return wait_tensor(output), None, None, None + + +def reduce_scatter_tensor_setup_context(ctx, inputs, output): + """ + Setup context for reduce_scatter_tensor backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (input, reduce_op, group_size, group_name) + output: Output from forward pass + """ + input, reduce_op, group_size, group_name = inputs + ctx.group_name = group_name + ctx.group_size = group_size + ctx.reduce_op = reduce_op.lower() + + +torch.library.register_autograd( + "_c10d_functional::reduce_scatter_tensor", + reduce_scatter_tensor_backward, + setup_context=reduce_scatter_tensor_setup_context, +) + + +def all_to_all_single_backward(ctx, grad_output: torch.Tensor): + """ + Backward for all_to_all_single: all_to_all with reversed split sizes. + + Forward does all-to-all with specified split sizes, backward reverses them. + + Args: + ctx: Context object with group_name, output_split_sizes, and input_split_sizes + grad_output: Gradient from downstream operations + + Returns: + Tuple of (grad_input, grad_output_split_sizes, grad_input_split_sizes, grad_group_name) + All except grad_input are None (not differentiable) + """ + group_name = ctx.group_name + output_split_sizes = ctx.output_split_sizes + input_split_sizes = ctx.input_split_sizes + + # Backward is all_to_all with reversed split sizes + output = torch.ops._c10d_functional.all_to_all_single( + grad_output.contiguous(), + input_split_sizes, # Reversed + output_split_sizes, # Reversed + group_name, + ) + return wait_tensor(output), None, None, None + + +def all_to_all_single_setup_context(ctx, inputs, output): + """ + Setup context for all_to_all_single backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (input, output_split_sizes, input_split_sizes, group_name) + output: Output from forward pass + """ + input, output_split_sizes, input_split_sizes, group_name = inputs + ctx.group_name = group_name + ctx.output_split_sizes = output_split_sizes + ctx.input_split_sizes = input_split_sizes + + +torch.library.register_autograd( + "_c10d_functional::all_to_all_single", + all_to_all_single_backward, + setup_context=all_to_all_single_setup_context, +) + +# Backward compatibility: register autograd for _c10d_functional_autograd ops +# These ops redirect to _c10d_functional but need their own autograd registration +torch.library.register_autograd( + "_c10d_functional_autograd::all_gather_into_tensor", + all_gather_into_tensor_backward, + setup_context=all_gather_into_tensor_setup_context, +) + +torch.library.register_autograd( + "_c10d_functional_autograd::reduce_scatter_tensor", + reduce_scatter_tensor_backward, + setup_context=reduce_scatter_tensor_setup_context, +) + +torch.library.register_autograd( + "_c10d_functional_autograd::all_to_all_single", + all_to_all_single_backward, + setup_context=all_to_all_single_setup_context, +) + + +def all_reduce_coalesced_backward(ctx, grad_outputs: list[torch.Tensor]): + """ + Backward for all_reduce_coalesced: all_reduce each gradient. + + Forward aggregates tensors, backward aggregates gradients. + + Args: + ctx: Context object with group_name and reduce_op + grad_outputs: Gradients from downstream operations (one per input tensor) + + Returns: + Tuple of (grad_inputs..., grad_reduce_op, grad_group_name) + grad_reduce_op and grad_group_name are None (not differentiable) + """ + group_name = ctx.group_name + reduce_op = ctx.reduce_op + + if reduce_op != "sum": + raise RuntimeError( + f"all_reduce_coalesced backward only supports 'sum' reduction, got '{reduce_op}'" + ) + + # Backward does all_reduce on list of gradients + grad_inputs = torch.ops._c10d_functional.all_reduce_coalesced( + [grad_output.contiguous() for grad_output in grad_outputs], + reduce_op, + group_name, + ) + return (list(map(wait_tensor, grad_inputs)), None, None) + + +def all_reduce_coalesced_setup_context(ctx, inputs, output): + """ + Setup context for all_reduce_coalesced backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (tensor_list, reduce_op, group_name) + output: Output from forward pass + """ + tensor_list, reduce_op, group_name = inputs + ctx.group_name = group_name + ctx.reduce_op = reduce_op.lower() + + +torch.library.register_autograd( + "_c10d_functional::all_reduce_coalesced", + all_reduce_coalesced_backward, + setup_context=all_reduce_coalesced_setup_context, +) + + +def all_gather_into_tensor_coalesced_backward(ctx, grad_outputs: list[torch.Tensor]): + """ + Backward for all_gather_into_tensor_coalesced: reduce_scatter each gradient. + + Forward gathers tensors from all ranks, backward scatters gradients back + with sum reduction. + + Args: + ctx: Context object with group_name and group_size + grad_outputs: Gradients from downstream operations (one per input tensor) + + Returns: + Tuple of (grad_inputs..., grad_group_size, grad_group_name) + grad_group_size and grad_group_name are None (not differentiable) + """ + group_name = ctx.group_name + group_size = ctx.group_size + + # Backward does reduce_scatter on list of gradients + grad_inputs = torch.ops._c10d_functional.reduce_scatter_tensor_coalesced( + [grad_output.contiguous() for grad_output in grad_outputs], + "sum", + group_size, + group_name, + ) + return (list(map(wait_tensor, grad_inputs)), None, None) + + +def all_gather_into_tensor_coalesced_setup_context(ctx, inputs, output): + """ + Setup context for all_gather_into_tensor_coalesced backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (tensor_list, group_size, group_name) + output: Output from forward pass + """ + tensor_list, group_size, group_name = inputs + ctx.group_name = group_name + ctx.group_size = group_size + + +torch.library.register_autograd( + "_c10d_functional::all_gather_into_tensor_coalesced", + all_gather_into_tensor_coalesced_backward, + setup_context=all_gather_into_tensor_coalesced_setup_context, +) + + +def reduce_scatter_tensor_coalesced_backward(ctx, grad_outputs: list[torch.Tensor]): + """ + Backward for reduce_scatter_tensor_coalesced: all_gather each gradient. + + Forward reduces and scatters tensors to ranks, backward gathers gradients + from all ranks. + + Args: + ctx: Context object with group_name, group_size, and reduce_op + grad_outputs: Gradients from downstream operations (one per input tensor) + + Returns: + Tuple of (grad_inputs..., grad_reduce_op, grad_group_size, grad_group_name) + grad_reduce_op, grad_group_size, grad_group_name are None (not differentiable) + """ + group_name = ctx.group_name + group_size = ctx.group_size + reduce_op = ctx.reduce_op + + # Lazy validation: check reduce_op only when backward is called + if reduce_op != "sum": + raise RuntimeError( + f"reduce_scatter_tensor_coalesced backward only supports 'sum' reduction, got '{reduce_op}'" + ) + + # Backward does all_gather on list of gradients + grad_inputs = torch.ops._c10d_functional.all_gather_into_tensor_coalesced( + [grad_output.contiguous() for grad_output in grad_outputs], + group_size, + group_name, + ) + return (list(map(wait_tensor, grad_inputs)), None, None, None) + + +def reduce_scatter_tensor_coalesced_setup_context(ctx, inputs, output): + """ + Setup context for reduce_scatter_tensor_coalesced backward. + + Args: + ctx: Context object to save state for backward + inputs: Tuple of (tensor_list, reduce_op, group_size, group_name) + output: Output from forward pass + """ + tensor_list, reduce_op, group_size, group_name = inputs + ctx.group_name = group_name + ctx.group_size = group_size + ctx.reduce_op = reduce_op.lower() + + +torch.library.register_autograd( + "_c10d_functional::reduce_scatter_tensor_coalesced", + reduce_scatter_tensor_coalesced_backward, + setup_context=reduce_scatter_tensor_coalesced_setup_context, +) + + +def permute_tensor( + self: torch.Tensor, + src_dst: list[int], + group: RANK_TYPES, + tag: str = "", +) -> torch.Tensor: + """ + Permutes the elements of the tensor according to the given source/destination pairs. `src_dst` should + be defined such that src_dst[m] == n means m sends to n. + + Group can be one of: + List[int]: ranks participating in the collective. + List[List[int]]: 2D mesh of ranks taking part of this collective in MPMD. + ProcessGroup: Will perform a collective using the ranks and tag of the PG. + DeviceMesh: Do a SPMD collective over all ranks of the mesh + (DeviceMesh, int): Do a MPMD collective over one + """ + t, rankset, group_size = _expand_group(group, tag) + local_pg = c10d._find_or_create_pg_by_ranks_and_tag(t, rankset, group_size) + + output_split_sizes = [0] * group_size + input_split_sizes = [0] * group_size + for src, dst in enumerate(src_dst): + if src == dist.get_rank(local_pg): + input_split_sizes[dst] = self.numel() + if dst == dist.get_rank(local_pg): + output_split_sizes[src] = self.numel() + + return all_to_all_single(self, output_split_sizes, input_split_sizes, group, tag) + + +class AsyncCollectiveTensor(torch.Tensor): + r""" + A Tensor wrapper subclass that is used to trigger a call to wait + prior to first use of the underlying tensor. + Use it inside functional collective pytorch wrappers like the following: + def functional_collective(self, group, tag): + tag, rankset, group_size = _expand_group(group, tag) + tensor = torch.ops.c10d_functional.{collective}(self, tag, rankset, group_size) + return _maybe_wrap_tensor(tensor) + """ + + elem: torch.Tensor + completed: bool + + __slots__ = ["elem", "completed"] + + @staticmethod + def __new__(cls, elem: torch.Tensor): + r = torch.Tensor._make_wrapper_subclass( + cls, + elem.size(), + strides=elem.stride(), + storage_offset=elem.storage_offset(), + dtype=elem.dtype, + layout=elem.layout, + device=elem.device, + requires_grad=elem.requires_grad, + ) + r.elem = elem + r.completed = False + return r + + def __tensor_flatten__(self): + return ["elem"], None + + def tolist(self): + return self.trigger_wait().tolist() + + @staticmethod + def __tensor_unflatten__(inner_tensors, meta, outer_size, outer_stride): + if meta is not None: + raise AssertionError( + "meta must be None for AsyncCollectiveTensor unflatten" + ) + elem = inner_tensors["elem"] + return AsyncCollectiveTensor(elem) + + def __coerce_same_metadata_as_tangent__( + self, expected_metadata: Any, expected_type: type | None = None + ): + if expected_type is not torch.Tensor: + return None + + return self.trigger_wait() + + def __repr__(self) -> str: # type: ignore[override] + return f"AsyncCollectiveTensor({self.trigger_wait()})" + + def trigger_wait(self): + if not self.completed: + out = wait_tensor(self.elem) + self.completed = True + return out + else: + return self.elem + + def wait(self) -> torch.Tensor: + return wait_tensor(self.elem) + + def _get_acs_underlying_tensor(self): + """This method enables _functional_collectives_impl to test if a tensor is an ACS""" + return self.elem + + @classmethod + def __torch_dispatch__(cls, func, types, args=(), kwargs=None): # type: ignore[override] + if func is torch.ops.aten.view.default: + # Fast handle aten.view as a lot of view related op goes to aten.view + # eventually, this avoids pytree slowdown + + res = func(args[0].elem, args[1]) + wrapper_res = AsyncCollectiveTensor(res) + return wrapper_res + + is_view_op = _is_view_op(func) + + def unwrap(e: AsyncCollectiveTensor): + # wait_tensor is idepotent and will do stream sync only once + if not is_view_op: + return e.trigger_wait() + return e.elem + + def wrap(e: torch.Tensor): + # wait_tensor is idepotent and will do stream sync only once + if isinstance(e, AsyncCollectiveTensor): + raise AssertionError( + "Cannot wrap an AsyncCollectiveTensor inside another AsyncCollectiveTensor" + ) + res = AsyncCollectiveTensor(e) + return res + + unwrapped_args = tree_map_only(AsyncCollectiveTensor, unwrap, args) + unwrapped_kwargs = tree_map_only(AsyncCollectiveTensor, unwrap, kwargs) + + # we don't wrap the result as it doesn't need to be waited on. + out = func(*unwrapped_args, **unwrapped_kwargs) + + # View ops don't require a sync, so we should re-wrap the outputs. + if is_view_op: + out = tree_map_only(torch.Tensor, wrap, out) + + return out + + def numpy(self): # type: ignore[override] + return self.wait().numpy() + + +""" +Utils and infrastructure for tracing support +""" + + +def _expand_group(group: RANK_TYPES, tag: str = "") -> tuple[str, list[int], int]: + """ + _expand_group desugars the different RANK_TYPES types into a canonical format that is traceable. + + By having this be part of the explicit eager codepath, we avoid having to specialize behavior inside + torchdynamo and can still interoperate with processgroup objects or other untraceable forms. + """ + # had to define this hack _inside_ expand_group to avoid + # graph_break [('torch.* op returned non-Tensor int + # caused by 'cast_*` functions being treated as 'torch.*' ops (iiuc) + if TYPE_CHECKING: + + def cast_listlistint(x): + return cast(list[list[int]], x) + + def cast_listint(x): + return cast(list[int], x) + + else: + # fake cast op for use at runtime since dynamo doesn't support real cast + # also, dynamo didn't like encountering 'typing' objects () + # NotImplementedError: argument of type: + def cast_listlistint(x): + return x + + def cast_listint(x): + return x + + rankset: list[int] + if isinstance(group, list): + if isinstance(group[0], list): + nested_list = cast_listlistint(group) + rankset = [] + group_size = -1 + for rs in nested_list: + rankset.extend(rs) + if group_size != -1 and group_size != len(rs): + raise ValueError( + f"group sizes must be identical found {group_size} and {len(rs)}" + ) + group_size = len(rs) + else: + rankset = cast_listint(group) + group_size = len(rankset) + elif isinstance(group, dist.ProcessGroup): + rankset = dist.get_process_group_ranks(group) + group_size = len(rankset) + tag = tag or c10d._get_group_tag(group) + elif isinstance(group, DeviceMesh): + if group.ndim != 1: + raise AssertionError( + "Only 1D mesh is supported, pass in (DeviceMesh, int) together if mesh > 1D" + ) + pg = group.get_group() + rankset = dist.get_process_group_ranks(pg) + group_size = len(rankset) + tag = tag or c10d._get_group_tag(pg) + elif isinstance(group, tuple): + if ( + len(group) == 2 + and isinstance(group[0], DeviceMesh) + and isinstance(group[1], int) + ): + dmesh = group[0] + dim = group[1] + pg = dmesh.get_group(dim) + rankset = dist.get_process_group_ranks(pg) + group_size = len(rankset) + tag = tag or c10d._get_group_tag(pg) + else: + raise ValueError("Invalid tuple for group must be (DeviceMesh, int)") + else: + raise ValueError( + "Invalid type for group, must be one of List, Processgroup, DeviceMesh or (DeviceMesh, int)." + ) + + return (tag, rankset, group_size) + + +def _resolve_group( + group: RANK_TYPES, tag: str = "" +) -> dist.ProcessGroup | c10d.GroupName: + """ + Given group in RANK_TYPES, return a ProcessGroup or group name. + """ + # `tag` will be deprecated. See details in: + # https://github.com/pytorch/pytorch/issues/93173#issuecomment-1907095208 + if isinstance(group, dist.ProcessGroup): + return group + elif isinstance(group, str): + # In some cases Dynamo doesn't like tracing through NewType constructors + # - so use a cast instead (the actual newtype representation is + # literally the underlying type so this is fine). I haven't been able to + # reproduce it in isolation (see T247631668). + # pyrefly: ignore [redundant-cast] + group_name = cast(c10d.GroupName, group) # c10d.GroupName(group) + return group_name + elif isinstance(group, DeviceMesh): + if group.ndim != 1: + raise AssertionError( + "Only 1D mesh is supported, pass in (DeviceMesh, int) together if mesh > 1D" + ) + if torch.compiler.config.compile_on_one_rank: + return torch.ops._dtensor.mesh_get_process_group(group, 0) + return group._dim_group_names[0] + elif isinstance(group, tuple): + if ( + len(group) == 2 + and isinstance(group[0], DeviceMesh) + and isinstance(group[1], int) + ): + dmesh = group[0] + dim = group[1] + if torch.compiler.config.compile_on_one_rank: + return torch.ops._dtensor.mesh_get_process_group(dmesh, dim) + return dmesh._dim_group_names[dim] + else: + raise ValueError( + f"Invalid tuple for group must be (DeviceMesh, int). Instead got {(type(group[0]), type(group[1]))}" + ) + elif isinstance(group, list): + if not is_torchdynamo_compiling(): + warnings.warn( + "The combination of ranks + tag as process group " + "identifier has been deprecated. Please switch to " + "using ProcessGroup, DeviceMesh, or group name instead.", + FutureWarning, + stacklevel=3, + ) + return c10d._resolve_group_name_by_ranks_and_tag( + # pyrefly: ignore [redundant-cast] + cast(list[int], group), + tag, + ) + else: + raise ValueError(f"Unsupported group type: {type(group)}, {group}") + + +def _resolve_group_name(group: RANK_TYPES, tag: str = "") -> c10d.GroupName: + """ + Given group in RANK_TYPES, return the group name. + """ + group = _resolve_group(group, tag) + if isinstance(group, str): + return c10d.GroupName(group) + else: + return group.group_name + + +@torch.library.custom_op( + "_c10d_functional::_wrap_tensor_autograd", + mutates_args=(), + schema="(Tensor input) -> Tensor", +) +def _wrap_tensor_autograd(input: torch.Tensor) -> torch.Tensor: + """ + Custom op that allows autograd to propagate + from a normal Tensor to an AsyncCollectiveTensor. + + This is the low-level implementation. Users should call _maybe_wrap_tensor directly. + + Args: + input: Input tensor to wrap in AsyncCollectiveTensor + + Returns: + AsyncCollectiveTensor wrapping the input (or wait_tensor result if tracing) + """ + return AsyncCollectiveTensor(input) + + +@_wrap_tensor_autograd.register_fake +def _(input: torch.Tensor) -> torch.Tensor: + """ + Meta kernel for _wrap_tensor_autograd. + """ + return torch.empty_like(input) + + +def _wrap_tensor_autograd_backward(ctx, grad_output: torch.Tensor): + """ + Backward for _wrap_tensor_autograd: identity (no-op). + + The wrapping is just for async optimization, gradients flow through unchanged. + + Args: + ctx: Context object (unused) + grad_output: Gradient from downstream operations + + Returns: + Gradient unchanged (identity) + """ + return grad_output + + +def _wrap_tensor_autograd_setup_context(ctx, inputs, output): + """ + Setup context for _wrap_tensor_autograd backward. + + Args: + ctx: Context object to save state for backward (nothing to save) + inputs: Tuple of (input,) + output: Output from forward pass + """ + return + + +_wrap_tensor_autograd.register_autograd( + _wrap_tensor_autograd_backward, + setup_context=_wrap_tensor_autograd_setup_context, +) + + +def _are_we_tracing() -> bool: + if is_torchdynamo_compiling(): + return True + # If fake mode is turned on, we are almost definitely compiling/tracing. + if torch._C._get_dispatch_mode(torch._C._TorchDispatchModeKey.FAKE) is not None: + return True + # See Note [enable_python_dispatcher in dynamo] + if torch._C._dispatch_tls_is_dispatch_key_included( + torch._C.DispatchKey.PythonDispatcher + ): + return True + return get_proxy_mode() is not None + + +def _maybe_wrap_tensor(self) -> torch.Tensor: + if _are_we_tracing(): + return wait_tensor(self) + return _wrap_tensor_autograd(self) + + +@contextlib.contextmanager +def allow_inflight_collective_as_graph_input_ctx(value: bool = True): + """ + Context manager to temporarily set whether inflight collectives are allowed as torch.compile graph inputs. + Common use case is when the collective is issued in eager (with `async_op=True`) but waited in compiled region: + ``` + def all_reduce_eager(x): + y = x * x + req = dist.all_reduce(y, op=dist.ReduceOp.SUM, async_op=True) + return y + + + @torch.compile(fullgraph=True) + def all_reduce_wait_compiled(y): + torch.ops.c10d_functional.wait_tensor(y) + return y * y + + + x = torch.ones(1280, 1280, device="cuda") + self.rank + # the context manager ensures that `wait_tensor(y)` will wait on the correct work object + with allow_inflight_collective_as_graph_input_ctx(): + y = all_reduce_eager(x) + z = all_reduce_wait_compiled(y) + ``` + With this context manager, when a collective is called, under the hood the work object of the collective + will be registered in the work registry, and the wait_tensor() in compiled region called on + the output tensor of the collective will wait on the correct work object. + """ + previous = torch._C._distributed_c10d._allow_inflight_collective_as_graph_input() + + try: + torch._C._distributed_c10d._set_allow_inflight_collective_as_graph_input(value) + yield + finally: + torch._C._distributed_c10d._set_allow_inflight_collective_as_graph_input( + previous + ) + + +def _make_all_gather_out_tensor(input, group_size): + out_size = list(input.size()) + if len(out_size) == 0: + out_size.append(group_size) + else: + out_size[0] *= group_size + out_tensor = input.new_empty(out_size) + return out_tensor + + +def _all_gather_into_tensor_coalesced_meta(self, tag, rankset, group_size): + return [_make_all_gather_out_tensor(t, group_size) for t in self] + + +# We now register meta kernels to deal with tracing +def _broadcast_meta(self, *args): + return torch.empty_like(self) + + +def _all_reduce_meta(self, *args): + return torch.empty_like(self, memory_format=torch.contiguous_format) + + +def _wait_tensor_meta(self, *args): + return torch.empty_like(self) + + +def _isend_meta(self, *args): + return torch.empty(0, dtype=self.dtype, device=self.device) + + +def _irecv_meta(self, *args): + return torch.empty_like(self) + + +def _batch_p2p_ops_meta(op_list, peer_list, tag_list, tensors, group_name): + return [ + t if op == "irecv" else torch.empty(0, dtype=t.dtype, device=t.device) + for op, t in zip(op_list, tensors) + ] + + +def _all_gather_into_tensor_meta(shard, tag, rankset, group_size): + return _make_all_gather_out_tensor(shard, group_size) + + +def _reduce_scatter_tensor_meta(input, reduce_op, tag, rankset, group_size): + out_size = list(input.size()) + out_size[0] //= group_size + return input.new_empty(out_size) + + +def _all_reduce_coalesced_meta(self, *args): + return [torch.empty_like(t) for t in self] + + +def _all_reduce__meta(inp, *args): + return inp + + +def _broadcast__meta(inp, *args): + return inp + + +def _all_reduce_coalesced__meta(inputs, *args): + return inputs + + +def _reduce_scatter_tensor_coalesced_meta(inputs, reduceOp, tag, rankset, group_size): + def mk_out_tensor(input): + out_size = list(input.size()) + out_size[0] //= group_size + out_tensor = input.new_empty(out_size) + return out_tensor + + return [mk_out_tensor(t) for t in inputs] + + +# NB: We often say all_to_all has dynamic output size, but this is not +# technically true: instead, what typically happens is you manually +# communicate the output_split_sizes ahead of time (which is dynamic), +# but then you pass those sizes explicitly, and the all to all itself +# isn't dynamic, it just follows the specified output splits +def _all_to_all_single_meta( + input, output_split_sizes, input_split_sizes, *args, **kwargs +): + if output_split_sizes is None: + return input.new_empty(input.size()) + else: + for s in output_split_sizes: + torch._check(s >= 0) + out_size = list(input.size()) + out_size[0] = sum(output_split_sizes) + return input.new_empty(out_size) + + +def _all_gather_into_tensor_out_native_meta(input, group_size, group_name, *, out): + return _make_all_gather_out_tensor(input, group_size) + + +def _all_gather_into_tensor_native_meta(input, group_size, group_name): + return _make_all_gather_out_tensor(input, group_size) + + +def _all_gather_into_tensor_coalesced_native_meta(inputs, group_size, group_name): + return [ + _all_gather_into_tensor_native_meta(input, group_size, group_name) + for input in inputs + ] + + +def _reduce_scatter_tensor_native_meta(inp, reduce_op, group_size, group_name): + shape = list(inp.size()) + shape[0] //= group_size + return inp.new_empty(shape) + + +def _reduce_scatter_tensor_out_native_meta( + inp, reduce_op, group_size, group_name, *, out +): + shape = list(inp.size()) + shape[0] //= group_size + return inp.new_empty(shape) + + +def _reduce_scatter_tensor_coalesced_native_meta( + inputs, reduce_op, group_size, group_name +): + return [ + _reduce_scatter_tensor_native_meta(inp, reduce_op, group_size, group_name) + for inp in inputs + ] + + +# Library MUST be defined at module scope or it doesn't work +lib_impl = torch.library.Library("_c10d_functional", "IMPL") +lib_impl.impl("all_reduce", _all_reduce_meta, "Meta") +lib_impl.impl("all_reduce_", _all_reduce__meta, "Meta") +lib_impl.impl("all_reduce_coalesced", _all_reduce_coalesced_meta, "Meta") +lib_impl.impl("all_reduce_coalesced_", _all_reduce_coalesced__meta, "Meta") +lib_impl.impl("wait_tensor", _wait_tensor_meta, "Meta") +lib_impl.impl("isend", _isend_meta, "Meta") +lib_impl.impl("irecv", _irecv_meta, "Meta") +lib_impl.impl("batch_p2p_ops", _batch_p2p_ops_meta, "Meta") + +lib_impl.impl( + "all_gather_into_tensor_out", _all_gather_into_tensor_out_native_meta, "Meta" +) +lib_impl.impl("all_gather_into_tensor", _all_gather_into_tensor_native_meta, "Meta") +lib_impl.impl( + "all_gather_into_tensor_coalesced", + _all_gather_into_tensor_coalesced_native_meta, + "Meta", +) +lib_impl.impl("reduce_scatter_tensor", _reduce_scatter_tensor_native_meta, "Meta") +lib_impl.impl( + "reduce_scatter_tensor_out", _reduce_scatter_tensor_out_native_meta, "Meta" +) +lib_impl.impl( + "reduce_scatter_tensor_coalesced", + _reduce_scatter_tensor_coalesced_native_meta, + "Meta", +) +lib_impl.impl("all_to_all_single", _all_to_all_single_meta, "Meta") +lib_impl.impl("broadcast", _broadcast_meta, "Meta") +lib_impl.impl("broadcast_", _broadcast__meta, "Meta") + +# Backward compatibility: register Meta impls for _c10d_functional_autograd ops +# These ops redirect to _c10d_functional but need their own Meta registration so +# torch.compile / FakeTensor mode can dispatch them. +lib_impl_autograd = torch.library.Library("_c10d_functional_autograd", "IMPL") +lib_impl_autograd.impl( + "all_gather_into_tensor", _all_gather_into_tensor_native_meta, "Meta" +) +lib_impl_autograd.impl( + "reduce_scatter_tensor", _reduce_scatter_tensor_native_meta, "Meta" +) +lib_impl_autograd.impl("all_to_all_single", _all_to_all_single_meta, "Meta") + +# Mark these ops as side effectful so that DCE does not remove communication +# whose result tensors are ignored by user code. +torch.fx.node.has_side_effect(torch.ops._c10d_functional.wait_tensor.default) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.wait_tensor) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.isend.default) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.isend) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.irecv.default) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.irecv) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.batch_p2p_ops.default) # type: ignore[has-type] +torch.fx.node.has_side_effect(torch.ops._c10d_functional.batch_p2p_ops) # type: ignore[has-type] + + +# Register legacy ops for backward compatibility +# TODO(yifu): remove these in functional collective beta release +legacy_lib = torch.library.Library("c10d_functional", "DEF") +legacy_lib_impl = torch.library.Library("c10d_functional", "IMPL") +ops_defs = [ + "broadcast(Tensor self, int src, str tag, int[] ranks, int group_size) -> Tensor", + "all_reduce(Tensor self, str reduceOp, str tag, int[] ranks, int group_size) -> Tensor", + "all_reduce_coalesced(Tensor[] self, str reduceOp, str tag, int[] ranks, int group_size) -> Tensor[]", + "wait_tensor(Tensor self) -> Tensor", + "isend(Tensor self, int dst, int tag, str group_name) -> Tensor", + "irecv(Tensor self, int src, int tag, str group_name) -> Tensor", + "batch_p2p_ops(str[] op_list, int[] peer_list, int[] tag_list, Tensor[] tensors, str group_name) -> Tensor[]", + "all_gather_into_tensor(Tensor shard, str tag, int[] ranks, int group_size) -> Tensor", + "all_gather_into_tensor_coalesced(Tensor[] input, str tag, int[] ranks, int group_size) -> Tensor[]", + "reduce_scatter_tensor(Tensor input, str reduceOp, str tag, int[] ranks, int group_size) -> Tensor", + "reduce_scatter_tensor_coalesced(Tensor[] inputs, str reduceOp, str tag, int[] ranks, int group_size) -> Tensor[]", + "all_to_all_single(Tensor input, SymInt[]? output_split_sizes, SymInt[]? input_split_sizes, str tag, int[] ranks, int group_size) -> Tensor", +] + +my_module = sys.modules[__name__] +for op_def in ops_defs: + op_name = op_def[0 : op_def.index("(")] + backend_impl = getattr(fun_col_impl, f"_{op_name}") + legacy_lib.define(op_def, tags=torch.Tag.pt2_compliant_tag) + legacy_lib_impl.impl(op_name, backend_impl, "CompositeImplicitAutograd") + + +""" +Dynamo Remappings allow seamless translation from non-functional collectives of supportable form into +functional collective calls followed by inplace copy ops, allowing them to be traced into a functional graph. + +We implement this by writing a decomposition and teaching dynamo how to associate it to a corresponding op via +the mapping dict below. + +These schemas intentionally match torch.distributed.distributed_c10d.* ops that we are trying to remap from +""" + + +def all_gather_tensor_inplace( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group=None, # TODO add a type, + async_op: bool = False, + tag: str = "", + gather_dim: int = 0, +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + return output_tensor.copy_(all_gather_single(input_tensor, gather_dim, group, tag)) + + +def reduce_scatter_tensor_inplace( + output: torch.Tensor, + input: torch.Tensor, + op: str = "sum", # TODO type is actually c10d ReduceOp. is this ok? + group=None, # TODO add a type + async_op: bool = False, + scatter_dim: int = 0, + tag: str = "", +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + return output.copy_(reduce_scatter_single(input, op, scatter_dim, group, tag)) + + +REDUCE_OP_TO_STR = { + dist.ReduceOp.SUM: "sum", + dist.ReduceOp.AVG: "avg", + dist.ReduceOp.PRODUCT: "product", + dist.ReduceOp.MIN: "min", + dist.ReduceOp.MAX: "max", + dist.ReduceOp.BAND: "band", + dist.ReduceOp.BOR: "bor", + dist.ReduceOp.BXOR: "bxor", +} + + +def all_reduce_inplace( + tensor: torch.Tensor, + op: str = "sum", + group=None, + async_op: bool = False, + tag: str = "", +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + return tensor.copy_(all_reduce(tensor, op, group, tag)) + + +def all_to_all_inplace( + output: torch.Tensor, + input: torch.Tensor, + output_split_sizes=None, + input_split_sizes=None, + group=None, + async_op=False, + tag: str = "", +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + return output.copy_( + all_to_all_single( + input, + output_split_sizes, + input_split_sizes, + group, + tag, + ) + ) + + +def all_gather_inplace( + tensor_list: list[torch.Tensor], + tensor: torch.Tensor, + group=None, + async_op=False, + tag: str = "", +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + if tensor.dim() != 0 and not all(t.size(0) == tensor.size(0) for t in tensor_list): + raise AssertionError("Remapping variable size all_gather is not yet supported") + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + output = all_gather_single(tensor, 0, group, tag) + + # Use aten.slice instead of aten.split because the latter causes + # tensor.shape(0) to be unnecessarily baked in when it's a SymInt. + output_splits = [] + offset = 0 + for t in tensor_list: + is_scalar = t.dim() == 0 + t_offset = 1 if is_scalar else t.size(0) + + out = output[offset] if is_scalar else output[offset : offset + t_offset] + output_splits.append(out) + + offset += t_offset + for dst, src in zip(tensor_list, output_splits): + dst.copy_(src) + return tensor_list + + +def reduce_scatter_inplace( + output: torch.Tensor, + input_list: list[torch.Tensor], + op: str = "sum", + group: dist.ProcessGroup | None = None, + async_op: bool = False, + tag: str = "", +): + if async_op: + raise AssertionError( + "Can't remap async version of inplace op to functional collective" + ) + if not all(t.size() == output.size() for t in input_list): + raise AssertionError( + "reduce_scatter requires every input_list element to have the same size as output" + ) + + group = group or dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + + if output.dim() == 0: + # scalars have no dim to scatter along; stack into 1-D then reshape back + result = reduce_scatter_single(torch.stack(input_list), op, 0, group, tag) + return output.copy_(result.reshape(output.shape)) + return output.copy_(reduce_scatter_single(torch.cat(input_list), op, 0, group, tag)) + + +def isend_inplace( + tensor: torch.Tensor, + dst: int | None = None, + tag: int = 0, + group: dist.ProcessGroup | None = None, + group_dst: int = -1, +): + if group is None: + group = dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + if group_dst != -1: + if dst is not None: + raise ValueError( + "Cannot specify both 'dst' and 'group_dst' args as per eager impl" + ) + local_dst = group_dst + elif dst is not None: + local_dst = c10d.get_group_rank(group, dst) + else: + raise ValueError("Must specify either 'dst' or 'group_dst'") + + group_name = _resolve_group_name(group) + tensor = torch.ops._c10d_functional.isend(tensor, local_dst, tag, group_name) + if _are_we_tracing(): + return tensor + return _maybe_wrap_tensor(tensor) + + +def irecv_inplace( + tensor: torch.Tensor, + src: int | None = None, + tag: int = 0, + group: dist.ProcessGroup | None = None, + group_src: int = -1, +): + if group is None: + group = dist.group.WORLD + if group is None: + raise AssertionError("group cannot be None") + if group_src != -1: + if src is not None: + raise ValueError( + "Cannot specify both 'src' and 'group_src' args as per eager impl" + ) + local_src = group_src + elif src is not None: + local_src = c10d.get_group_rank(group, src) + else: + raise ValueError("Must specify either 'src' or 'group_src'") + group_name = _resolve_group_name(group) + tensor = torch.ops._c10d_functional.irecv(tensor, local_src, tag, group_name) + return _maybe_wrap_tensor(tensor) + + +def batch_p2p_ops_inplace( + op_list: list[str], + peer_list: list[int], + tag_list: list[int], + tensors: list[torch.Tensor], + group_name: RANK_TYPES, +): + if not dist.is_initialized(): + raise AssertionError("torch.distributed must be initialized") + if group_name is None or group_name == "": + group_name = c10d._get_default_group() + resolved = _resolve_group(group_name) + group_name = resolved if isinstance(resolved, str) else resolved.group_name + tensors = torch.ops._c10d_functional.batch_p2p_ops( + op_list, peer_list, tag_list, tensors, group_name + ) + if _are_we_tracing(): + return [ + _maybe_wrap_tensor(t) if op == "irecv" else t + for op, t in zip(op_list, tensors) + ] + return list(map(_maybe_wrap_tensor, tensors)) + + +def _group_or_group_name( + group: dist.ProcessGroup | c10d.GroupName, +) -> dist.ProcessGroup | c10d.GroupName: + if isinstance(group, str): + return group + elif torch.compiler.config.compile_on_one_rank: + return group + else: + return group.group_name + + +from torch.distributed.distributed_c10d import ( # pyrefly: ignore # deprecated; pyrefly: ignore [deprecated] + _all_gather_base as legacy_all_gather_base, + _reduce_scatter_base as legacy_reduce_scatter_base, + all_gather as legacy_all_gather, + all_gather_into_tensor as legacy_allgather, + all_gather_single as legacy_allgather_single, + all_reduce as legacy_allreduce, + all_to_all_single as legacy_all_to_all_single, + batch_isend_irecv as legacy_batch_p2p_ops, + irecv as legacy_irecv, + isend as legacy_isend, + reduce_scatter as legacy_reduce_scatter, + reduce_scatter_single as legacy_reducescatter_single, + reduce_scatter_tensor as legacy_reducescatter, +) + + +# Dynamo remaps dist.* collectives to these wrappers via traceable_collective_remaps. +# Each wrapper calls the in-place functional collective and returns None, +# matching the return type of the original dist.* APIs when async_op=False. +# async_op=True already graph-breaks in CollectiveFunctionRewriteVariable. +# These must be module-level def statements (not closures from a decorator factory) +# because _traceable_collectives_source resolves Dynamo guard sources by looking up +# fn.__name__ as a module attribute — a def's __name__ matches its variable name +# automatically, whereas a closure's would not. +def _remapped_allgather(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError("_remapped_allgather should only be called during tracing") + all_gather_tensor_inplace(*args, **kwargs) + + +def _remapped_reducescatter(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError( + "_remapped_reducescatter should only be called during tracing" + ) + reduce_scatter_tensor_inplace(*args, **kwargs) + + +def _remapped_allreduce(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError("_remapped_allreduce should only be called during tracing") + all_reduce_inplace(*args, **kwargs) + + +def _remapped_all_to_all_single(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError( + "_remapped_all_to_all_single should only be called during tracing" + ) + all_to_all_inplace(*args, **kwargs) + + +def _remapped_all_gather(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError( + "_remapped_all_gather should only be called during tracing" + ) + all_gather_inplace(*args, **kwargs) + + +def _remapped_reduce_scatter(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError( + "_remapped_reduce_scatter should only be called during tracing" + ) + reduce_scatter_inplace(*args, **kwargs) + + +def _remapped_isend(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError("_remapped_isend should only be called during tracing") + return isend_inplace(*args, **kwargs) + + +def _remapped_irecv(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError("_remapped_irecv should only be called during tracing") + return irecv_inplace(*args, **kwargs) + + +def _remapped_batch_p2p_ops(*args, **kwargs): + if not _are_we_tracing(): + raise AssertionError( + "_remapped_batch_p2p_ops should only be called during tracing" + ) + return batch_p2p_ops_inplace(*args, **kwargs) + + +# This dict should contain sets of functions that dynamo is allowed to remap. +# Functions in this set should accept the same args/kwargs 1:1 as their mapping. +traceable_collective_remaps = { + legacy_allgather: _remapped_allgather, + legacy_allgather_single: _remapped_allgather, + legacy_reducescatter: _remapped_reducescatter, + legacy_reducescatter_single: _remapped_reducescatter, + legacy_allreduce: _remapped_allreduce, + legacy_all_to_all_single: _remapped_all_to_all_single, + legacy_all_gather: _remapped_all_gather, + legacy_reduce_scatter: _remapped_reduce_scatter, + legacy_reduce_scatter_base: _remapped_reducescatter, + legacy_all_gather_base: _remapped_allgather, + legacy_isend: _remapped_isend, + legacy_irecv: _remapped_irecv, + legacy_batch_p2p_ops: _remapped_batch_p2p_ops, +} + + +def _remap_traceable_collective( + func: Callable[..., Any], + args: Sequence[Any], + kwargs: dict[str, Any] | None, +) -> tuple[Callable[..., Any], tuple[Any, ...], dict[str, Any]] | None: + """Redirect a legacy ``torch.distributed`` collective to its functional + equivalent, following Dynamo's CollectiveFunctionRewriteVariable. + + Returns ``(mapped_func, (), kwargs)`` if ``func`` is a remappable collective, + else None. Shared by the make_fx compile_on_one_rank mode below and + non-strict export's _NonStrictTorchFunctionHandler. + """ + if func not in traceable_collective_remaps: + return None + import inspect + + mapped_func = traceable_collective_remaps[func] + bound = dict(inspect.signature(func).bind(*args, **(kwargs or {})).arguments) + if func in ( + torch.distributed.all_reduce, + torch.distributed.reduce_scatter, + torch.distributed.reduce_scatter_single, + torch.distributed.reduce_scatter_tensor, + torch.distributed._reduce_scatter_base, + ): + if "op" in bound: + bound["op"] = REDUCE_OP_TO_STR[bound["op"]] + return mapped_func, (), bound + + +class _LegacyToFunctionalCollectiveMode(torch.overrides.TorchFunctionMode): + """Redirect legacy ``torch.distributed`` collectives to functional collectives + while tracing, mirroring Dynamo and non-strict export. + + Used by ``make_fx`` under ``compile_on_one_rank``: the in-place ``c10d.*`` ops + bind the ProcessGroup as a torchbind constant that cannot be serialized, + whereas the functional collectives take the group as an op argument, so the + group (e.g. a ``DeviceMesh.get_group()`` -> ``mesh_get_process_group`` output) + flows into the graph instead of being baked in. + """ + + def __torch_function__( + self, + func: Callable[..., Any], + types: tuple[type, ...], + args: tuple[Any, ...] = (), + kwargs: dict[str, Any] | None = None, + ) -> Any: + remapped = _remap_traceable_collective(func, args, kwargs) + if remapped is not None: + mapped_func, mapped_args, mapped_kwargs = remapped + return mapped_func(*mapped_args, **mapped_kwargs) + return func(*args, **(kwargs or {})) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives_impl.py b/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives_impl.py new file mode 100644 index 0000000000000000000000000000000000000000..3d6979a330469b1ee1800bcb1ea5324f0e4fa642 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_functional_collectives_impl.py @@ -0,0 +1,137 @@ +# mypy: allow-untyped-defs + +import torch +import torch.distributed.distributed_c10d as c10d + + +""" +This file contains the op impls for the legacy (c10d_functional) functional collectives. +These impls simply call into the native (_c10d_functional) functional collectives. +""" + + +def _broadcast(input, src, tag, ranks, group_size): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.broadcast( + input, + src, + group_name, + ) + + +def _all_reduce(input, reduce_op, tag, ranks, group_size): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.all_reduce( + input, + reduce_op, + group_name, + ) + + +def _all_reduce_coalesced(inputs, reduce_op, tag, ranks, group_size): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.all_reduce_coalesced( + inputs, + reduce_op, + group_name, + ) + + +def _all_gather_into_tensor(input, tag, ranks, group_size): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.all_gather_into_tensor( + input, + group_size, + group_name, + ) + + +def _all_gather_into_tensor_coalesced(input, tag, ranks, group_size): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.all_gather_into_tensor_coalesced( + input, + group_size, + group_name, + ) + + +def _reduce_scatter_tensor( + input: torch.Tensor, + reduce_op: str, + tag: str, + ranks: list[int], + group_size: int, +): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.reduce_scatter_tensor( + input, + reduce_op, + group_size, + group_name, + ) + + +def _reduce_scatter_tensor_coalesced( + inputs: list[torch.Tensor], + reduce_op: str, + tag: str, + ranks: list[int], + group_size: int, +): + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.reduce_scatter_tensor_coalesced( + inputs, + reduce_op, + group_size, + group_name, + ) + + +def _all_to_all_single( + input: torch.Tensor, + output_split_sizes: list[int] | None, + input_split_sizes: list[int] | None, + tag: str, + ranks: list[int], + group_size: int, +): + if output_split_sizes is None or input_split_sizes is None: + if not (output_split_sizes is None and input_split_sizes is None): + raise AssertionError( + "output_split_sizes and input_split_sizes must either be " + "specified together or both set to None" + ) + output_split_sizes = [input.shape[0] // group_size] * group_size + input_split_sizes = output_split_sizes + + group_name = c10d._resolve_group_name_by_ranks_and_tag(ranks, tag) + return torch.ops._c10d_functional.all_to_all_single( + input, + output_split_sizes, + input_split_sizes, + group_name, + ) + + +def _wait_tensor(tensor: torch.Tensor) -> torch.Tensor: + return torch.ops._c10d_functional.wait_tensor(tensor) + + +def _isend(tensor: torch.Tensor, dst: int, tag: str, group_name): + return torch.ops._c10d_functional.isend(tensor, dst, tag, group_name) + + +def _irecv(tensor: torch.Tensor, src: int, tag: str, group_name): + return torch.ops._c10d_functional.irecv(tensor, src, tag, group_name) + + +def _batch_p2p_ops( + op_list: list[str], + peer_list: list[int], + tag_list: list[int], + tensors: list[torch.Tensor], + group_name: str, +): + return torch.ops._c10d_functional.batch_p2p_ops( + op_list, peer_list, tag_list, tensors, group_name + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f32b0ae4ccdf2834f64d4dbf3b2dd8fd65249f26 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__init__.py @@ -0,0 +1,2228 @@ +""" +A LocalTensor is a tensor subclass which simulates a tensor that is +distributed across SPMD ranks. A LocalTensor might be size N, but in fact +there are world_size shards/replicas of it stored internally. When you do a +plain PyTorch operation on it, we apply the operation to each shard; when you +do a collective, we do the mathematically equivalent operation on the local +shards. A LocalTensor is associated with a list of ranks which specify +which ranks it holds local tensors for. + +NB, this is NOT a DataParallel like abstraction where you can run operations +on multiple different GPUs. It is intended purely for *debugging* purposes, +the overhead is almost certainly too high to keep eight GPUs (even the C++ +autograd needs multithreading to keep up!) (It might potentially be possible +to trace through this with torch.compile and then compile it with CUDA graphs +but this is currently a non-goal.) + +We do not directly handle MPMD. However in practice even in SPMD you may +encounter divergence in behavior per rank (for example, uneven sharding +across ranks). To support scenarios like this, we provide a helper decorator +that allows you to run a function with no side effects for each LocalTensor +shard and combine results back into LocalTensor or LocalIntNode. + +NB: This is a torch dispatch Tensor subclass, as we want to assume that autograd +is SPMD, so we run it once, and dispatch the inner autograd calls to the individual +local shards. + +NOTE ABOUT MESH: This subclass requires collectives that are issued to it to +respect a DeviceMesh like abstraction. The reason for this is that when +DTensor issues us a collective for a particular rank, you will be asked to do +this on a specific process group which involves some ranks. However, this +will only be for the LOCAL PG that this particular rank is participating in; +there will be a bunch of other PGs for other nodes that you don't get to see. +We need to be able to reverse engineer all of the collectives that don't +involve the current local rank here to actually issue them. This can be done +two ways: (1) looking at the participating local ranks in the PG and computing +the complement which specifies all the other collectives you have to run, or +(2) retrieving the device mesh axis corresponding to the PG for this rank, and +then running all the fibers for this. +""" + +import contextlib +import copy +import functools +import importlib +import operator +import os +import sys +import threading +from ast import Call +from collections import defaultdict +from collections.abc import Callable, Generator, Sequence +from types import TracebackType +from typing import Any, Optional, ParamSpec, TypeVar, Union + + +try: + import numpy as np + + HAS_NUMPY = True +except ModuleNotFoundError: + HAS_NUMPY = False + np = None # type: ignore[assignment] + +import torch +import torch.distributed as dist +from torch import Size, SymBool, SymInt, Tensor +from torch._C import DispatchKey, DispatchKeySet, ScriptObject +from torch._export.wrappers import mark_subclass_constructor_exportable_experimental +from torch._ops import OpOverload +from torch._subclasses.fake_tensor import FakeTensor, FakeTensorMode +from torch.distributed import DeviceMesh, ProcessGroup +from torch.distributed._functional_collectives import AsyncCollectiveTensor +from torch.distributed.distributed_c10d import _get_default_group +from torch.fx.experimental._constant_symnode import ConstantIntNode +from torch.nested._internal.nested_int import NestedIntNode +from torch.utils import _pytree as pytree +from torch.utils._mode_utils import no_dispatch +from torch.utils._python_dispatch import ( + _get_current_dispatch_mode_stack, + return_and_correct_aliasing, + TorchDispatchMode, +) +from torch.utils.checkpoint import get_device_states, set_device_states + + +_R = TypeVar("_R") +_P = ParamSpec("_P") + +not_implemented_log = torch._logging.getArtifactLogger(__name__, "not_implemented") + + +from . import _c10d + + +def _is_in_fake_tensor_mode() -> bool: + return any( + isinstance(mode, FakeTensorMode) for mode in _get_current_dispatch_mode_stack() + ) + + +def _reduce_multidim_lists( + lists_to_reduce: list[Any], reduce_func: Callable[[list[Any]], Any] +) -> Any: + """ + Reduces a list of multi-dimensional lists, assuming they all have + the exact same shape. + + Args: + lists_to_reduce (list): A list where each item is a multi-dimensional + list (e.g., [md_list_1, md_list_2, ...]). + All inner md_lists must have the same shape. + reduce_func (callable): A function that takes an iterable (list) of + values and returns a single reduced value. + For example: sum, max, min, or + lambda x: sum(x) / len(x) for mean. + + Returns: + A single multi-dimensional list of the same shape as the inputs, + where each value is the result of the reduce_func. + + Raises: + ValueError: If the input list is empty or if shapes are inconsistent + (which may also raise IndexError or TypeError). + """ + if not lists_to_reduce: + raise ValueError("Input 'lists_to_reduce' cannot be empty.") + + # Get the first list to inspect its structure (shape) + first_list = lists_to_reduce[0] + + # Check if the first element of this list is *also* a list. + # This determines if we are at the base case or need to recurse. + if isinstance(first_list[0], list): + # --- RECURSIVE STEP --- + # The elements are lists, so we need to go one level deeper. + + # We find the number of sub-lists from the first list. + # (e.g., for [[1,2], [3,4]], this is 2) + num_sublists = len(first_list) + + result = [] + # Iterate by the index of the sub-lists (e.g., i = 0, then i = 1) + for i in range(num_sublists): + # Build a new list to pass to the recursive call. + # This list will contain the i-th sublist from *each* of the + # input lists. + # e.g., if lists_to_reduce = [ L1, L2 ] and i = 0, + # this creates [ L1[0], L2[0] ] + sublists_to_reduce = [l[i] for l in lists_to_reduce] + + # Recurse and append the result + result.append(_reduce_multidim_lists(sublists_to_reduce, reduce_func)) + return result + else: + # --- BASE CASE --- + # The elements are values (int, float, etc.), not lists. + # We are at the innermost dimension. + + # Find the number of values in the innermost list. + # (e.g., for [1, 2], this is 2) + num_values = len(first_list) + + result = [] + # Iterate by the index of the values (e.g., i = 0, then i = 1) + for i in range(num_values): + # Get the values at this specific position (i) from *all* + # input lists. + # e.g., if lists_to_reduce = [ [1,2], [10,20] ] and i = 0, + # this creates [ 1, 10 ] + values_at_pos = [l[i] for l in lists_to_reduce] + + # Apply the user-provided reduction function to this list of values + # and append the single result. + result.append(reduce_func(values_at_pos)) + return result + + +def _is_inplace_op(op: OpOverload | Callable[..., Any]) -> bool: + return ( + isinstance(op, OpOverload) + # Not precise heuristic to detect inplace operation + and op._schema.name[-1] == "_" + # Strengthen the heuristic to check that the first argument and return value are a write + and len(op._schema.arguments) > 0 + and op._schema.arguments[0].is_write + and len(op._schema.returns) > 0 + and op._schema.returns[0].is_write + ) + + +def _int_on_rank(i: "int | LocalIntNode | ConstantIntNode", r: int) -> int: + if isinstance(i, LocalIntNode): + return i._local_ints[r] + elif isinstance(i, ConstantIntNode): + return i.val + elif isinstance(i, int): + return i + else: + raise AssertionError(type(i)) + + +def _check_for_subclass(flat_args: Sequence[object]) -> bool: + return any(_check_for_subclass_arg(x) for x in flat_args) + + +def _check_for_subclass_arg(x: object) -> bool: + return ( + not isinstance(x, LocalTensor) + and isinstance(x, Tensor) + and type(x) + not in ( + Tensor, + FakeTensor, + torch.nn.Parameter, + torch.nn.Buffer, + ) + ) + + +def _map_to_rank_local_val(val: Any, rank: int) -> Any: + if isinstance(val, LocalTensor): + return val._local_tensors[rank] + if isinstance(val, SymInt): + if isinstance(val.node, LocalIntNode): + return val.node._local_ints[rank] + if isinstance(val.node, ConstantIntNode): + return val.node.val + return val + + +def _collect_accelerator_rng_states() -> dict[int, torch.Tensor]: + """ + Collects RNG state from all available accelerator devices. + + Returns: + List of RNG state tensors, one for each accelerator device. + Returns empty list if accelerator is not available. + """ + if not torch.accelerator.is_available(): + return {} + + if torch.accelerator.is_available(): + device_idx = torch.accelerator.current_device_index() + with torch.accelerator.device_index(device_idx): + return {device_idx: torch.get_device_module().get_rng_state()} + + return {} + + +def _set_accelerator_rng_states(rng_states: dict[int, torch.Tensor]) -> None: + """ + Sets RNG state for all accelerator devices from a list of states. + + Args: + rng_states: List of RNG state tensors to restore. + """ + if not torch.accelerator.is_available(): + return + + if torch.accelerator.is_available(): + for device_idx, device_rng_state in rng_states.items(): + with torch.accelerator.device_index(device_idx): + torch.get_device_module().set_rng_state(device_rng_state) + + +def _get_rng_state() -> tuple[torch.Tensor, dict[int, torch.Tensor]]: + """ + Gets CPU and accelerator (e.g., CUDA, XPU device) rng states from all devices. + """ + return (torch.get_rng_state(), _collect_accelerator_rng_states()) + + +def _set_rng_state( + cpu_state: torch.Tensor, accelerator_states: dict[int, torch.Tensor] +) -> None: + """ + Sets CPU and accelerator (e.g., CUDA, XPU device) rng states for all devices. If + the list of accelerator states is shorter than the number of devices only the + first len(accelerator_states) devices will get their rng state set. + """ + torch.set_rng_state(cpu_state) + _set_accelerator_rng_states(accelerator_states) + + +def _combine_int_rank_results(rank_results: dict[int, int]) -> int | torch.SymInt: + any_v = next(iter(rank_results.values())) + + if all(v == any_v for v in rank_results.values()): + return any_v + + return torch.SymInt(LocalIntNode(rank_results)) + + +def _combine_any_rank_results(rank_results: dict[int, Any]) -> Any: + any_v = next(iter(rank_results.values())) + + if isinstance(any_v, Tensor): + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(rank_results) + + if isinstance(any_v, int): + return _combine_int_rank_results(rank_results) + + if isinstance(any_v, torch.device): + if not all(v.type == any_v.type for v in rank_results.values()): + raise AssertionError("device type should be the same") + # Just use the first device - the device type is what matters, + # and LocalTensorMode runs on a single physical device anyway + return any_v + + if not all(v == any_v for v in rank_results.values()): + raise AssertionError( + "Non Tensor or int rank results must be equal for all ranks" + ) + + return any_v + + +def _combine_rank_results(rank_results: dict[int, Any], default: Any | None) -> Any: + rank_ids = rank_results.keys() + rank_value = rank_results[next(iter(rank_ids))] + + if isinstance(rank_value, (list, tuple)): + max_rank_result_len = max(len(v) for v in rank_results.values()) + ret_list = [] + for i in range(max_rank_result_len): + rank_col_results = { + r: v[i] if i < len(v) else default for r, v in rank_results.items() + } + ret_list.append(_combine_any_rank_results(rank_col_results)) + return type(rank_value)(ret_list) + else: + return _combine_any_rank_results(rank_results) + + +def _zero_sized_like(tensor: torch.Tensor, dim: int) -> torch.Tensor: + tensor_size = list(tensor.size()) + tensor_size[dim] = 0 + empty_tensor = torch.empty(*tensor_size, dtype=tensor.dtype, device=tensor.device) + return empty_tensor + + +def _for_each_rank_run_func( + func: OpOverload | Callable[..., Any], + ranks: frozenset[int], + args: Sequence[Any], + kwargs: dict[str, Any], + *, + alias: bool = True, +) -> Any: + flat_args, args_spec = pytree.tree_flatten((args, kwargs)) + flat_args = [ + a.wait() if isinstance(a, AsyncCollectiveTensor) else a for a in flat_args + ] + + lm = enabled_local_tensor_mode() + use_per_rank_rng = lm is not None and len(lm._per_rank_rng_states) > 0 + + global_rng_state = None if use_per_rank_rng else _get_rng_state() + + flat_rank_rets = {} + + default_value: Tensor | None = None + for r in sorted(ranks): + if use_per_rank_rng: + if lm is None: + raise AssertionError + if r in lm._per_rank_rng_states: + _set_rng_state(*lm._per_rank_rng_states[r]) + else: + if global_rng_state is None: + raise AssertionError + _set_rng_state(*global_rng_state) + + rank_flat_args = [_map_to_rank_local_val(a, r) for a in flat_args] + rank_args, rank_kwargs = pytree.tree_unflatten(rank_flat_args, args_spec) + if func is torch.ops.aten.hash_tensor.default and rank_args[0].numel() == 0: + # Special case for empty tensors, hash_tensor returns an empty tensor + rank_ret = torch.empty(0, dtype=torch.uint64, device=rank_args[0].device) + else: + rank_ret = func(*rank_args, **rank_kwargs) + flat_rank_rets[r] = rank_ret + + if use_per_rank_rng: + if lm is None: + raise AssertionError + lm._per_rank_rng_states[r] = _get_rng_state() + + if default_value is None and func is torch.ops.aten.split.Tensor: + # If split happens over the dimension smaller than the number of chunks + # it is possible that some ranks will produce shorter lists of chunks. + # In order to make the result across all ranks of the same length we + # append empty tensors (zero size on the split dimension). + tensor = rank_flat_args[0] + split_dim = 0 if len(rank_flat_args) < 3 else rank_flat_args[2] + default_value = _zero_sized_like(tensor, split_dim) + + if _is_inplace_op(func): + alias = False + # For the in-place ops return self + ret = args[0] + if isinstance(func, OpOverload) and torch.Tag.inplace_view in func.tags: + # Ensure that wrapper tensor size is synchronized with its local tensors + ret._sync_meta() + else: + ret = _combine_rank_results(flat_rank_rets, default_value) + + if alias: + return return_and_correct_aliasing(func, args, kwargs, ret) + else: + return ret + + +def _get_extra_dispatch_keys(t: torch.Tensor) -> DispatchKeySet: + extra_dispatch_keys = torch._C.DispatchKeySet.from_raw_repr(0) + if torch._C._dispatch_keys(t).has(torch._C.DispatchKey.Conjugate): + extra_dispatch_keys = extra_dispatch_keys.add(torch._C.DispatchKey.Conjugate) + if torch._C._dispatch_keys(t).has(torch._C.DispatchKey.Negative): + extra_dispatch_keys = extra_dispatch_keys.add(torch._C.DispatchKey.Negative) + return extra_dispatch_keys + + +class LocalIntNode: + """ + Like a LocalTensor, but for an int. We can't use a 0D tensor to represent this + because often only a SymInt is accepted where we wish to use this. + """ + + def __new__(cls, local_ints: dict[int, int]) -> "ConstantIntNode | LocalIntNode": # type: ignore[misc] + if len(set(local_ints.values())) == 1: + return ConstantIntNode(next(iter(local_ints.values()))) + return super().__new__(cls) + + def __init__(self, local_ints: dict[int, int]): + self._local_ints = local_ints + + def maybe_as_int(self) -> int | None: + return None + + def is_int(self) -> bool: + return True + + def is_float(self) -> bool: + return False + + def is_bool(self) -> bool: + return False + + def is_nested_int(self) -> bool: + return False + + def clone(self) -> "LocalIntNode": + return self + + def _str(self) -> str: + return f"LocalIntNode({self._local_ints})" + + def __str__(self) -> str: + return self._str() + + def __repr__(self) -> str: + return self._str() + + def _graph_repr(self) -> str: + return self._str() + + def is_symbolic(self) -> bool: + return False + + def is_constant(self) -> bool: + return False + + def sym_max( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + { + r: max(self._local_ints[r], _int_on_rank(other, r)) + for r in self._local_ints + } + ) + + def sym_min( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + { + r: min(self._local_ints[r], _int_on_rank(other, r)) + for r in self._local_ints + } + ) + + def sym_sum(self, other: Sequence[Any]) -> "LocalIntNode | ConstantIntNode": + t = LocalIntNode(dict.fromkeys(self._local_ints, 0)) + for o in other: + t = t.add(o) + return t + + def neg(self) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode({r: -self._local_ints[r] for r in self._local_ints}) + + def add( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] + _int_on_rank(other, r) for r in self._local_ints} + ) + + def sub( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] - _int_on_rank(other, r) for r in self._local_ints} + ) + + def mul( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] * _int_on_rank(other, r) for r in self._local_ints} + ) + + def floordiv( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] // _int_on_rank(other, r) for r in self._local_ints} + ) + + def mod( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] % _int_on_rank(other, r) for r in self._local_ints} + ) + + def int_floordiv( + self, other: "int | LocalIntNode | ConstantIntNode" + ) -> "LocalIntNode | ConstantIntNode": + return LocalIntNode( + {r: self._local_ints[r] // _int_on_rank(other, r) for r in self._local_ints} + ) + + def eq(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] == _int_on_rank(other, r) for r in self._local_ints} + return torch._C._get_constant_bool_symnode(len(r) == 1 and next(iter(r))) + + def ne(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] != _int_on_rank(other, r) for r in self._local_ints} + return torch._C._get_constant_bool_symnode(len(r) > 1 or next(iter(r))) + + def ge(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] >= _int_on_rank(other, r) for r in self._local_ints} + if len(r) != 1: + raise AssertionError((self, other)) + return torch._C._get_constant_bool_symnode(next(iter(r))) + + def le(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] <= _int_on_rank(other, r) for r in self._local_ints} + if len(r) != 1: + raise AssertionError((self, other)) + return torch._C._get_constant_bool_symnode(next(iter(r))) + + def gt(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] > _int_on_rank(other, r) for r in self._local_ints} + if len(r) != 1: + raise AssertionError((self, other)) + return torch._C._get_constant_bool_symnode(next(iter(r))) + + def lt(self, other: "int | LocalIntNode | ConstantIntNode") -> bool | SymBool: + r = {self._local_ints[r] < _int_on_rank(other, r) for r in self._local_ints} + if len(r) != 1: + raise AssertionError((self, other)) + return torch._C._get_constant_bool_symnode(next(iter(r))) + + def wrap_int(self, num: int) -> "LocalIntNode | ConstantIntNode": + return ConstantIntNode(num) + + +class _LocalDeviceHandle: + """ + Wrapper around device module (e.g., torch.cuda) with automatic LocalTensor semantics. + + This class wraps device modules and automatically handles per-rank operations in + LocalTensor mode: + - get_rng_state() returns a LocalTensor with per-rank states + - set_rng_state(LocalTensor) sets per-rank states + + When not in LocalTensor mode, it delegates directly to the underlying device handle. + """ + + def __init__(self, device_handle, device_type: str): + """ + Initialize the local device handle wrapper. + + Args: + device_handle: The underlying device module (e.g., torch.cuda) + device_type: Device type string (e.g., "cuda", "cpu") + """ + self._device_handle = device_handle + self._device_type = device_type + + def get_rng_state(self): + """ + Get RNG state, automatically returning LocalTensor in LocalTensor mode. + + Returns: + LocalTensor in LocalTensor mode, regular Tensor otherwise + """ + lm = enabled_local_tensor_mode() + if not lm: + return self._device_handle.get_rng_state() + + original_state = _get_rng_state() + per_rank_states = {} + + try: + for rank in lm.ranks: + # We need to set-then-get instead of directly copying lm._per_rank_rng_states[rank] + # because they have different structures: + # - lm._per_rank_rng_states[rank] is a tuple: (cpu_state, {device_idx: cuda_state}) + # - self._device_handle.get_rng_state() returns just the device-specific tensor + # So we temporarily restore the full RNG state (CPU + all CUDA devices) for this rank, + # then extract only the specific device's state tensor that we need. + if rank in lm._per_rank_rng_states: + _set_rng_state(*lm._per_rank_rng_states[rank]) + + per_rank_states[rank] = self._device_handle.get_rng_state() + finally: + _set_rng_state(*original_state) + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(per_rank_states) + + def set_rng_state(self, state): + """ + Set RNG state, automatically handling LocalTensor input. + + Args: + state: Regular Tensor or LocalTensor with per-rank states + """ + if isinstance(state, LocalTensor): + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError + + # Similar to get_rng_state but in reverse: we need to convert from + # device-specific tensor format to full state tuple format. + # - state._local_tensors[rank] contains just the device-specific RNG state tensor + # - lm._per_rank_rng_states[rank] needs a tuple: (cpu_state, {device_idx: cuda_state}) + # So we set the device's state with the rank-specific tensor, then _get_rng_state() + # captures both CPU and CUDA states into the tuple format that _per_rank_rng_states expects. + for rank, rank_state in state._local_tensors.items(): + self._device_handle.set_rng_state(rank_state.to("cpu")) + lm._per_rank_rng_states[rank] = _get_rng_state() + else: + self._device_handle.set_rng_state(state.to("cpu")) + + def __getattr__(self, name): + """Delegate all other attributes to the underlying device module.""" + return getattr(self._device_handle, name) + + +class _LocalOffsetBasedRNGTracker: + """ + LocalTensor-specific RNG tracker for DTensor random operations. + + This class manages per-rank RNG states when running in LocalTensor mode, + using _LocalPhiloxState to track different offsets for each virtual rank. + It is instantiated and used by OffsetBasedRNGTracker when in LocalTensor mode. + + Much of this is derived from OffsetBasedRNGTracker: + https://github.com/pytorch/pytorch/blob/402c46503002f98ccfc023a733081fb0719223a1/torch/distributed/tensor/_random.py#L182 + """ + + def __init__(self, device_type: str = "cuda"): + """Initialize the LocalTensor RNG tracker.""" + from torch.distributed.device_mesh import _get_device_handle + + self._device_type = device_type + self._device_handle = _LocalDeviceHandle( + _get_device_handle(device_type), device_type + ) + self.distribute_region_enabled = True + self._device_mesh = None + + @property + def _device(self): + return torch.device(self._device_type, torch.accelerator.current_device_index()) + + def _set_pre_op_offset(self, state, spec) -> None: + """Compute and set per-rank offsets before the random operation.""" + from torch.distributed.tensor._ops.utils import prod + from torch.distributed.tensor._utils import ( + _compute_local_shape_and_global_offset, + ) + from torch.distributed.tensor.placement_types import Shard + + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError + + state._per_rank_offsets = {} + + for rank in lm.ranks: + # compute this rank's coordinate in the mesh + mesh_coords = [] + for mesh_dim_idx in range(spec.mesh.ndim): + mesh_dim_size = spec.mesh.size(mesh_dim_idx) + # calculate rank's coordinate in this mesh dimension + num_chunks_after = 1 + for j in range(mesh_dim_idx + 1, spec.mesh.ndim): + num_chunks_after *= spec.mesh.size(j) + coord = (rank // num_chunks_after) % mesh_dim_size + mesh_coords.append(coord) + + # compute shard offset based on placements + from torch.distributed.tensor._random import ( + _calc_first_shard_size, + _calc_shard_info, + _calc_shard_linear_idx, + ) + + # Compute shard index and total number of shards on each tensor dim + shard_idx_by_dim, total_num_shards_by_dim = _calc_shard_info( + mesh_coords, spec + ) + + # compute shard linear index + shard_linear_idx = _calc_shard_linear_idx( + shard_idx_by_dim, total_num_shards_by_dim + ) + + # get current offset for this rank + current_offset = int( + state._per_rank_states[rank][8:].view(dtype=torch.int64).item() + ) + + local_shape = _calc_first_shard_size(spec) + # compute local size + local_size = prod(local_shape) + + # compute new offset (must be multiple of 4) + offset_incr = (shard_linear_idx * local_size + 3) // 4 * 4 + state._per_rank_offsets[rank] = current_offset + offset_incr + + def _set_post_op_offset(self, state, spec, old_offset) -> None: + """Set per-rank offsets after the random operation.""" + from torch.distributed.tensor._ops.utils import prod + + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError + + dtensor_shape = spec.shape + numel = prod(dtensor_shape) + # offset must be multiple of 4 + numel = (numel + 3) // 4 * 4 + + if not hasattr(state, "_per_rank_offsets"): + state._per_rank_offsets = {} + + # handle LocalIntNode old_offset (different values per rank) + if isinstance(old_offset, SymInt) and isinstance(old_offset.node, LocalIntNode): + for rank in lm.ranks: + rank_old_offset = old_offset.node._local_ints[rank] + state._per_rank_offsets[rank] = rank_old_offset + numel + else: + # same old_offset for all ranks + old_offset_int = ( + int(old_offset) if isinstance(old_offset, SymInt) else old_offset + ) + for rank in lm.ranks: + state._per_rank_offsets[rank] = old_offset_int + numel + + @contextlib.contextmanager + def _distribute_region(self, spec, generator=None): + """Context manager for LocalTensor mode distribute region.""" + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError + + # get base state + if generator is not None: + base_state_tensor = generator.get_state() + per_rank_states = {rank: base_state_tensor.clone() for rank in lm.ranks} + # pyrefly: ignore [bad-argument-type, bad-argument-count] + base_state_tensor = LocalTensor(per_rank_states) + else: + base_state_tensor = self._device_handle.get_rng_state() + + state = _LocalPhiloxState(base_state_tensor) + + if self.distribute_region_enabled: + # sync to rank 0's state if no explicit generator + if generator is None: + any_rank_state = lm._any_local_rng_state() + any_rank_cpu, any_rank_cuda = any_rank_state + + if self._device.type in {"cuda", "xpu"}: + if self._device.index not in any_rank_cuda: + raise AssertionError + any_rank_device_state = any_rank_cuda[self._device.index] + else: + any_rank_device_state = any_rank_cpu + + from torch.distributed.tensor._random import _PhiloxState + + any_rank_philox = _PhiloxState(any_rank_device_state) + state.seed = int(any_rank_philox.seed.item()) + state.offset = int(any_rank_philox.offset.item()) + + old_offset = state.offset + self._set_pre_op_offset(state, spec) + state.apply_to_local_tensor_mode(self._device_handle) + + try: + yield + finally: + self._set_post_op_offset(state, spec, old_offset) + state.apply_to_local_tensor_mode(self._device_handle) + else: + yield + + # maybe reset generator to rank 0's state + if generator is not None: + rank_0_state = state._per_rank_states[0] + generator.set_state(rank_0_state) + + +_LOCAL_TENSOR_ATTR_PREFIX = "_local_tensor_" + + +def _is_local_tensor_attr(attr: str) -> bool: + return attr.startswith(_LOCAL_TENSOR_ATTR_PREFIX) + + +def _to_local_tensor_attr(rank: int) -> str: + return f"{_LOCAL_TENSOR_ATTR_PREFIX}{rank}" + + +def _from_local_tensor_attr(attr: str) -> int: + if not _is_local_tensor_attr(attr): + raise AssertionError(f"Invalid local tensor attr {attr}") + return int(attr[len(_LOCAL_TENSOR_ATTR_PREFIX) :]) + + +def _all_elements_same(values: list[Any]) -> bool: + if not values: + return True + first_value = values[0] + return all(value == first_value for value in values) + + +def _compute_local_tensor_meta( + local_tensors: dict[int, torch.Tensor], +) -> tuple[ + list[torch.SymInt | int], + list[torch.SymInt | int], + torch.device, + torch.dtype, + torch.layout, + DispatchKeySet, +]: + """ + Computes the meta information for a LocalTensor from its local tensors. + """ + it = iter(local_tensors.values()) + first_local_tensor = next(it) + + first_shape = first_local_tensor.shape + first_stride = first_local_tensor.stride() + dtype = first_local_tensor.dtype + device = first_local_tensor.device + layout = first_local_tensor.layout + + extra_dispatch_keys = _get_extra_dispatch_keys(first_local_tensor) + + # Assert that all tensors have the same dtype, layout and dispatch keys. Due + # to uneven sharding, it is possible that tensors will have different shapes. + for local_tensor in it: + if dtype != local_tensor.dtype: + raise AssertionError( + "Tensors representing LocalTensor shards must have the same dtype" + ) + if layout != local_tensor.layout: + raise AssertionError( + "Tensors representing LocalTensor shards must have the same layout" + ) + if extra_dispatch_keys != _get_extra_dispatch_keys(local_tensor): + raise AssertionError( + "Tensors representing LocalTensor shards must have the " + "same set of extra dispatch keys" + ) + + # Compute shape/stride. We allow for non-SPMD'ness here + local_shapes: dict[int, dict[int, int]] = defaultdict(dict) # dim => rank => size + local_strides: dict[int, dict[int, int]] = defaultdict(dict) # dim => rank => size + for r, local_tensor in local_tensors.items(): + for d, size in enumerate(local_tensor.shape): + local_shapes[d][r] = size + local_strides[d][r] = local_tensor.stride(d) + shape = [ + ( + first_shape[d] + if _all_elements_same(list(local_shapes[d].values())) + else torch.SymInt(LocalIntNode(local_shapes[d])) + ) + for d in range(len(first_shape)) + ] + strides = [ + ( + first_stride[d] + if _all_elements_same(list(local_strides[d].values())) + else torch.SymInt(LocalIntNode(local_strides[d])) + ) + for d in range(len(first_shape)) + ] + return shape, strides, device, dtype, layout, extra_dispatch_keys + + +class LocalTensor(torch.Tensor): + """ + LocalTensor is a Tensor subclass that simulates a tensor distributed across multiple SPMD + (Single Program, Multiple Data) ranks. Each LocalTensor instance internally holds a mapping from + global rank ids to their corresponding local Tensor shards.Operations performed on a LocalTensor + are applied independently to each local shard, mimicking distributed computation. Collectives + and other distributed operations are handled by mapping them to the local shards as appropriate. + + Note: + This class is primarily intended for debugging and simulating distributed tensor computations + on a single process. + + """ + + # Map from global rank to the local tensor. + _local_tensors: dict[int, torch.Tensor] + # Precomputed for speed set of keys from the local tensor map. + _ranks: frozenset[int] + _size: list[torch.SymInt | int] + __slots__ = ["_local_tensors", "_ranks", "_size"] + + @staticmethod + @torch._disable_dynamo + def __new__( + cls, + local_tensors: dict[int, torch.Tensor], + requires_grad: bool = False, + ) -> "LocalTensor": + if any(t.requires_grad for t in local_tensors.values()): + raise AssertionError( + "Internal local_tensors require grad, but we will ignore those autograd graph. " + "Make a custom autograd function and make sure you detach the inner tensors." + ) + + if len(local_tensors) == 0: + raise ValueError("LocalTensor cannot be empty!") + + (shape, strides, device, dtype, layout, extra_dispatch_keys) = ( + _compute_local_tensor_meta(local_tensors) + ) + + r = torch.Tensor._make_wrapper_subclass( + cls, + shape, + strides=strides, + dtype=dtype, + device=device, + layout=layout, + # In place ops potentially change local tensor sizes (e.g. resize_). While + # executing an in-place op the return value must be the same as "self" input + # otherwise we can introduce errors due to tensor identity changes. Hence we + # need to be able to update wrapper subclass sizes after in-place ops. This + # dispatch policy allows us to do that. + dispatch_sizes_strides_policy="sizes", + requires_grad=requires_grad, + _extra_dispatch_keys=extra_dispatch_keys, + ) + # The wrapper has no real storage (data_ptr()=0). Prevent callers + # (e.g. Triton kernels) from silently reading the null pointer — + # turn it into a clear RuntimeError instead of a CUDA IMA. + torch._C._set_throw_on_mutable_data_ptr(r) + + local_tensors = { + r: v if not isinstance(v, AsyncCollectiveTensor) else v.wait() + for r, v in local_tensors.items() + } + r._local_tensors = local_tensors + r._ranks = frozenset(local_tensors.keys()) + r._size = shape + return r + + @torch._disable_dynamo + @mark_subclass_constructor_exportable_experimental # type: ignore[misc] + def __init__(self, *args: Any, **kwargs: Any): + super().__init__() + + def __deepcopy__(self, memo: dict[Any, Any] | None) -> "LocalTensor": + local_tensors_copy = { + r: copy.deepcopy(t, memo) for r, t in self._local_tensors.items() + } + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(local_tensors_copy, self.requires_grad) + + def __repr__(self) -> str: # type: ignore[override] + parts = [] + for k, v in self._local_tensors.items(): + parts.append(f" {k}: {v}") + tensors_str = ",\n".join(parts) + return f"LocalTensor(\n{tensors_str}\n)" + + def __getattr__(self, name: str) -> Any: + if _is_local_tensor_attr(name): + rank = _from_local_tensor_attr(name) + if rank not in self._ranks: + raise AttributeError(f"Local tensor has no knowledge of rank {rank}") + return self._local_tensors[rank] + return object.__getattribute__(self, name) + + def __tensor_flatten__(self) -> tuple[list[str], tuple[Any, ...]]: + """ + protocol to inform how to flatten a DTensor to local tensor + for PT2 tracing + """ + local_tensor_attrs = [_to_local_tensor_attr(r) for r in self._ranks] + return local_tensor_attrs, () + + @staticmethod + def __tensor_unflatten__( + inner_tensors: dict[str, Any], + flatten_spec: tuple[Any, ...], + outer_size: torch.Size, + outer_stride: tuple[int, ...], + ) -> "LocalTensor": + if flatten_spec is None: + raise AssertionError( + "Expecting spec to be not None from `__tensor_flatten__` return value!" + ) + local_tensors = { + _from_local_tensor_attr(a): t for a, t in inner_tensors.items() + } + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(local_tensors) + + @classmethod + @torch._disable_dynamo + def __torch_dispatch__( # type: ignore[override] + cls, + func: Any, + types: tuple[Any, ...], + args: tuple[Any, ...] = (), + kwargs: dict[str, Any] | None = None, + ) -> Any: + if kwargs is None: + kwargs = {} + + # This is horribly inefficient + flat_args, args_spec = pytree.tree_flatten((args, kwargs)) + local_tensor = None + for arg in flat_args: + if isinstance(arg, LocalTensor): + local_tensor = arg + break + + if local_tensor is None: + raise AssertionError("At least one of the arguments must be a LocalTensor") + + # Check for unrecognized tensor subclasses (but allow regular tensors and scalars) + has_unrecognized_types = _check_for_subclass(flat_args) + if has_unrecognized_types: + unrecognized_types = [ + type(x) for x in flat_args if _check_for_subclass_arg(x) + ] + not_implemented_log.debug( + "LocalTensor unrecognized subclass(es): %s", unrecognized_types + ) + return NotImplemented + + with LocalTensorMode(local_tensor._ranks): + return func(*args, **kwargs) + + def numpy(self, *, force: bool = False) -> Any: + if HAS_NUMPY: + return self.reconcile().numpy(force=force) + else: + raise RuntimeError("Numpy is not available") + + def contiguous( + self, + memory_format: torch.memory_format = torch.contiguous_format, + ) -> torch.Tensor: + return _LocalContiguous.apply(self, memory_format) + + def is_contiguous( + self, + memory_format: torch.memory_format = torch.contiguous_format, + ) -> bool: + return all( + t.is_contiguous(memory_format=memory_format) + for t in self._local_tensors.values() + ) + + def tolist(self) -> list[Any]: + """ + Try to reconcile, if successful convert to list, otherwise if dtype is integer, + convert to list of local integers. + """ + equal_obj = self._equal_local_tensors() + if isinstance(equal_obj, torch.Tensor): + return equal_obj.tolist() + if isinstance(equal_obj, torch.Size): + if not self.dtype.is_floating_point and not self.dtype.is_complex: + ranks = sorted(self._ranks) + local_lists = [self._local_tensors[r].tolist() for r in ranks] + return _reduce_multidim_lists( + local_lists, + lambda values: torch.SymInt( + LocalIntNode(dict(zip(ranks, values, strict=True))) + ), + ) + + raise RuntimeError("Cannot convert local tensor to list") + + def reconcile(self) -> torch.Tensor: + """ + Reconciles the LocalTensor into a single torch.Tensor by ensuring all local + shards are identical and returning a detached clone of one of them. + + Note: + This method is useful for extracting a representative tensor from a LocalTensor + when all shards are expected to be the same, such as after a collective operation + that synchronizes all ranks. + """ + + # Force all local tensor shards across ranks to be the same + equal_obj = self._equal_local_tensors() + if not isinstance(equal_obj, torch.Tensor): + raise AssertionError("LocalTensor shards must be the same to reconcile") + cl = equal_obj.clone().detach() + cl.requires_grad_(self.requires_grad) + return cl + + def _equal_local_tensors(self) -> torch.Tensor | torch.Size | None: + it = iter(self._local_tensors.values()) + t1 = next(it) + if all(t2.equal(t1) for t2 in it): + return t1 + if all(t2.shape == t1.shape for t2 in it): + return t1.shape + return None + + def _sync_meta(self) -> None: + with no_dispatch(): + (shape, strides, device, dtype, layout, extra_dispatch_keys) = ( + _compute_local_tensor_meta(self._local_tensors) + ) + self._size = shape + + +class _LocalContiguous(torch.autograd.Function): + """Autograd function for LocalTensor.contiguous() that preserves gradient flow.""" + + @staticmethod + def forward( # type: ignore[override] + ctx: torch.autograd.function.FunctionCtx, + input: LocalTensor, + memory_format: torch.memory_format, + ) -> LocalTensor: + # pyrefly: ignore [bad-argument-type] + return LocalTensor( + # pyrefly: ignore [bad-argument-count] + { + r: t.contiguous(memory_format=memory_format) + for r, t in input._local_tensors.items() + }, + input.requires_grad, + ) + + @staticmethod + def backward( # type: ignore[override] + ctx: torch.autograd.function.FunctionCtx, + grad_output: torch.Tensor, + ) -> tuple[torch.Tensor, None]: + return grad_output, None + + +# If set to `True` the LocalTensorMode stack will be created for the whole process, +# otherwise it will be created for each thread. +_PROCESS_MODE: bool = True +_PROCESS_LOCAL_TENSOR_MODE: list["LocalTensorMode"] = [] +# When running under local runner each thread must create its own local tensor mode +# so that they do not interfere with each other. +_THREAD_LOCAL_TENSOR_MODE: threading.local = threading.local() + + +def get_local_tensor_mode_list() -> list["LocalTensorMode"]: + global _PROCESS_MODE + if _PROCESS_MODE: + global _PROCESS_LOCAL_TENSOR_MODE + return _PROCESS_LOCAL_TENSOR_MODE + global _THREAD_LOCAL_TENSOR_MODE + if not hasattr(_THREAD_LOCAL_TENSOR_MODE, "value"): + _THREAD_LOCAL_TENSOR_MODE.value = [] + return _THREAD_LOCAL_TENSOR_MODE.value + + +# These methods are patched from DeviceMesh to the _LocalDeviceMesh versions. +_PATCHED_DEVICE_MESH_METHODS: Sequence[str] = ( + "get_coordinate", + "get_local_rank", + "get_rank", + "_is_current_rank_part_of_mesh", + "_sym_get_coordinate", +) + +# These random functions are also patched. +_PATCHED_RANDOM_FUNCTIONS: Sequence[tuple[str, str]] = ( + ("torch.random.manual_seed", "torch_manual_seed"), + ("torch.manual_seed", "torch_manual_seed"), + ("torch.random.initial_seed", "torch_initial_seed"), + ("torch.initial_seed", "torch_initial_seed"), +) + +_PATCHED_DIST_FUNCTIONS: Sequence[str] = ("get_rank",) + + +class LocalTensorMode(TorchDispatchMode): + """ + A TorchDispatchMode that simulates SPMD (Single Program, Multiple Data) execution + for LocalTensor objects across a set of ranks. + + LocalTensorMode enables PyTorch operations to be transparently applied to each + local shard of a LocalTensor, as if they were distributed across multiple ranks. + When active, this mode intercepts tensor operations and dispatches them to each + rank's local tensor, collecting and wrapping the results as LocalTensors. It also + handles collective operations by mapping them to local implementations. + + This mode is primarily intended for debugging and simulating distributed tensor + computations on a single process, rather than for high-performance distributed + training. It maintains a stack of active modes, patches DeviceMesh coordinate + resolution, and provides utilities for temporarily disabling the mode or mapping + functions over ranks. + """ + + @classmethod + def ignore_compile_internals(cls) -> bool: + return True + + # What ranks this local tensor mode is operating over + def __init__(self, ranks: int | frozenset[int]): + if isinstance(ranks, int): + # assume is world size + self.ranks = frozenset(range(ranks)) + else: + if not isinstance(ranks, frozenset): + raise AssertionError + self.ranks = ranks + self._disable = True + # Used to store the patched DeviceMesh methods + self._old_device_mesh_methods: dict[str, Callable[..., object]] | None = None + # Used to store the patched "random" functions + self._old_random_functions: dict[str, Callable[..., object]] = {} + # Used to store the patched dist functions + self._old_dist_functions: dict[str, Callable[..., object]] | None = None + self._per_rank_rng_states: dict[ + int, tuple[torch.Tensor, dict[int, torch.Tensor]] + ] = {} + # Cache for get_coordinate results, keyed by the inputs used to compute them: + # (ndim, flattened rank map, layout). + # Protected by _coordinate_cache_lock for thread safety in MPMD contexts. + self._coordinate_cache: dict[ + tuple[int, tuple, torch.distributed._mesh_layout._MeshLayout], list[SymInt] + ] = {} + self._coordinate_cache_lock = threading.Lock() + + def __enter__(self) -> "LocalTensorMode": + get_local_tensor_mode_list().append(self) + self.enable_() + + # _distribute_region will compute correct per-shard offsets + # but we want all ranks to start with the same state + if not _is_in_fake_tensor_mode(): + cpu_state, cuda_states = _get_rng_state() + for rank in self.ranks: + self._per_rank_rng_states[rank] = ( + cpu_state.clone(), + {idx: state.clone() for idx, state in cuda_states.items()}, + ) + + return super().__enter__() + + def __exit__( + self, + exc_type: type[BaseException] | None, + exc_val: BaseException | None, + exc_tb: TracebackType | None, + ) -> None: + local_tensor_mode_list = get_local_tensor_mode_list() + local_tensor_mode_list.pop() + self.disable_() + if len(local_tensor_mode_list) > 0: + if local_tensor_mode_list[-1]._disable: + local_tensor_mode_list[-1].disable_() + else: + local_tensor_mode_list[-1].enable_() + super().__exit__(exc_type, exc_val, exc_tb) + + def __torch_dispatch__( + self, + func: Any, + types: tuple[Any, ...], + args: tuple[Any, ...] = (), + kwargs: dict[str, Any] | None = None, + ) -> Any: + if kwargs is None: + kwargs = {} + + flat_args, args_spec = pytree.tree_flatten((args, kwargs)) + + # Find all LocalTensor arguments to determine ranks + local_tensors = [a for a in flat_args if isinstance(a, LocalTensor)] + + # Check for unrecognized tensor subclasses (but allow regular tensors and scalars) + has_unrecognized_types = _check_for_subclass(flat_args) + if has_unrecognized_types: + unrecognized_types = [ + type(x) for x in flat_args if _check_for_subclass_arg(x) + ] + not_implemented_log.debug( + "LocalTensorMode unrecognized subclass(es): %s", unrecognized_types + ) + return NotImplemented + + # device_mesh::_runtime_compute_coordinate_on_dim is checked before the + # no-LocalTensor early return because compile-on-one-rank calls it with + # only a plain mesh-tensor argument, but the result must still vary per + # simulated rank under LocalTensorMode. + if ( + not self._disable + and func.namespace == "device_mesh" + and func is torch.ops.device_mesh._runtime_compute_coordinate_on_dim.default + ): + from torch.fx.operator_schemas import normalize_function + + normalized = normalize_function( + func, + args, + kwargs, + normalize_to_only_use_kwargs=True, + ) + if normalized is None: + raise AssertionError( + "LocalTensorMode could not normalize arguments for " + "device_mesh::_runtime_compute_coordinate_on_dim" + ) + full_mesh = normalized.kwargs["full_mesh"] + index = normalized.kwargs["index"] + rank_results: dict[int, int] = {} + for r in self.ranks: + mesh_tensor = DeviceMesh._get_mesh_tensor_from_full_mesh( + full_mesh, current_rank=r + ) + mesh_coords = DeviceMesh._compute_coordinates_from_mesh(mesh_tensor, r) + if mesh_coords is None: + raise RuntimeError( + f"Rank {r} not found in mesh tensor for " + "_runtime_compute_coordinate_on_dim" + ) + rank_results[r] = mesh_coords[index] + return _combine_int_rank_results(rank_results) + + # Factory functions convert into LocalTensor, so we don't have to + # transmute a Tensor into a LocalTensor if mutation happens... + # But if you do an operation on a Tensor, do NOT wrap it into a + # LocalTensor. This helps prevent accidents when you're doing Tensor + # operations on the inner non-wrapped tensors. + if not local_tensors: + if self._disable or any(isinstance(a, Tensor) for a in flat_args): + return func(*args, **kwargs) + + # For LocalTensors, verify they have compatible ranks + for a in flat_args: + if isinstance(a, LocalTensor): + if a._ranks > self.ranks: + raise AssertionError( + f"Input LocalTensor {a} must be configured for a " + f"subset of the LocalTensorMode ranks {self.ranks}" + ) + + if func.overloadpacket == torch.ops.aten.dim: + return len(args[0]._size) + if func.overloadpacket == torch.ops.aten.sym_size: + return tuple(args[0]._size) + + if func.namespace == "c10d": + if func is torch.ops.c10d.allreduce_.default: + return _c10d._local_all_reduce_(*args, **kwargs) + elif func is torch.ops.c10d.allreduce_coalesced_.default: + return _c10d._local_allreduce_coalesced_(*args, **kwargs) + elif func is torch.ops.c10d.reduce_scatter_tensor_coalesced_.default: + return _c10d._local_reduce_scatter_tensor_coalesced_(*args, **kwargs) + elif func is torch.ops.c10d.scatter_.default: + return _c10d._local_scatter_(*args, **kwargs) + elif func is torch.ops.c10d.broadcast_.default: + return _c10d._local_broadcast_(*args, **kwargs) + elif func is torch.ops.c10d.allgather_.default: + return _c10d._local_all_gather_(*args, **kwargs) + elif func is torch.ops.c10d.allgather_into_tensor_coalesced_.default: + return _c10d._local_allgather_into_tensor_coalesced_(*args, **kwargs) + elif func is torch.ops.c10d._allgather_base_.default: + return _c10d._local_allgather_base_(*args, **kwargs) + elif func is torch.ops.c10d._reduce_scatter_base_.default: + return _c10d._local_reduce_scatter_base_(*args, **kwargs) + elif func is torch.ops.c10d.gather_.default: + return _c10d._local_gather_(*args, **kwargs) + elif func is torch.ops.c10d.alltoall_.default: + return _c10d._local_alltoall_(*args, **kwargs) + elif func is torch.ops.c10d.alltoall_base_.default: + return _c10d._local_alltoall_base_(*args, **kwargs) + elif func is torch.ops.c10d.barrier.default: + return _c10d._local_barrier(*args, **kwargs) + elif func is torch.ops.c10d.monitored_barrier_.default: + return _c10d._local_monitored_barrier_(*args, **kwargs) + elif func is torch.ops.c10d.send.default: + return _c10d._local_send(*args, **kwargs) + elif func is torch.ops.c10d.recv_.default: + return _c10d._local_recv_(*args, **kwargs) + elif func is torch.ops.c10d.recv_any_source_.default: + return _c10d._local_recv_any_source_(*args, **kwargs) + raise NotImplementedError(f"{func} not implemented") + + if func.namespace == "_c10d_functional" or func.namespace == "_dtensor": + if func is torch.ops._dtensor.shard_dim_alltoall.default: + return _c10d._local_functional_shard_dim_alltoall(*args, **kwargs) + elif func is torch.ops._c10d_functional.all_gather_into_tensor.default: + return _c10d._local_functional_all_gather_into_tensor(*args, **kwargs) + elif func is torch.ops._c10d_functional.reduce_scatter_tensor.default: + return _c10d._local_functional_reduce_scatter_tensor(*args, **kwargs) + elif func is torch.ops._c10d_functional.all_to_all_single.default: + return _c10d._local_functional_all_to_all_single(*args, **kwargs) + else: + with LocalTensorMode(self.ranks): + return func._op_dk( + DispatchKey.CompositeExplicitAutograd, *args, **kwargs + ) + + if func.namespace == "profiler": + return func(*args, **kwargs) + + if func.namespace == "_c10d_functional_autograd": + raise NotImplementedError(f"{func} not implemented") + + if func.namespace == "symm_mem": + raise NotImplementedError(f"{func} not implemented") + + return _for_each_rank_run_func(func, self.ranks, args, kwargs, alias=True) + + def disable_(self): + if self._disable: + return + + self._unpatch_device_mesh() + self._unpatch_random_functions() + self._unpatch_dist() + self._disable = True + + def enable_(self): + if not self._disable: + return + + self._patch_device_mesh() + self._patch_random_functions() + self._patch_dist() + self._disable = False + + @contextlib.contextmanager + def disable(self) -> Generator[None, None, None]: + """ + Disables LocalTensorMode temporarily. Primarily is intended to be used to perform + rank specific computations and merge results back before enabling LocalTensorMode back. + """ + + # don't unpatch again if already disabled + if self._disable: + try: + yield + finally: + # re-disable if the yield messed + # with the state + self.disable_() + return + + self.disable_() + try: + yield + finally: + self.enable_() + + def rank_map(self, cb: Callable[[int], Tensor]) -> LocalTensor: + """ + Creates a LocalTensor instance by mapping rank id to its local shard. + """ + + with self.disable(): + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor({r: cb(r) for r in self.ranks}) + + def tensor_map( + self, tensor: LocalTensor, cb: Callable[[int, Tensor], Tensor | None] + ) -> LocalTensor: + """ + Creates a LocalTensor instance by mapping rank id to its local shard. + """ + + with self.disable(): + results = {} + for r in self.ranks: + if r in tensor._local_tensors: + m = cb(r, tensor._local_tensors[r]) + if m is not None: + results[r] = m + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(results) + + def _any_local_rng_state(self) -> tuple[torch.Tensor, dict[int, torch.Tensor]]: + return self._per_rank_rng_states[next(iter(self.ranks))] + + def _patch_device_mesh(self) -> None: + if self._old_device_mesh_methods is not None: + raise AssertionError + saved = {} + for name in _PATCHED_DEVICE_MESH_METHODS: + saved[name] = getattr(DeviceMesh, name) + local = getattr(_LocalDeviceMesh, name) + setattr(DeviceMesh, name, local) + self._old_device_mesh_methods = saved + + def _unpatch_device_mesh(self) -> None: + saved, self._old_device_mesh_methods = self._old_device_mesh_methods, None + if saved is None: + raise AssertionError + for name, value in saved.items(): + setattr(DeviceMesh, name, value) + + def _patch_random_functions(self) -> None: + # TODO: This should either be removed or documented why it's necessary. + from torch.distributed.tensor import _random as dtensor_random + + for global_name, local_name in _PATCHED_RANDOM_FUNCTIONS: + if global_name in self._old_random_functions: + continue + mod_name, attr_name = global_name.rsplit(".", 1) + mod = importlib.import_module(mod_name) + old = getattr(mod, attr_name) + local = getattr(_LocalRandom, local_name) + setattr(mod, attr_name, local) + self._old_random_functions[global_name] = old + + def _unpatch_random_functions(self) -> None: + # TODO: This should either be removed or documented why it's necessary. + from torch.distributed.tensor import _random as dtensor_random + + for global_name, local_name in _PATCHED_RANDOM_FUNCTIONS: + value = self._old_random_functions.pop(global_name, None) + if value is not None: + mod_name, attr_name = global_name.rsplit(".", 1) + mod = importlib.import_module(mod_name) + setattr(mod, attr_name, value) + + def _patch_dist(self) -> None: + if self._old_dist_functions is not None: + raise AssertionError + import torch.distributed as dist + + saved = {} + for name in _PATCHED_DIST_FUNCTIONS: + saved[name] = getattr(dist, name) + local = getattr(_LocalDist, name) + setattr(dist, name, local) + self._old_dist_functions = saved + + def _unpatch_dist(self) -> None: + saved, self._old_dist_functions = self._old_dist_functions, None + if saved is None: + raise AssertionError + import torch.distributed as dist + + for name, value in saved.items(): + setattr(dist, name, value) + + +class _LocalRandom: + """ + Holds implementations of random functionality that must be patched while running + under LocalTensorMode. + """ + + @staticmethod + def torch_manual_seed(seed) -> torch._C.Generator: + """LocalTensor-aware version of torch.random.manual_seed.""" + if ( + (lm := enabled_local_tensor_mode()) + and isinstance(seed, torch.SymInt) + and isinstance(seed.node, LocalIntNode) + ): + from torch.random import _manual_seed_impl + + for rank in sorted(lm.ranks): + rank_seed = seed.node._local_ints[rank] + _manual_seed_impl(rank_seed) + lm._per_rank_rng_states[rank] = _get_rng_state() + return torch.random.default_generator + from torch.random import _manual_seed_impl + + result = _manual_seed_impl(seed) + + if lm is not None and len(lm._per_rank_rng_states) > 0: + cpu_state, cuda_states = _get_rng_state() + for rank in lm.ranks: + lm._per_rank_rng_states[rank] = ( + cpu_state.clone(), + {idx: state.clone() for idx, state in cuda_states.items()}, + ) + + return result + + @staticmethod + def torch_initial_seed(): + """LocalTensor-aware version of torch.random.initial_seed.""" + if lm := enabled_local_tensor_mode(): + if len(lm._per_rank_rng_states) == 0: + return torch.random.default_generator.initial_seed() + rank_seeds = {} + + for rank in sorted(lm.ranks): + _set_rng_state(*lm._per_rank_rng_states[rank]) + rank_seeds[rank] = torch.random.default_generator.initial_seed() + + local_int_node = LocalIntNode(rank_seeds) + return torch.SymInt(local_int_node) + + return torch.random.default_generator.initial_seed() + + +# Save the original get_coordinate method before any patching + + +class _LocalDeviceMesh: + """ + Holds implementations of DeviceMesh functionality that must be patched while running + under LocalTensorMode. + """ + + @staticmethod + def get_coordinate(self: DeviceMesh) -> list[SymInt] | None: + # NB: In order to support submeshes the code below recreates for each + # rank submesh with the same mesh dimensions as current mesh. We are + # doing this because when submesh is created it is created for a particular + # rank (therefore below we are patching get_rank method). We are trying to + # limit the invasiveness of local tensor. + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError("Unexpectedly not in LocalTensorMode") + + # Include all attributes used below in the cache key + cache_key = (self.ndim, self._flatten_rank_map, self._layout) + # Check cache first (fast path without lock) + if cache_key in lm._coordinate_cache: + return lm._coordinate_cache[cache_key] + + # Acquire lock for thread safety in MPMD contexts + with lm._coordinate_cache_lock: + # Double-check after acquiring lock + if cache_key in lm._coordinate_cache: + return lm._coordinate_cache[cache_key] + + coords: list[dict[int, int]] = [{} for _ in range(self.ndim)] + # Clone rank_map to avoid "Cannot set version_counter for inference tensor" + # error when running under torch.inference_mode() + rank_map = self._rank_map.clone() + for r in lm.ranks: + rank_tensor = self._layout.remap_to_tensor(rank_map) + rank_coords = (rank_tensor == r).nonzero().tolist() + if len(rank_coords) != 1: + raise AssertionError + for d, c in enumerate(rank_coords[0][1:]): + coords[d][r] = c + + out = [torch.SymInt(LocalIntNode(c)) for c in coords] + # Cache the result + lm._coordinate_cache[cache_key] = out + # The output contains coordinates for each of the ranks with respect to + # their meshes formed from root mesh and selecting the same dimensions + # as the current mesh. + return out # type: ignore[return-value] + + @staticmethod + def _is_current_rank_part_of_mesh(self: DeviceMesh) -> bool: + my_coordinate = self.get_coordinate() + return my_coordinate is not None + + @staticmethod + def _sym_get_coordinate(self: DeviceMesh, index: int) -> int: + my_coordinate = self.get_coordinate() + if my_coordinate is None: + raise AssertionError + return my_coordinate[index] + + @staticmethod + def get_rank(self) -> int | SymInt: + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError("Unexpectedly not in LocalTensorMode") + return torch.SymInt(LocalIntNode(local_ints={r: r for r in lm.ranks})) + + @staticmethod + def get_local_rank(self, mesh_dim: int | str | None = None) -> int | SymInt: + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError("Unexpectedly not in LocalTensorMode") + + if self.ndim > 1 and mesh_dim is None: + raise RuntimeError( + f"Found the DeviceMesh have {self.ndim} dimensions", + "Optional kwarg `mesh_dim` needs to be specified when device_mesh.ndim > 1.", + ) + elif mesh_dim is None: + mesh_dim = 0 + + if isinstance(mesh_dim, str): + mesh_dim = self._mesh_dim_names.index(mesh_dim) + + # Compute local rank for each global rank + # get_coordinate returns a list of SymInt, one per mesh dimension + # We need to extract the coordinate for the specified mesh_dim + coords = _LocalDeviceMesh.get_coordinate(self) + if coords is None: + raise AssertionError + return coords[mesh_dim] + + +class _LocalDist: + """ + Holds implementations of torch.distributed functions that must be patched while + running under LocalTensorMode. + """ + + @staticmethod + def get_rank(group: "ProcessGroup | None" = None) -> "int | SymInt": + import torch.distributed as dist + + lm = enabled_local_tensor_mode() + if lm is None: + raise AssertionError("Unexpectedly not in LocalTensorMode") + if group is None or group is dist.GroupMember.WORLD: + return torch.SymInt(LocalIntNode({r: r for r in lm.ranks})) + from ._c10d import _prepare_collective_groups + + ranks, group_offsets, _ = _prepare_collective_groups(group) + mapping: dict[int, int] = {} + for group_offset in group_offsets: + for local_rank, r in enumerate(ranks): + mapping[group_offset + r] = local_rank + return torch.SymInt(LocalIntNode(mapping)) + + +def reconcile_args(args: Any, kwargs: dict[str, Any] | None = None) -> Any: + """ + Reconciles arguments by converting any LocalTensor instances in the input + arguments to their underlying torch.Tensor representation. + + This function is typically used to prepare arguments for functions that + expect standard torch.Tensor objects, by flattening the input arguments, + replacing LocalTensor instances with their reconciled (standard tensor) + versions, and then reconstructing the original argument structure. + + Args: + args: Positional arguments, possibly containing LocalTensor instances. + kwargs: Keyword arguments, possibly containing LocalTensor instances. + + Returns: + Any: The arguments with all LocalTensor instances replaced by their reconciled torch.Tensor equivalents, + preserving the original structure. + """ + if kwargs is None: + kwargs = {} + flat_args, args_spec = pytree.tree_flatten((args, kwargs)) + reconciled_args = [ + a.reconcile() if isinstance(a, LocalTensor) else a for a in flat_args + ] + return pytree.tree_unflatten(reconciled_args, args_spec) + + +def local_tensor_mode() -> LocalTensorMode | None: + """ + Returns the current active LocalTensorMode if one exists. + + This function checks the global stack of LocalTensorMode instance. If there + is at least one LocalTensorMode active, it returns the most recently entered + (top of the stack) LocalTensorMode. If no LocalTensorMode is active, it returns None. + + Returns: + Optional[LocalTensorMode]: The current LocalTensorMode if active, else None. + """ + local_tensor_mode_list = get_local_tensor_mode_list() + if len(local_tensor_mode_list) > 0: + return local_tensor_mode_list[-1] + return None + + +def enabled_local_tensor_mode() -> LocalTensorMode | None: + """ + Returns the current active LocalTensorMode only if it's enabled. + + This is a convenience function that combines the common pattern of checking + if local_tensor_mode() is not None and not disabled. + + Returns: + Optional[LocalTensorMode]: The current LocalTensorMode if active and enabled, else None. + """ + lm = local_tensor_mode() + if lm is not None and not lm._disable: + return lm + return None + + +def maybe_run_for_local_tensor(func: Callable[_P, _R]) -> Callable[_P, _R]: + """ + Decorator that ensures a function is executed for each local tensor shard + when running under LocalTensorMode. If not in LocalTensorMode, the function + is executed normally. When in LocalTensorMode, the function is run for each + rank, and the results are collected appropriately. + + This decorator is useful for functions that exhibit non-SPMD behavior, such + as those requiring rank specific actions. For example, a function that computes + offset into input tensor based on rank. + + Note that the function being decorated must not have any side effects and + contain operations for a single rank only. For example, wrapping a function + that performs a collective operation will not work. + + Args: + func (Callable[..., Any]): The function to be decorated. + + Returns: + Callable[..., Any]: The wrapped function that handles LocalTensorMode logic. + """ + + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _R: + if not (lm := enabled_local_tensor_mode()): + return func(*args, **kwargs) + ret = None + with lm.disable(): + ret = _for_each_rank_run_func(func, lm.ranks, args, kwargs, alias=False) + + return ret + + return wrapper + + +def rank_map(cb: Callable[[int], Tensor]) -> Tensor: + """ + Creates a tensor by mapping a callback over the current rank. + + Under LocalTensorMode, calls cb(rank) for each simulated rank and returns + a LocalTensor. In real distributed (no LocalTensorMode), calls + cb(dist.get_rank()) and returns a plain Tensor. + """ + lm = enabled_local_tensor_mode() + if lm is not None: + return lm.rank_map(cb) + else: + return cb(dist.get_rank()) + + +def tensor_map(tensor: Tensor, cb: Callable[[int, Tensor], Tensor | None]) -> Tensor: + """ + Transforms a tensor by mapping a callback over the current rank and its + local shard. + + Under LocalTensorMode, calls cb(rank, shard) for each simulated rank and + returns a LocalTensor. In real distributed (no LocalTensorMode), calls + cb(dist.get_rank(), tensor) and returns the result directly. + """ + lm = enabled_local_tensor_mode() + if lm is not None: + if not isinstance(tensor, LocalTensor): + raise AssertionError(f"Expected LocalTensor, got {type(tensor)}") + return lm.tensor_map(tensor, cb) + else: + r = cb(dist.get_rank(), tensor) + if r is None: + raise AssertionError("callback returned None") + return r + + +def maybe_disable_local_tensor_mode() -> contextlib.AbstractContextManager: + """ + Context manager that disables LocalTensorMode for the duration of the context. + """ + lm = local_tensor_mode() + return lm.disable() if lm is not None else contextlib.nullcontext() + + +def maybe_enable_local_tracker( + device_type: str, distribute_region_enabled: bool, spec, generator +): + """ + Returns a context manager for LocalTensor-mode RNG tracking if local tensor mode is enabled. + + Args: + device_type: The device type (e.g., "cuda", "cpu") + distribute_region_enabled: Whether distribute region is enabled + spec: The DTensorSpec + generator: Optional torch.Generator + + Returns: + Context manager from local_tracker._distribute_region if local tensor mode is enabled, + otherwise None. + """ + if enabled_local_tensor_mode(): + local_tracker = _LocalOffsetBasedRNGTracker(device_type) + local_tracker.distribute_region_enabled = distribute_region_enabled + return local_tracker._distribute_region(spec, generator) + + return None + + +def get_generator_seed_for_device_type(device_type: str): + """ + Gets the generator seed for a specific device type, handling LocalTensor mode appropriately. + + Args: + device_type: The device type (e.g., "cuda", "cpu") + + Returns: + If in LocalTensor mode with per-rank RNG states: + - Returns int if all ranks have the same seed + - Returns SymInt(LocalIntNode) if ranks have different seeds + Otherwise: + - Returns int seed from the device's RNG state + """ + if lm := enabled_local_tensor_mode(): + if len(lm._per_rank_rng_states) == 0: + device_module = torch.get_device_module(device_type) + return device_module.get_rng_state()[:8].view(torch.int64).item() + device_module = torch.get_device_module(device_type) + + original_state = _get_rng_state() + + rank_seeds = {} + try: + for rank in sorted(lm.ranks): + _set_rng_state(*lm._per_rank_rng_states[rank]) + rank_seeds[rank] = int( + device_module.get_rng_state()[:8].view(torch.int64).item() + ) + finally: + # restore original state + _set_rng_state(*original_state) + + unique_seeds = set(rank_seeds.values()) + if len(unique_seeds) == 1: + return next(iter(unique_seeds)) + local_int_node = LocalIntNode(rank_seeds) + return torch.SymInt(local_int_node) + else: + device_module = torch.get_device_module(device_type) + return device_module.get_rng_state()[:8].view(torch.int64).item() + + +import threading +from queue import Queue + + +_LOCAL_RUNNER_MODE: "LocalRunnerMode | None" = None + + +class _ExceptionRaisingThread(threading.Thread): + def __init__( + self, group=None, target=None, name=None, args=(), kwargs=None, *, daemon=None + ): + super().__init__( + target=target, name=name, args=args, kwargs=kwargs, daemon=daemon + ) + self.exception: BaseException | None = None + + def run(self): + try: + super().run() + except BaseException as e: + self.exception = e + + def join(self, timeout=None): + super().join(timeout=timeout) + if self.exception: + raise self.exception + + +class LocalRunnerMode: + """ + A class for running multiple SPMD functions concurrently, however at any point + in time only one function can be running. The main use case for the local runner + mode is to enable SPMD functions to be able to use send and recv to communicate + with each other. Without local runner mode send and recv are not supported. + """ + + runner_context = threading.local() + + def __init__( + self, ranks: frozenset[int] | int, concurrency: int, fn: Callable[[int], None] + ): + if isinstance(ranks, int): + ranks = frozenset(range(ranks)) + self._ranks = ranks + self._fn = fn + self._run_lock = threading.Lock() + self._run_id = -1 + self._run_cond = threading.Condition(self._run_lock) + + self._recv_objects: dict[int, dict[int, Queue]] = { + dst: {src: Queue() for src in ranks} for dst in ranks + } + self._runners = [ + _ExceptionRaisingThread(target=self._run, args=(i,), name="LocalRunnerMode") + for i in range(concurrency) + ] + self._process_mode = True + + def __enter__(self) -> "LocalRunnerMode": + global _LOCAL_RUNNER_MODE + if _LOCAL_RUNNER_MODE is not None: + raise AssertionError("LocalRunnerMode is already running") + _LOCAL_RUNNER_MODE = self + + global _PROCESS_MODE + self._process_mode = _PROCESS_MODE + _PROCESS_MODE = False + for r in self._runners: + r.start() + return self + + def __exit__( + self, + exc_type: type[BaseException] | None, + exc_val: BaseException | None, + exc_tb: TracebackType | None, + ) -> None: + for r in self._runners: + r.join() + global _LOCAL_RUNNER_MODE + _LOCAL_RUNNER_MODE = None + + global _PROCESS_MODE + _PROCESS_MODE = self._process_mode + + def _run(self, id: int) -> None: + LocalRunnerMode.runner_context.id = id + # Only one thread can run at a time, hence must acquire the lock + try: + self._acquire_run_lock() + self._fn(id) + finally: + self._release_run_lock() + + def _acquire_run_lock(self) -> None: + self._run_lock.acquire() + self._run_id = LocalRunnerMode.runner_context.id + + def _release_run_lock(self) -> None: + self._run_id = -1 + self._run_lock.release() + + def _assert_holds_run_lock(self) -> None: + if self._run_id != LocalRunnerMode.runner_context.id: + raise AssertionError("Calling thread does not hold the run lock") + + def _get_recv_object(self, src: int, dst: int) -> object | None: + peers = [src] if src != -1 else list(self._ranks) + recv_objects = self._recv_objects[dst] + + for p in peers: + if not recv_objects[p].empty(): + return recv_objects[p].get() + + return None + + def _signal_send(self, src: int, dst: int, obj: object) -> None: + if obj is None: + raise AssertionError("Cannot signal None") + # Only a single thread a time executes so it is safe to mutate + # read objects queue (executing thread is already holding the lock) + self._recv_objects[dst][src].put(obj) + # Signal directly condition variable since the calling thread is already + # holding the lock + self._run_cond.notify_all() + + def _wait_recv(self, src: int, dst: int, post: Callable[[object], None]) -> None: + # Wait for the object to be available + while True: + obj = self._get_recv_object(src, dst) + if obj is not None: + post(obj) + # Note that we are not releasing the lock here, since the thread + # will continue to run and therefore must hold the lock + return + self._run_cond.wait() + + @staticmethod + def current() -> "LocalRunnerMode": + global _LOCAL_RUNNER_MODE + if _LOCAL_RUNNER_MODE is None: + raise AssertionError("LocalRunnerMode is not enabled") + return _LOCAL_RUNNER_MODE + + +class _LocalPhiloxState: + """ + LocalTensor-aware version of _PhiloxState that manages per-rank RNG states. + This class handles the case where the generator state is a LocalTensor, allowing + different offsets and seeds for different virtual ranks. + + Note: This is designed to be used as a drop-in replacement for _PhiloxState + when working with LocalTensors in the DTensor random ops implementation. + """ + + def __init__(self, state: torch.Tensor): + if not isinstance(state, LocalTensor): + raise AssertionError("_LocalPhiloxState requires a LocalTensor") + self._local_tensor = state + self._per_rank_states = { + rank: local_state.to("cpu") + for rank, local_state in state._local_tensors.items() + } + + @property + def state(self): + return LocalTensor(self._per_rank_states) # type: ignore[name-defined] + + @property + def offset(self) -> int | SymInt: + from torch.distributed.tensor._random import _PhiloxState + + offsets = {} + for rank, state in self._per_rank_states.items(): + rank_philox = _PhiloxState(state) + offsets[rank] = int(rank_philox.offset.item()) + + if len(set(offsets.values())) == 1: + return next(iter(offsets.values())) + + return SymInt(LocalIntNode(offsets)) + + @offset.setter + def offset(self, offset: int | SymInt) -> None: + from torch.distributed.tensor._random import _PhiloxState + + if isinstance(offset, SymInt) and isinstance(offset.node, LocalIntNode): + for rank, state in self._per_rank_states.items(): + rank_offset = offset.node._local_ints[rank] + rank_philox = _PhiloxState(state) + rank_philox.offset = torch.tensor([rank_offset], dtype=torch.int64) + else: + offset_int = int(offset) if isinstance(offset, SymInt) else offset + offset_tensor = torch.tensor([offset_int], dtype=torch.int64) + for state in self._per_rank_states.values(): + rank_philox = _PhiloxState(state) + rank_philox.offset = offset_tensor + + @property + def seed(self) -> int | SymInt: + from torch.distributed.tensor._random import _PhiloxState + + seeds = {} + for rank, state in self._per_rank_states.items(): + rank_philox = _PhiloxState(state) + seeds[rank] = int(rank_philox.seed.item()) + + if len(set(seeds.values())) == 1: + return next(iter(seeds.values())) + return SymInt(LocalIntNode(seeds)) + + @seed.setter + def seed(self, seed: int | SymInt) -> None: + from torch.distributed.tensor._random import _PhiloxState + + if isinstance(seed, SymInt) and isinstance(seed.node, LocalIntNode): + for rank, state in self._per_rank_states.items(): + rank_seed = seed.node._local_ints[rank] + rank_philox = _PhiloxState(state) + rank_philox.seed = torch.tensor([rank_seed], dtype=torch.int64) + else: + seed_int = int(seed) if isinstance(seed, SymInt) else seed + seed_tensor = torch.tensor([seed_int], dtype=torch.int64) + for state in self._per_rank_states.values(): + rank_philox = _PhiloxState(state) + rank_philox.seed = seed_tensor + + def apply_to_local_tensor_mode(self, device_handle) -> None: + """ + Apply per-rank RNG states to the LocalTensorMode's tracked states. + This updates both the device RNG state and the LocalTensorMode's _per_rank_rng_states. + + Args: + device_handle: The device handle to use for setting RNG state (_LocalDeviceHandle) + """ + if not enabled_local_tensor_mode(): + return + + if not hasattr(self, "_per_rank_offsets"): + raise AssertionError + + for rank in sorted(self._per_rank_states.keys()): + offset_value = self._per_rank_offsets[rank] + if isinstance(offset_value, SymInt): + if isinstance(offset_value.node, LocalIntNode): + offset_value = offset_value.node._local_ints[rank] + else: + offset_value = int(offset_value) + + offset_tensor = torch.tensor( + [offset_value], dtype=torch.uint64, device="cpu" + ).view(torch.uint8) + self._per_rank_states[rank][8:] = offset_tensor + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + device_handle.set_rng_state(LocalTensor(self._per_rank_states)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d13e8286e72d9c1fc05756126ecf784db96897e8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/__init__.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96f8a4c7437eba21f00d12064ac625e8d7e15f0c6469a7e0a0df8238ef2626a2 +size 108540 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/_c10d.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/_c10d.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0f8d1c7598a139fc5df23f46a7d627fe65b2b80b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/__pycache__/_c10d.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/_c10d.py b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/_c10d.py new file mode 100644 index 0000000000000000000000000000000000000000..0060ed80383cbdf91df6c77db0b94242ed7dc442 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_local_tensor/_c10d.py @@ -0,0 +1,1124 @@ +import functools +import math +import operator +from collections.abc import Callable, Sequence +from datetime import timedelta + +import torch +from torch._C import ScriptObject +from torch._C._distributed_c10d import FakeWork, PythonCallbackWork +from torch.distributed._mesh_layout import _FlatLayout +from torch.distributed.distributed_c10d import ( + _check_op, + _get_default_group, + _resolve_process_group, + GroupName, + ProcessGroup, + ReduceOp, + Work, +) + + +# NOTE: Most of the c10d collectives often take a Tensor[] (or Tensor[][]) +# when you would expect Tensor (or Tensor[]). In fact, there will only ever +# be one Tensor in this case; the old signature was to support dispatching a +# collective on multiple devices (ala DataParallel) but we don't support that +# API anymore. Note that we are not 100% consistent about this; some more +# modern collectives like _allgather_base_ got rid of the unnecessary list. +# When in doubt, consult the code that dispatches to the collective on the PG +# in distributed_c10d.py e.g., work = group.allgather([tensor_list], [tensor], +# opts) indicates it's always a list. + + +def _gcd_list(numbers: Sequence[int]) -> int: + return 0 if not numbers else functools.reduce(math.gcd, numbers) + + +def _indices_to_layout(indices: list[int]) -> tuple[tuple[int, ...], tuple[int, ...]]: + # Base case: A single index represents a point, not a dimension. + if len(indices) <= 1: + return (), () + + # The smallest stride is likely the GCD of the differences between consecutive indices. + # For a sorted, unique list, all differences will be positive. + diffs = [indices[i] - indices[i - 1] for i in range(1, len(indices))] + last_stride = _gcd_list(diffs) + + # This case should not be reached if indices are unique and sorted. + if last_stride == 0: + raise AssertionError("Cannot determine stride; indices may not be unique.") + + # Identify the starting index of each "row" in the last dimension. + # An index starts a new row if the preceding index (index - stride) is not present. + indices_set = set(indices) + higher_dim_indices = [indices[0]] + for index in indices[1:]: + if (index - last_stride) not in indices_set: + higher_dim_indices.append(index) + + # From the number of rows, we can deduce the shape of the last dimension. + if len(indices) % len(higher_dim_indices) != 0: + raise AssertionError( + "Indices do not form a regular grid. " + f"Found {len(higher_dim_indices)} subgroups for {len(indices)} total elements." + ) + last_shape = len(indices) // len(higher_dim_indices) + + # Recurse on the higher-dimensional indices (the start of each row). + higher_shapes, higher_strides = _indices_to_layout(higher_dim_indices) + + # Combine the results from the recursion with the current dimension's results. + final_shapes = higher_shapes + (last_shape,) + final_strides = higher_strides + (last_stride,) + + return final_shapes, final_strides + + +def _resolve_pg_or_name( + group: ScriptObject | ProcessGroup | str | GroupName, +) -> ScriptObject | ProcessGroup: + """Accept any of the group reps that may reach a functional collective. + + Under ``compile_on_one_rank=True`` the functional collective ops receive + the live ``ProcessGroup`` instead of a string group name, so we cannot + unconditionally call ``_resolve_process_group``. + """ + if isinstance(group, (ProcessGroup, ScriptObject)): + return group + # ``GroupName`` is ``NewType("GroupName", str)``, so wrapping ``group`` here + # is a runtime no-op. The cast exists purely so pyrefly accepts the + # narrowed ``str | GroupName`` argument against the declared parameter type. + return _resolve_process_group(GroupName(group)) + + +def _prepare_collective_groups( + process_group_so: ScriptObject | ProcessGroup, +) -> tuple[list[int], list[int], int]: + process_group = ( + ProcessGroup.unbox(process_group_so) + if isinstance(process_group_so, ScriptObject) + else process_group_so + ) + + ranks = torch.distributed.get_process_group_ranks(process_group) + if not ranks: + raise AssertionError + # TODO: We can handle permutations but the layout inference algorithm will + # lose the permutation so we will have to reapply it + if ranks != sorted(ranks): + raise AssertionError(ranks) + offset = ranks[0] + ranks = [r - offset for r in ranks] + + shape, strides = _indices_to_layout(ranks) + layout = _FlatLayout(shape, strides) + + global_pg = _get_default_group() + group_offsets = layout.complement(global_pg.size()).all_ranks_from_zero() + + return ranks, group_offsets, offset + + +# NB: There are two flavors of the collectives: regular and functional. Regular collectives +# allocate outputs to write the result to, accept process group and support async ops (return +# work object). Functional collectives expect the implementation to allocate outputs, accept +# process group name that must be resolved and do not support async ops (return output). +def _local_functional_all_gather_into_tensor( + tensor: torch.Tensor, + group_size: int, + group_name: GroupName | ProcessGroup, +) -> torch.Tensor: + # "all_gather_into_tensor(Tensor input, int group_size, str group_name) -> Tensor" + from . import LocalTensor + + ranks, group_offsets, offset = _prepare_collective_groups( + _resolve_pg_or_name(group_name) + ) + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + output_local_tensors: dict[int, torch.Tensor] = {} + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + + group_tensors = [] + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + + for rank in group_ranks: + group_tensors.append(tensor._local_tensors[rank]) + + gathered_tensor = torch.cat(group_tensors, dim=0) + + for rank in group_ranks: + output_local_tensors[rank] = gathered_tensor.clone() + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + output = LocalTensor(output_local_tensors) + + return output + + +def _local_functional_reduce_scatter_tensor( + tensor: torch.Tensor, + reduce_op: str, + group_size: int, + group_name: GroupName | ProcessGroup, +) -> torch.Tensor: + # "reduce_scatter_tensor(Tensor input, str reduce_op, int group_size, str group_name) -> Tensor" + from . import _zero_sized_like, LocalTensor + + ranks, group_offsets, offset = _prepare_collective_groups( + _resolve_pg_or_name(group_name) + ) + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + output_local_tensors: dict[int, torch.Tensor] = {} + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + + group_tensors = [] + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + + for rank in group_ranks: + group_tensors.append(tensor._local_tensors[rank]) + + reduced_tensor = _local_reduce(reduce_op, group_tensors) + + scattered_tensor = torch.split( + reduced_tensor, + reduced_tensor.size(0) // len(group_ranks), + dim=0, + ) + + for i, rank in enumerate(group_ranks): + if i < len(scattered_tensor): + output_local_tensors[rank] = scattered_tensor[i].clone() + else: + output_local_tensors[rank] = _zero_sized_like(reduced_tensor, 0) + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + output = LocalTensor(output_local_tensors) + + return output + + +def _local_functional_shard_dim_alltoall( + tensor: torch.Tensor, + gather_dim: int, + shard_dim: int, + group_name: GroupName | ProcessGroup, +) -> torch.Tensor: + # "shard_dim_alltoall(Tensor input, int gather_dim, int shard_dim, str group_name) -> Tensor" + from . import _zero_sized_like, LocalTensor + + ranks, group_offsets, offset = _prepare_collective_groups( + _resolve_pg_or_name(group_name) + ) + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + output_local_tensors: dict[int, torch.Tensor] = {} + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + + group_tensors = [] + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + + for rank in group_ranks: + group_tensors.append(tensor._local_tensors[rank]) + + gathered_tensor = torch.cat(group_tensors, dim=gather_dim) + + split_tensor = torch.split( + gathered_tensor, + gathered_tensor.size(shard_dim) // len(group_ranks), + dim=shard_dim, + ) + + for i, rank in enumerate(group_ranks): + if i < len(split_tensor): + output_local_tensors[rank] = split_tensor[i].clone() + else: + output_local_tensors[rank] = _zero_sized_like( + gathered_tensor, shard_dim + ) + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + output = LocalTensor(output_local_tensors) + + return output + + +def _local_functional_all_to_all_single( + tensor: torch.Tensor, + output_split_sizes: list[torch.SymInt], + input_split_sizes: list[torch.SymInt], + group_name: GroupName | ProcessGroup, +) -> torch.Tensor: + # "all_to_all_single(Tensor input, SymInt[] output_split_sizes, SymInt[] input_split_sizes, str group_name) -> Tensor" + from . import LocalIntNode, LocalTensor + + ranks, group_offsets, offset = _prepare_collective_groups( + _resolve_pg_or_name(group_name) + ) + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + split_local_sizes: dict[int, list[int]] = {} + for input_split_size in input_split_sizes: + if isinstance(input_split_size, torch.SymInt) and isinstance( + input_split_size.node, LocalIntNode + ): + local_ints = dict(input_split_size.node._local_ints.items()) + else: + local_ints = {rank: int(input_split_size) for rank in tensor._local_tensors} + for rank, split_size in local_ints.items(): + if rank not in split_local_sizes: + split_local_sizes[rank] = [] + split_local_sizes[rank].append(split_size) + + split_local_tensors: dict[int, list[torch.Tensor]] = {} + + for rank, split_sizes in split_local_sizes.items(): + split_local_tensors[rank] = list( + torch.split(tensor._local_tensors[rank], split_sizes) + ) + + output_local_tensors: dict[int, torch.Tensor] = {} + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + + if not all(rank in split_local_tensors for rank in group_ranks): + continue + + for i, dst in enumerate(group_ranks): + splits = [] + for src in group_ranks: + splits.append(split_local_tensors[src][i]) + output_local_tensors[dst] = torch.cat(splits) + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + output = LocalTensor(output_local_tensors) + + return output + + +def _local_broadcast_( + tensors: list[torch.Tensor], + process_group_so: ScriptObject, + root_rank: int, + root_tensor: int, + async_op: bool = True, + timeout: int = -1, +) -> tuple[list[torch.Tensor], ScriptObject]: + # "broadcast_(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int root_rank, int root_tensor, bool async_op=True, int timeout=-1) -> (Tensor[], __torch__.torch.classes.c10d.Work)" + from . import LocalTensor + + if len(tensors) != 1: + raise AssertionError + if root_tensor != 0: + raise AssertionError + tensor = tensors[0] + + ranks, group_offsets, offset = _prepare_collective_groups(process_group_so) + + # We're going to assume SPMD where for every rank group the root_rank is + # the same relative to others + relative_root_rank = root_rank - offset + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the broadcast on them + group_ranks = [group_offset + r for r in ranks] + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + + source_rank = group_offset + relative_root_rank + source_tensor = tensor._local_tensors[source_rank] + + # Broadcast the source tensor to all ranks in this group + for rank in group_ranks: + if source_rank != rank: + tensor._local_tensors[rank].copy_(source_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return (tensors, work_so) + + +def _local_reduce( + reduce_op: ReduceOp | str, + tensors: list[torch.Tensor], +) -> torch.Tensor: + if reduce_op == ReduceOp.SUM or reduce_op == "sum": + op = operator.add + elif reduce_op == ReduceOp.AVG or reduce_op == "avg": + op = None + elif reduce_op == ReduceOp.PRODUCT or reduce_op == "product": + op = operator.mul + elif reduce_op == ReduceOp.MIN or reduce_op == "min": + op = torch.minimum + elif reduce_op == ReduceOp.MAX or reduce_op == "max": + op = torch.maximum + elif reduce_op == ReduceOp.BAND or reduce_op == "band": + op = torch.bitwise_and + elif reduce_op == ReduceOp.BOR or reduce_op == "bor": + op = torch.bitwise_or + elif reduce_op == ReduceOp.BXOR or reduce_op == "bxor": + op = torch.bitwise_xor + elif reduce_op == ReduceOp.PREMUL_SUM or reduce_op == "premul_sum": + raise NotImplementedError("PREMUL_SUM: need to add binding for scaling factor") + else: + raise NotImplementedError(f"ReduceOp {reduce_op} not implemented") + + if reduce_op == ReduceOp.AVG or reduce_op == "avg": + return functools.reduce(operator.add, tensors) / len(tensors) + else: + if op is None: + raise AssertionError + return functools.reduce(op, tensors) + + +def _local_all_reduce_( + tensors: list[torch.Tensor], + process_group_so: ScriptObject, + reduce_op_so: ScriptObject, + sparse_indices: torch.Tensor | None = None, + async_op: bool = True, + timeout: int = -1, +) -> tuple[list[torch.Tensor], ScriptObject]: + # "allreduce_(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "__torch__.torch.classes.c10d.ReduceOp reduce_op, Tensor? sparse_indices, bool async_op=True, " + # "int timeout=-1) -> (Tensor[], __torch__.torch.classes.c10d.Work)"); + from . import LocalTensor + + if len(tensors) != 1: + raise AssertionError + tensor = tensors[0] + reduce_op = reduce_op_so.op() # type: ignore[attr-defined] + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the allreduce on them + group_ranks = [group_offset + r for r in ranks] + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + + # Collect tensors from the specified ranks in this group + group_tensors = [] + for rank in group_ranks: + group_tensors.append(tensor._local_tensors[rank]) + + # Perform the reduction operation + reduced_tensor = _local_reduce(reduce_op, group_tensors) + + # Update all tensors in the group with the reduced result + for rank in group_ranks: + tensor._local_tensors[rank].copy_(reduced_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return (tensors, work_so) + + +def _local_allreduce_coalesced_( + tensors: list[torch.Tensor], + process_group_so: ScriptObject, + reduce_op_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> ScriptObject: + # "allreduce_coalesced_(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "__torch__.torch.classes.c10d.ReduceOp reduce_op, bool async_op=True, int timeout=-1) -> __torch__.torch.classes.c10d.Work" + from . import LocalTensor + + reduce_op = reduce_op_so.op() # type: ignore[attr-defined] + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the allreduce on all tensors together + group_ranks = [group_offset + r for r in ranks] + + # For each tensor, perform the reduction operation + for tensor in tensors: + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + if not all(rank in tensor._local_tensors for rank in group_ranks): + continue + # Collect tensors from the specified ranks in this group + group_tensors = [] + for rank in group_ranks: + group_tensors.append(tensor._local_tensors[rank]) + + # Perform the reduction operation + reduced_tensor = _local_reduce(reduce_op, group_tensors) + + # Update all tensors in the group with the reduced result + for rank in group_ranks: + tensor._local_tensors[rank].copy_(reduced_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_reduce_scatter_tensor_coalesced_( + output_tensors: list[torch.Tensor], + input_tensors: list[torch.Tensor], + process_group_so: ScriptObject, + reduce_op_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> ScriptObject: + # "reduce_scatter_tensor_coalesced_(Tensor[] outputs, Tensor[] inputs, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, " + # "__torch__.torch.classes.c10d.ReduceOp reduce_op, bool async_op=True, " + # "int timeout=-1) -> __torch__.torch.classes.c10d.Work" + + from . import LocalTensor + + reduce_op = reduce_op_so.op() # type: ignore[attr-defined] + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the allreduce on all tensors together + group_ranks = [group_offset + r for r in ranks] + + # For each tensor, perform the reduction operation + for input_tensor, output_tensor in zip(input_tensors, output_tensors): + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + # Collect tensors from the specified ranks in this group + group_inputs = [] + for rank in group_ranks: + group_inputs.append(input_tensor._local_tensors[rank]) + + # Perform the reduction operation + reduced_input = _local_reduce(reduce_op, group_inputs) + + reduced_input_splits = torch.split( + reduced_input, reduced_input.size(0) // len(group_ranks), dim=0 + ) + + # Update all tensors in the group with the reduced result + for i, rank in enumerate(group_ranks): + output_tensor._local_tensors[rank].copy_(reduced_input_splits[i]) + + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_allgather_base_( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + process_group_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> tuple[torch.Tensor, ScriptObject]: + # "_allgather_base_(Tensor output_tensor, Tensor input_tensor, __torch__.torch.classes.c10d.ProcessGroup + # process_group, bool async_op=True, int timeout=-1) -> (Tensor, __torch__.torch.classes.c10d.Work)"); + from . import LocalTensor + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + gathered_tensors = [] + for rank_i in group_ranks: + gathered_tensors.append(input_tensor._local_tensors[rank_i]) + + gathered_tensor = torch.cat(gathered_tensors, dim=0) + + for rank_i in group_ranks: + output_tensor._local_tensors[rank_i].copy_(gathered_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return output_tensor, work_so + + +def _local_reduce_scatter_base_( # type: ignore[no-untyped-def] + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + process_group_so: ScriptObject, + reduce_op_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> tuple[torch.Tensor, ScriptObject]: + # "_reduce_scatter_base_(Tensor output_tensor, Tensor input_tensor, + # __torch__.torch.classes.c10d.ProcessGroup process_group, __torch__.torch.classes.c10d.ReduceOp reduce_op, + # bool async_op=True, int timeout=-1) -> (Tensor, __torch__.torch.classes.c10d.Work)" + + from . import LocalTensor + + reduce_op = reduce_op_so.op() # type: ignore[attr-defined] + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + for group_offset in group_offsets: + group_ranks = [group_offset + r for r in ranks] + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + gathered_tensors = [] + for rank_i in group_ranks: + gathered_tensors.append(input_tensor._local_tensors[rank_i]) + + reduced_tensor = _local_reduce(reduce_op, gathered_tensors) + + scattered_tensor = torch.split( + reduced_tensor, + reduced_tensor.size(0) // len(group_ranks), + dim=0, + ) + + for i, rank_i in enumerate(group_ranks): + output_tensor._local_tensors[rank_i].copy_(scattered_tensor[i].clone()) + + work = FakeWork() + work_so = Work.boxed(work) + return output_tensor, work_so + + +def _local_all_gather_( + output_tensors: list[list[torch.Tensor]], + input_tensors: list[torch.Tensor], + process_group_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> tuple[list[list[torch.Tensor]], ScriptObject]: + # "allgather_(Tensor[][] output_tensors, Tensor[] input_tensors, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, bool async_op=True, " + # "int timeout=-1) -> (Tensor[][], __torch__.torch.classes.c10d.Work)"); + + from . import LocalTensor + + if len(output_tensors) != 1: + raise AssertionError + if len(input_tensors) != 1: + raise AssertionError + + input_tensor = input_tensors[0] + # pyrefly: ignore [bad-assignment] + output_tensors = output_tensors[0] + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + for i in range(len(output_tensors)): + if not isinstance(output_tensors[i], LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the all_gather on them + group_ranks = [group_offset + r for r in ranks] + + # For each rank in the group, gather from their input tensor + for i, rank_i in enumerate(group_ranks): + # allgather object happens to create pure tensor, so we special case it here + source_tensor = input_tensor + if isinstance(input_tensor, LocalTensor): + source_tensor = input_tensor._local_tensors[rank_i] + # pyrefly: ignore [missing-attribute] + output_tensors[i].copy_(source_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + # pyrefly: ignore [bad-return] + return ([output_tensors], work_so) + + +def _local_allgather_into_tensor_coalesced_( + output_tensors: list[torch.Tensor], + input_tensors: list[torch.Tensor], + process_group_so: ScriptObject, + async_op: bool = True, +) -> ScriptObject: + # "allgather_into_tensor_coalesced_(Tensor[] outputs, Tensor[] inputs, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, bool async_op=True) " + # "-> __torch__.torch.classes.c10d.Work" + from . import LocalTensor + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + # Each output tensor should be sized to hold all gathered inputs + # outputs[i] will contain all inputs[i] from all ranks + if len(output_tensors) != len(input_tensors): + raise AssertionError( + f"Number of outputs ({len(output_tensors)}) must match " + f"number of inputs ({len(input_tensors)})" + ) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the allgather_into_tensor on them + group_ranks = [group_offset + r for r in ranks] + + # For each input/output pair + for input_tensor, output_tensor in zip(input_tensors, output_tensors): + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + # Gather input_tensor from all ranks into output_tensor + # The output should be a concatenation of all inputs along the first dimension + gathered_tensors = [] + for rank in group_ranks: + gathered_tensors.append(input_tensor._local_tensors[rank]) + + # Concatenate along first dimension and copy to output + if gathered_tensors: + concatenated = torch.cat(gathered_tensors, dim=0) + for rank in group_ranks: + output_tensor._local_tensors[rank].copy_(concatenated) + + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_gather_( + output_tensors: list[list[torch.Tensor]], + input_tensors: list[torch.Tensor], + process_group_so: ScriptObject, + root_rank: int, + async_op: bool = True, + timeout: int = -1, +) -> ScriptObject: + # "gather_(Tensor[][] output_tensors, Tensor[] input_tensors, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, int root_rank, " + # "bool async_op=True, int timeout=-1) -> __torch__.torch.classes.c10d.Work" + raise NotImplementedError( + "LocalTensor does not support MPMD operations like gather " + "(only root rank receives data). Use SPMD collective operations like allgather instead." + ) + + +def _local_scatter_( + output_tensors: list[torch.Tensor], + input_tensors: list[list[torch.Tensor]], + process_group_so: ScriptObject, + root_rank: int, + async_op: bool = True, + timeout: int = -1, +) -> tuple[list[torch.Tensor], ScriptObject]: + # "scatter_(Tensor[] output_tensors, Tensor[][] input_tensors, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, int root_rank, " + # "bool async_op=True, int timeout=-1) -> (Tensor[], __torch__.torch.classes.c10d.Work)"); + + from . import LocalTensor + + if len(output_tensors) != 1: + raise AssertionError + if len(input_tensors) != 1: + raise AssertionError + output_tensor = output_tensors[0] + # pyrefly: ignore [bad-assignment] + input_tensors = input_tensors[0] + + ranks, group_offsets, offset = _prepare_collective_groups(process_group_so) + + # We're going to assume SPMD where for every rank group the root_rank is + # the same relative to others + relative_root_rank = root_rank - offset + + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + if len(ranks) != len(input_tensors): + raise AssertionError((ranks, input_tensors)) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the scatter on them + group_ranks = [group_offset + r for r in ranks] + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + # Root rank scatters its input tensors to all ranks in this group + for i, rank in enumerate(group_ranks): + input_tensor = input_tensors[i] + if not isinstance(input_tensor, LocalTensor): + raise AssertionError + # Each rank i gets the i-th input tensor from the root + source_tensor = input_tensor._local_tensors[ + group_offset + relative_root_rank + ] + output_tensor._local_tensors[rank].copy_(source_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return (output_tensors, work_so) + + +def _local_alltoall_( + output_tensors: list[torch.Tensor], + input_tensors: list[torch.Tensor], + process_group_so: ScriptObject, + async_op: bool = True, + timeout: int = -1, +) -> tuple[list[torch.Tensor], ScriptObject]: + # "alltoall_(Tensor[] output_tensors, Tensor[] input_tensors, " + # "__torch__.torch.classes.c10d.ProcessGroup process_group, bool async_op=True, " + # "int timeout=-1) -> (Tensor[], __torch__.torch.classes.c10d.Work)"; + + from . import LocalTensor + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + if not (len(input_tensors) == len(output_tensors) == len(ranks)): + raise AssertionError( + f"Number of input tensors ({len(input_tensors)}), " + f"output tensors ({len(output_tensors)}), and ranks ({len(ranks)}) must match" + ) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the alltoall on them + group_ranks = [group_offset + r for r in ranks] + + # In alltoall, rank i sends input_tensors[j] to rank j and receives into output_tensors[i] from rank j + for i, rank_i in enumerate(group_ranks): + output_tensor = output_tensors[i] + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + for j, rank_j in enumerate(group_ranks): + input_tensor = input_tensors[j] + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + + # Rank i's j-th input tensor goes to rank j's i-th output tensor + source_tensor = input_tensor._local_tensors[rank_i] + output_tensor._local_tensors[rank_j].copy_(source_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return (output_tensors, work_so) + + +def _local_alltoall_base_( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + process_group_so: ScriptObject, + output_split_sizes: list[int], + input_split_sizes: list[int], + async_op: bool = True, + timeout: int = -1, +) -> ScriptObject: + # "alltoall_base_(Tensor output, Tensor input, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int[] output_split_sizes, int[] input_split_sizes, bool async_op=True, int timeout=-1) -> __torch__.torch.classes.c10d.Work"; + + from . import LocalTensor + + ranks, group_offsets, _offset = _prepare_collective_groups(process_group_so) + + if not isinstance(input_tensor, LocalTensor): + raise AssertionError("Input tensor must be a LocalTensor") + if not isinstance(output_tensor, LocalTensor): + raise AssertionError("Output tensor must be a LocalTensor") + # Convert split sizes to lists if they aren't already + if output_split_sizes is not None: + output_split_sizes = list(output_split_sizes) + if input_split_sizes is not None: + input_split_sizes = list(input_split_sizes) + + for group_offset in group_offsets: + # For the tensors in this group [group_offset + r for r in ranks] + # perform the alltoall_base on them + group_ranks = [group_offset + r for r in ranks] + + if not all(rank in input_tensor._local_tensors for rank in group_ranks): + continue + if not all(rank in output_tensor._local_tensors for rank in group_ranks): + continue + + for i, rank_i in enumerate(group_ranks): + # Split input tensor from rank_i according to input_split_sizes + rank_tensor = input_tensor._local_tensors[rank_i] + + if input_split_sizes is not None and len(input_split_sizes) > 0: + # Split the input tensor + input_splits = torch.split(rank_tensor, input_split_sizes, dim=0) + else: + # No split sizes specified, split evenly + split_size = rank_tensor.size(0) // len(group_ranks) + input_splits = torch.split(rank_tensor, split_size, dim=0) + + # Send each split to the corresponding rank + for j, rank_j in enumerate(group_ranks): + if j < len(input_splits): + split_tensor = input_splits[j] + + # Determine where to place this split in the output tensor + if output_split_sizes is not None and len(output_split_sizes) > 0: + # Calculate offset based on output split sizes + output_offset = sum(output_split_sizes[:i]) if i > 0 else 0 + end_offset = ( + output_offset + output_split_sizes[i] + if i < len(output_split_sizes) + else output_tensor._local_tensors[rank_j].size(0) + ) + else: + # No output split sizes, use even splits + split_size = output_tensor._local_tensors[rank_j].size( + 0 + ) // len(group_ranks) + output_offset = i * split_size + end_offset = min( + (i + 1) * split_size, + output_tensor._local_tensors[rank_j].size(0), + ) + + # Copy the split to the appropriate section of the output tensor + output_section = output_tensor._local_tensors[rank_j][ + output_offset:end_offset + ] + if split_tensor.numel() != output_section.numel(): + raise ValueError( + f"all_to_all_single: input split from rank {rank_i} to " + f"rank {rank_j} has {split_tensor.numel()} elements, but " + f"the corresponding output split has " + f"{output_section.numel()} elements" + ) + if output_section.numel() > 0: + # Reshape split_tensor to match output_section if necessary + if split_tensor.size() != output_section.size(): + split_tensor = split_tensor.view(output_section.size()) + output_section.copy_(split_tensor) + + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_barrier( + tensor: torch.Tensor, + process_group_so: ScriptObject, + device_ids: list[int], + async_op: bool = True, + timeout: int = -1, +) -> ScriptObject: + # "barrier(Tensor tensor, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int[] device_ids, bool async_op=True, int timeout=-1) -> __torch__.torch.classes.c10d.Work"; + + from . import LocalTensor + + # Barrier is a synchronization primitive - in local simulation, + # we don't need to do any actual work since all "ranks" are in the same process + # Just validate that the tensor is a LocalTensor + if not isinstance(tensor, LocalTensor): + raise AssertionError + + # In a real distributed setting, barrier would synchronize all processes + # In local simulation, this is essentially a no-op since all ranks are local + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_monitored_barrier_( + tensor: torch.Tensor, + process_group_so: ScriptObject, + device_ids: list[int], + timeout: int, + wait_all_ranks: bool, +) -> None: + # "monitored_barrier_(Tensor tensor, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int[] device_ids, int timeout, bool wait_all_ranks) -> ()"; + + from . import LocalTensor + + # Monitored barrier is a synchronization primitive with monitoring - in local simulation, + # we don't need to do any actual work since all "ranks" are in the same process + # Just validate that the tensor is a LocalTensor + if not isinstance(tensor, LocalTensor): + raise AssertionError + + # In a real distributed setting, monitored barrier would synchronize all processes + # and provide monitoring capabilities. In local simulation, this is essentially a no-op + # since all ranks are local and no actual synchronization is needed + return + + +def _local_send( + tensors: list[torch.Tensor], + process_group_so: ScriptObject, + dst: int, + tag: int, +) -> ScriptObject: + # "send(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int dst, int tag) -> __torch__.torch.classes.c10d.Work"; + + from . import LocalRunnerMode, LocalTensor + + if len(tensors) != 1: + raise AssertionError + tensor = tensors[0] + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a Tensor") + src = int(tensor.__src_rank__) + + LocalRunnerMode.current()._signal_send(src, dst, tensor._local_tensors[src]) + + work = FakeWork() + work_so = Work.boxed(work) + return work_so + + +def _local_recv_( + tensors: list[torch.Tensor], + process_group_so: ScriptObject, + src: int, + tag: int, +) -> ScriptObject: + # "recv_(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int src, int tag) -> __torch__.torch.classes.c10d.Work"; + from . import LocalRunnerMode, LocalTensor + + if len(tensors) != 1: + raise AssertionError + tensor = tensors[0] + + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a Tensor") + dst = int(tensor.__src_rank__) + + def _recv_and_store(timeout: timedelta) -> bool: + def _wait_and_store(obj: object) -> None: + if not isinstance(obj, torch.Tensor): + raise AssertionError("Expected to receive a Tensor") + if not isinstance(tensor, LocalTensor): + raise AssertionError("Input tensor must be a Tensor") + tensor._local_tensors[dst] = obj + + LocalRunnerMode.current()._wait_recv(src, dst, _wait_and_store) + return True + + work = PythonCallbackWork(_recv_and_store) + work_so = Work.boxed(work) + return work_so + + +def _local_recv_any_source_( + tensors: list[torch.Tensor], process_group_so: ScriptObject, tag: int +) -> ScriptObject: + # "recv_any_source_(Tensor[] tensors, __torch__.torch.classes.c10d.ProcessGroup process_group, " + # "int tag) -> __torch__.torch.classes.c10d.Work"; + + return _local_recv_(tensors, process_group_so, -1, tag) + + +def _attach_rank(tensor: torch.Tensor, rank: int) -> torch.Tensor: + """ + Attaches rank as an attribute to given tensor so that the send or recv implementation + knows which rank initiates the operation (note under local tensor mode ). + """ + from torch.distributed.tensor import DTensor + + if isinstance(tensor, DTensor): + tensor = tensor._local_tensor + + tensor.__src_rank__ = rank # type: ignore[attr-defined] + return tensor + + +def local_p2p_op( + dst: torch.SymInt, + tensor: torch.Tensor, + op: Callable[[torch.Tensor, int], Work | None], +) -> Work | list[Work | None] | None: + """ + Runs a point-to-point (P2P) operation for all combinations of source and destination ranks. + """ + _check_op(op) + + from . import LocalIntNode + + if not isinstance(dst.node, LocalIntNode): + raise AssertionError( + "Expected 'dst' to be a LocalIntNode where the value is the " + "destination rank and key is the source rank" + ) + + w = [] + for s, d in dst.node._local_ints.items(): + tensor = _attach_rank(tensor, s) + w.append(op(tensor, d)) + return w + + +def wait_all(work: Work | list[Work | None] | None) -> None: + """ + Waits for all work objects in the input to complete. + + A single Work object, None, or a list of Work objects (possibly containing None). + If None, does nothing. If a single Work, waits for it to complete. If a list, waits + for each non-None Work in the list to complete. + """ + + if work is None: + return + if isinstance(work, Work): + work = [work] + for w in work: + if w is None: + continue + w.wait() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_mesh_layout.py b/.venv/lib/python3.12/site-packages/torch/distributed/_mesh_layout.py new file mode 100644 index 0000000000000000000000000000000000000000..0f20c276027532c1576d18d747fcd71a1f92628f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_mesh_layout.py @@ -0,0 +1,403 @@ +""" +Definition of CuTe inspired Layouts for DeviceMesh internal bookkeeping and functions to manipulate them +""" + +import math +from collections.abc import Iterator, Sequence +from dataclasses import dataclass +from itertools import product +from typing import NoReturn, overload + +import torch +from torch.distributed._pycute import ( + as_tuple, + coalesce, + complement, + composition, + flatten, + IntTuple, + is_int, + is_tuple, + Layout, + make_layout, + match_structure, + suffix_product, +) + + +@dataclass(frozen=True) +class _FlatLayout: + """ + A canonical CuTe layout for a single dimension of a DeviceMesh + + Utility class for representing an integer layout by borrowing ideas from CuTe Layout Algebra. + See https://github.com/NVIDIA/cutlass/blob/main/media/docs/cpp/cute/02_layout_algebra.md for more details. + + Each layout is represented as a list of sizes and strides. We use it as a way for mechanical bookkeeping + of the integers such as ranks in a SPMD mesh, and the transformation on top of it. + + Lots of methods of layout like coalesce, composition, complement, etc. are borrowed from pycute. + https://github.com/NVIDIA/cutlass/blob/6dd13d42784ee5bfa232d2441e6b9a021c5c6290/python/pycute/layout.py#L137,L257 + + Note this is a CuTe-inspired layout, because CuTe uses co-lexicographic way in linearization while PyTorch + is using lexicographic. So even though the CuTe documentation can still be referenced, the implementation will be + different from that of PyCute's. + + This layout is _not_ itself subdivided into multiple dimensions. It might + internally sometimes use multidimensional tuple to represent "irregular" + layouts (e.g., flattening non-adjacent dims), but this should be considered + an opaque implementation detail. + + This class guarantees that all equivalent layouts are encoded as the same + normalized representation, and thus compare equal. This is achieved by + flattening and coalescing compatible adjacent dimensions (which includes + removing all dimensions of size 1). + + """ + + shape: tuple[int, ...] + stride: tuple[int, ...] + + def __init__(self, shape: IntTuple, stride: IntTuple | None = None) -> None: + if not is_tuple(shape) and not is_int(shape): + raise TypeError(f"shape must be a tuple or int, got {type(shape)}") + stride = stride if stride is not None else suffix_product(shape) + if not is_tuple(stride) and not is_int(stride): + raise TypeError(f"stride must be a tuple or int, got {type(stride)}") + if not match_structure(shape, stride): + raise ValueError(f"sizes {shape} and strides {stride} don't match") + + coalesced_layout = coalesce(Layout(shape, stride)) + flat_shape = flatten(coalesced_layout.shape) + flat_stride = flatten(coalesced_layout.stride) + + # pycute will preserve a size=1 dim if it's the only remaining dim, but + # we prefer to stick to the Tensor convention and make it 0-dimensional + if flat_shape == (1,) and flat_stride == (0,): + flat_shape = () + flat_stride = () + + # Set attributes using object.__setattr__ since frozen=True + object.__setattr__(self, "shape", flat_shape) + object.__setattr__(self, "stride", flat_stride) + + def __len__(self) -> NoReturn: + raise RuntimeError( + "You should never need to know the length of the internal representation of a FlatLayout" + ) + + def __getitem__(self, i: int) -> NoReturn: + raise RuntimeError( + "You should never need to index into the internal representation of a FlatLayout" + ) + + def to_pycute(self) -> Layout: + if not self.shape: + return Layout(1, 0) + return Layout(self.shape, self.stride) + + def numel(self) -> int: + return math.prod(self.shape) + + def composition(self, layout: "_MeshLayout") -> "_MeshLayout": + """ + By-dimension composition allows one layout to "select from" or "filter through" another layout. + Think of it as function composition: (self ∘ layout)(input) = self(layout(input)) + between two layouts. This function is a wrapper of pycute's composition. + + Mental model about how to understand the composition logic: + - The LEFT layout (self) defines the "output space" - what indices are possible + - The RIGHT layout (layout parameter) acts as a "selector" - which specific indices to pick + - The composition only generates indices that the left layout could originally produce, + but the right layout determines which indices to be picked. + - The stride of the composition layout will not be smaller than the stride of the right layout, + because when picking the indices the composition will at least follow the right layout's stride + to move forward. + + Example: + self = (6,2):(2,1) # sizes=(6,2), strides=(2,1) + layout = (3:2) # sizes=(3,), stride=(2,) + self o layout = (3:2) + + Returns: + Layout being composed. + """ + result = composition(self.to_pycute(), layout.to_pycute()) + result_axes = [ + _FlatLayout(shape, stride) + for shape, stride in zip(as_tuple(result.shape), as_tuple(result.stride)) + ] + return _MeshLayout(result_axes) + + def complement(self, world_size: int) -> "_FlatLayout": + """ + Compute the "complement layout" relative to a given world_size. + A complement layout fills in the "missing" factor so that: self repeat a layout of complement(self, world_size) + will get a complete world_size. We use ⊗ to denote the repeat operation. + + Example: + self = (4:1) # size=4, stride=1 + world_size = 8 + Then: + complete needed factor = 8 / 4 = 2 + complement(self, 8) = (2:1) + + Together they form: + (4:1) ⊗ (2:1) = (4,2):(2,1) + which has world_size = 4 * 2 = 8, as required. + + In distributed terms, complement() is often used to derive the "other" + rank grouping when splitting processes into 2D meshes. + + For a visualized explanation, see https://x.com/ezyang/status/1962364978393981433/ + """ + result = complement(self.to_pycute(), world_size) + return _FlatLayout(result.shape, result.stride) + + def all_ranks_from_zero(self) -> list[int]: + """ + This function computes the all ranks specified by the layout starting from zero. + + How it works: + 1. we enumerate every possible coordinate (like a nested for-loop). + If sizes = (2, 3), we get the following coordinates: + (0,0), (0,1), (0,2), (1,0), (1,1), (1,2) + + 2. For each coordinate, we compute a linear rank index as: + all_ranks_from_zero = sum(coord[i] * strides[i] for i in range(ndim)) + + Example A: + sizes = (2, 3) # 2 rows, 3 cols + strides = (3, 1) # row-major layout + coords = (0,0) -> 0*3 + 0*1 = 0 + (0,1) -> 0*3 + 1*1 = 1 + (0,2) -> 0*3 + 2*1 = 2 + (1,0) -> 1*3 + 0*1 = 3 + (1,1) -> 1*3 + 1*1 = 4 + (1,2) -> 1*3 + 2*1 = 5 + result = [0, 1, 2, 3, 4, 5] + + Example B: + sizes = (2, 3) + strides = (1, 2) # non-standard / strided layout + coords = (0,0) -> 0*1 + 0*2 = 0 + (0,1) -> 0*1 + 1*2 = 2 + (0,2) -> 0*1 + 2*2 = 4 + (1,0) -> 1*1 + 0*2 = 1 + (1,1) -> 1*1 + 1*2 = 3 + (1,2) -> 1*1 + 2*2 = 5 + result = [0, 2, 4, 1, 3, 5] + """ + return [ + sum(c * s for c, s in zip(coord, self.stride)) + for coord in product(*(range(s) for s in self.shape)) + ] + + def global_ranks(self, world_size: int) -> list[list[int]]: + """ + Build global ranks specified by the layout via two-level ranks composition. + + The nested list forms the Cartesian product of all ranks for one layout and offset + regarding filling up the world_size with the layout. + The final global ranks are the addition of these two. The result is a + list of lists: one sublist per layout. This rank list will be used to build + the communicator underlying the layout and the given `world_size`. + + Example: + world_size = 16 + self.size = 4 + self.stride = 1 + ranks = [0, 1, 2, 3] + offsets = [0, 4, 8, 12] + result = [ + [0+0, 0+1, 0+2, 0+3], # → [0, 1, 2, 3] + [4+0, 4+1, 4+2, 4+3], # → [4, 5, 6, 7] + [8+0, 8+1, 8+2, 8+3], # → [8, 9, 10,11] + [12+0, 12+1, 12+2, 12+3], # → [12,13,14,15] + ] + """ + return [ + [offset + rank for rank in self.all_ranks_from_zero()] + for offset in self.complement(world_size).all_ranks_from_zero() + ] + + def check_sorted(self) -> bool: + return tuple(sorted(self.stride, reverse=True)) == self.stride + + def check_orthogonal(self) -> bool: + """ + Check if the layout has any overlap between the ranks it generates. If there is overlap, + we return False, otherwise True. + + The layout is supposed to be injective i.e, aside from indice 0, indices from each + dim of the layout must be non-overlapping. + + Example 1 - Valid (no overlap): + Layout: sizes=(2,3), strides=(6,1) + - Dim 1: stride=1, span=3*1=3, covers indices [0,1,2] + - Dim 0: stride=6, span=2*6=12, covers indices [0,6] + → No overlap since 6 > 3 + + Example 2 - Invalid (overlap): + Layout: sizes=(2,3), strides=(2,1) + - Dim 1: stride=1, span=3*1=3, covers indices [0,1,2] + - Dim 0: stride=2, span=2*2=4, covers indices [0,2] + → Overlap! stride=2 < span=3, so indices [0,2] are duplicated + + Example 3 - Invalid (overlap): + Layout: sizes=(4,2), strides=(1,1) + - Dim 1: stride=1, span=4, covers indices [0,1,2,3] + - Dim 0: stride=1, span=2, covers indices [0,1] + → Overlap! stride is same for two dims, so indices [0,2] are duplicated + + Returns: + bool: True if no overlap, False if overlap detected + """ + if len(self.shape) < 2: + return True + stride, shape = zip(*sorted(zip(self.stride, self.shape), reverse=True)) + return all( + stride[i] % (stride[i + 1] * shape[i + 1]) == 0 + for i in range(len(stride) - 1) + ) + + @property + def sizes_and_strides(self) -> Iterator[tuple[int, int]]: + """Iterate over (size, stride) pairs for each dimension.""" + return zip(self.shape, self.stride) + + +@dataclass(frozen=True) +class _MeshLayout(Sequence[_FlatLayout]): + """ + A multi-dimensional structure consisting of a series of dimension-less layouts + + This class represents the layout of a full DeviceMesh, where the overall + top-level ndim and "logical" shape are well defined, but each individual + mesh axis is squashed and normalized into a canonical _FlatLayout. + + It only contains methods that need to make use of this multi-dimensional + structure (i.e., which access the ndim or the top-level sizes). Everything + else should go on _FlatLayout and accessed by first calling .collapse(). + + """ + + axes: tuple[_FlatLayout, ...] + + def __init__(self, axes: Sequence[_FlatLayout]) -> None: + object.__setattr__(self, "axes", tuple(axes)) + + @classmethod + def from_sizes_strides( + cls, sizes: tuple[int, ...], strides: tuple[int, ...] | None = None + ) -> "_MeshLayout": + if strides is None: + strides = flatten(suffix_product(sizes)) + if len(sizes) != len(strides): + raise ValueError( + f"sizes and strides must have the same length, got {len(sizes)} and {len(strides)}" + ) + axes = tuple(_FlatLayout((s,), (d,)) for s, d in zip(sizes, strides)) + return cls(axes) + + def __len__(self) -> int: + return len(self.axes) + + @overload + def __getitem__(self, i: int) -> _FlatLayout: ... + + @overload + def __getitem__(self, i: slice) -> "_MeshLayout": ... + + def __getitem__(self, i: int | slice) -> "_FlatLayout | _MeshLayout": + if isinstance(i, slice): + return _MeshLayout(self.axes[i]) + return self.axes[i] + + def __iter__(self) -> Iterator[_FlatLayout]: + return iter(self.axes) + + def to_pycute(self) -> Layout: + if len(self.axes) == 0: + return Layout(1, 0) + return make_layout(*(axis.to_pycute() for axis in self.axes)) + + @property + def top_level_sizes(self) -> tuple[int, ...]: + return tuple(axis.numel() for axis in self.axes) + + def numel(self) -> int: + return math.prod(self.top_level_sizes) + + def cosize(self) -> int: + return self.to_pycute().cosize() + + def collapse(self) -> _FlatLayout: + """ + Merge all axes into a single _FlatLayout. + + This is used to "forget" the multi-dimensional structure of this object + and recover a "flat" (and coalesced) representation. + """ + shapes = tuple(axis.shape for axis in self.axes) + strides = tuple(axis.stride for axis in self.axes) + return _FlatLayout(shapes, strides) + + def splice(self, start: int, end: int, layout: "_MeshLayout") -> "_MeshLayout": + """ + Replace (out-of-place) the start:end slice with the given list of layouts + + Returns the concatenation of self[:start] + layout + self[end:]. + """ + new_axes = list(self.axes) + new_axes[start:end] = list(layout.axes) + return _MeshLayout(new_axes) + + def remap_to_tensor(self, rank_map: torch.Tensor) -> torch.Tensor: + """ + Leverage layout as an index for mesh tensor that re-maps the indexes after layout + transformation to actual device ranks. + + With this method, the cute layout serves as the backend of indices bookkeeping for the + mesh tensor when it comes to flatten, unflatten and slicing operations. The actual mesh + tensor still represents the actual device assignment and ranks. We need this function + to specify device allocation and create backend for a mesh. Although any transform of mesh tensors + can be treated as a view or subset of mesh tensor, we do need to use the actual view or + sub-tensor for DeviceMesh and its backend creation. + + The shape of the `rank_map` must be 1D and contiguous. + + Examples: + + Case 1 - Consecutive ranks, full world: + original_mesh_tensor = [[0,1],[2,3]] # 2x2 mesh, ranks 0-3 + world_size = 4 + layout = Layout(2:2) + Return: [[0,2],[1,3]] + + Case 2 - Non-consecutive ranks: + original_mesh_tensor = [[10,20],[30,40]] # custom rank assignment + world_size = 4 + layout = Layout(2:2) + Return: [[[10,30],[20,40]]] + + Args: + rank_map: The concrete mesh tensor with actual device ranks + + Returns: + torch.Tensor: A tensor representing the actual device allocation from rank_map + """ + if rank_map.ndim != 1: + raise AssertionError + if not rank_map.is_contiguous(): + raise AssertionError + if rank_map.numel() < self.cosize(): + raise AssertionError + + self_layout = self.collapse() + complement_layout = self_layout.complement(rank_map.numel()) + + return rank_map.as_strided( + complement_layout.shape + self_layout.shape, + complement_layout.stride + self_layout.stride, + ).reshape(-1, *self.top_level_sizes) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_meta_registrations.py b/.venv/lib/python3.12/site-packages/torch/distributed/_meta_registrations.py new file mode 100644 index 0000000000000000000000000000000000000000..b7c0ab5fb31bde2e0144a23a0a5071c2e542f068 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_meta_registrations.py @@ -0,0 +1,55 @@ +import random + +import torch +from torch._C._distributed_c10d import FakeWork + + +used_ids: set[int] = set() + + +def generate_unique_id() -> int: + while True: + new_id = random.randint(1, 10**9) + if new_id not in used_ids: + used_ids.add(new_id) + return new_id + + +# Function to create and return FakeWork object +def create_fakework(args, return_first_arg=True): # type: ignore[no-untyped-def] + work = FakeWork() + work.seq_id = generate_unique_id() + fakework_script_obj = work.boxed() + return (args[0], fakework_script_obj) if return_first_arg else fakework_script_obj + + +# Dictionary mapping collective operations to their meta functions +# All 20 ops from torch.csrc.distributed.c10d.Ops.cpp are included +# _DEPRECATED_META_FUNCTIONS = { +# "allreduce_coalesced_": lambda *args: create_fakework(args, return_first_arg=False), +# "allgather_coalesced_": lambda *args: create_fakework(args, return_first_arg=False), +# "allgather_into_tensor_coalesced_": lambda *args: create_fakework(args, return_first_arg=False), +# "reduce_scatter_tensor_coalesced_": lambda *args: create_fakework(args, return_first_arg=False), +# } +_META_FUNCTIONS = { + "broadcast_": lambda *args: create_fakework(args), + "allreduce_": lambda *args: create_fakework(args), + "allgather_": lambda *args: create_fakework(args), + "_allgather_base_": lambda *args: create_fakework(args), + "reduce_scatter_": lambda *args: create_fakework(args), + "_reduce_scatter_base_": lambda *args: create_fakework(args), + "reduce_": lambda *args: create_fakework(args, return_first_arg=False), + "gather_": lambda *args: create_fakework(args, return_first_arg=False), + "scatter_": lambda *args: create_fakework(args), + "alltoall_": lambda *args: create_fakework(args), + "alltoall_base_": lambda *args: create_fakework(args, return_first_arg=False), + "barrier": lambda *args: create_fakework(args, return_first_arg=False), + "monitored_barrier_": lambda *args: None, + "send": lambda *args: create_fakework(args, return_first_arg=False), + "recv_": lambda *args: create_fakework(args, return_first_arg=False), + "recv_any_source_": lambda *args: create_fakework(args, return_first_arg=False), +} + +lib_impl = torch.library.Library("c10d", "IMPL") +for op, meta_func in _META_FUNCTIONS.items(): + lib_impl.impl(op, meta_func, "Meta") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f2661b8c6f307a8f419a6434c96e249fee48b19a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__init__.py @@ -0,0 +1 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..45f89e18f23595ad7ec914224f233a43b0ef78df Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/device_mesh.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/device_mesh.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6471e5238488272b392dfcd6ced29c80c4aa4410 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/__pycache__/device_mesh.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_ops/device_mesh.py b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/device_mesh.py new file mode 100644 index 0000000000000000000000000000000000000000..61d4312c3bb047cb2a77330e16cd2cda1f751488 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_ops/device_mesh.py @@ -0,0 +1,105 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import torch +from torch import SymInt + +from ..device_mesh import DeviceMesh + + +# Register custom operator +torch.library.define( + "device_mesh::_runtime_compute_coordinate_on_dim", + "(Tensor full_mesh, int index) -> SymInt", + tags=torch.Tag.pt2_compliant_tag, +) + + +@torch.library.register_fake("device_mesh::_runtime_compute_coordinate_on_dim") +def _runtime_compute_coordinate_on_dim_fake( + full_mesh: torch.Tensor, index: int +) -> SymInt: + from torch.fx.experimental.symbolic_shapes import _constrain_range_for_size + + ctx = torch._custom_op.impl.get_ctx() + shape_env = ctx._shape_env + + # Bypass allow_dynamic_output_shape_ops check by directly creating the symint. + # This is intentional - the coordinate is always valid and bounded. + sz = shape_env.create_unbacked_symint() + + # Apply size constraints - coordinate is bounded by mesh size on the given dimension. + # The full_mesh tensor has an extra batch dimension at the front, so the actual + # mesh dimensions start at index 1. mesh.size(index) = full_mesh.size(index + 1) + mesh_size = full_mesh.size(index + 1) + _constrain_range_for_size( + sz, min=0, max=mesh_size - 1 if isinstance(mesh_size, int) else None + ) + + try: + # Check if we're currently tracing in dynamo (as opposed to AOT or export). + in_dynamo = torch._dynamo.symbolic_convert.InstructionTranslator.current_tx() + except AttributeError: + in_dynamo = False + + if in_dynamo: + # During dynamo tracing, distributed ops are treated as atomic - so the + # rank SymInt may be computed but not traced into the graph (e.g., it + # affects tensor values but not shapes). Mark it as ignorable here; + # when we decompose these ops later (after dynamo), we'll create fresh + # SymInts that do get traced. + shape_env.ignorable_fresh_unbacked_symbols.append(sz.node._expr) + + return sz + + +@torch.library.impl( + "device_mesh::_runtime_compute_coordinate_on_dim", "CompositeExplicitAutograd" +) +def _runtime_compute_coordinate_on_dim_impl(full_mesh: torch.Tensor, index: int) -> int: + rank = torch.distributed.get_rank() + mesh = DeviceMesh._get_mesh_tensor_from_full_mesh(full_mesh) + mesh_coords = DeviceMesh._compute_coordinates_from_mesh(mesh, rank) + if mesh_coords is None: + raise AssertionError + return mesh_coords[index] + + +def _get_flattened_submesh_impl(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + from torch.distributed.tensor._redistribute import ( + _get_flattened_mesh_by_layout_impl, + ) + + result = _get_flattened_mesh_by_layout_impl(mesh, tuple(mesh_dims)) + if result is None: + raise ValueError(f"No flattened mesh found for mesh_dims={mesh_dims} on {mesh}") + return result + + +@torch.library.custom_op("device_mesh::_get_flattened_submesh", mutates_args=()) +def _get_flattened_submesh(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + return _get_flattened_submesh_impl(mesh, mesh_dims) + + +@_get_flattened_submesh.register_fake +def _get_flattened_submesh_fake(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + return _get_flattened_submesh_impl(mesh, mesh_dims) + + +def _get_submesh_impl(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + all_dim_names = mesh._mesh_dim_names + if all_dim_names is None: + raise ValueError(f"Cannot slice mesh without dim names: {mesh}") + dim_names = tuple(all_dim_names[i] for i in mesh_dims) + if len(dim_names) == 1: + return mesh[dim_names[0]] + return mesh[dim_names] + + +@torch.library.custom_op("device_mesh::_get_submesh", mutates_args=()) +def _get_submesh(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + return _get_submesh_impl(mesh, mesh_dims) + + +@_get_submesh.register_fake +def _get_submesh_fake(mesh: DeviceMesh, mesh_dims: list[int]) -> DeviceMesh: + return _get_submesh_impl(mesh, mesh_dims) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e13bcc86e5095a0762417cf0c6cfdaa20951ee5d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__init__.py @@ -0,0 +1,74 @@ +################################################################################################# +# +# Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are met: +# +# 1. Redistributions of source code must retain the above copyright notice, this +# list of conditions and the following disclaimer. +# +# 2. Redistributions in binary form must reproduce the above copyright notice, +# this list of conditions and the following disclaimer in the documentation +# and/or other materials provided with the distribution. +# +# 3. Neither the name of the copyright holder nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +################################################################################################# + +from .int_tuple import ( + as_tuple, + crd2crd, + crd2idx, + elem_scale, + flatten, + has_none, + idx2crd, + inner_product, + IntTuple, + is_int, + is_tuple, + match_structure, + product, + shape_div, + signum, + slice_, + suffix_product, + tuple_max, +) +from .layout import ( + coalesce, + complement, + composition, + cosize, + filter, + is_layout, + Layout, + LayoutBase, + left_inverse, + logical_divide, + logical_product, + make_layout, + right_inverse, + size, + slice_and_offset, + tiled_divide, + tiled_product, + zipped_divide, + zipped_product, +) +from .typing import Integer diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..00f3536186cb4506a7c312778db2b818de8fd3e1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/int_tuple.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/int_tuple.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cf557bec773331b37779ea5bd151c62d3bea6177 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/int_tuple.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/layout.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/layout.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..368fdebdca8188e03e222d7ef18282c98390ab29 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/layout.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/typing.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/typing.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ca3d2bed2a434606b2d3dcbf7c712c4ac7da0f85 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/__pycache__/typing.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/int_tuple.py b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/int_tuple.py new file mode 100644 index 0000000000000000000000000000000000000000..830810ab4878905f8971f3792a570dba31eff5b4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/int_tuple.py @@ -0,0 +1,285 @@ +################################################################################################# +# +# Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are met: +# +# 1. Redistributions of source code must retain the above copyright notice, this +# list of conditions and the following disclaimer. +# +# 2. Redistributions in binary form must reproduce the above copyright notice, +# this list of conditions and the following disclaimer in the documentation +# and/or other materials provided with the distribution. +# +# 3. Neither the name of the copyright holder nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +################################################################################################# + +""" +Functions for manipulating IntTuples +""" + +from functools import reduce +from itertools import chain +from typing import TypeAlias +from typing_extensions import TypeIs + +from .typing import Integer + + +# Type aliases for better readability +IntTuple: TypeAlias = int | tuple["IntTuple", ...] + + +def is_int(x: object) -> TypeIs[int]: + return isinstance(x, Integer) + + +def is_tuple(x: object) -> TypeIs[tuple]: + return isinstance(x, tuple) + + +def as_tuple(x: IntTuple) -> tuple[IntTuple, ...]: + if is_int(x): + return (x,) + return x + + +def match_structure(a: IntTuple, b: IntTuple) -> bool: + if is_int(a) and is_int(b): + return True + if is_tuple(a) and is_tuple(b): + return len(a) == len(b) and all(match_structure(x, y) for x, y in zip(a, b)) + return False + + +def flatten(t: IntTuple) -> tuple[int, ...]: + if is_tuple(t): + if len(t) == 0: + return () + else: + return tuple(i for a in t for i in flatten(a)) + else: + return (t,) + + +def signum(a: int) -> int: + return bool(a > 0) - bool(a < 0) + + +def product(a: IntTuple) -> int: + if is_tuple(a): + return reduce(lambda val, elem: val * product(elem), a, 1) + else: + return a + + +def inner_product(a: IntTuple, b: IntTuple) -> int: + if is_tuple(a) and is_tuple(b): # tuple tuple + if len(a) != len(b): + raise AssertionError + return sum(inner_product(x, y) for x, y in zip(a, b)) + else: # "int" "int" + if is_tuple(a) or is_tuple(b): + raise AssertionError + return a * b + + +def tuple_max(a: IntTuple) -> int: + if is_tuple(a): + return max(tuple_max(x) for x in a) + else: + return a + + +def elem_scale(a: IntTuple, b: IntTuple) -> IntTuple: + if is_tuple(a): + if is_tuple(b): # tuple tuple + if len(a) != len(b): + raise AssertionError + return tuple(elem_scale(x, y) for x, y in zip(a, b)) + else: # tuple "int" + raise AssertionError("Invalid combination: tuple with int") + else: + if is_tuple(b): # "int" tuple + return elem_scale(a, product(b)) + else: # "int" "int" + return a * b + + +# Inclusive prefix ceil div with output congruent to input a +def shape_div(a: IntTuple, b: IntTuple) -> IntTuple: + if is_tuple(a): + if is_tuple(b): # tuple tuple + if len(a) != len(b): + raise AssertionError + return tuple(shape_div(x, y) for x, y in zip(a, b)) + else: # tuple "int" + # r = [shape_div(a[0],b)] + [shape_div(a[i],b := shape_div(b, product(a[i-1]))) for i in range(1,len(a))] + r = [] + for v in a: + r.append(shape_div(v, b)) + b = shape_div(b, product(v)) + return tuple(r) + else: + if is_tuple(b): # "int" tuple + return shape_div(a, product(b)) + else: # "int" "int" + if not (a % b == 0 or b % a == 0): + raise AssertionError + return (a + b - 1) // b + + +# Exclusive suffix product with output congruent to input a (lexicographic) +def suffix_product(a: IntTuple, init: IntTuple = 1) -> IntTuple: + # TODO: With all these length asserts, may want to create a zip_strict wrapper. + if is_tuple(a): + if is_tuple(init): # tuple tuple + if len(a) != len(init): + raise AssertionError + return tuple(suffix_product(x, i) for x, i in zip(a, init)) + else: # tuple "int" + # Process from right to left for lexicographic ordering + # r = [prefix_product(a[len(a)-1],init)] + + # [prefix_product(a[i],init := init * product(a[i+1])) for i in range(len(a)-1,0)].reverse() + r = [] + + # Calculate products from right to left, appending to list + for i in range(len(a) - 1, -1, -1): + r.append(suffix_product(a[i], init)) + init = init * product(a[i]) + + # Reverse to get correct lexicographic order + r.reverse() + return tuple(r) + else: + if is_tuple(init): # "int" tuple + raise AssertionError("Invalid combination: int with tuple init") + else: # "int" "int" + return init + + +def idx2crd(idx: IntTuple, shape: IntTuple, stride: IntTuple | None = None) -> IntTuple: + if stride is None: + stride = suffix_product(shape) + + if is_tuple(idx): + if is_tuple(shape) and is_tuple(stride): # tuple tuple tuple + if not (len(idx) == len(shape) and len(stride) == len(shape)): + raise AssertionError + return tuple(idx2crd(i, s, d) for i, s, d in zip(idx, shape, stride)) + else: # tuple "int" "int" + raise AssertionError("Invalid combination: tuple with int stride") + else: + if is_tuple(shape) and is_tuple(stride): # "int" tuple tuple + if len(shape) != len(stride): + raise AssertionError + return tuple(idx2crd(idx, s, d) for s, d in zip(shape, stride)) + else: # "int" "int" "int" + if is_tuple(shape) or is_tuple(stride): + raise AssertionError + return (idx // stride) % shape # all are ints after type checks + + +def crd2idx( + crd: IntTuple | None, shape: IntTuple, stride: IntTuple | None = None +) -> int: + if stride is None: + stride = suffix_product(shape) + + if is_tuple(crd): + if is_tuple(shape) and is_tuple(stride): # tuple tuple tuple + if not (len(crd) == len(shape) and len(stride) == len(shape)): + raise AssertionError + return sum(crd2idx(c, s, d) for c, s, d in zip(crd, shape, stride)) + else: # tuple "int" "int" + raise AssertionError(f"Invalid combination: crd={crd}, shape={shape}") + else: + if crd is None: + crd = 0 + + if is_tuple(shape) and is_tuple(stride): # "int" tuple tuple + if len(shape) != len(stride): + raise AssertionError + result = 0 + # Process from right to left for lexicographic ordering + for i in range(len(shape) - 1, 0, -1): + result += crd2idx(crd % product(shape[i]), shape[i], stride[i]) + crd = crd // product(shape[i]) + if len(shape) > 0: + result += crd2idx(crd, shape[0], stride[0]) + return result + else: # "int" "int" "int" + if is_tuple(shape) or is_tuple(stride): + raise AssertionError + return crd * stride # all are ints after type checks + + +# Transform crd into the dst_shape's iteration space +def crd2crd( + crd: IntTuple, dst_shape: IntTuple, src_shape: IntTuple | None = None +) -> IntTuple: + if is_tuple(crd): + if is_tuple(dst_shape): # tuple tuple + if len(crd) != len(dst_shape): + raise AssertionError + return tuple(crd2crd(x, y) for x, y in zip(crd, dst_shape)) + else: # tuple "int" + # Ambiguous unless we have src_shape + if src_shape is None: + raise AssertionError + return crd2idx(crd, src_shape) + else: + if is_tuple(dst_shape): # "int" tuple + return idx2crd(crd, dst_shape) + else: # "int" "int" + if crd >= dst_shape: + raise AssertionError + return crd + + +# Filter trg according to crd: keep only elements of trg that are paired with None +def slice_(crd: tuple | int | None, trg: tuple | int) -> tuple | int: + if is_tuple(crd): + if is_tuple(trg): # tuple tuple + if len(crd) != len(trg): + raise AssertionError + # match C++ behavior of `filter_tuple` using `tuple_cat(...)` + return tuple( + chain( + *filter( # type: ignore[arg-type] # filter returns Iterator which is compatible + lambda x: x != (), + [slice_(c, s) for c, s in zip(crd, trg)], + ) + ) + ) + else: + raise AssertionError("Invalid combination: tuple crd with int trg") + elif crd is None: + # match C++ behavior `return cute::tuple{b};` + return (trg,) + else: + return () + + +# Determine if None appears at any of an int_tuples' terminals +def has_none(a: tuple | int | None) -> bool: + if is_tuple(a): + return any(has_none(v) for v in a) + else: + return a is None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/layout.py b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/layout.py new file mode 100644 index 0000000000000000000000000000000000000000..44d561fca471030cfe5d0874e4c34d19df620f7f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/layout.py @@ -0,0 +1,484 @@ +################################################################################################# +# +# Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are met: +# +# 1. Redistributions of source code must retain the above copyright notice, this +# list of conditions and the following disclaimer. +# +# 2. Redistributions in binary form must reproduce the above copyright notice, +# this list of conditions and the following disclaimer in the documentation +# and/or other materials provided with the distribution. +# +# 3. Neither the name of the copyright holder nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +################################################################################################# + +""" +Definition of CuTe Layouts and functions to manipulate them which works with the order +of lexicographic instead of co-lexicographic as implemented in the original layout.py +""" + +from itertools import chain +from typing import TypeAlias +from typing_extensions import Self, TypeIs + +from .int_tuple import ( + crd2idx, + flatten, + has_none, + IntTuple, + is_int, + is_tuple, + product, + slice_, + suffix_product, +) + + +# Type aliases +CoordinateType: TypeAlias = ( + int | IntTuple | tuple[object, ...] | None +) # Input for slice_ and crd2idx functions + + +class LayoutBase: + pass + + +def is_layout(x: object) -> TypeIs["Layout"]: + return isinstance(x, LayoutBase) + + +class Layout(LayoutBase): + def __init__(self, _shape: IntTuple, _stride: IntTuple | None = None) -> None: + self.shape = _shape + if _stride is None: + self.stride = suffix_product(self.shape) + else: + self.stride = _stride + + # operator == + def __eq__(self, other: object) -> bool: + if not isinstance(other, Layout): + return False + return self.shape == other.shape and self.stride == other.stride + + # operator len(L) (len [rank] like tuples) + def __len__(self) -> int: + if is_tuple(self.shape): + return len(self.shape) + else: + return 1 + + # operator () (map coord to idx) + def __call__(self, *args: CoordinateType) -> Self | int: + """ + Map a logical coordinate to a linear index (Coord has no Underscore slice operators) + OR + Slice the layout and return the sublayout (Coord has an Underscore slice op) + + Follow the same behavior of `Layout::operator(Coord const&)` in cute C++ + """ + if has_none(args): + if len(args) == 1: + return Layout(slice_(args[0], self.shape), slice_(args[0], self.stride)) + else: + return Layout(slice_(args, self.shape), slice_(args, self.stride)) + else: + if len(args) == 1: + return crd2idx(args[0], self.shape, self.stride) # type: ignore[arg-type] + else: + return crd2idx(args, self.shape, self.stride) # type: ignore[arg-type] + + # operator [] (get-i like tuples) + def __getitem__(self, i: int) -> Self: + if is_tuple(self.shape): + return Layout(self.shape[i], self.stride[i]) # type: ignore[index] + else: + if i != 0: + raise AssertionError + return Layout(self.shape, self.stride) + + # size(layout) Size of the domain + def size(self) -> int: + return product(self.shape) + + # cosize(layout) Size of the codomain + def cosize(self) -> int: + return self(self.size() - 1) + 1 # type: ignore[operator] + + # print and str + def __str__(self) -> str: + return f"{self.shape}:{self.stride}" + + # error msgs and representation + def __repr__(self) -> str: + return f"Layout({self.shape},{self.stride})" + + +# Type aliases +LayoutOrIntTuple: TypeAlias = Layout | IntTuple +LayoutProfile: TypeAlias = tuple[object, ...] | Layout | None +LayoutInput: TypeAlias = Layout | IntTuple | tuple[object, ...] | None + + +# Make Layout from a list of layouts (each layout its own mode in the result) +def make_layout(*layouts: Layout | tuple[Layout, ...]) -> Layout: + if len(layouts) == 1 and not is_layout(layouts[0]): + layouts = layouts[0] + + shape, stride = zip(*((a.shape, a.stride) for a in layouts)) # type: ignore[union-attr] + return Layout(shape, stride) + + +# Size of the domain +def size(layout: LayoutOrIntTuple) -> int: + if is_layout(layout): + return layout.size() + return product(layout) + + +# Size of the codomain +def cosize(layout: Layout) -> int: + return layout.cosize() + + +# Layout coalesce -- flatten and combine as many modes as possible while preserving the int-to-int function +def coalesce(layout: Layout, profile: LayoutProfile = None) -> Layout: + if is_tuple(profile): + if len(layout) < len(profile): + raise AssertionError + return make_layout( + # pyrefly: ignore [bad-argument-type] + chain( + (coalesce(layout[i], profile[i]) for i in range(len(profile))), # type: ignore[arg-type] + (layout[i] for i in range(len(profile), len(layout))), + ) + ) + + result_shape = [1] + result_stride = [0] + # Since we now follow lexicographic order, we need to process from right to left. + # And to make implementation more efficient, we append to the end of list and reverse it in the end. + for shape, stride in zip( + reversed(flatten(layout.shape)), reversed(flatten(layout.stride)) + ): + # skip their shape-1s + if shape == 1: + continue + # replace our shape-1 with anything + elif result_shape[-1] == 1: + result_shape[-1] = shape + result_stride[-1] = stride + # merge modes if the shape*stride match + elif result_shape[-1] * result_stride[-1] == stride: + result_shape[-1] = result_shape[-1] * shape + # append a new mode + else: + result_shape.append(shape) + result_stride.append(stride) + + if len(result_shape) == 1: + return Layout(result_shape[0], result_stride[0]) + else: + result_shape.reverse() + result_stride.reverse() + return Layout(tuple(result_shape), tuple(result_stride)) + + +# Layout filter -- replace all stride-0 modes with size-1 and then coalesce to remove them +def filter(layout: Layout, profile: LayoutProfile = None) -> Layout: + if is_tuple(profile): + if len(layout) < len(profile): + raise AssertionError + return make_layout( + # pyrefly: ignore [bad-argument-type] + chain( + (filter(layout[i], profile[i]) for i in range(len(profile))), # type: ignore[arg-type] + (layout[i] for i in range(len(profile), len(layout))), + ) + ) + + result_shape = [] + result_stride = [] + for shape, stride in zip(flatten(layout.shape), flatten(layout.stride)): + # skip their shape-1s and stride-0s + if not (shape == 1 or stride == 0): + result_shape.append(shape) + result_stride.append(stride) + + if len(result_shape) == 0: + return Layout(1, 0) + else: + return coalesce(Layout(tuple(result_shape), tuple(result_stride))) + + +# Layout composition +# Use tuples-of-layouts to perform this operation by-mode and None as no-op +def composition(layoutA: Layout, layoutB: LayoutInput) -> Layout: + if layoutB is None: + return layoutA + elif is_int(layoutB): + return composition(layoutA, Layout(layoutB)) + elif is_tuple(layoutB): + if len(layoutA) < len(layoutB): + raise AssertionError + return make_layout( + # pyrefly: ignore [bad-argument-type] + chain( + (composition(layoutA[i], layoutB[i]) for i in range(len(layoutB))), # type: ignore[arg-type] + (layoutA[i] for i in range(len(layoutB), len(layoutA))), + ) + ) + elif is_tuple(layoutB.shape): + return make_layout(composition(layoutA, layoutB_i) for layoutB_i in layoutB) # type: ignore[arg-type, attr-defined] + + if layoutB.stride == 0: + return Layout(layoutB.shape, 0) + else: + result_shape = [] + result_stride = [] + rest_shape = layoutB.shape + rest_stride = layoutB.stride + flat_A = coalesce(layoutA) + # when left layout is multi-dimensional sublayout, aka, self = (a,b,...,c):(x,y,...,z), layout = s:d, + # for integral s and d means that we want: + # (1) “remove” the first d elements from left, starting from rightmost. (This will increase the stride.) + # (2) “keep” the first s of those strided elements. (This does not affect the stride.) + # For example, if self = (6,2):(2,1), layout = (3:2) + # Step 1: remove the first 2 elements from self with stride increase, i.e., (6,2):(2,1) -> (6,1):(2,2) + # Step 2: keep the first 3 of those strided elements, i.e., (6,1):(2,2) -> (3,1):(2,2) + # Because we are going lexicographically, we go through left layout from right to left. + for curr_shape, curr_stride in zip( + reversed(flatten(flat_A.shape)[1:]), reversed(flatten(flat_A.stride)[1:]) + ): + if not (curr_shape % rest_stride == 0 or rest_stride % curr_shape == 0): # type: ignore[operator] + raise AssertionError + new_shape = min(max(1, curr_shape // rest_stride), rest_shape) # type: ignore[operator] + + if new_shape != 1: + result_shape.append(new_shape) # Append to end, will reverse later + result_stride.append(rest_stride * curr_stride) + + rest_shape = rest_shape // new_shape # type: ignore[operator] + rest_stride = -( + -rest_stride // curr_shape # type: ignore[operator] + ) # Python exclusive impl: "//" is always floor div so == ceil_div(abs(rest_stride), curr_shape) * signum(rest_stride) + + # When left has single-size sublayout or reach the last sublayout, aka, left = a:b, layout = s:d, + # the result is rather trivial: left o layout = a:b o s:d = s:(b*d). + # For example, if self = (6:2), layout = (3:2), the result is (3:(2*2)) = (3:4). + if rest_shape != 1 or len(result_shape) == 0: + result_shape.append(rest_shape) # Append to end, will reverse later + result_stride.append(rest_stride * flatten(flat_A.stride)[0]) + + # Reverse the lists because we build lists in reverse order (append to end), this way it is more efficient. + result_shape.reverse() + result_stride.reverse() + + if len(result_shape) == 1: + return Layout(result_shape[0], result_stride[0]) # type: ignore[arg-type] + else: + return Layout(tuple(result_shape), tuple(result_stride)) # type: ignore[arg-type] + + +# Layout complement +def complement(layout: LayoutOrIntTuple, max_idx: int = 1) -> Layout: + if is_int(layout): + return complement(Layout(layout)) + + result_shape = [] + result_stride = [] + current_idx = 1 + + sorted_DS = sorted(zip(flatten(layout.stride), flatten(layout.shape))) # type: ignore[union-attr] + for stride, shape in sorted_DS: + if stride == 0 or shape == 1: + continue + + in_bound = current_idx <= shape * stride + # To support symbolic value which can't be evaluated now + if (type(in_bound) is bool) and not in_bound: + raise AssertionError + + result_shape.append(stride // current_idx) + result_stride.append(current_idx) + current_idx = shape * stride + + result_shape.append((max_idx + current_idx - 1) // current_idx) # ceil_div + result_stride.append(current_idx) + # This is different from original pycute implementation, because we want to follow the lexicographic order here + # where the right-most dimension is the innermost dimension (smallest stride). + result_shape.reverse() + result_stride.reverse() + + return coalesce(Layout(tuple(result_shape), tuple(result_stride))) + + +# Layout right inverse +def right_inverse(layout: LayoutOrIntTuple | None) -> Layout | None: + if layout is None: + return None + elif is_int(layout): + return Layout(layout) + + result_shape = [] + result_stride = [] + current_idx = 1 + + flat_shape = flatten(layout.shape) # type: ignore[union-attr] + flat_stride = flatten(layout.stride) # type: ignore[union-attr] + sorted_DSA = sorted(zip(flat_stride, flat_shape, suffix_product(flat_shape))) # type: ignore[arg-type] + for stride, shape, rstride in sorted_DSA: + if shape == 1: + continue + if current_idx != stride: + break + + result_shape.append(shape) + result_stride.append(rstride) + current_idx = shape * stride + + result_shape.reverse() + result_stride.reverse() + return coalesce(Layout(tuple(result_shape), tuple(result_stride))) + + +# Layout left inverse +def left_inverse(layout: LayoutOrIntTuple | None) -> Layout | None: + if layout is None: + return None + elif is_int(layout): + return Layout(layout) + return right_inverse(make_layout(complement(layout), layout)) # type: ignore[arg-type] + + +# Split a layout by the composition of B and the "rest" +# Use tuples-of-layouts to perform this operation by-mode and None as no-op +def logical_divide(layoutA: Layout, layoutB: LayoutInput) -> Layout: + if layoutB is None: + return layoutA + elif is_int(layoutB): + return logical_divide(layoutA, Layout(layoutB)) + elif is_tuple(layoutB): + if len(layoutA) < len(layoutB): + raise AssertionError + return make_layout( + # pyrefly: ignore [bad-argument-type] + chain( + ( + logical_divide(layoutA[i], layoutB[i]) # type: ignore[arg-type] + for i in range(len(layoutB)) + ), + (layoutA[i] for i in range(len(layoutB), len(layoutA))), + ) + ) + + return composition( + layoutA, + make_layout(layoutB, complement(layoutB, size(layoutA))), + ) + + +# Reproduce a layoutA over a layoutB +# Use tuples-of-layouts to perform this operation by-mode and None as no-op +def logical_product(layoutA: Layout, layoutB: LayoutInput) -> Layout: + if layoutB is None: + return layoutA + elif is_int(layoutB): + return logical_divide(layoutA, Layout(layoutB)) + elif is_tuple(layoutB): + if len(layoutA) < len(layoutB): + raise AssertionError + return make_layout( + # pyrefly: ignore [bad-argument-type] + chain( + ( + logical_product(layoutA[i], layoutB[i]) # type: ignore[arg-type] + for i in range(len(layoutB)) + ), + (layoutA[i] for i in range(len(layoutB), len(layoutA))), + ) + ) + + return make_layout( + layoutA, + composition(complement(layoutA, size(layoutA) * cosize(layoutB)), layoutB), + ) + + +# Gather the modes from a hierarchical logical_divide or logical_product +def hier_unzip( + splitter: object, + layoutA: Layout, + layoutB: LayoutInput, +) -> Layout: + if layoutB is None: + return make_layout(Layout(1, 0), layoutA) + elif is_tuple(layoutB): + if len(layoutA) < len(layoutB): + raise AssertionError + # A layout with shape ((A,a),(B,b),(C,c)) + split = make_layout( + hier_unzip(splitter, layoutA[i], layoutB[i]) # type: ignore[arg-type] + for i in range(len(layoutB)) + ) + # Gather to shape ((A,B,C,...),(a,b,c,...,y,z)) + return make_layout( + make_layout(split[i][0] for i in range(len(layoutB))), # type: ignore[arg-type] + make_layout( + chain( # type: ignore[arg-type] + (split[i][1] for i in range(len(layoutB))), + (layoutA[i] for i in range(len(layoutB), len(layoutA))), + ) + ), + ) + + # splitter must return a rank-2 layout + return splitter(layoutA, layoutB) # type: ignore[operator] + + +# Apply logical divide hierarchically and gather the split modes into two modes +def zipped_divide(layoutA: Layout, layoutB: LayoutInput) -> Layout: + return hier_unzip(logical_divide, layoutA, layoutB) + + +# Perform logical divide hierarchically and gather tiles (B-layouts) into a new mode +def tiled_divide(layoutA: Layout, layoutB: LayoutInput) -> Layout: + result = zipped_divide(layoutA, layoutB) + return make_layout([result[0]] + [result[1][i] for i in range(len(result[1]))]) # type: ignore[arg-type] + + +# Apply logical product hierarchically and gather the split modes into two modes +def zipped_product(layoutA: Layout, layoutB: LayoutInput) -> Layout: + return hier_unzip(logical_product, layoutA, layoutB) + + +# Perform logical product hierarchically and gather tiles (B-layouts) into a new mode +def tiled_product(layoutA: Layout, layoutB: LayoutInput) -> Layout: + result = zipped_product(layoutA, layoutB) + return make_layout([result[0]] + [result[1][i] for i in range(len(result[1]))]) # type: ignore[arg-type] + + +def slice_and_offset(crd: tuple[object, ...], layout: Layout) -> tuple[Layout, int]: + return ( + Layout(slice_(crd, layout.shape), slice_(crd, layout.stride)), + crd2idx(crd, layout.shape, layout.stride), # type: ignore[arg-type] + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/typing.py b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/typing.py new file mode 100644 index 0000000000000000000000000000000000000000..5e6fe0a9c66e800186b4d84ef52cc12d6baeb1f6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_pycute/typing.py @@ -0,0 +1,42 @@ +################################################################################################# +# +# Copyright (c) 2023 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: BSD-3-Clause +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are met: +# +# 1. Redistributions of source code must retain the above copyright notice, this +# list of conditions and the following disclaimer. +# +# 2. Redistributions in binary form must reproduce the above copyright notice, +# this list of conditions and the following disclaimer in the documentation +# and/or other materials provided with the distribution. +# +# 3. Neither the name of the copyright holder nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +################################################################################################# + +from abc import ABC + + +class Integer(ABC): # noqa: B024 # Uses __subclasshook__ instead of abstract methods + @classmethod + def __subclasshook__(cls, c: type) -> bool: + if c in [bool, float]: + return False + + return issubclass(c, int) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_serialization.py b/.venv/lib/python3.12/site-packages/torch/distributed/_serialization.py new file mode 100644 index 0000000000000000000000000000000000000000..8f7043453be769cccfb70cd391dac87348508016 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_serialization.py @@ -0,0 +1,158 @@ +import pickle +from dataclasses import dataclass +from io import BufferedIOBase +from typing import Any + +import torch +import torch._weights_only_unpickler as _weights_only_unpickler +from torch.serialization import _load, _save, DEFAULT_PROTOCOL, MAP_LOCATION + + +__all__: list[str] = [] + + +@dataclass +class _Entry: + key: str + is_storage: bool + length: int + + +_weights_only_unpickler._add_safe_globals([_Entry]) + + +class _PseudoZipFile: + def __init__(self) -> None: + self.records: dict[str, tuple[object, int]] = {} + + def write_record(self, key: str, data: object, length: int) -> None: + self.records[key] = (data, length) + + def write_to(self, f: BufferedIOBase) -> None: + entries = [] + for key, (data, length) in self.records.items(): + entries.append( + _Entry( + key=key, + is_storage=isinstance(data, torch.UntypedStorage), + length=length, + ) + ) + + pickle.dump(entries, f, protocol=DEFAULT_PROTOCOL) + + for data, _ in self.records.values(): + if isinstance(data, bytes): + f.write(data) + elif isinstance(data, str): + f.write(data.encode("utf-8")) + elif isinstance(data, torch.UntypedStorage): + data._write_file(f, False, False, 1) + else: + raise TypeError(f"unknown type: {type(data)}") + + def read_from(self, f: BufferedIOBase) -> None: + entries = _weights_only_unpickler.load(f) + + for entry in entries: + data = f.read(entry.length) + if entry.is_storage: + if entry.length == 0: + storage = torch.UntypedStorage(0) + else: + storage = torch.frombuffer( + data, + dtype=torch.uint8, + ).untyped_storage() + + self.records[entry.key] = ( + storage, + entry.length, + ) + else: + self.records[entry.key] = (data, entry.length) + + def has_record(self, key: str) -> bool: + return key in self.records + + def get_record(self, key: str) -> object: + return self.records[key][0] + + def get_storage_from_record( + self, key: str, _length: int, _type: int + ) -> torch.Tensor: + return torch.tensor(self.records[key][0], dtype=torch.uint8) + + def serialization_id(self) -> str: + return "torchft" + + +def _streaming_save( + obj: object, + f: BufferedIOBase, + pickle_module: Any = pickle, + pickle_protocol: int = DEFAULT_PROTOCOL, +) -> None: + """ + Save the object to a file-like object in a streaming fashion compatible with + network sockets. + + This behaves similarly to :func:`torch.save` with a few notable differences: + + * A non-seekable file like object can be used when loading. + * No forwards/backwards compatibility is provided for the serialization + format. This is only intended to be used with a single version of PyTorch + with transient storage (i.e. sockets or temp files). + * mmap is not supported + + See :func:`torch.save` for more details on specific arguments. + """ + + zip_file = _PseudoZipFile() + _save( + obj, + zip_file=zip_file, + pickle_module=pickle_module, + pickle_protocol=pickle_protocol, + _disable_byteorder_record=False, + ) + zip_file.write_to(f) + + +def _streaming_load( + f: BufferedIOBase, + map_location: MAP_LOCATION = None, + pickle_module: Any = None, + *, + weights_only: bool = True, + **pickle_load_args: Any, +) -> object: + """ + Load the object from a file-like object in a streaming fashion compatible with + network sockets. + + See :func:`_streaming_save` for more details about the streaming behavior. + + See :func:`torch.load` for more details on specific arguments. + """ + if weights_only: + if pickle_module is not None: + raise RuntimeError( + "Can not safely load weights when explicit pickle_module is specified" + ) + pickle_module = _weights_only_unpickler + else: + if pickle_module is None: + pickle_module = pickle + + if "encoding" not in pickle_load_args: + pickle_load_args["encoding"] = "utf-8" + + zip_file = _PseudoZipFile() + zip_file.read_from(f) + return _load( + zip_file=zip_file, + map_location=map_location, + pickle_module=pickle_module, + **pickle_load_args, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..85a313c779e7aa87726f425146048fcd37efd261 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__init__.py @@ -0,0 +1 @@ +from .api import _shard_tensor, load_with_process_group, shard_module, shard_parameter diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7fcf77a3cbc9060ad1e32aad8a8a0dc90f3a01cb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2bc30ec356baed192f8fb5e44e804e777a9ca53c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4e198d13ff7b5b6c17276fde05b4d0ecf191095b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/common_op_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/common_op_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ce3f82268b8a7fd3a11ea08b810de41144c7e75e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/common_op_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/metadata.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/metadata.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6dd3fe0d72cf959c580ca0fd8b710c30470cf8a1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/metadata.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/op_registry_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/op_registry_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..63cfe7b53c3b3955791a1453026cac928c7c66d2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/op_registry_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/sharder.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/sharder.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c856cb4300d713c4a82afa96c7200aa4dcfbab86 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/__pycache__/sharder.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..6fd641b3f9443faa64b6b54c3ab209f8167a56f7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/_utils.py @@ -0,0 +1,32 @@ +from collections.abc import Sequence + +import torch +from torch.distributed._shard.metadata import ShardMetadata + + +DEPRECATE_MSG = "Please use DTensor instead and we are deprecating ShardedTensor." + + +def narrow_tensor_by_index( + tensor: torch.Tensor, + offsets: Sequence[int], + sizes: Sequence[int], +) -> torch.Tensor: + """ + Narrow the tensor according to ``offsets`` and ``sizes``. + """ + narrowed_tensor = tensor + for idx, (offset, size) in enumerate(zip(offsets, sizes)): + if size < tensor.size(idx): + # Reshape to get shard for this rank and we don't want autograd + # recording here for the narrow op and 'local_shard' should be a + # leaf variable in the autograd graph. + narrowed_tensor = narrowed_tensor.narrow(idx, offset, size) + return narrowed_tensor + + +def narrow_tensor(tensor: torch.Tensor, metadata: ShardMetadata) -> torch.Tensor: + """ + Narrow the tensor according to the metadata + """ + return narrow_tensor_by_index(tensor, metadata.shard_offsets, metadata.shard_sizes) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/api.py new file mode 100644 index 0000000000000000000000000000000000000000..82589119d7afa6086b6b6289954d88676516b620 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/api.py @@ -0,0 +1,305 @@ +# mypy: allow-untyped-defs +from contextlib import contextmanager + +import torch +import torch.distributed as dist +import torch.nn as nn +from torch.distributed import distributed_c10d +from torch.distributed._shard.sharded_tensor import ShardedTensor + +from .sharder import Sharder +from .sharding_plan import ShardingPlan +from .sharding_spec import ChunkShardingSpec, ShardingSpec + + +def _shard_tensor( + tensor: torch.Tensor, sharding_spec: ShardingSpec, src_rank=0, process_group=None +) -> ShardedTensor: + """ + Given a :class:`torch.Tensor`, it shards that tensor according to the provided + ``sharding_spec``. ``src_rank`` denotes the source rank which would be + used as the ground truth of the data which would be scattered as shards + across the rest of the ranks. + + Args: + tensor (:class:`torch.Tensor`): Tensor needs to be sharded. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + + Keyword args: + src_rank (int, optional): The source rank which is used as the ground truth of + the data for the parameter that would be sharded and scattered + across the rest of the ranks. + Default: 0. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Returns: + A :class:`ShardedTensor` sharded from the given tensor. + + .. warning:: + Only :class:`torch.distributed._shard.sharding_spec.ChunkShardingSpec` is + currently supported as the ``sharding_spec``. + """ + if not tensor.is_contiguous(): + raise ValueError("input tensor is not a contiguous Tensor") + + pg = ( + process_group + if process_group is not None + else distributed_c10d._get_default_group() + ) + world_size = dist.get_world_size(pg) + current_rank = dist.get_rank(pg) + + # Validate src_rank and sharding_spec are same across all ranks. + gathered_list = [None] * world_size + dist.all_gather_object(gathered_list, (src_rank, sharding_spec), group=pg) + + for idx, entry in enumerate(gathered_list): + if src_rank != entry[0]: # type: ignore[index] + raise ValueError( + f"src_rank={src_rank} on rank: {current_rank} does not " # type: ignore[index] + f"match with src_rank={entry[0]} on rank: {idx}" # type: ignore[index] + ) + if sharding_spec != entry[1]: # type: ignore[index] + raise ValueError( + f"sharding_spec={sharding_spec} on rank: {current_rank} does not " # type: ignore[index] + f"match with sharding_spec={entry[1]} on rank: {idx}" # type: ignore[index] + ) + + st = sharding_spec.shard(tensor, src_rank=src_rank, process_group=pg) + + return st + + +def shard_parameter( + module: torch.nn.Module, + param_name: str, + sharding_spec: ShardingSpec, + src_rank=0, + process_group=None, +): + """ + Given a :class:`torch.nn.Module`, a ``param_name`` for a parameter in that + module, it shards that parameter according to the provided + ``sharding_spec``. ``src_rank`` denotes the source rank which would be + used as the ground truth of the data which would be scattered as shards + across the rest of the ranks. + + This method replaces ``module.param_name`` with a + :class:`torch.distributed._sharded_tensor.ShardedTensor` + + Args: + module (:class:`torch.nn.Module`): Module whose parameter needs to be sharded. + param_name (str): Name of the parameter of ``module`` that needs to be sharded. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + + Keyword args: + src_rank (int, optional): The source rank which is used as the ground truth of + the data for the parameter that would be sharded and scattered + across the rest of the ranks. + Default: 0. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + .. warning:: + Only :class:`torch.distributed._shard.sharding_spec.ChunkShardingSpec` is + currently supported as the ``sharding_spec``. + """ + # Perform some validation first. + if not hasattr(module, param_name): + raise AttributeError(f"{module._get_name()} has no attribute `{param_name}`") + + tensor = getattr(module, param_name) + if not isinstance(tensor, torch.Tensor): + raise ValueError( + f"Expected {type(module).__name__}.{param_name} to be a Tensor, but found {type(tensor).__name__}" + ) + + if not tensor.is_contiguous(): + raise ValueError(f"param: {param_name} is not a contiguous Tensor") + + st = _shard_tensor(tensor, sharding_spec, src_rank, process_group) + + # Replace param with ShardedTensor. + module.register_parameter(param_name, nn.Parameter(st)) + + +# Tracks the current process group in the load context manager. +_CURRENT_PROCESS_GROUP: dist.ProcessGroup | None = None + + +@contextmanager +def load_with_process_group(process_group): + """ + Context manager to set the process group with which to load a ShardedTensor. + """ + global _CURRENT_PROCESS_GROUP + if _CURRENT_PROCESS_GROUP is not None: + raise RuntimeError( + 'ProcessGroup already set by previous "load_with_process_group" ' + "context manager" + ) + _CURRENT_PROCESS_GROUP = process_group + try: + yield process_group + finally: + _CURRENT_PROCESS_GROUP = None + + +def _get_current_process_group(): + """ + Retrieves the current process group set by ``load_with_process_group``. + If not set, it just returns the default group. + """ + global _CURRENT_PROCESS_GROUP + if _CURRENT_PROCESS_GROUP is None: + return distributed_c10d._get_default_group() + else: + return _CURRENT_PROCESS_GROUP + + +def _reshard_output( + module: torch.nn.Module, resharding_spec: ShardingSpec +) -> torch.nn.Module: + """ + Hook a module with output resharding in the forward pass according + to the given ``resharding_spec``. + + Args: + module (:class:`torch.nn.Module`): Module whose output needs to be resharded. + resharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): + The specification describing how the output of the module will be resharded. + + Returns: + A :class:`torch.nn.Module` object with reshard API hooked. + """ + + def hook_func(_module, _input, output): + if isinstance(output, ShardedTensor): + return output.reshard(resharding_spec) + return output + + module.register_forward_hook(hook_func) + return module + + +def _collect_local_shard(module: torch.nn.Module) -> torch.nn.Module: + """ + Hook a module with local shards collection in the forward pass. + + This API is typically used to convert a sharded representation back to data parallel + representation. In particular, it returns the local tensor for this Shard. If the + size along the sharding dimension for the local tensor is 1, this dimension is removed + from the final result. For example a [4, 16] ShardedTensor across 4 ranks is typically + a local Tensor of size [16] across each rank and not [1, 16] across each rank. + + Args: + module (:class:`torch.nn.Module`): Module whose output is ShardedTensor and the + local tensor value needs to be returned. + + Returns: + A :class:`torch.nn.Module` object with collection API hooked. + """ + + def hook_func(_module, _input, output): + if isinstance(output, ShardedTensor): + local_tensor = output.local_tensor() + # Squeeze the # of dimensions manually, only applicable to ChunkShardingSpec + sharding_spec = output._sharding_spec + if ( + isinstance(sharding_spec, ChunkShardingSpec) + and local_tensor.size(sharding_spec.dim) == 1 # type: ignore[attr-defined, arg-type] + ): + local_tensor = local_tensor.squeeze( + output._sharding_spec.dim # type: ignore[attr-defined] + ) + return local_tensor + + module.register_forward_hook(hook_func) + return module + + +def shard_module(module: nn.Module, plan: ShardingPlan, src_rank=0, process_group=None): + """ + Shards a given module according to the provided sharding `plan`. This method + first shards all the parameters according to the given sharding `plan`. Then if + `output_plan` and `return_local_tensor` are specified in the sharding `plan`, it + will tag the output of modules according `output_plan`, convert the module's + output back to data parallel according to `return_local_tensor`. + + Needs to be called on all ranks in an SPMD fashion. + + Args: + module (:class:`torch.nn.Module`): The module to apply sharding to + plan (:class:`torch.distributed._shard.sharding_plan.ShardingPlan`): + The ShardingPlan which specified param name to ShardingSpec to apply to + each parameter. + + Keyword args: + src_rank (int, optional): The source rank which is used as the ground truth of + the data for the module that would be sharded and scattered across the rest + of the ranks. + Default: 0. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + """ + # record Sharder paths for sanity check on the plan to ensure items in the plan + # does not conflict with the submodule tree that the Sharder is working with + sharder_paths = [] + for name, spec in plan.plan.items(): + if isinstance(spec, Sharder): + sharder_paths.append(name) + + # shard the parameter according to the ShardingPlan + for name, spec in plan.plan.items(): + if isinstance(spec, ShardingSpec): + # if found a sharding spec, try to shard the parameter + module_path, _, param_name = name.rpartition(".") + + for sharder_path in sharder_paths: + if module_path.startswith(sharder_path): + raise RuntimeError( + f"ShardingPlan is in-valid, trying to shard a parameter: {name}," + f" but there's already a Sharder entry for module {sharder_path}," + f" parameter sharding should not conflict with the submodule tree" + f" that a Sharder is working with!" + ) + + mod = module.get_submodule(module_path) + shard_parameter( + mod, param_name, spec, src_rank=src_rank, process_group=process_group + ) + elif isinstance(spec, Sharder): + parent_mod_path, _, _mod_name = name.rpartition(".") + if name == "": + raise KeyError("Module path must not be empty for custom sharder!") + mod = module.get_submodule(name) + parent_mod = module.get_submodule(parent_mod_path) + sharded_mod = spec.shard(mod) + # swap this submodule with the sharded module + parent_mod.mod_name = sharded_mod + else: + raise TypeError( + f"Only `ShardingSpec` and `Sharder` are supported to shard '{name}'" + ) + + # reshard output if there's an entry in `reshard_output` for this module + if plan.output_plan is not None: + for module_path, output_spec in plan.output_plan.items(): + if isinstance(output_spec, ShardingSpec): + mod = module.get_submodule(module_path) + _reshard_output(mod, output_spec) + else: + raise TypeError( + f"Only `ShardingSpec` is supported as output_plan for '{module_path}'" + ) + # convert the output back to data parallel for the modules appears in + # `return_local_tensor` of the plan, we will call `_collect_local_shard` + # to collect the local tensor for output of modules + if plan.return_local_tensor is not None: + for module_path in plan.return_local_tensor: + mod = module.get_submodule(module_path) + _collect_local_shard(mod) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..85915636a014640d8fff5a29db602c4a114f1b1d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__init__.py @@ -0,0 +1,19 @@ +# Keep old package for BC purposes, this file should be removed once +# everything moves to the `torch.distributed.checkpoint` package. +import sys +import warnings + +import torch +from torch.distributed.checkpoint import * # noqa: F403 + + +with warnings.catch_warnings(): + warnings.simplefilter("always") + warnings.warn( + "`torch.distributed._shard.checkpoint` will be deprecated, " + "use `torch.distributed.checkpoint` instead", + DeprecationWarning, + stacklevel=2, + ) + +sys.modules["torch.distributed._shard.checkpoint"] = torch.distributed.checkpoint diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..86178c693cbb3cc807187cb20a436dcd98527d90 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/checkpoint/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/common_op_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/common_op_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..c98b8c87ca2c7ceb1608a59673738a7e57333035 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/common_op_utils.py @@ -0,0 +1,64 @@ +# mypy: allow-untyped-defs + +import torch +from torch.utils import _pytree as pytree + + +def _basic_validation(op, args=(), kwargs=None): + """ + Common validation across all ops go in here. + """ + from torch.distributed._shard.sharded_tensor import ShardedTensor + + if len(args) == 0 and (kwargs is None or len(kwargs) == 0): + raise ValueError(f" No input for '{op.__name__}'!") + + # Validate types + has_distributed_tensor = False + + def is_distributed_tensor(e): + nonlocal has_distributed_tensor + if isinstance(e, ShardedTensor): + has_distributed_tensor = True + + pytree.tree_map_(is_distributed_tensor, args) + pytree.tree_map_(is_distributed_tensor, kwargs) + + if not has_distributed_tensor: + raise TypeError( + f"torch function '{op.__name__}', with args: {args} and " + f"kwargs: {kwargs} are called without any distributed tensor!" + ) + + # Validate all distributed tensors use the same PG. + cur_pg: torch.distributed.ProcessGroup | None = None + + def validate_pg(e): + nonlocal cur_pg + if isinstance(e, ShardedTensor): + if cur_pg is not None and e._process_group is not cur_pg: + raise RuntimeError( + "All distributed tensors should use the " + "same ProcessGroup if used together in an op." + ) + cur_pg = e._process_group + + pytree.tree_map_(validate_pg, args) + pytree.tree_map_(validate_pg, kwargs) + + +def _register_default_op(op, decorator): + @decorator(op) + def tensor_default_op(types, args=(), kwargs=None, pg=None): + """ + Handles ``__torch_function__`` dispatch for the default tensor ops that + behave the same as ``torch.Tensor`` such as ``torch.Tensor.shape`` or + ``torch.Tensor.dtype``. We simply lower to the real op call with + DisableTorchFunctionSubclass context like ``torch.Tensor.__torch_function__`` + to avoid recursions. + """ + if kwargs is None: + kwargs = {} + + with torch._C.DisableTorchFunctionSubclass(): + return op(*args, **kwargs) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/metadata.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/metadata.py new file mode 100644 index 0000000000000000000000000000000000000000..63ef073b1c494ab450bca79c83f3867548140fd8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/metadata.py @@ -0,0 +1,63 @@ +# mypy: allow-untyped-defs +from dataclasses import dataclass +from functools import reduce + +from torch.distributed.remote_device import _remote_device + + +@dataclass +class ShardMetadata: + """ + Represents a shard of the overall Tensor including its + offsets, lengths and device placement. + + Args: + shard_offsets(List[int]): Offsets in the original tensor indicating + the start offsets for this shard. Should have the same rank as + the original tensor. + shard_sizes(List[int]): Integers indicating the size of each + dimension for this shard. Should have the same rank as the + original tensor. + placement(:class:`torch.distributed._remote_device`): + Specifies the placement of this shard. + """ + + __slots__ = ["shard_offsets", "shard_sizes", "placement"] + + shard_offsets: list[int] + shard_sizes: list[int] + placement: _remote_device | None + + def __init__( + self, + shard_offsets: list[int], + shard_sizes: list[int], + placement: str | _remote_device | None = None, + ): + self.shard_offsets = shard_offsets + self.shard_sizes = shard_sizes + if isinstance(placement, str): + self.placement = _remote_device(placement) + else: + self.placement = placement + if len(self.shard_offsets) != len(self.shard_sizes): + raise ValueError( + f"shard_offsets and shard_sizes should have " + f"the same number of elements, found {len(self.shard_offsets)} " + f"and {self.shard_sizes} respectively" + ) + + for i in range(len(self.shard_offsets)): + if self.shard_offsets[i] < 0: + raise ValueError("shard_offsets should be >=0") + if self.shard_sizes[i] < 0: + raise ValueError("shard_sizes should be >= 0") + + def __hash__(self): + def _hash_reduce(a, b): + return (a << 8) + hash(b) + + res = reduce(_hash_reduce, self.shard_offsets, 37) + res = reduce(_hash_reduce, self.shard_sizes, res) + res = _hash_reduce(res, self.placement) + return res diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/op_registry_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/op_registry_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..12e0b1895e2f053e6c4a975cb6d3c0118baf50bb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/op_registry_utils.py @@ -0,0 +1,41 @@ +# mypy: allow-untyped-defs +import functools +from inspect import signature + +from .common_op_utils import _basic_validation + + +""" +Common utilities to register ops on ShardedTensor +and PartialTensor. +""" + + +def _register_op(op, func, op_table): + """ + Performs basic validation and registers the provided op in the given + op_table. + """ + if len(signature(func).parameters) != 4: + raise TypeError( + f"Custom sharded op function expects signature: " + f"(types, args, kwargs, process_group), but received " + f"signature: {signature(func)}" + ) + + op_table[op] = func + + +def _decorator_func(wrapped_func, op, op_table): + """ + Decorator function to register the given ``op`` in the provided + ``op_table`` + """ + + @functools.wraps(wrapped_func) + def wrapper(types, args, kwargs, process_group): + _basic_validation(op, args, kwargs) + return wrapped_func(types, args, kwargs, process_group) + + _register_op(op, wrapper, op_table) + return wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..effae2e3cd1b89027cf06bf6e603e6ca84551520 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__init__.py @@ -0,0 +1,53 @@ +from collections.abc import Iterator +from typing import Union + +import torch.nn as nn +from torch.distributed._shard.sharded_tensor import ShardedTensor + +from .api import ShardedOptimizer + + +def named_params_with_sharded_tensor( + module: nn.Module, + prefix: str = "", + recurse: bool = True, +) -> Iterator[tuple[str, nn.Parameter | ShardedTensor]]: + r"""Returns an iterator over module parameters (together with the + ShardedTensor parameters), yielding both the name of the parameter + as well as the parameter itself. This is typically passed to a + :class:torch.distributed._shard.sharded_optim.ShardedOptimizer + + Args: + prefix (str): prefix to prepend to all parameter names. + recurse (bool): if True, then yields parameters of this module + and all submodules. Otherwise, yields only parameters that + are direct members of this module. + + Yields: + (str, Union[Tensor, ShardedTensor]): Tuple containing + the name and parameter (or ShardedTensor parameter) + + Example:: + + >>> # xdoctest: +SKIP + >>> model = torch.nn.Linear(*linear_size) + >>> shard_parameter(model, "weight", spec) + >>> for name, param in named_params_with_sharded_tensor(model): + >>> if name in ['weight']: + >>> print(param.size()) + + """ + modules = module.named_modules(prefix=prefix) if recurse else [(prefix, module)] + + memo = set() + for mod_prefix, mod in modules: + # find all sharded tensor params + for name, val in vars(mod).items(): + if isinstance(val, ShardedTensor) and val not in memo: + memo.add(val) + name = mod_prefix + ("." if mod_prefix else "") + name + yield name, val + + # find all nn.Parameters + for name, val in module.named_parameters(): + yield name, val diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3b0956890afae0df5d3360f61f8fa604e8a0711a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f344dd3e2b9ca2ad467774bf99b014bcc6d6514b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/api.py new file mode 100644 index 0000000000000000000000000000000000000000..2989e85496090782fcdb39d0e6613b82155ea23c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_optim/api.py @@ -0,0 +1,102 @@ +# mypy: allow-untyped-defs +from collections.abc import Mapping +from typing import Any + +import torch.optim as optim +from torch import Tensor +from torch.distributed._shard.sharded_tensor import ShardedTensor + + +class ShardedOptimizer(optim.Optimizer): + def __init__( + self, + named_params: Mapping[str, Tensor | ShardedTensor], + optimizer_class, + *optimizer_args, + **optimizer_kwargs, + ): + """ + ShardedOptimizer collects all tensors and local shard tensors of + ShardedTensor, then use these tensors as ``params`` for optimizers + + Args: + named_params (Dict[str, Union[Tensor, ShardedTensor]]) : a Dict + of parameters, where key is the parameter key, value is either + Tensor or ShardedTensor parameter. + optimizer_class (torch.optim.Optimizer): the Optimizer to use + locally, i.e. torch.optim.SGD, torch.optim.Adagrad, etc. + *optimizer_args: the arguments to initialize the optimizer. + **optimizer_kwargs: the key-word arguments to initialize the optimizer. + + """ + tensors: list[Tensor] = [] + for value in named_params.values(): + if isinstance(value, ShardedTensor): + tensors.extend( + local_shard.tensor for local_shard in value.local_shards() + ) + else: + tensors.append(value) + + self.named_params = named_params + self._optim = optimizer_class(tensors, *optimizer_args, **optimizer_kwargs) + self.param_groups = self._optim.param_groups + self.state = self._optim.state + + def zero_grad(self, set_to_none: bool = True): # type: ignore[override] + r"""Resets the gradients of all optimized :class:`torch.Tensor` s. + + Args: + set_to_none (bool): instead of setting to zero, set the grads to None. + This will in general have lower memory footprint, and can modestly improve performance. + However, it changes certain behaviors. For example: + 1. When the user tries to access a gradient and perform manual ops on it, + a None attribute or a Tensor full of 0s will behave differently. + 2. If the user requests ``zero_grad(set_to_none=True)`` followed by a backward pass, ``.grad``\ s + are guaranteed to be None for params that did not receive a gradient. + 3. ``torch.optim`` optimizers have a different behavior if the gradient is 0 or None + (in one case it does the step with a gradient of 0 and in the other it skips + the step altogether). + """ + self._optim.zero_grad(set_to_none) + + def step(self, closure=None): + r"""Performs a single optimization step (parameter update). + + Args: + closure (Callable): A closure that reevaluates the model and + returns the loss. Optional for most optimizers. + + .. note:: + Unless otherwise specified, this function should not modify the + ``.grad`` field of the parameters. + """ + self._optim.step(closure) + + def state_dict(self) -> dict[str, Any]: + """ + Returned state and param_groups will contain parameter keys + instead of parameter indices like torch.optim.Optimizer. + This allows for advanced functionality like optimizer re-sharding to be implemented. + """ + # TODO: implement state_dict + raise NotImplementedError("ShardedOptimizer state_dict not implemented yet!") + + def load_state_dict(self, state_dict: Mapping[str, Any]): + r"""Loads the ShardedOptimizer state. + + Args: + state_dict (dict): ShardedOptimizer state. Should be an object returned + from a call to :meth:`state_dict`. + """ + # TODO: implement load_state_dict + raise NotImplementedError( + "ShardedOptimizer load_state_dict not implemented yet!" + ) + + def add_param_group(self, param_group: Any): + r"""Add a new param group""" + # TODO: implement add_param_group + raise NotImplementedError( + "ShardedOptimizer add_param_group not implemented yet!" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..b89c4826957e2f988b4a94a80ea4a82ae943af6a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__init__.py @@ -0,0 +1,490 @@ +# mypy: allow-untyped-defs +import functools +from typing import TYPE_CHECKING + +import torch +from torch.distributed._shard.op_registry_utils import _decorator_func + +from .api import ( + _CUSTOM_SHARDED_OPS, + _SHARDED_OPS, + Shard, + ShardedTensor, + ShardedTensorBase, + ShardedTensorMetadata, + TensorProperties, +) +from .metadata import ShardMetadata + + +if TYPE_CHECKING: + from torch.distributed._shard.sharding_spec import ShardingSpec +else: + ShardingSpec = "ShardingSpec" + + +def empty( + sharding_spec: ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Returns a :class:`ShardedTensor` filled with uninitialized data. + Needs to be called on all ranks in an SPMD fashion. + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a sequence of integers defining the shape of the output + tensor. Can be a variable number of arguments or a collection like a list or tuple. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + memory_format (:class:`torch.memory_format`, optional): the desired memory format of + returned Tensor. Default: ``torch.contiguous_format``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object on each rank + """ + return ShardedTensor( + sharding_spec, + *size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + + +def ones( + sharding_spec: ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Returns a :class:`ShardedTensor` with the scalar value 1. + Needs to be called on all ranks in an SPMD fashion. + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a sequence of integers defining the shape of the output + tensor. Can be a variable number of arguments or a collection like a list or tuple. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object on each rank + """ + return full( + sharding_spec, + size, + fill_value=1, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + + +def zeros( + sharding_spec: ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Returns a :class:`ShardedTensor` filled with the scalar value 0. + Needs to be called on all ranks in an SPMD fashion. + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a sequence of integers defining the shape of the output + tensor. Can be a variable number of arguments or a collection like a list or tuple. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object on each rank + """ + return full( + sharding_spec, + size, + fill_value=0, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + + +def full( + sharding_spec: ShardingSpec, + size, + fill_value, + *, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Creates a :class:`ShardedTensor` filled with fill_value. The tensor's dtype + is inferred from fill_value. If dtype is specified, it will override the + inferred type from fill_value. Needs to be called on all ranks in an SPMD fashion. + Args: + sharding_spec (:class:`torch.distributed._sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a list, tuple, or `torch.Size` of integers defining the shape of the + output tensor. + fill_value (Scalar) - the value to fill the output tensor with. + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + Returns: + A :class:`ShardedTensor` object on each rank + """ + sharded_tensor = ShardedTensor( + sharding_spec, + *size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + torch.nn.init.constant_(sharded_tensor, fill_value) # type: ignore[arg-type] + return sharded_tensor + + +def rand( + sharding_spec: ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Creates a :class:`ShardedTensor` filled with random numbers from a uniform distribution + on the interval :math:`[0, 1)`. The shape of the tensor is defined by the + variable argument `size`. Needs to be called on all ranks in an SPMD fashion. + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a list, tuple, or `torch.Size` of integers defining the shape of the + output tensor. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object on each rank + """ + sharded_tensor = ShardedTensor( + sharding_spec, + *size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + torch.nn.init.uniform_(sharded_tensor, 0, 1) # type: ignore[arg-type] + return sharded_tensor + + +def randn( + sharding_spec: ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, +) -> ShardedTensor: + """ + Creates a :class:`ShardedTensor` filled with random numbers from a uniform distribution + with mean `0` and variance `1` (also called standard normal distribution). The shape + of the tensor is defined by the variable argument `size`. Needs to be called on all ranks + in an SPMD fashion. + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a list, tuple, or `torch.Size` of integers defining the shape of the + output tensor. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object on each rank + """ + sharded_tensor = ShardedTensor( + sharding_spec, + *size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + pin_memory=pin_memory, + memory_format=memory_format, + process_group=process_group, + init_rrefs=init_rrefs, + ) + torch.nn.init.normal_(sharded_tensor, 0, 1) # type: ignore[arg-type] + return sharded_tensor + + +def init_from_local_shards( + local_shards: list[Shard], *global_size, process_group=None, init_rrefs=False +) -> ShardedTensor: + """ + Creates an :class:`ShardedTensor` from local shards and the global metadata. + Needs to be called on all ranks in an SPMD fashion. + + Args: + local_shards (List[:class `torch.distributed._shard.sharded_tensor.Shard`]): A list + of shards that represent the local shards on this rank. + global_size (int...): a list, tuple, or `torch.Size` of integers defining the + shape of the overall sharded tensor. + + Keyword args: + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` object handle on this rank + + + Examples: + Suppose we want construct a sharded tensor on two ranks, global size = (10, 5), + each shard have a (5, 5) local tensor, we can do it like below: + + on rank 0: + >>> # xdoctest: +SKIP("not distributed") + >>> local_shard_metadata = ShardMetadata( + >>> shard_offsets=[0, 0], + >>> shard_lengths=[5, 5], + >>> placement="rank:0/cuda:0" + >>> ) + >>> local_shards = [Shard(torch.randn(5, 5), local_shard_metadata)] + >>> sharded_tensor = init_from_local_shards(local_shards, [10, 5]) + + on rank 1: + >>> # xdoctest: +SKIP("not distributed") + >>> local_shard_metadata = ShardMetadata( + >>> shard_offsets=[5, 0], + >>> shard_lengths=[5, 5], + >>> placement="rank:1/cuda:1" + >>> ) + >>> local_shards = [Shard(torch.randn(5, 5), local_shard_metadata)] + >>> sharded_tensor = init_from_local_shards(local_shards, [10, 5]) + """ + return ShardedTensor._init_from_local_shards( + local_shards, *global_size, process_group=process_group, init_rrefs=init_rrefs + ) + + +def state_dict_hook(module, destination, prefix, local_metadata): + """ + Hook to add ShardedTensor to Module's ``state_dict``. Needs to be + registered to the Module using + :meth:`torch.nn.Module._register_state_dict_hook`. + """ + for submodule_name, submodule in module.named_modules(): + for attr_name, attr in submodule.__dict__.items(): + if isinstance(attr, ShardedTensor): + mod_prefix = prefix + submodule_name + key = mod_prefix + ("." if mod_prefix else "") + attr_name + destination[key] = attr + + +def pre_load_state_dict_hook( + module, + state_dict, + prefix, + local_metadata, + strict, + missing_keys, + unexpected_keys, + error_msgs, +): + """ + Pre-load state dict hook to add ShardedTensor to the module. + """ + for submodule_name, submodule in module.named_modules(): + for attr_name in submodule.__dict__: + mod_prefix = prefix + submodule_name + key = mod_prefix + ("." if mod_prefix else "") + attr_name + if key in state_dict: + if isinstance(state_dict[key], ShardedTensor): + setattr(submodule, attr_name, state_dict[key]) + + +def custom_sharded_op_impl(func): + """ + Provides a way for users to write their own custom sharded operator. This + can be used to override existing ShardedTensor operators or write a new + one not supported by ShardedTensor. If the operator in question is covered + by ``__torch_function__`` dispatch and has a ShardedTensor as any of its + parameters, the function provided will be invoked for that operator. + + Example:: + >>> # xdoctest: +SKIP + >>> @custom_sharded_op_impl(torch.nn.functional.linear) + >>> def my_custom_sharded_linear(types, args, kwargs, process_group): + >>> ... + >>> # xdoctest: +SKIP("Undefined variables") + >>> input = torch.rand(10, 32) + >>> weight = sharded_tensor.rand(32, 16) + >>> bias = torch.rand(16) + >>> # This will call 'my_custom_sharded_linear' + >>> torch.nn.functional.linear(input, weight, bias) + + The types, args and kwargs parameters are the same parameters that are + passed to ``__torch_function__`` dispatch API + (https://pytorch.org/docs/stable/notes/extending.html#extending-torch). + There is an additional ``process_group`` parameter which is the + process_group used for the ShardedTensor and can be used by + implementations for communications within a sharded implementation. + + Args: + func(Callable): Torch function for which we want to provide a sharded + implementation (ex: torch.nn.functional.linear) + """ + return functools.partial(_decorator_func, op=func, op_table=_CUSTOM_SHARDED_OPS) + + +def _sharded_op_impl(func): + """ + Decorator to register a default sharded op. + """ + return functools.partial(_decorator_func, op=func, op_table=_SHARDED_OPS) + + +# Import all builtin sharded ops +from ._ops import * # noqa: F403 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..99b99f03f84363da528f4661f67661cd3d084929 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2fc7ad3bd8debdf66573fd668ff6c4db710f77de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logger.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logger.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7378e5548c49258aaef04df6053f7d7640c10c6a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logger.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logging_handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logging_handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e6a93675ca0c463ccb6eaa76b321bd34611bd1df Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/logging_handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/metadata.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/metadata.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d67150c45ec08407d672c68fe5a754fd90afa199 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/metadata.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/reshard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/reshard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d1b1c5bc4f75f531b355a21eb452610448a65989 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/reshard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/shard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/shard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..afaefd9dd0de6af2b3a0aff8e1938a3d6b11b11f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/shard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2fd7c54fc7a2c2ce7ca8c07bec96d31ff999eed5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..be6d01fc8e54ee214fafa847c9261db375d8b87e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__init__.py @@ -0,0 +1,13 @@ +import torch.distributed._shard.sharded_tensor._ops.misc_ops +import torch.distributed._shard.sharded_tensor._ops.tensor_ops + +# Import all ChunkShardingSpec ops +from torch.distributed._shard.sharding_spec.chunk_sharding_spec_ops.embedding import ( + sharded_embedding, +) +from torch.distributed._shard.sharding_spec.chunk_sharding_spec_ops.embedding_bag import ( + sharded_embedding_bag, +) + +from .binary_cmp import allclose, equal +from .init import constant_, kaiming_uniform_, normal_, uniform_ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c4b6eec26c9d3f7160375dc8ca787b95c2e2a3dd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/_common.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/_common.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ea64dab246bfa05ad99877541ce0ae76a819b240 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/_common.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/binary_cmp.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/binary_cmp.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..64e017d196f78a8b278b9fbc5fcee7bce9dfc4c2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/binary_cmp.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/init.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/init.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cb2e25f4a659ca7cfca926f661a68c1b369a48d5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/init.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/misc_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/misc_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a2896b72b6c34487382e6d6375f4501f59a51228 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/misc_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/tensor_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/tensor_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d2eb2b2abe2240f098f2afa0904a8fc203d038be Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/__pycache__/tensor_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/_common.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/_common.py new file mode 100644 index 0000000000000000000000000000000000000000..433f6b810cb25a2cbac8e6ccc26c8f309d130f6e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/_common.py @@ -0,0 +1,115 @@ +# mypy: allow-untyped-defs +import functools + +from torch.distributed._shard.common_op_utils import _basic_validation +from torch.distributed._shard.sharded_tensor import ( + _sharded_op_impl, + Shard, + ShardedTensor, +) + + +def _sharded_op_common(op, early_stop_func, extra_check): + """ + Inject sharded tensor op registration with common logics executed before + different behaviors are done on either local shards or a local tensor. + + Example:: + >>> # xdoctest: +SKIP("Undefined variables") + >>> op = torch.transpose + >>> @_sharded_op_impl(op) + >>> @_sharded_op_common(op, early_stop_func, extra_check) + >>> def sharded_tensor_op(types, args, kwargs, process_group): + >>> ... + >>> + >>> st = sharded_tensor.rand(32, 16) + >>> st.transpose(1, 2) + >>> # This will call '_sharded_op_common' + + Args: + op: The op to be registered and applied to all shards of the st. + early_stop_func (Callable, optional): the func for early stop. + Default: if ``None``, no early stop. + extra_check (Callable, optional): the func for extra condition check. + Default: if ``None``, no extra check. + + Return: + func (Callable): Torch function for which we want to provide a sharded + implementation (ex: torch.transpose) + """ + + def decorator_sharded_func(wrapped_func): + @functools.wraps(wrapped_func) + def wrapper(types, args=(), kwargs=None, pg=None): + _basic_validation(op, args, kwargs) + + # pyrefly: ignore [bad-index] + st = args[0] + if kwargs is None: + kwargs = {} + if extra_check: + extra_check(*args, **kwargs) + if early_stop_func: + early_stop = early_stop_func(*args, **kwargs) + if early_stop: + return st + return wrapped_func(types, args, kwargs, pg) + + return wrapper + + return decorator_sharded_func + + +def _register_sharded_op_on_local_shards( + op, early_stop_func=None, extra_check=None, customized_func=None +): + """ + Handles ``__torch_function__`` dispatch for ops which are performed on + each shard of the sharded tensor such as elementwise op like + ``torch.nn.functional.gelu`` or ``torch.nn.functional.relu``. + + For more complicated ops, a customized func can be used to generate + the new shards and sharded tensor size. + + This function expects that the original ShardingSpec for the ShardedTensor + is preserved irrespective of whether or not a customized function is used. + + Args: + op: The op to be registered and applied to all shards of the st. + early_stop_func (Callable, optional): the func for early stop. + Default: if ``None``, no early stop. + extra_check (Callable, optional): the func for extra condition check. + Default: if ``None``, no extra check. + customized_func (Callable, optional): the func for customized logic + to generate new shards and sharded tensor size. + Default: if ``None``, we simply lower to the real op call with + all local shards of the st. + + Return: + func (Callable): registered implementation for sharded op for + ``__torch_function__`` dispatch. + """ + + @_sharded_op_impl(op) + @_sharded_op_common(op, early_stop_func, extra_check) + def sharded_tensor_op_on_local_shards(types, args=(), kwargs=None, pg=None): + # pyrefly: ignore [bad-index] + st = args[0] + st_metadata = st.metadata() + local_shards = st.local_shards() + local_shards_new = [] + if customized_func: + local_shards_new, st_metadata = customized_func(args, kwargs, pg) + else: + for local_shard in local_shards: + args = (local_shard.tensor, *args[1:]) + local_shards_new.append( + Shard(op(*args, **kwargs), local_shard.metadata) + ) + return ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards_new, + st_metadata, + process_group=pg, + init_rrefs=st._init_rrefs, + sharding_spec=st.sharding_spec(), + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/binary_cmp.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/binary_cmp.py new file mode 100644 index 0000000000000000000000000000000000000000..3e940c5e4faed918f6fd1eab71d9fc6e83c4660c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/binary_cmp.py @@ -0,0 +1,82 @@ +# mypy: allow-untyped-defs +import torch +import torch.distributed as dist +import torch.distributed.distributed_c10d as distributed_c10d +from torch.distributed._shard.sharded_tensor import _sharded_op_impl, ShardedTensor + + +def _communicate_result(result, pg): + # Gather results from all ranks. + if result: + result_tensor = torch.ones( + 1, device=torch.device(torch.accelerator.current_device_index()) + ) + else: + result_tensor = torch.zeros( + 1, device=torch.device(torch.accelerator.current_device_index()) + ) + + dist.all_reduce(result_tensor, group=pg) + + expected_result = torch.ones( + 1, device=torch.device(torch.accelerator.current_device_index()) + ) * dist.get_world_size(pg) + + return torch.equal(result_tensor, expected_result) + + +def binary_cmp(cmp_fun, types, args, kwargs=None, process_group=None): + if len(args) != 2: + raise ValueError(f"Expected two arguments for torch.{cmp_fun.__name__}") + + st1 = args[0] + st2 = args[1] + if not (isinstance(st1, ShardedTensor) and isinstance(st2, ShardedTensor)): + raise TypeError( + f"Both arguments to torch.{cmp_fun.__name__} need to be of type ShardedTensor" + ) + + # Verify same PG + if st1._process_group != st2._process_group: + return False + + if distributed_c10d._rank_not_in_group( + st1._process_group + ) or distributed_c10d._rank_not_in_group(st2._process_group): + return distributed_c10d._rank_not_in_group( + st1._process_group + ) == distributed_c10d._rank_not_in_group(st2._process_group) + + # Verify metadata + if st1.metadata() != st2.metadata(): + return _communicate_result(False, st1._process_group) + + # Verify number of local shards + st1_local_shards = st1.local_shards() + st2_local_shards = st2.local_shards() + if len(st1_local_shards) != len(st2_local_shards): + return _communicate_result(False, st1._process_group) + + # kwargs must be dict-like + if kwargs is None: + kwargs = {} + # Verify each local shard + for idx in range(len(st1_local_shards)): + if st1_local_shards[idx].metadata != st2_local_shards[idx].metadata: + return _communicate_result(False, st1._process_group) + if not cmp_fun( + st1_local_shards[idx].tensor, st2_local_shards[idx].tensor, **kwargs + ): + return _communicate_result(False, st1._process_group) + + return _communicate_result(True, st1._process_group) + + +@_sharded_op_impl(torch.equal) +def equal(types, args, kwargs, process_group): + return binary_cmp(torch.equal, types, args, kwargs, process_group) + + +@_sharded_op_impl(torch.allclose) +def allclose(types, args, kwargs, process_group): + return binary_cmp(torch.allclose, types, args, kwargs, process_group) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/init.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/init.py new file mode 100644 index 0000000000000000000000000000000000000000..58365997b8154b2a93be6d1e1d5b28488348bb34 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/init.py @@ -0,0 +1,164 @@ +# mypy: allow-untyped-defs +import torch +import torch.distributed._shard.sharded_tensor as sharded_tensor +from torch.distributed._shard.sharded_tensor import _sharded_op_impl + + +def validate_param(param, param_name): + if param is None: + raise ValueError(f"param: {param_name} shouldn't be None!") + + +@_sharded_op_impl(torch.nn.init.uniform_) +def uniform_(types, args=(), kwargs=None, pg=None): + r""" + Fills the Tensor in tensor.local_shards with values drawn from the uniform + distribution :math:`\mathcal{U}(a, b)`. + Args: + tensor: tensor sharded across devices + a: the lower bound of the uniform distribution + b: the upper bound of the uniform distribution + """ + validate_param(kwargs, "kwargs") + # pyrefly: ignore [unsupported-operation] + sharded_tensor = kwargs["tensor"] + validate_param(sharded_tensor, "tensor") + # pyrefly: ignore [unsupported-operation] + a = kwargs["a"] + validate_param(a, "a") + # pyrefly: ignore [unsupported-operation] + b = kwargs["b"] + validate_param(b, "b") + + for shard in sharded_tensor.local_shards(): + torch.nn.init.uniform_(shard.tensor, a=a, b=b) + return sharded_tensor + + +@_sharded_op_impl(torch.nn.init.normal_) +def normal_(types, args=(), kwargs=None, pg=None): + r""" + Fills the Tensors in tensor.local_shards with values drawn from the normal + distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)`. + Args: + tensor: tensor sharded across devices + mean: the mean of the normal distribution + std: the standard deviation of the normal distribution + """ + validate_param(kwargs, "kwargs") + # pyrefly: ignore [unsupported-operation] + sharded_tensor = kwargs["tensor"] + validate_param(sharded_tensor, "tensor") + # pyrefly: ignore [unsupported-operation] + mean = kwargs["mean"] + validate_param(mean, "mean") + # pyrefly: ignore [unsupported-operation] + std = kwargs["std"] + validate_param(std, "std") + + for shard in sharded_tensor.local_shards(): + torch.nn.init.normal_(shard.tensor, mean=mean, std=std) + return sharded_tensor + + +@_sharded_op_impl(torch.nn.init.kaiming_uniform_) +def kaiming_uniform_(types, args=(), kwargs=None, pg=None): + r""" + Fills the Tensors in tensor.local_shards with values according to the method + described in `Delving deep into rectifiers: Surpassing human-level + performance on ImageNet classification` - He, K. et al. (2015), using a + uniform distribution. The resulting tensor will have values sampled from + :math:`\mathcal{U}(-\text{bound}, \text{bound})` where + .. math:: + \text{bound} = \text{gain} \times \sqrt{\frac{3}{\text{fan\_mode}}} + Also known as He initialization. + Args: + tensor: tensor sharded across devices + a: the negative slope of the rectifier used after this layer (only + used with ``'leaky_relu'``) + mode: either ``'fan_in'`` (default) or ``'fan_out'``. Choosing ``'fan_in'`` + preserves the magnitude of the variance of the weights in the + forward pass. Choosing ``'fan_out'`` preserves the magnitudes in the + backwards pass. + nonlinearity: the non-linear function (`nn.functional` name), + recommended to use only with ``'relu'`` or ``'leaky_relu'`` (default). + """ + validate_param(kwargs, "kwargs") + # pyrefly: ignore [unsupported-operation] + sharded_tensor = kwargs["tensor"] + validate_param(sharded_tensor, "tensor") + # pyrefly: ignore [unsupported-operation] + a = kwargs["a"] + validate_param(a, "a") + # pyrefly: ignore [unsupported-operation] + mode = kwargs["mode"] + validate_param(mode, "mode") + # pyrefly: ignore [unsupported-operation] + nonlinearity = kwargs["nonlinearity"] + validate_param(nonlinearity, "nonlinearity") + + for shard in sharded_tensor.local_shards(): + torch.nn.init.kaiming_uniform_( + shard.tensor, a=a, mode=mode, nonlinearity=nonlinearity + ) + return sharded_tensor + + +@_sharded_op_impl(torch.nn.init.constant_) +def constant_(types, args=(), kwargs=None, pg=None): + r""" + Fills the input ShardedTensor with the value \text{val}val. + Args: + tensor: tensor sharded across devices + val: the value to fill the tensor with + """ + validate_param(kwargs, "kwargs") + # pyrefly: ignore [unsupported-operation] + sharded_tensor = kwargs["tensor"] + validate_param(sharded_tensor, "tensor") + # pyrefly: ignore [unsupported-operation] + val = kwargs["val"] + validate_param(val, "val") + for shard in sharded_tensor.local_shards(): + torch.nn.init.constant_(shard.tensor, val=val) + return sharded_tensor + + +tensor_like_creation_op_map = { + torch.full_like: sharded_tensor.full, + torch.empty_like: sharded_tensor.empty, + torch.zeros_like: sharded_tensor.zeros, + torch.ones_like: sharded_tensor.ones, + torch.rand_like: sharded_tensor.rand, + torch.randn_like: sharded_tensor.randn, +} + + +# tensor ops that behave the same as the default tensor +def register_tensor_creation_op(op): + @_sharded_op_impl(op) + def tensor_creation_op(types, args=(), kwargs=None, pg=None): + """ + Handles ``__torch_function__`` dispatch for tensor creation ops that + takes a ShardedTensor as argument, such as ``torch.zeros_like`` or + ``torch.full_like``. + """ + creation_op = tensor_like_creation_op_map.get(op) + if creation_op is None: + raise RuntimeError(f"Tensor creation {op} not supported!") + if kwargs is None: + kwargs = {} + + # pyrefly: ignore [bad-index] + st = args[0] + + new_st = creation_op(st.sharding_spec(), st.size(), *args[1:], **kwargs) # type: ignore[operator] + return new_st + + +register_tensor_creation_op(torch.full_like) +register_tensor_creation_op(torch.empty_like) +register_tensor_creation_op(torch.zeros_like) +register_tensor_creation_op(torch.ones_like) +register_tensor_creation_op(torch.rand_like) +register_tensor_creation_op(torch.randn_like) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/misc_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/misc_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..8b84c1684c32456989e3998b3d4c30c34cb5dbf4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/misc_ops.py @@ -0,0 +1,12 @@ +# mypy: allow-untyped-defs +import torch +from torch.distributed._shard.sharded_tensor import _sharded_op_impl + + +# This is used by `_apply()` within module.py to set new +# parameters after apply a certain method, we should follow +# the future behavior of overwriting the existing tensor +# instead of doing in-place change using `.data = `. +@_sharded_op_impl(torch._has_compatible_shallow_copy_type) +def tensor_has_compatible_shallow_copy_type(types, args=(), kwargs=None, pg=None): + return False diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/tensor_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/tensor_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..4dc81f0d174453e561b6d34c8f4c330521a63b84 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/_ops/tensor_ops.py @@ -0,0 +1,222 @@ +# mypy: allow-untyped-defs +import copy + +import torch +from torch.distributed._shard.common_op_utils import _register_default_op +from torch.distributed._shard.sharded_tensor import ( + _sharded_op_impl, + Shard, + ShardedTensor, +) + +from ._common import _register_sharded_op_on_local_shards + + +# Tensor properties access +_register_default_op(torch.Tensor.shape.__get__, _sharded_op_impl) # type: ignore[attr-defined] +_register_default_op(torch.Tensor.dtype.__get__, _sharded_op_impl) # type: ignore[attr-defined] +_register_default_op(torch.Tensor.layout.__get__, _sharded_op_impl) # type: ignore[attr-defined] +_register_default_op(torch.Tensor.size, _sharded_op_impl) +_register_default_op(torch.Tensor.dim, _sharded_op_impl) +_register_default_op(torch.Tensor.ndim.__get__, _sharded_op_impl) # type: ignore[attr-defined] +_register_default_op(torch.Tensor.is_contiguous, _sharded_op_impl) +_register_default_op(torch.Tensor.contiguous, _sharded_op_impl) +_register_default_op(torch.Tensor.is_floating_point, _sharded_op_impl) + +# __reduce_ex__ to dispatch to get_state/set_state +_register_default_op(torch.Tensor.__reduce_ex__, _sharded_op_impl) + +# autograd related properties +_register_default_op(torch.Tensor.requires_grad.__get__, _sharded_op_impl) # type: ignore[attr-defined] +# TODO: set grad with a ShardedTensor that consists of all local grads +_register_default_op(torch.Tensor.grad.__get__, _sharded_op_impl) # type: ignore[union-attr] +_register_default_op(torch.Tensor.grad_fn.__get__, _sharded_op_impl) # type: ignore[union-attr] +_register_default_op(torch.Tensor.is_leaf.__get__, _sharded_op_impl) # type: ignore[attr-defined] + + +# device property is ambiguous as from a global perspective, +# ShardedTensor.device consists of multiple devices (might even across hosts) +# We choose to return the current device of the local tensor to represent +# the device property on each rank +@_sharded_op_impl(torch.Tensor.device.__get__) +def tensor_device(types, args=(), kwargs=None, pg=None): + # pyrefly: ignore [bad-index] + self_st = args[0] + # Validate types + if not isinstance(self_st, ShardedTensor): + raise TypeError("input needs to be a ShardedTensor") + dev: torch.device + if self_st._local_shards: + dev = self_st._local_shards[0].tensor.device + elif pg and pg._get_backend_name() == "gloo": + dev = torch.device("cpu") + else: + dev = torch.device(torch.cuda.current_device()) + return dev + + +@_sharded_op_impl(torch.Tensor.is_meta.__get__) # type: ignore[attr-defined] +def st_is_meta(types, args=(), kwargs=None, pg=None): + # pyrefly: ignore [bad-index] + return args[0].local_tensor().is_meta + + +def sharded_type_as_check(*args, **kwargs): + """ + Perform extra checks for the sharded_type_as op such as the input needs to + be either a Tensor or ShardedTensor. + + Args: same as ``torch.Tensor.type_as``. + + Return: None + """ + if len(args) < 2: + raise ValueError("Needs to give a tensor to cast type as!") + if not isinstance(args[1], torch.Tensor) and not isinstance(args[1], ShardedTensor): + raise ValueError("Needs to give a Tensor or ShardedTensor to cast type as!") + + +def same_dtype(*args, **kwargs): + """ + When the dtype is the same, return the original ShardedTensor. + + Args: same as ``torch.Tensor.type_as``. + + Return (bool): Whether to return early or not. + """ + return args[0].dtype == args[1].dtype + + +def sharded_type_as(args, kwargs, pg): + """ + Handles ``__torch_function__`` dispatch for the ``torch.Tensor.type_as`` op. + + Args: same as ``torch.Tensor.type_as``. + + Return: + new_local_shards (List[Shard]): Local shards for the new sharded tensor. + st_meta (ShardedTensorMetadata): Metadata of the new sharded tensor. + """ + st = args[0] + tensor = args[1] + if isinstance(tensor, ShardedTensor): + tensor = tensor.local_tensor() + new_local_shards = [ + Shard(shard.tensor.type_as(tensor), shard.metadata) + for shard in st.local_shards() + ] + st_meta = copy.deepcopy(st._metadata) + st_meta.tensor_properties.dtype = tensor.dtype + return new_local_shards, st_meta + + +_register_sharded_op_on_local_shards( + torch.Tensor.type_as, + early_stop_func=same_dtype, + extra_check=sharded_type_as_check, + customized_func=sharded_type_as, +) + + +def sharded_deepcopy(args, kwargs, pg): + # NOTE: we directly implement deepcopy magic method + # instead of using the default tensor.__deepcopy__ + # and implement clone(). This is because the default + # tensor deepcopy copies every attribute, but the + # process_group in ShardedTensor cannot be deep copied. + self_st = args[0] + new_local_shards = copy.deepcopy(self_st.local_shards()) + new_metadata = copy.deepcopy(self_st.metadata()) + return new_local_shards, new_metadata + + +_register_sharded_op_on_local_shards( + torch.Tensor.__deepcopy__, + customized_func=sharded_deepcopy, +) + + +@_sharded_op_impl(torch.Tensor.copy_) +def sharded_inplace_copy(types, args, kwargs, pg): + # NOTE: inplace op don't need to rewrap + kwargs = {} if kwargs is None else kwargs + self_st = args[0] + new_st = args[1] + nonblocking = kwargs.get("non_blocking", False) + for local_shard, new_shard in zip(self_st.local_shards(), new_st.local_shards()): + if local_shard.metadata != new_shard.metadata: + raise RuntimeError( + "inplace copy can only happen between two ShardedTensor with same metadata!" + ) + for local_shard, new_shard in zip(self_st.local_shards(), new_st.local_shards()): + local_shard.tensor.copy_(new_shard.tensor, nonblocking) + + return self_st + + +def sharded_clone(args, kwargs, pg): + self_st = args[0] + desire_memory_format = kwargs.get("memory_format", None) + if desire_memory_format and desire_memory_format != torch.preserve_format: + raise RuntimeError("Only support torch.preserve_format for ShardedTensor!") + cloned_local_shards = [ + Shard( + local_shard.tensor.clone(memory_format=desire_memory_format), + metadata=copy.deepcopy(local_shard.metadata), + ) + for local_shard in self_st.local_shards() + ] + new_metadata = copy.deepcopy(self_st.metadata()) + return cloned_local_shards, new_metadata + + +_register_sharded_op_on_local_shards( + torch.Tensor.clone, + customized_func=sharded_clone, +) + + +def sharded_detach(args, kwargs, pg): + self_st = args[0] + detached_local_shards = [ + Shard( + local_shard.tensor.detach(), + metadata=copy.deepcopy(local_shard.metadata), + ) + for local_shard in self_st.local_shards() + ] + new_metadata = copy.deepcopy(self_st.metadata()) + new_metadata.tensor_properties.requires_grad = False + return detached_local_shards, new_metadata + + +_register_sharded_op_on_local_shards( + torch.Tensor.detach, + customized_func=sharded_detach, +) + + +@_sharded_op_impl(torch.Tensor.requires_grad_) +def tensor_requires_grad_set(types, args=(), kwargs=None, pg=None): + # pyrefly: ignore [bad-index] + self_st = args[0] + # Validate types + if not isinstance(self_st, ShardedTensor): + raise TypeError("input needs to be a ShardedTensor") + + if kwargs is None: + kwargs = {} + + requires_grad = args[1] if len(args) > 1 else kwargs.get("requires_grad", True) + if requires_grad == self_st.requires_grad: + return self_st + + for local_shard in self_st.local_shards(): + local_shard.tensor.requires_grad_(requires_grad) + + # update the wrapper class property + with torch._C.DisableTorchFunctionSubclass(): + self_st.requires_grad_(requires_grad) + # update the metadata in the meanwhile + self_st._metadata.tensor_properties.requires_grad = requires_grad + return self_st diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/api.py new file mode 100644 index 0000000000000000000000000000000000000000..c97eaef573adecc7e22f172eb787ea8b32d7482b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/api.py @@ -0,0 +1,1372 @@ +# mypy: allow-untyped-defs +from __future__ import annotations # type: ignore[attr-defined] + +import copy +import operator +import threading +import warnings +import weakref +from dataclasses import dataclass +from functools import reduce +from typing import cast, TYPE_CHECKING +from typing_extensions import deprecated + +import torch +import torch.distributed as dist +import torch.distributed._shard.sharding_spec as shard_spec +from torch._utils import _get_device_module +from torch.distributed import distributed_c10d, rpc +from torch.distributed._shard._utils import DEPRECATE_MSG +from torch.distributed._shard.sharding_spec._internals import ( + check_tensor, + validate_non_overlapping_shards_metadata, +) +from torch.distributed._shard.sharding_spec.api import ( + _dispatch_custom_op, + _has_custom_op, +) +from torch.distributed.remote_device import _remote_device +from torch.utils import _pytree as pytree + +from .metadata import ShardedTensorMetadata, TensorProperties +from .reshard import reshard_local_shard, reshuffle_local_shard +from .shard import Shard +from .utils import ( + _flatten_tensor_size, + _parse_and_validate_remote_device, + _validate_output_tensor_for_gather, + build_global_metadata, + build_metadata_from_local_shards, +) + + +if TYPE_CHECKING: + from collections.abc import Callable, Sequence + + from torch.distributed._shard.metadata import ShardMetadata + + +# Tracking for sharded tensor objects. +_sharded_tensor_lock = threading.Lock() +_sharded_tensor_current_id = 0 +_sharded_tensor_map: dict[int, weakref.ReferenceType[ShardedTensor]] = {} + +# Default sharded ops +_SHARDED_OPS: dict[Callable, Callable] = {} + +# Customized user ops +_CUSTOM_SHARDED_OPS: dict[Callable, Callable] = {} + + +def _register_remote_shards( + sharded_tensor_id: int, rrefs: list[rpc.RRef[Shard]], rpc_rank: int +): + with _sharded_tensor_lock: + if sharded_tensor_id not in _sharded_tensor_map: + raise RuntimeError( + f"Could not find sharded_tensor_id: {sharded_tensor_id} in map: {_sharded_tensor_map.keys()}" + ) + + sharded_tensor = _sharded_tensor_map[sharded_tensor_id]() + if sharded_tensor is None: + raise RuntimeError("ShardedTensor weakref has been deallocated") + else: + sharded_tensor._register_remote_shards(rrefs, rpc_rank) + + +class ShardedTensorBase(torch.Tensor): + _sharding_spec: shard_spec.ShardingSpec + _metadata: ShardedTensorMetadata + _local_shards: list[Shard] + + def __new__(cls, sharding_spec: shard_spec.ShardingSpec, *size, **kwargs): + # Use __new__ to construct a wrapper tensor, for recording tensor + # properties and logging purposes. + torch._C._log_api_usage_once("torch.distributed._shard.sharded_tensor") + + # check sharding spec and build sharded tensor metadata + if not isinstance(sharding_spec, shard_spec.ShardingSpec): + raise ValueError(f"Expecting ShardingSpec but got: {type(sharding_spec)}") + + sizes = _flatten_tensor_size(size) + dtype = kwargs["dtype"] + layout = kwargs["layout"] + pin_memory = kwargs["pin_memory"] + requires_grad = kwargs["requires_grad"] + + if dtype is None: + dtype = torch.get_default_dtype() + + tensor_properties = TensorProperties( + dtype, layout, requires_grad, pin_memory=pin_memory + ) + sharded_tensor_metadata = sharding_spec.build_metadata( + sizes, tensor_properties=tensor_properties + ) + + r = torch.Tensor._make_wrapper_subclass( + cls, + sizes, + dtype=dtype, + layout=layout, + pin_memory=pin_memory, + requires_grad=requires_grad, + ) + # set sharding spec + r._sharding_spec = sharding_spec + # set metadata + r._metadata = sharded_tensor_metadata + # set local shards + r._local_shards = [] + return r + + def metadata(self) -> ShardedTensorMetadata: + """ + Returns a :class:`ShardedTensorMetadata` object corresponding to the + metadata for the entire tensor. + """ + return self._metadata + + def local_shards(self) -> list[Shard]: + """ + Returns a list of :class:`Shard' corresponding to the + local shards for this rank. Returns an empty list if the current rank + does not host any shards for this Tensor. + """ + return self._local_shards + + @classmethod + def _init_from_local_shards_and_global_metadata( + cls, + local_shards: list[Shard], + sharded_tensor_metadata: ShardedTensorMetadata, + sharding_spec=None, + ) -> ShardedTensorBase: + """ + Initialize a ShardedTensorBase with local shards and a global + ShardedTensorMetadata built on each rank. + Warning: This API is experimental and subject to change. It does + not do cross rank validations, and fully rely on the user + for the correctness of sharded_tensor_metadata on each rank + """ + shards_metadata = sharded_tensor_metadata.shards_metadata + tensor_properties = sharded_tensor_metadata.tensor_properties + + if len(shards_metadata) == 0: + raise ValueError("shards_metadata must not be empty!") + + if tensor_properties.layout != torch.strided: + raise ValueError("Only torch.strided layout is currently supported") + + if sharding_spec is None: + spec = shard_spec._infer_sharding_spec_from_shards_metadata(shards_metadata) + else: + spec = sharding_spec + + sharded_tensor_base = ShardedTensorBase.__new__( + ShardedTensor, + spec, + sharded_tensor_metadata.size, + dtype=tensor_properties.dtype, + layout=tensor_properties.layout, + pin_memory=tensor_properties.pin_memory, + requires_grad=tensor_properties.requires_grad, + ) + + # check if shards_metadata have overlap shards + validate_non_overlapping_shards_metadata(shards_metadata) + + # check if the shards_metadata is compatible with overall size of the sharded tensor. + check_tensor(shards_metadata, list(sharded_tensor_metadata.size)) + + # done validation, add local_shards + sharded_tensor_base._local_shards = local_shards + return sharded_tensor_base + + @classmethod + def __torch_dispatch__(cls, func, types, args=(), kwargs=None): # type: ignore[override] + raise RuntimeError( + f"A {cls.__name__} object is being used from c++ while calling {func.__module__}.{func.__name__} " + "but the there is no custom __torch_dispatch__ implementation for it." + ) + + +class ShardedTensor(ShardedTensorBase): + """ + ShardedTensor is an torch.Tensor subclass to represent Tensors that are sharded + across multiple devices and multiple processes. + + ShardedTensor is initialized in an SPMD like fashion where each rank + initializes the ShardedTensor. The ShardedTensor object on each rank + then only stores the local shard for the Tensor and provides global + metadata for all the shards. + + ShardedTensor doesn't provide any Tensor like operations but is a wrapper + providing the Tensor representing the local shard and the global metadata. + Using these, users can build their custom distributed._sharded computations + on top of this primitive. The local shards are all initialized using the + create_op specified by tensor_init_params.create_op, e.g., torch.ones, or + torch.empty + + Args: + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The specification + describing how to shard the Tensor. + size (int...): a sequence of integers defining the shape of the output + tensor. Can be a variable number of arguments or a collection like a list or tuple. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned Tensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned tensor. Default: ``False``. + pin_memory (bool, optional): If set, returned tensor would be allocated in + the pinned memory. Works only for CPU tensors. Default: ``False``. + memory_format (:class:`torch.memory_format`, optional): the desired memory format of + returned Tensor. Default: ``torch.contiguous_format``. + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + .. note:: ShardedTensor uses collectives to do various operations, i.e. it + uses all_gather to do cross rank validations. For NCCL-based process + groups, internal tensor representations of objects must be moved to the + GPU device before communication takes place. In this case, the device + used is given by ``torch.cuda.current_device()`` and it is the user's + responsibility to ensure that this is set so that each rank has an + individual GPU, via ``torch.cuda.set_device()`` + + """ + + def __new__(cls, sharding_spec: shard_spec.ShardingSpec, *size, **kwargs): + self = super().__new__(cls, sharding_spec, *size, **kwargs) + return self + + def __init__( + self, + sharding_spec: shard_spec.ShardingSpec, + *size, + dtype=None, + layout=torch.strided, + requires_grad=False, + pin_memory=False, + memory_format=torch.contiguous_format, + process_group=None, + init_rrefs=False, + ): + # prepare initialization, initialize fields like + # _process_group, _local_shards, etc. + self._prepare_init(process_group=process_group, init_rrefs=init_rrefs) + + if layout != torch.strided: + raise ValueError("Only torch.strided layout is currently supported") + + if memory_format != torch.contiguous_format: + raise ValueError( + "Only torch.contiguous_format memory_format is currently supported" + ) + + self._metadata.tensor_properties.memory_format = memory_format + + current_rank = dist.get_rank() # global rank + + for shard_metadata in self._metadata.shards_metadata: + rank, device = _parse_and_validate_remote_device( + self._process_group, shard_metadata.placement + ) + if rank == current_rank: + local_tensor = _create_tensor_from_params( + shard_metadata.shard_sizes, + local_device=device, + tensor_properties=self._metadata.tensor_properties, + ) + self._local_shards.append(Shard(local_tensor, shard_metadata)) + + # do post initialization (i.e. register sharded_tensor_id, initialize_rpc) + self._post_init() + + def _prepare_init(self, process_group=None, init_rrefs=False): + self._init_rrefs = init_rrefs + self._sharded_tensor_id = None + + self._process_group = self._normalize_pg(process_group) + self._remote_shards: dict[int, list[rpc.RRef[Shard]]] = {} + + def _post_init(self): + # Initialize RPC if available. + if self._init_rrefs: + with _sharded_tensor_lock: + global _sharded_tensor_current_id, _sharded_tensor_map + self._sharded_tensor_id = _sharded_tensor_current_id + _sharded_tensor_map[self._sharded_tensor_id] = weakref.ref(self) + _sharded_tensor_current_id += 1 + + if not rpc._is_current_rpc_agent_set(): + raise RuntimeError( + "RPC Framework needs to be initialized using" + " torch.distributed.rpc.init_rpc if init_rrefs is set to True" + ) + self._init_rpc() + + def __del__(self): + # Clean up the global map. + with _sharded_tensor_lock: + global _sharded_tensor_current_id, _sharded_tensor_map + if ( + hasattr(self, "_sharded_tensor_id") + and self._sharded_tensor_id in _sharded_tensor_map + ): + _sharded_tensor_map.pop(self._sharded_tensor_id) # type: ignore[call-overload] + + def _init_rpc(self): + # Validate PG and RPC ranks match. + pg_rank = dist.get_rank() + rpc_rank = rpc.get_worker_info().id + if pg_rank != rpc_rank: + raise ValueError( + f"Default ProcessGroup and RPC ranks must be " + f"the same for ShardedTensor, found process group rank: " + f"{pg_rank} and RPC rank: {rpc_rank}" + ) + + self._remote_shards = {} + + # Gather all the sharded tensor ids. + worker_infos = rpc._get_current_rpc_agent().get_worker_infos() + rank_to_name = {} + name_to_rank = {} + + for worker_info in worker_infos: + rank_to_name[worker_info.id] = worker_info.name + name_to_rank[worker_info.name] = worker_info.id + + all_tensor_ids = rpc.api._all_gather(self._sharded_tensor_id) + + # Share the local shards to the entire world. + futs = [] + rpc_rank = rpc.get_worker_info().id + for rank in range(dist.get_world_size()): + # Skip self. + if rank == dist.get_rank(): + continue + + if len(self.local_shards()) != 0: + rrefs: list[rpc.RRef[Shard]] = [ + rpc.RRef(shard) for shard in self.local_shards() + ] + fut = rpc.rpc_async( + rank, + _register_remote_shards, + args=(all_tensor_ids[rank_to_name[rank]], rrefs, rpc_rank), + ) + futs.append(fut) + + torch.futures.wait_all(futs) + + # Barrier for all RPCs to finish on all ranks. + rpc.api._all_gather(None) + + def _get_preferred_device(self) -> torch.device: + """ + Return the preferred device to be used when creating tensors for collectives. + This method takes into account the associated process group + """ + backend = dist.get_backend(self._process_group) + if backend == dist.Backend.NCCL: + return torch.device(torch.cuda.current_device()) + elif backend == dist.Backend.GLOO: + return torch.device("cpu") + else: + backend_config = dist.BackendConfig(backend) + for device, backend_str in backend_config.get_device_backend_map().items(): + if backend_str == backend and device != "cpu": + return torch.device( + device, _get_device_module(device).current_device() + ) + return torch.device("cpu") + + def gather( # type: ignore[override] + self, + dst: int = 0, + out: torch.Tensor | None = None, + enforce_dtype: bool = False, + dtype: torch.dtype | None = None, + ) -> None: + """ + Creates a full :class:`Tensor` on rank ``dst`` by gathering all shards of the + sharded tensor. + + The API needs to be called on all ranks in SPMD fashion. All ranks should have + the same ``dst``. ``out`` should be a tensor of the same size as the overall + size of the sharded tensor on ``dst`` and ``None`` on all other ranks. + + Args: + dst(int): The rank where full tensor is constructed. + Default: 0 + out (:class `torch.Tensor`, optional): The output full tensor. + Must be provided ONLY on ``dst`` rank. + Default: ``None`` + enforce_dtype (bool): Deprecated, please use dtype instead. Force the + gathered tensors to be the same type as input and output. + dtype (torch.dtype): Force the gathered tensors to be this dtype. + Default: ``None`` + """ + + def shard_size(shard_md): + return reduce(operator.mul, shard_md.shard_sizes) # type: ignore[attr-defined] + + if enforce_dtype: + warnings.warn( + "`enforce_dtype` is deprecated. Please use `dtype` instead.", + FutureWarning, + stacklevel=2, + ) + + rank = dist.get_rank(self._process_group) + full_size = self.metadata().size + _validate_output_tensor_for_gather(rank, dst, full_size, out) + + local_shards = self.local_shards() + world_size = dist.get_world_size(self._process_group) + rank_sizes = [0 for _ in range(world_size)] + max_rank_size = 0 + shard_placement: dict[ShardMetadata, tuple[int, int]] = {} + # collect sizes + for shard_md in self.metadata().shards_metadata: + shard_rank = cast(_remote_device, shard_md.placement).rank() + if shard_rank is None: + raise AssertionError + + shard_placement[shard_md] = (shard_rank, rank_sizes[shard_rank]) + rank_sizes[shard_rank] += shard_size(shard_md) + max_rank_size = max(max_rank_size, rank_sizes[shard_rank]) + + gather_list: list[torch.Tensor] | None + if rank == dst: + if out is None: + raise AssertionError + if enforce_dtype: + # enforce_dtype is deprecated. Do it for backward compatibility. + dtype = out.dtype + # TODO make it as a view of out tensor + gather_list = [ + torch.empty((max_rank_size,), device=out.device, dtype=dtype) + for _ in range(world_size) + ] + else: + gather_list = None + + with torch.no_grad(): + if enforce_dtype and len(local_shards) > 0: + # enforce_dtype is deprecated. Do it for backward compatibility. + dtype = local_shards[0].tensor.dtype + data = torch.empty( + max_rank_size, device=self._get_preferred_device(), dtype=dtype + ) + + for shard in local_shards: + src = shard.tensor.flatten() + if src.nelement() == 0: + warnings.warn( + "Gathering a tensor with zero elements on rank " + str(rank), + stacklevel=2, + ) + continue + shard_offset = shard_placement[shard.metadata][1] + data[shard_offset : shard_offset + src.numel()].copy_(src) + + dist.gather( + tensor=data, + gather_list=gather_list, + dst=dst, + group=self._process_group, + ) + if rank != dst: + return + # In _validate_output_tensor_for_gather, we raise if out == None and rank == dst + out = cast(torch.Tensor, out) + if gather_list is None: + raise AssertionError + + full_size = self.metadata().size + dims = len(full_size) + for shard_md in self.metadata().shards_metadata: + rank, rank_offset = shard_placement[shard_md] + tensor = gather_list[rank] + tensor = tensor[rank_offset : rank_offset + shard_size(shard_md)] + tensor = tensor.view(shard_md.shard_sizes) + + out_narrow_view = out + for dim in range(dims): + out_narrow_view = out_narrow_view.narrow( + dim, + shard_md.shard_offsets[dim], + shard_md.shard_sizes[dim], + ) + + out_narrow_view.copy_(tensor) + + def cpu( + self, memory_format=torch.preserve_format, process_group=None + ) -> ShardedTensor: + """ + Returns a copy of this object in CPU memory. + + If this ShardedTensor is already on CPU memory, then no copy is + performed and original object is returned. + + .. note:: When moving a ShardedTensor from GPU to CPU, the ShardedTensor might + need to be managed by a different type of ProcessGroup(i.e. ProcessGroupGloo), + it is the user's responsibility to explicitly pass in a new process_group that + is compatible with CPU. + """ + # TODO: make this a __torch_function__ op once ShardedTensor becomes a + # torch.Tensor subclass, see https://github.com/pytorch/pytorch/issues/75402 + if ( + memory_format != torch.preserve_format + and memory_format != torch.contiguous_format + ): + raise RuntimeError( + "Only `torch.contiguous_format` or " + "`torch.preserve_format` is supported!" + ) + all_on_cpu = True + for meta in self.metadata().shards_metadata: + all_on_cpu &= meta.placement.device().type == "cpu" # type: ignore[union-attr] + + # if every shard is already on CPU, return the original object + if all_on_cpu: + return self + + # if not, returns a copy of this object on CPU + list_shards: list[Shard] = [] + # move all local shards to cpu, and change metadata + for shard in self._local_shards: + cpu_tensor = shard.tensor.cpu(memory_format=memory_format) # type: ignore[call-arg] + metadata = copy.deepcopy(shard.metadata) + metadata.placement._device = torch.device("cpu") # type: ignore[union-attr] + list_shards.append(Shard(cpu_tensor, metadata)) + + st_meta = copy.deepcopy(self.metadata()) + for meta in st_meta.shards_metadata: + if meta.placement.device().type != "cpu": # type: ignore[union-attr] + meta.placement._device = torch.device("cpu") # type: ignore[union-attr] + + pg = self._process_group if process_group is None else process_group + st_cpu = ShardedTensor._init_from_local_shards_and_global_metadata( + list_shards, + sharded_tensor_metadata=st_meta, + process_group=pg, + init_rrefs=self._init_rrefs, + ) + return st_cpu + + def cuda( + self, + device=None, + non_blocking=False, + memory_format=torch.preserve_format, + process_group=None, + ) -> ShardedTensor: + """ + Returns a copy of this object in CUDA memory, if the original ShardedTensor + is on CPU, we will move the local shard to the current GPU device of each + process in a SPMD fashion. + If this ShardedTensor is already on CUDA memory and local shards on each rank are + already on current device, we still returns a new ShardedTensor object with new + metadata, but no underlying data movements are performed. + .. note:: When moving a ShardedTensor from CPU to GPU, the ShardedTensor might + need to be managed by a different type of ProcessGroup(i.e. ProcessGroupNCCL), + it is the user's responsibility to explicitly pass in a new process_group that + is compatible with GPU. + """ + if ( + memory_format != torch.preserve_format + and memory_format != torch.contiguous_format + ): + raise RuntimeError( + "Only `torch.contiguous_format` or " + "`torch.preserve_format` is supported!" + ) + + if device is not None: + device = torch.device(device) if isinstance(device, str) else device + if not ( + isinstance(device, torch.device) + and device.index == torch.accelerator.current_device_index() + ): + raise AssertionError( + f"Only device without device id (e.g. 'cpu' or '{device.type}') " + f"is expected for ShardedTensor!" + ) + current_device = torch.device(torch.accelerator.current_device_index()) + # returns a copy of ShardedTensor on accelerator current device + list_shards: list[Shard] = [] + # move all local shards to current device, and change metadata + # if local shards already on the current device, there's no + # real data movement, only the metadata are copied. + for shard in self._local_shards: + new_tensor = shard.tensor.to( + device=current_device, + non_blocking=non_blocking, + memory_format=memory_format, + ) # type: ignore[call-arg] + metadata = copy.deepcopy(shard.metadata) + metadata.placement._device = current_device # type: ignore[union-attr] + + list_shards.append(Shard(new_tensor, metadata)) + + st_meta = copy.deepcopy(self.metadata()) + for meta in st_meta.shards_metadata: + if meta.placement.device().type != current_device.type: # type: ignore[union-attr] + meta.placement._device = current_device # type: ignore[union-attr] + + pg = self._process_group if process_group is None else process_group + # we need to use `init_from_local_shards` to communicate between ranks + # and update the sharding spec/shards metadata. + return ShardedTensor._init_from_local_shards_and_global_metadata( + list_shards, + sharded_tensor_metadata=st_meta, + process_group=pg, + init_rrefs=self._init_rrefs, + ) + + def to(self, *args, **kwargs) -> ShardedTensor: + current_device: torch.device + if self._local_shards: + current_device = self._local_shards[0].tensor.device + elif self._process_group._get_backend_name() == "gloo": + current_device = torch.device("cpu") + else: + current_device = torch.device(torch.accelerator.current_device_index()) + current_dtype = self.dtype + device_to = current_device + dtype_to = current_dtype + if len(args) == 1: + if isinstance(args[0], torch.dtype): + dtype_to = args[0] + elif isinstance(args[0], torch.device): + device_to = args[0] + elif isinstance(args[0], (str, int)): + device_to = torch.device(args[0]) + elif isinstance(args[0], torch.Tensor): + dtype_to = args[0].dtype + device_to = args[0].device + else: + raise RuntimeError(f"ShardedTensor.to() have wrong arguments: {args}") + elif len(args) == 2: + device_to, dtype_to = args + else: + dtype_to = kwargs.get("dtype", current_dtype) + device_to = kwargs.get("device", current_device) + + device_to = ( + torch.device(device_to) if isinstance(device_to, (str, int)) else device_to + ) + + if device_to.type != "cpu": + # if device_to set to accelerator, set to current device even + # if user specifies the device index. + current_idx = torch.accelerator.current_device_index() + if device_to.index != current_idx: + warnings.warn( + "ShardedTensor.to only move tensor to its current device" + "If you want to put to different device, use `reshard` instead.", + stacklevel=2, + ) + device_to = torch.device(current_idx) + + copy_tensor = kwargs.get("copy", False) + non_blocking = kwargs.get("non_blocking", False) + memory_format = kwargs.get("memory_format", torch.preserve_format) + process_group = kwargs.get("process_group") + + if ( + not copy_tensor + and dtype_to == current_dtype + and device_to == current_device + ): + # already have correct dtype and device, return itself + return self + + # returns a copy of ShardedTensor on target device + list_shards: list[Shard] = [] + + for shard in self._local_shards: + new_tensor = shard.tensor.to( # type: ignore[call-overload] + device=device_to, + dtype=dtype_to, + non_blocking=non_blocking, + copy=copy_tensor, + memory_format=memory_format, + ) + metadata = copy.deepcopy(shard.metadata) + if metadata.placement is not None: + metadata.placement._device = device_to + list_shards.append(Shard(new_tensor, metadata)) + + # update metadata + st_meta = copy.deepcopy(self.metadata()) + st_meta.tensor_properties.dtype = dtype_to + for meta in st_meta.shards_metadata: + meta.placement._device = device_to # type: ignore[union-attr] + + pg = self._process_group if process_group is None else process_group + # we need to use `init_from_local_shards` to communicate between ranks + # and update the sharding spec/shards metadata. + st_to = ShardedTensor._init_from_local_shards_and_global_metadata( + list_shards, + sharded_tensor_metadata=st_meta, + process_group=pg, + init_rrefs=self._init_rrefs, + ) + return st_to + + @classmethod + def _normalize_pg( + cls, process_group: dist.ProcessGroup | None + ) -> dist.ProcessGroup: + if process_group is not None: + return process_group + return distributed_c10d._get_default_group() + + @classmethod + def _init_from_local_shards( + cls, + local_shards: list[Shard], + *global_size, + process_group=None, + init_rrefs=False, + ): + # recalc metadata handles special ST creation cases like each rank only has tensor available + # caller need to provide None on the unknown dimension of the global size + # We will change None into zeros and go through the same amount of checks as before to create ST + # and use all_gather to calculate the offsets and global size for metadata + # It is compatible with the current use case since, conventionally we don't pass None as global size + # Therefore the old path won't trigger the new feature + recalc_metadata = False + for dim in global_size: + if dim is None: + recalc_metadata = True + if recalc_metadata: + global_size = tuple( + 0 if dim_size is None else dim_size for dim_size in global_size + ) + # STEP 1: Validate the Shardmetadatas locally + process_group = cls._normalize_pg(process_group) + current_rank = dist.get_rank() # intentional to get global rank + world_size = dist.get_world_size(process_group) + + local_sharded_tensor_metadata: ShardedTensorMetadata | None = None + global_tensor_size = _flatten_tensor_size(global_size) + + if len(local_shards) > 0: + local_sharded_tensor_metadata = build_metadata_from_local_shards( + local_shards, global_tensor_size, current_rank, process_group + ) + + # STEP 2. Validate metadata across ranks, and build a global sharded tensor + # metadata by gathering local ShardedTensorMetadata + gathered_metadatas: list[ShardedTensorMetadata | None] = [] + if world_size > 1: + gathered_metadatas = [None for _ in range(world_size)] + + dist.all_gather_object( + gathered_metadatas, local_sharded_tensor_metadata, group=process_group + ) + else: + gathered_metadatas = [local_sharded_tensor_metadata] + + global_sharded_tensor_metadata = build_global_metadata( + gathered_metadatas, recalc_metadata=recalc_metadata + ) + if recalc_metadata: + # for recalc use cases, we only support rw for now, limit the blast radius + # will modify here once we support more sharding type + if not ( + len(local_shards) > 0 + and len(global_sharded_tensor_metadata.shards_metadata) > current_rank + ): + raise AssertionError( + f"# for metadata recalculation, local_shards must be larger than 0 " + f"actual:{len(local_shards)}, # glb metadata must be greater than any rank id, " + f"# metadata:{len(global_sharded_tensor_metadata.shards_metadata)}, rank id:{current_rank}" + ) + local_md = [ + shard_md + for shard_md in global_sharded_tensor_metadata.shards_metadata + if shard_md.placement.rank() == current_rank + ] + if len(local_md) != 1: + raise AssertionError( + f"should has and only has one metadata for local rank, actual:{local_md}" + ) + local_shards[0].metadata = local_md[0] + tensor_properties = global_sharded_tensor_metadata.tensor_properties + + # STEP 3: Validation done, create the actual ShardedTensor and populate fields + # prepare initialization + spec = shard_spec._infer_sharding_spec_from_shards_metadata( + global_sharded_tensor_metadata.shards_metadata + ) + sharded_tensor = cls.__new__( + cls, + spec, + global_sharded_tensor_metadata.size, + dtype=tensor_properties.dtype, + layout=tensor_properties.layout, + pin_memory=tensor_properties.pin_memory, + requires_grad=tensor_properties.requires_grad, + ) + sharded_tensor._prepare_init(process_group=process_group, init_rrefs=init_rrefs) + + # attach local_shards to the ShardedTensor created + sharded_tensor._local_shards = local_shards + + # run post initialization, i.e. map registration, rpc initialization + sharded_tensor._post_init() + return sharded_tensor + + @classmethod + @deprecated(DEPRECATE_MSG, category=FutureWarning) + def _init_from_local_tensor( + cls, + local_tensor: torch.Tensor, + sharding_spec: shard_spec.ShardingSpec, + *global_size: Sequence[int], + process_group: dist.ProcessGroup | None = None, + init_rrefs=False, + ) -> ShardedTensor: + """ + Initialize a ShardedTensor given only one local tensor, global sharded tensor + size and sharding spec on each rank. + + Args: + local_tensor (Tensor): Single tensor of local shard stored in each rank. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): + The specification describing how to shard the Tensor. + global_size (Sequence[int]): Size of the sharded tensor. + process_group (ProcessGroup, optional): The process group to aggregate on. + Default: None + init_rrefs (bool, optional): Whether or not to initialize + :class:`torch.distributed.rpc.RRef`s pointing to remote shards. + Need to initialize the RPC Framework if specified as ``True``. + Default: ``False``. + + Returns: + A :class:`ShardedTensor` sharded based on the given sharding_spec with local + tensor stored in the current rank. + + Examples: + >>> # xdoctest: +SKIP + >>> # All tensors below are of torch.int64 type. + >>> # We have 2 process groups, 2 ranks. + >>> tensor = torch.arange(2, dtype=torch.int64) + 1 + 2 * rank + >>> local_tensor = torch.unsqueeze(torch.cat([tensor, tensor + 2])) + >>> local_tensor + tensor([[1, 2, 3, 4]]) # Rank 0 + tensor([[3, 4, 5, 6]]) # Rank 1 + >>> sharding_dim = 0 + >>> sharding_spec = ChunkShardingSpec( + dim=sharding_dim, + placements=[ + "rank:0/cuda:0", + "rank:1/cuda:1", + ], + ) + >>> st = ShardedTensor._init_from_local_tensor( + ... local_tensor, sharding_spec, [2, 4] + ... ) + >>> st + ShardedTensor( + ShardedTensorMetadata( + shards_metadata=[ + ShardMetadata(shard_offsets=[0, 0], shard_sizes=[1, 4], placement=rank:0/cuda:0), + ShardMetadata(shard_offsets=[1, 0], shard_sizes=[1, 4], placement=rank:1/cuda:1), + ], + size=torch.Size([2, 4]) + ) + >>> st.local_tensor() + tensor([1, 2, 3, 4]) # Rank 0 + tensor([3, 4, 5, 6]) # Rank 1 + + Warning: This API is experimental and subject to change. It lacks of a fully across + rank validations, and we only validate the local shard on the current rank. + We fully rely on the user to ensure local tensor is sharded based on the + sharding spec. + """ + if not local_tensor.is_contiguous(): + raise ValueError("local_tensor is not a contiguous Tensor.") + + global_tensor_size = _flatten_tensor_size(global_size) + tensor_properties = TensorProperties( + dtype=local_tensor.dtype, + layout=local_tensor.layout, + requires_grad=local_tensor.requires_grad, + memory_format=torch.contiguous_format, + pin_memory=local_tensor.is_pinned(), + ) + sharded_tensor_metadata = sharding_spec.build_metadata( + global_tensor_size, tensor_properties + ) + + process_group = cls._normalize_pg(process_group) + current_rank = dist.get_rank() # intentional to get global rank + + local_shards: list[Shard] = [] + for shard_metadata in sharded_tensor_metadata.shards_metadata: + rank, _device = _parse_and_validate_remote_device( + process_group, shard_metadata.placement + ) + if rank == current_rank: + local_shards.append(Shard(local_tensor, shard_metadata)) + + # TODO: figure out what the API should behave when some rank have no shard + # see https://github.com/pytorch/pytorch/issues/7313 + return ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards, + sharded_tensor_metadata, + process_group=process_group, + init_rrefs=init_rrefs, + sharding_spec=sharding_spec, + ) + + @classmethod + def _init_from_local_shards_and_global_metadata( # type: ignore[override] + cls, + local_shards: list[Shard], + sharded_tensor_metadata: ShardedTensorMetadata, + process_group=None, + init_rrefs=False, + sharding_spec=None, + ) -> ShardedTensor: + """ + Initialize a ShardedTensor with local shards and a global + ShardedTensorMetadata built on each rank. + + Warning: This API is experimental and subject to change. It does + not do cross rank validations, and fully rely on the user + for the correctness of sharded_tensor_metadata on each rank + """ + process_group = cls._normalize_pg(process_group) + current_rank = dist.get_rank() # intentional to get global rank + + shards_metadata = sharded_tensor_metadata.shards_metadata + + local_shard_metadatas = [] + + # collect local shard metadatas from the global sharded_tensor_metadata + for shard_metadata in shards_metadata: # type: ignore[attr-defined] + rank, local_device = _parse_and_validate_remote_device( + process_group, shard_metadata.placement + ) + + if current_rank == rank: + local_shard_metadatas.append(shard_metadata) + + if len(local_shards) != len(local_shard_metadatas): + raise RuntimeError( + f"Number of local shards ({len(local_shards)}) does not match number of local " + f"shards metadata in sharded_tensor_metadata ({len(local_shard_metadatas)}) " + f"on rank ({current_rank}) " + ) + + shards_metadata = sharded_tensor_metadata.shards_metadata + tensor_properties = sharded_tensor_metadata.tensor_properties + + if len(shards_metadata) == 0: + raise ValueError("shards_metadata must not be empty!") + + if tensor_properties.layout != torch.strided: + raise ValueError("Only torch.strided layout is currently supported") + + if sharding_spec is None: + spec = shard_spec._infer_sharding_spec_from_shards_metadata(shards_metadata) + else: + spec = sharding_spec + + sharded_tensor = ShardedTensor.__new__( + ShardedTensor, + spec, + sharded_tensor_metadata.size, + dtype=tensor_properties.dtype, + layout=tensor_properties.layout, + pin_memory=tensor_properties.pin_memory, + requires_grad=tensor_properties.requires_grad, + ) + + def _raise_if_mismatch(expected, actual, prop_name, rank, is_property=False): + tensor_property_or_metadata = ( + "tensor property" if is_property else "local ShardMetadata" + ) + if expected != actual: + raise ValueError( + f"Local shards' tensor {prop_name} property is incompatible with " + f"{tensor_property_or_metadata} on rank {rank}: " + f"{tensor_property_or_metadata} {prop_name}={expected}, " + f"local shard tensor {prop_name}={actual}." + ) + + for shard in local_shards: + shard_meta = shard.metadata + local_shard_tensor = shard.tensor + placement = shard_meta.placement + if placement is None: + raise AssertionError("Must specify placement for `Shard`!") + rank = placement.rank() + local_device = placement.device() + + _raise_if_mismatch( + tensor_properties.layout, + local_shard_tensor.layout, + "layout", + rank, + True, + ) + if not local_shard_tensor.is_contiguous(): + raise ValueError( + "Only torch.contiguous_format memory_format is currently supported" + ) + + _raise_if_mismatch( + shard_meta.shard_sizes, + list(local_shard_tensor.size()), + "size", + rank, + ) + _raise_if_mismatch( + tensor_properties.pin_memory, + local_shard_tensor.is_pinned(), + "pin_memory", + rank, + True, + ) + _raise_if_mismatch(local_device, local_shard_tensor.device, "device", rank) + _raise_if_mismatch( + tensor_properties.dtype, + local_shard_tensor.dtype, + "dtype", + rank, + True, + ) + _raise_if_mismatch( + tensor_properties.requires_grad, + local_shard_tensor.requires_grad, + "requires_grad", + rank, + True, + ) + + # check if shards_metadata have overlap shards + validate_non_overlapping_shards_metadata(shards_metadata) + + # check if the shards_metadata is compatible with overall size of the sharded tensor. + check_tensor(shards_metadata, list(sharded_tensor_metadata.size)) + + # done validation, add local_shards + sharded_tensor._local_shards = local_shards + sharded_tensor._prepare_init(process_group=process_group, init_rrefs=init_rrefs) + + # run post initialization, i.e. map registration, rpc initialization + sharded_tensor._post_init() + return sharded_tensor + + def sharding_spec(self) -> shard_spec.ShardingSpec: + """ + Returns the ShardingSpec for the tensor. + """ + return self._sharding_spec + + @deprecated(DEPRECATE_MSG, category=FutureWarning) + def reshard(self, resharding_spec: shard_spec.ShardingSpec) -> ShardedTensor: + """ + Reshard a sharded tensor given the ``resharding_spec``. For now, we only support + single local shard. + + If ``resharding_spec`` is same as the original one, this becomes a no-op. + If only ``resharding_spec`` shares the same sharding dim with the original one, + we swap local shards directly. + For more generic cases, we merge different shards across different ranks and split + the local shards based on the ``resharding_spec`` via `all_to_all` collective API. + + Args: + resharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor is sharded. + + Returns: + A :class:`ShardedTensor` object whose local shards are resharded. + + Examples: + >>> # xdoctest: +SKIP + >>> # We have 2 process groups, 2 ranks. + >>> tensor = torch.arange(4, dtype=torch.int64) + 1 + 2 * rank + >>> tensor = torch.stack([tensor, tensor]) + >>> tensor + tensor([[1, 2, 3, 4], [1, 2, 3, 4]]) # Rank 0 + tensor([[3, 4, 5, 6], [3, 4, 5, 6]]) # Rank 1 + tensor([[5, 6, 7, 8], [5, 6, 7, 8]]) # Rank 2 + tensor([[7, 8, 9, 10], [7, 8, 9, 10]]) # Rank 3 + >>> sharding_dim = 0 + >>> spec = ChunkShardingSpec( + dim=sharding_dim, + placements=[ + "rank:0/cuda:0", + "rank:1/cuda:1", + "rank:2/cuda:2", + "rank:3/cuda:3", + ], + ) + >>> current_offsets = [0] * 2 + >>> current_offsets[0] = rank * 2 + >>> shard_metadata = ShardMetadata( + shard_offsets=copy.deepcopy(current_offsets), + shard_sizes=tensor.size(), + placement=spec.placements[rank], + ) + >>> local_shards = [ + Shard( + tensor=tensor, + metadata=shard_metadata, + ) + ] + >>> st = ShardedTensor._init_from_local_shards(local_shards, tensor.size()) + >>> sharding_dim = 1 + >>> resharding_spec = ChunkShardingSpec( + dim=sharding_dim, + placements=[ + "rank:0/cuda:0", + "rank:1/cuda:1", + "rank:2/cuda:2", + "rank:3/cuda:3", + ], + ) + >>> st.reshard(resharding_spec) + >>> tensor = st.local_shards()[0].tensor + >>> tensor + tensor([[1], [1], [3], [3], [5], [5], [7], [7]]) # Rank 0 + tensor([[2], [2], [4], [4], [6], [6], [8], [8]]) # Rank 1 + tensor([[3], [3], [5], [5], [7], [7], [9], [9]]) # Rank 2 + tensor([[4], [4], [6], [6], [8], [8], [10], [10]]) # Rank 3 + """ + if not isinstance( + resharding_spec, shard_spec.ChunkShardingSpec + ) or not isinstance(self._sharding_spec, shard_spec.ChunkShardingSpec): + raise NotImplementedError("Only ChunkShardingSpec supported for reshard.") + + num_local_shards = len(self.local_shards()) + if num_local_shards != 1: + raise NotImplementedError( + f"Only single local shard supported for reshard. Number of shards: {num_local_shards}" + ) + + if self._sharding_spec.dim == resharding_spec.dim: # type: ignore[attr-defined] + if self._sharding_spec.placements == resharding_spec.placements: # type: ignore[attr-defined] + return self + else: + local_shards, shards_metadata = reshuffle_local_shard( + self.local_tensor(), + self.size(), # type: ignore[arg-type] + self._sharding_spec, + resharding_spec, + self._process_group, + ) + else: + local_shards, shards_metadata = reshard_local_shard( + self.local_tensor(), + self.size(), # type: ignore[arg-type] + self._sharding_spec, + resharding_spec, + self._process_group, + ) + self._local_shards = local_shards + self._metadata.shards_metadata = shards_metadata + self._sharding_spec = resharding_spec + return self + + def local_tensor(self) -> torch.Tensor: + """ + Return local tensor for a sharded_tensor. For now we only support single local shard. + + Returns: + A :class:`torch.Tensor` of the local shard. + """ + num_local_shards = len(self.local_shards()) + if num_local_shards != 1: + raise NotImplementedError( + f"Only single local shard is supported. Number of shards: {num_local_shards}" + ) + return self.local_shards()[0].tensor + + @classmethod + @deprecated(DEPRECATE_MSG, category=FutureWarning) + def __torch_function__(cls, func, types, args=(), kwargs=None): + def dispatch(st: ShardedTensor, func: Callable): + # Dispatch to custom user provided op first if it exists. + if func in _CUSTOM_SHARDED_OPS: + return _CUSTOM_SHARDED_OPS[func](types, args, kwargs, st._process_group) + + # Dispatch to custom sharding spec op if it has one. + if _has_custom_op(st._sharding_spec, func): + return _dispatch_custom_op( + st._sharding_spec, func, types, args, kwargs, st._process_group + ) + + if func in _SHARDED_OPS: + return _SHARDED_OPS[func](types, args, kwargs, st._process_group) + + raise RuntimeError( + f"torch function '{func.__name__}', with args: {args} and " + f"kwargs: {kwargs} not supported for ShardedTensor!" + ) + + # Find ShardedTensor instance to get process_group and sharding_spec. + st_instance = None + + def find_sharded_tensor(e): + nonlocal st_instance + if st_instance is None and isinstance(e, ShardedTensor): + st_instance = e + + pytree.tree_map_(find_sharded_tensor, args) + pytree.tree_map_(find_sharded_tensor, kwargs) + + if st_instance is not None: + return dispatch(st_instance, func) + + raise RuntimeError( + f"torch function '{func.__name__}', with args: {args} and " + f"kwargs: {kwargs} not supported for ShardedTensor!" + ) + + def is_pinned(self) -> bool: # type: ignore[override] + """ + Returns True if the sharded tensor (each local shard) resides in pinned memory. + """ + return self._metadata.tensor_properties.pin_memory + + def _register_remote_shards( + self, remote_shards: list[rpc.RRef[Shard]], rpc_rank: int + ): + self._remote_shards[rpc_rank] = remote_shards + + def remote_shards(self) -> dict[int, list[rpc.RRef[Shard]]]: + """ + Returns a Dict[int, RRef] with keys being the RPC rank and values + being RRefs to shards on that rank. Need to initialize the + RPC framework for this functionality. + + Raises an exception if ShardedTensor was created with ``init_rrefs=False`` + """ + if not self._init_rrefs: + raise RuntimeError( + "ShardedTensor created with init_rrefs=False, no RRefs to remote shards available" + ) + return self._remote_shards + + def __hash__(self): + return id(self) + + def __repr__(self) -> str: # type: ignore[override] + return f"ShardedTensor({self._metadata})" + + @dataclass + class ProcessGroupState: + """ + State for ser-de of process group + """ + + local_rank: int + global_rank: int + local_world_size: int + global_world_size: int + + def __getstate__(self): + pg_state = ShardedTensor.ProcessGroupState( + distributed_c10d.get_rank(self._process_group), + distributed_c10d.get_rank(), + distributed_c10d.get_world_size(self._process_group), + distributed_c10d.get_world_size(), + ) + + return ( + self._local_shards, + self._metadata, + pg_state, + self._sharding_spec, + self._init_rrefs, + ) + + def __setstate__(self, state): + self._sharded_tensor_id = None + if not distributed_c10d.is_initialized(): + raise RuntimeError( + "Need to initialize default process group using " + '"init_process_group" before loading ShardedTensor' + ) + + ( + self._local_shards, + self._metadata, + pg_state, + self._sharding_spec, + self._init_rrefs, + ) = state + + # Setup process group + from torch.distributed._shard.api import _get_current_process_group + + self._process_group = _get_current_process_group() + + # Validate process group. + local_rank = distributed_c10d.get_rank(self._process_group) + if pg_state.local_rank != local_rank: + raise RuntimeError( + f"Local rank at save time was {pg_state.local_rank}, but at " + f"load time was {local_rank}" + ) + + global_rank = distributed_c10d.get_rank() + if pg_state.global_rank != global_rank: + raise RuntimeError( + f"Global rank at save time was {pg_state.global_rank}, but at " + f"load time was {global_rank}" + ) + + local_world_size = distributed_c10d.get_world_size(self._process_group) + if pg_state.local_world_size != local_world_size: + raise RuntimeError( + f"Local world size at save time was {pg_state.local_world_size}, " + f"but at load time was {local_world_size}" + ) + + global_world_size = distributed_c10d.get_world_size() + if pg_state.global_world_size != global_world_size: + raise RuntimeError( + f"Global world size at save time was {pg_state.global_world_size}, " + f"but at load time was {global_world_size}" + ) + + self._post_init() + + +def _create_tensor_from_params( + *size, local_device, tensor_properties: TensorProperties +): + """Helper to construct tensor from size, device and common params.""" + dtype = tensor_properties.dtype + layout = tensor_properties.layout + requires_grad = tensor_properties.requires_grad + memory_format = tensor_properties.memory_format + pin_memory = tensor_properties.pin_memory + + return torch.empty( + *size, + dtype=dtype, + layout=layout, + device=local_device, + requires_grad=requires_grad, + memory_format=memory_format, + pin_memory=pin_memory, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logger.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logger.py new file mode 100644 index 0000000000000000000000000000000000000000..ff8cb4d18fb180ea620dd8daad60b5771a9688be --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logger.py @@ -0,0 +1,35 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging + +from torch.distributed._shard.sharded_tensor.logging_handlers import _log_handlers + + +__all__: list[str] = [] + + +def _get_or_create_logger() -> logging.Logger: + logging_handler, log_handler_name = _get_logging_handler() + logger = logging.getLogger(f"sharding-spec-{log_handler_name}") + logger.setLevel(logging.DEBUG) + formatter = logging.Formatter( + "%(asctime)s %(filename)s:%(lineno)s %(levelname)s p:%(processName)s t:%(threadName)s: %(message)s" + ) + logging_handler.setFormatter(formatter) + logger.propagate = False + logger.addHandler(logging_handler) + return logger + + +def _get_logging_handler( + destination: str = "default", +) -> tuple[logging.Handler, str]: + log_handler = _log_handlers[destination] + log_handler_name = type(log_handler).__name__ + return (log_handler, log_handler_name) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logging_handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logging_handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..ed6832fd1ae834b6365a6b005b07bbbfffe90726 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/logging_handlers.py @@ -0,0 +1,16 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging + + +__all__: list[str] = [] + +_log_handlers: dict[str, logging.Handler] = { + "default": logging.NullHandler(), +} diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/metadata.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/metadata.py new file mode 100644 index 0000000000000000000000000000000000000000..466ca1a0c519ce4cc4ee24fae98ff4ddfbee300a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/metadata.py @@ -0,0 +1,94 @@ +# mypy: allow-untyped-defs +from dataclasses import dataclass, field +from enum import Enum + +import torch +from torch.distributed._shard.metadata import ShardMetadata + + +class MEM_FORMAT_ENCODING(Enum): + TORCH_CONTIGUOUS_FORMAT = 0 + TORCH_CHANNELS_LAST = 1 + TORCH_PRESERVE_FORMAT = 2 + + +@dataclass +class TensorProperties: + """Properties used to create :class:`Tensor`""" + + # Regular tensor fields + dtype: torch.dtype = field(default=torch.get_default_dtype()) + layout: torch.layout = field(default=torch.strided) + requires_grad: bool = False + memory_format: torch.memory_format = field(default=torch.contiguous_format) + pin_memory: bool = False + + def __getstate__(self): + # Since torch.memory_format cannot be pickled! + memory_format = self.memory_format + if memory_format == torch.contiguous_format: + mem_format_encoding = MEM_FORMAT_ENCODING.TORCH_CONTIGUOUS_FORMAT + elif memory_format == torch.channels_last: + mem_format_encoding = MEM_FORMAT_ENCODING.TORCH_CHANNELS_LAST + elif memory_format == torch.preserve_format: + mem_format_encoding = MEM_FORMAT_ENCODING.TORCH_PRESERVE_FORMAT + else: + raise RuntimeError(f"Invalid torch.memory_format: {memory_format}") + + return ( + self.dtype, + self.layout, + self.requires_grad, + mem_format_encoding, + self.pin_memory, + ) + + def __setstate__( + self, + state, + ): + ( + self.dtype, + self.layout, + self.requires_grad, + mem_format_encoding, + self.pin_memory, + ) = state + + if mem_format_encoding == MEM_FORMAT_ENCODING.TORCH_CONTIGUOUS_FORMAT: + memory_format = torch.contiguous_format + elif mem_format_encoding == MEM_FORMAT_ENCODING.TORCH_CHANNELS_LAST: + memory_format = torch.channels_last + elif mem_format_encoding == MEM_FORMAT_ENCODING.TORCH_PRESERVE_FORMAT: + memory_format = torch.preserve_format + else: + raise RuntimeError( + f"Invalid torch.memory_format encoding: {mem_format_encoding}" + ) + + self.memory_format = memory_format + + @staticmethod + def create_from_tensor(tensor: torch.Tensor) -> "TensorProperties": + return TensorProperties( + dtype=tensor.dtype, + layout=tensor.layout, + requires_grad=tensor.requires_grad, + memory_format=torch.contiguous_format, + pin_memory=tensor.is_pinned(), + ) + + +@dataclass +class ShardedTensorMetadata: + """ + Represents metadata for :class:`ShardedTensor` + """ + + # Metadata about each shard of the Tensor + shards_metadata: list[ShardMetadata] = field(default_factory=list) + + # Size of each dim of the overall Tensor. + size: torch.Size = field(default=torch.Size([])) + + tensor_properties: TensorProperties = field(default_factory=TensorProperties) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/reshard.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/reshard.py new file mode 100644 index 0000000000000000000000000000000000000000..fa3dace4b8e62145b4d949765642e2a5b1b425ac --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/reshard.py @@ -0,0 +1,243 @@ +# mypy: allow-untyped-defs +import copy + +import torch +import torch.distributed as dist +import torch.distributed._shard.sharding_spec as shard_spec +from torch._C._distributed_c10d import ProcessGroup +from torch.distributed._shard.metadata import ShardMetadata +from torch.distributed._shard.sharding_spec._internals import ( + get_chunked_dim_size, + get_split_size, +) +from torch.distributed.nn.functional import all_to_all, all_to_all_single + +from .shard import Shard + + +def get_idx_from_placements(placements, current_rank) -> int: + """ + Return the position of the current rank in the given placements. + + Args: + placements(List[Union[_remote_device, str]]): + Specifies the placement of each shard of the Tensor. The size of + the list represents the number of shards to be created. This could + be a list of + :class:`torch.distributed._remote_device`'s. This list + could also contain a string which represents remote + device as accepted by + :class:`torch.distributed._remote_device` + current_rank (int): number of current device. + + Returns: + An int which contains the position of current device in the placement list. + """ + for idx, placement in enumerate(placements): # type: ignore[attr-defined] + if current_rank == placement.rank(): # type: ignore[union-attr] + return idx + raise RuntimeError("current_rank not in the placement.") + + +def build_reshard_metadata( + st_size: torch.Size, + sharding_spec: shard_spec.ShardingSpec, + world_size: int, +) -> tuple[list[ShardMetadata], list[int]]: + """ + Based the given sharding spec, we calculate the offset and local shard size. + We then build a ShardMetadata on top of the calculation result. + + Args: + st_size (torch.Size): The size of the sharded tensor. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor is sharded. + world_size (int): number of ranks. + + Returns: + A Tuple of the following: + A List[`ShardMetadata`] which contains the metadata for the shard, including + offsets, lengths and device placement. + A List[int] which contains the ranks in the order of placement. + """ + shard_dim = int(sharding_spec.dim) # type: ignore[attr-defined] + shards_metadata = [None] * world_size + ranks = [] + offsets = [0] * len(st_size) + split_size = get_split_size(st_size[shard_dim], world_size) + for idx, placement in enumerate(sharding_spec.placements): # type: ignore[attr-defined] + ranks.append(placement.rank()) + sharded_dim_size = get_chunked_dim_size(st_size[shard_dim], split_size, idx) + local_tensor_size = list(st_size) + local_tensor_size[shard_dim] = sharded_dim_size + shards_metadata[placement.rank()] = ShardMetadata( # type: ignore[call-overload] + shard_offsets=copy.deepcopy(offsets), + shard_sizes=local_tensor_size, + placement=placement, + ) + offsets[shard_dim] += sharded_dim_size + return shards_metadata, ranks # type: ignore[return-value] + + +def reshuffle_local_shard( + local_shard: torch.Tensor, + st_size: torch.Size, + sharding_spec: shard_spec.ShardingSpec, + resharding_spec: shard_spec.ShardingSpec, + pg: ProcessGroup, +) -> tuple[list[Shard], list[ShardMetadata]]: + """ + Reshuffle the local shard directly when the reshard dim is same as the original + sharding dim. Logically we do this in two step: + 1. To collect all shards based on original sharding spec. + 2. Reshard the tensor based on the given resharding spec. + + In reality, we consolidate the two steps into one by sending the local tensor to + the new shard directly based on the resharding spec. + + Args: + local_shard (Tensor): Local tensor stored in the current rank. + st_size (torch.Size): The size of the sharded tensor. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor is sharded originally. + resharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor will be resharded. + pg (ProcessGroup): The process group to aggregate on. + + Returns: + A Tuple of the following: + A List[`Shard`] which contains the local tensor and its metadata. + A List[`ShardMetadata`] which contains the metadata for the shard, including + offsets, lengths and device placement. + """ + current_rank = dist.get_rank(pg) + world_size = dist.get_world_size(pg) + # Build shards_metadata first. + shards_metadata, ranks = build_reshard_metadata( + st_size, resharding_spec, world_size + ) + # Get input split size for all2all. + reshard_dim = int(resharding_spec.dim) # type: ignore[attr-defined] + split_size = get_split_size(st_size[reshard_dim], world_size) + input_split_sizes = [0] * world_size + idx = get_idx_from_placements(sharding_spec.placements, current_rank) # type: ignore[attr-defined] + new_rank = resharding_spec.placements[idx].rank() # type: ignore[union-attr, attr-defined] + input_split_sizes[new_rank] = local_shard.size(reshard_dim) + # Get output split size for all2all. + output_split_sizes = [0] * world_size + new_idx = ranks.index(current_rank) + sharded_dim_size = get_chunked_dim_size(st_size[reshard_dim], split_size, new_idx) + output_split_sizes[new_rank] = sharded_dim_size + # Get gathered_input for all2all. + local_shard = local_shard.transpose(0, reshard_dim).contiguous() + gathered_input_size = list(local_shard.size()) + gathered_input_size[0] = sharded_dim_size + gathered_input = torch.empty( + gathered_input_size, device=local_shard.device, dtype=local_shard.dtype + ) + # all2all. + local_shard = all_to_all_single( + gathered_input, + local_shard, + input_split_sizes=input_split_sizes, + output_split_sizes=output_split_sizes, + group=pg, + ) + local_tensor = local_shard.transpose(0, reshard_dim).contiguous() + local_shards = [Shard(local_tensor, shards_metadata[current_rank])] + return local_shards, shards_metadata + + +def reshard_local_shard( + local_tensor: torch.Tensor, + st_size: torch.Size, + sharding_spec: shard_spec.ShardingSpec, + resharding_spec: shard_spec.ShardingSpec, + pg: ProcessGroup, +) -> tuple[list[Shard], list[ShardMetadata]]: + """ + Reshard a sharded tensor given the ``resharding_spec``. When the reshard dim is + different from the original sharding dim, we need to do two steps logically: + 1. To collect all shards based on original sharding spec. + 2. Reshard the tensor based on the given resharding spec. + + In reality, we consolidate the two steps into one by sending each rank the new + shard based on the resharding spec. + + Args: + local_tensor (Tensor): Local tensor stored in the current rank. + st_size (torch.Size): The size of the sharded tensor. + sharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor is sharded originally. + resharding_spec (:class:`torch.distributed._shard.sharding_spec.ShardingSpec`): The + specification describing how the tensor will be resharded. + pg (ProcessGroup): The process group to aggregate on. + + Returns: + A Tuple of the following: + A List[`Shard`] which contains the local tensor and its metadata. + A List[`ShardMetadata`] which contains the metadata for the shard, including + offsets, lengths and device placement. + """ + current_rank = dist.get_rank(pg) + world_size = dist.get_world_size(pg) + current_sharding_dim = int(sharding_spec.dim) # type: ignore[attr-defined] + reshard_dim = int(resharding_spec.dim) # type: ignore[attr-defined] + + # Build shards_metadata first. + shards_metadata, ranks = build_reshard_metadata( + st_size, resharding_spec, world_size + ) + + # Compute expected size + input_split_sizes = [ + metadata.shard_sizes[reshard_dim] for metadata in shards_metadata + ] + rearrange_input = any(ranks[i] > ranks[i + 1] for i in range(len(ranks) - 1)) + + if rearrange_input: + # Need to re-arrange reshard_dim of local_tensor before all2all. + indices: list[int] = [] + for metadata in shards_metadata: + offset_start_idx = metadata.shard_offsets[reshard_dim] + split_size = metadata.shard_sizes[reshard_dim] + indices += range(offset_start_idx, offset_start_idx + split_size) + local_tensor = local_tensor.index_select( + reshard_dim, torch.tensor(indices, device=local_tensor.device) + ) + + # Because reshard_dim != original shard_dim. We need to compute the + # size of tensor from each rank. + output_tensor_list = [torch.tensor(1)] * world_size + split_size = get_split_size(st_size[current_sharding_dim], world_size) + rearrange_output_list = False + indices = [] + for idx, placement in enumerate(sharding_spec.placements): # type: ignore[attr-defined] + sharded_dim_size = get_chunked_dim_size( + st_size[current_sharding_dim], split_size, idx + ) + output_tensor_size = list(st_size) + output_tensor_size[current_sharding_dim] = sharded_dim_size + output_tensor_size[reshard_dim] = input_split_sizes[current_rank] + output_tensor_list[placement.rank()] = torch.empty( # type: ignore[union-attr, index] + output_tensor_size, device=local_tensor.device, dtype=local_tensor.dtype + ) + indices.append(placement.rank()) # type: ignore[union-attr, index, arg-type] + if idx != placement.rank(): # type: ignore[union-attr] + rearrange_output_list = True + + # Perform autograd enabled all2all. + input_tensor_tuple = torch.split(local_tensor, input_split_sizes, dim=reshard_dim) + input_tensor_list = [tensor.contiguous() for tensor in input_tensor_tuple] + output_tensor_list = all_to_all( + output_tensor_list, + input_tensor_list, + group=pg, + ) + + if rearrange_output_list: + # Need to re-arrange original shard_dim of output_tensor_list. + output_tensor_list = [output_tensor_list[idx] for idx in indices] # type: ignore[call-overload] + local_tensor = torch.cat(output_tensor_list, dim=current_sharding_dim) + local_shards = [Shard(local_tensor, shards_metadata[current_rank])] + return local_shards, shards_metadata diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/shard.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/shard.py new file mode 100644 index 0000000000000000000000000000000000000000..2d9d4357436a6c15f590a4db486d9d54b6d6ca57 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/shard.py @@ -0,0 +1,61 @@ +from dataclasses import dataclass + +import torch +from torch.distributed._shard.metadata import ShardMetadata +from torch.distributed.remote_device import _remote_device + + +@dataclass +class Shard: + """ + Container which holds the data for a shard as a Tensor and also + the associated metadata for that shard. + + Args: + tensor(torch.Tensor): Local tensor for the shard. + metadata(:class `torch.distributed._shard.sharded_tensor.ShardMetadata`): + The metadata for the shard, including offsets, lengths and device placement. + """ + + __slots__ = ["tensor", "metadata"] + tensor: torch.Tensor + metadata: ShardMetadata + + def __post_init__(self) -> None: + # verification between local tensor and metadata + if list(self.tensor.size()) != self.metadata.shard_sizes: + raise ValueError( + "Shard tensor size does not match with metadata.shard_lengths! " + f"Found shard tensor size: {list(self.tensor.size())}, " + f"metadata.shard_lengths: {self.metadata.shard_sizes}, " + ) + placement_device = self.metadata.placement + if ( + placement_device is not None + and placement_device.device() != self.tensor.device + ): + raise ValueError( + f"Local shard tensor device does not match with local Shard's placement! " + f"Found local shard tensor device: {self.tensor.device}, " + f"local shard metadata placement device: {placement_device.device()}" + ) + + @classmethod + def from_tensor_and_offsets( + cls, tensor: torch.Tensor, shard_offsets: list[int], rank: int + ) -> "Shard": + """ + Creates a Shard of a ShardedTensor from a local torch.Tensor, shard_offsets and rank. + + Args: + tensor(torch.Tensor): Local tensor for the shard. + shard_offsets(List[int]): List of integers specify the offset + of the shard on each dimension. + rank(int): Specify the rank for the shard. + """ + shard_sizes = list(tensor.size()) + placement = _remote_device(f"rank:{rank}/{str(tensor.device)}") + shard_meta = ShardMetadata( + shard_offsets=shard_offsets, shard_sizes=shard_sizes, placement=placement + ) + return Shard(tensor, shard_meta) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..a2a6c3aca95d4cea9d7ff7a8edd56570c2aef266 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharded_tensor/utils.py @@ -0,0 +1,327 @@ +# mypy: allow-untyped-defs +import collections.abc +import copy +import itertools +from collections.abc import Sequence +from typing import TYPE_CHECKING + +import torch +from torch.distributed import distributed_c10d as c10d, rpc +from torch.distributed._shard.sharding_spec._internals import ( + check_tensor, + validate_non_overlapping_shards_metadata, +) + +from .metadata import ShardedTensorMetadata, TensorProperties +from .shard import Shard + + +if TYPE_CHECKING: + from torch.distributed._shard.metadata import ShardMetadata + + +def _parse_and_validate_remote_device(pg, remote_device): + if remote_device is None: + raise ValueError("remote device is None") + + worker_name = remote_device.worker_name() + rank = remote_device.rank() + device = remote_device.device() + + # Validate rank, skip validation if rank is not part of process group. + if rank is not None and not c10d._rank_not_in_group(pg): + pg_global_ranks = c10d.get_process_group_ranks(pg) + if rank not in pg_global_ranks: + raise ValueError( + f"Global rank {rank} does not exist in input process group: {pg_global_ranks}" + ) + + if worker_name is not None: + if not rpc._is_current_rpc_agent_set(): + raise RuntimeError( + f"RPC framework needs to be initialized for using worker names: {worker_name}" + ) + + workers = rpc._get_current_rpc_agent().get_worker_infos() + for worker in workers: + if worker.name == worker_name: + return worker.id, device + + raise ValueError(f"Invalid worker name: {worker_name}") + + return rank, device + + +def _validate_output_tensor_for_gather( + my_rank: int, + dst_rank: int, + size: torch.Size, + dst_tensor: torch.Tensor | None, +) -> None: + if dst_rank == my_rank: + if dst_tensor is None: + raise ValueError( + f"Argument ``dst_tensor`` must be specified on destination rank {dst_rank}" + ) + if tuple(size) != (dst_tensor.size()): + raise ValueError( + f"Argument ``dst_tensor`` have size {tuple(dst_tensor.size())}," + f"but should be {tuple(size)}" + ) + elif dst_tensor: + raise ValueError( + "Argument ``dst_tensor`` must NOT be specified on non-destination ranks." + ) + + +def _flatten_tensor_size(size) -> torch.Size: + """ + Checks if tensor size is valid, then flatten/return a torch.Size object. + """ + if len(size) == 1 and isinstance(size[0], collections.abc.Sequence): + dims = list(*size) + else: + dims = list(size) + + for dim in dims: + if not isinstance(dim, int): + raise TypeError(f"size has to be a sequence of ints, found: {dims}") + + return torch.Size(dims) + + +def _raise_if_mismatch(expected, actual, prop_name, ranks, is_local=True): + if is_local: + if not isinstance(ranks, int): + raise AssertionError + if expected != actual: + raise ValueError( + f"Local shards' tensor {prop_name} property need to be the same on rank:{ranks}! " + f"Found one local shard tensor {prop_name}={expected}, " + f"the other local shard tensor {prop_name}={actual}." + ) + else: + # compare failure check across ranks, ranks list should have two rank + if len(ranks) != 2: + raise AssertionError + if expected != actual: + raise ValueError( + f"ShardedTensor {prop_name} property does not match from different ranks! " + f"Found {prop_name}={expected} on rank:{ranks[0]}, " + f"and {prop_name}={actual} on rank:{ranks[1]}." + ) + + +def build_metadata_from_local_shards( + local_shards: list[Shard], + global_size: torch.Size, + current_rank: int, + pg: c10d.ProcessGroup, +) -> ShardedTensorMetadata: + if len(local_shards) <= 0: + raise AssertionError("must have local shards!") + local_shard_metadatas: list[ShardMetadata] = [] + + first_shard_dtype = local_shards[0].tensor.dtype + first_shard_layout = local_shards[0].tensor.layout + first_shard_requires_grad = local_shards[0].tensor.requires_grad + first_shard_is_pinned = local_shards[0].tensor.is_pinned() + + # 1). Validate local tensors and associated metadata + for local_shard in local_shards: + local_shard_tensor = local_shard.tensor + local_shard_meta = local_shard.metadata + local_shard_metadatas.append(local_shard_meta) + rank, local_device = _parse_and_validate_remote_device( + pg, local_shard_meta.placement + ) + + if ( + local_shard_tensor.layout != torch.strided + or local_shard_tensor.layout != first_shard_layout + ): + raise ValueError( + f"Only torch.strided layout is currently supported, but found " + f"{local_shard_tensor.layout} on rank:{current_rank}!" + ) + + if not local_shard_tensor.is_contiguous(): + raise ValueError( + "Only torch.contiguous_format memory_format is currently supported!" + ) + + if rank != current_rank: + raise ValueError( + f"Local shard metadata's rank does not match with the rank in its process group! " + f"Found current rank in the process group: {current_rank}, " + f"local ShardMetadata placement's rank: {rank}" + ) + if local_shard_tensor.device != local_device: + raise ValueError( + f"Local shard tensor device does not match with local Shard's placement! " + f"Found local shard tensor device: {local_shard_tensor.device}, " + f"local shard metadata placement device: {local_device}" + ) + + _raise_if_mismatch( + local_shard_meta.shard_sizes, + list(local_shard_tensor.size()), + "size", + current_rank, + ) + _raise_if_mismatch( + local_shard_tensor.is_pinned(), + first_shard_is_pinned, + "pin_memory", + current_rank, + ) + _raise_if_mismatch( + local_shard_tensor.dtype, first_shard_dtype, "dtype", current_rank + ) + _raise_if_mismatch( + local_shard_tensor.requires_grad, + first_shard_requires_grad, + "requires_grad", + current_rank, + ) + + # 2). Build a "local" ShardedTensorMetadata with all local shards on this rank, then + # do all_gather to collect local_sharded_tensor_metadata from all ranks + local_tensor_properties = TensorProperties( + dtype=first_shard_dtype, + layout=first_shard_layout, + requires_grad=first_shard_requires_grad, + memory_format=torch.contiguous_format, + pin_memory=first_shard_is_pinned, + ) + + local_sharded_tensor_metadata = ShardedTensorMetadata( + shards_metadata=local_shard_metadatas, + size=global_size, + tensor_properties=local_tensor_properties, + ) + + return local_sharded_tensor_metadata + + +def build_global_metadata( + gathered_metadatas: Sequence[ShardedTensorMetadata | None], + recalc_metadata: bool = False, +): + global_sharded_tensor_metadata = None + global_metadata_rank = 0 + + # pyrefly: ignore [bad-assignment] + for rank, rank_metadata in enumerate(gathered_metadatas): + if rank_metadata is None: + continue + + if global_sharded_tensor_metadata is None: + global_sharded_tensor_metadata = copy.deepcopy(rank_metadata) + global_metadata_rank = rank + else: + _raise_if_mismatch( + global_sharded_tensor_metadata.size, + rank_metadata.size, + "global_size", + [global_metadata_rank, rank], + is_local=False, + ) + + # don't need to check layout and memory format as we already checked in local shards validation stage + _raise_if_mismatch( + global_sharded_tensor_metadata.tensor_properties.dtype, + rank_metadata.tensor_properties.dtype, + "dtype", + [global_metadata_rank, rank], + is_local=False, + ) + + _raise_if_mismatch( + global_sharded_tensor_metadata.tensor_properties.requires_grad, + rank_metadata.tensor_properties.requires_grad, + "requires_grad", + [global_metadata_rank, rank], + is_local=False, + ) + + _raise_if_mismatch( + global_sharded_tensor_metadata.tensor_properties.pin_memory, + rank_metadata.tensor_properties.pin_memory, + "pin_memory", + [global_metadata_rank, rank], + is_local=False, + ) + # pass all validations, extend shards metadata + global_sharded_tensor_metadata.shards_metadata.extend( + rank_metadata.shards_metadata + ) + + if global_sharded_tensor_metadata is not None: + if recalc_metadata: + recalc_global_sharded_tensor_metadata( + global_sharded_tensor_metadata, + 0, # sharded on 0th dim + ) + + # check if shards_metadata have overlap shards + validate_non_overlapping_shards_metadata( + global_sharded_tensor_metadata.shards_metadata + ) + + # check if the shards_metadata is compatible with global size of the sharded tensor. + check_tensor( + global_sharded_tensor_metadata.shards_metadata, + global_sharded_tensor_metadata.size, + ) + else: + raise ValueError("ShardedTensor have no local shards on all ranks!") + + return global_sharded_tensor_metadata + + +def recalc_global_sharded_tensor_metadata( + global_sharded_tensor_metadata: ShardedTensorMetadata, sharded_dim: int +) -> None: + # recalculate global ShardedTensorMetadata + + # reorder here in case shard metadata is not sorted on sharded_dim + placement_idx_pairs = [] + for i, shard_metadata in enumerate(global_sharded_tensor_metadata.shards_metadata): + if shard_metadata.placement: + placement_idx_pairs.append((shard_metadata.placement.rank(), i)) + else: + raise AssertionError( + "currently only support rw, it should always have valid rank info" + ) + sorted_idx = sorted(placement_idx_pairs) + shard_sizes = [ + global_sharded_tensor_metadata.shards_metadata[idx].shard_sizes[sharded_dim] + for _, idx in sorted_idx + ] + cum_sum = [0] + list(itertools.accumulate(shard_sizes)) + + for shard_id, shard_metadata in enumerate( + global_sharded_tensor_metadata.shards_metadata + ): + # update shard offset for each shard on the sharded dimension + shard_metadata.shard_offsets[sharded_dim] = cum_sum[shard_id] + for other_dim in range( + len(global_sharded_tensor_metadata.shards_metadata[0].shard_sizes) + ): + if other_dim != sharded_dim: + # shard offset for each shard on the unsharded dimension + shard_metadata.shard_offsets[other_dim] = 0 + + # update global size for ShardedTensorMetadata + global_size_list = [] + for other_dim in range( + len(global_sharded_tensor_metadata.shards_metadata[0].shard_sizes) + ): + if other_dim != sharded_dim: + global_size_list.append( + global_sharded_tensor_metadata.shards_metadata[0].shard_sizes[other_dim] + ) + else: + global_size_list.append(cum_sum[-1]) + global_sharded_tensor_metadata.size = torch.Size(global_size_list) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharder.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharder.py new file mode 100644 index 0000000000000000000000000000000000000000..5d91ec15775bea870b81c4b10fb1443a3fba0977 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharder.py @@ -0,0 +1,29 @@ +import abc + +import torch.nn as nn + + +class Sharder(abc.ABC): + """ + This is an interface which allows user to create more advanced + sharding strategies that are not easily be composed by the + `ShardingSpec`. + + :class:`torch.distributed._shard.sharding_plan.ShardingPlan` could + take an object of the `Sharder` and call `shard` to shard the module, + then replace the original module with sharded module returned. + """ + + @abc.abstractmethod + def shard(self, module: nn.Module) -> nn.Module: + """ + Shard a module base on the implementation of this method, and + return the sharded version of the module. + + Args: + module (:class:`torch.nn.Module`): + The module to apply sharding to. + Returns: + A :class:`torch.nn.Module` object that represents a module + that's already been sharded. + """ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..325f7d7eb47b96a79fdc10cc2d1f072cdec9b4ce --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__init__.py @@ -0,0 +1 @@ +from .api import ShardingPlan, ShardingPlanner diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2ab40a5ae070174e70f35f8c380a7d8c5900da9e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c9b7185faf777c76ef66ef8574d2fd9fa67df2a9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/api.py new file mode 100644 index 0000000000000000000000000000000000000000..e4bbff548b2b7b69a7f3fdafcf730467fca12a7d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_plan/api.py @@ -0,0 +1,86 @@ +import abc +from dataclasses import dataclass + +import torch.nn as nn +from torch.distributed._shard.sharder import Sharder +from torch.distributed._shard.sharding_spec import ShardingSpec + + +@dataclass +class ShardingPlan: + """ + Representation of a sharding plan, describes how to shard a module + across hosts. `plan` is used to shard module parameters according to the spec provided, + `output_plan` and `return_local_tensor` are optional, they are used to specify the output + layout of a module with a spec, and when to convert back to data parallel fashion. + + Args: + plan (Dict[str, Union[:class:`torch.distributed._shard.sharding_spec.ShardingSpec`, + :class:`torch.distributed._shard.sharder.Sharder`]): + a dict describes how to shard a module, there're currently two ways to shard a module: + 1. directly shard a module parameter by a `ShardingSpec`, keyed by the name of + a parameter to a `ShardingSpec`. + 2. shard a submodule by applying a `Sharder` on it, keyed by the name of a module + to a `Sharder` object. + output_plan (Dict[str, :class:`torch.distributed._shard.sharding_spec.ShardingSpec`), optional): + a dict specifies the layout of a module's output which produces a ShardedTensor, + keyed by the name of module to ShardingSpec("" in key means the root module). + Default: `None` + return_local_tensor (List[str], optional): a list of string, each element enables + a module's sharded output to be returned as a Tensor from its local shards to + ensure further processing in a data parallel fashion. ("" in list means the + root module). + Default: None + Example: + Suppose we want to shard a module with two linear layers and then run it with DDP, we also + want to convert the output of the second linear layer back to DDP, we can do it as follows: + + >>> # xdoctest: +REQUIRES(module:torch._C._distributed_c10d) + >>> class MyModule(nn.Module): + >>> def __init__(self) -> None: + >>> super().__init__() + >>> self.fc1 = nn.Linear() + >>> self.gelu = nn.GELU() + >>> self.fc2 = nn.Linear() + >>> self.relu = nn.Linear() + >>> + >>> def forward(self, input): + >>> return self.relu(self.fc2(self.gelu(self.fc1(input)))) + + + >>> # xdoctest: +SKIP("Undefined spec1, spec2) + >>> sharding_plan = ShardingPlan( + >>> plan={ + >>> "fc1.weight": spec1, + >>> "fc2.weight": spec2 + >>> }, + >>> output_plan={ + >>> "fc2": output_spec + >>> }, + >>> return_local_tensor=["fc2"] + >>> ) + """ + + plan: dict[str, ShardingSpec | Sharder] + output_plan: dict[str, ShardingSpec] | None = None + return_local_tensor: list[str] | None = None + + +class ShardingPlanner(abc.ABC): + """ + Default ShardingPlanner interface, can be extended and + implement advanced sharding strategies. + """ + + @abc.abstractmethod + def build_plan(self, module: nn.Module) -> ShardingPlan: + """ + Given an nn.Module, define how to shard the module across + ranks, return a ShardingPlan + Args: + module (:class:`torch.nn.Module`): + The module to apply sharding to. + Returns: + A :class:`torch.distributed._shard.sharding_plan.ShardingPlan` object that + represents how to shard the module. + """ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..bfd3f0a7581e8c4352eba843af6d3751bee7f387 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__init__.py @@ -0,0 +1,10 @@ +from torch.distributed._shard.metadata import ShardMetadata + +from .api import ( + _infer_sharding_spec_from_shards_metadata, + DevicePlacementSpec, + EnumerableShardingSpec, + PlacementSpec, + ShardingSpec, +) +from .chunk_sharding_spec import ChunkShardingSpec as ChunkShardingSpec diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d646c0331d9cbc7f575d2a4b1ea20e12305d1beb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/_internals.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/_internals.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..056bd84f7fa41429a746d9c92e249c12929a5c40 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/_internals.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ff082425b2c893b2408bcf50590aea611342edcb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/chunk_sharding_spec.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/chunk_sharding_spec.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3880bf2bdcd0994bba1d76635935ab5b5f815ba1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/__pycache__/chunk_sharding_spec.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/_internals.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/_internals.py new file mode 100644 index 0000000000000000000000000000000000000000..0e0070cd5f435fd4fae7e1bfb26d859b920fa8f4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/_internals.py @@ -0,0 +1,244 @@ +# mypy: allow-untyped-defs +import math +import sys +from bisect import bisect_right, insort + +from torch.distributed._shard.metadata import ShardMetadata + + +def _check_shard_metadata_pair_overlap(shard1: ShardMetadata, shard2: ShardMetadata): + """ + Checks if two shards overlap. + """ + + # For each dim of each shard, check if one shard resides on the other + # end of second shard with respect to that dim. As an example for a 2D + # shard, we would check if one shard is above or on the left of the + # other shard. + ndims = len(shard1.shard_offsets) + for i in range(ndims): + if shard1.shard_offsets[i] >= shard2.shard_offsets[i] + shard2.shard_sizes[i]: + return False + if shard2.shard_offsets[i] >= shard1.shard_offsets[i] + shard1.shard_sizes[i]: + return False + + return True + + +def _find_nd_overlapping_shards( + shards: list[ShardMetadata], sharded_dims: list[int] +) -> tuple[int, int] | None: + """Find overlapping shards using sweep-line algorithm.""" + if len(shards) <= 1: + return None + + dims = len(sharded_dims) + if dims == 0: + return None + + sweep_dim_idx = 0 + if dims > 1: + max_size = 0 + for i, dim in enumerate(sharded_dims): + dim_size = shards[0].shard_offsets[dim] + shards[0].shard_sizes[dim] + if dim_size > max_size: + max_size = dim_size + sweep_dim_idx = i + sweep_dim = sharded_dims[sweep_dim_idx] + + sorted_indices = sorted( + range(len(shards)), + key=lambda idx: ( + shards[idx].shard_offsets[sweep_dim], + *(shards[idx].shard_offsets[d] for d in sharded_dims if d != sweep_dim), + ), + ) + active: list[tuple[int, int]] = [] + + for idx in sorted_indices: + current = shards[idx] + start = current.shard_offsets[sweep_dim] + end = start + current.shard_sizes[sweep_dim] + + cutoff = bisect_right(active, (start, sys.maxsize)) + if cutoff: + del active[:cutoff] + + for _, other_idx in active: + other = shards[other_idx] + + if _check_shard_metadata_pair_overlap(current, other): + return (other_idx, idx) + insort(active, (end, idx)) + return None + + +def _find_1d_overlapping_shards( + shards: list[ShardMetadata], dim: int +) -> tuple[int, int] | None: + # (begin, end, index_in_shards). Begin and end are inclusive. + intervals = [ + (s.shard_offsets[dim], s.shard_offsets[dim] + s.shard_sizes[dim] - 1, i) + for i, s in enumerate(shards) + ] + intervals.sort() + for i in range(len(shards) - 1): + if intervals[i][1] >= intervals[i + 1][0]: + return (intervals[i][2], intervals[i + 1][2]) + return None + + +def validate_non_overlapping_shards_metadata(shards: list[ShardMetadata]): + """ + Ensures none of the shards overlap with each other. + + Args: + shards(List[ShardMetadata]): List of :class:`ShardMetadata` objects representing + each shard. + Raises: + ``ValueError`` if there's overlap in any two shards. + """ + if not shards or len(shards) == 1: + return + + sharded_dims: list[int] = [] + for dim in range(len(shards[0].shard_offsets)): + for i in range(1, len(shards)): + if ( + shards[i].shard_offsets[dim] != shards[0].shard_offsets[dim] + or shards[i].shard_sizes[dim] != shards[0].shard_sizes[dim] + ): + sharded_dims.append(dim) + break + + pair: tuple[int, int] | None = None + if len(sharded_dims) == 0: + # if shard is all zeros, we should consider as pass + all_zeros: bool = all( + # strictly limited all offsets to be 0 to pass + # could loosen it later on + shard.shard_offsets == [0] * len(shards[0].shard_offsets) + and math.prod(shard.shard_sizes) == 0 # one dimension is 0 + for shard in shards + ) + if all_zeros: + return + # All shards are the same, all dims are not partitioned. Choose any 2. + pair = (0, 1) + elif len(sharded_dims) == 1: + # Shards are partitioned over only one dimension. Overlap can be found + # using a O(nlogn) overlapping interval algorithm. + pair = _find_1d_overlapping_shards(shards, sharded_dims[0]) + else: + # Shards are partitioned over more than one dimension. + # Use sweep-line algorithm for O(n log n) complexity. + pair = _find_nd_overlapping_shards(shards, sharded_dims) + + if pair: + raise ValueError(f"Shards {shards[pair[0]]} and {shards[pair[1]]} overlap") + + +def check_tensor(shards_metadata, tensor_dims) -> None: + """ + Checks if the shards_metadata is compatible with the provided tensor dims. + + Args: + shards_metadata(List[ShardMetadata]): List of :class:`ShardMetadata` + objects representing each shard of the tensor. + tensor_dims(Sequence of int): Dimensions of tensor to verify + Raises: + ``ValueError`` if not compatible. + """ + + # If the tensor's volume matches the total volume of all shards and + # all shard boundaries are within tensor dims, we have a compatible + # sharding spec for this tensor. Note that we have already verified + # we don't have overlapping shards. + tensor_rank = len(tensor_dims) + shards_rank = len(shards_metadata[0].shard_offsets) + if tensor_rank != shards_rank: + raise ValueError( + f"Rank of tensor is {tensor_rank}, but shards rank is {shards_rank}" + ) + + total_shard_volume = 0 + for shard in shards_metadata: + shard_volume = 1 + for i, shard_length in enumerate(shard.shard_sizes): + shard_volume *= shard_length + if shard.shard_offsets[i] + shard.shard_sizes[i] > tensor_dims[i]: + raise ValueError( + f"Shard offset {shard.shard_offsets[i]} and length " + f"{shard.shard_sizes[i]} exceeds tensor dim: {tensor_dims[i]} for shard {shard}" + ) + total_shard_volume += shard_volume + + tensor_volume = 1 + for size in tensor_dims: + tensor_volume *= size + + if total_shard_volume != tensor_volume: + # TODO: Can we improve this error message to point out the gaps? + raise ValueError( + f"Total volume of shards: {total_shard_volume} " + f"does not match tensor volume: {tensor_volume}, in other words " + f"all the individual shards do not cover the entire tensor" + ) + + +def get_split_size(dim_size, chunks): + """ + Computes the split size inline with ``torch.chunk`` + + Args: + dim_size(int): Size of the dimension being chunked. + chunks(int): Number of chunks to create for ``dim_size``. + + Returns: + An int indicating the split size to use. + """ + return (dim_size + chunks - 1) // chunks + + +def get_chunked_dim_size(dim_size, split_size, idx): + """ + Computes the dim size of the chunk for provided ``idx`` given ``dim_size`` + and ``split_size``. + + Args: + dim_size(int): Size of the dimension being chunked. + split_size(int): The chunk size for each chunk of ``dim_size``. + idx(int): The index of chunk whose dim size is being requested. + + Returns: + An int indicating the dim size of the chunk. + """ + return max(min(dim_size, split_size * (idx + 1)) - split_size * idx, 0) + + +def get_chunk_sharding_params(sharding_dim_size, world_size, spec, rank): + """ + Generate the start pos and offset length for the current rank for + chunk sharding. + + Args: + sharding_dim_size(int): The dimension length which we shard on. + world_size(int): number of ranks. + spec (:class:`torch.distributed._shard.sharding_spec.ChunkShardingSpec`): + sharding spec. + rank(int): # of cuda process. + + Returns: + start_pos(int): start position of sharded tensor on the given rank. + chunk_size(int): chunk size of sharded tensor on the given rank. + """ + split_size = get_split_size(sharding_dim_size, world_size) + current_offsets = 0 + start_pos = current_offsets + for idx, placement in enumerate(spec.placements): + chunk_size = get_chunked_dim_size(sharding_dim_size, split_size, idx) + if rank == placement.rank(): + start_pos = current_offsets + break + current_offsets += chunk_size + return start_pos, chunk_size # type: ignore[possibly-undefined] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/api.py new file mode 100644 index 0000000000000000000000000000000000000000..460286aeb1da0d85b1f8a9ed572977050e3aa925 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/api.py @@ -0,0 +1,264 @@ +# mypy: allow-untyped-defs +import functools +import operator +from abc import ABC, abstractmethod +from collections.abc import Callable +from dataclasses import dataclass +from typing import TYPE_CHECKING + +import torch +import torch.distributed._shard.sharded_tensor.metadata as sharded_tensor_meta +from torch.distributed._shard.metadata import ShardMetadata +from torch.distributed._shard.op_registry_utils import _decorator_func + +from ._internals import ( + check_tensor, + get_chunked_dim_size, + get_split_size, + validate_non_overlapping_shards_metadata, +) + + +if TYPE_CHECKING: + # Only include ShardedTensor when do type checking, exclude it + # from run-time to resolve circular dependency. + from torch.distributed._shard.sharded_tensor import ShardedTensor + + +class PlacementSpec(ABC): # noqa: B024 + """ + Base class representing the placement of an entity. Subclasses of this + class can be used to specify customized placements which might not be + covered by existing APIs. + """ + + +@dataclass +class DevicePlacementSpec(PlacementSpec): + """ + Associates placement of an entity with a single device. + + Args: + device(:class:`torch.distributed._remote_device`): The device to place the entity on. + """ + + device: torch.distributed._remote_device + + def __post_init__(self): + if not isinstance(self.device, torch.distributed._remote_device): + self.device = torch.distributed._remote_device(self.device) + + +class ShardingSpec(ABC): + """ + Base class representing sharding specifications. + """ + + @abstractmethod + def build_metadata( + self, + tensor_sizes: torch.Size, + tensor_properties: sharded_tensor_meta.TensorProperties, + ) -> sharded_tensor_meta.ShardedTensorMetadata: + """ + Given a global tensor size, define how to shard a tensor like this shape + across ranks, return ShardedTensorMetadata + Args: + tensor_sizes (:class:`torch.Size`): + The tensor shape to shard on, a `torch.Size` object that represents the + tensor shape to be sharded according to the ShardingSpec. + tensor_properties(:class:`torch.distributed._shard.sharded_tensor.TensorProperties): + Tensor properties used to create a ShardedTensor. + Returns: + A :class:`ShardedTensorMetadata` object that encodes the information about + the layout of the ShardedTensor and its properties. + """ + + @abstractmethod + def shard( + self, tensor: torch.Tensor, src_rank: int = 0, process_group=None + ) -> "ShardedTensor": + """ + Given a global tensor on src_rank, shard this tensor + across ranks within the process group, return a ShardedTensor. + Args: + tensor (:class:`torch.Tensor`): Tensor needs to be sharded. + Keyword args: + src_rank (int, optional): The source rank which is used as the ground truth of + the data for the parameter that would be sharded and scattered + across the rest of the ranks. + Default: 0. + process_group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + Returns: + A :class:`ShardedTensor` sharded from the given tensor. + """ + + +# Ops customized for a particular ShardingSpec. +_CUSTOM_SHARDING_SPEC_OPS: dict[str, dict[Callable, Callable]] = {} + + +def _has_custom_op(sharding_spec, op): + """ + Returns whether or not the ShardingSpec has a custom op implementation. + """ + class_name = type(sharding_spec).__qualname__ + return ( + class_name in _CUSTOM_SHARDING_SPEC_OPS + and op in _CUSTOM_SHARDING_SPEC_OPS[class_name] + ) + + +def _dispatch_custom_op( + sharding_spec, op: Callable, types, args, kwargs, process_group +): + """ + Calls the custom op for this ShardingSpec if it exists. + """ + class_name = type(sharding_spec).__qualname__ + if not _has_custom_op(sharding_spec, op): + raise RuntimeError(f"Custom op: {op} not registered for {class_name}") + func = _CUSTOM_SHARDING_SPEC_OPS[class_name][op] + return func(types, args, kwargs, process_group) + + +def custom_sharding_spec_op(sharding_spec_class, func): + """ + Decorator to allow custom registration of ops. + Args: + sharding_spec_class(type): The ShardingSpec for which we need to add this custom op. + func(Callable): The op to override (ex: torch.bmm) + """ + class_name = sharding_spec_class.__qualname__ + if class_name not in _CUSTOM_SHARDING_SPEC_OPS: + _CUSTOM_SHARDING_SPEC_OPS[class_name] = {} + return functools.partial( + _decorator_func, op=func, op_table=_CUSTOM_SHARDING_SPEC_OPS[class_name] + ) + + +@dataclass +class EnumerableShardingSpec(ShardingSpec): + """ + This is a type of PlacementSpec that allows users to specify a generic + sharding scheme by enumerating exactly how each shard is laid out. + + Args: + shards(List[ShardMetadata]): List of :class:`ShardMetadata` objects representing + each shard. Note that none of the shards should overlap. + """ + + shards: list[ShardMetadata] + + def __post_init__(self): + if len(self.shards) == 0: + raise ValueError(f"Empty shard list provided: {self.shards}") + + # Validate each shard has same rank. + rank = -1 + for shard in self.shards: + if rank != -1 and rank != len(shard.shard_offsets): + raise ValueError( + f"Found inconsistent ranks for shards: {rank} and {len(shard.shard_offsets)}" + ) + rank = len(shard.shard_offsets) + + validate_non_overlapping_shards_metadata(self.shards) + + def build_metadata( + self, + tensor_sizes: torch.Size, + tensor_properties: sharded_tensor_meta.TensorProperties, + ) -> sharded_tensor_meta.ShardedTensorMetadata: + # check if shards form a valid tensor + check_tensor(self.shards, tensor_sizes) + return sharded_tensor_meta.ShardedTensorMetadata( + self.shards, tensor_sizes, tensor_properties + ) + + def shard( + self, tensor: torch.Tensor, src_rank: int = 0, process_group=None + ) -> "ShardedTensor": + # TODO: figure out a generic and efficient way to scatter the shards for EnumerableShardingSpec + raise NotImplementedError("EnumerableShardingSpec.shard not implemented yet!") + + +def _infer_sharding_spec_from_shards_metadata(shards_metadata): + """ + Infer the sharding spec from the metadata of each shard of a ShardedTensor. + If the tensor is sharded only on one dimension, we can then verify whether it's + a ChunkShardingSpec or not. The way to verify it is to first get the total length + and perform a chunk sharding with the given placements to see if we can have the + same chunk size as the given shards_metadata. If not, we assume it's enum sharded. + + Args: + shards_metadata (List[ShardMetadata]): List of Metadata of local shards. + + Returns: + A :class:`torch.distributed._shard.sharding_spec.ShardingSpec` object of sharding + spec for one sharded tensor. + """ + placements = [] + chunk_sharding_dim = None + chunk_offset_list = [] + shard_size_list = [] + shard_offset_list = [] + # collect local shard metadata from the global sharded_tensor_metadata + for shard_metadata in shards_metadata: # type: ignore[attr-defined] + placements.append(shard_metadata.placement) + local_offsets = shard_metadata.shard_offsets + chunk_offset_list.append(sum(local_offsets)) + shard_size_list.append(shard_metadata.shard_sizes) + shard_offset_list.append(shard_metadata.shard_offsets) + shard_dims = [idx for idx, e in enumerate(local_offsets) if e != 0] + # If the offset is [0, 0, ..., 0] (all zeros), + # we cannot decide whether how the tensor is sharded. + if len(shard_dims) == 0: + continue + # If the offset is [0, N, .,0, M, 0, .., 0], + # we are sure it's sharded by more than one dimension. + if len(shard_dims) != 1: + chunk_sharding_dim = None + break + # If the offset is [0, 0, .,0, M, 0, .., 0], aka, it's sharded by just + # one dimension, we need to make sure all ranks share the same dimension. + if not chunk_sharding_dim: + chunk_sharding_dim = shard_dims[0] + elif chunk_sharding_dim != shard_dims[0]: + chunk_sharding_dim = None + break + + if chunk_sharding_dim is not None: + # Ensure we infer the correct placement order from offsets + placements = [ + x + for _, x in sorted( + zip(chunk_offset_list, placements), key=operator.itemgetter(0) + ) + ] + + from .chunk_sharding_spec import ChunkShardingSpec + + chunk_spec = ChunkShardingSpec( + dim=chunk_sharding_dim, + placements=placements, + ) + + shard_sizes = sorted([x[chunk_sharding_dim] for x in shard_size_list]) + shard_total_length = sum(shard_sizes) + shard_offsets = sorted([x[chunk_sharding_dim] for x in shard_offset_list]) + + chunks = len(placements) + split_size = get_split_size(shard_total_length, chunks) + chunk_shard_sizes = sorted( + [ + get_chunked_dim_size(shard_total_length, split_size, idx) + for idx in range(chunks) + ] + ) + # Should match ChunkShardingSpec offsets calculation + chunk_shard_offsets = [split_size * idx for idx in range(chunks)] + if shard_sizes == chunk_shard_sizes and shard_offsets == chunk_shard_offsets: + return chunk_spec + return EnumerableShardingSpec(shards_metadata) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec.py new file mode 100644 index 0000000000000000000000000000000000000000..d30b8a15585d12d374f00f7ad2b1c3cd572029f5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec.py @@ -0,0 +1,230 @@ +# mypy: allow-untyped-defs +from dataclasses import dataclass +from typing import cast, TYPE_CHECKING + +import torch +import torch.distributed as dist +import torch.distributed._shard.sharded_tensor.metadata as sharded_tensor_meta +import torch.distributed.distributed_c10d as distributed_c10d +from torch.distributed._shard._utils import narrow_tensor +from torch.distributed._shard.metadata import ShardMetadata +from torch.distributed._shard.sharded_tensor.shard import Shard +from torch.distributed._shard.sharded_tensor.utils import ( + _parse_and_validate_remote_device, +) + +from ._internals import get_chunked_dim_size, get_split_size +from .api import ShardingSpec + + +if TYPE_CHECKING: + # Only include ShardedTensor when do type checking, exclude it + # from run-time to resolve circular dependency. + from torch.distributed._shard.sharded_tensor import ShardedTensor + + +@dataclass +class ChunkShardingSpec(ShardingSpec): + """ + This is a type of PlacementSpec that defines the placement as being sharded + across multiple devices. In particular, it represents sharding a Tensor + along a single dimension into equal chunks (similar to :meth:`torch.chunk`). + + The semantics of how a tensor is partitioned is inline with + :meth:`torch.chunk`, where ``dim`` in torch.chunk corresponds to the + specified ``dim`` and ``chunks`` in torch.chunk is the number of elements + in the placement specified. + + Args: + dim (int): + The dimension to shard on, an integer representing the dimension. + placement(List[Union[_remote_device, str]]): + Specifies the placement of each shard of the Tensor. The size of + the list represents the number of shards to be created. This could + be a list of + :class:`torch.distributed._remote_device`'s. This list + could also contain a string which represents remote + device as accepted by + :class:`torch.distributed._remote_device` + """ + + ShardingDim = int | str + + dim: ShardingDim + placements: list[torch.distributed._remote_device | str] + + def __post_init__(self): + self._verify_dim(self.dim) + for i, remote_device in enumerate(self.placements): + if not isinstance(remote_device, torch.distributed._remote_device): + self.placements[i] = torch.distributed._remote_device(remote_device) + + @staticmethod + def _verify_dim(dim): + # Validate the sharding spec. + # TODO: support named dimension + if isinstance(dim, str): + raise NotImplementedError( + "ChunkShardingSpec does not support named dimension yet!" + ) + + if not isinstance(dim, int): + raise ValueError(f"Sharding dim needs to be an integer, found: {dim}") + + def build_metadata( + self, + tensor_sizes: torch.Size, + tensor_properties: sharded_tensor_meta.TensorProperties, + ) -> sharded_tensor_meta.ShardedTensorMetadata: + tensor_num_dim = len(tensor_sizes) + + self._verify_dim(self.dim) + if self.dim >= tensor_num_dim or self.dim < -tensor_num_dim: # type: ignore[operator] + raise ValueError(f"Invalid sharding dim: {self.dim}") + + shards_metadata = [] + sharding_dim_size = tensor_sizes[self.dim] # type: ignore[index] + chunks = len(self.placements) + split_size = get_split_size(sharding_dim_size, chunks) + for idx, placement in enumerate(self.placements): + # generate ShardMetadata for each placement device + chunked_dim_size = get_chunked_dim_size(sharding_dim_size, split_size, idx) + shard_size = list(tensor_sizes) + current_offsets = [0] * tensor_num_dim + current_offsets[self.dim] = split_size * idx # type: ignore[index] + shard_size[self.dim] = chunked_dim_size # type: ignore[index] + + shard_metadata = ShardMetadata( + shard_offsets=current_offsets, + shard_sizes=shard_size, + placement=placement, + ) + shards_metadata.append(shard_metadata) + + return sharded_tensor_meta.ShardedTensorMetadata( + shards_metadata, tensor_sizes, tensor_properties + ) + + def shard( + self, tensor: torch.Tensor, src_rank: int = 0, process_group=None + ) -> "ShardedTensor": + """ + Args: + src_rank: group rank relative to ``process_group`` + + N.B. If ``process_group`` is None, ``src_rank`` is a global rank. + """ + # relative imports to avoid circular dependency + from torch.distributed._shard.sharded_tensor import ShardedTensor + + tensor_properties = sharded_tensor_meta.TensorProperties( + dtype=tensor.dtype, + layout=tensor.layout, + requires_grad=tensor.requires_grad, + memory_format=torch.contiguous_format, + pin_memory=tensor.is_pinned(), + ) + current_rank = dist.get_rank(process_group) + current_global_rank = dist.get_rank() + tensor_meta = self.build_metadata(tensor.size(), tensor_properties) + local_shards = [] + local_tensor = None + local_metadata = None + + tensors_to_scatter = cast( + list[torch.Tensor | None], + [None] * dist.get_world_size(process_group), + ) + + sharding_dim_size = tensor.size()[self.dim] # type: ignore[index] + chunks = len(self.placements) + split_size = get_split_size(sharding_dim_size, chunks) + scatter_shape = list(tensor.size()) + scatter_shape[self.dim] = split_size # type: ignore[index] + + for shard_meta in tensor_meta.shards_metadata: + remote_global_rank, device = _parse_and_validate_remote_device( + process_group, shard_meta.placement + ) + if current_rank == src_rank: + # Reshape to get shard for this rank and we don't want autograd + # recording here for the narrow op and 'local_shard' should be a + # leaf variable in the autograd graph. + narrowed_tensor = narrow_tensor(tensor, shard_meta) + if shard_meta.shard_sizes[self.dim] < split_size: # type: ignore[index] + # for the last shard that might be smaller to other shards + # resize the narrowed tensor to the same size and use it for + # the scatter collective as dist.scatter requires same size + # inputs on every rank + tensor_to_scatter = ( + narrowed_tensor.detach().clone().resize_(scatter_shape) + ) + else: + tensor_to_scatter = narrowed_tensor.detach().clone( + memory_format=torch.contiguous_format + ) + + tensors_to_scatter[ + # pyrefly: ignore [bad-argument-type] + dist.get_group_rank(process_group, remote_global_rank) + ] = tensor_to_scatter + + if current_global_rank == remote_global_rank: + local_tensor = torch.empty( + scatter_shape, + dtype=tensor.dtype, + layout=tensor.layout, + device=device, + ) + local_metadata = shard_meta + + # each rank should have local_tensor and local_metadata initialized if we build + # the metadata list in a correct way. + if local_tensor is None: + raise AssertionError + if local_metadata is None: + raise AssertionError + + # Scatter the shards to all ranks in the pg + # scatter takes the global rank as ``src`` + src_for_scatter = src_rank + if ( + process_group is not None + and process_group is not distributed_c10d._get_default_group() + ): + src_for_scatter = distributed_c10d.get_global_rank( + process_group, src_for_scatter + ) + + tensors_to_scatter_: list[torch.Tensor] | None = None + if current_rank == src_rank: + tensors_to_scatter_ = [] + for t in tensors_to_scatter: + if not isinstance(t, torch.Tensor): + raise AssertionError + tensors_to_scatter_.append(t) + + dist.scatter( + local_tensor, + scatter_list=tensors_to_scatter_, + src=src_for_scatter, + group=process_group, + ) + + if list(local_tensor.size()) != local_metadata.shard_sizes: + # detach again after receiving to ensure local shards remain a leaf node + local_tensor = local_tensor.resize_(local_metadata.shard_sizes).detach() + + # Sync requires_grad to local_shard. + local_tensor.requires_grad = tensor.requires_grad + + local_shards.append(Shard(tensor=local_tensor, metadata=local_metadata)) + + st = ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards, tensor_meta, process_group=process_group + ) + + # Manually set sharding_spec + st._sharding_spec = self + + return st diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b69366eedfc36443ec84c4394f8f93378450b9e0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/_common.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/_common.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b0fd7bf6028b91d1a0546ff6b111f13d0c1650b0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/_common.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..992bf4c1152aad75c77c56583454ee64d49c1df2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding_bag.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding_bag.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e5e14bade5f24d09126dd44d4ef36b5dd514419f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/__pycache__/embedding_bag.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/_common.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/_common.py new file mode 100644 index 0000000000000000000000000000000000000000..eecafaf8a4ec19627106046f096e3fbfc96a0a4a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/_common.py @@ -0,0 +1,350 @@ +# mypy: allow-untyped-defs + +import torch +import torch.distributed as dist +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed._shard.sharded_tensor._ops._common import _sharded_op_common +from torch.distributed._shard.sharding_spec import ChunkShardingSpec +from torch.distributed._shard.sharding_spec._internals import ( + get_chunk_sharding_params, + get_chunked_dim_size, + get_split_size, +) +from torch.distributed._shard.sharding_spec.api import custom_sharding_spec_op +from torch.distributed.nn.functional import ( + _all_gather_base, + all_reduce, + all_to_all_single, +) + + +def _chunk_sharding_spec_check(spec, op): + """ + For the given op implementation check if the sharding spec is ChunkShardingSpec. + """ + if not isinstance(spec, ChunkShardingSpec): + raise NotImplementedError( + f"Only ChunkShardingSpec supported for '{op.__name__}'." + ) + + +def _register_sharded_op_on_local_tensor( + op, early_stop_func=None, extra_check=None, customized_func=None +): + """ + Handles ``__torch_function__`` dispatch for ops which are performed on + the single local tensor of the sharded tensor such as op like + ``torch.nn.functional.softmax`` or ``torch.Tensor.view``. + + For more complicated ops, a customized func can be used to generate + the new local tensor, sharding spec and sharded tensor size. + + Args: + op: The op to be registered and applied to all shards of the st. + early_stop_func (Callable, optional): the func for early stop. + Default: if ``None``, no early stop. + extra_check (Callable, optional): the func for extra condition check. + Default: if ``None``, no extra check. + customized_func (Callable, optional): the func for customized logic + to generate the new local tensor, sharding spec and sharded tensor size. + Default: if ``None``, we simply lower to the real op call with + the single local tensor of the st. + + Return: + func (Callable): registered implementation for sharded op for + ``__torch_function__`` dispatch. + """ + + @custom_sharding_spec_op(ChunkShardingSpec, op) + @_sharded_op_common(op, early_stop_func, extra_check) + def sharded_tensor_op_on_local_tensor(types, args=(), kwargs=None, pg=None): + # pyrefly: ignore [bad-index] + st = args[0] + sharding_spec = st.sharding_spec() + if len(st.local_shards()) != 1: + raise TypeError( + f"torch function '{op.__name__}', with args: {args} and " + f"kwargs: {kwargs} only supported for single local tensor!" + ) + st_size = st.size() + if customized_func: + local_tensor, sharding_spec, st_size = customized_func(args, kwargs, pg) + else: + args = (st.local_tensor(), *args[1:]) + local_tensor = op(*args, **kwargs) + return ShardedTensor._init_from_local_tensor( + local_tensor.contiguous(), + sharding_spec, + st_size, # type: ignore[arg-type] + process_group=pg, + init_rrefs=st._init_rrefs, + ) + + +def _handle_col_wise_sharding_base( + op_func, + col_dim, + input, + world_size, + weight, + local_shard, + pg, + gathered_inputs, + mode=None, + gathered_per_sample_weights=None, + gathered_offsets=None, + padding_idx=None, +): + """ + For col-wise sharding of weight, lots of logic are common. + So we extract the common logic and put in this function: + Step 1. To get input from each rank and + Step 2. To perform the op on the concatenated tensor. + Step 3. To distribute results to each rank with col rearrangement. + Step 4. To concatenate all results from all ranks. + + Args: + op_func: operator which is applied to the input tensor. + col_dim: dim of result tensor after the operation. + input: tensor to be applied op on. + world_size: number of ranks. + weight: sharded weight tensor. + local_shard: col-wise sharded weight tensor. + pg: process group. + gathered_inputs: list of inputs from all ranks. If specified, we + don't need to communicate with each rank any more. + mode: aggregation mode of EmbeddingBag. + gathered_per_sample_weights: per_sample_weights across all ranks. + gathered_offsets: offsets across all ranks. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + Note that the embedding vector at padding_idx is + excluded from the reduction. + + Return: final result of input being applied with the op. + """ + # run the operator's function for all the inputs. + results = [] + for i, inp in enumerate(gathered_inputs): + if op_func is torch.nn.functional.embedding_bag: + result = op_func( + inp, + local_shard, + offsets=gathered_offsets[i] if gathered_offsets is not None else None, + # pyrefly: ignore [bad-argument-type] + mode=mode, + per_sample_weights=gathered_per_sample_weights[i] + if gathered_per_sample_weights is not None + else None, + padding_idx=padding_idx, + ) + elif op_func is torch.nn.functional.embedding: + result = op_func( + inp, + local_shard, + padding_idx=padding_idx, + ) + else: + result = op_func(inp, local_shard) + results.append(torch.transpose(result, 0, col_dim)) + + # Distribute results to each rank with col rearrangement. + output = _result_distribute_with_col_rearrange( + results, input, world_size, weight, pg + ) + + # transpose the output and return result. + return torch.transpose(output, 0, col_dim) + + +def _result_distribute_with_col_rearrange(results, input, world_size, weight, pg): + """ + For col-wise sharding of weight, we need to distribute + results to each rank. We do them in this function. + Note that, if the index in the Sharding Spec is not equal to + the rank number, we need to do the rearrangement based on the + order given by the Sharding Spec (placement). + + Args: + results: results from ops applied to inputs from all ranks. + We need to distribute them back to their original ranks. + input: tensor to be applied op to. + world_size: number of ranks. + weight: sharded weight tensor. + pg: process group. + + Return: column rearranged result. + """ + # Process results and outputs for all2all. + sharding_dim = weight._sharding_spec.dim + sharding_dim_size = weight.size(sharding_dim) + dims = list(results[0].size()) + dims[0] = sharding_dim_size + combined_results = torch.cat(results) + output = torch.empty( + *dims, device=combined_results.device, dtype=combined_results.dtype + ) + + # Compute output splits + split_size = get_split_size(sharding_dim_size, world_size) + output_split_sizes = [0] * world_size + for idx, placement in enumerate(weight._sharding_spec.placements): + output_split_sizes[placement.rank()] = get_chunked_dim_size( + sharding_dim_size, split_size, idx + ) + + # distribute the outputs using all2all. + output = all_to_all_single( + output, combined_results, output_split_sizes=output_split_sizes, group=pg + ) + + # Check if we need to rearrange columns appropriately for output. + rearrange_columns = any( + idx != placement.rank() + for idx, placement in enumerate(weight._sharding_spec.placements) + ) + if not rearrange_columns: + return output + + indices = [] + for placement in weight._sharding_spec.placements: + dim_size = output_split_sizes[placement.rank()] + start = sum( + split_size if i < placement.rank() else 0 + for i, split_size in enumerate(output_split_sizes) + ) + indices += list(range(start, start + dim_size)) + + return output.index_select(0, torch.tensor(indices, device=output.device)) + + +def _handle_max_norm_col_wise( + max_norm, + norm_type, + local_shard, + input, + world_size, + gathered_inputs, + pg, +): + """ + For col-wise sharding of weight, we need to aggregate the + norm across all ranks before we can perform the proper re-norm. + Note that, the max_norm logic is only applied to the embedding + indices that are looked up and not the whole shard. + + Args: + max_norm: If given, each embedding vector with norm larger + than max_norm is renormalized to have norm max_norm. + Note: this will modify weight in-place. + norm_type: The p in the p-norm to compute for the max_norm option. + local_shard: col-wise shared local weight used for lookup. + input: tensor to be applied op to. + world_size: number of ranks. + gathered_inputs: list of inputs from all ranks. + pg: process group. + + Return: + local_shard_norm_renormed: local_shard re-normed to max_norm if the norm is larger + than it. + + """ + norm_type = norm_type if norm_type is not None else 2.0 + unique_inp = torch.unique(torch.cat(gathered_inputs)) + local_shard_sum = torch.sum( + torch.pow(torch.abs(local_shard), norm_type), dim=1, dtype=local_shard.dtype + ) + # For col-wise sharding, we need to first aggregate the powered sum + # from each rank first and then calculate the norm. + local_shard_sum = all_reduce(local_shard_sum, group=pg) + local_shard_norm = torch.pow(local_shard_sum, 1.0 / norm_type) + max_norm_tensor = torch.full( + (local_shard.size(0),), + float("inf"), + dtype=local_shard.dtype, + device=input.device, + ) + max_norm_tensor[unique_inp] = max_norm + local_shard_t = local_shard.t().contiguous() + normalized_tensor = torch.where( + local_shard_norm > max_norm_tensor, max_norm_tensor, local_shard_norm + ) + # Make sure divisor is not zero. + local_shard_norm[local_shard_norm == 0.0] = 1.0 + local_shard_norm_renormed = ( + torch.div(torch.mul(local_shard_t, normalized_tensor), local_shard_norm) + .t() + .contiguous() + ) + return local_shard_norm_renormed + + +def _all_gather_base_input(input, pg): + """ + Use _all_gather_base to get a concatenated input from each rank. + + Args: + input: tensor to be applied op on. + pg: process group. + + Returns: + gathered_inputs: input gathered from each rank and concat by dim 0. + """ + # allgather the inputs first. + gather_inp_size = list(input.size()) + gather_inp_size[0] = input.size(0) * dist.get_world_size(pg) + gather_inp = torch.empty(gather_inp_size, device=input.device, dtype=input.dtype) + return _all_gather_base(gather_inp, input, group=pg) + + +def _handle_row_wise_mask(gather_inp, padding_idx, weight, world_size, rank): + """ + Mask the input for embedding look-up for IDs which are not stored + on the current rank. This function also adjust the ``padding_idx`` + so that it is only used on the rank where the corresponding row is + stored. + + Note that, with ``max_norm`` flag on, only weights of rows being + looked up will be re-normed. So we need an extra row for masked ID + so that it does not affect the final result and ``max_norm``. + + Args: + gather_inp: tensor to be applied op on gathered from all ranks. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + Note that the embedding vector at padding_idx is + excluded from the reduction. + weight: weight tensor of Embedding look-up table. + world_size: number of ranks. + rank: # of cuda process. + + Returns: + lookup_input: Tensor of masked input. + padding_idx: adjusted padding_idx. + padding_row: The extra row we used during lookup so that + looking up does not affect ``max_norm``. + """ + (start_pos, chunk_size) = get_chunk_sharding_params( + weight.size(0), world_size, weight._sharding_spec, rank + ) + mask = (gather_inp < start_pos) | (gather_inp >= start_pos + chunk_size) + lookup_input = gather_inp.clone() - start_pos + lookup_input[mask] = chunk_size + if ( + padding_idx is not None + and padding_idx >= start_pos + and padding_idx < (start_pos + chunk_size) + ): + padding_idx = padding_idx - start_pos + else: + padding_idx = None + + # When max_norm is set, it will only re-norm the row being looked up. + padding_row = torch.zeros( + 1, weight.size(1), device=gather_inp.device, dtype=weight.dtype + ) + return lookup_input, padding_idx, padding_row diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding.py new file mode 100644 index 0000000000000000000000000000000000000000..ea4579359cfaf03e4f71e8f890c78dcf7c68abe7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding.py @@ -0,0 +1,294 @@ +# mypy: allow-untyped-defs + +import torch +import torch.distributed as dist +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed._shard.sharding_spec import ChunkShardingSpec +from torch.distributed._shard.sharding_spec.api import custom_sharding_spec_op +from torch.distributed.nn.functional import all_gather, reduce_scatter + +from ._common import ( + _all_gather_base_input, + _handle_col_wise_sharding_base, + _handle_max_norm_col_wise, + _handle_row_wise_mask, +) + + +@custom_sharding_spec_op(ChunkShardingSpec, torch.nn.functional.embedding) +def sharded_embedding(types, args, kwargs, pg): + """ + Handles ``__torch_function__`` dispatch for ``torch.nn.functional.embedding``. + This method computes a sharded embedding lookup and has the following limitations: + + 1. Supports only sharding of ``weight``. + 2. Supports only ``ChunkShardingSpec``. + 3. Supports only a single local shard per rank. + 4. Supports all specs except for scale_grad_by_freq, sparse, etc. + + Based on the dimension that the weight is sharded on, there are two + algorithms: + + ROWWISE SHARDING + ================ + For row-wise sharding the weight is sharded on dimension 0. + + The overall algorithm can be best explained with an example. Let's assume + the dims for input are (4 x 6) and W are (10 x 17) and W is sharded across + 4 GPUs creating 3 shard of (3 x 17) and 1 shard of (1 x 17). + The algorithm is as follows: + + 1. First the input is all gathered to all ranks, since this is SPMD and + input is actually sharded across all ranks. The inputs then become a + 4 (4 x 6) tensor on each rank. For example if the given input is + tensor([[6, 5, 2, 9, 6, 3], + [3, 1, 2, 4, 7, 6], + [4, 0, 4, 9, 8, 9], + [8, 6, 6, 4, 6, 1]]) + on rank 0. + Then on every rank, we will have this tensor. + If input itself is already replicated, no all-gather will be done. + 2. Next, we mask the ID which are not stored on that rank. + For example on rank 0, we store ID [0, 1, 2]. We only keep the ID + inside the set of numbers. The rest of them will be masked to an extra row. + The masked matrix will be used for embedding look up and is like: + tensor([[4, 4, 2, 4, 4, 4], + [4, 1, 2, 4, 4, 4], + [4, 0, 4, 4, 4, 4], + [4, 4, 4, 4, 4, 1]]) + The reason of having an extra row (aka, number 4 in the example) is + because when max_norm is specified only weight which has looked will + be re-normed so mask IDs whose embeddings are not stored in current + rank will to an extra row will ensure max_norm still works as expected. + 3. If max_norm is specified, the extra row guarantees that the mask ID will + not affect the behavior of weight re-norm. + + COLWISE SHARDING + ================ + For col-wise sharding the weight is sharded on dimension 1. + + The overall algorithm can be best explained with an example. Let's assume + the dims for input are (4 x 6) and W are (16 x 17) and W is sharded across + 4 GPUs creating 3 shards of (16 x 5) and 1 shard of (16 x 2). + The algorithm is as follows: + + 1. First the input is broadcasted to all ranks, since this is SPMD we + actually do an all_gather for all the inputs resulting in 4 (4 x 6) + inputs on each rank. + 2. Next we perform local embedding lookup operation by apply each + input (4 x 6) with the local shard (16 x 5) ((16 x 2) for the last). + This results in 4 (5 x 6 x 4) ((2 x 6 x 4) for the last) matrices + on each rank. We transpose dim 0 and dim 2. + 3. Next, we concat these 4 matrices and perform an all2all to share the + appropriate (5 x 6 x 4) or (2 x 6 x 4) matrices to each rank. + 4. Now, each rank receives a (17 x 6 x 4) matrix which is basically the + size of the result we need. + 5. If placements are not in order any appropriate rearrangement of columns + are done for the (17 x 6 x 4) matrix and finally we transpose the + dim 0 and dim 2 again. + 6. If max_norm is specified, we manually sum up the norm and renorm. Because + the renorm must be in place, we need to override the local_shard to mimic + this behavior. + """ + # Validate input params + _validate_embedding_param(args, kwargs) + + input = args[0] + weight = args[1] + max_norm = kwargs.get("max_norm") + norm_type = kwargs.get("norm_type") + padding_idx = kwargs.get("padding_idx") + + local_shard = weight.local_tensor().contiguous() + sharding_dim = weight._sharding_spec.dim + world_size = dist.get_world_size(pg) + rank = dist.get_rank(pg) + + if sharding_dim == 1: + output, local_shard = _handle_col_wise_sharding( + input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, pg + ) + weight.local_shards()[0].tensor = local_shard + return output + elif sharding_dim == 0: + return _handle_row_wise_sharding( + input, + world_size, + weight, + local_shard, + max_norm, + norm_type, + padding_idx, + rank, + pg, + ) + else: + raise RuntimeError( + f"nn.Embedding weight sharded on dim {sharding_dim} not supported!" + ) + + +def _validate_embedding_param(args, kwargs): + """ + Validate input params of sharded embedding op. + + Args: + input: list of ID used for lookup. + weight: sharded weight tensor. + kwargs: same as normal Embedding. + + Return: None. + """ + + input = args[0] + weight = args[1] + max_norm = kwargs.get("max_norm") + scale_grad_by_freq = kwargs.get("scale_grad_by_freq") + sparse = kwargs.get("sparse") + + # Validate types + if not isinstance(input, torch.Tensor): + raise TypeError("input need to be torch.Tensor") + if not isinstance(weight, ShardedTensor): + raise TypeError("weight needs to be ShardedTensor") + weight_size = weight.size() + if len(weight_size) != 2: + raise ValueError("Weight needs to have exactly 2 dims") + if int(torch.min(input).item()) < 0: + raise ValueError( + "Index out of range in Input %d %d", + int(torch.min(input).item()), + weight_size[1], + ) + if int(torch.max(input).item()) >= weight_size[0]: + raise ValueError( + "Index out of range in Input %d %d", + int(torch.max(input).item()), + weight_size[1], + ) + if scale_grad_by_freq: + raise RuntimeError( + 'nn.Embedding weight sharded with flag on "scale_grad_by_freq" not supported!' + ) + if sparse: + raise RuntimeError( + 'nn.Embedding weight sharded with flag on "sparse" not supported!' + ) + if max_norm and max_norm <= 0.0: + raise ValueError('"max_norm" must be larger than zero!') + + if not isinstance(weight._sharding_spec, ChunkShardingSpec): + raise ValueError("Only ChunkShardingSpec supported for ShardedTensor ops!") + if len(weight.local_shards()) != 1: + raise ValueError("Only one local shard supported!") + + +def _handle_col_wise_sharding( + input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, pg +): + """ + Entry-point function to handle the logic of col-wise sharding of weight + for embedding. (Detailed explanations of the logic can be found in + the comment for sharded_embedding.) + + Args: + input: list of ID used for lookup and aggregation. + world_size: number of ranks. + weight: sharded weight tensor. + local_shard: col-wise shared local weight used for lookup. + max_norm: If given, each embedding vector with norm larger + than max_norm is renormalized to have norm max_norm. + Note: this will modify weight in-place. + norm_type: The p in the p-norm to compute for the max_norm option. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + pg: process group. + + Returns: final result of lookup. + """ + # allgather the inputs first for non Replicated Tensor. + gathered_inputs = all_gather(input, group=pg) + + if max_norm is not None: + # max_norm changes the weight in-place + local_shard = _handle_max_norm_col_wise( + max_norm, norm_type, local_shard, input, world_size, gathered_inputs, pg + ) + + output = _handle_col_wise_sharding_base( + torch.nn.functional.embedding, + len(input.size()), + input, + world_size, + weight, + local_shard, + pg, + gathered_inputs, + padding_idx=padding_idx, + ) + return (output, local_shard) + + +def _handle_row_wise_sharding( + input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, rank, pg +): + """ + Entry-point function to handle the logic of row-wise sharding of weight + for embedding. (Detailed explanations of the logic can be found in + the comment for sharded_embedding.) + + Args: + input: list of ID used for lookup and aggregation. + world_size: number of ranks. + weight: sharded weight tensor. + local_shard: row-wise shared local weight used for lookup. + max_norm: If given, each embedding vector with norm larger + than max_norm is renormalized to have norm max_norm. + Note: this will modify weight in-place. + norm_type: The p in the p-norm to compute for the max_norm option. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + rank: # of cuda process. + pg: process group. + + Returns: final result of lookup. + """ + # allgather the inputs first for non Replicated Tensor. + gather_inp = _all_gather_base_input(input, pg) + + # Mask the input according to sharding spec. + lookup_input, padding_idx, padding_row = _handle_row_wise_mask( + gather_inp, padding_idx, weight, world_size, rank + ) + + # When input is a large tensor, the value of weight is changed. + # This is a walk-around for now. GH issue: #81717 + if max_norm is not None: + torch.nn.functional.embedding( + torch.unique(lookup_input)[:-1], + local_shard, + padding_idx=padding_idx, + max_norm=max_norm, + norm_type=norm_type, + ) + max_norm = None + + local_input_embeddings = torch.nn.functional.embedding( + lookup_input, + torch.cat([local_shard, padding_row]), + padding_idx=padding_idx, + max_norm=max_norm, + norm_type=norm_type, + ) + + # TODO: Make the result a PartialTensor. + local_shards = local_input_embeddings.chunk(pg.size()) + return reduce_scatter( + torch.empty_like(local_shards[0]), + list(local_shards), + group=pg, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding_bag.py b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding_bag.py new file mode 100644 index 0000000000000000000000000000000000000000..5708406de35a346a225fc59c3fbb9cc73157dbf3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/embedding_bag.py @@ -0,0 +1,477 @@ +# mypy: allow-untyped-defs + +from typing import cast + +import torch +import torch.distributed as dist +from torch._C._distributed_c10d import ReduceOp +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed._shard.sharding_spec import ChunkShardingSpec +from torch.distributed._shard.sharding_spec.api import custom_sharding_spec_op +from torch.distributed.nn.functional import all_gather, reduce_scatter + +from ._common import ( + _all_gather_base_input, + _handle_col_wise_sharding_base, + _handle_max_norm_col_wise, + _handle_row_wise_mask, +) + + +@custom_sharding_spec_op(ChunkShardingSpec, torch.nn.functional.embedding_bag) +def sharded_embedding_bag(types, args, kwargs, pg): + """ + Handles ``__torch_function__`` dispatch for ``torch.nn.functional.embedding_bag``. + This method computes a sharded embedding bag aggregation and has the following limitations: + + 1. Supports only sharding of ``weight``. + 2. Supports only ``ChunkShardingSpec``. + 3. Supports only a single local shard per rank. + 4. Supports all specs except for scale_grad_by_freq, sparse, etc. + + Based on the dimension that the weight is sharded on, there are two + algorithms: + + ROWWISE SHARDING + ================ + For row-wise sharding the weight is sharded on dimension 0. + + The overall algorithm can be best explained with an example. Let's assume + the dims for input are (4 x 6) and W are (16 x 17) and W is sharded across + 4 GPUs creating 4 shard of (4 x 17). + The algorithm is as follows: + + 1. First the input is all gathered to all ranks, since this is SPMD and + input is actually sharded across all ranks. The inputs then become a + 4 (4 x 6) tensor on each rank. For example if the given input is + tensor([[6, 5, 2, 9, 6, 3], + [3, 1, 2, 4, 7, 6], + [4, 0, 4, 9, 8, 9], + [8, 6, 6, 4, 6, 1]]) + on rank 0. + Then on every rank, we will have this tensor. + If input itself is already replicated, no all-gather will be done. + 2. Next, we mask the ID which are not stored on that rank. + For example on rank 0, we store ID [0, 1, 2]. We only keep the ID + inside the set of numbers. The rest of them will be masked to an extra row. + The masked matrix will be used for embedding look up and is like: + tensor([[4, 4, 2, 4, 4, 4], + [4, 1, 2, 4, 4, 4], + [4, 0, 4, 4, 4, 4], + [4, 4, 4, 4, 4, 1]]) + 3. If ``max_norm`` is specified, the extra row guarantees that the mask ID will + not affect the behavior of weight re-norm. + 4. The example above only happens in one rank and each rank does a very similar thing. + For "Mean" mode we need to divide by either column size (2D) or the interval length + defined by the offset (excluding the row specified in ``padding_idx``). + We also need to mask the nonexistent row to neg Inf so that negative value does not + gets wiped out in the "Max" mode. + + COLWISE SHARDING + ================ + For col-wise sharding the weight is sharded on dimension 1. + + The overall algorithm can be best explained with an example. Let's assume + the dims for input are (4 x 6) and W are (16 x 17) and W is sharded across + 4 GPUs creating 3 shards of (16 x 5) and 1 shard of (16 x 2). + The algorithm is as follows: + + 1. First the input is broadcasted to all ranks, since this is SPMD we + actually do an all_gather for all the inputs resulting in 4 (4 x 6) + inputs on each rank. + 2. Next we perform local embedding bag operation under the given mode by + apply each input (4 x 6) with the local shard (16 x 5) ((16 x 2) for the last). + This results in 4 (5 x 4) ((2 x 4) for the last) matrices on each rank. + We transpose the aggregation result. + 3. Next, we concatenate these 4 matrices and perform an all2all to share the + appropriate (5 x 4) or (2 x 4) matrices to each rank. + 4. Now, each rank receives a (17 x 4) matrix which is basically the + size of the result we need. + 5. If placements are not in order any appropriate rearrangement of columns + are done for the (17 x 4) matrix and finally we transpose the output again. + 6. If max_norm is specified, we manually sum up the norm and renorm. Because + the renorm must be in place, we need to override the local_shard to mimic + this behavior. + """ + # Validate input params + _validate_embedding_bag_param(args, kwargs) + + input = args[0] + weight = args[1] + offsets = kwargs.get("offsets") + per_sample_weights = kwargs.get("per_sample_weights") + mode = kwargs.get("mode") + max_norm = kwargs.get("max_norm") + norm_type = kwargs.get("norm_type") + include_last_offset = kwargs.get("include_last_offset") + padding_idx = kwargs.get("padding_idx") + + local_shard = weight.local_tensor().contiguous() + sharding_dim = weight._sharding_spec.dim + world_size = dist.get_world_size(pg) + rank = dist.get_rank(pg) + if include_last_offset: + offsets = offsets[:-1] + + if sharding_dim == 1: + output, local_shard = _handle_col_wise_sharding( + input, + world_size, + weight, + local_shard, + offsets, + per_sample_weights, + mode, + max_norm, + norm_type, + padding_idx, + pg, + ) + weight.local_shards()[0].tensor = local_shard + return output + elif sharding_dim == 0: + return _handle_row_wise_sharding( + input, + world_size, + weight, + local_shard, + offsets, + per_sample_weights, + mode, + max_norm, + norm_type, + padding_idx, + rank, + pg, + ) + else: + raise RuntimeError( + f"nn.EmbeddingBag weight sharded on dim {sharding_dim} not supported!" + ) + + +def _validate_embedding_bag_param(args, kwargs): + """ + Validate input params of sharded embeddingBag op. + + Args: + input: list of ID used for lookup and aggregation. + weight: sharded weight tensor. + kwargs: same as normal EmbeddingBag. + + Return: None. + """ + + input = args[0] + weight = args[1] + offsets = kwargs.get("offsets") + per_sample_weights = kwargs.get("per_sample_weights") + mode = kwargs.get("mode") + max_norm = kwargs.get("max_norm") + scale_grad_by_freq = kwargs.get("scale_grad_by_freq") + sparse = kwargs.get("sparse") + include_last_offset = kwargs.get("include_last_offset") + + # Validate types + if not isinstance(input, torch.Tensor): + raise TypeError("input need to be torch.Tensor") + if offsets is not None and not isinstance(offsets, torch.Tensor): + raise TypeError("offsets need to be torch.Tensor") + if per_sample_weights is not None and not isinstance( + per_sample_weights, torch.Tensor + ): + raise TypeError("per_sample_weights need to be torch.Tensor") + if not isinstance(weight, ShardedTensor): + raise TypeError("weight needs to be ShardedTensor") + if len(input.size()) > 2: + raise ValueError("Input more than 2 dims not supported") + weight_size = weight.size() + if len(weight_size) != 2: + raise ValueError("Weight needs to have exactly 2 dims") + if int(torch.min(input).item()) < 0: + raise ValueError( + "Index out of range in Input %d %d", + int(torch.min(input).item()), + weight_size[1], + ) + if int(torch.max(input).item()) >= weight_size[0]: + raise ValueError( + "Index out of range in Input %d %d", + int(torch.max(input).item()), + weight_size[1], + ) + if offsets is not None and len(input.size()) != 1: + raise ValueError("Input dimension needs to be exactly 1 dim") + if len(input.size()) == 1 and offsets is None: + raise ValueError("offsets is required for 1D input") + if per_sample_weights is not None and per_sample_weights.size() != input.size(): + raise ValueError( + f"per_sample_weights size {per_sample_weights.size()} not equal to input size {input.size()}" + ) + if mode is None: + mode = "mean" + if mode not in ["sum", "mean", "max"]: + raise ValueError(f"mode '{mode}' is not supported") + if scale_grad_by_freq: + raise RuntimeError( + 'nn.Embedding weight sharded with flag on "scale_grad_by_freq" not supported!' + ) + if sparse: + raise RuntimeError( + 'nn.Embedding weight sharded with flag on "sparse" not supported!' + ) + if include_last_offset and offsets is None: + raise ValueError('offsets is required for flag "include_last_offset"!') + if include_last_offset and cast(list[int], offsets)[-1] != input.size(0): + raise ValueError( + 'offsets need to have the input size in the end when the flag "include_last_offset" is on!' + ) + + if max_norm and max_norm <= 0.0: + raise ValueError('"max_norm" must be larger than zero!') + + if not isinstance(weight._sharding_spec, ChunkShardingSpec): + raise ValueError("Only ChunkShardingSpec supported for ShardedTensor ops!") + if len(weight.local_shards()) != 1: + raise ValueError("Only one local shard supported!") + + +def _handle_col_wise_sharding( + input, + world_size, + weight, + local_shard, + offsets, + per_sample_weights, + mode, + max_norm, + norm_type, + padding_idx, + pg, +): + """ + Entry-point function to handle the logic of col-wise sharding of weight + for embeddingBag. (Detailed explanations of the logic can be found in + the comment for sharded_embedding_bag.) + + Args: + input: list of ID used for lookup and aggregation. + world_size: number of ranks. + weight: sharded weight tensor. + local_shard: col-wise shared local weight used for lookup. + offsets: list of start positions of each bag for 1D input. + per_sample_weights: weights for weighted sum mode. + mode: aggregation method of each bag. + max_norm: If given, each embedding vector with norm larger + than max_norm is renormalized to have norm max_norm. + Note: this will modify weight in-place. + norm_type: The p in the p-norm to compute for the max_norm option. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + Note that the embedding vector at padding_idx is + excluded from the reduction. + pg: process group. + + Return: + output: final result of lookup and aggregation. + local_shard: col-wise shared local weight used for lookup. + If max_norm, this will be the renormed weight. + """ + # allgather the special input of embedding bag first. + ( + gathered_inputs, + gathered_per_sample_weights, + gathered_offsets, + ) = _all_gather_embedding_bag_input(input, per_sample_weights, offsets, pg) + + if max_norm is not None: + # max_norm changes the weight in-place + local_shard = _handle_max_norm_col_wise( + max_norm, norm_type, local_shard, input, world_size, gathered_inputs, pg + ) + + output = _handle_col_wise_sharding_base( + torch.nn.functional.embedding_bag, + 1, + input, + world_size, + weight, + local_shard, + pg, + gathered_inputs, + mode=mode, + gathered_per_sample_weights=gathered_per_sample_weights, + gathered_offsets=gathered_offsets, + padding_idx=padding_idx, + ) + return (output, local_shard) + + +def _handle_row_wise_sharding( + input, + world_size, + weight, + local_shard, + offsets, + per_sample_weights, + mode, + max_norm, + norm_type, + padding_idx, + rank, + pg, +): + """ + Entry-point function to handle the logic of row-wise sharding of weight + for embeddingBag. (Detailed explanations of the logic can be found in + the comment for sharded_embedding_bag.) + + Args: + input: list of ID used for lookup and aggregation. + world_size: number of ranks. + weight: sharded weight tensor. + local_shard: row-wise shared local weight used for lookup. + offsets: list of start positions of each bag for 1D input. + per_sample_weights: weights for weighted sum mode. + mode: aggregation method of each bag. + max_norm: If given, each embedding vector with norm larger + than max_norm is renormalized to have norm max_norm. + Note: this will modify weight in-place. + norm_type: The p in the p-norm to compute for the max_norm option. + padding_idx: If specified, the entries at padding_idx do + not contribute to the gradient; therefore, the embedding + vector at padding_idx is not updated during training, + i.e. it remains as a fixed "pad". + Note that the embedding vector at padding_idx is + excluded from the reduction. + rank: # of cuda process. + pg: process group. + + Returns: + gathered_output: final result of lookup and aggregation. + """ + if input.dim() > 1 and per_sample_weights is None: + # allgather the inputs first for non Replicated Tensor. + gather_inp = _all_gather_base_input(input, pg) + else: + ( + gathered_inputs, + gathered_per_sample_weights, + gathered_offsets, + ) = _all_gather_embedding_bag_input(input, per_sample_weights, offsets, pg) + cat_dim = 0 if input.dim() != 1 else -1 + gather_inp = torch.cat(gathered_inputs, dim=cat_dim) + if per_sample_weights is not None: + per_sample_weights = torch.cat(gathered_per_sample_weights, dim=cat_dim) + offset_add = 0 if input.dim() > 1 else input.size(0) + if offsets is not None: + offsets_list = torch.cat( + [gathered_offsets[i] + (offset_add * i) for i in range(pg.size())], + dim=cat_dim, + ) + + # Mask the input according to sharding spec. + lookup_input, padding_local, padding_row = _handle_row_wise_mask( + gather_inp, padding_idx, weight, world_size, rank + ) + if mode == "max": + padding_row[:] = -float("Inf") + + # When input is a large tensor, the value of weight is changed. + # This is a walk-around for now. GH issue: #81717. + if max_norm is not None: + torch.nn.functional.embedding_bag( + torch.unique(lookup_input)[:-1], + local_shard, + offsets=torch.tensor([0], device=local_shard.device, dtype=torch.long), + mode=mode, + per_sample_weights=None, + max_norm=max_norm, + norm_type=norm_type, + padding_idx=padding_local, + ) + max_norm = None + result = torch.nn.functional.embedding_bag( + lookup_input, + torch.cat([local_shard, padding_row]), + offsets=offsets_list if offsets is not None else offsets, # type: ignore[possibly-undefined] + mode=mode if mode != "mean" else "sum", + per_sample_weights=per_sample_weights, + max_norm=max_norm, + norm_type=norm_type, + padding_idx=padding_local, + ) + + op = ReduceOp.SUM if mode != "max" else ReduceOp.MAX + # TODO: Make the result a PartialTensor and move the logic below there. + local_shards = result.chunk(pg.size()) + result = reduce_scatter( + torch.empty_like(local_shards[0]), + list(local_shards), + op=op, + group=pg, + ) + + # For Mean, we cannot do the division until very end because the sum of means + # not equal to the mean of sum. (Divisor is different) + if mode == "mean": + if input.dim() > 1: + padding_idx = padding_idx if padding_idx is not None else -1 + split_sizes = torch.sum( + torch.ne(input, padding_idx), dim=-1, dtype=local_shard.dtype + ) + else: + split_sizes = torch.cat( + ( + offsets[1 : offsets.size(0)] - offsets[0:-1], + (input.size(0) - offsets[-1]).unsqueeze(0), + ), + dim=-1, + ) + return torch.div(result, split_sizes.unsqueeze(1)) + + # Return the appropriate local result. + return result + + +def _all_gather_embedding_bag_input(input, per_sample_weights, offsets, pg): + """ + In case we need to gather input and all other parameters of embeddingBag + ops, we need to stack all input together to perform ``all_gather`` + collective communication just once. + + Note that since offsets does not share the same size as input and + is always smaller than input, we resize it during the communication. + + Args: + input: tensor to be applied op on. + per_sample_weights: weights for weighted sum mode. + offsets: when input is 1D. offsets determines the starting + index position of each bag (sequence) in input. + pg: process group. + + Returns: + gathered_inputs: list of input tensor gathered from each rank. + gathered_per_sample_weights: list of per_sample_weights from each rank. + gathered_offsets: list of offsets from each rank. + """ + input_to_gather = [input] + if per_sample_weights is not None: + input_to_gather.append(per_sample_weights) + if offsets is not None: + input_to_gather.append(offsets.clone().resize_(input.size())) + gathered_inputs = all_gather(torch.stack(input_to_gather), group=pg) + + gathered_per_sample_weights = None + if per_sample_weights is not None: + gathered_per_sample_weights = [t[1] for t in gathered_inputs] + gathered_offsets = None + if offsets is not None: + idx = 2 if per_sample_weights is not None else 1 + gathered_offsets = [ + t[idx].resize_(offsets.size()).to(offsets.dtype) for t in gathered_inputs + ] + gathered_inputs = [t[0].to(input.dtype) for t in gathered_inputs] + return gathered_inputs, gathered_per_sample_weights, gathered_offsets diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..24de2628c0ab9ceb89fa28b52753a421b58b56c2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__init__.py @@ -0,0 +1,21 @@ +# Keep old package for BC purposes, this file should be removed once +# everything moves to the `torch.distributed._shard` package. +import sys +import warnings + +import torch +from torch.distributed._shard.sharded_tensor import * # noqa: F403 + + +with warnings.catch_warnings(): + warnings.simplefilter("always") + warnings.warn( + "`torch.distributed._sharded_tensor` will be deprecated, " + "use `torch.distributed._shard.sharded_tensor` instead", + DeprecationWarning, + stacklevel=2, + ) + +sys.modules["torch.distributed._sharded_tensor"] = ( + torch.distributed._shard.sharded_tensor +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..48b1a16c4805a1a166ee6090be45202965034ff0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_sharded_tensor/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..c74dd3633e0f5e8436b844fd2d14f3bdb00635b7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__init__.py @@ -0,0 +1,22 @@ +# Keep old package for BC purposes, this file should be removed once +# everything moves to the `torch.distributed._shard` package. +import sys +import warnings + +import torch +from torch.distributed._shard.sharding_spec import * # noqa: F403 + + +with warnings.catch_warnings(): + warnings.simplefilter("always") + warnings.warn( + "`torch.distributed._sharding_spec` will be deprecated, " + "use `torch.distributed._shard.sharding_spec` instead", + DeprecationWarning, + stacklevel=2, + ) + +import torch.distributed._shard.sharding_spec as _sharding_spec + + +sys.modules["torch.distributed._sharding_spec"] = _sharding_spec diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..203a7e742b3f3fa3113d38895c76585d6d51f01c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_sharding_spec/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_state_dict_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_state_dict_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..225860cd2a83fdaedb1958e8529ea93b2a947a1e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_state_dict_utils.py @@ -0,0 +1,829 @@ +# mypy: allow-untyped-defs +import copy +import io +import math +import weakref +from collections.abc import Callable, Mapping, MutableMapping +from typing import Any, cast, NamedTuple, TYPE_CHECKING + +import torch +import torch.cuda._pin_memory_utils as pin_memory_utils +import torch.distributed as dist +import torch.nn.functional as F +from torch.distributed._functional_collectives import AsyncCollectiveTensor + + +if dist.is_available() or TYPE_CHECKING: + from torch.distributed import distributed_c10d + from torch.distributed._shard.sharded_tensor import ShardedTensor + from torch.distributed.tensor import distribute_tensor, DTensor, Replicate + from torch.distributed.tensor._utils import compute_local_shape_and_global_offset + + +def _identity_func( + obj: torch.Tensor, + pg: dist.ProcessGroup | None, + device: torch.device | None, + companion_obj: Any, +) -> torch.Tensor: + return obj + + +def _all_gather_sharded_tensor( + sharded_tensor: "ShardedTensor", + pg: dist.ProcessGroup | None = None, + device: torch.device | None = None, +) -> torch.Tensor: + if pg is None: + pg = distributed_c10d._get_default_group() + world_size = dist.get_world_size(pg) + shards = sharded_tensor.local_shards() + dim_0_size = sharded_tensor.size()[0] # type: ignore[index] + tensor_numel = sharded_tensor.size().numel() # type: ignore[union-attr] + chunk_size = math.ceil(dim_0_size / world_size) * tensor_numel // dim_0_size + pg_device = ( + distributed_c10d._get_pg_default_device(pg) if device is None else device + ) + if shards: + local_tensor = shards[0].tensor.flatten() + if local_tensor.device.type != pg_device.type: + local_tensor = local_tensor.to(pg_device) + num_padding = chunk_size - local_tensor.numel() + if num_padding > 0: + local_tensor = F.pad(local_tensor, [0, num_padding]) + else: + local_tensor = torch.zeros( + chunk_size, dtype=sharded_tensor.dtype, device=pg_device + ) + + tensor = torch.empty( + chunk_size * world_size, + dtype=local_tensor.dtype, + device=pg_device, + ) + dist.all_gather_single(tensor, local_tensor, group=pg) + + tensor = tensor.narrow(0, 0, tensor_numel).reshape(sharded_tensor.size()) + return tensor + + +class CompanionMismatch(Exception): + pass + + +def _iterate_state_dict( + iter_object: Any, + sharded_tensor_func: Callable, + dtensor_func: Callable, + tensor_func: Callable, + *, + pg: dist.ProcessGroup | None = None, + device: torch.device | None = None, + cpu_offload: bool = False, + companion_obj: Any = None, + ranks_only: tuple[int, ...] = (), + type_check: bool = True, + non_blocking: bool = True, +) -> dict[str, Any]: + """Iterate through the state dict, applying the given functions to each tensor type. + + Args: + iter_object (Any): the target state_dict. + sharded_tensor_func (Callable): the function to apply to ShardedTensor + dtensor_func (Callable): the function to apply to DTensor + tensor_func (Callable): the function to apply to Tensor + pg (Optional[dist.ProcessGroup]): process group passed to tensor functions + device (Optional[torch.device]): device passed to tensor functions + cpu_offload (bool): whether to offload the tensors to CPU memory. This option is ignored + if a companion_obj is supplied. + companion_obj (Any): A companion object to the state dict. If this object + is supplied, we attempt to copy the tensor to the companion object. + ranks_only (Tuple[int, ...]): if this tuple is empty, all ranks will + have the same state_dicts. Otherwise only ranks that in ``ranks_only`` + have the same state_dicts. Other ranks will get empty state_dicts. + type_check (bool): check if the instance data type is a supported type + that can be saved by DCP. The current supported data types are + torch.Tensor, DTensor, int, float, str, list, dict, None. + non_blocking (bool): whether to use non-blocking copy when copying to the companion object. + """ + # TODO: should we use pytree? + cpu_device = torch.device("cpu") + if isinstance(iter_object, ShardedTensor): + ret = sharded_tensor_func(iter_object, pg, device, companion_obj) + elif isinstance(iter_object, DTensor): + ret = dtensor_func(iter_object, pg, device, companion_obj) + elif isinstance(iter_object, torch.Tensor): + ret = tensor_func(iter_object, pg, device, companion_obj) + elif ( + isinstance(iter_object, (int, float, str, bytes, io.BytesIO)) + or iter_object is None + ): + ret = iter_object + elif isinstance(iter_object, dict): + if companion_obj is not None and ( + not isinstance(companion_obj, dict) + or set(companion_obj.keys()) != set(iter_object.keys()) + ): + msg = ( + "" + if isinstance(companion_obj, dict) + else f"{set(companion_obj.keys())=} {set(iter_object.keys())=}" + ) + raise CompanionMismatch(msg) + + ret = { + key: _iterate_state_dict( + value, + sharded_tensor_func, + dtensor_func, + tensor_func, + pg=pg, + device=device, + cpu_offload=cpu_offload, + companion_obj=companion_obj[key] if companion_obj is not None else None, + ranks_only=ranks_only, + type_check=type_check, + non_blocking=non_blocking, + ) + for key, value in iter_object.items() + } + elif isinstance(iter_object, (list, tuple)): + if companion_obj is not None and ( + not isinstance(companion_obj, (list, tuple)) + or len(companion_obj) != len(iter_object) + ): + raise CompanionMismatch + + ret = [ + _iterate_state_dict( + v, + sharded_tensor_func, + dtensor_func, + tensor_func, + pg=pg, + device=device, + cpu_offload=cpu_offload, + companion_obj=companion_obj[idx] if companion_obj is not None else None, + ranks_only=ranks_only, + type_check=type_check, + non_blocking=non_blocking, + ) + for idx, v in enumerate(iter_object) + ] + if isinstance(iter_object, tuple): + ret = tuple(ret) + elif not type_check: + ret = copy.deepcopy(iter_object) + else: + raise ValueError(f"Unexpected value type {type(iter_object)}") + + if not ranks_only or dist.get_rank(pg) in ranks_only: + if isinstance(ret, torch.Tensor): + if cpu_offload and companion_obj is None: + ret = ret.to(cpu_device) + + if companion_obj is not None: + if isinstance(companion_obj, DTensor): + if not isinstance(ret, DTensor): + raise AssertionError( + "ret must be a DTensor when companion_obj is a DTensor" + ) + companion_obj._local_tensor.copy_( + ret._local_tensor, non_blocking=non_blocking + ) + elif isinstance(companion_obj, ShardedTensor): + if not isinstance(ret, ShardedTensor): + raise AssertionError( + "ret must be a ShardedTensor when companion_obj is a ShardedTensor" + ) + for idx, shard in enumerate(companion_obj.local_shards()): + shard.tensor.copy_( + ret.local_shards()[idx].tensor, non_blocking=non_blocking + ) + else: + companion_obj.copy_(ret, non_blocking=non_blocking) + ret = companion_obj + else: + ret = {} if isinstance(ret, dict) else None + + # pyrefly: ignore [bad-return] + return ret + + +def _gather_state_dict( + state_dict: dict[str, Any], + *, + pg: dist.ProcessGroup | None = None, + device: torch.device | None = None, + cpu_offload: bool = False, + ranks_only: tuple[int, ...] = (), + type_check: bool = True, +) -> dict[str, Any]: + """ + Given a state_dict, this API gathers all the ShardedTensors or DTensors in + the state_dict. + + + Args: + state_dict (Dict[str, Any]): the target sharded state_dict. + pg (Optional[dist.ProcessGroup]): the process group that is used to + gather ShardedTensor. Note that gathering a DTensor will use + the DeviceMesh. So this argument will be ignored when gathering a + DTensor. + device: (Optional[torch.device]): the device that is used to + perform allgather for ShardedTensor. Note that gathering a DTensor + will use the DeviceMesh. So this argument will be ignored when + gathering a DTensor. + cpu_offload (bool): whether to offload the tensors to CPU memory. The + default value is False. + ranks_only: (Tuple[int, ...]): if this tuple is empty, all ranks will + have the same state_dicts. Otherwise only ranks that in ``ranks_only`` + have the same state_dicts. Other ranks will get empty state_dicts. + type_check: (bool): check if the instance data type is a supported type + that can be saved by DCP. The current supported data types are + torch.Tensor, DTensor, int, float, str, list, dict, None. + + Returns: + The gathered state dictionary. + """ + + def sharded_tensor_func(value, pg, device, companion_obj): + # ShardedTensor does not seem to record the original device type. + # So if the tensor is moved to CPU, we won't know the original type. + # As a result, we have to rely on the user to tell us the correct one. + cpu_device = torch.device("cpu") + output_tensor = _all_gather_sharded_tensor(value, pg, device) + local_shard_device = ( + value.local_shards()[0].tensor.device + if value.local_shards() + else cpu_device + ) + if output_tensor.device != local_shard_device: + value = output_tensor.to(local_shard_device) + else: + value = output_tensor + return value + + def dtensor_func(value, pg, device, companion_obj): + if value.device != value.device_mesh.device_type: + value = value.to(value.device_mesh.device_type) + # FSDP all_gather: [Shard(0)] -> [Replicate()] + # HSDP all_gather: [Replicate(), Shard(0)] -> [Replicate(), Replicate()] + # 2D FSDP + TP all_gather: + # - [Shard(0), Shard(n)] -> [Replicate(), Replicate()] + # - [Shard(0), Replicate()] -> [Replicate(), Replicate()] + placements = [Replicate() for _ in value.placements] + value = value.redistribute( + device_mesh=value.device_mesh, + placements=placements, + ) + # Call `wait()` to force the tensor to be synchronous with respect + # to the main stream. + # See the discussion in https://github.com/pytorch/pytorch/pull/117799. + value = value.to_local() + if isinstance(value, AsyncCollectiveTensor): + value = value.wait() + return value + + return _iterate_state_dict( + state_dict, + sharded_tensor_func, + dtensor_func, + _identity_func, + pg=pg, + device=device, + cpu_offload=cpu_offload, + ranks_only=ranks_only, + type_check=type_check, + ) + + +def _offload_state_dict_to_cpu( + state_dict: dict[str, Any], + *, + ranks_only: tuple[int, ...] = (), + type_check: bool = True, +) -> dict[str, Any]: + """ + Given a state_dict, this API offload all the tensors to CPU memory. + + Args: + state_dict (Dict[str, Any]): the target state_dict. + pg (Optional[dist.ProcessGroup]): the process group that is used to + gather ShardedTensor. Note that gathering a DTensor will use + the DeviceMesh. So this argument will be ignored when gathering a + DTensor. + ranks_only: (Tuple[int, ...]): if this tuple is empty, all ranks will + have the same state_dicts. Otherwise only ranks that in ``ranks_only`` + have the same state_dicts. Other ranks will get empty state_dicts. + type_check: (bool): check if the instance data type is a supported type + that can be saved by DCP. The current supported data types are + torch.Tensor, DTensor, int, float, str, list, dict, None. + + Returns: + The gathered state dictionary. + """ + + ret = _iterate_state_dict( + state_dict, + _identity_func, + _identity_func, + _identity_func, + pg=None, + device=None, + cpu_offload=True, + ranks_only=ranks_only, + type_check=type_check, + ) + return ret + + +@torch.no_grad() +def _copy_state_dict( + state_dict: dict[str, Any], + copy_state_dict: dict[str, Any], + non_blocking: bool = False, + type_check: bool = True, +) -> dict[str, Any]: + """ + Copies all tensors in a given state dict into a different state_dict with the + same structure. Additionally, a copied state dict with the same value references + is returned. Editing the keys on this state dict will not affect the + passed in copy_state_dict (but the value references are the same). + + .. warning:: + It is expected by this function that state_dict and copy_state_dict share + the same structure and data types. + + .. warning:: + The current supported data types are + torch.Tensor, DTensor, int, float, str, list, dict, None. + + Args: + state_dict (Dict[str, Any]): the target state_dict. + copy_state_dict (Dict[str, Any]): + The state dict we are copying into. This state_dict must have exactly + the same structure as the source `state_dict`. + non_blocking: (bool): Whether copy ops should be performed asynchronously + type_check (bool): check if the instance data type is a supported type + that can be saved by DCP. The current supported data types are + torch.Tensor, DTensor, int, float, str, list, dict, None. + + Returns: + State Dict copy + """ + + return _iterate_state_dict( + state_dict, + _identity_func, + _identity_func, + _identity_func, + pg=None, + device=None, + cpu_offload=False, + ranks_only=(), + companion_obj=copy_state_dict, + type_check=type_check, + non_blocking=non_blocking, + ) + + +@torch.no_grad() +def _create_cpu_state_dict( + state_dict: dict[str, Any], pin_memory: bool = False, share_memory: bool = False +) -> dict[str, Any]: + """ + Given a state_dict, create another state_dict with the same structure and elements. + However, all tensors in the returned state_dict are new tensors on CPU. These + tensors can be placed on pin_memory or share_memory based on the provided arguments. + + .. warning:: + Setting both `pin_memory` and `share_memory` to True significantly increases the + latency of this method because of the nuances which require us to register memory + as pinned directly as opposed to relying on the pin_memory cache allocator. This + option should only be used for long lived tensors which are required to be shared. + This is not the case as long as at least one of `pin_memory` or `share_memory` is + set to False. + + """ + + def tensor_func( + obj: torch.Tensor, + pg: dist.ProcessGroup | None, + device: torch.device | None, + _: Any, + ) -> torch.Tensor: + if len(obj.size()) == 0: + return torch.tensor(0, dtype=obj.dtype) + + # sometimes, a tensor might have non-zero size and 0 numel. In this case, pinning memory will fail + # so we take a best guess at how to replicate the tensor below to maintain symmetry in the returned + # state dict. + if obj.numel() == 0 or obj.data_ptr() == 0: + t = torch.zeros_like(obj, device="cpu") + if share_memory: + t = t.share_memory_() + return t + + if share_memory: + t = torch.empty(*tuple(obj.size()), dtype=obj.dtype) + t = t.share_memory_() + if pin_memory: + pin_memory_utils.pin_memory(t.data_ptr(), t.numel() * t.element_size()) + weakref.finalize(t, pin_memory_utils.unpin_memory, t.data_ptr()) + + return t + elif pin_memory: + return torch.empty(*tuple(obj.size()), dtype=obj.dtype).pin_memory() + else: + return torch.empty(*tuple(obj.size()), dtype=obj.dtype) + + def dtensor_func( + obj: DTensor, + pg: dist.ProcessGroup | None, + device: torch.device | None, + _: Any, + ) -> DTensor: + if len(obj.size()) == 0: + return obj + + if obj.device != torch.device("cpu"): + ret = cast(DTensor, obj.to(device="cpu")) + else: + ret = copy.deepcopy(obj) + ret._local_tensor = tensor_func(ret._local_tensor, pg, device, None) + return ret + + def sharded_tensor_func( + obj: ShardedTensor, + pg: dist.ProcessGroup | None, + device: torch.device | None, + _: Any, + ) -> ShardedTensor: + if not obj.local_shards(): + return obj + + if obj.device != torch.device("cpu"): + ret = obj.to(device="cpu") + else: + ret = copy.deepcopy(obj) + + for shards in ret.local_shards(): + shards.tensor = tensor_func(shards.tensor, pg, device, None) + + return ret + + ret = _iterate_state_dict( + state_dict, + sharded_tensor_func, + dtensor_func, + tensor_func, + pg=None, + device=None, + cpu_offload=False, + ranks_only=(), + type_check=False, + ) + return ret + + +def _check_state_dict_similarity( + state_dict: dict[str, Any], + compared_state_dict: dict[str, Any], +) -> bool: + """ + Given two state_dicts, check if the structures are the same. And + if a [key, tensor] pair exist in one state_dict there must be + a corresponding pair, [key, other_tensor], in the other state_dict, + where tensor and other_tensor have the same size and dtype. + + Return the check result. + """ + + def tensor_func( + obj: torch.Tensor, + pg: dist.ProcessGroup | None, + device: torch.device | None, + companion_obj: Any, + ) -> torch.Tensor: + if companion_obj.dtype != obj.dtype or companion_obj.size() != obj.size(): + raise CompanionMismatch + return obj + + try: + _iterate_state_dict( + state_dict, + _identity_func, + _identity_func, + tensor_func, + pg=None, + device=None, + cpu_offload=False, + ranks_only=(), + companion_obj=compared_state_dict, + type_check=False, + ) + except CompanionMismatch: + return False + + return True + + +class _TensorInfo(NamedTuple): + size: torch.Size + dtype: torch.dtype + + +def _broadcast_tensors( + full_state_dict: dict[str, Any], + local_state_dict: dict[str, Any], + keys: list[str], + device: torch.device, + pg: dist.ProcessGroup | None = None, +) -> None: + if pg is None: + pg = dist.distributed_c10d._get_default_group() + pg_device = ( + device + if device.type in {pg_device.type for pg_device in pg._device_types} + else pg._device_types[0] + ) + + tensors: list[torch.Tensor] = [] + for key in keys: + if dist.get_rank() == 0: + full_state = full_state_dict[key] + if not isinstance(full_state, torch.Tensor): + raise AssertionError("full_state must be a torch.Tensor") + full_tensor = full_state.detach().to(pg_device) + else: + tensor_info = full_state_dict[key] + full_tensor = torch.empty( + size=tensor_info.size, + device=pg_device, + dtype=tensor_info.dtype, + ) + + tensors.append(full_tensor) + + if (local_state := local_state_dict.get(key)) is None: + continue + + local_state_dict[key] = ( + (local_state, full_tensor) + if isinstance(local_state, DTensor) + else full_tensor + ) + + if len(tensors) > 1: + dist._broadcast_coalesced(pg, tensors, 500, 0) + else: + dist.broadcast(tensors[0], src=0, group=pg) + + if pg_device != device: + for key, full_tensor in zip(keys, tensors): + if (local_state := local_state_dict.get(key)) is not None: + local_state_dict[key] = ( + (local_state[0], full_tensor.to(device)) + if ( + isinstance(local_state, tuple) + and isinstance(local_state[0], DTensor) + ) + else full_tensor.to(device) + ) + + _distribute_tensors(local_state_dict, keys, device, pg) + + +def _distribute_tensors( + local_state_dict: dict[str, Any], + keys: list[str], + device: torch.device, + pg: dist.ProcessGroup | None = None, +) -> None: + if pg is None: + pg = dist.distributed_c10d._get_default_group() + for key in keys: + _local_state = local_state_dict.get(key) + if _local_state is None or torch.is_tensor(_local_state): + continue + + local_state = _local_state[0] + full_tensor = _local_state[1] + + shape, offset = compute_local_shape_and_global_offset( + full_tensor.shape, local_state.device_mesh, local_state.placements + ) + slices = [ + slice(cur_offset, cur_offset + cur_shape) + for cur_shape, cur_offset in zip(shape, offset) + ] + if local_state.is_meta: + # Use .clone() here rather than view to clone and return only the sliced portion, minimizing memory access and cost. + local_tensor = full_tensor[tuple(slices)].detach().clone() + # TODO: currently, we cannot handle strided sharding if the dp dimension is not even. For example, + # one of the case that is not yet supported is when placements = (Shard(0), _StridedShard(0, sf=2)). + ret = DTensor.from_local( + local_tensor, + local_state.device_mesh, + local_state.placements, + shape=local_state.shape, + stride=local_state.stride(), + ) + else: + ret = local_state + # Copy full_tensor[slices] into local_state.to_local() to reduce memory footprint. + ret.to_local().copy_(full_tensor[tuple(slices)]) + local_state_dict[key] = ret + + +def _broadcast_state_dict( + full_state_dict: dict[str, Any], + local_state_dict: dict[str, Any], + device: torch.device, + pg: dist.ProcessGroup | None = None, + strict: bool = False, + cpu_offload: bool = False, +) -> None: + # Broadcast from rank0's `full_state_dict` to all ranks' `local_state_dict`. + # If strict is True, any keys in `local_state_dict` but not in `full_state_dict` + # will be removed from `local_state_dict`. + ret = {} + if dist.get_rank() == 0: + for key, value in full_state_dict.items(): + if not torch.is_tensor(value): + ret[key] = value + elif value.dim() == 0: + ret[key] = value.cpu() + else: + ret[key] = _TensorInfo(value.size(), value.dtype) + + broadcast_list = [ret] + dist.broadcast_object_list(broadcast_list, src=0, group=pg) + ret = broadcast_list[0] + # Gather values + keys = [] + local_state_dict_keys = set(local_state_dict.keys()) + global_keys = set() + for key, value in ret.items(): + global_keys.add(key) + if not isinstance(value, _TensorInfo): + if key in local_state_dict: + local_state_dict[key] = value + continue + + if dist.get_rank() == 0: + ret[key] = full_state_dict[key] + + keys.append(key) + # Broadcast every tensor to avoid OOM for now. + if len(keys) >= 1: + _broadcast_tensors(ret, local_state_dict, keys, device, pg) + if cpu_offload: + for key in keys: + local_state_dict[key] = local_state_dict[key].cpu() + keys.clear() + + if strict: + if missing_keys := (local_state_dict_keys - global_keys): + for key in missing_keys: + local_state_dict.pop(key) + + if keys: + _broadcast_tensors(ret, local_state_dict, keys, device, pg) + if cpu_offload: + for key in keys: + local_state_dict[key] = local_state_dict[key].cpu() + + +def _distribute_state_dict( + full_state_dict: dict[str, Any], + local_state_dict: dict[str, Any], + device: torch.device, + pg: dist.ProcessGroup | None = None, +) -> None: + # Full_state_dict = True, broadcast_from_rank0 = False here. Each rank has + # full_state_dict. Skip the broadcast in ``_broadcast_state_dict`` and + # distribute tensors in each rank + for key, value in full_state_dict.items(): + if key not in full_state_dict: + continue + if not torch.is_tensor(value): + local_state_dict[key] = value + elif value.dim() == 0: + local_state_dict[key] = value.cpu() + else: + if not isinstance(value, torch.Tensor): + raise AssertionError("value must be a torch.Tensor") + local_state = local_state_dict.get(key) + if local_state is None: + continue + elif isinstance(local_state, DTensor): + local_state_dict[key] = distribute_tensor( + value.detach().to(device), + local_state.device_mesh, + local_state.placements, + ) + else: + local_state_dict[key] = value.detach().to(device) + + +# These APIs are from torch.distributed.checkpoint. +# TODO: We should consolidate the code here as some not all modules can depend on +# DCP. +PATH_ITEM = str | int +OBJ_PATH = tuple[PATH_ITEM, ...] +FLATTEN_MAPPING = dict[str, OBJ_PATH] +STATE_DICT_TYPE = dict[str, Any] +CONTAINER_TYPE = MutableMapping[PATH_ITEM, Any] + + +def _traverse_state_dict( + state_dict: STATE_DICT_TYPE, + visitor: Callable[[OBJ_PATH, Any], None], +) -> None: + """ + Invoke ``visitor`` for each value recursively in ``state_dict``. + Mapping, list, and tuple will be flattened and other value types are treated + as the terminal values and will invoke ``visitor``. + """ + + def _traverse_obj(path: OBJ_PATH, value: Any) -> None: + if isinstance(value, Mapping): + for k, v in value.items(): + _traverse_obj(path + (str(k),), v) + elif isinstance(value, (list, tuple)): + for i, v in enumerate(value): + _traverse_obj(path + (i,), v) + else: + visitor(path, value) + + for key, value in state_dict.items(): + _traverse_obj((str(key),), value) + + +def _flatten_state_dict( + state_dict: STATE_DICT_TYPE, +) -> tuple[STATE_DICT_TYPE, FLATTEN_MAPPING]: + """ + Flatten ``state_dict`` made of nested dicts and lists into a top level dictionary. + + Use ``unflatten_state_dict`` to revert this process. + Returns: + A tuple with the flatten state_dict and a mapping from original to new state_dict. + N.B. The new keys are derived from the object paths, joined by dot. + For example: ``{ 'a': {'b':...}}`` results in the key `a.b`. + """ + flattened: STATE_DICT_TYPE = {} + mappings: FLATTEN_MAPPING = {} + + def flat_copy(path: OBJ_PATH, value: Any) -> None: + new_fqn = ".".join(map(str, path)) + if new_fqn in flattened: + raise ValueError(f"duplicated flatten key {new_fqn}") + flattened[new_fqn] = value + mappings[new_fqn] = path + + _traverse_state_dict(state_dict, flat_copy) + return flattened, mappings + + +def _set_element(root_dict: STATE_DICT_TYPE, path: OBJ_PATH, value: Any) -> None: + """Set ``value`` in ``root_dict`` along the ``path`` object path.""" + cur_container = cast(CONTAINER_TYPE, root_dict) + + def extend_list(lst: list[Any], idx: int) -> None: + while len(lst) <= idx: + lst.append(None) + + for i in range(1, len(path)): + prev_key = path[i - 1] + key = path[i] + def_val: CONTAINER_TYPE | list[Any] = {} if type(key) is str else [] + + if isinstance(cur_container, Mapping): + cur_container = cast( + CONTAINER_TYPE, cur_container.setdefault(prev_key, def_val) + ) + else: + # pyrefly: ignore [bad-argument-type] + extend_list(cur_container, prev_key) + if cur_container[prev_key] is None: + cur_container[prev_key] = def_val + cur_container = cur_container[prev_key] + + key = path[-1] + if type(key) is int: + extend_list(cast(list[Any], cur_container), key) + + cur_container[key] = value + + +def _unflatten_state_dict( + state_dict: STATE_DICT_TYPE, mapping: FLATTEN_MAPPING +) -> STATE_DICT_TYPE: + """Restore the original nested state_dict according to ``mapping`` and the flattened ``state_dict``.""" + nested: STATE_DICT_TYPE = {} + for key, value in state_dict.items(): + _set_element(nested, mapping[key], value) + return nested diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..7edd0ef75f4d17f6ca26a2f4c1c4adefc30d1a23 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__init__.py @@ -0,0 +1,2683 @@ +from __future__ import annotations + +import math +import os +import socket +import uuid +from collections.abc import Callable, Generator +from contextlib import contextmanager +from datetime import timedelta +from enum import Enum +from typing import Any, Literal +from typing_extensions import deprecated + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.distributed_c10d as c10d +from torch._C._autograd import DeviceType +from torch._C._distributed_c10d import _SymmetricMemory, Work as _Work +from torch._prims_common import make_contiguous_strides_for +from torch.utils._triton import has_triton + + +_group_name_to_store: dict[str, c10d.Store] = {} + + +@deprecated( + "`enable_symm_mem_for_group` is deprecated. There is no need to call this function anymore.", + category=FutureWarning, +) +def enable_symm_mem_for_group(group_name: c10d.GroupName) -> None: + """ + Enables symmetric memory for a process group. + + Args: + group_name (str): the name of the process group. + """ + if group_name in _group_name_to_store: + return + + group = c10d._resolve_process_group(group_name) + global_ranks = sorted(c10d._world.pg_group_ranks[group].keys()) + # Different subgroups with the same name should use different stores + global_ranks_str = "_".join(map(str, global_ranks)) + store = c10d.PrefixStore( + f"symmetric_memory-{global_ranks_str}", + c10d._get_process_group_store(group), + ) + _group_name_to_store[group_name] = store + _SymmetricMemory.set_group_info( + group_name, + group.rank(), + group.size(), + store, + ) + + +_is_test_mode: bool = False +_mocked_group_names: set[str] | None = None + + +@contextmanager +def _test_mode(group_names: set[str] | None = None) -> Generator[None, None, None]: + """ + Forces ``is_symm_mem_enabled_for_group()`` to return ``True`` and the ops + defined in the ``symm_mem`` namespace to use fallback implementations. + + The context manager is not thread safe. + """ + global _is_test_mode + global _mocked_group_names + prev = _is_test_mode + prev_group_names = _mocked_group_names + try: + _is_test_mode = True + _mocked_group_names = group_names + yield + finally: + _is_test_mode = prev + _mocked_group_names = prev_group_names + + +def is_symm_mem_enabled_for_group(group_name: c10d.GroupName) -> bool: + """ + Check if symmetric memory is enabled for a process group. + + Args: + group_name (str): the name of the process group. + """ + if _is_test_mode: + return _mocked_group_names is None or group_name in _mocked_group_names + return group_name in _group_name_to_store + + +_group_name_to_workspace_tensor: dict[str, torch.Tensor | None] = {} + + +def get_symm_mem_workspace( + group_name: c10d.GroupName, min_size: int +) -> _SymmetricMemory: + """ + Get the symmetric memory workspace associated with the process group. If + ``min_size`` is greater than the workspace associated with ``group_name``, + the workspace will be re-allocated and re-rendezvous'd. + + Args: + group_name (str): the name of the process group. + min_size (int): the size requirement for the workspace in bytes. + + Returns: + _SymmetricMemory: the symmetric memory workspace associated with the + group. + """ + tensor = _group_name_to_workspace_tensor.get(group_name) + size = tensor.numel() * tensor.element_size() if tensor is not None else 0 + if tensor is None or size < min_size: + if ( + torch.accelerator.is_available() + and torch.accelerator.current_stream().is_capturing() + ): + curr_size = 0 if tensor is None else tensor.numel() * tensor.element_size() + raise RuntimeError( + f"get_symm_mem_workspace(): the requested size ({min_size} bytes) " + "is greater than the size of the currently allocated workspace " + f"({curr_size} bytes). It's currently not possible to expand the " + "workspace size during graph capture. Please invoke " + f'`get_symm_mem_workspace(group_name="{group_name}", ' + f'min_size="{min_size}")` before initiating the graph capture ' + "and try again." + ) + tensor = _SymmetricMemory.empty_strided_p2p( + (max(size, min_size),), + [1], + torch.uint8, + torch.device(torch.accelerator.current_device_index()), + group_name, + ) + _group_name_to_workspace_tensor[group_name] = tensor + return _SymmetricMemory.rendezvous(tensor) + + +_backend_streams: dict[int, torch.Stream] = {} + + +def _get_backend_stream(priority: int = 0) -> torch.Stream: + if priority not in _backend_streams: + _backend_streams[priority] = torch.Stream(priority=priority) + return _backend_streams[priority] + + +def _pipelined_multi_all_gather_and_consume( + shard: list[torch.Tensor], + shard_consumer: Callable[[list[torch.Tensor], int], None], + ag_out: list[torch.Tensor], + group_name: c10d.GroupName, + ag_out_needed: bool = True, +) -> None: + """ + Perform the following logic with micro-pipelined computation and + communication: + + gathered = [ + all_gather_single(x, gather_dim=0, group=group) + for x in shard + ] + + shards = [[] for _ in range(group_size)] + for x in ag_out: + for i, y in enumerate(x.chunk(group_size)): + shards[i].append(y) + + for src_rank, shard in enumerate(shards): + shard_consumer(shard, src_rank) + """ + p2p_workspace_size_req = 0 + for x in shard: + p2p_workspace_size_req += x.numel() * x.element_size() + symm_mem = get_symm_mem_workspace(group_name, min_size=p2p_workspace_size_req) + group_size = symm_mem.world_size + rank = symm_mem.rank + + symm_mem.barrier(channel=0) + backend_stream = _get_backend_stream() + backend_stream.wait_stream(torch.accelerator.current_stream()) + + for x, y in zip(shard, ag_out): + if not x.is_contiguous(): + raise AssertionError( + "_pipelined_all_gather_and_consume: all tensors " + "in `shard` must be contiguous" + ) + if not y.is_contiguous(): + raise AssertionError( + "_pipelined_all_gather_and_consume: all tensors " + "in `ag_out` must be contiguous" + ) + if x.shape[0] * group_size != y.shape[0]: + raise AssertionError + if x.shape[1:] != y.shape[1:]: + raise AssertionError + + def copy_shard(dst: list[torch.Tensor], src: list[torch.Tensor]) -> None: + for d, s in zip(dst, src): + d.copy_(s) + + def get_p2p_bufs(remote_rank: int) -> list[torch.Tensor]: + offset_bytes = 0 + bufs = [] + for x in shard: + buf = symm_mem.get_buffer( + remote_rank, + x.shape, + x.dtype, + storage_offset=offset_bytes // x.element_size(), + ) + bufs.append(buf) + offset_bytes += buf.numel() * buf.element_size() + return bufs + + local_p2p_bufs = get_p2p_bufs(rank) + + # shards[i] => shard from rank i + shards: list[list[torch.Tensor]] = [[] for _ in range(group_size)] + for x in ag_out: + for i, y in enumerate(x.chunk(group_size)): + shards[i].append(y) + + # Parallelization strategy: after each rank copies its shard into its local + # p2p buffer, every rank issues independent p2p copy -> shard_consumer + # sequences to two streams. In addition to computation/communication + # overlapping, the strategy allows for computation/computation overlapping, + # greatly reducing quantization inefficiency. + # + # Notation: + # - "mv" for the copy to local buffer + # - "cp" for p2p copies + # - "b" for barriers + # + # Constraints: + # - The GPU scheduler may or may not overlap "mv" with the first shard_consumer. + # - "cp" from different streams cannot overlap. + # + # Ideal scenario 0 - "mv" overlaps with the first shard_consumer: + # + # stream 0: [ shard_consumer ][ cp ][ shard_consumer ] + # stream 1: [ mv ][b][ cp ][ shard_consumer ] + # + # Ideal scenario 1 - "mv" is scheduled before the first shard_consumer: + # + # stream 0: [ shard_consumer ][ cp ][ shard_consumer ] + # stream 1: [ mv ][b][ cp ][ shard_consumer ] + # + # Suboptimal scenario 0 - "mv" is scheduled after the first shard_consumer: + # + # stream 0: [ shard_consumer ] [ cp ][ shard_consumer ] + # stream 1: [ mv ][b][ cp ][ shard_consumer ] + # + # Suboptimal scenario 0 - "b" is scheduled after the first shard_consumer: + # + # stream 0: [ shard_consumer ] [ cp ][ shard_consumer ] + # stream 1: [ mv ] [b][ cp ][ shard_consumer ] + # + # We haven't yet figured out a way to ensure "mv" and "b" are either + # overlapped with or scheduled before the first shard_consumer. Thus, to + # prevent suboptimal scenarios, we are giving up the chance to overlap "mv" + # and "b" with the first shard_consumer for now. + copy_shard(dst=local_p2p_bufs, src=shard) + symm_mem.barrier(channel=1) + backend_stream.wait_stream(torch.accelerator.current_stream()) + + # At this point, all ranks have copied their local shard to + # their local p2p buffer. Each rank can now copy and consume + # remote shards. + shard_consumer(shard, rank) + + for step in range(1, group_size): + if step % 2 == 0: + stream = torch.accelerator.current_stream() + else: + stream = backend_stream + remote_rank = (step + rank) % group_size + remote_p2p_bufs = get_p2p_bufs(remote_rank) + with stream: + copy_shard(dst=shards[remote_rank], src=remote_p2p_bufs) + shard_consumer(shards[remote_rank], remote_rank) + + if ag_out_needed: + # Copy from input to the all-gather output. Opportunistically overlap + # it with the last shard_consumer. + if group_size % 2 == 0: + stream = torch.accelerator.current_stream() + else: + stream = backend_stream + with stream: + copy_shard(dst=shards[rank], src=shard) + + torch.accelerator.current_stream().wait_stream(backend_stream) + symm_mem.barrier(channel=0) + + +def _pipelined_all_gather_and_consume( + shard: torch.Tensor, + shard_consumer: Callable[[torch.Tensor, int], None], + ag_out: torch.Tensor, + group_name: c10d.GroupName, + ag_out_needed: bool = True, +) -> None: + """ + Perform the following logic with micro-pipelined computation and + communication: + + ag_out = all_gather_single(shard, gather_dim=0, group=group) + shards = ag_out.chunk(group.size()) + for src_rank, shard in enumerate(shards): + shard_consumer(shard, src_rank) + """ + + def adapter(shard: list[torch.Tensor], rank: int) -> None: + shard_consumer(shard[0], rank) + + _pipelined_multi_all_gather_and_consume( + [shard], + adapter, + [ag_out], + group_name, + ag_out_needed, + ) + + +def _pipelined_produce_and_all2all( + chunk_producer: Callable[[int, torch.Tensor], None], + output: torch.Tensor, + group_name: c10d.GroupName, + out_chunk_dim: int = 0, +) -> None: + """ + Perform the following logic with micro-pipelined computation and + communication: + + chunks = [ + chunk_producer(dst_rank, chunks[dst_rank]) + for dst_rank in range(group_size): + ] + dist.all_to_all_single(output=output, input=torch.cat(chunks)) + """ + out_chunks = output.chunk( + c10d._get_group_size_by_name(group_name), dim=out_chunk_dim + ) + p2p_workspace_size_req = out_chunks[0].numel() * out_chunks[0].element_size() * 2 + symm_mem = get_symm_mem_workspace(group_name, min_size=p2p_workspace_size_req) + group_size = symm_mem.world_size + rank = symm_mem.rank + + symm_mem.barrier(channel=0) + backend_stream = _get_backend_stream() + backend_stream.wait_stream(torch.accelerator.current_stream()) + + def get_p2p_buf(rank: int, idx: int) -> torch.Tensor: + if idx not in (0, 1): + raise AssertionError + offset = 0 if idx == 0 else out_chunks[0].numel() + return symm_mem.get_buffer( + rank, out_chunks[0].shape, out_chunks[0].dtype, offset + ) + + # Prepare two local p2p buffers, so that a remote rank can pull the result + # of step [i] in one p2p buffer while the local rank can compute the + # result of step [i+1] and write it directly the other p2p buffer. + local_p2p_buf_0 = get_p2p_buf(rank, 0) + local_p2p_buf_1 = get_p2p_buf(rank, 1) + + for step in range(1, group_size): + remote_rank = (rank - step) % group_size + if step % 2 == 0: + stream = torch.accelerator.current_stream() + p2p_buf = local_p2p_buf_1 + remote_p2p_buf = get_p2p_buf(remote_rank, 1) + else: + stream = backend_stream + p2p_buf = local_p2p_buf_0 + remote_p2p_buf = get_p2p_buf(remote_rank, 0) + with stream: + # Parallelization strategy: every rank issues independent compute + # -> barrier -> p2p copy sequences on two streams. In addition to + # computation/communication overlapping, the strategy allows for + # computation/computation overlapping, greatly reducing + # quantization inefficiency. + # + # Ideally, stream activities would look like this ("b" for + # barriers, "cp" for p2p copies): + # + # [rank 0] + # stream 0: [ chunk_producer ][b][ cp ][ chunk_producer ][b][ cp ] + # stream 1: [ chunk_producer ][b][ cp ][ chunk_producer ][b][ cp ] + # + # [rank 1] + # stream 0: [ chunk_producer ][b][ cp ][ chunk_producer ][b][ cp ] + # stream 1: [ chunk_producer ][b][ cp ][ chunk_producer ][b][ cp ] + # + # Note that the barriers synchronize streams with the same ID + # across ranks. They don't synchronize streams on the same rank. + # + # Since the work on both streams is independent, there's no + # guarantee that the chunk_producer from stream 0 or stream 1 will + # be scheduled first. If there is a scheduling mismatch across + # ranks, the barrier forces all ranks to wait for the slowest. + # + # When scheduling mismatches occur among ranks, the stream + # activities might look like this (note that p2p copies from + # different streams cannot overlap with each other): + # + # [rank 0] + # stream 0: [ chunk_producer ][b ][ cp ][ chunk_producer ][b ][ cp ] + # stream 1: [ chunk_producer ][b] [ cp ][ chunk_producer ][b] [ cp ] + # + # [rank 1] + # stream 0: [ chunk_producer ][b] [ cp ][ chunk_producer ][b] [ cp ] + # stream 1: [ chunk_producer ][b ][ cp ][ chunk_producer ][b ][ cp ] + # + # To prevent this, we need to ensure that the chunk_producer on + # stream 1 gets scheduled first on every rank. Without access to + # the underlying kernels, CUDA offers no API to control the + # scheduling order of two independent, overlapping kernels. Our + # solution is to issue a small sleep kernel in stream 0. The sleep + # duration is insignificant, but having an extra task in stream 0 + # will almost guarantee that the chunk_producer on stream 1 gets + # scheduled first. Once the first chunk_producer is scheduled in + # the correct order, there's very little room for the scheduling + # order of subsequent kernels to be inconsistent across ranks. + if step == 2 and torch.cuda.is_available(): + torch.cuda._sleep(100) + chunk_producer((rank + step) % group_size, p2p_buf) + symm_mem.barrier(channel=step % 2) + out_chunks[remote_rank].copy_(remote_p2p_buf) + # The local P2P buffer can only be overwritten by the next + # chunk_producer after all peers have finished reading from it. + symm_mem.barrier(channel=step % 2) + + # If the sleep wasn't issued in the above loop, do it now. + if group_size == 2 and torch.cuda.is_available(): + torch.cuda._sleep(100) + + chunk_producer(rank, out_chunks[rank]) + torch.accelerator.current_stream().wait_stream(backend_stream) + symm_mem.barrier(channel=0) + + +def reduce_partials( + partials: torch.Tensor, + *, + dim: int, + reduce_op: str, + output_dtype: torch.dtype, + group_size: int, +) -> torch.Tensor: + if output_dtype in (torch.float16, torch.bfloat16): + normalized_dim = dim if dim >= 0 else dim + partials.dim() + # The common RS layout is [rank, local_M, N]. Fuse the final rank + # reduction into one kernel so low-precision partials accumulate in fp32 + # and round only once when written to the output dtype. + if normalized_dim == 0: + reduced = triton_reduce_partials_first_dim( + partials, + reduce_op=reduce_op, + output_dtype=output_dtype, + group_size=group_size, + ) + if reduced is not None: + return reduced + if reduce_op == "sum": + return torch.sum(partials, dim=dim) + if reduce_op == "avg": + return torch.mean(partials, dim=dim) + if reduce_op == "sum": + return torch.sum(partials, dim=dim) + if reduce_op == "avg": + return torch.mean(partials, dim=dim) + raise ValueError("reduce_op must be sum or avg") + + +def triton_reduce_partials_first_dim( + partials: torch.Tensor, + *, + reduce_op: str, + output_dtype: torch.dtype, + group_size: int, +) -> torch.Tensor | None: + if ( + partials.device.type != "cuda" + or not has_triton() + or not partials.is_contiguous() + or partials.dim() < 2 + or partials.shape[0] != group_size + or output_dtype not in (torch.float16, torch.bfloat16) + or partials.dtype not in (torch.float16, torch.bfloat16) + ): + return None + + out = partials.new_empty(partials.shape[1:], dtype=output_dtype) + shard_elems = out.numel() + block = 1024 + import triton + + from torch.distributed._symmetric_memory._triton_kernels import ( + reduce_partials_first_dim_kernel, + ) + + grid = (triton.cdiv(shard_elems, block),) + reduce_partials_first_dim_kernel[grid]( + partials, + out, + shard_elems, + group_size, + reduce_op == "avg", + BLOCK=block, + ) + return out + + +lib = torch.library.Library("symm_mem", "DEF") +lib.define( + "fused_all_gather_matmul(" + "Tensor A, Tensor[] Bs, int gather_dim, str group_name, *, bool return_A = True) -> (Tensor?, Tensor[])", + tags=[torch._C.Tag.needs_fixed_stride_order], +) +lib.define( + "fused_all_gather_scaled_matmul(" + "Tensor A, Tensor[] Bs, Tensor A_scale, Tensor[] B_scales, " + "int gather_dim, str group_name, " + "Tensor?[] biases, " + "Tensor?[] result_scales, " + "ScalarType?[] out_dtypes, " + "bool[] use_fast_accum) -> (Tensor, Tensor[])", + tags=[torch._C.Tag.needs_fixed_stride_order], +) +lib.define( + "fused_matmul_reduce_scatter(Tensor A, Tensor B, str reduce_op, int scatter_dim, str group_name) -> Tensor", + tags=[torch._C.Tag.needs_fixed_stride_order], +) +lib.define( + "fused_scaled_matmul_reduce_scatter(" + "Tensor A, Tensor B, Tensor A_scale, Tensor B_scale, " + "str reduce_op, int orig_scatter_dim, int scatter_dim_after_maybe_reshape, str group_name, SymInt[]? output_shape, " + "Tensor? bias = None, " + "Tensor? result_scale = None, " + "ScalarType? out_dtype = None, " + "bool use_fast_accum = False) -> Tensor", + tags=[torch._C.Tag.needs_fixed_stride_order], +) +lib.define("_low_contention_all_gather(Tensor tensor, str group_name) -> Tensor") +lib.define( + "_low_contention_reduce_scatter(Tensor tensor, str reduce_op, str group_name) -> Tensor" +) +# Copy-engine multicast low-contention all-gather that allocates symm_mem output. +lib.define( + "_low_contention_all_gather_ce_multicast(Tensor tensor, str group_name) -> Tensor" +) +# Out variant for preallocated symm_mem output, including CUDA graph capture. +lib.define( + "_low_contention_all_gather_ce_multicast_out(" + "Tensor tensor, str group_name, Tensor(a!) out) -> Tensor(a!)" +) + +lib.define("get_remote_tensors(Tensor x, str group_name) -> Tensor[]") +""" +Given a local tensor and a group name, return a tuple of tensors that are +symmetric on other devices. The returned tensors are ordered by rank IDs. The +length of the tuple equals to the size of the group. + +Note: this API works only when `world_within_direct_access()` returns True, i.e. +only when the group is within NVLink domain or similar. It does not work across +network interfaces. +""" + + +@torch.library.impl(lib, "get_remote_tensors", "CUDA") +@torch.library.impl(lib, "get_remote_tensors", "XPU") +def _get_remote_tensors_default( + local: torch.Tensor, group_name: c10d.GroupName +) -> tuple[torch.Tensor, ...]: + hdl = rendezvous(local, group_name) + if hdl is None: + raise ValueError("Tensor is not allocated from Symmetric Memory") + + return tuple( + hdl.get_remote_tensor(peer, local.size(), local.dtype) + for peer in range(hdl.world_size) + ) + + +@torch.library.impl(lib, "get_remote_tensors", "Meta") +def _get_remote_tensors_meta( + local: torch.Tensor, group_name: c10d.GroupName +) -> tuple[torch.Tensor, ...]: + group = c10d._resolve_process_group(group_name) + return tuple(torch.empty_like(local) for _ in range(group.size())) + + +class _ScaleMode(Enum): + UNSCALED = "unscaled" + TENSOR_WISE = "tensor-wise" + ROW_WISE_SHARDED = "row-wise-sharded" + ROW_WISE_REPLICATED = "row-wise-replicated" + + +def _check_and_verify_fp8_all_gather_scale_mode( + shard: torch.Tensor, scale: torch.Tensor | None, gather_dim: int, group_size: int +) -> _ScaleMode: + full_shape = list(shard.shape) + full_shape[gather_dim] *= group_size + + if scale is None: + return _ScaleMode.UNSCALED + elif scale.shape[:-1] == shard.shape[:-1] and scale.shape[-1] == 1: + # Row-wise scaling + # + # NOTE: when the last dim of both A_shard and A_scale is one, we can't + # tell if A_scale is replicated tensor-wise scale or sharded row-wise + # scale. Treating it as row-wise scaling for safety. + return _ScaleMode.ROW_WISE_SHARDED + elif scale.numel() == 1: + return _ScaleMode.TENSOR_WISE + elif list(scale.shape[:-1]) == full_shape[:-1]: + return _ScaleMode.ROW_WISE_REPLICATED + else: + raise ValueError( + "Invalid scale shape for fp8 all-gather " + f"(shard shape: {shard.shape}, scale shape: {scale.shape})" + ) + + +def _fused_all_gather_matmul_impl( + mm_out_op: torch._ops.OpOverload, + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + A_scale: torch.Tensor | None, + kwargs_list: list[dict[str, Any]], + out_dtypes: list[torch.dtype | None], + gather_dim: int, + group_name: c10d.GroupName, + return_A: bool, +) -> tuple[torch.Tensor | None, list[torch.Tensor]]: + if A_shard.dim() < 2: + raise ValueError("A_shard must be a matrix") + for B in Bs: + if B.dim() != 2: + raise ValueError("B must be a matrix") + if len(out_dtypes) != len(Bs): + raise ValueError("len(out_types) must be the same as len(Bs)") + if len(kwargs_list) != len(Bs): + raise ValueError("len(kwargs_list) must be the same as len(Bs)") + if gather_dim < 0 or gather_dim >= A_shard.dim(): + raise ValueError("Invalid gather_dim") + + group = c10d._resolve_process_group(group_name) + + if gather_dim == A_shard.ndim - 1 or gather_dim == -1: + return _fused_all_gather_matmul_last_gather_dim_impl( + mm_out_op, + A_shard, + Bs, + A_scale, + kwargs_list, + out_dtypes, + gather_dim, + group_name, + return_A, + ) + + # Move the gather_dim to the front and flatten the tensor into a 2D matrix. + # The flattened tensor doesn't need to be contiguous (for computation + # efficiency), as _pipelined_all_gather_and_consume guarantees that shards + # passed to shard_consumer are contiguous. + A_shard_flat = A_shard.movedim(gather_dim, 0) + leading_dims = [group.size()] + list(A_shard_flat.shape[:-1]) + A_shard_flat = A_shard_flat.flatten(0, -2) + + # Helper function for reverting the above transformation + def unflatten(t: torch.Tensor) -> torch.Tensor: + return t.view(*leading_dims, -1).flatten(0, 1).movedim(0, gather_dim) + + A_flat = A_shard_flat.new_empty( + A_shard_flat.shape[0] * group.size(), + A_shard_flat.shape[1], + ) + + outputs = [ + A_flat.new_empty(A_flat.shape[0], B.shape[1], dtype=out_dtype or B.dtype) + for B, out_dtype in zip(Bs, out_dtypes) + ] + output_shards = [output.chunk(group.size()) for output in outputs] + + scale_mode = _check_and_verify_fp8_all_gather_scale_mode( + shard=A_shard, scale=A_scale, gather_dim=gather_dim, group_size=group.size() + ) + + # Computing block-wise matmul along the first dim of A + if scale_mode == _ScaleMode.ROW_WISE_SHARDED: + if A_scale is None: + raise AssertionError + A_scale_shard = A_scale.movedim(gather_dim, 0).flatten(0, -2) + A_scale_flat = A_scale_shard.new_empty( + A_scale_shard.shape[0] * group.size(), + A_scale_shard.shape[1], + ) + + def row_wise_sharded_consumer(shard: list[torch.Tensor], rank: int) -> None: + for idx, (B, kwargs) in enumerate(zip(Bs, kwargs_list)): + mm_out_op( + shard[0], + B, + scale_a=shard[1], + **kwargs, + out=output_shards[idx][rank], + ) + + _pipelined_multi_all_gather_and_consume( + [A_shard_flat, A_scale_shard], + row_wise_sharded_consumer, + [A_flat, A_scale_flat], + group_name, + return_A, + ) + elif scale_mode == _ScaleMode.ROW_WISE_REPLICATED: + if A_scale is None: + raise AssertionError + A_scale_shards = ( + A_scale.movedim(gather_dim, 0).flatten(0, -2).chunk(group.size()) + ) + + def row_wise_replicated_consumer(shard: torch.Tensor, rank: int) -> None: + for idx, (B, kwargs) in enumerate(zip(Bs, kwargs_list)): + mm_out_op( + shard, + B, + scale_a=A_scale_shards[rank], + **kwargs, + out=output_shards[idx][rank], + ) + + _pipelined_all_gather_and_consume( + A_shard_flat, + row_wise_replicated_consumer, + A_flat, + group_name, + return_A, + ) + else: + if scale_mode == _ScaleMode.TENSOR_WISE: + if A_scale is None: + raise AssertionError + for kwargs in kwargs_list: + kwargs["scale_a"] = A_scale + else: + if scale_mode != _ScaleMode.UNSCALED: + raise AssertionError + + def default_consumer(shard: torch.Tensor, rank: int) -> None: + for idx, (B, kwargs) in enumerate(zip(Bs, kwargs_list)): + mm_out_op(shard, B, **kwargs, out=output_shards[idx][rank]) + + _pipelined_all_gather_and_consume( + A_shard_flat, + default_consumer, + A_flat, + group_name, + return_A, + ) + + A = unflatten(A_flat) if return_A else None + return A, [unflatten(output) for output in outputs] + + +def _pipelined_all_gather_and_consume_last_dim( + shard: torch.Tensor, + shard_consumer: Callable[[torch.Tensor, int], None], + ag_out: torch.Tensor, + group_name: c10d.GroupName, + ag_out_needed: bool = True, +) -> None: + p2p_workspace_size_req = 0 + p2p_workspace_size_req = shard.numel() * shard.element_size() + symm_mem = get_symm_mem_workspace(group_name, min_size=p2p_workspace_size_req) + group_size = symm_mem.world_size + rank = symm_mem.rank + + symm_mem.barrier(channel=0) + backend_stream = _get_backend_stream() + backend_stream.wait_stream(torch.accelerator.current_stream()) + + def copy_shard(dst: torch.Tensor, src: torch.Tensor) -> None: + dst.copy_(src) + + def get_p2p_buf(remote_rank: int) -> torch.Tensor: + buf = symm_mem.get_buffer( + remote_rank, + shard.shape, + shard.dtype, + ) + return buf + + local_p2p_buf = get_p2p_buf(rank) + + shards = ag_out.chunk(group_size) + + copy_shard(dst=local_p2p_buf, src=shard) + symm_mem.barrier(channel=1) + backend_stream.wait_stream(torch.accelerator.current_stream()) + + # At this point, all ranks have copied their local shard to + # their local p2p buffer. Each rank can now copy and consume + # remote shards. + shard_consumer(shard, rank) + + for step in range(1, group_size): + if step % 2 == 0: + stream = torch.accelerator.current_stream() + else: + stream = backend_stream + remote_rank = (step + rank) % group_size + remote_p2p_buf = get_p2p_buf(remote_rank) + with stream: + copy_shard(dst=shards[remote_rank], src=remote_p2p_buf) + shard_consumer(shards[remote_rank], remote_rank) + + if ag_out_needed: + # Copy from input to the all-gather output. Opportunistically overlap + # it with the last shard_consumer. + if group_size % 2 == 0: + stream = torch.accelerator.current_stream() + else: + stream = backend_stream + with stream: + copy_shard(dst=shards[rank], src=shard) + + torch.accelerator.current_stream().wait_stream(backend_stream) + symm_mem.barrier(channel=0) + + +def _fused_all_gather_matmul_last_gather_dim_impl( + mm_out_op: torch._ops.OpOverload, + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + A_scale: torch.Tensor | None, + kwargs_list: list[dict[str, Any]], + out_dtypes: list[torch.dtype | None], + gather_dim: int, + group_name: c10d.GroupName, + return_A: bool, +) -> tuple[torch.Tensor | None, list[torch.Tensor]]: + group = c10d._resolve_process_group(group_name) + group_size = group.size() + + B_shards = [B.chunk(group.size()) for B in Bs] + + leading_dims = list(A_shard.shape[:-1]) + A_shard_flat = A_shard.flatten(0, -2) + + def unflatten(t: torch.Tensor) -> torch.Tensor: + return t.view(*leading_dims, -1) + + A_flat_out = A_shard_flat.new_empty( + A_shard_flat.shape[0] * group.size(), + A_shard_flat.shape[1], + ) + + outputs = [ + torch.empty( + (A_shard_flat.shape[0], B.shape[1]), + dtype=out_dtype or B.dtype, + device=A_shard.device, + ) + for B, out_dtype in zip(Bs, out_dtypes) + ] + + first = True + events = [torch.Event() for _ in outputs] + + def default_consumer(shard: torch.Tensor, rank: int) -> None: + nonlocal first + for out, event, B_shard, kwargs in zip(outputs, events, B_shards, kwargs_list): + event.wait() + if first: + torch.ops.aten.mm.out(shard, B_shard[rank], **kwargs, out=out) + else: + out.addmm_(shard, B_shard[rank]) + event.record() + + first = False + + _pipelined_all_gather_and_consume_last_dim( + A_shard_flat, + default_consumer, + A_flat_out, + group_name, + return_A, + ) + ret_A = None + if return_A: + # This path is inefficient and will be filtered out at passes stage + # Added only for completeness. + A_split_cat_out_flat = torch.cat(A_flat_out.chunk(group_size), dim=-1) + ret_A = unflatten(A_split_cat_out_flat) + + return ret_A, [unflatten(output) for output in outputs] + + +@torch.library.impl(lib, "fused_all_gather_matmul", "Meta") +def _fused_all_gather_matmul_fallback( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + gather_dim: int, + group_name: c10d.GroupName, + *, + return_A: bool = True, +) -> tuple[torch.Tensor | None, list[torch.Tensor]]: + group_size = c10d._get_group_size_by_name(group_name) + A = torch.ops._c10d_functional.all_gather_into_tensor( + A_shard.contiguous(), group_size, group_name + ) + A = torch.ops._c10d_functional.wait_tensor(A) + if gather_dim == A.ndim - 1 or gather_dim == -1: + A_splits = A.chunk(group_size) + A_mm = torch.cat(A_splits, dim=-1) + res = [torch.matmul(A_mm, B) for B in Bs] + if return_A: + return A_mm, res + else: + return None, res + + A = A.view(group_size, *A_shard.shape).movedim(gather_dim + 1, 1).flatten(0, 1) + res = [torch.matmul(A, B).movedim(0, gather_dim) for B in Bs] + if return_A: + return A.movedim(0, gather_dim), res + else: + return None, res + + +@torch.library.impl(lib, "fused_all_gather_matmul", "CUDA") +@torch.library.impl(lib, "fused_all_gather_matmul", "XPU") +def _fused_all_gather_matmul( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + gather_dim: int, + group_name: c10d.GroupName, + *, + return_A: bool = True, +) -> tuple[torch.Tensor | None, list[torch.Tensor]]: + """ + Perform the following logic with micro-pipelined computation and + communication: + + all_gather_single(A_shard, gather_dim, group_name) @ B + + Optimal stride order for A_shard - if A_shard.movedim(gather_dim, 0) is + contiguous, no extra copy is required for input layout transformation. + Otherwise A_shard needs to be copied once. + """ + if _is_test_mode: + return _fused_all_gather_matmul_fallback( + A_shard, Bs, gather_dim, group_name, return_A=return_A + ) + + if _should_use_fused_all_gather_matmul_native(A_shard, Bs, gather_dim, group_name): + group = c10d._resolve_process_group(group_name) + leading_dims = list(A_shard.shape[:-1]) + leading_dims[0] *= group.size() + A, out = _fused_all_gather_matmul_native( + A_shard.flatten(0, -2), Bs[0], group_name + ) + return A.view(*leading_dims, -1), [out.view(*leading_dims, -1)] + + if _should_use_multimem_all_gather_matmul( + A_shard, gather_dim, group_name, return_A + ): + return None, _multimem_all_gather_matmul(A_shard, Bs, group_name) + + with torch.profiler.record_function("fused_all_gather_matmul"): + return _fused_all_gather_matmul_impl( + torch.ops.aten.mm.out, + A_shard, + Bs, + None, + [{} for B in Bs], + [B.dtype for B in Bs], + gather_dim, + group_name, + return_A, + ) + + +def _should_use_fused_all_gather_matmul_native( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + gather_dim: int, + group_name: c10d.GroupName, +) -> bool: + group = c10d._resolve_process_group(group_name) + local_M = math.prod(A_shard.shape[:-1]) + + return ( + "TORCH_SYMM_MEM_ENABLE_NATIVE_ASYNC_TP" in os.environ + and torch.cuda.is_available() + and A_shard.device.type == "cuda" + and A_shard.is_contiguous() + and gather_dim == 0 + # _async_input_mm requires local_M to be divisible by world_size. + and local_M % group.size() == 0 + # _async_input_mm outperforms the decomposition-based approach when the + # global M is small. + and 2048 < local_M * group.size() <= 4096 + # _async_input_mm only supports a single B. + and len(Bs) == 1 + ) + + +def _fused_all_gather_matmul_native( + A_shard: torch.Tensor, + B: torch.Tensor, + group_name: c10d.GroupName, +) -> tuple[torch.Tensor, torch.Tensor]: + symm_mem = rendezvous(A_shard, group_name) + if symm_mem is None: + symm_mem = get_symm_mem_workspace( + group_name, A_shard.numel() * A_shard.element_size() + ) + symm_mem.barrier() + buf = symm_mem.get_buffer(symm_mem.rank, A_shard.shape, A_shard.dtype) + buf.copy_(A_shard) + A_shard = buf + + rank = symm_mem.rank + world_size = symm_mem.world_size + + current_stream = torch.cuda.current_stream() + backend_stream = _get_backend_stream(priority=-1) + + symm_mem.barrier() + backend_stream.wait_stream(current_stream) + current_stream.wait_stream(backend_stream) + + A = A_shard.new_empty(A_shard.shape[0] * world_size, A_shard.shape[1]) + A_signals = torch.zeros(world_size, dtype=torch.uint32, device=A_shard.device) + A_shards = A.chunk(world_size) + + A_shards[rank].copy_(A_shard) + if not torch.cuda.is_current_stream_capturing(): + _SymmetricMemory.stream_write_value32(A_signals, rank, 1) + else: + _SymmetricMemory.memset32(A_signals, offset=rank, val=1, count=1) + + out = torch.ops.symm_mem._async_input_mm(A, B, A_signals, rank) + for step in range(1, world_size): + src_rank = (rank + step) % world_size + src_buf = symm_mem.get_buffer(src_rank, A_shard.shape, A_shard.dtype) + with backend_stream: + A_shards[src_rank].copy_(src_buf) + if not torch.cuda.is_current_stream_capturing(): + # cuStreamWriteValue32 issues a system level fence before the write + _SymmetricMemory.stream_write_value32(A_signals, src_rank, 1) + else: + _SymmetricMemory.memset32(A_signals, offset=src_rank, val=1, count=1) + + current_stream.wait_stream(backend_stream) + backend_stream.wait_stream(current_stream) + + symm_mem.barrier() + return A, out + + +def _should_use_multimem_all_gather_matmul( + A_shard: torch.Tensor, + gather_dim: int, + group_name: c10d.GroupName, + return_A: bool, +) -> bool: + group = c10d._resolve_process_group(group_name) + local_M = math.prod(A_shard.shape[:-1]) + has_multicast_support = ( + A_shard.device.type == "cuda" + and _SymmetricMemory.has_multicast_support( + DeviceType.CUDA, A_shard.device.index + ) + ) + + return ( + has_multicast_support + and not return_A + and A_shard.is_contiguous() + and gather_dim == 0 + # The heuristic is empirical. We could refine it with a more + # sophisticated perf model. + and local_M * group.size() <= 2048 + ) + + +def _multimem_all_gather_matmul( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + group_name: c10d.GroupName, +) -> list[torch.Tensor]: + group = c10d._resolve_process_group(group_name) + A_shape = torch.Size((A_shard.shape[0] * group.size(), *A_shard.shape[1:])) + symm_mem = get_symm_mem_workspace( + group_name, A_shape.numel() * A_shard.element_size() + ) + A = symm_mem.get_buffer(symm_mem.rank, A_shape, A_shard.dtype) + torch.ops.symm_mem.multimem_all_gather_out(A_shard, group_name, A) + return [torch.matmul(A, B) for B in Bs] + + +@torch.library.impl(lib, "fused_all_gather_scaled_matmul", "Meta") +def _fused_all_gather_scaled_matmul_fallback( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + A_scale: torch.Tensor, + B_scales: list[torch.Tensor], + gather_dim: int, + group_name: c10d.GroupName, + biases: list[torch.Tensor | None], + result_scales: list[torch.Tensor | None], + out_dtypes: list[torch.dtype | None], + use_fast_accum: list[bool], +) -> tuple[torch.Tensor, list[torch.Tensor]]: + out_dtypes = _maybe_convert_scalar_types_to_dtypes(out_dtypes) + + group_size = c10d._get_group_size_by_name(group_name) + A = torch.ops._c10d_functional.all_gather_into_tensor( + A_shard.contiguous(), group_size, group_name + ) + A = torch.ops._c10d_functional.wait_tensor(A) + A = A.view(group_size, *A_shard.shape).movedim(gather_dim + 1, 1).flatten(0, 1) + + scale_mode = _check_and_verify_fp8_all_gather_scale_mode( + shard=A_shard, scale=A_scale, gather_dim=gather_dim, group_size=group_size + ) + if scale_mode == _ScaleMode.ROW_WISE_SHARDED: + A_scale_shard = A_scale + A_scale = torch.ops._c10d_functional.all_gather_into_tensor( + A_scale.contiguous(), group_size, group_name + ) + A_scale = torch.ops._c10d_functional.wait_tensor(A_scale) + A_scale = ( + A_scale.view(group_size, *A_scale_shard.shape) + .movedim(gather_dim + 1, 1) + .flatten(0, -2) + ) + elif scale_mode == _ScaleMode.ROW_WISE_REPLICATED: + A_scale = A_scale.movedim(gather_dim, 0).flatten(0, -2) + else: + if scale_mode != _ScaleMode.TENSOR_WISE: + raise AssertionError + + def scaled_matmul( + A: torch.Tensor, + B: torch.Tensor, + A_scale: torch.Tensor, + B_scale: torch.Tensor, + bias: torch.Tensor | None, + result_scale: torch.Tensor | None, + out_dtype: torch.dtype | None, + use_fast_accum: bool, + ) -> torch.Tensor: + leading_dims = A.shape[:-1] + res = torch.ops.aten._scaled_mm( + A.flatten(0, -2), + B, + A_scale, + B_scale, + bias, + result_scale, + out_dtype=out_dtype, + use_fast_accum=use_fast_accum, + ) + return res.unflatten(0, leading_dims) + + return A.movedim(0, gather_dim), [ + scaled_matmul( + A, B, A_scale, B_scale, bias, result_scale, out_dtype, fast_accum + ).movedim(0, gather_dim) + for B, B_scale, bias, result_scale, out_dtype, fast_accum in zip( + Bs, B_scales, biases, result_scales, out_dtypes, use_fast_accum + ) + ] + + +@torch.library.impl(lib, "fused_all_gather_scaled_matmul", "CUDA") +@torch.library.impl(lib, "fused_all_gather_scaled_matmul", "XPU") +def _fused_all_gather_scaled_matmul( + A_shard: torch.Tensor, + Bs: list[torch.Tensor], + A_scale: torch.Tensor, + B_scales: list[torch.Tensor], + gather_dim: int, + group_name: c10d.GroupName, + biases: list[torch.Tensor | None], + result_scales: list[torch.Tensor | None], + out_dtypes: list[torch.dtype | None], + use_fast_accum: list[bool], +) -> tuple[torch.Tensor, list[torch.Tensor]]: + """ + Perform the following logic with micro-pipelined computation and + communication: + + A = all_gather_single(A_shard, gather_dim, group_name) + leading_dims = A.shape[:-1] + res = torch.ops.aten._scaled_mm(A.flatten(0, -2), B, A_scale, B_scale) + res = res.unflatten(0, leading_dims) + + The input `A_scale` can be tensor-wise, row-wise-sharded or + row-wise-replicated. + + Optimal stride order for `A_shard` - if `A_shard.movedim(gather_dim, 0)` is + contiguous, no extra copy is required for input layout transformation. + Otherwise A_shard needs to be copied once. + """ + out_dtypes = _maybe_convert_scalar_types_to_dtypes(out_dtypes) + + if len(biases) != len(Bs): + raise ValueError("len(biases) must be the same as len(Bs)") + if len(result_scales) != len(Bs): + raise ValueError("len(result_scales) must be the same as len(Bs)") + if len(out_dtypes) != len(Bs): + raise ValueError("len(out_dtypes) must be the same as len(Bs)") + if len(use_fast_accum) != len(Bs): + raise ValueError("len(use_gast_accum_list) must be the same as len(Bs)") + + if _is_test_mode: + return _fused_all_gather_scaled_matmul_fallback( + A_shard, + Bs, + A_scale, + B_scales, + gather_dim, + group_name, + biases, + result_scales, + out_dtypes, + use_fast_accum, + ) + + with torch.profiler.record_function("fused_all_gather_scaled_matmul"): + A, res = _fused_all_gather_matmul_impl( + torch.ops.aten._scaled_mm.out, + A_shard, + Bs, + A_scale, + [ + { + "scale_b": B_scale, + "bias": bias, + "scale_result": result_scale, + "out_dtype": out_dtype, + "use_fast_accum": fast_accum, + } + for B_scale, bias, result_scale, out_dtype, fast_accum in zip( + B_scales, biases, result_scales, out_dtypes, use_fast_accum + ) + ], + out_dtypes, + gather_dim, + group_name, + True, + ) + if A is None: + raise AssertionError + return A, res + + +def make_contiguous_for_perm( + t: torch.Tensor, + perm: list[int], +) -> torch.Tensor: + """ + Restride `t` such that `t.permute(perm)` is contiguous. + """ + inv_perm = [0] * len(perm) + for i, p in enumerate(perm): + inv_perm[p] = i + return t.permute(perm).contiguous().permute(inv_perm) + + +def restride_A_shard_for_fused_all_gather_matmul( + t: torch.Tensor, + gather_dim: int, +) -> torch.Tensor: + """ + Restride the `A_shard` arg of `fused_all_gather_matmul` for optimal perf. + See the doc for `fused_all_gather_matmul` for detail. + """ + perm = list(range(len(t.shape))) + perm.insert(0, perm.pop(gather_dim)) + return make_contiguous_for_perm(t, perm) + + +@torch.library.impl(lib, "fused_matmul_reduce_scatter", "CUDA") +@torch.library.impl(lib, "fused_matmul_reduce_scatter", "XPU") +def _fused_matmul_reduce_scatter( + A: torch.Tensor, + B: torch.Tensor, + reduce_op: str, + scatter_dim: int, + group_name: c10d.GroupName, +) -> torch.Tensor: + """ + Perform the following logic with micro-pipelined computation and + communication: + + reduce_scatter_single(A @ B, reduce_op, scatter_dim, group_name) + + Optimal stride order for A - if A.movedim(scatter_dim, 0) is contiguous, no + extra copy is required for input layout transformation. Otherwise A needs + to be copied once. + """ + if _is_test_mode: + return _fused_matmul_reduce_scatter_fallback( + A, B, reduce_op, scatter_dim, group_name + ) + + with torch.profiler.record_function("fused_matmul_reduce_scatter"): + return _fused_matmul_reduce_scatter_impl( + mm_out_op=torch.ops.aten.mm.out, + A=A, + B=B, + kwargs={}, + out_dtype=A.dtype, + reduce_op=reduce_op, + scatter_dim=scatter_dim, + group_name=group_name, + ) + + +@torch.library.impl(lib, "fused_matmul_reduce_scatter", "Meta") +def _fused_matmul_reduce_scatter_fallback( + A: torch.Tensor, + B: torch.Tensor, + reduce_op: str, + scatter_dim: int, + group_name: c10d.GroupName, +) -> torch.Tensor: + res = funcol.reduce_scatter_single(A @ B, reduce_op, scatter_dim, group_name) + res = funcol.wait_tensor(res) + return res + + +def _fused_matmul_reduce_scatter_impl( + mm_out_op: torch._ops.OpOverload, + A: torch.Tensor, + B: torch.Tensor, + kwargs: dict[str, Any], + out_dtype: torch.dtype | None, + reduce_op: str, + scatter_dim: int, + group_name: c10d.GroupName, +) -> torch.Tensor: + if A.dim() < 2: + raise ValueError("A_shard must be a matrix") + if scatter_dim < 0 or scatter_dim >= A.dim(): + raise ValueError("Invalid gather_dim") + if B.dim() != 2: + raise ValueError("B must be a matrix") + if reduce_op not in ("sum", "avg"): + raise ValueError("reduce_op must be sum or avg") + group = c10d._resolve_process_group(group_name) + out_shape = [*A.shape[:-1], B.shape[1]] + out_shape[scatter_dim] //= group.size() + output_dtype = out_dtype or A.dtype + + if scatter_dim == A.ndim - 1: + B_shards = B.chunk(group.size(), dim=B.ndim - 1) + A_flat = A.flatten(0, -2) + + def _chunk_producer(rank: int, out: torch.Tensor) -> None: + mm_out_op(A_flat, B_shards[rank], **kwargs, out=out) + + leading_dims = list(A.shape[:-1]) + + stacked_partials = torch.empty( + (A_flat.shape[0], B.shape[1]), + dtype=out_dtype or A.dtype, + device=A.device, + ) + + _pipelined_produce_and_all2all( + _chunk_producer, + stacked_partials, + group_name, + out_chunk_dim=1, + ) + + stacked_partials_view = stacked_partials.reshape( + *leading_dims, group.size(), -1 + ) + return reduce_partials( + stacked_partials_view, + dim=-2, + reduce_op=reduce_op, + output_dtype=output_dtype, + group_size=group.size(), + ) + + # Move the scatter_dim to the front and flatten the tensor into a 2D matrix + x = A.movedim(scatter_dim, 0) + leading_dims = [group.size()] + list(x.shape[:-1]) + leading_dims[1] //= group.size() + x = x.flatten(0, -2) + A_shards = x.chunk(group.size()) + + # Computing block-wise matmul along the first dim of A + def chunk_producer(rank: int, out: torch.Tensor) -> None: + mm_out_op(A_shards[rank], B, **kwargs, out=out) + + stacked_partials = x.new_empty(x.shape[0], B.shape[1], dtype=output_dtype) + + _pipelined_produce_and_all2all( + chunk_producer, + stacked_partials, + group_name, + ) + + # Ensures that the transpose and reduction produce contiguous result + # in a single reduction kernel. + return reduce_partials( + stacked_partials.view(*leading_dims, -1) + .movedim(1, scatter_dim + 1) + .movedim(0, scatter_dim), + dim=scatter_dim, + reduce_op=reduce_op, + output_dtype=output_dtype, + group_size=group.size(), + ) + + +@torch.library.impl(lib, "fused_scaled_matmul_reduce_scatter", "CUDA") +@torch.library.impl(lib, "fused_scaled_matmul_reduce_scatter", "XPU") +def _fused_scaled_matmul_reduce_scatter( + A: torch.Tensor, + B: torch.Tensor, + A_scale: torch.Tensor, + B_scale: torch.Tensor, + reduce_op: str, + orig_scatter_dim: int, + scatter_dim_after_maybe_reshape: int, + group_name: c10d.GroupName, + output_shape: list[int], + bias: torch.Tensor | None = None, + result_scale: torch.Tensor | None = None, + out_dtype: torch.dtype | None = None, + use_fast_accum: bool = False, +) -> torch.Tensor: + if _is_test_mode: + return _fused_scaled_matmul_reduce_scatter_fallback( + A, + B, + A_scale, + B_scale, + reduce_op, + orig_scatter_dim, + scatter_dim_after_maybe_reshape, + group_name, + output_shape, + bias, + result_scale, + out_dtype, + use_fast_accum, + ) + with torch.profiler.record_function("fused_scaled_matmul_reduce_scatter"): + return _fused_scaled_matmul_reduce_scatter_impl( + mm_out_op=torch.ops.aten._scaled_mm.out, + A=A, + B=B, + A_scale=A_scale, + kwargs={ + "scale_b": B_scale, + "bias": bias, + "scale_result": result_scale, + "out_dtype": out_dtype, + "use_fast_accum": use_fast_accum, + }, + out_dtype=out_dtype, + reduce_op=reduce_op, + orig_scatter_dim=orig_scatter_dim, + scatter_dim_after_maybe_reshape=scatter_dim_after_maybe_reshape, + group_name=group_name, + output_shape=output_shape, + ) + + +@torch.library.impl(lib, "fused_scaled_matmul_reduce_scatter", "Meta") +def _fused_scaled_matmul_reduce_scatter_fallback( + A: torch.Tensor, + B: torch.Tensor, + A_scale: torch.Tensor, + B_scale: torch.Tensor, + reduce_op: str, + orig_scatter_dim: int, + scatter_dim_after_maybe_reshape: int, + group_name: c10d.GroupName, + output_shape: list[int], + bias: torch.Tensor | None = None, + result_scale: torch.Tensor | None = None, + out_dtype: torch.dtype | None = None, + use_fast_accum: bool = False, +) -> torch.Tensor: + if A_scale.numel() > 1: + if A_scale.shape[:-1] != A.shape[:-1]: + raise ValueError( + "For row-wise scaling, the leading dims of A_scale " + "must match the leading dims of A " + f"(A shape: {A.shape}, A_scale shape: {A_scale.shape})" + ) + A_scale = A_scale.flatten(0, -2).contiguous() + elif A_scale.numel() != 1: + raise ValueError( + "Invalid A_scale shape " + f"(A shape: {A.shape}, A_scale shape: {A_scale.shape})" + ) + + C = torch._scaled_mm( + A.flatten(0, -2).contiguous(), + B, + A_scale, + B_scale, + bias, + result_scale, + out_dtype, + use_fast_accum, + ) + C = C.view(*output_shape[:-1], B.shape[1]) + res = funcol.reduce_scatter_single( + C, + reduce_op, + orig_scatter_dim, # need original scatter dim for 3D+ output tensor here + group_name, + ) + res = funcol.wait_tensor(res) + return res + + +def _fused_scaled_matmul_reduce_scatter_impl( + mm_out_op: torch._ops.OpOverload, + A: torch.Tensor, + B: torch.Tensor, + A_scale: torch.Tensor, + kwargs: dict[str, Any], + out_dtype: torch.dtype | None, + reduce_op: str, + orig_scatter_dim: int, + scatter_dim_after_maybe_reshape: int, + group_name: c10d.GroupName, + output_shape: list[int], +) -> torch.Tensor: + if A.dim() < 2: + raise ValueError("A_shard must be a matrix") + if ( + scatter_dim_after_maybe_reshape < 0 + or scatter_dim_after_maybe_reshape >= A.dim() + ): + raise ValueError("Invalid scatter dim for 2D tensor input to scaled_mm") + if orig_scatter_dim < 0 or orig_scatter_dim >= len(output_shape): + raise ValueError("Invalid scatter dim for 3D+ output tensor") + if B.dim() != 2: + raise ValueError("B must be a matrix") + if reduce_op not in ("sum", "avg"): + raise ValueError("reduce_op must be sum or avg") + output_dtype = out_dtype or A.dtype + + group = c10d._resolve_process_group(group_name) + + # Move scatter to first dim, then shard the tensor along the first dim, so the chunk producer + # can perform matmuls along the first dim. + A_with_scatter_dim_0 = A.movedim(scatter_dim_after_maybe_reshape, 0) + + # To handle case where A is 3D+, reshape to 2D to prepare for mm which requires 2D inputs. + A_2D_with_scatter_dim_0 = A_with_scatter_dim_0.flatten(0, -2) + + # Partition A along the first dim to prepare for sharding across TP process group. + A_shards = A_2D_with_scatter_dim_0.chunk(group.size()) + + # Now that 'A' is sharded along the first dim, we need to update its scale(s) accordingly. + # How we do this depends on if we are using tensorwise scaling, rowwise scaling, or no scaling. + tensorwise_scaling = A_scale is not None and A_scale.numel() == 1 + rowwise_scaling = A_scale is not None and A_scale.numel() > 1 + + # For tensorwise scaling, the scale should be replicated so each shard has a copy. + if tensorwise_scaling: + A_scale_shards = [A_scale] * group.size() + + # For rowwise scaling, we need to move the scatter dim to the first dim to match the + # dim swap of the 'A' tensor. Then we can shard the scales along the first dim, just like + # the 'A' tensor. + elif rowwise_scaling: + if A_scale.shape[:-1] != A.shape[:-1]: + raise ValueError( + "For row-wise scaling, the leading dims of A_scale " + "must match the leading dims of A " + f"(A shape: {A.shape}, A_scale shape: {A_scale.shape})" + ) + A_scale = ( + A_scale.movedim(scatter_dim_after_maybe_reshape, 0) + .contiguous() + .flatten(0, -2) + ) + A_scale_shards = list(A_scale.chunk(group.size())) + # cuBLAS's row-wise kernel requires scales to be aligned to 16 bytes. + # When we slice them we might break this and need to reallocate them. + A_scale_shards = [ + t if t.data_ptr() % 16 == 0 else t.clone() for t in A_scale_shards + ] + else: + raise ValueError("A_scale cannot be none for scaled_mm") + + # Computing block-wise matmul along the first dim of A + def chunk_producer(rank: int, out: torch.Tensor) -> None: + mm_out_op(A_shards[rank], B, scale_a=A_scale_shards[rank], **kwargs, out=out) + + # Stacked partials will be the 2D outputs of the pipelined scaled mm, and will + # have the shape (A_with_scatter_dim_0_tensor.shape[0], B.shape[1]) to align with the formula: + # (a*b,c) @ (c,d) = (a*b,d) + stacked_partials = A_with_scatter_dim_0.new_empty( + A_2D_with_scatter_dim_0.shape[0], B.shape[1], dtype=output_dtype + ) + + # Execute the pipelined mm/scaled_mm. + _pipelined_produce_and_all2all( + chunk_producer, + stacked_partials, + group_name, + ) + + # We now need to transform the *unreduced* stacked 2D partial mm outputs to an *unreduced* 3D+ output, + # then reduce-scatter. To do this, we first need to determine the shape of the unreduced 3D+ output, + # to reshape our stacked partials so we can apply the reduce-scatter. + # + # The *unreduced* 3D+ tensor will have dim 0 = `group_size`, as we have `group_size` instances of + # stacked partial outputs. The next dims will be A's leading dims (sharded along the original scatter dim), + # as it was the left operand of the mm op. We can use -1 as the final dim of the view to populate the rest. + stacked_partials_3D_leading_dims = [group.size()] + list( + # We use A from after the dim swap 0<=>scatter_dim, but before the flatten, + # to get the leading dims of the 3D+ view of stacked partials. + A_with_scatter_dim_0.shape[:-1] + ) + + # The `group_size` leading dim has been prepended to `stacked_partials_3D_leading_dims`, + # to capture the partial output from each rank. We need to divide the sharding/scatter dim + # by the group size. If the original scatter dim was 0, then it is now dim 1 in this + # tensor, since this new `group_size` dim was prepended. + stacked_partial_scatter_dim = orig_scatter_dim if orig_scatter_dim > 0 else 1 + stacked_partials_3D_leading_dims[stacked_partial_scatter_dim] //= group.size() + + # Ensures that the transpose and reduction produce contiguous result + # in a single reduction kernel. + reduced_out = reduce_partials( + # View 2D stacked partials as 3D+ tensor of shape (`group_size`, ...) + stacked_partials.view(*stacked_partials_3D_leading_dims, -1) + # We originally swapped 0<=>scatter_dim_after_maybe_reshape. Now after + # prepending the `group_size` dim, to undo this original swap, we + # must swap 1<=>scatter_dim_after_maybe_reshape+1. + .movedim(1, scatter_dim_after_maybe_reshape + 1), + # Reduce along the `group_size` dim (0). + dim=0, + reduce_op=reduce_op, + output_dtype=output_dtype, + group_size=group.size(), + ) + + # Output shape must be scattered along original scatter dim as well. + output_shape[orig_scatter_dim] //= group.size() + out = reduced_out.view(*output_shape) + return out + + +def restride_A_for_fused_matmul_reduce_scatter( + t: torch.Tensor, + scatter_dim: int, +) -> torch.Tensor: + """ + Restride the `A_shard` arg of `fused_matmul_reduce_scatter` for optimal + perf. See the doc for `fused_matmul_reduce_scatter` for detail. + """ + perm = list(range(len(t.shape))) + perm.insert(0, perm.pop(scatter_dim)) + return make_contiguous_for_perm(t, perm) + + +def _maybe_convert_scalar_types_to_dtypes( + scalar_types: list[Any], +) -> list[torch.dtype | None]: + """ + When a list of `torch.dtype`s is passed through the dispatcher as + `ScalarType[]`, it is converted to a list of scalar type enum values. This + function converts it back to a list of `torch.dtype`s. + """ + # Order defined in https://github.com/pytorch/pytorch/blob/344defc9733a45fee8d0c4d3f5530f631e823196/c10/core/ScalarType.h + _SCALAR_TYPE_TO_DTYPE = { + 0: torch.uint8, + 1: torch.int8, + 2: torch.short, + 3: torch.int, + 4: torch.int64, + 5: torch.half, + 6: torch.float, + 7: torch.double, + 8: torch.complex32, + 9: torch.complex64, + 10: torch.complex128, + 11: torch.bool, + 12: torch.qint8, + 13: torch.quint8, + 14: torch.qint32, + 15: torch.bfloat16, + 16: torch.float8_e5m2, + 17: torch.float8_e4m3fn, + 18: torch.float8_e5m2fnuz, + 19: torch.float8_e4m3fnuz, + } + if any(not isinstance(x, (type(None), int)) for x in scalar_types): + return scalar_types + + dtypes: list[torch.dtype | None] = [] + for scalar_type in scalar_types: + if scalar_type is None: + dtypes.append(scalar_type) + elif scalar_type not in _SCALAR_TYPE_TO_DTYPE: + raise ValueError(f"Unrecognized scalar type {scalar_type}") + else: + dtypes.append(_SCALAR_TYPE_TO_DTYPE[scalar_type]) + return dtypes + + +class Work(_Work): + def __init__(self) -> None: + super().__init__() + self.event = torch.Event() + self.event.record() + + def wait(self, timeout: timedelta = timedelta(seconds=0)) -> bool: + self.event.wait() + return True + + +""" +NOTE [low-contention collectives] +When a collective is overlapped with abundant compute, it makes sense to +prioritize reducing the contention between the collective and the overlapped +compute, even at the cost of a slightly slower collective. + +Common collective implementations (e.g., NCCL without user buffer +registration) optimize for throughput with no ambient compute. However, such +implementations may not be optimal when they are overlapped with compute: +- These implementations typically fuse the entire collective into a single +kernel and reserve SM resources based on the most demanding portion of the +collective, even when a large portion of the collective does not require this +much resource. +- These implementations often use SM-based P2P copy as opposed to copy +engine-based P2P copy. Copy engine-based P2P copy may not have a significant +advantage when there's no ambient compute. However, it may significantly +improve overall resource utilization in the presence of ambient compute. + +When overlapped with intensive compute (e.g., persistent matmul kernels), the +SM-usage of a collective can lead to inefficient overlapping. + +Low-contention collectives achieve their goals with the following strategies: +- Use copy engine-based copy whenever possible. +- Break down portions of a collective with different resource requirements +into multiple kernels. This improves the overlapping efficiency at the cost +of additional launching overhead. +""" + + +@torch.library.impl(lib, "_low_contention_all_gather", "Meta") +def _low_contention_all_gather_meta( + tensor: torch.Tensor, + group_name: c10d.GroupName, +) -> torch.Tensor: + group_size = c10d._get_group_size_by_name(group_name) + return tensor.new_empty(tensor.shape[0] * group_size, *tensor.shape[1:]) + + +@torch.library.impl(lib, "_low_contention_all_gather", "CUDA") +def _low_contention_all_gather( + tensor: torch.Tensor, + group_name: c10d.GroupName, +) -> torch.Tensor: + """ + Performs all-gather with symmetric memory in a low-contention fashion. + + When `tensor` is already in symmetric memory: + - The collective is carried out without using SMs. + - No symmetric memory workspace is required. + + When `tensor` is not in symmetric memory: + - An extra SM-based copy is performed to copy the input data into the + symmetric memory workspace. + - Symmetric memory workspace size requirement: the size of `tensor`. + """ + symm_mem = rendezvous(tensor, group_name) + if symm_mem is not None: + input_is_symm_mem = True + else: + symm_mem = get_symm_mem_workspace( + group_name, tensor.numel() * tensor.element_size() + ) + input_is_symm_mem = False + + rank = symm_mem.rank + world_size = symm_mem.world_size + + output = tensor.new_empty(tensor.shape[0] * world_size, *tensor.shape[1:]) + chunks = output.chunk(world_size) + + _get_backend_stream().wait_stream(torch.accelerator.current_stream()) + with _get_backend_stream(): + if not input_is_symm_mem: + local_buf = symm_mem.get_buffer(rank, tensor.shape, tensor.dtype) + local_buf.copy_(tensor) + # pull + symm_mem.barrier() + for step in range(world_size): + remote_rank = (rank - step) % world_size + src_buf = symm_mem.get_buffer(remote_rank, tensor.shape, tensor.dtype) + chunks[remote_rank].copy_(src_buf) + symm_mem.barrier() + torch._C._distributed_c10d._register_work(output, Work()) + return output + + +def _require_multicast(device_index: int, op_name: str) -> None: + if not _SymmetricMemory.has_multicast_support(DeviceType.CUDA, device_index): + raise RuntimeError(f"{op_name} requires multicast support (NVSwitch / NVLS).") + + +# Use one dedicated signal-pad channel for CE multicast, separate from existing +# low-contention collectives. Its two barriers are ordered on the same stream; +# if a peer has not consumed the first signal yet, the second barrier's CAS put +# spins until that peer slot is cleared. +_CE_MULTICAST_BARRIER_CHANNEL = 3 + + +def _lc_ag_out_shape(tensor: torch.Tensor, world_size: int) -> tuple[int, ...]: + if tensor.dim() == 0: + return (world_size,) + return (tensor.shape[0] * world_size, *tensor.shape[1:]) + + +def _check_lc_ag_out( + tensor: torch.Tensor, + output: torch.Tensor, + world_size: int, + op_name: str, +) -> None: + expected_shape = _lc_ag_out_shape(tensor, world_size) + if tuple(output.shape) != expected_shape: + raise RuntimeError( + f"{op_name}: expected out shape {expected_shape}, " + f"but got {tuple(output.shape)}." + ) + if output.dtype != tensor.dtype: + raise RuntimeError( + f"{op_name}: expected out dtype {tensor.dtype}, but got {output.dtype}." + ) + if output.device != tensor.device: + raise RuntimeError( + f"{op_name}: expected out device {tensor.device}, but got {output.device}." + ) + if not output.is_contiguous(): + raise RuntimeError(f"{op_name}: out must be contiguous.") + + +def _check_lc_signal_pad_capacity(symm_mem: _SymmetricMemory) -> None: + required_bytes = (_CE_MULTICAST_BARRIER_CHANNEL + 1) * symm_mem.world_size * 4 + actual_bytes = _SymmetricMemory.signal_pad_size + if actual_bytes < required_bytes: + raise RuntimeError( + f"low_contention all-gather requires signal_pad_size >= " + f"{required_bytes} bytes for world_size={symm_mem.world_size}, but " + f"the current size is {actual_bytes} bytes." + ) + + +@torch.library.impl(lib, "_low_contention_all_gather_ce_multicast", "Meta") +def _low_contention_all_gather_ce_multicast_meta( + tensor: torch.Tensor, + group_name: c10d.GroupName, +) -> torch.Tensor: + world_size = c10d._get_group_size_by_name(group_name) + return tensor.new_empty(_lc_ag_out_shape(tensor, world_size)) + + +@torch.library.impl(lib, "_low_contention_all_gather_ce_multicast", "CUDA") +def _low_contention_all_gather_ce_multicast( + tensor: torch.Tensor, + group_name: c10d.GroupName, +) -> torch.Tensor: + """Direct-input copy-engine multicast all-gather into a fresh output.""" + device_index = ( + tensor.device.index + if tensor.device.index is not None + else torch.cuda.current_device() + ) + _require_multicast(device_index, "_low_contention_all_gather_ce_multicast") + world_size = c10d._get_group_size_by_name(group_name) + out_shape = _lc_ag_out_shape(tensor, world_size) + if torch.cuda.is_current_stream_capturing(): + raise RuntimeError( + "_low_contention_all_gather_ce_multicast cannot allocate its output " + "during CUDA graph capture. Use " + "_low_contention_all_gather_ce_multicast_out with a preallocated " + "symmetric-memory output." + ) + device = torch.device("cuda", device_index) + with torch.cuda.use_mem_pool(get_mem_pool(device)): + output = torch.empty_strided( + out_shape, + make_contiguous_strides_for(out_shape), + dtype=tensor.dtype, + device=device, + ) + return _low_contention_all_gather_ce_multicast_impl(tensor, group_name, output) + + +@torch.library.impl(lib, "_low_contention_all_gather_ce_multicast_out", "Meta") +def _low_contention_all_gather_ce_multicast_out_meta( + tensor: torch.Tensor, + group_name: c10d.GroupName, + out: torch.Tensor, +) -> torch.Tensor: + world_size = c10d._get_group_size_by_name(group_name) + _check_lc_ag_out(tensor, out, world_size, "_low_contention_all_gather_ce_multicast") + return out + + +@torch.library.impl(lib, "_low_contention_all_gather_ce_multicast_out", "CUDA") +def _low_contention_all_gather_ce_multicast_out( + tensor: torch.Tensor, + group_name: c10d.GroupName, + out: torch.Tensor, +) -> torch.Tensor: + """Direct-input copy-engine multicast all-gather into caller-provided output.""" + world_size = c10d._get_group_size_by_name(group_name) + _check_lc_ag_out(tensor, out, world_size, "_low_contention_all_gather_ce_multicast") + return _low_contention_all_gather_ce_multicast_impl(tensor, group_name, out) + + +def _low_contention_all_gather_ce_multicast_impl( + tensor: torch.Tensor, + group_name: c10d.GroupName, + output: torch.Tensor, +) -> torch.Tensor: + device_index = ( + output.device.index + if output.device.index is not None + else torch.cuda.current_device() + ) + _require_multicast(device_index, "_low_contention_all_gather_ce_multicast") + symm_mem = rendezvous(output, group_name) + if symm_mem is None: + raise RuntimeError( + "ce_multicast output must be allocated from symmetric memory" + ) + _check_lc_signal_pad_capacity(symm_mem) + + rank = symm_mem.rank + shard_bytes = tensor.numel() * tensor.element_size() + + ag_input = tensor if tensor.is_contiguous() else tensor.contiguous() + symm_mem.barrier(channel=_CE_MULTICAST_BARRIER_CHANNEL) + torch.ops.symm_mem.memcpy_to_multicast_( + output, ag_input, rank * shard_bytes, group_name + ) + symm_mem.barrier(channel=_CE_MULTICAST_BARRIER_CHANNEL) + return output + + +@torch.library.impl(lib, "_low_contention_reduce_scatter", "Meta") +def _low_contention_reduce_scatter_meta( + tensor: torch.Tensor, + reduce_op: str, + group_name: c10d.GroupName, +) -> torch.Tensor: + group_size = c10d._get_group_size_by_name(group_name) + return tensor.unflatten(0, (group_size, -1)).mean(dim=0) + + +def _low_contention_reduce_scatter_with_symm_mem_input( + tensor: torch.Tensor, + reduce_op: str, + symm_mem: _SymmetricMemory, +) -> torch.Tensor: + rank = symm_mem.rank + world_size = symm_mem.world_size + + if tensor.shape[0] % world_size != 0: + raise AssertionError + a2a_res = torch.empty_like(tensor) + chunks = a2a_res.chunk(world_size) + + _get_backend_stream().wait_stream(torch.accelerator.current_stream()) + with _get_backend_stream(): + # pull + offline reduction + symm_mem.barrier() + for step in range(world_size): + remote_rank = (rank - step) % world_size + src_buf = symm_mem.get_buffer( + remote_rank, + chunks[0].shape, + chunks[0].dtype, + chunks[0].numel() * rank, + ) + chunks[remote_rank].copy_(src_buf) + symm_mem.barrier() + + ret = a2a_res.unflatten(0, (world_size, -1)) + if reduce_op == "sum": + ret = ret.sum(dim=0) + elif reduce_op == "avg": + ret = ret.mean(dim=0) + else: + raise ValueError(f"reduce_op ({reduce_op}) is not supported") + torch._C._distributed_c10d._register_work(ret, Work()) + return ret + + +def _low_contention_reduce_scatter_with_workspace( + tensor: torch.Tensor, + reduce_op: str, + workspace: _SymmetricMemory, +) -> torch.Tensor: + rank = workspace.rank + world_size = workspace.world_size + + if tensor.shape[0] % world_size != 0: + raise AssertionError + chunks = tensor.chunk(world_size) + + _get_backend_stream().wait_stream(torch.accelerator.current_stream()) + with _get_backend_stream(): + # push + offline reduction + workspace.barrier() + for step in range(world_size): + remote_rank = (rank - step) % world_size + dst_buf = workspace.get_buffer( + remote_rank, chunks[0].shape, chunks[0].dtype, chunks[0].numel() * rank + ) + dst_buf.copy_(chunks[remote_rank]) + workspace.barrier() + + buf = workspace.get_buffer(rank, tensor.shape, tensor.dtype) + ret = buf.unflatten(0, (world_size, -1)) + if reduce_op == "sum": + ret = ret.sum(dim=0) + elif reduce_op == "avg": + ret = ret.mean(dim=0) + else: + raise ValueError(f"reduce_op ({reduce_op}) is not supported") + torch._C._distributed_c10d._register_work(ret, Work()) + return ret + + +@torch.library.impl(lib, "_low_contention_reduce_scatter", "CUDA") +def _low_contention_reduce_scatter( + tensor: torch.Tensor, + reduce_op: str, + group_name: c10d.GroupName, +) -> torch.Tensor: + """ + Performs reduce-scatter with symmetric memory in a low-contention fashion. + + This implementation performs a P2P-based all-to-all followed by an offline + reduction. + + When `tensor` is already in symmetric memory: + - Pull-based all-to-all is used. + - No symmetric memory workspace is required. + + When `tensor` is not in symmetric memory: + - Push-based all-to-all is used. + - Symmetric memory workspace size requirement: the size of `tensor`. + + SM-usage: + - SM-based copy of the rank's own chunk for the all-to-all. + - Reduction on the all-to-all result. + + TODO(yifu): the SM-based copy can be avoided with a list-based reduction + kernel. + """ + symm_mem = rendezvous(tensor, group_name) + if symm_mem is not None: + return _low_contention_reduce_scatter_with_symm_mem_input( + tensor, reduce_op, symm_mem + ) + else: + workspace = get_symm_mem_workspace( + group_name, tensor.numel() * tensor.element_size() + ) + return _low_contention_reduce_scatter_with_workspace( + tensor, reduce_op, workspace + ) + + +@torch.library.impl(lib, "all_to_all_vdev_2d", "Meta") +def _all_to_all_vdev_2d_meta( + input: torch.Tensor, + out: torch.Tensor, + in_splits: torch.Tensor, + out_splits_offsets: torch.Tensor, + group_name: c10d.GroupName, + major_align: int | None = None, +) -> None: + return None + + +@torch.library.impl(lib, "all_to_all_vdev_2d_offset", "Meta") +def _all_to_all_vdev_2d_offset_meta( + input: torch.Tensor, + out: torch.Tensor, + in_splits_offsets: torch.Tensor, + out_splits_offsets: torch.Tensor, + group_name: c10d.GroupName, +) -> None: + return None + + +# ============================================================================= +# User-facing APIs +# ============================================================================= + + +from collections.abc import Sequence +from typing import overload, TYPE_CHECKING, Union + + +if TYPE_CHECKING: + from torch._C._distributed_c10d import ProcessGroup + from torch.types import _device, _dtype, _int + + +_use_implicit_mempool: bool | None = None # type: ignore[assignment] + + +def _should_use_implicit_mempool() -> bool: + r""" + Check if the implicit memory pool should be used for symmetric memory allocations. + + Returns: + bool: True if the implicit memory pool should be used, False otherwise. + + By default, use implicit memory pool for `symm_mem.empty`. Users can + disable this by setting the environment variable `TORCH_SYMMMEM_IMPLICIT_POOL` to `0`. + """ + global _use_implicit_mempool + if _use_implicit_mempool is None: + _use_implicit_mempool = os.getenv("TORCH_SYMMMEM_IMPLICIT_POOL", "1") == "1" + + return _use_implicit_mempool + + +@overload +def empty( + *size: _int, dtype: _dtype | None = None, device: _device | None = None +) -> torch.Tensor: ... + + +@overload +# pyrefly: ignore [inconsistent-overload] +def empty( + size: Sequence[_int], + *, + dtype: _dtype | None = None, + device: _device | None = None, +) -> torch.Tensor: ... + + +def empty( # type: ignore[misc] + *size: Any, + dtype: _dtype | None = None, + device: _device | None = None, +) -> torch.Tensor: + r""" + Similar to :func:`torch.empty()`. The returned tensor can be used by + :func:`torch._distributed._symmetric_memory.rendezvous()` to establish a + symmetric memory tensor among participating processes. + + .. note:: + This is a host-synchronous allocation. Together with + :func:`rendezvous`, it is intended as an initialization-time + operation: allocate a symmetric memory tensor once and reuse it, + rather than allocating in hot code paths. + + Args: + size (int...): a sequence of integers defining the shape of the output tensor. + Can be a variable number of arguments or a collection like a list or tuple. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned tensor. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + device (:class:`torch.device`, optional): the desired device of returned tensor. + Default: if ``None``, uses the current device for the default tensor type + (see :func:`torch.set_default_device`). :attr:`device` will be the CPU + for CPU tensor types, the current CUDA device for CUDA tensor types, + and the current XPU device for XPU tensor types. + """ + if len(size) == 1 and isinstance(size[0], Sequence): + size = tuple(size[0]) + else: + size = tuple(size) + + if dtype is None: + dtype = torch.get_default_dtype() + + if device is None: + device = torch.get_default_device() + else: + device = torch.device(device) + + stride = torch._prims_common.make_contiguous_strides_for(size) + + if _should_use_implicit_mempool() and device.type == "cuda": + # Allocate tensor from an implicit memory pool + mempool = get_mem_pool(device) + # TODO: this path can be made device-agnostic if `use_mem_pool` is + # elevated from torch.cuda to torch accelerator. + with torch.cuda.use_mem_pool(mempool): + return _SymmetricMemory.empty_strided_p2p(size, stride, dtype, device) + else: + return _SymmetricMemory.empty_strided_p2p(size, stride, dtype, device) + + +def _resolve_group_name(group: c10d.GroupName | ProcessGroup) -> c10d.GroupName: + from torch._C._distributed_c10d import ProcessGroup + + if isinstance(group, str): + return c10d.GroupName(group) + if isinstance(group, ProcessGroup): + return group.group_name + raise TypeError(f"unsupported group type: {type(group)}") + + +def rendezvous( + tensor: torch.Tensor, group: c10d.GroupName | ProcessGroup +) -> _SymmetricMemory: + r""" + rendezvous(tensor, group) -> _SymmetricMemory + + Establish a symmetric memory tensor among participating processes. This is + a collective operation. + + .. note:: + This is a host-blocking initialization operation: the first rendezvous + of a tensor performs handle exchange and mapping across processes, and + synchronizes the host with the device. It cannot be ordered onto a + CUDA stream or captured in a CUDA graph. Rendezvous a buffer once and + reuse the returned handle rather than calling this in hot code paths; + subsequent calls on the same tensor return the cached handle. + + Args: + tensor (:class:`torch.Tensor`): the local tensor used to establish the symmetric memory tensor. + It must be allocated via :func:`torch._distributed._symmetric_memory.empty()`. The shape, + dtype, and device type must be identical across all participating processes. + group (Union[str, :class:`torch.distributed.ProcessGroup`]): The group identifying the + participating processes. This can be either a group name or a process group object. + """ + group_name = _resolve_group_name(group) + return _SymmetricMemory.rendezvous(tensor, group_name) + + +def is_nvshmem_available() -> bool: + r""" + is_nvshmem_available() -> bool + + Check if NVSHMEM (CUDA) or rocSHMEM (ROCm) is available in the current + build and usable at runtime. On ROCm, rocSHMEM ``VERSION`` must be at + least 3.3.0 (see ``rocshmem/rocshmem.hpp``). + """ + try: + from torch._C._distributed_c10d import _is_nvshmem_available + except ImportError: + # Not all builds have NVSHMEM support. + return False + + # Check if NVSHMEM is available on current system. + return _is_nvshmem_available() + + +def set_backend(name: Literal["NVSHMEM", "CUDA", "NCCL"]) -> None: + r""" + Set the backend for symmetric memory allocation. This is a global setting + and affects all subsequent calls to + :func:`torch._distributed._symmetric_memory.empty()`. Note that the backend + cannot be changed once a symmetric memory tensor has been allocated. + + Args: + backend (str): the backend for symmetric memory allocation. Currently, + only `"NVSHMEM"`, `"CUDA"`, `"NCCL"` are supported. + """ + _SymmetricMemory.set_backend(name) + + +def get_backend(device: _device) -> str | None: + r""" + Get the backend for symmetric memory allocation for a given device. If not + found, return None. + + Args: + device (`torch.device` or str): the device for which to get the backend. + """ + return _SymmetricMemory.get_backend(torch.device(device)) + + +def get_mempool_allocator(device: _device): # type: ignore[no-untyped-def] + r""" + Get the MemPool allocator for symmetric memory for a given device. + + Args: + device (`torch.device` or str): the device for which to get the MemPool + allocator. + """ + return _SymmetricMemory.get_mempool_allocator(torch.device(device)) + + +def set_signal_pad_size(size: int) -> None: + r""" + Set the signal pad size for future symmetric memory allocations. + + Signal pads are P2P-accessible memory regions used for synchronization in + symmetric memory. This function allows users to configure + the signal pad size to be proportional to their workload requirements. + + .. warning:: + This must be called before any symmetric memory allocations are made. + The size cannot be changed after allocations have been performed. + + Args: + size (int): the signal pad size in bytes. The size should be + proportional to the number of blocks launched and the world size. + + Example:: + + >>> # doctest: +SKIP + >>> # Set a larger signal pad size before any allocations + >>> torch.distributed._symmetric_memory.set_signal_pad_size(1024 * 1024) # 1MB + """ + _SymmetricMemory.signal_pad_size = size + + +def get_signal_pad_size() -> int: + r""" + Get the current signal pad size for symmetric memory allocations. + + Returns the user-configured size if set via :func:`set_signal_pad_size`, + otherwise returns the default size. + + Returns: + int: the signal pad size in bytes. + + Example:: + + >>> # doctest: +SKIP + >>> size = torch.distributed._symmetric_memory.get_signal_pad_size() + >>> print(f"Signal pad size: {size} bytes") + """ + return _SymmetricMemory.signal_pad_size + + +# An internal map from device to the symmetric memory pool for that device. +_symm_mem_pools: dict[_device, torch.cuda.MemPool] = {} + + +def get_mem_pool(device: _device) -> torch.cuda.MemPool: + """ + Get the symmetric memory pool for a given device. If not found, create a new + pool. + + The tensor allocations with this pool must be symmetric across ranks. The + allocated tensors can be used with symmetric operations, for example, + operations defined under `torch.ops.symm_mem`. + + Args: + device (`torch.device` or str): the device for which to get the symmetric memory pool. + + Returns: + `torch.cuda.MemPool`: the symmetric memory pool for the given device. + + Example:: + + >>> # doctest: +SKIP + >>> pool = torch.distributed._symmetric_memory.get_mem_pool("cuda:0") + >>> with torch.cuda.use_mem_pool(pool): + >>> tensor = torch.randn(1000, device="cuda:0") + >>> tensor = torch.ops.symm_mem.one_shot_all_reduce(tensor, "sum", group_name) + + """ + # This function is a wrapper around the `torch.cuda.MemPool` constructor. + # Due to special requirements of SymmetricMemory, we preset certain options for the pool. + # - use_on_oom=False: we don't want to lend the space of the pool for + # non-symmetric allocations because this could desync the allocation state + # across ranks. + # - no_split=True: we don't want to split segments, because today a segment + # is associated with a signal pad, if two allocated tensors share a segment + # and their kernels concurrently use (the same) signal pad, this could cause + # undefined behaviors. We could consider relaxing this in the future if we + # establish stream tracking and implicit synchronization around an + # allocation. + if device not in _symm_mem_pools: + allocator = get_mempool_allocator(device) + # Create a new pool with the given allocator and the preset options. + _symm_mem_pools[device] = torch.cuda.MemPool( + allocator, + use_on_oom=False, + no_split=True, + ) + + return _symm_mem_pools[device] + + +def _cuda_get_out( + dst: torch.Tensor, hdl: _SymmetricMemory, offset: int, size: int, peer: int +) -> None: + storage_offset = hdl.offset // dst.element_size() + offset + remote_src = hdl.get_buffer(peer, (size,), dst.dtype, storage_offset) + dst.view(-1)[:size].copy_(remote_src) + + +# One-sided communication APIs. +def get( + dst: torch.Tensor, + hdl: _SymmetricMemory, + peer: int, + offset: int = 0, +) -> None: + r""" + get(dst, hdl, peer, offset=0) -> () + + Copy ``dst.numel()`` elements starting at ``offset`` from ``peer``'s + symmetric allocation into local ``dst`` using one-sided symmetric memory + access. + + ``hdl`` is the symmetric memory handle returned by + :func:`torch.distributed._symmetric_memory.rendezvous`; the remote source + is ``peer``'s allocation backing that handle. The number of elements copied + is inferred from ``dst``; pass a view (e.g. ``dst[:n]``) to fill only part + of a tensor. ``offset`` is expressed in elements of ``dst``'s dtype. + ``dst`` can be a regular CUDA tensor or a symmetric-memory tensor; it must + be on the same device as ``hdl`` and backed by contiguous memory. The copy + is issued on the current CUDA stream. + + Args: + dst (Tensor): local destination tensor. + hdl (SymmetricMemory): handle whose peer allocation is the remote + source. + peer (int): rank to copy from. + offset (int, optional): element offset into the peer allocation to + start reading from. Defaults to ``0``. + """ + if dst.device != hdl.device: + raise ValueError("get: dst must be on the same device as hdl") + if not dst.is_contiguous(): + raise ValueError("get: dst must be backed by contiguous memory") + if offset < 0: + raise ValueError("get: offset must be non-negative") + if peer < 0 or peer >= hdl.world_size: + raise ValueError("get: invalid peer") + size = dst.numel() + element_size = dst.element_size() + if hdl.offset % element_size != 0: + raise RuntimeError("get: handle offset is not element-aligned") + start = hdl.offset + offset * element_size + end = start + size * element_size + if start > hdl.buffer_size or end > hdl.buffer_size: + raise ValueError("get: requested range exceeds symmetric allocation") + + backend = get_backend(dst.device) + if backend == "CUDA": + _cuda_get_out(dst, hdl, offset, size, peer) + return + + # `hdl` is a pybind `_SymmetricMemory` object. Dispatcher expects the + # TorchBind custom class type `__torch__.torch.classes.c10d.SymmetricMemory`. + # Convert via `.boxed()`. + hdl_boxed = hdl.boxed() if hasattr(hdl, "boxed") else hdl + if backend == "NVSHMEM": + torch.ops.symm_mem.nvshmem_get_out(dst, hdl_boxed, offset, size, peer) + elif backend == "NCCL": + torch.ops.symm_mem.nccl_get_out(dst, hdl_boxed, offset, size, peer) + else: + raise ValueError(f"get: unsupported backend: {backend}") + + +def put_signal(src: torch.Tensor, hdl: _SymmetricMemory, peer: int) -> None: + r""" + put_signal(src, hdl, peer) -> None + + Put data to a peer's symmetric memory and signal the peer. + + Args: + src (torch.Tensor): the source tensor to read data from. + hdl (SymmetricMemory): the symmetric memory to put data to. + peer (int): the peer to put data to. + """ + backend = get_backend(src.device) + # `hdl` is a pybind `_SymmetricMemory` object. Dispatcher expects the + # TorchBind custom class type `__torch__.torch.classes.c10d.SymmetricMemory`. + # Convert via `.boxed()`. + hdl_boxed = hdl.boxed() if hasattr(hdl, "boxed") else hdl + if backend == "NCCL": + torch.ops.symm_mem.nccl_put_signal(src, hdl_boxed, peer) + # TODO: other backends' dispatch goes here + else: + raise ValueError(f"put_signal: unsupported backend: {backend}") + + +def wait_signal(hdl: _SymmetricMemory, peer: int) -> None: + r""" + wait_signal(hdl, peer) -> None + + Wait for a signal from a peer. + + Args: + hdl (SymmetricMemory): the symmetric memory handle on which to wait for a signal. + peer (int): the peer to wait for a signal from. + """ + backend = get_backend(hdl.device) + # See note in `put_signal` about `_SymmetricMemory` vs TorchBind type. + hdl_boxed = hdl.boxed() if hasattr(hdl, "boxed") else hdl + if backend == "NCCL": + torch.ops.symm_mem.nccl_wait_signal(hdl_boxed, peer) + # TODO: other backends' dispatch goes here + else: + raise ValueError(f"wait_signal: unsupported backend: {backend}") + + +def reduce_scatter_offset( + input: torch.Tensor, + out: list[torch.Tensor], + group: str, + *, + dim: int, + offsets: list[int] | None = None, + dst_ranks: list[int] | None = None, + red_op: str = "sum", +) -> None: + r""" + reduce_scatter_offset(input, out, group, *, dim, offsets, dst_ranks, red_op='sum') -> None + + Simultaneously reduce N blocks of a 2-D ``input`` tensor from a symmetric + memory buffer, routing each block to a specific destination rank. Only + ``dst_ranks[i]`` writes the reduced result for block ``i``; the result is + written to a contiguous output tensor, with the same shape as block ``i``. + + The ``dim`` argument controls which dimension is sharded: + + - ``dim=0`` (row sharding): block ``i`` spans + ``input[offsets[i-1] : offsets[i], :]``. Each ``out[j]`` has shape + ``(size_j, input.size(1))``. + - ``dim=1`` (column sharding): block ``i`` spans + ``input[:, offsets[i-1] : offsets[i]]``. Each ``out[j]`` has shape + ``(input.size(0), size_j)``. + + Blocks are described by ``offsets``, an inclusive prefix-sum of block sizes + along ``dim`` (first block starts at index 0 by convention). Block offsets + can be even or uneven; when uneven, the following condition must be met: for + each ``j``, the ``j``-th owned block must have the same size across all + ranks (so that ``out[j]`` has a uniform shape); different ``j``'s may + differ. + + Args: + input (Tensor): 2-D tensor allocated via symmetric memory (innermost + dimension must be contiguous). + out (list[Tensor]): Output tensors for this rank's owned blocks. Must + have length equal to the number of blocks owned by this rank (i.e. + the count of ``i`` where ``dst_ranks[i] == my_rank``). Each + ``out[j]`` must be contiguous with the same dtype as ``input``. + group (str): The name of the ``ProcessGroup`` to perform the operation on. + dim (int): Dimension along which blocks are defined (0 or 1). + offsets (list[int] | None): Inclusive prefix-sum of block sizes along + ``dim``, length N. If not provided, ``input.size(dim)`` is divided + into equal-size blocks based on the size of the ``group``. + dst_ranks (list[int] | None): Destination rank for each block. If not + provided, blocks are distributed round-robin across ranks. + red_op (str): Reduction operation; currently only ``'sum'`` is supported. + + Example:: + + >>> # doctest: +SKIP + >>> # Each rank holds a Grouped GEMM gradient buffer in symmetric memory. + >>> # The buffer has W experts laid out as equal column blocks; each expert + >>> # is reduced to a specific rank (dst_ranks[i] == i % world_size). + >>> buf = symm_mem.empty(H, W * C, dtype=torch.bfloat16, device="cuda") + >>> symm_mem.rendezvous(buf, group=group_name) + >>> offsets = [i * C for i in range(1, W + 1)] # inclusive prefix-sum + >>> dst_ranks = [i % world_size for i in range(W)] + >>> n_owned = sum(r == rank for r in dst_ranks) + >>> out = [torch.empty(H, C, dtype=torch.bfloat16, device="cuda") for _ in range(n_owned)] + >>> symm_mem.reduce_scatter_offset(buf, out, group_name, dim=1, offsets=offsets, dst_ranks=dst_ranks) + """ + backend = get_backend(input.device) + if backend == "NCCL": + torch.ops.symm_mem.nccl_reduce_scatter_offset( + input, out, group, dim, offsets, dst_ranks, red_op + ) + else: + raise NotImplementedError( + f"reduce_scatter_offset: unsupported backend: {backend}" + ) + + +def is_symm_mem_tensor(tensor: torch.Tensor) -> bool: + r""" + is_symm_mem_tensor(tensor) -> bool + + Returns ``True`` if ``tensor`` was allocated via symmetric memory + (i.e. via :func:`torch.distributed._symmetric_memory.empty` or + :meth:`_SymmetricMemory.empty_strided_p2p`). + + This is a non-collective, O(1) check. + + Args: + tensor (:class:`torch.Tensor`): the tensor to check. + """ + return _SymmetricMemory.is_symm_mem_tensor(tensor) + + +def all_to_all_nd( + input: torch.Tensor, + out: torch.Tensor, + scatter_dim: int, + gather_dim: int, + *, + group: str, +) -> None: + r""" + all_to_all_nd(input, out, scatter_dim, gather_dim, *, group) -> None + + Permute-free all-to-all: shards along ``scatter_dim`` of ``input`` are exchanged + so each rank receives one shard per peer; results are laid out along ``gather_dim`` + of ``out``. Supported pairs are ``(scatter_dim=1, gather_dim=0)`` and + ``(scatter_dim=0, gather_dim=1)``. + + For ``scatter_dim=1``, ``gather_dim=0`` (let ``G`` be the group size): ``input`` is + ``[rows, G * local_cols]`` or equivalently ``[rows, G, local_cols]``; + each rank ``r`` reads column block ``r`` from every peer and writes peer ``j`` into + ``out[j]``. ``out`` must be contiguous with shape ``[G, rows, local_cols]`` or the + flattened equivalent ``[G * rows, local_cols]``. + + For ``scatter_dim=0``, ``gather_dim=1`` (``G`` is the group size): ``input`` is + ``[G * local_rows, cols]`` or equivalently ``[G, local_rows, cols]``; + each rank ``r`` reads row block ``r`` from every peer; peer ``j`` is stored in + ``out[:, j, :]``. ``out`` must be contiguous with shape ``[local_rows, G, cols]`` or + ``[local_rows, G * cols]`` (flattened gather and inner dims). + + Args: + input (Tensor): For ``(scatter_dim=1, gather_dim=0)``, 2-D ``[rows, G*local_cols]`` or + 3-D ``[rows, G, local_cols]`` in symmetric memory (innermost dimension contiguous). + For ``(scatter_dim=0, gather_dim=1)``, 2-D ``[G*local_rows, cols]`` or 3-D + ``[G, local_rows, cols]`` (same layout). + out (Tensor): Contiguous buffer (see shapes above; 2-D allowed where noted). + scatter_dim (int): ``0`` or ``1`` — dimension along which the input is partitioned. + gather_dim (int): ``0`` or ``1`` — dimension along which peer chunks are + concatenated in ``out``. + group (str): The name of the ``ProcessGroup`` to perform the operation on. + """ + backend = get_backend(input.device) + if backend == "NCCL": + torch.ops.symm_mem.nccl_all_to_all_nd( + input, out, scatter_dim, gather_dim, group + ) + else: + raise NotImplementedError(f"all_to_all_nd: unsupported backend: {backend}") + + +__all__ = [ + "empty", + "is_symm_mem_tensor", + "rendezvous", + "is_nvshmem_available", + "set_backend", + "get_backend", + "get", + "set_signal_pad_size", + "get_signal_pad_size", + "get_mem_pool", + "reduce_scatter_offset", + "all_to_all_nd", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ceb259d8bdd6373503ab3fea7865235b8f6e58c8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/__init__.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9237cb6186b2c1d24eb5c1014aeb821a28e5ec2795a2d42897c99d3344c151b2 +size 106447 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nccl.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nccl.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..79e1b24ea98ad2e853176db2a434a86d54794176 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nccl.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nvshmem_triton.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nvshmem_triton.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2ab0f694be02ee2cdf9b913d0c06f0cd7cd691ba Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_nvshmem_triton.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_rocshmem_triton.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_rocshmem_triton.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cf24ac721df86256eed23e35d42b8d7f044fa057 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_rocshmem_triton.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..af72f50a68529d12cf521daf7c2446e582010c35 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..594c88be96812338889448f60933c88097ad8917 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_shmem_triton_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_triton_kernels.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_triton_kernels.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..92e7210ad3c2538fe58f9d9d842833593952ca56 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/__pycache__/_triton_kernels.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nccl.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nccl.py new file mode 100644 index 0000000000000000000000000000000000000000..b083788e1419570679871511ba8acc19041622d5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nccl.py @@ -0,0 +1,111 @@ +from __future__ import annotations + +from typing import TYPE_CHECKING +from typing_extensions import Self + +import torch + + +if TYPE_CHECKING: + from torch.types import _device + + +__all__ = ["NcclCommRegistration", "register_external_nccl_comm"] + + +class NcclCommRegistration: + r""" + Handle returned by :func:`register_external_nccl_comm`. + + Keeps an externally-owned NCCL communicator published in PyTorch's + symmetric memory registry for as long as this object is alive. Dropping it + (via ``del``, exiting its ``with`` block, or calling :meth:`unregister`) + removes the entry so a successor producer can register cleanly under the + same ``group_name``. + + The comm's lifetime is *not* owned by this object; the registration only + publishes a borrowed pointer. Optionally a strong reference to the source + comm object is held so a stray ``del comm`` cannot dangle the registered + pointer for the lifetime of the registration. + """ + + def __init__( + self, + group_name: str, + device: torch.device, + comm: object | None = None, + ) -> None: + self._group_name = group_name + self._device = device + self._comm = comm + self._active = True + + def unregister(self) -> None: + """Remove the registration. Idempotent.""" + if not self._active: + return + # Imported lazily: only present in NCCL builds with symmetric-memory + # device support. + from torch._C._distributed_c10d import _unregister_external_nccl_comm + + _unregister_external_nccl_comm(self._group_name, self._device) + self._active = False + self._comm = None + + def __enter__(self) -> Self: + return self + + def __exit__(self, *exc: object) -> None: + self.unregister() + + def __del__(self) -> None: + try: + self.unregister() + except Exception: + # Best-effort cleanup; never raise from __del__ (e.g. during + # interpreter teardown the C extension may already be gone). + pass + + +def register_external_nccl_comm( + group_name: str, + comm_ptr: int, + device: _device, + comm: object | None = None, +) -> NcclCommRegistration: + r""" + Publish an externally-owned ``ncclComm_t`` into PyTorch's symmetric memory + registry under ``group_name``. + + This lets producers that are not a ``ProcessGroupNCCL`` (e.g. torchcomms + backends) supply the host NCCL communicator that + :func:`rendezvous` needs, without symmetric memory having to + ``dynamic_cast`` to a specific process-group implementation. After this + call, ``rendezvous(tensor, group=group_name)`` on a tensor in a process + group registered under ``group_name`` will find this comm. + + Args: + group_name (str): the process-group name to register the comm under. + comm_ptr (int): the host ``ncclComm_t`` as an opaque integer pointer + (e.g. from ``torchcomms`` ``get_nccl_comm_ptr()``). + device (`torch.device` or str): the CUDA device the comm belongs to. + comm (object, optional): the source comm object. When provided, a + strong reference is held by the returned registration so the comm + cannot be garbage-collected while still registered. + + Returns: + NcclCommRegistration: keep this alive for as long as symmetric memory + should use this comm; drop it (or call ``unregister()``) to remove the + registration. + + .. note:: + This is only available in NCCL builds with symmetric-memory device + support; otherwise it raises :class:`ImportError`. + """ + from torch._C._distributed_c10d import ( + _register_external_nccl_comm as _register_external_nccl_comm_impl, + ) + + device = torch.device(device) + _register_external_nccl_comm_impl(group_name, comm_ptr, device) + return NcclCommRegistration(group_name, device, comm) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nvshmem_triton.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nvshmem_triton.py new file mode 100644 index 0000000000000000000000000000000000000000..7084ee26aca8f41c77e8b187b5ff7a4068d930fc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_nvshmem_triton.py @@ -0,0 +1,1220 @@ +import logging +import os +import subprocess +import sysconfig +from typing import Any + +import torch.distributed as dist +from torch.utils._triton import has_triton + + +logger = logging.getLogger(__name__) + + +class NvshmemLibFinder: + """ + A class to find path to the NVSHMEM device library. + + Environment variable: + + `NVSHMEM_LIB_DIR` (Optional[str]): The directory where the NVSHMEM device + library is located. If not provided, it will use the default path where + NVSHMEM wheel is installed, or search for the library in common system + paths. + """ + + # Class variable to store the found library path for reuse + found_device_lib_path: str | None = None + + @classmethod + def find_device_library(cls) -> str: + """ + Find the path to the NVSHMEM device library. + + Returns: + str: The path to libnvshmem_device.bc (included). + """ + if cls.found_device_lib_path is not None: + # Return the cached path if it exists + return cls.found_device_lib_path + + # First, check if the user has specified a custom library path + user_lib_dir = os.environ.get("NVSHMEM_LIB_DIR", None) + if user_lib_dir is not None: + lib_path = os.path.join(user_lib_dir, "libnvshmem_device.bc") + if not os.path.exists(lib_path): + raise RuntimeError( + f"NVSHMEM device library not found at specified path: {user_lib_dir}" + ) + cls.found_device_lib_path = lib_path + return lib_path + + # Otherwise, search for the library in the default installation paths + paths = [ + os.path.join(sysconfig.get_path("purelib"), "nvidia", "nvshmem", "lib") + ] + + # Add common system installation paths + common_paths = [ + "/usr/local/lib", + "/usr/lib", + "/opt/nvidia/nvshmem/lib", + ] + paths.extend(common_paths) + + try: + import torch + + torch_lib = os.path.join(os.path.dirname(torch.__file__), "lib") + so_path = os.path.join(torch_lib, "libtorch_nvshmem.so") + + if os.path.exists(so_path): + try: + result = subprocess.run( + ["readelf", "-d", so_path], + capture_output=True, + text=True, + check=True, + ) + + for line in result.stdout.splitlines(): + if ("RPATH" in line or "RUNPATH" in line) and "[" in line: + rpath = line.split("[", 1)[1].split("]", 1)[0] + for p in rpath.split(":"): + p = p.strip().replace("$ORIGIN", torch_lib) + if p and p not in paths: + paths.append(p) + except subprocess.CalledProcessError: + pass + + except ImportError: + pass + + for path in paths: + device_lib = os.path.join(path, "libnvshmem_device.bc") + if os.path.exists(device_lib): + cls.found_device_lib_path = device_lib + return device_lib + + raise RuntimeError(f"NVSHMEM device library not found. Searched: {paths}") + + +def enable_triton(lib_dir: str | None = None) -> dict[str, str]: + raise NotImplementedError( + "`enable_triton` is deprecated. " + "If you need NVSHMEM device function support for Triton, " + "please use `@requires_nvshmem` to decorate your Triton kernel. ", + ) + + +class NvshmemKernelRegistry: + """ + A class to register kernel functions that ** require NVSHMEM initialization ** + """ + + # Class variable to store the functions to be initialized + _to_init: dict[str, Any] = {} + + @classmethod + def register(cls, name: str) -> None: + """ + Register a kernel function with the given name. + + Args: + name (str): The name of the kernel function. + """ + cls._to_init.setdefault(name) + + @classmethod + def deregister(cls, name: str) -> None: + """ + Deregister a kernel function with the given name. + + Args: + name (str): The name of the kernel function. + """ + cls._to_init.pop(name, None) + + @classmethod + def has(cls, name: str) -> bool: + """ + Check if a kernel function with the given name is registered. + + Args: + name (str): The name of the kernel function. + + Returns: + bool: True if the kernel function is registered, False otherwise. + """ + return name in cls._to_init + + +def _nvshmem_init_hook(*args, **kwargs) -> None: # type: ignore[no-untyped-def] + """ + A hook function to initialize the CUModule created by `triton.jit` with + NVSHMEM device context + """ + from torch._C._distributed_c10d import _nvshmemx_cumodule_init + + jit_function = kwargs["fn"].jit_function + fn_name = jit_function.fn.__name__ + + # Only initialize NVSHMEM module for kernels registered via @requires_nvshmem + if NvshmemKernelRegistry.has(fn_name): + key = kwargs["key"] + device = kwargs["compile"]["device"] + jit_function = kwargs["fn"].jit_function + kernel_cache = jit_function.device_caches[device][0] + kernel = kernel_cache.get(key, None) + if kernel is not None: + _ = kernel.run + # Initialize NVSHMEM for the CU module + _nvshmemx_cumodule_init(kernel.module) + else: + logger.warning( + f"It seems Triton hasn't created a kernel for function {fn_name}. " # noqa: G004 + "Please report this issue to Triton." + ) + + +if has_triton(): + from triton.runtime.jit import JITFunction, KernelInterface + + # Create a new Callable class that follows the KernelInterface protocol so + # that the Callable works with the subscript operator, e.g. `foo[(1, 1)]` + class GridCallableWithExtern(KernelInterface): + """ + `KernelInterface` invokes `self.run` in `__getitem__`, i.e. []. We + implement a `run` method by directing the call to `JITFunction.run`, + with added extern_libs kwarg, so that users don't have to pass it + """ + + def __init__(self, jit_func: JITFunction, extern_libs: dict[str, str]) -> None: + self.jit_func = jit_func + self.extern_libs = extern_libs + + def run(self, *args, **kwargs): # type: ignore[no-untyped-def] + # Call the JITFunction.run with added extern_libs kwarg + return self.jit_func.run(*args, **kwargs, extern_libs=self.extern_libs) + + +def requires_nvshmem( # type: ignore[no-untyped-def] + jit_func, # JITFunction created by triton.jit +): + """ + A decorator to register a Triton kernel function that requires NVSHMEM initialization. + + Example usage: + ``` + @requires_nvshmem + @triton.jit + def foo(...): + ... + ``` + + If you would like to specify a path to the NVSHMEM device library other + than standard search locations, you can use the following environment + variable: + ``` + export NVSHMEM_LIB_DIR=/path/to/nvshmem/lib + ``` + """ + + import triton + from triton.runtime.jit import JITFunction + + if not isinstance(jit_func, JITFunction): + raise TypeError(f"Expected a JITFunction, but got {type(jit_func)}") + + # Find the NVSHMEM device library + lib_path = NvshmemLibFinder.find_device_library() + extern_libs = {"libnvshmem_device": lib_path} + + # Register the JITFunction with the kernel registry as "to be initialized" + NvshmemKernelRegistry.register(jit_func.fn.__name__) + + # Register the NVSHMEM init function as a post-compile hook. + # [Note] This is a global setting (due to lack of Triton API exposure). To + # avoid initializing Triton kernels that do not require NVSHMEM, filtering + # is performed in the hook function itself by checking against + # NvshmemKernelRegistry. + triton.knobs.runtime.jit_post_compile_hook = _nvshmem_init_hook + + return GridCallableWithExtern(jit_func, extern_libs) + + +if has_triton(): + import triton + import triton.language as tl + from triton.language import core + + @triton.jit # type: ignore[misc] + def put(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """ + Put tensor data from local PE to a remote PE. + + This high-level function provides a tensor-aware interface for NVSHMEM put + operations. It automatically handles type checking and size calculations, making + the API more ergonomic and type-safe. + + Args: + dest: Destination tensor on the remote PE. Type must match source. + source: Source tensor on the local PE containing data to be copied. + nelems: Number of elements to transfer. + pe: PE number of the remote PE (0 ≤ pe < nvshmem_n_pes()). + + Notes: + - Performs compile-time type checking between dest and source tensors. + - Automatically calculates byte size from tensor type and element count. + - This is a blocking operation that returns after data has been copied out + of the source array on the local PE. + - The operation does not guarantee delivery to the destination PE. + Use nvshmem_fence() for ordering or nvshmem_quiet() for completion. + + Example: + ``` + # Transfer 100 elements to PE 1 + nvshmem.put(dest_tensor, src_tensor, 100, 1) + ``` + """ + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return putmem_block_extern_wrapper( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def putmem_block_extern_wrapper(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + """Low-level extern wrapper for NVSHMEM put""" + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), # dest ptr + core.dtype("int64"), # source ptr + core.dtype("int64"), # size in bytes + core.dtype("int32"), # pe number + ): ("nvshmemx_putmem_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit # type: ignore[misc] + def get(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """ + Get tensor data from a remote PE to local PE. + + This high-level function provides a tensor-aware interface for NVSHMEM get + operations. It automatically handles type checking and size calculations, making + the API more ergonomic and type-safe. + + Args: + dest: Destination tensor on the local PE. Type must match source. + source: Source tensor on the remote PE containing data to be copied. + nelems: Number of elements to transfer. + pe: PE number of the remote PE (0 ≤ pe < nvshmem_n_pes()). + + Notes: + - Performs compile-time type checking between dest and source tensors. + - Automatically calculates byte size from tensor type and element count. + - This is a blocking operation that returns after data has been delivered + to the destination array on the local PE. + - The destination data is guaranteed to be available for use after the call returns. + + Example: + ``` + # Get 100 elements from PE 0 + nvshmem.get(dest_tensor, src_tensor, 100, 0) + ``` + """ + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return getmem_block_extern_wrapper( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def getmem_block_extern_wrapper(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + """Low-level extern wrapper for NVSHMEM get""" + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), # dest ptr + core.dtype("int64"), # source ptr + core.dtype("int64"), # size in bytes + core.dtype("int32"), # pe number + ): ("nvshmemx_getmem_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit # type: ignore[misc] + def get_nbi(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """ + Get tensor data from a remote PE to local PE, non-blocking. + + Different from the `get` function, this function returns after + initiating the operation. The operation is considered complete after a + subsequent call to `quiet`. + + Args: + dest: Destination tensor on the local PE. Type must match source. + source: Source tensor on the remote PE containing data to be copied. + nelems: Number of elements to transfer. + pe: PE number of the remote PE (0 ≤ pe < nvshmem_n_pes()). + + Notes: + - Performs compile-time type checking between dest and source tensors. + - Automatically calculates byte size from tensor type and element count. + + Example: + ``` + # Get 100 elements from PE 0 + nvshmem.get_nbi(dest, src, 100, 0) + # Some independent computation which overlaps with the get operation + ... + # Wait for completion of the get operation + nvshmem.quiet() + ``` + """ + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return getmem_block_extern_wrapper( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def getmem_nbi_block_extern_wrapper(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + """Low-level extern wrapper for NVSHMEM get""" + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), # dest ptr + core.dtype("int64"), # source ptr + core.dtype("int64"), # size in bytes + core.dtype("int32"), # pe number + ): ("nvshmemx_getmem_nbi_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit # type: ignore[misc] + def putmem_signal_block( # type: ignore[no-untyped-def] + dst, + src, + size_bytes, + signal, + sig_val, + sig_op, + pe, + ): # type: ignore[no-untyped-def] + """ + Put data to remote PE with atomic signal operation using block-scoped operation. + + This function copies data from the local PE to the remote PE and then + atomically updates a signal variable on the remote PE to indicate completion. + This enables efficient point-to-point synchronization between PEs. + + Args: + dst (tensor): A tensor on calling PE symmetric to the destination tensor on remote PE. + src (tensor): Local tensor containing the source data. + size_bytes (int64): Number of bytes to transfer. Must be positive. + signal (tensor): Symmetric signal pad with remote PE. + Must be 8-byte aligned symmetric memory. + signal (int64): Value to be used in the signal operation. + sig_op (int32): Signal operation type. Common values: + - NVSHMEM_SIGNAL_SET (0): Atomic set operation + - NVSHMEM_SIGNAL_ADD (5): Atomic add operation + pe (int32): PE number of the remote PE (0 ≤ pe < nvshmem_n_pes()). + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a blocking operation that returns after data has been copied out + of the source array and the signal has been updated on the remote PE. + - The signal update is performed atomically with respect to other signal + operations and synchronization routines. + - The signal variable must be of type uint64_t in symmetric memory. + - Use with nvshmem_signal_wait_until() for synchronization. + + Example: + ``` + # Transfer data and set completion flag to 1 + NVSHMEM_SIGNAL_SET = 0 + nvshmem.putmem_signal_block( + dst_ptr, src_ptr, 1024, sig_ptr, 1, NVSHMEM_SIGNAL_SET, target_pe + ) + ``` + """ + # Ensure sig_val is 64 bits + sig_val = 0 << 32 | sig_val + return putmem_signal_block_extern_wrapper( + dst.to(tl.int64), + src.to(tl.int64), + size_bytes.to(tl.int64), + signal.to(tl.int64), + sig_val.to(tl.uint64), + sig_op, + pe, + ) + + @core.extern + def putmem_signal_block_extern_wrapper( # type: ignore[no-untyped-def] + dst, + src, + size_bytes, + signal, + sig_val, + sig_op, + pe, + _semantic=None, + ): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [dst, src, size_bytes, signal, sig_val, sig_op, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("uint64"), + core.dtype("int32"), + core.dtype("int32"), + ): ("nvshmemx_putmem_signal_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + # Wait and Signal Operations + + @triton.jit # type: ignore[misc] + def wait_until(ivar, cmp_op, cmp_val): # type: ignore[no-untyped-def] + """ + Wait until a tensor variable meets a specified condition. + + This high-level function provides a tensor-aware interface for NVSHMEM wait_until + operations. It automatically handles tensor address extraction, making + the API more ergonomic and type-safe. + + Args: + ivar_tensor: Tensor to monitor (typically int64/uint64) in symmetric memory. + cmp: Comparison operator. Common values: + - NVSHMEM_CMP_EQ (0): Wait until ivar == cmp_val + - NVSHMEM_CMP_NE (1): Wait until ivar != cmp_val + - NVSHMEM_CMP_GT (2): Wait until ivar > cmp_val + - NVSHMEM_CMP_GE (3): Wait until ivar >= cmp_val + - NVSHMEM_CMP_LT (4): Wait until ivar < cmp_val + - NVSHMEM_CMP_LE (5): Wait until ivar <= cmp_val + cmp_val: Value to compare against. + + Notes: + - This is a blocking operation that will wait indefinitely until the + condition is satisfied. + - The tensor must be in symmetric memory and accessible from other PEs. + + Example: + ``` + # Wait until flag tensor becomes 1 (set by another PE) + NVSHMEM_CMP_EQ = 0 + nvshmem.wait_until_tensor(flag_tensor, NVSHMEM_CMP_EQ, 1) + ``` + """ + tl.static_assert( + ivar.type.element_ty.itemsize == 4, + "wait_until expects a 32-bit type for the synchronization variable", + ) + return wait_until_extern_wrapper(ivar.to(tl.int64), cmp_op, cmp_val) + + @core.extern + def wait_until_extern_wrapper(ivar, cmp, cmp_val, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [ivar, cmp, cmp_val], + { + ( + core.dtype("int64"), + core.dtype("int32"), + core.dtype("int32"), + ): ("nvshmem_int_wait_until", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit # type: ignore[misc] + def signal_wait_until(signal, cmp, cmp_val): # type: ignore[no-untyped-def] + """ + Wait until a signal variable meets a specified condition. + + This function blocks the calling thread until the value at the specified + signal variable satisfies the given comparison condition. Signal variables + are special uint64_t symmetric objects used for efficient synchronization + with signal operations. + + Args: + signal (tensor): Symmetric signal tensor with remote PE. + Must be 8-byte aligned symmetric memory. + cmp (int32): Comparison operator. Common values: + - NVSHMEM_CMP_EQ (0): Wait until signal == cmp_val + - NVSHMEM_CMP_NE (1): Wait until signal != cmp_val + - NVSHMEM_CMP_GT (2): Wait until signal > cmp_val + - NVSHMEM_CMP_GE (3): Wait until signal >= cmp_val + - NVSHMEM_CMP_LT (4): Wait until signal < cmp_val + - NVSHMEM_CMP_LE (5): Wait until signal <= cmp_val + cmp_val (int64): Value to compare against. + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a blocking operation designed specifically for signal variables. + - Signal variables are updated atomically by putmem_signal operations. + - More efficient than wait_until for signal-based synchronization patterns. + - Ensures the signal update is fully complete before returning. + - Commonly used with putmem_signal_block for producer-consumer patterns. + + Example: + ``` + # Wait for signal to be set to completion value + NVSHMEM_CMP_EQ = 0 + nvshmem.signal_wait_until(signal_ptr, NVSHMEM_CMP_EQ, 42) + ``` + """ + cmp_val = 0 << 32 | cmp_val + return signal_wait_until_extern_wrapper( + signal.to(tl.int64), cmp, cmp_val.to(tl.uint64) + ) + + @core.extern + def signal_wait_until_extern_wrapper(signal, cmp, cmp_val, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [signal, cmp, cmp_val], + { + ( + core.dtype("int64"), + core.dtype("int32"), + core.dtype("uint64"), + ): ("nvshmem_signal_wait_until", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def signal_op(sig_addr, signal, sig_op, pe, _semantic=None): # type: ignore[no-untyped-def] + """ + Perform an atomic signal operation on a remote PE. + + This function atomically updates a signal variable on the specified remote PE + using the given operation and value. This enables efficient point-to-point + synchronization and notification between PEs. + + Args: + sig_addr (int64): Symmetric address of the signal variable (uint64_t) on the remote PE. + Must be 8-byte aligned symmetric memory. + signal (int64): Value to be used in the signal operation. + sig_op (int32): Signal operation type. Common values: + - NVSHMEM_SIGNAL_SET (0): Atomically set sig_addr = signal + - NVSHMEM_SIGNAL_ADD (5): Atomically set sig_addr += signal + pe (int32): PE number of the remote PE (0 ≤ pe < nvshmem_n_pes()). + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a one-sided operation - the remote PE does not need to participate. + - The signal operation is performed atomically on the remote PE. + - Can be used with signal_wait_until() on the remote PE for synchronization. + - Provides low-overhead notification mechanism between PEs. + - The signal variable must be of type uint64_t in symmetric memory. + + Example: + ```python + # Atomically set remote signal to 1 to notify completion + NVSHMEM_SIGNAL_SET = 0 + nvshmem.signal_op(remote_signal_ptr, 1, NVSHMEM_SIGNAL_SET, target_pe) + ``` + """ + return core.extern_elementwise( + "", + "", + [sig_addr, signal, sig_op, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int32"), + core.dtype("int32"), + ): ("nvshmemx_signal_op", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + # Memory Ordering Operations + @core.extern + def fence(_semantic=None): # type: ignore[no-untyped-def] + """ + Ensure ordering of put operations to each remote PE. + + This function provides a memory fence that ensures point-to-point ordering + of remote memory operations. Put operations issued before the fence are + guaranteed to be ordered before put operations issued after the fence, + when targeting the same remote PE. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Status code (0 for success). + + Notes: + - This provides weaker ordering guarantees than quiet(). + - Operations to each PE are ordered, but operations to different PEs + may still be reordered relative to each other. + - Does not guarantee completion of operations, only ordering. + - Non-blocking operations are not ordered by fence - use quiet() instead. + - Essential for ensuring correct ordering in communication patterns. + + Memory Ordering Guarantees: + - Put operations before fence() → ordered before → Put operations after fence() + - Ordering is maintained per-destination-PE basis + - Remote PEs can observe the enforced ordering + + Example: + ``` + # Ensure first put completes before second put to same PE + nvshmem.put(dst, src, nelems, target_pe) + nvshmem.fence() # Enforce ordering + nvshmem.put(dst2, src2, nelems, target_pe) + ``` + """ + return core.extern_elementwise( + "", + "", + [], + { + (): ("nvshmem_fence", core.dtype("int32")), + }, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def quiet(_semantic=None): # type: ignore[no-untyped-def] + """ + Wait for completion of all outstanding put operations. + + This function blocks until all outstanding remote memory operations issued + by the calling PE have completed. It provides stronger guarantees than + fence() by ensuring both ordering and completion of all operations. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a blocking operation that waits for completion. + - Ensures all previous put operations have been delivered to their destinations. + - Provides global ordering - operations to ALL PEs are ordered. + - Required to complete non-blocking operations. + - More expensive than fence() but provides stronger guarantees. + + Memory Ordering Guarantees: + - All put operations before quiet() are completed before any operations after quiet() + - Operations are visible to all PEs as having occurred before subsequent operations + - Both blocking and non-blocking operations are completed + + Example: + ``` + # Ensure all data transfers complete before setting completion flag + nvshmem.putmem_block(data_ptr, src_ptr, data_size, target_pe) + nvshmem.quiet() # Wait for data transfer completion + nvshmem.putmem_block( + flag_ptr, flag_src_ptr, 8, target_pe + ) # Signal completion + ``` + """ + return core.extern_elementwise( + "", + "", + [], + { + (): ("nvshmem_quiet", core.dtype("int32")), + }, + is_pure=False, + _semantic=_semantic, + ) + + # PE Information Operations + @core.extern + def my_pe(_semantic=None): # type: ignore[no-untyped-def] + """ + Get the PE number of the calling PE. + + This function returns the unique identifier (PE number) of the current + processing element within the NVSHMEM job. PE numbers range from 0 to + nvshmem_n_pes() - 1. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: PE number of the calling PE (0 ≤ pe < nvshmem_n_pes()). + + Notes: + - This is a pure function that returns the same value throughout execution. + - PE numbering starts from 0 and is contiguous. + - Each PE has a unique identifier within the NVSHMEM job. + - Can be called from both host and device code. + - Essential for implementing PE-specific logic and communication patterns. + + Example: + ``` + # Get current PE number for conditional logic + pe = nvshmem.my_pe() + if pe == 0: + # Root PE logic + pass + else: + # Non-root PE logic + pass + ``` + """ + return core.extern_elementwise( + "", + "", + [], + {(): ("nvshmem_my_pe", core.dtype("int32"))}, + is_pure=True, + _semantic=_semantic, + ) + + @core.extern + def n_pes(_semantic=None): # type: ignore[no-untyped-def] + """ + Get the total number of PEs in the NVSHMEM job. + + This function returns the total count of processing elements (PEs) + participating in the current NVSHMEM job. This value remains constant + throughout the execution of the program. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Total number of PEs in the job (always ≥ 1). + + Notes: + - This is a pure function that returns the same value throughout execution. + - The value is determined at NVSHMEM initialization and never changes. + - Valid PE numbers range from 0 to n_pes() - 1. + - Can be called from both host and device code. + - Essential for implementing collective operations and communication patterns. + + Example: + ``` + # Broadcast from root to all other PEs + total_pes = nvshmem.n_pes() + my_rank = nvshmem.my_pe() + + if my_rank == 0: + # Send to all other PEs + for peer in range(1, total_pes): + nvshmem.putmem_block(dst_ptr, src_ptr, size, peer) + ``` + """ + return core.extern_elementwise( + "", + "", + [], + {(): ("nvshmem_n_pes", core.dtype("int32"))}, + is_pure=True, + _semantic=_semantic, + ) + + # Synchronization Operations + @core.extern + def barrier_all(_semantic=None): # type: ignore[no-untyped-def] + """ + Synchronize all PEs with completion guarantee. + + This function creates a barrier across all PEs in the NVSHMEM job. It ensures + that all local and remote memory updates issued before the barrier by any PE + are completed before any PE exits the barrier. This provides both + synchronization and memory consistency. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a collective operation - all PEs must participate. + - Stronger guarantee than sync_all() - ensures completion of remote operations. + - Blocks until all PEs reach the barrier AND all memory operations complete. + - Must be called from kernels launched with cooperative launch. + - Provides full memory consistency across all PEs. + - More expensive than sync_all() due to completion guarantees. + + Memory Consistency Guarantees: + - All memory updates before barrier_all() are visible to all PEs + - All remote memory operations are completed before any PE continues + - Provides a global synchronization point with memory ordering + + Example: + ``` + # Ensure all PEs complete their work before proceeding + # All PEs execute this - it's a collective operation + nvshmem.barrier_all() + # At this point, all previous operations are complete on all PEs + ``` + """ + return core.extern_elementwise( + "", + "", + [], + {(): ("nvshmem_barrier_all", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def sync_all(_semantic=None): # type: ignore[no-untyped-def] + """ + Synchronize all PEs with local completion guarantee. + + This function creates a lightweight synchronization barrier across all PEs. + It ensures that all local store operations issued before the sync are + visible to other PEs, but does not guarantee completion of remote memory + operations initiated by the calling PE. + + Args: + _semantic: Optional semantic information for Triton compilation. + + Returns: + int32: Status code (0 for success). + + Notes: + - This is a collective operation - all PEs must participate. + - Lighter weight than barrier_all() - only ensures local store visibility. + - Does not guarantee completion of remote memory updates initiated locally. + - Must be called from kernels launched with cooperative launch. + - Suitable when only synchronization (not completion) is needed. + - More efficient than barrier_all() for synchronization-only patterns. + + Memory Consistency Guarantees: + - Local store operations are visible to other PEs + - Does NOT ensure completion of outgoing remote operations + - Provides synchronization point without full completion overhead + + Example: + ``` + # Lightweight synchronization between PEs + # All PEs execute this - it's a collective operation + nvshmem.sync_all() + # Local stores are visible, but remote ops may still be in flight + ``` + """ + return core.extern_elementwise( + "", + "", + [], + {(): ("nvshmem_sync_all", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + # Collective Operations (mem-based APIs - sizes in bytes) + @triton.jit # type: ignore[misc] + def alltoall(team, dest, source, nelems_per_pe): # type: ignore[no-untyped-def] + """ + All-to-all tensor exchange between PEs in a team. + + This high-level function provides a tensor-aware interface for NVSHMEM alltoall + operations. Each PE sends nelems_per_pe elements to every other PE and receives + the same amount from every other PE. + + Args: + team: Team handle for the collective operation. Use 0 for NVSHMEM_TEAM_WORLD. + dest: Destination tensor. Must be large enough for nelems_per_pe * n_pes elements. + source: Source tensor containing data for all PEs. Must contain nelems_per_pe * n_pes elements. + nelems_per_pe: Number of elements to exchange with each PE. + + Notes: + - Performs compile-time type checking between dest and source tensors. + - Automatically calculates byte size from tensor type and element count. + - This is a collective operation - all PEs in the team must participate. + - Data layout: source=[data_for_pe0, data_for_pe1, ...], dest=[data_from_pe0, data_from_pe1, ...] + + Example: + ``` + # Each PE exchanges 10 elements with every other PE + nvshmem.alltoall(0, dest_tensor, src_tensor, 10) + ``` + """ + tl.static_assert(dest.type == source.type) + size_bytes_per_pe = nelems_per_pe * dest.type.element_ty.itemsize + return alltoallmem_block_extern_wrapper( + team, dest.to(tl.int64), source.to(tl.int64), size_bytes_per_pe.to(tl.int64) + ) + + @core.extern # type: ignore[misc] + def alltoallmem_block_extern_wrapper( + team: Any, dest: Any, source: Any, size_bytes: Any, _semantic: Any = None + ) -> None: + """Low-level extern wrapper for NVSHMEM alltoall""" + return core.extern_elementwise( + "", + "", + [team, dest, source, size_bytes], + { + ( + core.dtype("int32"), # team handle + core.dtype("int64"), # dest ptr + core.dtype("int64"), # source ptr + core.dtype("int64"), # size in bytes + ): ("nvshmemx_alltoallmem_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit # type: ignore[misc] + def broadcast(team, dest, source, nelems, pe_root): # type: ignore[no-untyped-def] + """ + Broadcast tensor data from a root PE to all other PEs in a team. + + This high-level function provides a tensor-aware interface for NVSHMEM broadcast + operations. It automatically handles type checking and size calculations, making + the API more ergonomic and type-safe. + + Args: + team: Team handle for the collective operation. Use 0 for NVSHMEM_TEAM_WORLD. + dest: Destination tensor with type information. All PEs receive data here. + source: Source tensor on the root PE. Type must match dest. + nelems: Number of elements to broadcast. + pe_root: PE number of the root PE that provides the source data. + + Notes: + - Performs compile-time type checking between dest and source tensors. + - Automatically calculates byte size from tensor type and element count. + - This is a collective operation - all PEs in the team must participate. + - Must be called from kernels launched with cooperative launch. + + Example: + ``` + # Broadcast 100 elements from PE 0 to all PEs + nvshmem.broadcast(0, dest_tensor, src_tensor, 100, 0) + ``` + """ + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return broadcastmem_block_extern_wrapper( + team, dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe_root + ) + + @core.extern # type: ignore[misc] + def broadcastmem_block_extern_wrapper( + team: Any, + dest: Any, + source: Any, + size_bytes: Any, + pe_root: Any, + _semantic: Any = None, + ) -> None: + """Low-level extern wrapper for NVSHMEM broadcast""" + return core.extern_elementwise( + "", + "", + [team, dest, source, size_bytes, pe_root], + { + ( + core.dtype("int32"), # team handle + core.dtype("int64"), # dest ptr + core.dtype("int64"), # source ptr + core.dtype("int64"), # size in bytes + core.dtype("int32"), # pe_root + ): ("nvshmemx_broadcastmem_block", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + # Reduction Operation + @triton.jit # type: ignore[misc] + def reduce(team, dest, source, nreduce, operation: tl.constexpr): # type: ignore[no-untyped-def] + """ + Performs a collective reduction on tensors across a team of PEs. + + This high-level function provides a tensor-aware interface for NVSHMEM + reduction operations. It automatically infers the data type from the + input tensors and calls the appropriate underlying NVSHMEM function. + + Args: + team: The team handle for the collective (0 for NVSHMEM_TEAM_WORLD). + dest: Destination tensor for the reduction results. + source: Source tensor containing data to be reduced. Must be the same type as dest. + nreduce: The number of elements in the source tensor to reduce. + operation: The reduction operation to perform ("sum", "max", "min", "prod"). + + Notes: + - Performs compile-time type checking between dest and source tensors. + - This is a collective operation that must be called by all PEs in the team. + - Requires a cooperative grid launch. + + Example: + ``` + # Perform a sum reduction on two tensors + nvshmem.reduce(0, dest_tensor, src_tensor, 100, "sum") + ``` + """ + tl.static_assert(dest.type == source.type) + dtype = dest.type.element_ty + return reduce_extern_wrapper( + team, + dest.to(tl.int64), + source.to(tl.int64), + nreduce.to(tl.int64), + operation, + dtype, + ) + + @core.extern # type: ignore[misc] + def reduce_extern_wrapper( + team: Any, + dest: Any, + source: Any, + nreduce: Any, + operation: str, + dtype: Any, + _semantic: Any = None, + ) -> None: + """ + Low-level extern wrapper for NVSHMEM reduction operations. + + This function provides a generic interface to NVSHMEM reduction operations, + automatically selecting the appropriate NVSHMEM function based on the data type + and operation specified. + Args: + team (int64): The team handle (0 for NVSHMEM_TEAM_WORLD). + dest (pointer): Destination pointer where reduction results are stored. + source (pointer): Source pointer containing data to be reduced. + nreduce (int64): Number of elements to reduce. + operation (str): Reduction operation ("sum", "max", "min", "prod"). + dtype: Data type specification - accepts torch.dtype, tl.dtype, str, or constexpr. + _semantic: Optional semantic information for Triton compilation. + + Raises: + ValueError: If the operation is not supported. + TypeError: If the data type is not supported. + + Example: + nvshmem.reduce(0, dest_ptr, src_ptr, 100, "sum", torch.float32) + """ + # Mapping from Triton dtype names to NVSHMEM typenames + DTYPE_TO_NVSHMEM_MAP = { + "int8": "int8", + "int16": "int16", + "int32": "int32", + "int64": "int64", + "uint8": "uint8", + "uint16": "uint16", + "uint32": "uint32", + "uint64": "uint64", + "fp16": "half", + "bf16": "bfloat16", + "fp32": "float", + "fp64": "double", + } + + # Triton dtype names are standardized as fp16, bf16, fp32, etc. + dtype_name = str(dtype).replace("tl.", "") + + if dtype_name not in DTYPE_TO_NVSHMEM_MAP: + raise TypeError( + f"Unsupported reduction dtype: {dtype_name}. Supported dtypes: {list(DTYPE_TO_NVSHMEM_MAP.keys())}" + ) + + # Extract operation name from constexpr if needed + op_name = operation.value if hasattr(operation, "value") else operation + + # Validate operation is supported + supported_ops = {"sum", "max", "min", "prod"} + if op_name not in supported_ops: + raise ValueError( + f"Unsupported reduction operation: '{op_name}'. Supported ops are {supported_ops}" + ) + + # Map to NVSHMEM typename and validate dtype is supported + nvshmem_typename = DTYPE_TO_NVSHMEM_MAP.get(dtype_name) + if nvshmem_typename is None: + raise TypeError( + f"Unsupported reduction dtype: {dtype_name}. Supported dtypes are {list(DTYPE_TO_NVSHMEM_MAP.keys())}" + ) + + # Generate NVSHMEM function name + nvshmem_func = f"nvshmem_{nvshmem_typename}_{op_name}_reduce" + + # Define function signature - all parameters are int64 in Triton (they are just ptrs) + signature = ( + core.dtype("int32"), # team handle + core.dtype("int64"), # destination pointer + core.dtype("int64"), # source pointer + core.dtype("int64"), # number of elements + ) + + return core.extern_elementwise( + "", + "", + [team, dest, source, nreduce], + {signature: (nvshmem_func, core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + # Utility for inspecting Triton kernels + + triton_kernels: dict = {} + + def _log_triton_kernel(kernel) -> None: # type: ignore[no-untyped-def] + import atexit + import tempfile + + if dist.is_initialized() and dist.get_rank() != 0: + return + + def on_exit() -> None: + logger.info("PTX files:") + for kernel in triton_kernels: + with tempfile.NamedTemporaryFile(delete=False) as f: + f.write(kernel.asm["ptx"].encode("utf-8")) + logger.info(f"+- {kernel.name}: {f.name}") # noqa: G004 + + if len(triton_kernels) == 0: + atexit.register(on_exit) + + if kernel not in triton_kernels: + triton_kernels[kernel] = None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_rocshmem_triton.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_rocshmem_triton.py new file mode 100644 index 0000000000000000000000000000000000000000..af75abb442e8af716553e3aae56aac2bdff0cbea --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_rocshmem_triton.py @@ -0,0 +1,459 @@ +""" +rocSHMEM Triton integration for AMD GPU (ROCm) builds. + +This module provides the rocSHMEM-specific counterpart to the NVSHMEM Triton +integration in _nvshmem_triton.py. It is only imported on ROCm builds +(i.e. when ``torch.version.hip is not None``). +""" + +import logging +import os +import sysconfig +from typing import Any + +import torch +from torch.distributed._symmetric_memory._shmem_triton_utils import ( + run_shmem_init_hook, + ShmemKernelRegistry, +) +from torch.utils._triton import has_triton + + +logger = logging.getLogger(__name__) + + +class RocshmemLibFinder: + """ + Find the architecture-specific rocSHMEM device bitcode library. + + Environment variable: + ``ROCSHMEM_LIB_DIR`` (Optional[str]): directory containing + ``librocshmem_device_{arch}.bc``. When not set, the standard + ROCm installation at ``/opt/rocm/lib`` is searched. + + Example:: + export ROCSHMEM_LIB_DIR=/opt/rocm/lib + """ + + found_device_lib_path: str | None = None + + @classmethod + def find_device_library(cls) -> str: + if cls.found_device_lib_path is not None: + return cls.found_device_lib_path + + if not torch.cuda.is_available(): + raise RuntimeError( + "ROCm/CUDA not available — cannot detect GPU architecture" + ) + + props = torch.cuda.get_device_properties(0) + # gcnArchName returns e.g. "gfx942:sramecc+:xnack-" + arch = props.gcnArchName.split(":")[0] + logger.info("Detected GPU architecture: %s", arch) + + lib_name = f"librocshmem_device_{arch}.bc" + + search_paths = [ + os.path.join(sysconfig.get_path("purelib"), "amd", "rocshmem", "lib"), + os.path.join(os.environ.get("ROCM_PATH") or "/opt/rocm", "lib"), + "/usr/local/lib", + "/usr/lib", + ] + + user_lib_dir = os.environ.get("ROCSHMEM_LIB_DIR") + if user_lib_dir is not None: + lib_path = os.path.join(user_lib_dir, lib_name) + if not os.path.exists(lib_path): + raise RuntimeError( + f"rocSHMEM device library not found at ROCSHMEM_LIB_DIR: {lib_path}" + ) + logger.info("Found rocSHMEM device library: %s", lib_path) + cls.found_device_lib_path = lib_path + return lib_path + + lib_path = None + for path in search_paths: + candidate = os.path.join(path, lib_name) + if os.path.exists(candidate): + lib_path = candidate + break + + if lib_path is None: + raise RuntimeError( + f"rocSHMEM device library '{lib_name}' not found.\n" + f"Searched: {search_paths}\n" + "Set ROCSHMEM_LIB_DIR to the directory containing it." + ) + logger.info("Found rocSHMEM device library: %s", lib_path) + cls.found_device_lib_path = lib_path + return lib_path + + +class RocshmemKernelRegistry(ShmemKernelRegistry): + """Track Triton kernels that need rocSHMEM HIP-module initialization.""" + + _to_init: dict[str, Any] = {} + + +def _rocshmem_init_hook(*args, **kwargs) -> None: # type: ignore[no-untyped-def] + """ + Post-compile hook that initializes rocSHMEM device context in the + compiled HIP module using the existing c10d NVSHMEM entrypoint. + """ + from torch._C._distributed_c10d import _nvshmemx_cumodule_init + + run_shmem_init_hook( + kwargs=kwargs, + registry=RocshmemKernelRegistry, + module_init=_nvshmemx_cumodule_init, + logger=logger, + ) + + +def requires_rocshmem( # type: ignore[no-untyped-def] + jit_func, +): + """ + Decorator to mark a Triton kernel as requiring rocSHMEM device APIs. + + Finds the architecture-specific rocSHMEM bitcode library, registers + the kernel for post-compile HIP-module initialization, and wraps the + function so that ``extern_libs`` is injected automatically. + + Example:: + + @requires_rocshmem + @triton.jit + def my_kernel(...): + pe = rocshmem_my_pe() + rocshmem_putmem_wg(dest, src, nbytes, target_pe) + + Set ``ROCSHMEM_LIB_DIR`` to override the default library search path. + """ + from torch.distributed._symmetric_memory._shmem_triton_utils import ( + build_requires_shmem_decorator, + ) + + return build_requires_shmem_decorator( + jit_func=jit_func, + find_device_library=RocshmemLibFinder.find_device_library, + extern_libs_key="rocshmem", + registry=RocshmemKernelRegistry, + init_hook=_rocshmem_init_hook, + error_prefix="@requires_rocshmem", + ) + + +if has_triton(): + import triton + import triton.language as tl + from triton.language import core + + # ----------------------------------------------------------------------- + # rocSHMEM device API — Triton-callable device functions. + # + # All symbol names are the direct rocSHMEM device-bitcode names; no + # NVSHMEM → rocSHMEM translation layer is needed here. + # + # RMA (block/wg-scoped): + # nvshmemx_putmem_block → rocshmem_putmem_wg + # nvshmemx_getmem_block → rocshmem_getmem_wg + # nvshmemx_getmem_nbi_block → rocshmem_getmem_nbi_wg + # nvshmemx_putmem_signal_block → rocshmem_putmem_signal_wg + # Wait / signal: + # nvshmem_int_wait_until → rocshmem_int_wait_until + # nvshmem_signal_wait_until → rocshmem_uint64_wait_until + # Memory ordering, PE info, barriers: names are identical. + # ----------------------------------------------------------------------- + + @triton.jit + def put(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """Put *nelems* elements from local *source* to *dest* on remote *pe*.""" + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return _putmem_wg( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def _putmem_wg(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int32"), + ): ("rocshmem_putmem_wg", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def get(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """Get *nelems* elements from *source* on remote *pe* into local *dest* (blocking).""" + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return _getmem_wg( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def _getmem_wg(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int32"), + ): ("rocshmem_getmem_wg", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def get_nbi(dest, source, nelems, pe): # type: ignore[no-untyped-def] + """Non-blocking get; call quiet() for completion.""" + tl.static_assert(dest.type == source.type) + nbytes = nelems * dest.type.element_ty.itemsize + return _getmem_nbi_wg( + dest.to(tl.int64), source.to(tl.int64), nbytes.to(tl.int64), pe + ) + + @core.extern + def _getmem_nbi_wg(dest, source, size_bytes, pe, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [dest, source, size_bytes, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int32"), + ): ("rocshmem_getmem_nbi_wg", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def putmem_signal_block( # type: ignore[no-untyped-def] + dst, + src, + size_bytes, + signal, + sig_val, + sig_op, + pe, + ): + """Put data to remote PE and atomically update a signal variable.""" + sig_val = 0 << 32 | sig_val + return _putmem_signal_wg( + dst.to(tl.int64), + src.to(tl.int64), + size_bytes.to(tl.int64), + signal.to(tl.int64), + sig_val.to(tl.uint64), + sig_op, + pe, + ) + + @core.extern + def _putmem_signal_wg( # type: ignore[no-untyped-def] + dst, + src, + size_bytes, + signal, + sig_val, + sig_op, + pe, + _semantic=None, + ): + return core.extern_elementwise( + "", + "", + [dst, src, size_bytes, signal, sig_val, sig_op, pe], + { + ( + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("int64"), + core.dtype("uint64"), + core.dtype("int32"), + core.dtype("int32"), + ): ("rocshmem_putmem_signal_wg", core.dtype("int32")) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def wait_until(ivar, cmp_op, cmp_val): # type: ignore[no-untyped-def] + """Block until *ivar* satisfies the comparison condition.""" + tl.static_assert( + ivar.type.element_ty.itemsize == 4, + "wait_until expects a 32-bit type for the synchronization variable", + ) + return _int_wait_until(ivar.to(tl.int64), cmp_op, cmp_val) + + @core.extern + def _int_wait_until(ivar, cmp, cmp_val, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [ivar, cmp, cmp_val], + { + (core.dtype("int64"), core.dtype("int32"), core.dtype("int32")): ( + "rocshmem_int_wait_until", + core.dtype("int32"), + ) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def signal_wait_until(signal, cmp, cmp_val): # type: ignore[no-untyped-def] + """Block until a uint64 signal variable satisfies the comparison condition.""" + cmp_val = 0 << 32 | cmp_val + return _uint64_wait_until(signal.to(tl.int64), cmp, cmp_val.to(tl.uint64)) + + @core.extern + def _uint64_wait_until(signal, cmp, cmp_val, _semantic=None): # type: ignore[no-untyped-def] + return core.extern_elementwise( + "", + "", + [signal, cmp, cmp_val], + { + (core.dtype("int64"), core.dtype("int32"), core.dtype("uint64")): ( + "rocshmem_uint64_wait_until", + core.dtype("int32"), + ) + }, + is_pure=False, + _semantic=_semantic, + ) + + @triton.jit + def signal_op(sig_addr, signal, sig_op, pe): # type: ignore[no-untyped-def] + """Not available in current rocSHMEM device bitcode.""" + tl.static_assert( + False, + "rocshmem has no device-bitcode equivalent for signal_op. " + "Use rocshmem_uint64_atomic_set or rocshmem_uint64_atomic_add instead.", + ) + + @core.extern + def fence(_semantic=None): # type: ignore[no-untyped-def] + """Ensure ordering of issued remote-memory operations to each target PE.""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_fence", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def quiet(_semantic=None): # type: ignore[no-untyped-def] + """Wait for completion of all outstanding remote-memory operations.""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_quiet", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def my_pe(_semantic=None): # type: ignore[no-untyped-def] + """Return the PE number of the calling PE.""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_my_pe", core.dtype("int32"))}, + is_pure=True, + _semantic=_semantic, + ) + + @core.extern + def n_pes(_semantic=None): # type: ignore[no-untyped-def] + """Return the total number of PEs.""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_n_pes", core.dtype("int32"))}, + is_pure=True, + _semantic=_semantic, + ) + + @core.extern + def barrier_all(_semantic=None): # type: ignore[no-untyped-def] + """Barrier across all PEs with completion guarantee (workgroup-scoped).""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_barrier_all_wg", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + @core.extern + def sync_all(_semantic=None): # type: ignore[no-untyped-def] + """Lightweight synchronization barrier across all PEs (workgroup-scoped).""" + return core.extern_elementwise( + "", + "", + [], + {(): ("rocshmem_sync_all_wg", core.dtype("int32"))}, + is_pure=False, + _semantic=_semantic, + ) + + # Collective stubs: rocSHMEM *_wg collectives are not exposed in current + # device bitcode, so Triton kernels cannot call them yet. + + @triton.jit + def alltoall(team, dest, source, nelems_per_pe): # type: ignore[no-untyped-def] + """Not available: rocshmem_alltoallmem_wg is not in current device bitcode.""" + tl.static_assert( + False, + "rocshmem_alltoallmem_wg is not available in current device bitcode. " + "Use host-side rocshmem_alltoallmem_on_stream instead.", + ) + + @triton.jit + def broadcast(team, dest, source, nelems, pe_root): # type: ignore[no-untyped-def] + """Not available: rocshmem_broadcastmem_wg is not in current device bitcode.""" + tl.static_assert( + False, + "rocshmem_broadcastmem_wg is not available in current device bitcode. " + "Use host-side rocshmem_broadcastmem_on_stream instead.", + ) + + @triton.jit + def reduce(team, dest, source, nreduce, operation: tl.constexpr): # type: ignore[no-untyped-def] + """Not available: rocshmem team reduce wg ops are not in current device bitcode.""" + tl.static_assert( + False, + "rocshmem team reduce is not available in current device bitcode. " + "Use host-side rocshmem reduce API instead.", + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton.py new file mode 100644 index 0000000000000000000000000000000000000000..d1dcf01833ccddc259ef65f80cb9e9052ab6d867 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton.py @@ -0,0 +1,37 @@ +""" +SHMEM Triton backend dispatch. + +Thin abstraction layer that routes to the correct platform-specific module: + - ROCm - _rocshmem_triton + - CUDA - _nvshmem_triton + + +End users import from here to get backend-agnostic functionality: + - get_shmem_backend_module(): returns the active backend module + - requires_shmem(): backend-agnostic @requires_shmem decorator +""" + +import torch + + +def get_shmem_backend_module(): + if torch.version.hip is not None: + from torch.distributed._symmetric_memory import _rocshmem_triton + + return _rocshmem_triton + from torch.distributed._symmetric_memory import _nvshmem_triton + + return _nvshmem_triton + + +def requires_shmem(jit_func): # type: ignore[no-untyped-def] + """ + Backend-agnostic Triton decorator for SHMEM kernels. + + Delegates to ``@requires_rocshmem`` on ROCm and ``@requires_nvshmem`` + on CUDA. + """ + backend = get_shmem_backend_module() + if torch.version.hip is not None: + return backend.requires_rocshmem(jit_func) + return backend.requires_nvshmem(jit_func) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..b9abf5dac6a976d13b15682b8313fdd32279edcd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_shmem_triton_utils.py @@ -0,0 +1,101 @@ +""" +Shared utilities for SHMEM Triton backends (_nvshmem_triton, _rocshmem_triton). + +This is a leaf module — it does NOT import any backend or dispatch module, +so backend modules can safely import from here without circular dependencies. +""" + +from collections.abc import Callable +from typing import Any + +from torch.utils._triton import has_triton + + +class ShmemKernelRegistry: + """Base registry tracking Triton kernels that need SHMEM module init.""" + + _to_init: dict[str, Any] = {} + + @classmethod + def register(cls, name: str) -> None: + cls._to_init.setdefault(name) + + @classmethod + def deregister(cls, name: str) -> None: + cls._to_init.pop(name, None) + + @classmethod + def has(cls, name: str) -> bool: + return name in cls._to_init + + +def run_shmem_init_hook( + *, + kwargs: dict[str, Any], + registry: type[ShmemKernelRegistry], + module_init: Callable[[Any], None], + logger: Any, +) -> None: + """Common post-compile hook logic shared by NVSHMEM and rocSHMEM backends.""" + jit_function = kwargs["fn"].jit_function + fn_name = jit_function.fn.__name__ + + if not registry.has(fn_name): + return + + key = kwargs["key"] + device = kwargs["compile"]["device"] + kernel_cache = jit_function.device_caches[device][0] + kernel = kernel_cache.get(key, None) + if kernel is not None: + _ = kernel.run + module_init(kernel.module) + else: + logger.warning( + "It seems Triton hasn't created a kernel for function %s. " + "Please report this issue to Triton.", + fn_name, + ) + + +if has_triton(): + from triton.runtime.jit import JITFunction, KernelInterface + + class GridCallableWithExtern(KernelInterface): + """ + KernelInterface invokes self.run in __getitem__, i.e. []. + We direct the call to JITFunction.run with extern_libs injected, + so that users don't have to pass it. + """ + + def __init__(self, jit_func: JITFunction, extern_libs: dict[str, str]) -> None: + self.jit_func = jit_func + self.extern_libs = extern_libs + + def run(self, *args, **kwargs): # type: ignore[no-untyped-def] + return self.jit_func.run(*args, **kwargs, extern_libs=self.extern_libs) + + def build_requires_shmem_decorator( # type: ignore[no-untyped-def] + *, + jit_func, + find_device_library: Callable[[], str], + extern_libs_key: str, + registry: type[ShmemKernelRegistry], + init_hook: Callable[..., None], + error_prefix: str, + ): + """Build the @requires_ decorator for a Triton JIT function.""" + import triton + from triton.runtime.jit import JITFunction + + if not isinstance(jit_func, JITFunction): + raise TypeError( + f"{error_prefix} must be applied to a @triton.jit function, " + f"got {type(jit_func)}" + ) + + lib_path = find_device_library() + extern_libs = {extern_libs_key: lib_path} + registry.register(jit_func.fn.__name__) + triton.knobs.runtime.jit_post_compile_hook = init_hook + return GridCallableWithExtern(jit_func, extern_libs) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_triton_kernels.py b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_triton_kernels.py new file mode 100644 index 0000000000000000000000000000000000000000..700562a2661adc08cbb02346bdf4286b28c2b42e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_symmetric_memory/_triton_kernels.py @@ -0,0 +1,30 @@ +from torch.utils._triton import has_triton + + +if has_triton(): + import triton + import triton.language as tl + + @triton.jit + def reduce_partials_first_dim_kernel( + partials, + out, + shard_elems: tl.constexpr, + group_size: tl.constexpr, + do_avg: tl.constexpr, + BLOCK: tl.constexpr, + ): + offsets = tl.program_id(0) * BLOCK + tl.arange(0, BLOCK) + mask = offsets < shard_elems + acc = tl.zeros((BLOCK,), tl.float32) + for rank in tl.static_range(0, group_size): + vals = tl.load( + partials + rank * shard_elems + offsets, + mask=mask, + other=0.0, + ) + acc += vals.to(tl.float32) + if do_avg: + acc /= group_size + # The accumulation stays in fp32 and is rounded exactly once, here. + tl.store(out + offsets, acc.to(out.dtype.element_ty), mask=mask) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..272a696d69fdb715fc34e1d3a570551cbfa96f17 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__init__.py @@ -0,0 +1,45 @@ +""" +NOTICE: DTensor has moved to torch.distributed.tensor + +This file is a shim to redirect to the new location, and +we keep the old import path starts with `_tensor` for +backward compatibility. We will remove this folder once +we resolve all the BC issues. +""" + +import sys +from importlib import import_module + + +submodules = [ + # TODO: _shards_wrapper/_utils here mainly for checkpoint BC, remove them + "_shards_wrapper", + "_utils", + "experimental", + "device_mesh", +] + +# Redirect imports +for submodule in submodules: + full_module_name = f"torch.distributed.tensor.{submodule}" + sys.modules[f"torch.distributed._tensor.{submodule}"] = import_module( + full_module_name + ) + +from torch.distributed.tensor import ( + DeviceMesh, + distribute_module, + distribute_tensor, + DTensor, + empty, + full, + init_device_mesh, + ones, + Partial, + Placement, + rand, + randn, + Replicate, + Shard, + zeros, +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a5e65a44136425fbbdf7532792c55b6ddf40385d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bf63387c9307246674aa7a30eebe2302d9eedc54 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/placement_types.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/placement_types.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1ebc495d5ec0e3e43b7f7724c1206c4b9a8e207d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/__pycache__/placement_types.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/api.py new file mode 100644 index 0000000000000000000000000000000000000000..621c700e77ff90eb18a71e7767b37f79dd24dd96 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/api.py @@ -0,0 +1,9 @@ +""" +NOTE: torch.distributed._tensor has been moved to torch.distributed.tensor. +The imports here are purely for backward compatibility. We will remove these +imports in a few releases + +TODO: throw warnings when this module imported +""" + +from torch.distributed.tensor._api import * # noqa: F403 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/placement_types.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/placement_types.py new file mode 100644 index 0000000000000000000000000000000000000000..e7dac1b76b52b91424d2406ff66db9be706c1057 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tensor/placement_types.py @@ -0,0 +1,10 @@ +""" +NOTE: torch.distributed._tensor has been moved to torch.distributed.tensor. +The imports here are purely for backward compatibility. We will remove these +imports in a few releases + +TODO: throw warnings when this module imported +""" + +from torch.distributed.tensor._dtensor_spec import * # noqa: F403 +from torch.distributed.tensor.placement_types import * # noqa: F403 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_token_switch.py b/.venv/lib/python3.12/site-packages/torch/distributed/_token_switch.py new file mode 100644 index 0000000000000000000000000000000000000000..c904edf2c7158f895583623bf8a0dce607699fa9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_token_switch.py @@ -0,0 +1,343 @@ +# mypy: allow-untyped-defs +from __future__ import annotations + +import abc +import ctypes +import importlib.util +import os +from dataclasses import dataclass +from typing import Any, TYPE_CHECKING + +import torch + + +if TYPE_CHECKING: + from torch.distributed.distributed_c10d import ProcessGroup + + +# Keeps preloaded shared libraries resident for the process lifetime. +_NCCL_EP_KEEPALIVE: list[ctypes.CDLL] = [] + + +def _find_pkg_dir(name: str) -> str | None: + # Locate an installed package's directory without importing it. find_spec + # raises (rather than returning None) when a dotted name's parent namespace + # is absent, so guard that. + try: + spec = importlib.util.find_spec(name) + except ModuleNotFoundError: + return None + if spec is None or not spec.submodule_search_locations: + return None + return spec.submodule_search_locations[0] + + +def _prepare_nccl4py() -> None: + # Dynamic (USE_SYSTEM_NCCL=ON / wheel) build only: the extension NEEDED-links + # libnccl_ep.so, which the nccl4py wheel provides at runtime. Point nccl-ep's + # JIT at nccl4py's EP headers (NCCL_EP_HOME) and the nvidia.nccl wheel's NCCL + # headers (NCCL_HOME), and make libnccl_ep resolvable, before importing the + # extension (its libnccl dependency is already loaded by torch). The static + # build bakes all of this in and never reaches here. + nccl_pkg = _find_pkg_dir("nccl") + if nccl_pkg is None: + raise ImportError( + "TokenSwitchNCCL needs the 'nccl4py' package for this build's " + "libnccl_ep.so and runtime JIT headers. Install it with " + "`pip install nccl4py==0.3.1`." + ) + ep_dir = os.path.join(nccl_pkg, "ep") + if "NCCL_EP_HOME" not in os.environ: + ep_headers = os.path.join(ep_dir, "include", "nccl_ep") + if not os.path.isdir(ep_headers): + raise ImportError( + f"nccl4py at {nccl_pkg} is missing the EP JIT headers expected " + f"at {ep_headers}; reinstall nccl4py or set NCCL_EP_HOME." + ) + os.environ["NCCL_EP_HOME"] = ep_dir + + # Point the JIT at NCCL headers. libnccl.so itself needs no preload here: a + # USE_SYSTEM_NCCL=ON torch (the only build that reaches this path) already + # dynamically loaded the system libnccl, so libnccl_ep.so's NEEDED + # libnccl.so.2 resolves to that already-loaded copy. + nvidia_nccl = _find_pkg_dir("nvidia.nccl") + if nvidia_nccl is not None and os.path.isdir(os.path.join(nvidia_nccl, "include")): + os.environ.setdefault("NCCL_HOME", nvidia_nccl) + + # nccl4py ships libnccl_ep unversioned (libnccl_ep.so) while the extension's + # NEEDED entry is its SONAME libnccl_ep.so.0; load it by path to register + # that SONAME for the import below. + lib = os.path.join(ep_dir, "lib", "libnccl_ep.so") + if os.path.isfile(lib): + _NCCL_EP_KEEPALIVE.append(ctypes.CDLL(lib, mode=ctypes.RTLD_LOCAL)) + + +def _import_nccl_ep() -> Any: + # The EP bindings live in the optional torch._nccl_ep extension (USE_NCCL_EP). + # A USE_SYSTEM_NCCL=OFF build links libnccl_ep statically and bakes its JIT + # header paths, so the extension imports directly -- self-contained, no + # nccl4py. A USE_SYSTEM_NCCL=ON build NEEDED-links libnccl_ep from the nccl4py + # wheel, so the first import fails until nccl4py is set up; fall back to that + # and retry. + try: + # pyrefly: ignore [missing-import] # built only with USE_NCCL_EP + import torch._nccl_ep as _ep + + return _ep + except ImportError: + pass + + _prepare_nccl4py() + try: + # pyrefly: ignore [missing-import] # built only with USE_NCCL_EP + import torch._nccl_ep as _ep + except ImportError as e: + raise ImportError( + "torch._nccl_ep is unavailable; this PyTorch was not built with " + "USE_NCCL_EP (or, for a USE_SYSTEM_NCCL=ON build, the nccl4py wheel " + "is missing)." + ) from e + + return _ep + + +@dataclass(frozen=True, slots=True) +class Routing: + handle: object + topk_idx: torch.Tensor + + +class _DispatchAutograd(torch.autograd.Function): + @staticmethod + def forward( + ctx: Any, + ts: TokenSwitch, + routing: Routing, + tokens: torch.Tensor, + topk_weights: torch.Tensor, + max_recv_tokens: int, + ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]: + _N, H = tokens.shape + K = topk_weights.shape[1] + out_tokens = tokens.new_zeros(max_recv_tokens, H) + out_topk_weights = topk_weights.new_zeros(max_recv_tokens, K) + out_topk_idx = routing.topk_idx.new_zeros(max_recv_tokens, K) + ts._dispatch( + routing, tokens, topk_weights, out_tokens, out_topk_weights, out_topk_idx + ) + ctx.ts = ts + ctx.routing = routing + ctx.tokens_shape = tokens.shape + return out_tokens, out_topk_weights, out_topk_idx + + @staticmethod + # pyrefly: ignore [bad-override] + def backward( + ctx: Any, + grad_out_tokens: torch.Tensor, + grad_out_topk_weights: torch.Tensor, + grad_out_topk_idx: torch.Tensor, + ) -> tuple[None, None, torch.Tensor, None, None]: + grad_tokens = grad_out_tokens.new_zeros(ctx.tokens_shape) + ctx.ts._combine(ctx.routing, grad_out_tokens.contiguous(), grad_tokens) + return None, None, grad_tokens, None, None + + +class _CombineAutograd(torch.autograd.Function): + @staticmethod + def forward( + ctx: Any, + ts: TokenSwitch, + routing: Routing, + expert_tokens: torch.Tensor, + ) -> torch.Tensor: + N = routing.topk_idx.shape[0] + H = expert_tokens.shape[1] + out_tokens = expert_tokens.new_zeros(N, H) + ts._combine(routing, expert_tokens, out_tokens) + ctx.ts = ts + ctx.routing = routing + ctx.expert_shape = expert_tokens.shape + ctx.expert_dtype = expert_tokens.dtype + ctx.top_k = routing.topk_idx.shape[1] + return out_tokens + + @staticmethod + # pyrefly: ignore [bad-override] + def backward( + ctx: Any, grad_out_tokens: torch.Tensor + ) -> tuple[None, None, torch.Tensor]: + M, H = ctx.expert_shape + N = grad_out_tokens.shape[0] + K = ctx.top_k + dtype = ctx.expert_dtype + # ncclEpDispatch requires the output buffer sized to the group's + # max_recv_tokens_per_rank, regardless of what shape expert_tokens had + # in forward (often a slice like out_tokens[:M]). Allocate full-size, + # run dispatch, then slice to ctx.expert_shape so the returned grad + # matches the input that produced it. + max_recv = ctx.ts._max_recv_tokens_per_rank + grad_expert_full = grad_out_tokens.new_zeros(max_recv, H).to(dtype) + dummy_weights = grad_out_tokens.new_zeros(N, K, dtype=torch.float32) + dummy_out_weights = grad_out_tokens.new_zeros(max_recv, K, dtype=torch.float32) + dummy_out_idx = ctx.routing.topk_idx.new_zeros(max_recv, K) + ctx.ts._dispatch( + ctx.routing, + grad_out_tokens.to(dtype).contiguous(), + dummy_weights, + grad_expert_full, + dummy_out_weights, + dummy_out_idx, + ) + return None, None, grad_expert_full[:M].contiguous() + + +class TokenSwitch(abc.ABC): + """Abstract token routing switch (e.g. expert-parallel dispatch / combine). + + Typical usage: :meth:`create_routing`, then :meth:`dispatch` / :meth:`combine`. + """ + + @abc.abstractmethod + def create_routing( + self, + topk_idx: torch.Tensor, + per_expert_token_counts: torch.Tensor | None = None, + ) -> Routing: + """Create expert routing for the current phase (e.g. top-k indices). + + ``per_expert_token_counts`` is optional 1D int32, length >= local experts: + output buffer for per-expert receive counts (NCCL EP ``RECV_EXPERT_COUNTER``). + """ + raise NotImplementedError + + @abc.abstractmethod + def _dispatch( + self, + routing: Routing, + tokens: torch.Tensor, + topk_weights: torch.Tensor, + out_tokens: torch.Tensor, + out_topk_weights: torch.Tensor, + out_topk_idx: torch.Tensor, + ) -> None: + raise NotImplementedError + + @abc.abstractmethod + def _combine( + self, + routing: Routing, + expert_tokens: torch.Tensor, + out_tokens: torch.Tensor, + ) -> None: + raise NotImplementedError + + def dispatch( + self, + routing: Routing, + tokens: torch.Tensor, + topk_weights: torch.Tensor, + max_recv_tokens: int | None = None, + *, + out: tuple[torch.Tensor, torch.Tensor, torch.Tensor] | None = None, + ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]: + """Route tokens to experts. + + With ``out=(out_tokens, out_topk_weights, out_topk_idx)``: writes to the provided + buffers and returns them; no autograd support. + Without ``out``: allocates output buffers and returns + ``(out_tokens, out_topk_weights, out_topk_idx)`` with autograd support. + ``max_recv_tokens`` is required when ``out`` is not provided. + ``topk_weights`` receives no gradient (routing metadata). + """ + if out is not None: + self._dispatch(routing, tokens, topk_weights, *out) + return out + if max_recv_tokens is None: + raise ValueError("max_recv_tokens is required when out= is not provided") + return _DispatchAutograd.apply( + self, routing, tokens, topk_weights, max_recv_tokens + ) # type: ignore[return-value] + + def combine( + self, + routing: Routing, + expert_tokens: torch.Tensor, + *, + out: torch.Tensor | None = None, + ) -> torch.Tensor: + """Gather expert outputs back to token order. + + With ``out=out_tokens``: writes to the provided buffer and returns it; + no autograd support. + Without ``out``: allocates an output buffer and returns it with autograd support. + """ + if out is not None: + self._combine(routing, expert_tokens, out) + return out + return _CombineAutograd.apply(self, routing, expert_tokens) # type: ignore[return-value] + + +class TokenSwitchNCCL(TokenSwitch): + """Token switch backed by NCCL EP (:func:`ncclEpCreateGroup` / dispatch / combine).""" + + def __init__( + self, + process_group: ProcessGroup, + num_experts: int, + max_dispatch_tokens_per_rank: int, + max_recv_tokens_per_rank: int, + max_token_bytes: int, + ) -> None: + self._ep = _import_nccl_ep() + self._max_recv_tokens_per_rank = max_recv_tokens_per_rank + self._group = self._ep._NcclEpGroup.create( + process_group, + num_experts, + max_dispatch_tokens_per_rank, + max_recv_tokens_per_rank, + max_token_bytes, + ) + + def create_routing( + self, + topk_idx: torch.Tensor, + per_expert_token_counts: torch.Tensor | None = None, + ) -> Routing: + """Create expert routing for this phase; pass to :meth:`dispatch` / :meth:`combine`.""" + handle = self._ep._NcclEpHandle.create( + self._group, + topk_idx, + per_expert_token_counts, + ) + return Routing(handle=handle, topk_idx=topk_idx) + + def _dispatch( + self, + routing: Routing, + tokens: torch.Tensor, + topk_weights: torch.Tensor, + out_tokens: torch.Tensor, + out_topk_weights: torch.Tensor, + out_topk_idx: torch.Tensor, + ) -> None: + self._ep._nccl_ep_dispatch( + routing.handle, + tokens, + topk_weights, + out_tokens, + out_topk_weights, + out_topk_idx, + ) + + def _combine( + self, + routing: Routing, + expert_tokens: torch.Tensor, + out_tokens: torch.Tensor, + ) -> None: + self._ep._nccl_ep_combine( + routing.handle, + expert_tokens, + out_tokens, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..22e974cdd64f1082e7a89e441eb8c90163f56d3b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__init__.py @@ -0,0 +1,12 @@ +from .fsdp2_mem_tracker import FSDPMemTracker +from .mem_tracker import MemTracker +from .memory_tracker import MemoryTracker +from .mod_tracker import ModTracker +from .runtime_estimator import RuntimeEstimator +from .sac_estimator import ( + MSPS, + SACEstimator, + SACGreedyOrderMeta, + SACStats, + SACTradeOffStats, +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..39cc19885e7aba6361c58355002269712fe7538b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/common_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/common_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3d202ea6780cf98744b26bc9487d70864b6070c3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/common_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fake_collectives.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fake_collectives.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..88a8a635fc6c7da43df49b7c6f1d5df73fae1086 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fake_collectives.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fsdp2_mem_tracker.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fsdp2_mem_tracker.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ae79c1316f6365f9f7261f7a1a25ee2681e1ec72 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/fsdp2_mem_tracker.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/ilp_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/ilp_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..11af4b9702e881fa43a2544f9f40f727eecf2662 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/ilp_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mem_tracker.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mem_tracker.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..36d4d3a2e29c156ba603b6d9389f936ea5c98c1f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mem_tracker.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/memory_tracker.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/memory_tracker.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fe7c81a46d24b2ed48ee7293237973cd2216ee34 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/memory_tracker.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mod_tracker.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mod_tracker.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c615ba1e019ac2163856d9e8a49b6bc50dc93d0e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/mod_tracker.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/runtime_estimator.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/runtime_estimator.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1299cd4e679b40dc398d12a528ddee92d0087412 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/runtime_estimator.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_estimator.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_estimator.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0950849ab3b2fe568bcbbe26bb9f35e84e94a671 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_estimator.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_ilp.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_ilp.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dfc84d8f6cfc5103cf502c76be43fa5d2ed6b4cf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/__pycache__/sac_ilp.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/common_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/common_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..fdfb649e15e706a063190ac4060707ceca4220f2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/common_utils.py @@ -0,0 +1,43 @@ +import warnings + +import torch +from torch._custom_class_base import CustomClassBase +from torch.utils._python_dispatch import is_traceable_wrapper_subclass + + +def get_untyped_storages(t: torch.Tensor) -> set[torch.UntypedStorage]: + """ + Recursively extracts untyped storages from a tensor or its subclasses. + + Args: + t (torch.Tensor): The tensor to extract storages from. + + Returns: + Set[torch.UntypedStorage]: A set of untyped storages. + """ + unflattened_tensors = [t] + flattened_tensor_storages = set() + while len(unflattened_tensors) > 0: + obj = unflattened_tensors.pop() + if is_traceable_wrapper_subclass(obj): + attrs, _ = obj.__tensor_flatten__() + for attr in attrs: + match getattr(obj, attr): + case torch.Tensor() as v: + unflattened_tensors.append(v) + case CustomClassBase(): + pass + case unexpected: + raise AssertionError( + f"expected Tensor or CustomClassBase, got {type(unexpected)}" + ) + else: + if not hasattr(obj, "untyped_storage"): + warnings.warn( + f"Expected a tensor or a traceable wrapper-subclass of tensor, but got {type(obj)}", + category=UserWarning, + stacklevel=2, + ) + else: + flattened_tensor_storages.add(obj.untyped_storage()) + return flattened_tensor_storages diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fake_collectives.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fake_collectives.py new file mode 100644 index 0000000000000000000000000000000000000000..f8893f3b973e771b9c585db3a6e30109ffde5070 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fake_collectives.py @@ -0,0 +1,270 @@ +from typing import Any + +import torch +from torch._C._distributed_c10d import ( + _resolve_process_group, + FakeWork, + ProcessGroup, + Work, +) +from torch.utils._pytree import tree_map_only + + +c10d = torch.ops.c10d +_c10d_functional = torch.ops._c10d_functional +_c10d_functional_autograd = torch.ops._c10d_functional_autograd +_dtensor = torch.ops._dtensor + +# List of collective operation functions including functional collectives +# Note: The following collectives might be deprecated soon hence not adding them +# depcreated_non_functional_collectives = [ +# c10d.allreduce_coalesced_.default, +# c10d.reduce_scatter_tensor_coalesced_.default, +# c10d.allgather_into_tensor_coalesced_.default, +# c10d.allgather_coalesced_.default, +# ] +non_functional_collectives: set[torch._ops.OpOverload] = { + c10d.broadcast_.default, + c10d.allreduce_.default, + c10d.reduce_.default, + c10d.send.default, + c10d.recv_.default, + c10d.recv_any_source_.default, + c10d.allgather_.default, + c10d.reduce_scatter_.default, + c10d._reduce_scatter_base_.default, + c10d._allgather_base_.default, + c10d.gather_.default, + c10d.scatter_.default, + c10d.alltoall_.default, + c10d.alltoall_base_.default, + c10d.barrier.default, + c10d.monitored_barrier_.default, +} +functional_collectives: set[torch._ops.OpOverload] = { + _c10d_functional.broadcast.default, + _c10d_functional.all_reduce.default, + _c10d_functional.all_gather_into_tensor.default, + _c10d_functional.reduce_scatter_tensor.default, + _c10d_functional.reduce_scatter_tensor_out.default, + _c10d_functional.all_to_all_single.default, + _c10d_functional_autograd.all_to_all_single.default, + _c10d_functional.wait_tensor.default, + _c10d_functional.all_reduce_.default, + _c10d_functional.all_reduce_coalesced.default, + _c10d_functional.all_reduce_coalesced_.default, + _c10d_functional.all_gather_into_tensor_out.default, + _c10d_functional.all_gather_into_tensor_coalesced.default, + _c10d_functional_autograd.all_gather_into_tensor.default, + _c10d_functional.reduce_scatter_tensor_coalesced.default, + _c10d_functional_autograd.reduce_scatter_tensor.default, + _c10d_functional.broadcast_.default, + _c10d_functional.isend.default, + _c10d_functional.irecv.default, + _c10d_functional.batch_p2p_ops.default, + _dtensor.shard_dim_alltoall.default, +} + +sync_ops: set[torch._ops.OpOverload] = { + c10d.barrier.default, + c10d.monitored_barrier_.default, + _c10d_functional.wait_tensor.default, +} + +collective_ops = set.union(functional_collectives, non_functional_collectives) + + +class CollectiveOp: + # Static sets for performance optimization + PG_ARG_1 = { + c10d.broadcast_.default, + c10d.allreduce_.default, + c10d.reduce_.default, + c10d.send.default, + c10d.recv_.default, + c10d.recv_any_source_.default, + c10d.barrier.default, + # c10d.allreduce_coalesced_.default + } + + PG_ARG_2 = { + c10d.allgather_.default, + c10d._allgather_base_.default, + c10d.reduce_scatter_.default, + c10d._reduce_scatter_base_.default, + c10d.gather_.default, + c10d.scatter_.default, + c10d.alltoall_.default, + c10d.alltoall_base_.default, + # c10d.allgather_coalesced_.default, + # c10d.allgather_into_tensor_coalesced_.default + # c10d.reduce_scatter_tensor_coalesced_.default + } + + PG_ARG_3 = { + _c10d_functional.broadcast.default, + _c10d_functional.broadcast_.default, + _c10d_functional.all_reduce.default, + _c10d_functional.all_reduce_.default, + _c10d_functional.all_reduce_coalesced.default, + _c10d_functional.all_reduce_coalesced_.default, + _c10d_functional.all_gather_into_tensor.default, + _c10d_functional.all_gather_into_tensor_out.default, + _c10d_functional_autograd.all_gather_into_tensor.default, + _c10d_functional.all_gather_into_tensor_coalesced.default, + } + + PG_ARG_4 = { + _c10d_functional.reduce_scatter_tensor.default, + _c10d_functional.reduce_scatter_tensor_coalesced.default, + _c10d_functional_autograd.reduce_scatter_tensor.default, + _c10d_functional.all_to_all_single.default, + _c10d_functional_autograd.all_to_all_single.default, + _c10d_functional.isend.default, + _c10d_functional.irecv.default, + _dtensor.shard_dim_alltoall.default, + } + + PG_ARG_5 = { + _c10d_functional.batch_p2p_ops.default, + } + + WK_ARG_1 = { + c10d.broadcast_.default, + c10d.allreduce_.default, + c10d.allgather_.default, + c10d.reduce_scatter_.default, + c10d._reduce_scatter_base_.default, + c10d._allgather_base_.default, + c10d.scatter_.default, + c10d.alltoall_.default, + _c10d_functional.isend.default, + _c10d_functional.irecv.default, + } + + WK = { + c10d.send.default, + c10d.recv_.default, + c10d.recv_any_source_.default, + c10d.reduce_.default, + c10d.gather_.default, + c10d.alltoall_base_.default, + c10d.barrier.default, + } + + COMM_TENSOR_ARG_0 = { + c10d.allreduce_.default, + c10d.send.default, + c10d.recv_.default, + c10d.recv_any_source_.default, + c10d.allgather_.default, + c10d.gather_.default, + c10d.reduce_.default, + c10d.broadcast_.default, + _c10d_functional.all_reduce_coalesced.default, + _c10d_functional.all_reduce_coalesced_.default, + # c10d.allreduce_coalesced_.default + # c10d.allgather_coalesced_.default + # c10d.allgather_into_tensor_coalesced_.default, + } + + COMM_TENSOR_ARG_1 = { + c10d.reduce_scatter_.default, + c10d.scatter_.default, + # c10d.reduce_scatter_tensor_coalesced_.default, + } + + COMM_TENSOR_ARG_RES = { + _c10d_functional.all_gather_into_tensor.default, + _c10d_functional_autograd.all_gather_into_tensor.default, + } + + COMM_TENSOR_SINGLE_UNTYPED_STORAGE = { + c10d._allgather_base_.default, + _c10d_functional.broadcast.default, + _c10d_functional.broadcast_.default, + _c10d_functional.all_reduce.default, + _c10d_functional.all_reduce_.default, + _c10d_functional.reduce_scatter_tensor.default, + _c10d_functional_autograd.reduce_scatter_tensor.default, + } + + COMM_TENSOR_ARG_0_AND_RES = { + _c10d_functional.all_to_all_single.default, + _c10d_functional_autograd.all_to_all_single.default, + _dtensor.shard_dim_alltoall.default, + } + + COMM_TENSOR_RES_SUM = { + _c10d_functional.all_gather_into_tensor_coalesced.default, + _c10d_functional.reduce_scatter_tensor_coalesced.default, + } + + @staticmethod + def sum_tensors(arg: Any) -> int: + """Calculate total memory consumed by the tensors in the argument.""" + total_memory = 0 + + def sum_bytes(t: torch.Tensor) -> None: + nonlocal total_memory + total_memory += t.untyped_storage().nbytes() + + tree_map_only(torch.Tensor, sum_bytes, arg) + return total_memory + + @staticmethod + def get_process_group(func, args) -> ProcessGroup: # type: ignore[no-untyped-def] + """Retrieve the process group for collective operations, except `wait_tensor`.""" + if func in CollectiveOp.PG_ARG_1: + return ProcessGroup.unbox(args[1]) + if func in CollectiveOp.PG_ARG_2: + return ProcessGroup.unbox(args[2]) + if func in CollectiveOp.PG_ARG_3: + return _resolve_process_group(args[2]) + if func in CollectiveOp.PG_ARG_4: + return _resolve_process_group(args[3]) + if func in CollectiveOp.PG_ARG_5: + return _resolve_process_group(args[4]) + raise TypeError(f"Func {func} not found in {collective_ops}") + + @staticmethod + def get_comm_tensor_size(func, res, args, kwargs) -> int: # type: ignore[no-untyped-def] + """Compute the communication tensor size, except for `wait_tensor`, `barrier`, and `monitored_barrier`.""" + if func in CollectiveOp.COMM_TENSOR_ARG_0: + return CollectiveOp.sum_tensors(args[0]) + if func in CollectiveOp.COMM_TENSOR_ARG_1: + return CollectiveOp.sum_tensors(args[1]) + if func in CollectiveOp.COMM_TENSOR_ARG_RES: + return res.untyped_storage().nbytes() + if func in CollectiveOp.COMM_TENSOR_SINGLE_UNTYPED_STORAGE: + return args[0].untyped_storage().nbytes() + if func is c10d._reduce_scatter_base_.default: + return args[1].untyped_storage().nbytes() + if func is c10d.alltoall_.default: + # TODO(@sanketpurandare) - Confirm size computation + return max( + CollectiveOp.sum_tensors(args[0]), CollectiveOp.sum_tensors(args[1]) + ) + if func is c10d.alltoall_base_.default: + # TODO(@sanketpurandare) - Confirm size computation + return max( + args[0].untyped_storage().nbytes(), args[1].untyped_storage().nbytes() + ) + if func == _c10d_functional.all_gather_into_tensor_out.default: + return args[-1].untyped_storage().nbytes() + if func in CollectiveOp.COMM_TENSOR_RES_SUM: + return CollectiveOp.sum_tensors(res) + if func in CollectiveOp.COMM_TENSOR_ARG_0_AND_RES: + # TODO(@sanketpurandare) - Confirm size computation + return args[0].untyped_storage().nbytes() + res.untyped_storage().nbytes() + if func is _c10d_functional.batch_p2p_ops.default: + return CollectiveOp.sum_tensors(args[3]) + raise TypeError(f"Unknown function: {func} in {collective_ops}") + + @staticmethod + def get_work(func, res) -> Work: # type: ignore[no-untyped-def] + if func in CollectiveOp.WK: + return FakeWork.unbox(res) + elif func in CollectiveOp.WK_ARG_1: + return FakeWork.unbox(res[1]) + raise TypeError(f"Func {func} not found in {collective_ops}") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fsdp2_mem_tracker.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fsdp2_mem_tracker.py new file mode 100644 index 0000000000000000000000000000000000000000..c74f50d29821da65803f138846caae9861f77414 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/fsdp2_mem_tracker.py @@ -0,0 +1,595 @@ +from collections.abc import Callable +from copy import deepcopy +from enum import auto, Enum +from functools import partial, wraps +from typing import Any, NamedTuple, TYPE_CHECKING, TypeVar +from typing_extensions import ParamSpec, TypeVarTuple, Unpack + +import torch +import torch.distributed._tools.fake_collectives +from torch import nn, optim +from torch._guards import active_fake_mode +from torch.distributed._tools.mem_tracker import _RefType, _State, MemTracker +from torch.distributed.fsdp import FSDPModule +from torch.distributed.fsdp._fully_shard._fsdp_param_group import FSDPParamGroup +from torch.distributed.tensor import DTensor +from torch.utils._python_dispatch import TorchDispatchMode +from torch.utils._pytree import tree_map_only +from torch.utils.weak import WeakIdKeyDictionary, weakref + + +if TYPE_CHECKING: + from torch.utils.hooks import RemovableHandle + +_TOTAL_KEY = "Total" + +__all__ = ["FSDPMemTracker"] + +_P = ParamSpec("_P") +_R = TypeVar("_R") +_Ts = TypeVarTuple("_Ts") + +c10d = torch.ops.c10d + + +class _FSDPRefType(_RefType): + """ + Enumerates categories of memory usage in FSDP modules, including parameters, gradients, activations, + and optimizer states. + + Attributes: + SHARDED_PARAM (str): Memory usage of sharded parameters. + UNSHARDED_PARAM (str): Memory usage of unsharded parameters. + SHARDED_GRAD (str): Memory usage of sharded gradients corresponding to the sharded parameters. + UNSHARDED_GRAD (str): Memory usage of unsharded gradients corresponding to the unsharded parameters. + ACT (str): Memory usage of activations and tensors from forward and AC recomputation. + TEMP (str): Memory usage of temporary tensors during the backward pass including gradients of activations. + ALL_GATHER (str): Memory usage of all_gather output tensor. + REDUCE_SCATTER (str): Memory usage of reduce_scatter input tensor. + OPT (str): Memory usage of tensors storing optimizer states. + INP (str): Memory usage of input tensors. + """ + + SHARDED_PARAM = "Sharded Param" + UNSHARDED_PARAM = "Unsharded Param" + BUFFER = "Buffer" + SHARDED_GRAD = "Sharded Grad" + UNSHARDED_GRAD = "Unsharded Grad" + ACT = "Activation" + TEMP = "Temp" + ALL_GATHER = "All Gather" + REDUCE_SCATTER = "Reduce Scatter" + OPT = "OptState" + INP = "Inputs" + + +class _SavedFSDPMethods(NamedTuple): + pre_backward: Callable + post_backward: Callable + + +class _FSDPModState(_State): + """ + Enumerates the states of FSDP modules during the forward and backward passes. + """ + + BEF_PRE_FW = "Before Pre-Forward" + AFT_PRE_FW = "After Pre-Forward" + BEF_POST_FW = "Before Post-Forward" + AFT_POST_FW = "After Post-Forward" + BEF_PRE_BW = "Before Pre-Backward" + AFT_PRE_BW = "After Pre-Backward" + BEF_POST_BW = "Before Post-Backward" + AFT_POST_BW = "After Post-Backward" + PRE_FW_AC = "Pre-Forward AC" + POST_FW_AC = "Post-Forward AC" + PEAK_FW = "Peak Forward" + PEAK_BW = "Peak Backward" + + +class _FSDPModMemStats: + """ + A class to store the memory statistics of an FSDP module. + + Args: + mod_fqn (str): The fully qualified name of the FSDP module. + + Attributes: + snapshots (Dict[_FSDPModState, Dict[torch.device, Dict[str, int]]]): A dictionary of memory snapshots + of the module at different states as defined by ``_FSDPModState``. Each key is a device, and + each value is another dictionary with keys as memory reference types defined by ``_FSDPRefType`` and + values as the memory consumed in bytes. + + """ + + def __init__(self, mod_fqn: str) -> None: + self.mod_fqn = mod_fqn + self.local_peak: dict[torch.device, int] = {} + self.snapshots: dict[ + _FSDPModState, list[dict[torch.device, dict[str, int]]] + ] = {} + + +class _FSDPState(Enum): + PRE_FW = auto() + FW = auto() + POST_FW = auto() + PRE_BW = auto() + BW = auto() + POST_BW = auto() + + +class FSDPMemTracker(MemTracker): + """ + A ``TorchDispatchMode`` based context manager that extends ``torch.distributed._tools.mem_tracker.MemTracker`` to track + and categorize the peak memory and module-wise memory usage of FSDP modules. + + It tracks the peak memory usage across all the devices of all the FSDP modules in the module tree and categorizes + the tensor memory usage as defined by ``_FSDPRefType``. Further, it captures memory `snapshots` at different stages of + the module execution defined by ``_FSDPModState``. + + Attributes: + memory_tracking: A weakref key dictionary to store the memory statistics of each module. Each key is a reference + to a module, and each value is a ``_FSDPModMemStats`` object that stores the memory statistics of the module. + + Args: + mod (torch.nn.Module): The root FSDP module to be tracked. + optm (torch.optim.Optimizer, optional): The optimizer to be tracked. + + Note: Please refer to ``torch.distributed._tools.mem_tracker.MemTracker`` to learn about the limitations. + + Example usage + + .. code-block:: python + + module = ... + optimizer = ... + inp = ... + fmt = FSDPMemTracker(module, optimizer) + fmt.track_inputs((inp,)) + with fmt: + optimizer.zero_grad() + loss = module(inp) + print("After Forward:") + fmt.display_snapshot("current") + loss.backward() + optimizer.step() + fmt.display_snapshot("peak") + fmt.display_modulewise_snapshots(depth=3, units="MB") + + """ + + def __init__( + self, + mod: torch.nn.Module, + optm: torch.optim.Optimizer | None = None, + ) -> None: + super().__init__() + if not isinstance(mod, FSDPModule): + raise AssertionError("FSDPMemTracker only supports FSDP modules") + self._root_mod = mod + self._optm = optm + self._fsdp_mod_to_saved_methods: WeakIdKeyDictionary = WeakIdKeyDictionary() + self._fsdp_state: _FSDPState = _FSDPState.PRE_FW + self._ref_class: type[_RefType] = _FSDPRefType + + def _instrument_fsdp_sharded_params_grads( + self, fsdp_param_group: FSDPParamGroup + ) -> None: + # Track sharded params and grads after initialization + for fsdp_param in fsdp_param_group.fsdp_params: + self._update_and_maybe_create_winfos( + fsdp_param.sharded_param, + _FSDPRefType.SHARDED_PARAM, + ) + sharded_grad = fsdp_param.sharded_param.grad + if sharded_grad is not None: + self._update_and_maybe_create_winfos( + sharded_grad, + _FSDPRefType.SHARDED_GRAD, + ) + + def _fsdp_state_pre_forward( + self, + fsdp_mod: FSDPModule, + orig_fsdp_state_pre_fw: Callable[_P, tuple[tuple[Unpack[_Ts]], dict[str, Any]]], + ) -> Callable[_P, tuple[tuple[Unpack[_Ts]], dict[str, Any]]]: + # We capture memory snapshots before and after ``FSDPState._pre_forward`` to attribute the `unsharded` params + # and `all_gather` buffers. There are three cases: + # Case 1: If the module is not in the ``memory_tracking`` dictionary, create a new ``_FSDPModMemStats`` + # instance for the module and add it to the ``memory_tracking`` dictionary. + # Case 2: If the module is already in the ``memory_tracking`` dictionary and we are in backward, this means + # we are in the AC region. We check if this is the top most module in the AC region. If it is, + # we store a weak reference and set the flag ``_in_ac`` to True. + # Case 3: If the module is already in the ``memory_tracking`` dictionary and we are in forward, this means + # this module is called for the second time. If it is a root module, that means we are in the next + # iteration and we error out. If it is not a root module, that means it's a submodule that is being + # used multiple times in the same iteration, which we allow and track. + # For Case 1 and 3, we also initialize the ``local_peak`` and ``PEAK_FW`` snapshot for the module. + # For Case 2 we only capture 1 snapshot after ``FSDPState._pre_forward`` runs because it is a no-op. + @wraps(orig_fsdp_state_pre_fw) + def inner( + *args: _P.args, **kwargs: _P.kwargs + ) -> tuple[tuple[Unpack[_Ts]], dict[str, Any]]: + self._fsdp_state = _FSDPState.PRE_FW + mod_fqn = self._mod_tracker.get_known_fqn(fsdp_mod) + if mod_fqn is None: + raise AssertionError + if fsdp_mod not in self.memory_tracking: + mod_stat = _FSDPModMemStats(mod_fqn) + self.memory_tracking[fsdp_mod] = mod_stat + snapshot = self.get_tracker_snapshot() + mod_stat.local_peak = { + dev: dev_snap[_TOTAL_KEY] for dev, dev_snap in snapshot.items() + } + mod_stat.snapshots.setdefault(_FSDPModState.PEAK_FW, []).append( + snapshot + ) + mod_stat.snapshots.setdefault(_FSDPModState.BEF_PRE_FW, []).append( + deepcopy(snapshot) + ) + elif not self._mod_tracker.is_bw: + parents = self._mod_tracker.parents - {mod_fqn} + if len(parents) == 1 and "Global" in parents: + raise NotImplementedError( + "FSDPMemTracker does not support memory tracking for multiple iterative calls." + " Either use ``reset_mod_stats`` to clear module memory stats for the previous iteration" + " or file a github issue if you need this feature." + ) + + # pyrefly: ignore [bad-assignment] + args, kwargs = orig_fsdp_state_pre_fw(*args, **kwargs) + + fsdp_state = fsdp_mod._get_fsdp_state() + if fsdp_param_group := fsdp_state._fsdp_param_group: + for fsdp_param in fsdp_param_group.fsdp_params: + self._update_and_maybe_create_winfos( + fsdp_param.unsharded_param, + _FSDPRefType.UNSHARDED_PARAM, + ) + mod_stat = self.memory_tracking[fsdp_mod] + if self._mod_tracker.is_bw: + state = _FSDPModState.PRE_FW_AC + if self._ac_mod is None: + self._ac_mod = weakref.ref(fsdp_mod) + self._in_ac = True + else: + state = _FSDPModState.AFT_PRE_FW + mod_stat.snapshots.setdefault(state, []).append(self.get_tracker_snapshot()) + self._fsdp_state = _FSDPState.FW + # pyrefly: ignore [bad-return] + return args, kwargs + + return inner + + def _fsdp_state_post_forward( + self, + fsdp_mod: FSDPModule, + orig_fsdp_state_post_fw: Callable[_P, _R], + ) -> Callable[_P, _R]: + # We capture memory snapshots before and after ``FSDPState._post_forward`` to capture the resharded state + # if ``reshard_after_forward`` is not ``False``. There are two cases: + # Case 1: This is called in backward, which means we are in the AC region. If this is the top most module + # in the AC region, we set the flag ``_in_ac`` to False. + # Case 2: This is called in forward. + @wraps(orig_fsdp_state_post_fw) + def inner(*args: _P.args, **kwargs: _P.kwargs) -> _R: + mod_stat = self.memory_tracking[fsdp_mod] + if self._mod_tracker.is_bw: + state = _FSDPModState.POST_FW_AC + if self._ac_mod is not None and self._ac_mod() is fsdp_mod: + self._ac_mod = None + self._in_ac = False + else: + state = _FSDPModState.BEF_POST_FW + mod_stat.snapshots.setdefault(state, []).append(self.get_tracker_snapshot()) + self._fsdp_state = _FSDPState.POST_FW + + output = orig_fsdp_state_post_fw(*args, **kwargs) + + if not self._mod_tracker.is_bw: + mod_stat.snapshots.setdefault(_FSDPModState.AFT_POST_FW, []).append( + self.get_tracker_snapshot() + ) + return output + + return inner + + def _fsdp_param_group_pre_backward( + self, + fsdp_mod: FSDPModule, + orig_fsdp_param_group_pre_backward: Callable[_P, Any], + ) -> Callable[_P, None]: + # We capture memory snapshots before and after ``FSDPParamGroup.pre_backward`` to capture the pre-fetching + # and unsharding of params. We also initialize ``local_peak`` and ``PEAK_BW`` snapshot for the module. + @wraps(orig_fsdp_param_group_pre_backward) + def inner(*args: _P.args, **kwargs: _P.kwargs) -> None: + self._fsdp_state = _FSDPState.PRE_BW + mod_stat = self.memory_tracking[fsdp_mod] + snapshot = self.get_tracker_snapshot() + mod_stat.local_peak = { + dev: dev_snap[_TOTAL_KEY] for dev, dev_snap in snapshot.items() + } + mod_stat.snapshots.setdefault(_FSDPModState.PEAK_BW, []).append(snapshot) + mod_stat.snapshots.setdefault(_FSDPModState.BEF_PRE_BW, []).append( + deepcopy(snapshot) + ) + orig_fsdp_param_group_pre_backward(*args, **kwargs) + + mod_stat.snapshots.setdefault(_FSDPModState.AFT_PRE_BW, []).append( + self.get_tracker_snapshot() + ) + self._fsdp_state = _FSDPState.BW + + return inner + + def _fsdp_param_group_post_backward( + self, + fsdp_mod: FSDPModule, + orig_fsdp_param_group_post_backward: Callable[_P, Any], + ) -> Callable[_P, None]: + # We capture the memory snapshots before and after ``FSDPParamGroup.post_backward`` to track and attribute + # the `unsharded` grads before the post backward and then `sharded` grads and `reduce_scatter` buffers + # after the post backward. + @wraps(orig_fsdp_param_group_post_backward) + def inner(*args: _P.args, **kwargs: _P.kwargs) -> None: + fsdp_state = fsdp_mod._get_fsdp_state() + if fsdp_param_group := fsdp_state._fsdp_param_group: + for fsdp_param in fsdp_param_group.fsdp_params: + unsharded_grad = fsdp_param._unsharded_param.grad + if unsharded_grad is not None: + self._update_and_maybe_create_winfos( + unsharded_grad, + _FSDPRefType.UNSHARDED_GRAD, + update_existing=True, + ) + + mod_stat = self.memory_tracking[fsdp_mod] + mod_stat.snapshots.setdefault(_FSDPModState.BEF_POST_BW, []).append( + self.get_tracker_snapshot() + ) + self._fsdp_state = _FSDPState.POST_BW + orig_fsdp_param_group_post_backward(*args, **kwargs) + + if fsdp_param_group := fsdp_state._fsdp_param_group: + for fsdp_param in fsdp_param_group.fsdp_params: + sharded_grad = fsdp_param.sharded_param.grad + if sharded_grad is not None: + self._update_and_maybe_create_winfos( + sharded_grad, + _FSDPRefType.SHARDED_GRAD, + ) + + mod_stat.snapshots.setdefault(_FSDPModState.AFT_POST_BW, []).append( + self.get_tracker_snapshot() + ) + + return inner + + def _instrument_fsdp_module(self) -> None: + # We uninstall the existing `FSDPState._pre_forward` and `FSDPState._post_forward` hooks and install + # our own hooks that wrap them. We choose this over monkey-patching `FSDPParamGroup.pre_forward` and + # `FSDPParamGroup.post_forward` because during AC these won't be called. + # TODO(@sanketpurandare): This will need to be modified after this PR (https://github.com/pytorch/pytorch/pull/127786) + # lands. For backward we monkey-patch the `FSDPParamGroup.pre_backward` and `FSDPParamGroup.post_backward`. + + # get the unique _MultiHandlers/RemoveHandlers and store in dictionary + # the _MultiHandlers object will only need to be grabbed once. + unique_handlers: dict[RemovableHandle, bool] = {} + + for module in self._root_mod.modules(): + if isinstance(module, FSDPModule): + fsdp_state = module._get_fsdp_state() + if fsdp_param_group := fsdp_state._fsdp_param_group: + if not unique_handlers.get(fsdp_state._pre_forward_hook_handle): + unique_handlers[fsdp_state._pre_forward_hook_handle] = True + if not unique_handlers.get(fsdp_state._post_forward_hook_handle): + unique_handlers[fsdp_state._post_forward_hook_handle] = True + # call remove on the handles once + for f_hook_handle in unique_handlers: + f_hook_handle.remove() + + for module in self._root_mod.modules(): + if isinstance(module, FSDPModule): + fsdp_state = module._get_fsdp_state() + if fsdp_param_group := fsdp_state._fsdp_param_group: + self._instrument_fsdp_sharded_params_grads(fsdp_param_group) + fsdp_state._pre_forward_hook_handle = ( + module.register_forward_pre_hook( + self._fsdp_state_pre_forward( + module, fsdp_state._pre_forward + ), + prepend=True, + with_kwargs=True, + ) + ) + + fsdp_state._post_forward_hook_handle = module.register_forward_hook( + self._fsdp_state_post_forward(module, fsdp_state._post_forward), + prepend=False, + always_call=True, + ) + self._fsdp_mod_to_saved_methods[module] = _SavedFSDPMethods( + fsdp_param_group.pre_backward, + fsdp_param_group.post_backward, + ) + fsdp_param_group.pre_backward = self._fsdp_param_group_pre_backward( # type: ignore[assignment] + module, fsdp_param_group.pre_backward + ) + fsdp_param_group.post_backward = ( # type: ignore[assignment] + self._fsdp_param_group_post_backward( + module, fsdp_param_group.post_backward + ) + ) + + for buffer in self._root_mod.buffers(): + self._update_and_maybe_create_winfos( + buffer, + _FSDPRefType.BUFFER, + ) + + def _instrument_optimizer(self) -> None: + # Register a hook on the optimizer step to track the optimizer states. + # The pre-hook is to set the flag ``_in_opt`` to True. The post-hook unsets the flag, + # and also tracks any optimizer states that are created during the optimizer step. + if self._optm is not None: + self._track_optimizer_states(_FSDPRefType.OPT, self._optm) + + def _opt_step_pre_hook( + optimizer: optim.Optimizer, args: Any, kwargs: Any + ) -> None: + self._in_opt = True + + def _opt_step_post_hook( + optimizer: optim.Optimizer, args: Any, kwargs: Any + ) -> None: + self._track_optimizer_states(_FSDPRefType.OPT, optimizer) + self._in_opt = False + + self._optimizer_hook_handles = ( + self._optm.register_step_pre_hook(_opt_step_pre_hook), + self._optm.register_step_post_hook(_opt_step_post_hook), + ) + + def _register_module_and_optimizer_hooks(self) -> None: + self._instrument_fsdp_module() + self._instrument_optimizer() + + def _deregister_module_and_optimizer_hooks(self) -> None: + for ( + fsdp_mod, + saved_methods, + ) in self._fsdp_mod_to_saved_methods.items(): + fsdp_state = fsdp_mod._get_fsdp_state() + fsdp_state._pre_forward_hook_handle.remove() + fsdp_state._post_forward_hook_handle.remove() + fsdp_state._pre_forward_hook_handle = fsdp_mod.register_forward_pre_hook( + fsdp_state._pre_forward, prepend=True, with_kwargs=True + ) + fsdp_state._post_forward_hook_handle = fsdp_mod.register_forward_hook( + fsdp_state._post_forward, prepend=False + ) + if fsdp_param_group := fsdp_state._fsdp_param_group: + fsdp_param_group.pre_backward = saved_methods.pre_backward + fsdp_param_group.post_backward = saved_methods.post_backward + self._fsdp_mod_to_saved_methods.clear() + + if self._optimizer_hook_handles is not None: + for handle in self._optimizer_hook_handles: + handle.remove() + self._optimizer_hook_handles = None + + def track_inputs(self, inputs: tuple[Any, ...]) -> None: + """ + This is used to track the input tensors to the model and annotate them as ``Inputs``. + Args: + inputs (Tuple[Any]): A tuple containing the input data. This can include tensors + as well as other data types. Only tensors will be tracked. + """ + + def _track_inputs(t: torch.Tensor) -> None: + self._update_and_maybe_create_winfos( + t, + _FSDPRefType.INP, + ) + + tree_map_only(torch.Tensor, _track_inputs, inputs) + + def track_external( + self, *external: nn.Module | optim.Optimizer | torch.Tensor + ) -> None: + """This is no-op for ``FSDPMemTracker``""" + + def __enter__(self) -> "FSDPMemTracker": + if self._depth == 0: + # None in eager, a FakeTensorMode instance in SAC. Used in + # __torch_dispatch__ to skip DTensor propagation ops. + self._fake_mode_on_entry = active_fake_mode() + self._register_module_and_optimizer_hooks() + self._track_resize() + self._peak_mem_snap = self.get_tracker_snapshot() + self._peak_mem = { + dev: dev_snap[_TOTAL_KEY] + for dev, dev_snap in self._peak_mem_snap.items() + } + self._mod_tracker.__enter__() + TorchDispatchMode.__enter__(self) + self._depth += 1 + return self + + def __exit__(self, *args: Any) -> None: + self._depth -= 1 + if self._depth == 0: + self._deregister_module_and_optimizer_hooks() + self._restore_resize() + self._mod_tracker.__exit__(*args) + TorchDispatchMode.__exit__(self, *args) + + def __torch_dispatch__(self, func, types, args=..., kwargs=None): # type: ignore[no-untyped-def] + # When running this mode with DTensor, ordinarily all modes will + # run **before** subclasses get a chance to run. + # Returning NotImplemented here gives us a chance to let DTensor + # run and desugar into local tensor ops, before `MemTracker` sees them. + if any(t == DTensor for t in types): + return NotImplemented + if ( + func is torch.ops._c10d_functional.wait_tensor.default + and active_fake_mode() + ): + # N.B: This is a hacky way to override the Meta IMPL of wait_tensor. The original impl returns + # a new tensor which does not happen in eager mode, when a wait_tensor is called. + # pyrefly: ignore [unsupported-operation] + res = args[0] + else: + res = func(*args, **kwargs or {}) + # DTensor sharding propagation may use a nested FakeTensorMode to + # compute output metadata (shapes, placements). The real memory + # usage comes from the local op on shard data, which + # runs outside DTensor's fake mode. We use identity comparison + # against the fake mode at tracker entry to skip propagation ops + # while still tracking local ops in both eager and SAC contexts. + # - Eager (entry mode is None): real ops have no fake mode + # (None is None), DTensor propagation ops are skipped (B is not None). + # - SAC (entry mode is A): SAC ops run under mode A (A is A), + # DTensor propagation ops are skipped (B is not A). + if active_fake_mode() is self._fake_mode_on_entry: + # If we are tracking an optimizer state, we use the optimizer reference type. + # If we are in backward region and not in AC region, we use the backward reference type. + # Else we use the forward reference type. + if self._in_opt: + reftype = _FSDPRefType.OPT + elif self._mod_tracker.is_bw and not self._in_ac: + reftype = _FSDPRefType.TEMP + else: + reftype = _FSDPRefType.ACT + if func is c10d._allgather_base_.default and self._fsdp_state in [ + _FSDPState.PRE_FW, + _FSDPState.PRE_BW, + ]: + # pyrefly: ignore [unsupported-operation] + output_tensor = args[0] + self._update_and_maybe_create_winfos( + output_tensor, + _FSDPRefType.ALL_GATHER, + update_existing=True, + ) + if ( + func is c10d._reduce_scatter_base_.default + and self._fsdp_state == _FSDPState.POST_BW + ): + # pyrefly: ignore [unsupported-operation] + input_tensor = args[1] + self._update_and_maybe_create_winfos( + input_tensor, + _FSDPRefType.REDUCE_SCATTER, + update_existing=True, + ) + + tree_map_only(torch.Tensor, partial(self._track, reftype), res) + peak_state = ( + _FSDPModState.PEAK_BW + if self._mod_tracker.is_bw + else _FSDPModState.PEAK_FW + ) + self._update_peak_stats(peak_state) + return res diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/ilp_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/ilp_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..15a80ed554d88b74a2ac19d11af4418af2c2ec8d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/ilp_utils.py @@ -0,0 +1,293 @@ +import copy +from collections import OrderedDict +from typing import cast, TypedDict + +import numpy as np + +import torch +from torch.distributed._tools.mem_tracker import ( + _MemRefType, + _ModMemStats, + _ModState, + MemTracker, +) +from torch.distributed._tools.runtime_estimator import RuntimeEstimator +from torch.distributed._tools.sac_estimator import SACEstimator, SACTradeOffStats + + +class ModOrder(TypedDict): + fw_pre_order: list[str] + bw_pre_order: list[str] + fw_post_order: list[str] + bw_post_order: list[str] + + +class ModRuntime(TypedDict): + fw: float + bw: float + + +class ModStats(TypedDict): + fqn: str + # per-module params + param_per_module: int + # per-module grads + grad_per_module: int + # total accumulated gradients up to and including this module + grad_total: int + # per module fw activation size (excluding input and output) + act_fw_per_module: int + # per module bw activation size during peak_bw + act_bw_per_module: int + # per module activation grad size during peak_bw + act_grad_per_module: int + # total activation size up to but excluding the current module + # includes input of the current module (i.e., output of previous module) + act_total: int + # Inputs to the module + input_per_module: int + # Outputs of the module + output_per_module: int + # Total fw run-time of the module + fw_runtime_per_module: float + # Total bw run-time of the module + bw_runtime_per_module: float + # Is this module a leaf module + is_leaf: bool + # Total ac run-time of the module + sac_runtime: float + # Total ac_memory for the module + sac_memory: int + # Number of piecewise-linear functions used for approximating ac tradeoff curve + n_segments: int + # Slopes of the of piecewise-linear functions + slopes: list[float] + # Intercepts of the of piecewise-linear functions + intercepts: list[float] + # X breakpoints of the of piecewise-linear functions + breakpoints: list[float] + # Original trade-off curves + tradeoff_curve: OrderedDict[float, float] + + +class ModuleInfo(TypedDict): + mod_order: ModOrder + mod_stats: list[ModStats] + + +def aggregate_stats( + model: torch.nn.Module, + mem_tracker: MemTracker, + runtime_estimator: RuntimeEstimator, + sac_estimator: SACEstimator, + dev: torch.device, +) -> ModuleInfo: + """ + Collect modulewise stats for a given model, including memory, runtime, and AC tradeoff stats. + + Args: + model: nn.Module object + runtime_estimator: RuntimeEstimator object with runtime stats + mem_tracker: MemTracker object with memory stats + sac_estimator: SACEstimator object with AC tradeoff stats + dev: device the model was run on (used to extract memory stats from MemTracker) + + Returns: + ModuleInfo: A dictionary with module order and module stats. + """ + + # Memory stats + mod_mem_stats: dict[torch.nn.Module, _ModMemStats] = dict( + copy.deepcopy(mem_tracker.memory_tracking) + ) + + # Runtime stats + mod_runtime_stats: dict[str, ModRuntime] = { + fqn: {"fw": v["fw"], "bw": v["bw"]} + for fqn, v in runtime_estimator.mod_runtimes.items() + } + + # Module order + mod_order: ModOrder = { + "fw_pre_order": list(runtime_estimator.mod_fw_pre_order), + "bw_pre_order": list(runtime_estimator.mod_bw_pre_order), + "fw_post_order": list(runtime_estimator.mod_fw_post_order), + "bw_post_order": list(runtime_estimator.mod_bw_post_order), + } + + # Selective Activation Checkpointing stats + sac_estimator.pwlf_sac_tradeoff_curve() + mod_sac_tradeoff_stats: dict[str, SACTradeOffStats] = copy.deepcopy( + sac_estimator.sac_mod_tradeoff_stats + ) + + module_info: ModuleInfo = { + "mod_order": mod_order, + "mod_stats": [], + } + + for mod in model.modules(): + if mod_mem_stat := mod_mem_stats.get(mod): + if tradeoff_stats := mod_sac_tradeoff_stats.get(mod_mem_stat.mod_fqn): + sac_runtime = tradeoff_stats.sac_runtime + sac_memory = tradeoff_stats.sac_memory + n_segments = tradeoff_stats.n_segments + slopes = tradeoff_stats.slopes + intercepts = tradeoff_stats.intercepts + breakpoints = tradeoff_stats.fit_breaks + tradeoff_curve = tradeoff_stats.tradeoff_curve + is_leaf = False + else: + sac_runtime = sac_memory = n_segments = 0 + slopes = intercepts = breakpoints = [] + tradeoff_curve: OrderedDict[float, float] = OrderedDict() # type: ignore[no-redef] + is_leaf = True + mod_stat: ModStats = { + "fqn": mod_mem_stat.mod_fqn, + "param_per_module": mod_mem_stat.parameter_mem, + "grad_per_module": mod_mem_stat.parameter_mem, + "grad_total": mod_mem_stat.snapshots[_ModState.PRE_BW][-1][dev][ + _MemRefType.GRAD + ], + "act_fw_per_module": max( + 0, + mod_mem_stat.snapshots[_ModState.POST_FW][-1][dev][_MemRefType.ACT] + - mod_mem_stat.snapshots[_ModState.PRE_FW][-1][dev][_MemRefType.ACT] + - mod_mem_stat.output_mem, + ), + "act_bw_per_module": max( + 0, + mod_mem_stat.snapshots[_ModState.PEAK_BW][-1][dev][_MemRefType.ACT], + ), + "act_grad_per_module": ( + mod_mem_stat.snapshots[_ModState.PEAK_BW][-1][dev][_MemRefType.TEMP] + - mod_mem_stat.snapshots[_ModState.PRE_BW][-1][dev][ + _MemRefType.TEMP + ] + ), + "act_total": mod_mem_stat.snapshots[_ModState.POST_FW][-1][dev][ + _MemRefType.ACT + ], + "input_per_module": mod_mem_stat.input_mem, + "output_per_module": mod_mem_stat.output_mem, + "fw_runtime_per_module": mod_runtime_stats[mod_mem_stat.mod_fqn]["fw"], + "bw_runtime_per_module": mod_runtime_stats[mod_mem_stat.mod_fqn]["bw"], + "is_leaf": is_leaf, + "sac_runtime": sac_runtime, + "sac_memory": sac_memory, + "n_segments": n_segments, + "slopes": slopes, + "intercepts": intercepts, + "breakpoints": breakpoints, + "tradeoff_curve": tradeoff_curve, + } + module_info["mod_stats"].append(mod_stat) + + return module_info + + +class Node(ModStats): + index: int # index according to forward pre-order + pos_fw_post_order: int # index according to forward post-order + + +class Graph: + def __init__(self, n: int) -> None: + self.nodes: list[Node] = [] + self.name2node: dict[str, Node] = {} + self.ad_matrix = np.zeros((n, n)) + self.fw_post_order: list[str] = [] + + def add_node(self, node: Node) -> None: + self.nodes.append(node) + self.name2node[node["fqn"]] = node + + +def parse_module_info(module_info: ModuleInfo) -> Graph: + """ + Parse module info and create a graph (tree) of modules. The graph will be + used by MILP solver to find optimal SAC and/or FSDP configurations. + """ + mod_stats = module_info["mod_stats"] + fw_pre_order = module_info["mod_order"]["fw_pre_order"] + # assertion and number of nodes + if len(mod_stats) != len(fw_pre_order): + raise AssertionError + n_nodes = len(mod_stats) + + # create graph + g = Graph(n_nodes) + g.fw_post_order = module_info["mod_order"]["fw_post_order"] + + # sort the modules by pre-order and add them to the graph + module_info["mod_stats"] = sorted( + mod_stats, key=lambda x: fw_pre_order.index(x["fqn"]) + ) + for i, one_mod_stats in enumerate(mod_stats): + node: Node = cast(Node, one_mod_stats) + node["index"] = i + node["pos_fw_post_order"] = g.fw_post_order.index(node["fqn"]) + g.add_node(node) + + # set up ancestor-descendant matrix + for i in range(n_nodes): + for j in range(i, n_nodes): + if is_self_or_submodule(g.nodes[j]["fqn"], g.nodes[i]["fqn"]): + g.ad_matrix[i][j] = 1 + else: + break + + return g + + +def is_self_or_submodule(name_descendant: str, name_ancestor: str) -> bool: + """ + check if name_descendant is a submodule of name_ancestor, or if they are the same + """ + return name_descendant == name_ancestor or name_ancestor + "." in name_descendant + + +def is_submodule(name_descendant: str, name_ancestor: str) -> bool: + """ + if name_descendant is a submodule of name_ancestor, but not the same + """ + return name_ancestor + "." in name_descendant + + +def display_bytes(b: int, unit: str = "MiB") -> str: + """ + return a string that represent the number of bytes in a desired unit + """ + if unit == "KiB": + return f"{b / 2**10:.2f} KiB" + if unit == "MiB": + return f"{b / 2**20:.2f} MiB" + if unit == "GiB": + return f"{b / 2**30:.2f} GiB" + return f"{b:.2f} bytes" + + +def get_peak_memory_runtime_baseline(graph: Graph) -> tuple[int, float]: + """ + Get the baseline peak memory and runtime. + Baseline here means there is no FSDP or AC. + Memory includes the parameters, gradients, activations, and activation gradients. + Memory does not include e.g., optimizer states, embedding tables, etc. + + Returns: + int: peak memory in bytes + float: compute time in ms + """ + P_1 = graph.nodes[0]["param_per_module"] + num_nodes = len(graph.nodes) + peak_mem = 0 + for i in range(num_nodes): + TG_i = graph.nodes[i]["grad_total"] + AG_i = graph.nodes[i]["act_grad_per_module"] + TA_i = graph.nodes[i]["act_total"] + peak_mem = max(peak_mem, P_1 + TG_i + AG_i + TA_i) + compute_time = ( + graph.nodes[0]["fw_runtime_per_module"] + + graph.nodes[0]["bw_runtime_per_module"] + ) + return (peak_mem, compute_time) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mem_tracker.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mem_tracker.py new file mode 100644 index 0000000000000000000000000000000000000000..57d4fec5f09bfee9afedde03d7abb3c14786400f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mem_tracker.py @@ -0,0 +1,956 @@ +import math +import os +import re +import warnings +from collections.abc import Callable +from copy import deepcopy +from enum import auto, Enum +from functools import partial, wraps +from typing import Any, TYPE_CHECKING +from typing_extensions import Self + +import torch +import torch.distributed._tools.fake_collectives +from torch import nn, optim +from torch._guards import active_fake_mode +from torch.distributed._tools.common_utils import get_untyped_storages +from torch.distributed._tools.mod_tracker import ModTracker +from torch.distributed.tensor import DTensor +from torch.optim.optimizer import ( + register_optimizer_step_post_hook, + register_optimizer_step_pre_hook, +) +from torch.utils._python_dispatch import TorchDispatchMode +from torch.utils._pytree import tree_flatten, tree_map_only +from torch.utils.weak import WeakIdKeyDictionary, weakref + + +if TYPE_CHECKING: + from torch.utils.hooks import RemovableHandle + +# This value is hard-coded here: +# https://github.com/pytorch/pytorch/blob/5fba5d83f0703ff8077ab65448a998e9ad6598fd/c10/cuda/CUDACachingAllocator.cpp#L117 +_PYTORCH_MIN_ALLOCATE = ( + 2**9 if int(os.environ.get("PYTORCH_NO_CUDA_MEMORY_CACHING", 0)) == 0 else 1 +) +_TOTAL_KEY = "Total" + +__all__ = ["MemTracker"] + + +class _RefType(str, Enum): + """Base Class for defining memory reference types, categorizing tensors based on their usage within a model.""" + + +class _State(str, Enum): + """Base Class for defining module state to capture snapshots .""" + + +class _MemRefType(_RefType): + """ + An enum to define memory reference types, categorizing tensors based on their usage within a model. + + - PARAM: Tensors registered as nn.Parameter within modules. + - BUFFER: Tensors registered as nn.Buffer within modules. + - GRAD: Gradients associated with parameters. + - ACT: Tensors produced during the forward pass and recomputation in activation checkpointing. + - TMP: Temporary memory used during the backward pass, including gradients of activations. + - OPT: Tensors holding optimizer states. + - OTH: Tensors registered via `track_external` that do not fit the above categories. + """ + + PARAM = "Parameter" + BUFFER = "Buffer" + GRAD = "Gradient" + ACT = "Activation" + TEMP = "Temp" + OPT = "Optstate" + OTH = "Other" + + +class _ModState(_State): + """ + An enum to define the state of a module. + + - PRE_FW: The module is about to run the forward pass. + - POST_FW: The module has finished running the forward pass. + - PEAK_FW: The module has reached the peak memory usage during the forward pass. + - PRE_BW: The module is about to run the backward pass. + - PRE_FW_AC: The module is about to run the forward pass with activation checkpointing. + - POST_FW_AC: The module has finished running the forward pass with activation checkpointing. + - POST_BW: The module has finished running the backward pass. + - PEAK_BW: The module has reached the peak memory usage during the backward pass. + """ + + PRE_FW = "Pre-Forward" + POST_FW = "Post-Forward" + PEAK_FW = "Peak-Forward" + PRE_BW = "Pre-Backward" + PRE_FW_AC = "Pre-Forward-AC" + POST_FW_AC = "Post-Forward-AC" + POST_BW = "Post-Backward" + PEAK_BW = "Peak-Backward" + + +class _ModMemStats: + """ + A class to store the memory statistics of a module. + + Args: + mod_fqn (str): The fully qualified name of the module. + Attributes: + mod_fqn (str): The fully qualified name of the module. + parameter_mem (int): The memory usage of the parameters of the module. + buffer_mem (int): The memory usage of the buffers of the module. + input_mem (int): The memory usage of the inputs to the module. + output_mem (int): The memory usage of the outputs from the module. + snapshots (Dict[_ModState, Dict[torch.device, Dict[str, int]]]): A dictionary of memory snapshots + of the module at different states defined by ``_ModState``. + Note: + The memory snapshot is stored as a dictionary - Dict[torch.device, Dict[str, int]], where each key is a device, + and each value is another dictionary with keys as memory reference types defined by `_MemRefType` and + values as the memory consumed in bytes. + """ + + def __init__(self, mod_fqn: str): + self.mod_fqn = mod_fqn + self.parameter_mem: int + self.buffer_mem: int + self.input_mem: int + self.output_mem: int + self.local_peak: dict[torch.device, int] = {} + self.snapshots: dict[_ModState, list[dict[torch.device, dict[str, int]]]] = {} + + +class _WeakRefInfo: + """ + Manages memory statistics and device attributes for tensor storages. + """ + + def __init__( + self, size: int, element_size: int, device: torch.device, reftype: _RefType + ) -> None: + """ + Initializes the ``_WeakRefInfo`` object with tensor storage properties. + + Args: + size (int): The number of elements in the tensor storage. + element_size (int): The size of each element in the tensor storage. + device (torch.device): The device on which the tensor is allocated. + reftype (_RefType): The reference type of the tensor. + """ + self.size = size + self.element_size = element_size + self.reftype = reftype + self.device = device + self.mem_consumed = self._calculate_mem_consumed() + + def _calculate_mem_consumed(self) -> int: + """ + Calculates the memory consumed by the tensor storage, considering device-specific allocation rules. + + Returns: + int: The memory consumed in bytes. + """ + mem = self.size * self.element_size + if self.device.type == "cuda": + return math.ceil((mem) / _PYTORCH_MIN_ALLOCATE) * _PYTORCH_MIN_ALLOCATE + return mem + + def update_mem_consumed(self, st: torch.UntypedStorage) -> int: + """ + Updates and returns the memory consumed if the storage size has changed. + + Args: + st (torch.UntypedStorage): The tensor storage to check for size updates. + + Returns: + int: The updated memory consumed in bytes. + """ + if st.size() != self.size: + self.size = st.size() + self.mem_consumed = self._calculate_mem_consumed() + return self.mem_consumed + + @classmethod + def create_winfo( + cls, + st: torch.UntypedStorage, + device: torch.device, + reftype: _RefType, + callback: Callable[[Self, weakref.ref], Any] | None = None, + ) -> tuple[Self, weakref.ref]: + """ + Creates a new ``_WeakRefInfo`` instance and a weak reference to a ``torch.UntypedStorage`` object, + optionally attaching a callback to the weak reference. + + Args: + st (torch.UntypedStorage): The storage object for which to create the weak reference info. + device (torch.device): The device associated with the storage object. + reftype (_RefType): The type of reference, used to categorize the storage. + callback (Optional[Callable[[Self, weakref.ref]]]): A callback function that is called when + the storage object is about to be finalized (garbage collected). The callback function + should accept two arguments: the ``_WeakRefInfo`` instance and the weak reference to the storage. + Returns: + Tuple[Self, weakref.ref]: A tuple containing the newly created ``_WeakRefInfo`` instance and the + weak reference to the storage object. The weak reference may have an attached callback if provided. + """ + + winfo = cls(st.size(), st.element_size(), device, reftype) + w_st = weakref.ref(st, partial(callback, winfo) if callback else None) + return winfo, w_st + + +def _get_mem_divisor(units: str) -> int: + unit_dict = {"B": 1, "KiB": 2**10, "MiB": 2**20, "GiB": 2**30} + if units in unit_dict: + return unit_dict[units] + else: + raise ValueError( + f"Unsupported unit: {units}. Supported units are: {', '.join(unit_dict.keys())}" + ) + + +def _rounding_fn(value: int, divisor: int, precision: int) -> float | int: + return value if divisor == 1 else round(value / divisor, precision) + + +def _print_snapshot(snapshot: dict[torch.device, dict[str, int]], units: str) -> None: + if len(snapshot) == 0: + print("No memory tracked.") + return + divisor = _get_mem_divisor(units) + for dev, dev_snap in snapshot.items(): + if _rounding_fn(dev_snap[_TOTAL_KEY], divisor, 2) <= 0: + continue + print( + f"Device: {dev}", + *( + f"\t{k.value}: {_rounding_fn(v, divisor, 2)} {units}" + if isinstance(k, _RefType) + else f"\t{k}: {_rounding_fn(v, divisor, 2)} {units}" + for k, v in dev_snap.items() + ), + sep="\n", + ) + + +def _print_snapshot_tabular( + snapshot: dict[torch.device, dict[str, int]], units: str +) -> None: + if len(snapshot) == 0: + print("No memory tracked.") + return + try: + from tabulate import tabulate + except ImportError as err: + raise ImportError( + "Please install tabulate to use the tabulate option." + ) from err + divisor = _get_mem_divisor(units) + table_data = [] + key_list = list(next(iter(snapshot.values())).keys()) + headers = ["Device"] + [ + f"{key.value}" if isinstance(key, _RefType) else f"{key}" for key in key_list + ] + + for dev, dev_snap in snapshot.items(): + if _rounding_fn(dev_snap[_TOTAL_KEY], divisor, 2) <= 0: + continue + row = [str(dev)] + row.extend(f"{_rounding_fn(v, divisor, 2)} {units}" for v in dev_snap.values()) + table_data.append(row) + print(tabulate(table_data, headers=headers, tablefmt="rst")) + + +def _print_state_snapshots( + snapshots: dict[_State, list[dict[torch.device, dict[str, int]]]], units: str +) -> None: + for state, snapshot_list in snapshots.items(): + print(f"{state.value}") + for i, snapshot in enumerate(snapshot_list): + print(f"# {i + 1}:") + _print_snapshot(snapshot, units) + print() + + +def _print_state_snapshots_tabular( + snapshots: dict[_State, list[dict[torch.device, dict[str, int]]]], units: str +) -> None: + try: + from tabulate import tabulate + except ImportError as err: + raise ImportError( + "Please install tabulate to use the tabulate option." + ) from err + + table_data = [] + last_state_call = None + divisor = _get_mem_divisor(units) + for state, snapshot_list in snapshots.items(): + for i, snapshot in enumerate(snapshot_list): + state_call = f"{state.value} # {i + 1}" + for dev, dev_snap in snapshot.items(): + if _rounding_fn(dev_snap[_TOTAL_KEY], divisor, 2) <= 0: + continue + row = { + "State & Call": ( + state_call if state_call != last_state_call else "" + ), + "Device": str(dev), + } + last_state_call = state_call + for k, v in dev_snap.items(): + row[f"{k.value}" if isinstance(k, _RefType) else f"{k}"] = ( + f"{_rounding_fn(v, divisor, 2)} {units}" + ) + table_data.append(row) + print(tabulate(table_data, headers="keys", tablefmt="rst")) + + +class _UpdateType(Enum): + # These are used for tracking updates to the continuously maintained memory snapshot. + # ADD - When a new tensor storage is tracked + # DEL - When a tensor storage is about to be finalized (garbage collected). + # REF - When a tensor reference is updated, for instance, the gradients are marked as + # generic backward reference types until the grad_hook categorizes them as gradients. + # SIZE - When a tensor's storage is resized. + ADD = auto() + DEL = auto() + REF = auto() + SIZE = auto() + + +class MemTracker(TorchDispatchMode): + """ + A TorchDispatchMode to track, categorize and attribute the tensor memory created or accessed within its context. + + It categorizes the tracked tensors as parameters, buffers, activations, gradients, temporary memory and optimizer states + as defined by ``_MemRefType`` within its context. It captures memory `snapshots` for the modules, called within its context, + at various states defined by ``_ModState``. + + Attributes: + memory_tracking: A weakref key dictionary to store the memory statistics of each module. Each key + is a reference to a module, and each value is a ``_ModMemStats`` object that stores the memory + statistics of the module. + + Note: + The MemTracker should be used as a context manager. The modules, optimizers, and any other tensors created within + the context of MemTracker will be tracked by default. Any tensors or stateful objects such as modules, optimizers etc. + that need to be tracked but are created outside the MemTracker should be registered using the `track_external` method. + The `track_external` method should be called before the MemTracker is used. Any tensors created outside the ``MemTracker`` + and not supplied to the `track_external` method will not be tracked by the ``MemTracker``. + + Example usage: + + .. code-block:: python + + module = ... + optimizer = ... + inp = ... + mem_tracker = MemTracker() + mem_tracker.track_external(module, optimizer, inp) + with mem_tracker as mt: + loss = module(inp) + print("After Forward:") + mt.display_snapshot("current") + loss.backward() + optimizer.step() + optimizer.zero_grad() + mt.display_snapshot("peak") + mt.display_modulewise_snapshots(depth=3, units="MiB") + + Known Limitations: + - The ``MemTracker`` does not track memory for tensors that bypass the ``TorchDispatchMode`` ex. under ``no_dispatch``. + - Resizing tensor storages directly by using non-Tensor methods other than using ``torch.Untyped_Storage.resize_`` + is not tracked. File a Github issue if you have use-cases for this. + - If the tensors are not traceable or wrappable subclasses of ``torch.Tensor``, then the tracker does not know how to + track their storages. File a Github issue if you have use-cases for this. + - During AC in the backward pass there might be misattribution between activation and temp memory, but the peak memory + will be tracked accurately. This will be fixed in the next update by hooking intricately with ``torch.utils.checkpoint``. + """ + + def __init__(self) -> None: + self.memory_tracking = WeakIdKeyDictionary() + self._curr_mem_snap: dict[torch.device, dict[str, int]] = {} + self._peak_mem: dict[torch.device, int] = {} + self._peak_mem_snap: dict[torch.device, dict[str, int]] = {} + self._param_to_grad_hook_handles = WeakIdKeyDictionary() + self._optimizer_hook_handles: tuple[RemovableHandle, RemovableHandle] | None = ( + None + ) + # Dictionary to store the ``_WeakRefInfo`` instances corresponding to each tensor's storage. + self._WINFO = WeakIdKeyDictionary() + self._mod_tracker = ModTracker() + # This is a general memory tracker which can be used with any ``_RefType`` subclass + self._ref_class: type[_RefType] = _MemRefType + # Flags to track if we are in the AC region or optimizer step region + self._in_opt: bool = False + self._in_ac: bool = False + # Weak references to the topmost AC module currently active + self._ac_mod: weakref.ref | None = None + self._orig_resize = torch.UntypedStorage.resize_ + self._depth = 0 + + def _update_snap( + self, + u_type: _UpdateType, + winfo: _WeakRefInfo, + old_mem_consumed: int | None = None, + old_reftype: _RefType | None = None, + ) -> None: + # Initialize a flag to track if the total memory might drop to zero after updates. + maybe_zero = False + # Ensure the device entry exists in the current memory snapshot, initializing if necessary. + # pyrefly: ignore [no-matching-overload] + dev_snap = self._curr_mem_snap.setdefault( + winfo.device, dict.fromkeys(self._ref_class, 0) + ) + dev_snap.setdefault(_TOTAL_KEY, 0) + # Handle different types of updates based on the update type (`u_type`). + if u_type == _UpdateType.ADD: + # Increase the memory consumed for the specific reference type and update the total. + dev_snap[winfo.reftype] += winfo.mem_consumed + dev_snap[_TOTAL_KEY] += winfo.mem_consumed + elif u_type == _UpdateType.DEL: + # Decrease the memory consumed for the specific reference type and reduce the total. + dev_snap[winfo.reftype] -= winfo.mem_consumed + dev_snap[_TOTAL_KEY] -= winfo.mem_consumed + maybe_zero = True + elif u_type == _UpdateType.REF: + if old_reftype is None: + raise AssertionError + # Adjust memory consumption between two reference types within the same device. + dev_snap[old_reftype] -= winfo.mem_consumed + dev_snap[winfo.reftype] += winfo.mem_consumed + elif u_type == _UpdateType.SIZE: + if old_mem_consumed is None: + raise AssertionError + # Adjust the memory consumed for a reference type due to a change in size. + change = winfo.mem_consumed - old_mem_consumed + dev_snap[winfo.reftype] += change + dev_snap[_TOTAL_KEY] += change + maybe_zero = True + else: + raise ValueError(f"Invalid update type: {u_type}") + # Check if the total memory for the device has dropped to zero. + if maybe_zero: + if self._curr_mem_snap[winfo.device][_TOTAL_KEY] == 0: + # Remove the device entry from the memory snapshot if the total memory is zero. + del self._curr_mem_snap[winfo.device] + + def _update_and_maybe_create_winfos( + self, + t: torch.Tensor, + reftype: _RefType, + update_existing: bool = False, + ) -> set[_WeakRefInfo]: + sts = get_untyped_storages(t) + winfos = set() + for st in sts: + # Attempt to retrieve existing ``_WeakRefInfo`` and its weak reference from the tracking dictionary. + winfo, _ = self._WINFO.get(st, (None, None)) + if winfo is not None: + # If ``_WeakRefInfo`` exists, check if the reference type needs to be updated. + old_reftype = winfo.reftype + if old_reftype != reftype: + # Update the reference type and apply changes via ``_update_snap``. + winfo.reftype = reftype + self._update_snap(_UpdateType.REF, winfo, old_reftype=old_reftype) + winfos.add(winfo) + elif update_existing: + # If no existing ``_WeakRefInfo`` is found and update_existing is True, raise an error. + raise KeyError("No existing winfo found") + else: + # If no existing _WeakRefInfo is found and update_existing is False, create a new ``_WeakRefInfo``. + winfo, w_st = _WeakRefInfo.create_winfo( + st, t.device, reftype, self._delete_callback + ) + # Store the new ``_WeakRefInfo`` and its weak reference in the tracking dictionary. + self._WINFO[st] = (winfo, w_st) + # Update the snapshot for the newly added ``_WeakRefInfo``. + if winfo.mem_consumed > 0: + self._update_snap(_UpdateType.ADD, winfo) + winfos.add(winfo) + return winfos + + def _delete_callback(self, winfo: _WeakRefInfo, w_st: weakref.ref) -> None: + # Callback to be called when the storage object corresponding to the ``_WeakRefInfo`` + # instance is about to be finalized. + if winfo.mem_consumed > 0: + self._update_snap(_UpdateType.DEL, winfo) + + def _track_resize(self) -> None: + # Need to monkey-patch this because ``torch.UntypedStorage.resize_`` is not captured + # by ``TorchDispatchMode``. + @wraps(self._orig_resize) + def resize_(st: torch.UntypedStorage, size: int) -> None: + self._orig_resize(st, size) + winfo, _ = self._WINFO.get(st, (None, None)) + if winfo is not None and winfo.size != st.size(): + old_mem_consumed = winfo.mem_consumed + winfo.update_mem_consumed(st) + self._update_snap( + _UpdateType.SIZE, winfo, old_mem_consumed=old_mem_consumed + ) + + torch.UntypedStorage.resize_ = resize_ # type: ignore[method-assign, assignment] + + def _restore_resize(self) -> None: + torch.UntypedStorage.resize_ = self._orig_resize # type: ignore[method-assign] + + def _update_peak_stats(self, peak_state: _State) -> None: + # We first capture the current memory snapshot of the current tracker state then, + # We step through each of the modules we have tracked so far in ``memory_tracking`` + # and check if it is currently active by querying ``_mod_tracker.parents`` + # If it is active, we update the per device peak memory usage for the module + # corresponding to the ``_State`` which can be ``PEAK_FW`` or ``PEAK_BW``. + curr_snap = self._curr_mem_snap + + for mod_stats in self.memory_tracking.values(): + if mod_stats.mod_fqn in self._mod_tracker.parents: + if peak_state in mod_stats.snapshots: + for dev, dev_snap in curr_snap.items(): + if mod_stats.local_peak.get(dev, 0) < dev_snap[_TOTAL_KEY]: + mod_stats.local_peak[dev] = dev_snap[_TOTAL_KEY] + mod_stats.snapshots[peak_state][-1][dev] = deepcopy( + dev_snap + ) + + for dev, dev_snap in curr_snap.items(): + if self._peak_mem.get(dev, 0) < dev_snap[_TOTAL_KEY]: + self._peak_mem[dev] = dev_snap[_TOTAL_KEY] + self._peak_mem_snap[dev] = deepcopy(dev_snap) + + def _track(self, reftype: _RefType, t: torch.Tensor) -> None: + # Get the storages of the tensor and check if we have already tracked them. + # If yes, then check if the storage size has changed and update the current snapshot. + # Else create a new ``_WeakRefInfo`` instance and add it to the dictionary. + sts = get_untyped_storages(t) + for st in sts: + winfo, _ = self._WINFO.get(st, (None, None)) + if winfo is not None: + if winfo.size != st.size(): + old_mem_consumed = winfo.mem_consumed + winfo.update_mem_consumed(st) + self._update_snap( + _UpdateType.SIZE, winfo, old_mem_consumed=old_mem_consumed + ) + return + else: + winfo, w_st = _WeakRefInfo.create_winfo( + st, t.device, reftype, self._delete_callback + ) + self._WINFO[st] = (winfo, w_st) + # Update the current snapshot for the newly added ``_WeakRefInfo``. + if winfo.mem_consumed > 0: + self._update_snap(_UpdateType.ADD, winfo) + + def get_tracker_snapshot( + self, type: str = "current" + ) -> dict[torch.device, dict[str, int]]: + """ + Capture a snapshot of the memory usage breakdown per device, based on the specified type. + + Args: + type (str): The type of snapshot to capture. Can be "current" for the current memory usage or "peak" for the + peak memory usage. Defaults to "current". + Returns: + Dict[torch.device, Dict[str, int]]: A dictionary where each key is a torch.device, and each value is another + dictionary. This inner dictionary has keys representing memory reference + types as defined in ``_MemRefType`` and values representing the amount of + memory consumed in bytes. + Raises: + ValueError: If an invalid type is specified. + """ + if type == "current": + return deepcopy(self._curr_mem_snap) + elif type == "peak": + return deepcopy(self._peak_mem_snap) + else: + raise ValueError(f"Invalid type {type}") + + def _track_module_params_and_buffers( + self, module: nn.Module, install_grad_hooks: bool = True + ) -> tuple[int, int]: + # Track the parameters and buffers of the module if not already tracked. + # If the parameters have gradients, track the gradients as well. + # If install_grad_hooks is True, install a gradient hook on the parameters + # to track the gradients, if it has not already been installed. + # Return the total memory consumed by the parameters and buffers. + def _grad_hook(grad: torch.Tensor) -> None: + self._update_and_maybe_create_winfos( + grad, + _MemRefType.GRAD, + ) + + param_memory = 0 + for param in module.parameters(): + winfos = self._update_and_maybe_create_winfos( + param, + _MemRefType.PARAM, + ) + param_memory += sum(winfo.mem_consumed for winfo in winfos) + if param.grad is not None: + self._update_and_maybe_create_winfos( + param.grad, + _MemRefType.GRAD, + ) + if ( + self._param_to_grad_hook_handles.get(param, None) is None + and install_grad_hooks + ): + grad_hook_handle = param.register_hook(_grad_hook) + post_acc_grad_hook_handle = param.register_post_accumulate_grad_hook( + lambda p: (_grad_hook(p.grad)) + ) + self._param_to_grad_hook_handles[param] = ( + grad_hook_handle, + post_acc_grad_hook_handle, + ) + buffer_memory = 0 + for buffer in module.buffers(): + winfos = self._update_and_maybe_create_winfos( + buffer, + _MemRefType.BUFFER, + ) + buffer_memory += sum(winfo.mem_consumed for winfo in winfos) + return (param_memory, buffer_memory) + + def _track_inputs_or_outputs(self, args: Any) -> int: + # Calculate the memory consumed by the inputs or outputs of the module. + input_or_output_memory = 0 + + def add_inps_or_outs(t: torch.Tensor) -> None: + nonlocal input_or_output_memory + sts = get_untyped_storages(t) + for st in sts: + winfo, _ = self._WINFO.get(st, (None, None)) + if winfo is not None: + input_or_output_memory += winfo.mem_consumed + + tree_map_only(torch.Tensor, add_inps_or_outs, args) + return input_or_output_memory + + def _pre_fw_hook(self, module: nn.Module, inputs: Any) -> None: + # This is installed as a pre-fwd user hook with ``ModTracker.`` Based on the following cases we + # set the state and capture the memory snapshot for the module. + # Case 1: If the module is not in the ``memory_tracking`` dictionary, we track the parameters, buffers, + # input and output memory of the module. Create a new ``_ModMemStats`` instance for the module + # and add it to the ``memory_tracking`` dictionary. + # Case 2: If the module is already in the ``memory_tracking`` dictionary and we are in backward, this means + # we are in the AC region. We check if this is the top most module in the AC region. If it is, + # we store a weak reference and set the flag ``_in_ac`` to True. + # Case 3: If the module is already in the ``memory_tracking`` dictionary and we are in forward, this means + # this module is called for the second time. If it is a root module, that means we are in the next + # iteration and we error out. If it is not a root module, that means it's a submodule that is being + # used multiple times in the same iteration, which we allow and track. + # For Case 1 and 3, we also initialize the ``local_peak`` and ``PEAK_FW`` snapshot for the module. + mod_name = self._mod_tracker.get_known_fqn(module) + if mod_name is None: + raise AssertionError + if module not in self.memory_tracking: + mod_stats = _ModMemStats(mod_name) + param_mem, buffer_mem = self._track_module_params_and_buffers( + module, install_grad_hooks=True + ) + input_mem = self._track_inputs_or_outputs(inputs) + mod_stats.parameter_mem = param_mem + mod_stats.buffer_mem = buffer_mem + mod_stats.input_mem = input_mem + self.memory_tracking[module] = mod_stats + state = _ModState.PRE_FW + + elif self._mod_tracker.is_bw: + mod_stats = self.memory_tracking[module] + state = _ModState.PRE_FW_AC + if self._ac_mod is None: + self._ac_mod = weakref.ref(module) + self._in_ac = True + else: + parents = set(self._mod_tracker.parents) - {mod_name} + if len(parents) == 1 and "Global" in parents: + raise NotImplementedError( + "MemTracker does not support memory tracking for multiple iterative calls." + " Either use ``reset_mod_stats`` to clear module memory stats for the previous iteration" + " or file a github issue if you need this feature." + ) + mod_stats = self.memory_tracking[module] + state = _ModState.PRE_FW + input_mem = self._track_inputs_or_outputs(inputs) + mod_stats.mod_fqn = mod_name + mod_stats.input_mem = input_mem + + mem_snapshot = self.get_tracker_snapshot() + if state == _ModState.PRE_FW: + mod_stats.local_peak = { + dev: dev_snap[_TOTAL_KEY] for dev, dev_snap in mem_snapshot.items() + } + mod_stats.snapshots.setdefault(_ModState.PEAK_FW, []).append(mem_snapshot) + mod_stats.snapshots.setdefault(state, []).append(deepcopy(mem_snapshot)) + + def _post_fw_hook(self, module: nn.Module, inputs: Any, outputs: Any) -> None: + # This is installed as a post-fwd user hook with ``ModTracker``. Based on the following cases we + # set the state and capture the memory snapshot for the module. + # Case 1: This is called in backward, which means we are in the AC region. If this is the top most module + # in the AC region, we set the flag ``_in_ac`` to False. + # Case 2: This is called in forward so we calculate the output memory + # of the module and update its mod_stats. + mod_stats = self.memory_tracking[module] + if self._mod_tracker.is_bw: + state = _ModState.POST_FW_AC + if self._ac_mod is not None and self._ac_mod() is module: + self._ac_mod = None + self._in_ac = False + else: + state = _ModState.POST_FW + output_mem = self._track_inputs_or_outputs(outputs) + mod_stats.output_mem = output_mem + mod_stats.snapshots.setdefault(state, []).append(self.get_tracker_snapshot()) + + def _pre_bw_hook(self, module: nn.Module, args: Any) -> None: + # This is installed as a pre-bwd user hook with ``ModTracker``. We set the state and capture the + # snapshot for the module. We also initialize the ``local_peak`` and ``PEAK_BW`` snapshot for it. + # If the module is None, we skip the hook. + # This can happen since this installed inside a multi-grad hook on the module's output tensors + # and the module itself may not be alive during backward. + if module is None: + warnings.warn("Module is None. Skipping PRE_BW hook.", stacklevel=2) + return + mod_stats = self.memory_tracking[module] + mem_snapshot = self.get_tracker_snapshot() + mod_stats.local_peak = { + dev: dev_snap[_TOTAL_KEY] for dev, dev_snap in mem_snapshot.items() + } + mod_stats.snapshots.setdefault(_ModState.PEAK_BW, []).append(mem_snapshot) + mod_stats.snapshots.setdefault(_ModState.PRE_BW, []).append( + deepcopy(mem_snapshot) + ) + + def _post_bw_hook(self, module: nn.Module, args: Any) -> None: + # This is installed as a post-bwd user hook with ``ModTracker``. We set the state and capture the + # snapshot for the module if it is not None. + # This can happen since this installed inside a multi-grad hook on the module's input tensors + # and the module itself may not be alive during backward. + if module is None: + warnings.warn("Module is None. Skipping POST_BW hook.", stacklevel=2) + return + mod_stats = self.memory_tracking[module] + mod_stats.snapshots.setdefault(_ModState.POST_BW, []).append( + self.get_tracker_snapshot() + ) + + def _track_optimizer_states( + self, reftype: _RefType, optimizer: optim.Optimizer + ) -> None: + for states in optimizer.state.values(): + for val in states.values(): + if isinstance(val, torch.Tensor): + self._update_and_maybe_create_winfos( + val, + reftype, + ) + + def _register_global_optimizer_hook(self) -> None: + # Register a hook on the optimizer step to track the optimizer states. + # The pre-hook is to set the flag ``_in_opt`` to True. The post-hook unsets the flag, + # and also tracks any optimizer states that are created during the optimizer step. + def _opt_step_pre_hook( + optimizer: optim.Optimizer, args: Any, kwargs: Any + ) -> None: + self._in_opt = True + + def _opt_step_post_hook( + optimizer: optim.Optimizer, args: Any, kwargs: Any + ) -> None: + self._track_optimizer_states(_MemRefType.OPT, optimizer) + self._in_opt = False + + self._optimizer_hook_handles = ( + register_optimizer_step_pre_hook(_opt_step_pre_hook), + register_optimizer_step_post_hook(_opt_step_post_hook), + ) + + def _deregister_param_and_optimizer_hooks(self) -> None: + for ( + grad_hook_handle, + post_acc_grad_hook_handle, + ) in self._param_to_grad_hook_handles.values(): + grad_hook_handle.remove() + post_acc_grad_hook_handle.remove() + self._param_to_grad_hook_handles.clear() + + if self._optimizer_hook_handles is not None: + for handle in self._optimizer_hook_handles: + handle.remove() + self._optimizer_hook_handles = None + + def track_external( + self, *external: nn.Module | optim.Optimizer | torch.Tensor + ) -> None: + """ + Track tensors and stateful objects like modules, optimizers etc. that are created outside the MemTracker. + + This method should be called before the ``MemTracker`` is used. Any tensors that are not module parameters, buffers, + gradients activations, or optimizer states will be categorized as ``Other``. If you want them categorized with a + custom name, please file a GitHub issue. Any tensors created outside the MemTracker and not supplied to this + method will not be tracked by ``MemTracker``. + + Args: + *external (Union[nn.Module, optim.Optimizer, torch.Tensor]): The external modules, optimizers, and + tensors to be tracked. + """ + flat_external, _ = tree_flatten(external) + for obj in flat_external: + if isinstance(obj, torch.Tensor): + self._update_and_maybe_create_winfos( + obj, + _MemRefType.OTH, + ) + elif isinstance(obj, torch.nn.Module): + self._track_module_params_and_buffers(obj, install_grad_hooks=False) + elif isinstance(obj, optim.Optimizer): + self._track_optimizer_states(_MemRefType.OPT, obj) + elif obj is None: + continue + else: + raise TypeError( + f"Object of type {type(obj)} is not supported for tracking. " + f"Only stateful objects like modules, optimizers, and tensors are supported." + ) + + def display_snapshot( + self, type: str = "current", units: str = "B", tabulate: bool = False + ) -> None: + """ + Display the memory usage breakdown snapshot of the tracker based on the specified type and units. + + Keyword args: + type (str): The type of snapshot to display. Can be "current" for the current memory usage or "peak" for the + peak memory usage. Defaults to "current". + units (str): The units to use for displaying memory usage. Defaults to "B". Supports ["B", "KiB", "MiB", "GiB"]. + tabulate (bool): Whether to display the snapshot in a tabular format. Defaults to False. + """ + snapshot = self.get_tracker_snapshot(type) + if tabulate: + _print_snapshot_tabular(snapshot, units) + else: + _print_snapshot(snapshot, units) + + def display_modulewise_snapshots( + self, depth: int = 2, units: str = "B", tabulate: bool = False + ) -> None: + """ + Print per device memory breakdown snapshot for each module called within MemTracker. + + Snapshots are displayed for the states defined by ``_ModState``. + The module hierarchy is displayed up to the specified depth. + + Keyword Args: + depth (int, optional): The depth of the module hierarchy to display. Defaults to 2. + units (str, optional): The units to use for memory tracking. Defaults to "B". Supports ["B", "KiB", "MiB", "GiB"]. + tabulate (bool, optional): Whether to display the snapshot in a tabular format. Defaults to False. + """ + + def natural_sort_key(s: str) -> list[int | str]: + return [ + int(text) if text.isdigit() else text.lower() + for text in re.split("([0-9]+)", s) + ] + + for mod_stats in sorted( + self.memory_tracking.values(), + key=lambda m_stats: natural_sort_key(m_stats.mod_fqn), + ): + mod_fqn = mod_stats.mod_fqn + mod_depth = mod_fqn.count(".") + 1 + if mod_depth > depth: + continue + print(f"Module: {mod_fqn}") + if tabulate: + _print_state_snapshots_tabular(mod_stats.snapshots, units) + else: + _print_state_snapshots(mod_stats.snapshots, units) + + def reset_mod_stats(self) -> None: + """ + Reset all the module memory stats. Clears ``memory_tracking`` dictionary. + """ + self.memory_tracking.clear() + + def __enter__(self) -> "MemTracker": + if self._depth == 0: + # None in eager, a FakeTensorMode instance in SAC. Used in + # __torch_dispatch__ to skip DTensor propagation ops. + self._fake_mode_on_entry = active_fake_mode() + self._register_global_optimizer_hook() + self._mod_tracker.register_user_hooks( + self._pre_fw_hook, + self._post_fw_hook, + self._pre_bw_hook, + self._post_bw_hook, + ) + self._track_resize() + self._peak_mem_snap = self.get_tracker_snapshot() + self._peak_mem = { + dev: dev_snap[_TOTAL_KEY] + for dev, dev_snap in self._peak_mem_snap.items() + } + self._mod_tracker.__enter__() + super().__enter__() + self._depth += 1 + return self + + # pyrefly: ignore [bad-override] + def __exit__(self, *args: Any) -> None: + self._depth -= 1 + if self._depth == 0: + self._deregister_param_and_optimizer_hooks() + self._mod_tracker.clear_user_hooks() + self._restore_resize() + self._mod_tracker.__exit__(*args) + super().__exit__(*args) + + def __torch_dispatch__(self, func, types, args=(), kwargs=None): # type: ignore[no-untyped-def] + # When running this mode with DTensor, ordinarily all modes will + # run **before** subclasses get a chance to run. + # Returning NotImplemented here gives us a chance to let DTensor + # run and desugar into local tensor ops, before `MemTracker` sees them. + if any(t == DTensor for t in types): + return NotImplemented + if ( + func is torch.ops._c10d_functional.wait_tensor.default + and active_fake_mode() + ): + # N.B: This is a hacky way to override the Meta IMPL of wait_tensor. The original impl returns + # a new tensor which does not happen in eager mode, when a wait_tensor is called. + # pyrefly: ignore [bad-index] + res = args[0] + else: + res = func(*args, **kwargs or {}) + # DTensor sharding propagation may use a nested FakeTensorMode to + # compute output metadata (shapes, placements). The real memory + # usage comes from the local op on shard data, which + # runs outside DTensor's fake mode. We use identity comparison + # against the fake mode at tracker entry to skip propagation ops + # while still tracking local ops in both eager and SAC contexts. + # - Eager (entry mode is None): real ops have no fake mode + # (None is None), DTensor propagation ops are skipped (B is not None). + # - SAC (entry mode is A): SAC ops run under mode A (A is A), + # DTensor propagation ops are skipped (B is not A). + if active_fake_mode() is self._fake_mode_on_entry: + # If we are tracking an optimizer state, we use the optimizer reference type. + # If we are in backward region and not in AC region, we use the backward reference type. + # Else we use the forward reference type. + if self._in_opt: + reftype = _MemRefType.OPT + elif self._mod_tracker.is_bw and not self._in_ac: + reftype = _MemRefType.TEMP + else: + reftype = _MemRefType.ACT + tree_map_only(torch.Tensor, partial(self._track, reftype), res) + peak_state = ( + _ModState.PEAK_BW if self._mod_tracker.is_bw else _ModState.PEAK_FW + ) + self._update_peak_stats(peak_state) + return res diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/memory_tracker.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/memory_tracker.py new file mode 100644 index 0000000000000000000000000000000000000000..b421267b25e1c08eb42cdcffa718c5272865d079 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/memory_tracker.py @@ -0,0 +1,302 @@ +# mypy: allow-untyped-defs +import operator +import pickle +from collections import defaultdict +from collections.abc import Callable, Sequence +from itertools import chain +from typing import Any, no_type_check, TYPE_CHECKING + +import torch +import torch.nn as nn +from torch.utils._python_dispatch import TorchDispatchMode + + +if TYPE_CHECKING: + from torch.utils.hooks import RemovableHandle + + +BYTES_PER_MB = 1024 * 1024.0 + + +class MemoryProfileDispatchMode(TorchDispatchMode): + """Run in ``TorchDispatchMode`` to get memory stats at operator level.""" + + def __init__(self, memory_tracker) -> None: + self.memory_tracker = memory_tracker + + def __torch_dispatch__(self, func, types, args=..., kwargs=None): + rs = func(*args, **kwargs) + if func is torch.ops.aten.detach.default: + return rs + func_name: str = ( + self.memory_tracker._cur_module_name + + "." + + func.__name__ + + "_" + + str(self.memory_tracker._operator_names[func.__name__]) + ) + self.memory_tracker._operator_names[func.__name__] = ( + self.memory_tracker._operator_names[func.__name__] + 1 + ) + self.memory_tracker._record_memory_stats(func_name) + + return rs + + +class MemoryTracker: + """ + Collect and plot the memory stats at operator level. + + Includes ``memories_allocated``, ``memories_active`` and ``memories_reserved``. + It also prints a summary for the top 20 operators that generate the most memories. + + Example usage: + + >>> # xdoctest: +SKIP(failing) + >>> net.cuda() + >>> input = input.cuda() + + >>> mem_tracker = MemoryTracker() + >>> mem_tracker.start_monitor(net) + + >>> net.zero_grad(True) + >>> loss = net(input) + >>> if isinstance(loss, dict): + >>> loss = loss['out'] + >>> loss.sum().backward() + >>> net.zero_grad(set_to_none=True) + + >>> mem_tracker.stop() + >>> mem_tracker.summary() + >>> mem_tracker.show_traces() + """ + + def __init__(self) -> None: + torch._C._log_api_usage_once("torch.distributed.memory_tracker") + self._hooks: list[RemovableHandle] = [] + self._operator_names: dict[str, int] = defaultdict(int) + self.memories_allocated: dict[int, dict[str, float]] = defaultdict() + self.memories_active: dict[int, dict[str, float]] = defaultdict() + self.memories_reserved: dict[int, dict[str, float]] = defaultdict() + self._markers: dict[str, int] = defaultdict(int) + self._cur_module_name: str = "" + self._op_index: int = 0 + self._num_alloc_retries: int = 0 + self._device_module = torch.get_device_module() + + @no_type_check + def start_monitor(self, root_module: nn.Module) -> None: + """ + Register module hooks and entering ``MemoryProfileDispatchMode``. + + This enables operator level memory stats can be tracked during module runtime. + """ + self._clear_state() + root_module.__setattr__("_memory_tracker_is_root", True) + for name, m in root_module.named_modules(): + if m is not root_module: + m.__setattr__("_memory_tracker_is_root", False) + # fused_proxy_group does not support hooks + if ".fused_proxy_grouped_embedding_bag" in name: + continue + # hook ordering with other hooks added by users is not managed, so + # the memory stats tracked here may not completely accurate. + h1 = m.register_forward_pre_hook(self._create_pre_forward_hook(name)) + h2 = m.register_forward_hook(self._create_post_forward_hook(name)) + # it does not work well with jagged tensor somehow, the root cause is not + # clear and remove it for now as it does not really capture important info. + # h3 = m.register_backward_hook(self._create_backward_hook(name)) + self._hooks.extend([h1, h2]) + self._device_module.empty_cache() + if getattr(self, "profile_mode", None) is not None: + raise AssertionError + self.profile_mode = MemoryProfileDispatchMode(self) + self.profile_mode.__enter__() + + @no_type_check + def stop(self) -> None: + """ + Remove module hooks and exit ``MemoryProfileDispatchMode`` to stop tracking memory stats at operator level. + + Get some aggregated stats when the memory_tracker() is enabled, like ``num_alloc_retries``. + """ + self._num_alloc_retries = self._device_module.memory_stats().get( + "num_alloc_retries", 0 + ) + + for h in self._hooks: + h.remove() + self._hooks.clear() + if getattr(self, "profile_mode", None) is None: + raise AssertionError + self.profile_mode.__exit__(None, None, None) + self.profile_mode = None + + @no_type_check + def summary(self, top: int = 20) -> None: + """ + Print out the top operators that generate the most memories. + + The number of the top operators can be configured. + """ + op_diff: dict[str, float] = defaultdict(float) + op_name, previous_allocated_memory = self.memories_allocated[0] + for i in range(1, self._op_index): + op_name, current_allocated_memory = self.memories_allocated[i] + op_diff[op_name] = current_allocated_memory - previous_allocated_memory + previous_allocated_memory = current_allocated_memory + + print("------------------------------------------------") + print(f"The number of alloc retries are: {self._num_alloc_retries}") + print(f"Top {top} ops that generates memory are:") + for k, v in sorted(op_diff.items(), key=operator.itemgetter(1), reverse=True)[ + :top + ]: + print(f"{k}: {v}MB") + print("------------------------------------------------") + + @no_type_check + def show_traces(self, path: str = "") -> None: + import matplotlib.pyplot as plt + + def _plot_figure(x, y_values, labels): + min_val = min(chain.from_iterable(y_values)) * 0.999 + max_val = max(chain.from_iterable(y_values)) * 1.001 + plt.figure() + for y, label in zip(y_values, labels): + plt.plot(x, y, label=label) + plt.xlabel("# Operator Calls") + plt.ylabel("Memory (MB)") + plt.legend() + for marker_name, marker in self._markers.items(): + if marker_name == "fw_bw_boundary": + plt.plot( + [marker, marker], + [min_val, max_val], + "r", + lw=2, + label=marker_name, + ) + else: + plt.plot( + [marker, marker], + [min_val, max_val], + "k-", + lw=2, + label=marker_name, + ) + + if path != "": + self.load(path) + + y_1 = [gb for (name, gb) in self.memories_allocated.values()] + y_2 = [gb for (name, gb) in self.memories_active.values()] + y_3 = [gb for (name, gb) in self.memories_reserved.values()] + x = list(range(len(y_1))) + # Split figures when there is big difference between + # "reserved_memory" and "allocated_memory" or "active_memory". + _plot_figure( + x, + [list(y_1), list(y_2), list(y_3)], + ["allocated_memory", "active_memory", "reserved_memory"], + ) + _plot_figure(x, [list(y_1)], ["allocated_memory"]) + _plot_figure(x, [list(y_2)], ["active_memory"]) + _plot_figure(x, [list(y_3)], ["reserved_memory"]) + + def save_stats(self, path: str) -> None: + """Save the stats using pickle during runtime if users want to plot the traces in other places like notebook.""" + stats = { + "memories_allocated": self.memories_allocated, + "memories_active": self.memories_active, + "memories_reserved": self.memories_reserved, + "markers": self._markers, + "num_alloc_retries": self._num_alloc_retries, + } + + with open(path, "wb") as f: + pickle.dump(stats, f, pickle.HIGHEST_PROTOCOL) + + def load(self, path: str) -> None: + """Load the pickled memory stats to plot the traces or print the summary.""" + with open(path, "rb") as f: + stats = pickle.load(f) + + self.memories_allocated = stats["memories_allocated"] + self.memories_active = stats["memories_active"] + self.memories_reserved = stats["memories_reserved"] + self._markers = stats["markers"] + self._num_alloc_retries = stats["num_alloc_retries"] + + def _create_pre_forward_hook(self, name: str) -> Callable: + """Prefix operator name with current module and 'forward', and insert 'fw_start' marker at forward pass start.""" + + def _pre_forward_hook(module: nn.Module, inputs: Any) -> None: + self._cur_module_name = f"{name}.forward" + if ( + hasattr(module, "_memory_tracker_is_root") + and module._memory_tracker_is_root + ): + self._add_marker("fw_start") + + return _pre_forward_hook + + def _create_post_forward_hook(self, name: str) -> Callable: + """Insert the marker 'fw_bw_boundary' at the boundary of forward and backward pass.""" + + def _post_forward_hook( + module: nn.Module, + inputs: Sequence[torch.Tensor], + outputs: Sequence[torch.Tensor], + ) -> None: + if ( + hasattr(module, "_memory_tracker_is_root") + and module._memory_tracker_is_root + ): + self._add_marker("fw_bw_boundary") + + return _post_forward_hook + + def _create_backward_hook(self, name: str) -> Callable: + """Insert the current module name with backward prefix for the operator name.""" + + def _backward_hook( + module: nn.Module, grad_input: torch.Tensor, grad_output: torch.Tensor + ) -> None: + self._cur_module_name = f"{name}.backward" + + return _backward_hook + + @no_type_check + def _record_memory_stats(self, fn_name: str) -> None: + """ + Record current memory allocated, current memory active and current memory reserved. + + The memory stats dict is indexed with ``self._op_index``. + """ + memory_allocated: float = self._device_module.memory_allocated() / BYTES_PER_MB + memory_reserved: float = self._device_module.memory_reserved() / BYTES_PER_MB + memory_active: float = ( + self._device_module.memory_stats().get("active_bytes.all.current", 0) + / BYTES_PER_MB + ) + self.memories_allocated[self._op_index] = (fn_name, memory_allocated) + self.memories_reserved[self._op_index] = (fn_name, memory_reserved) + self.memories_active[self._op_index] = (fn_name, memory_active) + self._op_index += 1 + + def _add_marker(self, marker_name: str) -> None: + """Set the marker's x-axis value.""" + marker_val = len(self.memories_allocated.values()) + self._markers[marker_name] = marker_val + + def _clear_state(self) -> None: + """Clear states when start_monitor() is called.""" + self._operator_names.clear() + self.memories_allocated.clear() + self.memories_active.clear() + self.memories_reserved.clear() + self._markers.clear() + self._cur_module_name = "" + self._op_index = 0 + self._num_alloc_retries = 0 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mod_tracker.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mod_tracker.py new file mode 100644 index 0000000000000000000000000000000000000000..271383a19b88058e35485ad16d592d6b7653b404 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/mod_tracker.py @@ -0,0 +1,259 @@ +# mypy: allow-untyped-defs +import warnings +import weakref +from collections.abc import Callable + +import torch +from torch.autograd.graph import register_multi_grad_hook +from torch.nn.modules.module import ( + register_module_forward_hook, + register_module_forward_pre_hook, +) +from torch.utils._pytree import tree_flatten + + +__all__ = ["ModTracker"] + + +class ModTracker: + """ + ``ModTracker`` is a context manager that tracks the nn.Module hierarchy during execution + so that other system can query which Module is currently being executed (or its backward is being + executed). + + You can access the ``parents`` attribute on this context manager to get the set of all the + Modules currently being executed via their fqn (fully qualified name, also used as the key within + the state_dict). + You can access the ``is_bw`` attribute to know if you are currently running in backward or not. + + Note that ``parents`` is never empty and always contains the "Global" key. The ``is_bw`` flag + will remain ``True`` after the forward until another Module is executed. If you need it to be + more accurate, please submit an issue requesting this. Adding a map from fqn to the module instance + is possible but not done yet, please submit an issue requesting this if you need it. + + Example usage + + .. code-block:: python + + mod = torch.nn.Linear(2, 2) + + with ModTracker() as tracker: + # Access anything during the forward pass + def my_linear(m1, m2, bias): + print(f"Current modules: {tracker.parents}") + return torch.mm(m1, m2.t()) + bias + + torch.nn.functional.linear = my_linear + + mod(torch.rand(2, 2)) + + """ + + parents: set[str] + """ + A Set containing the fqn for each module currently running their forward + """ + + def __init__(self): + self.parents = {"Global"} + self._active_module_cnt = {} + self._known_modules: weakref.WeakKeyDictionary = weakref.WeakKeyDictionary() + self._seen_modules: weakref.WeakSet = weakref.WeakSet() + self._has_callback = False + self._post_bw_callbacks_to_enqueue: list[Callable] = [] + self._user_pre_fw_hook = None + self._user_post_fw_hook = None + self._user_pre_bw_hook = None + self._user_post_bw_hook = None + + def _maybe_set_engine_callback(self): + # This assumes no concurrent calls to backward + if self._has_callback: + return + + for post_bw_callback in reversed(self._post_bw_callbacks_to_enqueue): + torch.autograd.Variable._execution_engine.queue_callback(post_bw_callback) + self._post_bw_callbacks_to_enqueue.clear() + + def callback(): + self.parents = {"Global"} + self._has_callback = False + + torch.autograd.Variable._execution_engine.queue_callback(callback) + self._has_callback = True + + @property + def is_bw(self): + """ + A boolean marking if this is currently running during the backward pass or not + """ + return torch._C._current_graph_task_id() != -1 + + def get_known_fqn(self, mod): + """ + Return the fqn for the given module if it is known to the ``ModTracker``, otherwise ``None``. + """ + return self._known_modules.get(mod, None) + + def register_user_hooks( + self, + pre_fw_hook: Callable | None = None, + post_fw_hook: Callable | None = None, + pre_bw_hook: Callable | None = None, + post_bw_hook: Callable | None = None, + ): + """ + Registers user-specified hooks to be called before/after the forward/backward pass for each + module tracked by the ``ModTracker``. One or more can be ``None``. + Args: + pre_fw_hook (Callable, optional): A hook to be called before the forward pass for the + module. It should have the following signature: + pre_fw_hook (module, input) -> None + post_fw_hook (Callable, optional): A hook to be called after the forward pass for the + module. It should have the following signature: + post_fw_hook (module, input, output) -> None + pre_bw_hook (Callable, optional): A multi-grad hook to be called on all the outputs of + the module that require gradients. It should have the following signature: + pre_bw_hook (module, grad_output) -> None + post_bw_hook (Callable, optional): A multi-grad hook to be called on all the inputs of + the module that require gradients. It should have the following signature: + post_bw_hook (module, grad_input) -> None + Raises: + AssertionError: If a new hook is provided when one is already registered. + Note: + If the module is not alive during the backward pass, the pre_bw_hook and post_bw_hook will + receive None as the module argument. + The module fqn will be present in the ``parents`` attribute when each of the hooks is called. + Hooks are intended to be used as markers only not to modify the inputs/outputs. + """ + + def set_hook(hook, user_hook, hook_name): + if hook is not None and user_hook is not None: + raise AssertionError( + f"Only one {hook_name} can be registered at a time" + f" Clear the existing hook by calling ``clear_user_hooks`` before registering a new one" + ) + return hook + + self._user_pre_fw_hook = set_hook( + pre_fw_hook, self._user_pre_fw_hook, "pre_fw_hook" + ) + self._user_post_fw_hook = set_hook( + post_fw_hook, self._user_post_fw_hook, "post_fw_hook" + ) + self._user_pre_bw_hook = set_hook( + pre_bw_hook, self._user_pre_bw_hook, "pre_bw_hook" + ) + self._user_post_bw_hook = set_hook( + post_bw_hook, self._user_post_bw_hook, "post_bw_hook" + ) + + def clear_user_hooks(self): + """ + Clears the user specified hooks registered with ``register_user_hooks`` + """ + self._user_pre_fw_hook = None + self._user_post_fw_hook = None + self._user_pre_bw_hook = None + self._user_post_bw_hook = None + + def _get_mod_name(self, mod): + if mod not in self._known_modules: + self._known_modules[mod] = type(mod).__name__ + mod_name = self._known_modules[mod] + if mod not in self._seen_modules: + for name, submod in mod.named_children(): + self._known_modules[submod] = f"{mod_name}.{name}" + self._get_mod_name(submod) + self._seen_modules.add(mod) + return mod_name + + def _get_append_fn(self, w_mod, name, is_bw): + def fn(*args): + if is_bw: + self._maybe_set_engine_callback() + if name in self.parents and not self.is_bw: + + def custom_formatwarning(msg, category, filename, lineno, line=None): + return f"{filename}:{lineno}: {category.__name__}: {msg} \n" + + # pyrefly: ignore [bad-assignment] + warnings.formatwarning = custom_formatwarning + warnings.warn( + "The module hierarchy tracking maybe be messed up." + " Please file a bug to PyTorch, if it is the case.", + stacklevel=2, + ) + if name not in self.parents: + self._active_module_cnt[name] = 1 + self.parents.add(name) + else: + self._active_module_cnt[name] += 1 + + if self._user_pre_bw_hook is not None and is_bw: + self._user_pre_bw_hook(w_mod(), args) + + return fn + + def _get_pop_fn(self, w_mod, name, is_bw): + def fn(*args): + if self._user_post_bw_hook is not None and is_bw: + self._user_post_bw_hook(w_mod(), args) + if name in self.parents: + self._active_module_cnt[name] -= 1 + if self._active_module_cnt[name] == 0: + self.parents.remove(name) + elif not self.is_bw: + # Due to some input/output not requiring gradients, we cannot enforce + # proper nesting in backward + raise RuntimeError( + "The Module hierarchy tracking is wrong. Report a bug to PyTorch" + ) + + return fn + + def _fw_pre_hook(self, mod, input): + if torch._dynamo.eval_frame._is_in_optimized_module(): + return + + name = self._get_mod_name(mod) + w_mod = weakref.ref(mod) + self._get_append_fn(w_mod, name, False)() + if self._user_pre_fw_hook is not None: + self._user_pre_fw_hook(mod, input) + args, _ = tree_flatten(input) + tensors = [a for a in args if isinstance(a, torch.Tensor) and a.requires_grad] + if not self.is_bw: + if tensors: + register_multi_grad_hook(tensors, self._get_pop_fn(w_mod, name, True)) + else: + self._post_bw_callbacks_to_enqueue.append( + self._get_pop_fn(w_mod, name, True) + ) + + def _fw_post_hook(self, mod, input, output): + if torch._dynamo.eval_frame._is_in_optimized_module(): + return + + name = self._get_mod_name(mod) + w_mod = weakref.ref(mod) + if self._user_post_fw_hook is not None: + self._user_post_fw_hook(mod, input, output) + self._get_pop_fn(w_mod, name, False)() + args, _ = tree_flatten(output) + tensors = [a for a in args if isinstance(a, torch.Tensor) and a.requires_grad] + if not self.is_bw and tensors: + register_multi_grad_hook( + tensors, self._get_append_fn(w_mod, name, True), mode="any" + ) + + def __enter__(self): + self._fw_pre_handle = register_module_forward_pre_hook(self._fw_pre_hook) + self._fw_post_handle = register_module_forward_hook( + self._fw_post_hook, always_call=True + ) + return self + + def __exit__(self, *args): + self._fw_pre_handle.remove() + self._fw_post_handle.remove() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/runtime_estimator.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/runtime_estimator.py new file mode 100644 index 0000000000000000000000000000000000000000..9289a280c5aaa835766e8881bca44746b2336d87 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/runtime_estimator.py @@ -0,0 +1,416 @@ +# Owner(s): ["module: unknown"] +from collections import defaultdict +from typing import Any, TYPE_CHECKING +from typing_extensions import Self + +import torch +import torch.utils._pytree as pytree +from torch._guards import active_fake_mode +from torch._subclasses.fake_tensor import FakeTensorMode +from torch.distributed._tools.mod_tracker import ModTracker +from torch.utils._mode_utils import no_dispatch +from torch.utils._python_dispatch import TorchDispatchMode +from torch.utils._runtime_estimation import ( + _FLOAT_TYPES, + _IGNORE_OPS, + _VIEW_OPS, + get_compute_time, + get_transfer_time, +) + + +if TYPE_CHECKING: + from collections.abc import Callable + +__all__ = ["RuntimeEstimator"] + + +class RuntimeEstimator(TorchDispatchMode): + """ + Estimates the GPU runtime in milliseconds using various estimation methods under the ``FakeTensorMode``. + + This class provides a ``TorchDispatchMode`` based context manager that can be used to estimate the eager + runtime of PyTorch functions. It supports two estimation modes, benchmarking (`operator-level-benchmark`) and + roofline cost modeling (`operator-level-cost-model`). + For modules executed under this context manager, it aggregates the forward and backward operation runtimes + and also records their execution orders. + + Attributes: + mod_runtimes (Dict[str, Dict[str, float]]): A dictionary of module runtimes. The key to the outer dictionary + is the fully qualified name (FQN) of the module. For each module the forward and backward runtimes of the + operations are aggregated in the inner dictionary keyed by 'fw' and 'bw'. + mod_fw_pre_order (List[str]): List of module FQNs in pre-forward execution order. + mod_bw_pre_order (List[str]): List of module FQNs in pre-backward execution order. + mod_fw_post_order (List[str]): List of module FQNs in post-forward execution order. + mod_bw_post_order (List[str]): List of module FQNs in post-backward execution order. + total_runtime (float): The total estimated runtime in milliseconds. + + Note: + 1) The benchmarking estimate mode will execute kernels on GPU and assumes that every operation can run in + isolation without causing an OOM error. It is also designed to be used only under ``FakeTensorMode``. + 2) Currently wrapper tensor sub-classes such as ``DTensor`` won't produce correct estimates. We plan to support + them in future PRs. + 3) We only estimate the compute time, if your code has communication, it will not be considered. Again, we will + support this in future PRs. + + Example usage: + + .. code-block:: python + + runtime_estimator = RuntimeEstimator() + with FakeTensorMode(): + module = ... + optimizer = ... + inp = ... + with runtime_estimator(estimate_mode_type="operator-level-cost-model"): + loss = module(inp) + loss.backward() + optimizer.step() + optimizer.zero_grad() + runtime_estimator.display_modulewise_stats() + """ + + _no_fallback_kernel: set[torch._ops._OpNamespace] = set() + fake_mode: FakeTensorMode + gpu_type: str | None = None + + def __init__(self, gpu_type: str | None = None) -> None: + """ + Args: + gpu_type (str | None): Optional datasheet device name (e.g. + ``"NVIDIA H100"``) to pin the roofline peak FLOPS and DRAM + bandwidth to instead of querying the current device. This + makes ``operator-level-cost-model`` estimates deterministic + and hardware-independent. Only affects the roofline estimator. + """ + super().__init__() + self._gpu_type = gpu_type + self._estimate: Callable + self._estimate_mode_type: str + self._mod_tracker = ModTracker() + self.mod_runtimes: dict[str, dict[str, float]] = defaultdict( + lambda: defaultdict(lambda: 0.0) + ) + self.mod_fw_pre_order: list[str] = [] + self.mod_bw_pre_order: list[str] = [] + self.mod_fw_post_order: list[str] = [] + self.mod_bw_post_order: list[str] = [] + self.total_runtime: float = 0.0 + + # Adapted from: https://github.com/pytorch/pytorch/blob/9b902b3ee3bd608a19543362b66bf06c373dd374/torch/_subclasses/fake_tensor.py#L1969 + # NB: returns fake tensors + @classmethod + def _maybe_run_and_benchmark_fallback_kernel( # type: ignore[no-untyped-def] + cls, + func, + args, + kwargs, + orig_not_implemented_exception, + ): + """ + Runs and benchmarks a fallback kernel for a given function. + + Args: + func (Callable): The function to benchmark. + args (Tuple): The arguments to pass to the function. + kwargs (Dict[str, Any]): The keyword arguments to pass to the function. + orig_not_implemented_exception (Exception): The original exception to raise if the fallback kernel + is not implemented. + + Returns: + Tuple[Any, float]: A tuple containing the result of the function and + the mean operation time in milliseconds. + """ + # these should all be supported, just to be safe + # avoid fallback for operators which inplace modify metadata + # because the input fake tensors would be unmodified + if torch.Tag.inplace_view in func.tags: # type: ignore[attr-defined] + raise orig_not_implemented_exception + + inp_impls = {} + flat_args, args_spec = pytree.tree_flatten((args, kwargs)) + # Don't use in_kernel_invocation_manager(fake_mode) as we want to do + # REAL compute (not with meta device) + with no_dispatch(): + + def to_real_tensor(e): # type: ignore[no-untyped-def] + if cls.fake_mode.is_our_fake(e): + if e.dtype in _FLOAT_TYPES: + out = torch.rand_like(e, device=e.fake_device) + else: + out = torch.ones_like(e, device=e.fake_device) + if e.is_sparse: + out._coalesced_(e.is_coalesced()) + inp_impls[id(out)] = e + return out + return e + + flat_args = [to_real_tensor(a) for a in flat_args] + args, kwargs = pytree.tree_unflatten(flat_args, args_spec) + r = func(*args, **kwargs) + warmup_iters, actual_iters = 2, 3 + for _ in range(warmup_iters): + func(*args, **kwargs) + start_event = torch.Event(enable_timing=True) + end_event = torch.Event(enable_timing=True) + start_event.record(torch.accelerator.current_stream()) + for _ in range(actual_iters): + func(*args, **kwargs) + end_event.record(torch.accelerator.current_stream()) + torch.accelerator.synchronize() + cuda_time = start_event.elapsed_time(end_event) + mean_op_time = cuda_time / actual_iters + + storages = set() + + for e in flat_args: + if isinstance(e, torch.Tensor): + if not e.is_sparse: + storages.add(e._typed_storage()._cdata) + + # TODO: also check metadata change on inputs + # proper aliasing/metadata relationship between outputs and inputs will + # not be set up, bc of conversion to device, unless we can reuse an + # input impl + + def map_out(e): # type: ignore[no-untyped-def] + if id(e) not in inp_impls and ( + isinstance(e, torch.Tensor) + and not e.is_sparse + and e._typed_storage()._cdata in storages + ): + raise orig_not_implemented_exception + + if isinstance(e, torch.Tensor): + if id(e) in inp_impls: + return inp_impls[id(e)] + else: + return cls.fake_mode.fake_tensor_converter.from_real_tensor( + cls.fake_mode, e + ) + else: + return e + + return (pytree.tree_map(map_out, r), mean_op_time) + + @classmethod + def _benchmark_estimate(cls, func, args, kwargs) -> tuple[Any, float]: # type: ignore[no-untyped-def] + """ + Estimates the runtime of a function using benchmarking. + + Args: + func: The function to estimate. + args: The arguments to pass to the function. + kwargs: The keyword arguments to pass to the function. + res: The result of the function. + + Returns: + Tuple[Any, float]: A tuple containing the result of the function and + the mean operation time in milliseconds. + """ + if not isinstance(cls.fake_mode, FakeTensorMode): + raise AssertionError( + "Initialize/Assign FakeTensorMode before using this function" + ) + mean_op_time = 0.0 + if func._overloadpacket not in _VIEW_OPS: + try: + res, mean_op_time = cls._maybe_run_and_benchmark_fallback_kernel( + func, + args, + kwargs, + NotImplementedError, + ) + return (res, mean_op_time) + except NotImplementedError: + cls._no_fallback_kernel.add(func._overloadpacket) + res = func(*args, **kwargs or {}) + return (res, mean_op_time) + + # Adapted from: https://github.com/pytorch/pytorch/blob/9b902b3ee3bd608a19543362b66bf06c373dd374/torch/_inductor/scheduler.py#L589 + @classmethod + def _roofline_estimate(cls, func, args, kwargs) -> tuple[Any, float]: # type: ignore[no-untyped-def] + """ + Estimates the runtime of a function using a roofline cost model. + + Args: + func: The function to estimate. + args: The arguments to pass to the function. + kwargs: The keyword arguments to pass to the function. + out: The output of the function. + + Returns: + Tuple[Any, float]: A tuple containing the result of the function and + the mean operation time in milliseconds. + """ + if not torch.accelerator.is_available(): + raise AssertionError( + "Roofline estimation needs to access CUDA capabilities to make estimations" + ) + + # Roofline Cost Model Explanation + + # The roofline cost model estimates the execution time of an operator based on + # the device's empirical maximum FLOPs/sec (pi) and device DRAM bandwidth (beta). + + # Variables: + # - pi: Maximum empirical FLOPs/sec of the device + # - beta: Maximum empirical device DRAM bandwidth (bytes/sec) of the device + # - I: Arithmetic intensity of the operator (FLOPs/bytes) + # - op_flops: FLOPs required by the operator + # - op_bytes: Bytes transferred to and from DRAM for the operator + + # Calculation Steps: + # 1. Calculate arithmetic intensity: I = op_flops / op_bytes + # 2. Calculate estimated FLOPs/sec: est_flops_sec = min(pi, beta * I) + # 3. Calculate estimated operator time: estimated_op_time = op_flops / est_flops_sec + # This simplifies to: estimated_op_time = max(op_flops / pi, op_flops / (beta * I)) + # Further simplifying: estimated_op_time = max(op_flops / pi, op_bytes / beta) + + # Simplified Formulas: + # - compute_time = op_flops / pi + # - transfer_time = op_bytes / beta + # - estimated_op_time = max(compute_time, transfer_time) + + kwargs = kwargs if kwargs else {} + out = func(*args, **kwargs) + op_time = 0.0 + func_packet = func._overloadpacket + if func_packet not in _IGNORE_OPS: + flat_args_kwargs, args_spec = pytree.tree_flatten((args, kwargs)) + flat_outs, out_spec = pytree.tree_flatten(out) + transfer_time = get_transfer_time( + flat_args_kwargs, flat_outs, gpu_type=cls.gpu_type + ) + + out_dtypes = { + t.dtype + for t in flat_outs + if isinstance(t, torch.Tensor) and t.dtype in _FLOAT_TYPES + } + + args, kwargs = pytree.tree_unflatten(flat_args_kwargs, args_spec) + out = pytree.tree_unflatten(flat_outs, out_spec) + + compute_time = get_compute_time( + func_packet, args, kwargs, out, out_dtypes, gpu_type=cls.gpu_type + ) + # We get the estimated time as the max of the transfer time and + # compute time. We divide by 1e6 to get the time in ms + op_time = max(transfer_time, compute_time) / 1e6 + + return (out, op_time) + + def display_modulewise_stats(self, depth: int = 2) -> None: + """ + Displays module-wise statistics collected by ``RuntimeEstimator``. + + Prints the pre-forward and pre-backward execution orders. + Displays the module-wise forward and backward runtimes in milliseconds. + + Args: + depth (int): The maximum depth of module hierarchy to display (default to 2). + """ + print("Pre-Forward Execution Order: ") + for mod_fqn in self.mod_fw_pre_order: + mod_depth = mod_fqn.count(".") + 1 + if mod_depth > depth: + continue + print(mod_fqn) + print("Pre-Backward Execution Order: ") + for mod_fqn in self.mod_bw_pre_order: + mod_depth = mod_fqn.count(".") + 1 + if mod_depth > depth: + continue + print(mod_fqn) + for mod_fqn, runtimes in self.mod_runtimes.items(): + mod_depth = mod_fqn.count(".") + 1 + if mod_depth > depth: + continue + print( + f"{mod_fqn} fw: {runtimes.get('fw', 0.0):.3f}ms bw: {runtimes.get('bw', 0.0):.3f}ms" + ) + + def __torch_dispatch__(self, func, types, args=..., kwargs=None): # type: ignore[no-untyped-def] + # TODO: @sanketpurandare: Flatten tensors by desugaring the tensor subclasses + # TODO: @sanketpurandare: Add logic for incorporating communication time + res, op_time = self._estimate(func, args, kwargs) + for par in self._mod_tracker.parents: + if self._mod_tracker.is_bw: + self.mod_runtimes[par]["bw"] += op_time + else: + self.mod_runtimes[par]["fw"] += op_time + self.total_runtime += op_time + return res + + def __call__(self, estimate_mode_type: str) -> Self: + """ + Sets the estimate mode type. + + Currently supported modes: + - "operator-level-benchmark": Estimates runtime using operator benchmarking. + - "operator-level-cost-model": Estimates runtime using roofline cost model. + + Args: + estimate_mode_type (str): The type of estimate mode to use. + + Returns: + RuntimeEstimator: The runtime estimator instance. + + Raises: + NotImplementedError: If the estimate mode type is not supported. + """ + if estimate_mode_type == "operator-level-benchmark": + self._estimate = RuntimeEstimator._benchmark_estimate + elif estimate_mode_type == "operator-level-cost-model": + self._estimate = RuntimeEstimator._roofline_estimate + else: + raise NotImplementedError( + f"estimate_mode_type {estimate_mode_type} not supported" + ) + self._estimate_mode_type = estimate_mode_type + return self + + def __enter__(self) -> Self: + fake_mode = active_fake_mode() + if not isinstance(fake_mode, FakeTensorMode): + raise AssertionError( + "No FakeTensorMode found, designed to be used under FakeTensorMode" + ) + RuntimeEstimator.fake_mode = fake_mode + RuntimeEstimator.gpu_type = self._gpu_type + self.total_runtime = 0.0 + self.mod_runtimes = defaultdict(lambda: defaultdict(lambda: 0.0)) + self.mod_fw_pre_order.clear() + self.mod_bw_pre_order.clear() + self.mod_fw_post_order.clear() + self.mod_bw_post_order.clear() + self._mod_tracker.register_user_hooks( + pre_fw_hook=lambda mod, inp: self.mod_fw_pre_order.append( + self._mod_tracker.get_known_fqn(mod) + ), + pre_bw_hook=lambda mod, g_out: self.mod_bw_pre_order.append( + self._mod_tracker.get_known_fqn(mod) + ), + post_fw_hook=lambda mod, inp, out: self.mod_fw_post_order.append( + self._mod_tracker.get_known_fqn(mod) + ), + post_bw_hook=lambda mod, g_inp: self.mod_bw_post_order.append( + self._mod_tracker.get_known_fqn(mod) + ), + ) + self._mod_tracker.__enter__() + super().__enter__() + return self + + # pyrefly: ignore [bad-override] + def __exit__(self, *args: Any) -> None: + print( + f"Estimated ({self._estimate_mode_type})" + f"total_time: {self.total_runtime:.3f} ms" + ) + if len(self._no_fallback_kernel) > 0: + print("no_fallback_kernel: ", list(self._no_fallback_kernel)) + super().__exit__(*args) + self._mod_tracker.clear_user_hooks() + self._mod_tracker.__exit__() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_estimator.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_estimator.py new file mode 100644 index 0000000000000000000000000000000000000000..edd390bd565c92b9cdd8ee9cc34c984a202f391c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_estimator.py @@ -0,0 +1,973 @@ +import math +import os +import sys +from collections import OrderedDict +from dataclasses import astuple, dataclass +from typing import Any, NamedTuple +from typing_extensions import Self + +import torch +from torch import nan, nn, UntypedStorage +from torch._guards import active_fake_mode +from torch._subclasses.fake_tensor import FakeTensorMode +from torch.distributed._tools.common_utils import get_untyped_storages +from torch.distributed._tools.mod_tracker import ModTracker +from torch.distributed._tools.runtime_estimator import RuntimeEstimator +from torch.testing._internal.composite_compliance import ( + is_inplace, + is_inplace_view_fn, + is_view_fn, +) +from torch.utils._python_dispatch import TorchDispatchMode +from torch.utils._pytree import tree_flatten +from torch.utils.checkpoint import SAC_IGNORED_OPS + + +__all__ = ["SACEstimator", "SACStats", "MSPS", "SACTradeOffStats", "SACGreedyOrderMeta"] +aten = torch.ops.aten + +_ADDITIONAL_IGNORED_OPS = { + aten.lift_fresh.default, # type: ignore[attr-defined] + torch.ops.profiler._record_function_exit._RecordFunction, # type: ignore[attr-defined] + aten.clone.default, # type: ignore[attr-defined] # seems needed for torch.compile +} +OPS_TO_ALWAYS_SKIP = SAC_IGNORED_OPS | _ADDITIONAL_IGNORED_OPS +# This value is hard-coded here: +# https://github.com/pytorch/pytorch/blob/5fba5d83f0703ff8077ab65448a998e9ad6598fd/c10/cuda/CUDACachingAllocator.cpp#L117 +_PYTORCH_MIN_ALLOCATE = ( + 2**9 if int(os.environ.get("PYTORCH_NO_CUDA_MEMORY_CACHING", 0)) == 0 else 1 +) + + +def _display_stats_tabular(headers: list[str], table_data: list[list[Any]]) -> None: + try: + from tabulate import tabulate + except ImportError as err: + raise ImportError("Please install tabulate.") from err + + # Use tabulate to print the table + print(tabulate(table_data, headers=headers, tablefmt="rst")) + + +# Based on: +# https://github.com/facebookresearch/xformers/blob/main/xformers/checkpoint.py#L71 +@dataclass +class _SACMetadata: + """ + Stores metadata for a single operator for SAC. + + Attributes: + func (Any): The operator function. + time_taken (float): The time taken by the operator. + memory_used (float): The memory used by the operator. + curr_idx (int): The current operator index. + output_ids (Tuple[int, ...]): The storage IDs of the operator's outputs. + inplace_info (Tuple[int, ...]): Tuple of self and parent operator for in-place operator. + is_view_like (bool): Whether the operator is view-like. + is_rand_op (bool): Whether the operator is a random operator. + """ + + func: Any + time_taken: float + memory_used: float + curr_idx: int + output_ids: tuple[int, ...] + inplace_info: tuple[int, ...] + is_view_like: bool + is_rand_op: bool + + +@dataclass +class _SACModMetadata: + """ + Stores metadata for a module for SAC. + + Attributes: + start_idx (int): The starting index of the module's operators. + force_store_random (bool): Whether to force store random operators in the module. + sac_metadata (List[_SACMetadata]): List of metadata for each operator in the module. + """ + + start_idx: int + force_store_random: bool + sac_metadata: list[_SACMetadata] + + +@dataclass +class SACStats: + """ + A class for storing Activation Checkpointing statistics corresponding to a module. + + Attributes: + func_names (List[str]): List of operator names. + runtimes (List[float]): List of operator runtimes in millliseconds. + memory (List[int]): List of operator memory usage in bytes. + view_like_ops (List[int]): Indices of view-like operators. + rand_ops (List[int]): Indices of random operators. + saved_autograd_ops (List[int]): Indices of operator results saved by autograd engine. + inplace_ops (List[Tuple[int, int]]): Tuple of indices of op and its first parent for Inplace operators. + force_store_random (bool): Whether to force store random operator results. + """ + + func_names: list[str] + runtimes: list[float] + memory: list[int] + view_like_ops: list[int] + rand_ops: list[int] + saved_autograd_ops: list[int] + inplace_ops: list[tuple[int, int]] + force_store_random: bool + + +class MSPS(NamedTuple): + """ + Represents Memory and Runtime Statistics for an operator/operator group. + + Attributes: + func_names (set[str]): Set of operator/operator group names. + op_idx (int): Operator index (group head index in case of operator groups). + memory (int): Memory usage in bytes. + runtime (float): Runtime in milliseconds. + msps (float): Memory per second calculated as memory/runtime. + """ + + func_names: set[str] + op_idx: int + memory: int + runtime: float + msps: float + + +@dataclass +class SACTradeOffStats: + """ + Stores statistics for activation-checkpointing trade-off. + + Attributes: + n_segments (int): Number of piecewise linear segments fitted to the trade-off curve. + slopes (List[float]): Slopes of the pieces of linear segments fitted to the trade-off curve. + intercepts (List[float]): Intercepts of the of the pieces of linear segments fitted to the trade-off curve. + fit_breaks (List[float]): Breakpoints of the of the pieces of linear segments fitted to the trade-off curve. + tradeoff_curve (OrderedDict[float, float]): Trade-off curve data of memory discarded vs recomputation time. + sac_memory (int): Total memory of operations available for activation checkpointing in bytes. + sac_runtime (float): Total runtime of operations available for activation checkpointing in milliseconds. + """ + + n_segments: int + slopes: list[float] + intercepts: list[float] + fit_breaks: list[float] + tradeoff_curve: OrderedDict[float, float] + sac_memory: int + sac_runtime: float + + +@dataclass +class SACGreedyOrderMeta: + """ + Stores metadata for Greedy-order SAC. + + Attributes: + recomputed_ops (set[int]): Set of operator indices to be recomputed. + stored_ops (set[int]): Set of operator indices to be stored. + inplace_op_groups (dict[int, set[int]]): Dictionary of inplace operator groups from group-head to operators. + random_ops_group (dict[int, set[int]]): Dictionary of random op group head to random ops. + msps_meta (list[MSPS]): List of Memory and Runtime Statistics for operators. + """ + + recomputed_ops: set[int] + stored_ops: set[int] + inplace_op_groups: dict[int, set[int]] + random_ops_group: dict[int, set[int]] + msps_meta: list[MSPS] + + +class SACEstimator(TorchDispatchMode): + """ + Estimates the memory and recomputation time trade-offs for applying Selective Activation Checkpointing (SAC). + + This class provides a ``TorchDispatchMode`` based context manager that can be used to estimate the memory and + runtime trade-offs of functions or ``torch.nn.Module``s for Selective Activation Checkpointing (SAC). It provides + detailed statistics and metadata information for operators of each module and provides a greedy order for selecting + the operators to be recomputed/checkpointed. It also constructs the per-module trade-off graph of discarded memory + vs recomputation time for the obtained greedy order. Using ``RuntimeEstimator`` under the hood, it supports two + estimation modes, `operator-level-benchmark` and (`operator-level-cost-model` (roofline model). + + Attributes: + sac_mod_stats (Dict[str, SACStats]): Dictionary from module FQN (fully qualified name) to ``SACStats``. + sac_mod_tradeoff_stats (Dict[str, SACTradeOffStats]): Dictionary from module FQN to ``SACTradeOffStats``. + sac_mod_greedy_order_meta (Dict[str, SACGreedyOrderMeta]): Dictionary from module FQN to ``SACGreedyOrderMeta``. + + Note: + 1) This class is designed to be used under ``FakeTensorMode``. + 2) Currently, it only supports estimation of compute time and memory usage, and does not consider communication. + + Example usage: + + .. code-block:: python + + sac_estimator = SACEstimator() + with FakeTensorMode(): + module = ... + inp = ... + with sac_estimator("operator-level-cost-model"): + output = module(inp) + sac_estimator.display_modulewise_sac_stats(depth=4, print_tabular=True) + """ + + def __init__(self, gpu_type: str | None = None) -> None: + """ + Args: + gpu_type (str | None): Optional datasheet device name (e.g. + ``"NVIDIA H100"``) to pin the roofline estimates to instead of + querying the current device. See ``RuntimeEstimator``. + """ + self.sac_mod_stats: dict[str, SACStats] = {} + self.sac_mod_tradeoff_stats: dict[str, SACTradeOffStats] = {} + self.sac_mod_greedy_order_meta: dict[str, SACGreedyOrderMeta] = {} + self._mod_tracker = ModTracker() + self._sac_metadata: list[_SACMetadata] = [] + self._sac_mod_metadata: dict[str, _SACModMetadata] = {} + self._leaf_modules: set[str] = set() + self._saved_tensor_hook_ctx = torch.autograd.graph.saved_tensors_hooks( + self._pack_hook, lambda x: x + ) + self._saved_tensor_ids: set[int] = set() + self._gpu_type = gpu_type + self._estimate_runtime = RuntimeEstimator._roofline_estimate + + def _pack_hook(self, x: torch.Tensor) -> torch.Tensor: + # Hook function to track underlying storage IDs of tensors + # Updates the _saved_tensor_ids set with the IDs of the tensor's storages + # Used in conjunction with torch.autograd.graph.saved_tensors_hooks + untyped_storages = get_untyped_storages(x) + storage_ids = (hash(st) for st in untyped_storages) + self._saved_tensor_ids.update(storage_ids) + return x + + def _pre_fw_hook(self, mod: nn.Module, inputs: Any) -> None: + # Pre-forward hook function to prepare module metadata + # Tracks module FQN, force store random flag, and ``SACModMetadata`` + # Initializes metadata for non-leaf modules, marks leaf modules + mod_fqn = self._mod_tracker.get_known_fqn(mod) + if mod_fqn is None: + raise AssertionError + num_children = sum(1 for _ in mod.children()) + if num_children > 0: + force_store_random = self._get_force_store_random(inputs) + self._sac_mod_metadata[mod_fqn] = _SACModMetadata( + start_idx=len(self._sac_metadata), + force_store_random=force_store_random, + sac_metadata=[], + ) + else: + self._leaf_modules.add(mod_fqn) + + def _post_fw_hook(self, mod: nn.Module, inputs: Any, outputs: Any) -> None: + # 1. Retrieves the module's FQN and checks if it's a leaf module + # 2. If not a leaf module, computes: + # - ``SACStats`` using the module's metadata and force store random flag + # - ``SACGreedyOrderMeta`` using the computed SAC statistics + mod_fqn = self._mod_tracker.get_known_fqn(mod) + if mod_fqn is None: + raise AssertionError + if mod_fqn in self._leaf_modules: + return + else: + self.sac_mod_stats[mod_fqn] = self._get_sac_stats( + data=self._sac_mod_metadata[mod_fqn].sac_metadata, + force_store_random=self._sac_mod_metadata[mod_fqn].force_store_random, + ) + self.sac_mod_greedy_order_meta[mod_fqn] = self._get_greedy_order_meta( + self.sac_mod_stats[mod_fqn] + ) + + def _get_force_store_random(self, inputs: Any) -> bool: + flat_inputs, _ = tree_flatten(inputs) + return all(not isinstance(x, torch.Tensor) for x in flat_inputs) + + def _get_sac_stats( + self, data: list[_SACMetadata], force_store_random: bool + ) -> SACStats: + # 1. Ignore the operations that should be skipped by SAC such as aten.detach.default because autograd + # inserts those during backward and it breaks the fwd-bwd alignment + filtered_data = [x for x in data if x.func not in OPS_TO_ALWAYS_SKIP] + + ( + ops, + runtimes_, + memory_, + new_ids, + output_ids, + inplace_ops_, + view_like_ops_, + rand_ops_, + ) = zip(*[astuple(x) for x in filtered_data], strict=True) + + # 2. Extract the metadata information + runtimes = list(runtimes_) + memory = list(memory_) + func_names = [op._overloadpacket.__name__ for op in ops] + view_like_ops = [i for i, x in enumerate(view_like_ops_) if x] + rand_ops = [i for i, x in enumerate(rand_ops_) if x] + saved_autograd_ops = [ + i + for i, out_ids in enumerate(output_ids) + if set(out_ids).issubset(self._saved_tensor_ids) + ] + + # 3. Remap the inplace indices as we have removed OPS_TO_ALWAYS_SKIP + # FIXME @sanketpurandare: Fix this by changing the parent of the inplace-op + # to itself if the original parent is in OPS_TO_ALWAYS_SKIP. + try: + inplace_ops = [tuple(map(new_ids.index, x)) for x in inplace_ops_ if x] + except ValueError as err: + raise ValueError( + f"The remapping of inplace ops failed since one of the inplace op parents" + f" must have been present in {OPS_TO_ALWAYS_SKIP}" + ) from err + + # 4. The last operation is always stored as the output of the checkpoint + # block, so we can avoid recomputing it. We set the memory to zero + # instead of adding a new constraint because we want both the 0 and 1 + # endpoints for memory_budget to be valid + # FIXME @sanketpurandare: this heuristic for finding the last non-view non-inplace op + # might not always be correct, which would yield suboptimal policies + last_op = len(ops) - 1 + skip_ops_ = set(view_like_ops) | set({x[0] for x in inplace_ops}) + reversed_skip_ops = sorted(skip_ops_, reverse=True) + for op in reversed_skip_ops: + if op == last_op: + last_op -= 1 + + memory[last_op] = 0 + + # 5. Create a single ``SACStats`` object for the entire block of ``_SACMetadata``. + return SACStats( + func_names=func_names, + runtimes=runtimes, + memory=memory, + view_like_ops=view_like_ops, + rand_ops=rand_ops, + saved_autograd_ops=saved_autograd_ops, + inplace_ops=inplace_ops, # type: ignore[arg-type] + force_store_random=force_store_random, + ) + + def _get_inplace_metadata( + self, func: Any, out_storages: set[UntypedStorage] + ) -> tuple[int, tuple[int, ...], dict[str, tuple[int, ...]]]: + # 1. Get the current index of the metadata obtained so far + curr_idx = len(self._sac_metadata) + # 2. Get the set of active modules that are not leaf + active_mod_fqns: set[str] = { + par for par in self._mod_tracker.parents if par not in self._leaf_modules + } + # 3. Output ids are the identifies of the storage objects corresponding to the tensors + output_ids = tuple(hash(st) for st in out_storages) + # 4. If the function is not inplace, return + if not is_inplace(func): + return curr_idx, output_ids, dict.fromkeys(active_mod_fqns, ()) + + op_idx = curr_idx + # 5. Initialize the parent op ids of the inplace op for each of the active modules + mod_op_parent_idxs: dict[str, int] = dict.fromkeys(active_mod_fqns, -1) + for i, d in enumerate(self._sac_metadata): + # 6. Find the first occurrence of a tensor corresponding to each module that + # shares the same storage as the current tensor + past_output_ids = d.output_ids + if set(output_ids).issubset(set(past_output_ids)): + for mod_fqn, op_parent_idx in mod_op_parent_idxs.items(): + if op_parent_idx == -1: + if acm_stats := self._sac_mod_metadata.get(mod_fqn, None): + if i >= acm_stats.start_idx: + mod_op_parent_idxs[mod_fqn] = i + else: + if mod_fqn != "Global": + raise AssertionError + mod_op_parent_idxs[mod_fqn] = i + # 7. If no parent tensor is found, then it's probably an inplace op on the arguments + # so one can just store the current-op idx as parent idx + for mod_fqn, op_parent_idx in mod_op_parent_idxs.items(): + if op_parent_idx < 0: + mod_op_parent_idxs[mod_fqn] = op_idx + mod_inplace_info = { + mod_fqn: (op_idx, mod_op_parent_idxs[mod_fqn]) + for mod_fqn in active_mod_fqns + } + return curr_idx, output_ids, mod_inplace_info # type: ignore[return-value] + + def __torch_dispatch__( # type: ignore[no-untyped-def] + self, func, types, args=..., kwargs=None + ): + # 1. Get the runtime estimate + out, op_time = self._estimate_runtime(func, args, kwargs) + flat_outs, _ = tree_flatten(out) + out_storages_cuda: set[UntypedStorage] = set() + out_storages_cpu: set[UntypedStorage] = set() + cuda_devices: set[torch.device] = set() + for o in flat_outs: + if isinstance(o, torch.Tensor): + if o.device.type == "cuda": + out_storages_cuda.update(get_untyped_storages(o)) + cuda_devices.add(o.device) + else: + out_storages_cpu.update(get_untyped_storages(o)) + + # Check if there's more than 1 CUDA device + if len(cuda_devices) > 1: + raise AssertionError( + f"{func.__name__}'s output has more than 1 CUDA devices {cuda_devices}" + ) + + # 2. Get the memory consumed by output + nbytes_cuda = sum( + math.ceil(st.nbytes() / _PYTORCH_MIN_ALLOCATE) * _PYTORCH_MIN_ALLOCATE + for st in out_storages_cuda + ) + nbytes_cpu = sum(st.nbytes() for st in out_storages_cpu) + nbytes = nbytes_cuda + nbytes_cpu + # 3. Get the current operator index, output storage identifiers and inplace metadata + out_storages = out_storages_cuda | out_storages_cpu + curr_idx, output_ids, mod_inplace_info = self._get_inplace_metadata( + func, out_storages + ) + # 4. Determine if the function is in-place, random-op or a view-like + is_view_like = is_view_fn(func) or is_inplace_view_fn(func) + is_rand_op = torch.Tag.nondeterministic_seeded in func.tags + if is_view_like: + nbytes = 0 + # sdpa has non-deterministic seed, but might be deterministic + # if no dropout is applied + if func.overloadpacket.__name__ == "_scaled_dot_product_flash_attention": + # pyrefly: ignore [missing-attribute] + is_rand_op = kwargs.get("dropout_p", 0) != 0 + # 5. Create metadata information per active non-leaf module + for mod_fqn in self._mod_tracker.parents: + if mod_fqn in self._leaf_modules: + continue + acm = _SACMetadata( + func=func, + time_taken=op_time, + memory_used=nbytes, + curr_idx=curr_idx, + output_ids=output_ids, + inplace_info=mod_inplace_info[mod_fqn], + is_view_like=is_view_like, + is_rand_op=is_rand_op, + ) + if acm_stats := self._sac_mod_metadata.get(mod_fqn, None): + acm_stats.sac_metadata.append(acm) + else: + if mod_fqn != "Global": + raise AssertionError(f"Module {mod_fqn} not found in AC Mod Stats") + self._sac_metadata.append(acm) + + return out + + def _get_greedy_order_meta(self, sac_stats: SACStats) -> SACGreedyOrderMeta: + # An inplace-op group is a set of inplace-ops that operate on the same underlying tensor storage. + # 1. inplace_op_groups: A dictionary from the top-most parent of inplace-ops to the inplace-ops in the group + # The top-most op can itself be an inplace-op or can be a non-inplace op. + # 2. inplace_op_to_group_head: A dictionary that maps all the inplace-ops to their respective group heads. + inplace_op_groups: dict[int, set[int]] = {} + inplace_op_to_group_head: dict[int, int] = dict(sac_stats.inplace_ops) + + # Initialize inplace_op_groups using inplace_op_to_group_head + for op_idx, group_head_idx in inplace_op_to_group_head.items(): + op_group = inplace_op_groups.setdefault(group_head_idx, {group_head_idx}) + op_group.add(op_idx) + + # Like inplace ops, all of the random ops in the function/module should all be either recomputed or saved + # as a group. This is because, they affect the random seed generator. If force_store_random is set True, + # all of the random ops will be stored by default. For ease of manageability, we store the top-most random op + # as the leader of the random_ops_group. + random_ops_group: dict[int, set[int]] = {} + random_group_head_idx = min(sac_stats.rand_ops, default=-1) + has_rand_ops = bool(sac_stats.rand_ops) + if has_rand_ops: + random_ops_group[random_group_head_idx] = set(sac_stats.rand_ops) + + # 1. Random ops are stored if force_store_random is set + # 2. View-like ops are recomputed by default + # 3. For inplace_op_groups: + # a) If the head of this group is an inplace op, then we have to store the entire group. + # b) If any op in the group is random and force_store_random is set, then entire group will be stored. + # c) If none of ops in the group are random and the head of the group is not an in-place op, then + # this group can be considered for recomputation in its entirety + stored_ops: set[int] = set() + recomputed_ops: set[int] = set() + # Case 1: + if has_rand_ops and sac_stats.force_store_random: + stored_ops.add(random_group_head_idx) + # Case 2: + recomputed_ops.update(set(sac_stats.view_like_ops)) + + for group_head_idx, op_group in inplace_op_groups.items(): + # Case 3a: + if group_head_idx in inplace_op_to_group_head: + stored_ops.add(group_head_idx) + # Case 3b: + if ( + sac_stats.force_store_random & len(op_group & set(sac_stats.rand_ops)) + > 0 + ): + stored_ops.add(group_head_idx) + + # The potential recompute candidates are populated as: + recompute_candidates: set[int] = set() + # 1) The random group head if it is not stored + if has_rand_ops and random_group_head_idx not in stored_ops: + recompute_candidates.add(random_group_head_idx) + # 2) The in-place op group heads that are not stored + recompute_candidates.update(set(inplace_op_groups.keys()) - stored_ops) + # 3) The non-inplace and non-random ops that are neither stored nor recomputed by default + recompute_candidates.update( + set(range(len(sac_stats.memory))) + - recomputed_ops + - stored_ops + - set(inplace_op_to_group_head.keys()) + - set(sac_stats.rand_ops) + ) + + # We define msps for a recomp candidate as the ratio of memory/runtime aka memory savings per second + msps_meta: list[MSPS] = [] + for cand_idx in recompute_candidates: + op_indices = {cand_idx} + if cand_idx in inplace_op_groups: + op_indices.update(inplace_op_groups[cand_idx]) + if has_rand_ops and cand_idx == random_group_head_idx: + op_indices.update(sac_stats.rand_ops) + + mem = sum(sac_stats.memory[op_idx] for op_idx in op_indices) + runtime = sum(sac_stats.runtimes[op_idx] for op_idx in op_indices) + func_names = {sac_stats.func_names[op_idx] for op_idx in op_indices} + msps = (mem / runtime) if runtime > 0 else sys.float_info.max + msps_meta.append(MSPS(func_names, cand_idx, mem, runtime, msps)) + # We choose candidates to be recomputed based on increasing msps + msps_meta.sort(key=lambda x: x.msps, reverse=True) + return SACGreedyOrderMeta( + recomputed_ops, stored_ops, inplace_op_groups, random_ops_group, msps_meta + ) + + def _get_sac_tradeoff_pwlf_stats( + self, + sac_stats: SACStats, + greedy_order_meta: SACGreedyOrderMeta, + n_segments: int = 2, + save_tradeoff_graph: bool = False, + filename: str = "ac_tradeoff", + ) -> SACTradeOffStats: + try: + import numpy as np # type: ignore[import-not-found] + import pwlf # type: ignore[import-untyped, import-not-found] + except ImportError as err: + raise ImportError("Please install pwlf and numpy package.") from err + + stored_ops, recomputed_ops, inplace_op_groups, random_ops_group, msps_meta = ( + greedy_order_meta.stored_ops, + greedy_order_meta.recomputed_ops, + greedy_order_meta.inplace_op_groups, + greedy_order_meta.random_ops_group, + greedy_order_meta.msps_meta, + ) + # 1. Initialize the discarded memory and recomputation runtime to sum of already chosen recomputed_ops + recomp_indices: set[int] = set() + for r_idx in recomputed_ops: + recomp_indices.add(r_idx) + if r_idx in inplace_op_groups: + recomp_indices.update(inplace_op_groups[r_idx]) + if r_idx in random_ops_group: + recomp_indices.update(random_ops_group[r_idx]) + + discarded_mem = sum(sac_stats.memory[op_idx] for op_idx in recomp_indices) + recomp_runtime = sum(sac_stats.runtimes[op_idx] for op_idx in recomp_indices) + # 2. Initialize the max recomputation time and total recomputation memory + sac_runtime = sum(sac_stats.runtimes) + sac_memory = sum(sac_stats.memory) + # 3. Tradeoff curve stores the KV pair of the discarded memory to total memory and, + # recomputation time to total runtime incurred. + delta = 1e-2 + tradeoff_curve = OrderedDict() + # 4. Initialize the trade-off curve with the stats of already chosen recomputed_ops + tradeoff_curve[(discarded_mem / sac_memory) + delta] = ( + recomp_runtime / sac_runtime + ) + # 5. Update the trade-off curve with memory and runtime stats of SAC candidates in the + # greedy order of their ``MSPS``. + for cand in msps_meta: + discarded_mem += cand.memory + recomp_runtime += cand.runtime + tradeoff_curve[(discarded_mem / sac_memory) + delta] = ( + recomp_runtime / sac_runtime + ) + # 6. Finally, we add the memory and recomputation time of the always stored ops. + stored_indices: set[int] = set() + for s_idx in stored_ops: + stored_indices.add(s_idx) + if s_idx in inplace_op_groups: + stored_indices.update(inplace_op_groups[s_idx]) + if s_idx in random_ops_group: + stored_indices.update(random_ops_group[s_idx]) + discarded_mem += sum(sac_stats.memory[op_idx] for op_idx in stored_indices) + recomp_runtime += sum(sac_stats.runtimes[op_idx] for op_idx in stored_indices) + tradeoff_curve[(discarded_mem / sac_memory) + delta] = ( + recomp_runtime / sac_runtime + ) + x_ = list(tradeoff_curve.keys()) + y_ = list(tradeoff_curve.values()) + # 7. We shift the y values to left and x values to right to upperbound the trade-off function + # TODO: Write a better explanation why this needs to be done + x = x_[: len(x_) - 1] + y = y_[1:] + tradeoff_pwlf = pwlf.PiecewiseLinFit(x, y) + # 8. Fit a piecewise linear function with the specified number of segments to the trade-off curve. + n_segments = max(min(len(x) - 2, n_segments), 1) + tradeoff_pwlf.fit(n_segments=n_segments) + + # save prediction graph + def save_prediction_graph( + pwlf_: pwlf.PiecewiseLinFit, x: list[float], y: list[float], filename: str + ) -> None: + try: + import matplotlib.pyplot as plt # type: ignore[import-not-found] + import numpy as np # type: ignore[import-not-found] + except ImportError as err: + raise ImportError( + "Install matplotlib and numpy using pip: pip install matplotlib numpy" + ) from err + # predict for the determined points + xHat = np.linspace(min(x), max(x), num=10000) + yHat = pwlf_.predict(xHat) + + # plot the results + plt.figure() + plt.plot(x, y, "o", label="Shifted") + plt.plot(xHat, yHat, "-", label="Predicted") + plt.plot(x_, y_, "x", label="Original") + plt.ylabel("Recomp time / Total recomp time") + plt.xlabel("Memory discarded / Total memory") + plt.legend() + plt.title(f"{filename}") + plt.suptitle( + f"Total Memory = {sac_memory} B Total Runtime = {sac_runtime:.4f} ms", + fontsize=10, + ) + folder_name = "tradeoff_graphs" + if not os.path.exists(folder_name): + os.makedirs(folder_name) + # Save the plots in the folder + plt.savefig(os.path.join(folder_name, f"{filename}.png")) + + if save_tradeoff_graph: + save_prediction_graph(tradeoff_pwlf, x, y, filename) + # 9. Obtain the slopes, intercepts and breakpoints of the fitted piecewise linear functions + slopes = tradeoff_pwlf.calc_slopes().tolist() + if not ( + isinstance(tradeoff_pwlf.intercepts, np.ndarray) + and isinstance(tradeoff_pwlf.fit_breaks, np.ndarray) + ): + raise AssertionError + intercepts = tradeoff_pwlf.intercepts.tolist() + fit_breaks = tradeoff_pwlf.fit_breaks.tolist() + return SACTradeOffStats( + n_segments=n_segments, + slopes=slopes, + intercepts=intercepts, # type: ignore[arg-type] + fit_breaks=fit_breaks, # type: ignore[arg-type] + tradeoff_curve=tradeoff_curve, + sac_memory=sac_memory, + sac_runtime=sac_runtime, + ) + + def display_sac_stats( + self, sac_stats: SACStats, print_tabular: bool = False + ) -> None: + """ + Displays the SAC statistics. + + Args: + sac_stats (SACStats): The SAC statistics to display. + print_tabular (bool, optional): Whether to print the statistics in a tabular format. Defaults to False. + + Prints: + 1. Total Memory: The total memory usage in bytes. + 2. Total Runtime: The total runtime in milliseconds. + 3. Store Random: A flag indicating whether to force store random operator results. + + Followed by a table with the following columns: + 1. Op Idx: The operator index. + 2. Op Name: The operator name. + 3. Runtimes (ms): The operator runtime in milliseconds. + 4. Memory (B): The operator memory usage in bytes. + 5. View-like: A flag indicating whether the operator is view-like. + 6. Random: A flag indicating whether the operator is random. + 7. Saved Autograd: A flag indicating whether the operator's result is saved by autograd engine. + 8. In-place: The index of the operator's first parent, or None if not in-place. + + If print_tabular is True, the table is printed in a tabular format. + Otherwise, the table is printed in a plain text format. + """ + print( + f"Total Memory: {sum(sac_stats.memory)} B Total Runtime: {sum(sac_stats.runtimes)} ms" + f" Store Random: {sac_stats.force_store_random}" + ) + table_data = [] + op_parent = dict(sac_stats.inplace_ops) + for i, fn_name in enumerate(sac_stats.func_names): + row = [ + str(i), + fn_name, + f"{sac_stats.runtimes[i]:.4f}", + str(sac_stats.memory[i]), + str(i in sac_stats.view_like_ops), + str(i in sac_stats.rand_ops), + str(i in sac_stats.saved_autograd_ops), + str(op_parent.get(i)), + ] + table_data.append(row) + # Define headers + headers = [ + "Op Idx", + "Op Name", + "Runtimes(ms)", + "Memory (B)", + "View-like", + "Random", + "Saved Autograd", + "In-place", + ] + if print_tabular: + _display_stats_tabular(headers, table_data) + else: + max_widths = [0 for _ in range(len(headers))] + table_data.insert(0, headers) + for row in table_data: + for i, elem in enumerate(row): + max_widths[i] = max(max_widths[i], len(elem)) + for row in table_data: + print( + "\t".join( + [f"{elem:<{max_widths[i]}}" for i, elem in enumerate(row)] + ) + ) + + def display_sac_tradeoff_stats( + self, + greedy_order_meta: SACGreedyOrderMeta, + sac_stats: SACStats, + print_tabular: bool = False, + ) -> None: + """ + Displays the SAC trade-off statistics. + + Args: + greedy_order_meta (SACGreedyOrderMeta): The SAC greedy order metadata. + sac_stats (SACStats): The SAC statistics. + print_tabular (bool, optional): Whether to print the statistics in a tabular format. Defaults to False. + + Prints: + A table with the following columns: + 1. Op Id(s): The operator index(es). + 2. Op Name(s): The operator name(s). + 3. Discarded Mem (%): The percentage of discarded memory. + 4. Discarded Mem (B): The discarded memory in bytes. + 5. Recomp time (%): The percentage of recomputed time. + 6. Recomp time (ms): The recomputed time in milliseconds. + 7. MSPS: The memory per second. + 8. Always Stored: A flag indicating whether the operator is always stored. + 9. Always Recomputed: A flag indicating whether the operator is always recomputed. + + If print_tabular is True, the table is printed in a tabular format. + Otherwise, the table is printed in a plain text format. + """ + table_data = [] + total_memory, total_runtime = sum(sac_stats.memory), sum(sac_stats.runtimes) + discarded_mem: int = 0 + recomp_runtime: float = 0.0 + + def append_row( + op_indices: set[int], + func_names: set[str], + msps: float | None = None, + stored: bool | None = False, + recomputed: bool | None = False, + ) -> None: + row = [ + str(op_indices), + str(func_names), + f"{discarded_mem / total_memory:.4f}", + str(discarded_mem), + f"{recomp_runtime / total_runtime:.4f}", + str(recomp_runtime), + f"{msps:.2e}" if msps is not None else str(nan), + str(stored), + str(recomputed), + ] + table_data.append(row) + + stored_ops, recomputed_ops, inplace_op_groups, random_ops_group, msps_meta = ( + greedy_order_meta.stored_ops, + greedy_order_meta.recomputed_ops, + greedy_order_meta.inplace_op_groups, + greedy_order_meta.random_ops_group, + greedy_order_meta.msps_meta, + ) + + for op_idx in recomputed_ops: + op_indices: set[int] = {op_idx} + if op_idx in inplace_op_groups: + op_indices.update(inplace_op_groups[op_idx]) + if op_idx in random_ops_group: + op_indices.update(random_ops_group[op_idx]) + discarded_mem += sum(sac_stats.memory[i] for i in op_indices) + recomp_runtime += sum(sac_stats.runtimes[i] for i in op_indices) + func_names = {sac_stats.func_names[i] for i in op_indices} + append_row(op_indices, func_names, recomputed=True) + + for cand in msps_meta: + discarded_mem += cand.memory + recomp_runtime += cand.runtime + op_indices = {cand.op_idx} + if cand.op_idx in inplace_op_groups: + op_indices.update(inplace_op_groups[cand.op_idx]) + if cand.op_idx in random_ops_group: + op_indices.update(random_ops_group[cand.op_idx]) + append_row(op_indices, cand.func_names, msps=cand.msps) + + for op_idx in stored_ops: + op_indices = {op_idx} + if op_idx in inplace_op_groups: + op_indices.update(inplace_op_groups[op_idx]) + if op_idx in random_ops_group: + op_indices.update(random_ops_group[op_idx]) + discarded_mem += sum(sac_stats.memory[i] for i in op_indices) + recomp_runtime += sum(sac_stats.runtimes[i] for i in op_indices) + func_names = {sac_stats.func_names[i] for i in op_indices} + append_row(op_indices, func_names, stored=True) + + headers = [ + "Op Id(s)", + "Op Name(s)", + "Discarded Mem (%)", + "Discarded Mem (B)", + "Recomp time (%)", + "Recomp time (ms)", + "MSPS", + "Always Stored", + "Always Recomputed", + ] + if print_tabular: + _display_stats_tabular(headers, table_data) + else: + max_widths = [0 for _ in range(len(headers))] + table_data.insert(0, headers) + for row in table_data: + for i, elem in enumerate(row): + max_widths[i] = max(max_widths[i], len(elem)) + for row in table_data: + print( + "\t".join( + [f"{elem:<{max_widths[i]}}" for i, elem in enumerate(row)] + ) + ) + + def pwlf_sac_tradeoff_curve( + self, + n_segments: int = 2, + save_tradeoff_graphs: bool = False, + ) -> None: + """ + Fits a piecewise linear function with the specified sumber of segments to the SAC trade-off curve of + discarded memory vs recomputation time. + + Args: + n_segments (int, optional): The number of segments to be used for fitting the piecewise linear function to + the trade-off curve. Defaults to 2. + save_tradeoff_graphs (bool, optional): Whether to save the trade-off graphs to file. Defaults to False. + + If save_tradeoff_graphs is True, the trade-off graphs are saved to file using the module FQN as the filename. + """ + for mod_fqn, sac_stats in self.sac_mod_stats.items(): + self.sac_mod_tradeoff_stats[mod_fqn] = self._get_sac_tradeoff_pwlf_stats( + sac_stats=sac_stats, + greedy_order_meta=self.sac_mod_greedy_order_meta[mod_fqn], + n_segments=n_segments, + save_tradeoff_graph=save_tradeoff_graphs, + filename=mod_fqn, + ) + + def display_modulewise_sac_stats( + self, depth: int = 2, print_tabular: bool = False + ) -> None: + """ + Displays the SAC and trade-off statistics for each module. + + Args: + depth (int, optional): The maximum depth of modules to display. Defaults to 2. + print_tabular (bool, optional): Whether to print the statistics in a tabular format. Defaults to False. + + Prints: + For each module with depth less than or equal to the specified depth: + 1. The SAC statistics for the module (using display_sac_stats). + 2. The SAC trade-off statistics for the module (using display_sac_tradeoff_stats). + + If print_tabular is True, the statistics are printed in a tabular format. + Otherwise, the statistics are printed in a plain text format. + """ + for mod_fqn, sac_stats in self.sac_mod_stats.items(): + mod_depth = mod_fqn.count(".") + 1 + if mod_depth > depth: + continue + print(f"Module: {mod_fqn}") + self.display_sac_stats(sac_stats, print_tabular) + print(f"AC Trade-off for Module: {mod_fqn} MSPS = Memory/Runtime") + self.display_sac_tradeoff_stats( + self.sac_mod_greedy_order_meta[mod_fqn], sac_stats, print_tabular + ) + + def __call__(self, estimate_mode_type: str) -> Self: + """ + Sets the estimate mode type. + + Currently supported modes: + - "operator-level-benchmark": Estimates runtime using operator benchmarking. + - "operator-level-cost-model": Estimates runtime using roofline cost model. + + Args: + estimate_mode_type (str): The type of estimate mode to use. + + Returns: + SACEstimator: The SAC estimator instance. + + Raises: + NotImplementedError: If the estimate mode type is not supported. + """ + if estimate_mode_type == "operator-level-benchmark": + self._estimate_runtime = RuntimeEstimator._benchmark_estimate + elif estimate_mode_type == "operator-level-cost-model": + self._estimate_runtime = RuntimeEstimator._roofline_estimate + else: + raise NotImplementedError( + f"estimate_mode_type {estimate_mode_type} not supported" + ) + return self + + def __enter__(self) -> Self: # type: ignore[no-untyped-def] + fake_mode = active_fake_mode() + if not isinstance(fake_mode, FakeTensorMode): + raise AssertionError("SAC Estimator should be called in FakeTensorMode") + RuntimeEstimator.fake_mode = fake_mode + RuntimeEstimator.gpu_type = self._gpu_type + self._mod_tracker.register_user_hooks( + pre_fw_hook=self._pre_fw_hook, + post_fw_hook=self._post_fw_hook, + ) + self._mod_tracker.__enter__() + self._saved_tensor_hook_ctx.__enter__() + return super().__enter__() + + def __exit__(self, *args: Any) -> None: # type: ignore[no-untyped-def] + self._saved_tensor_hook_ctx.__exit__() + self._mod_tracker.__exit__(*args) + super().__exit__(*args) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_ilp.py b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_ilp.py new file mode 100644 index 0000000000000000000000000000000000000000..7aad006c204d91654cee6bf1aa8efbb290719fca --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/_tools/sac_ilp.py @@ -0,0 +1,297 @@ +import logging +import math +from enum import IntEnum + +from torch.distributed._tools.ilp_utils import Graph, is_submodule +from torch.distributed._tools.sac_estimator import SACStats + + +try: + from pulp import ( # type: ignore[import-untyped,import-not-found] + lpDot, + LpInteger, + LpMaximize, + LpMinimize, + LpProblem, + LpStatus, + lpSum, + LpVariable, + PULP_CBC_CMD, + value, + ) +except ImportError as err: + raise ImportError( + "Please install pulp package. See: https://github.com/coin-or/pulp." + ) from err + +# Create a logger object +logger = logging.getLogger(__name__) + +# Set the logging level to INFO +logger.setLevel(logging.INFO) + + +def sac_milp( + graph: Graph, + memory_budget: float, + world_size: int = 1, + ac_units: list[str] | None = None, + fsdp_units: list[str] | None = None, +) -> tuple[dict[str, float], float, int]: + """ + MILP to decide which modules to AC and how much memory to discard. + The objective is to minimize recomputation time. + The constraint is to ensure peak memory is under budget. + + Args: + graph: graph representation of the model as a module submodule tree + where each node is a submodule with memory & runtime stats + memory_budget: memory budget in GiB + world_size: number of GPUs. In the case of FSDP, world_size will be + used to compute the amount of parameter and gradient memory on each rank + ac_units: a list of user-specified AC units. + fsdp_units: a list of FSDP units. AC units cannot be supermodules of FSDP units. + + Returns: + Dict[str, float]: the optimal SAC solution, mapping from module fqn to + the percentage of activation memory to **discard** + float: the recomputation time of the optimal SAC solution + int: upper bound on the peak memory of the optimal SAC solution. + note that value of -1 means that the ILP solver failed to find a solution. + + """ + num_nodes = len(graph.nodes) + max_runtime = max( + (graph.nodes[i]["sac_runtime"] for i in range(num_nodes)), default=1 + ) + M = max(10**2, 2 * max_runtime) + MEM_MULTIPLIER = 2**30 + + # Create a MILP problem + prob = LpProblem("SAC", LpMinimize) + + # Create decision variables + # y_i: indicator for if module i is AC'ed + y = LpVariable.matrix("y", list(range(num_nodes)), 0, 1, LpInteger) + # r_i: percentage of discarded activation memory + r = LpVariable.matrix("r", list(range(num_nodes)), 0, 1) + # d_i: discarded activation memory for module i + d = LpVariable.matrix("d", list(range(num_nodes)), 0) + # a_i: total activation memory at module i + a = LpVariable.matrix("a", list(range(num_nodes)), 0) + # m_i: memory at module i, combining parameters, gradients, and activations + m = LpVariable.matrix("m", list(range(num_nodes)), 0) + # rcp_i: percentage of recomputation time + rcp = LpVariable.matrix("rcp", list(range(num_nodes)), 0) + # rct_i: recomputation time for module i (in ms) + rct = LpVariable.matrix("rct", list(range(num_nodes)), 0) + # max_m: peak memory + max_m = LpVariable("max_m", 0) + + # Add constraints + # [Constraint] User specified AC units + if ac_units: + ac_units_set = set(ac_units) + for i in range(num_nodes): + if graph.nodes[i]["fqn"] not in ac_units_set: + prob += y[i] == 0 + + # [Constraint] AC units cannot be supmodules of user specified FSDP units + if fsdp_units: + for i in range(num_nodes): + if any( + is_submodule(fsdp_unit, graph.nodes[i]["fqn"]) + for fsdp_unit in fsdp_units + ): + prob += y[i] == 0 + + # [Constraint] No nested AC units + for i in range(num_nodes): + for j in range(i + 1, num_nodes): + if graph.ad_matrix[i][j] == 1: + prob += y[i] + y[j] <= 1 + + # [Constraint] Do not AC leaf modules + for i in range(num_nodes): + if graph.nodes[i]["is_leaf"]: + prob += y[i] == 0 + + # [Constraint] Express amount of discarded activation memory + for i in range(num_nodes): + # There are two measures for activation memory: ACM and IA + # 1. IA is the activation memory saved when not using AC + # 2. ACM is the total activation memory, including those + # that are not typically saved when not using AC + # Note: ACM >= IA + if (not graph.nodes[i]["is_leaf"]) and graph.nodes[i][ + "sac_memory" + ] < graph.nodes[i]["act_fw_per_module"]: + logger.warning("For module {%s}: ", graph.nodes[i]["fqn"]) + logger.warning( + "activation memory from memory tracker is {%d},", + graph.nodes[i]["act_fw_per_module"], + ) + logger.warning( + "activation memory from SAC estimator is {%d}.", + graph.nodes[i]["sac_memory"], + ) + logger.warning("Something is wrong. Please check!") + logger.warning("Overriding the latter with the former.") + graph.nodes[i]["sac_memory"] = graph.nodes[i]["act_fw_per_module"] + ACM_i = graph.nodes[i]["sac_memory"] / MEM_MULTIPLIER + IA_i = graph.nodes[i]["act_fw_per_module"] / MEM_MULTIPLIER + prob += d[i] == ACM_i * r[i] - (ACM_i - IA_i) * y[i] + + # [Constraint] Ensure correctness of r_i + # There are two parts to its correctness + # 1. r_i > 0 only if y_i == 1 (discard only if it is an AC unit) + # 2. r_i needs to be large enough to cover the difference between + # ACM and IA. Otherwise, we are not saving any memory + for i in range(num_nodes): + prob += y[i] >= r[i] + if graph.nodes[i]["is_leaf"]: + continue + ACM_i = graph.nodes[i]["sac_memory"] / MEM_MULTIPLIER + IA_i = graph.nodes[i]["act_fw_per_module"] / MEM_MULTIPLIER + prob += r[i] >= (ACM_i - IA_i) / ACM_i * y[i] + + # [Constraint] Express total activation memory in the backward pass + for i in range(num_nodes): + AG_i = graph.nodes[i]["act_grad_per_module"] / MEM_MULTIPLIER + TA_i = graph.nodes[i]["act_total"] / MEM_MULTIPLIER + # related to discarded amount of memory + pos = graph.nodes[i]["pos_fw_post_order"] + coeff = [0] * num_nodes + for p in range(pos): + j = graph.name2node[graph.fw_post_order[p]]["index"] + coeff[j] = 1 + prob += a[i] == TA_i + AG_i - lpDot(coeff, d) + + # [Constraint] Express the total amount of memory at each module + # Note that unsharded parameters and gradients are not included here + P_1 = graph.nodes[0]["param_per_module"] / MEM_MULTIPLIER + for i in range(num_nodes): + TG_i = graph.nodes[i]["grad_total"] / MEM_MULTIPLIER + prob += m[i] == a[i] + (P_1 + TG_i) / world_size + + # [Constraint] Express peak memory + for i in range(num_nodes): + prob += max_m >= m[i] + + # [Constraint] Express percentage of recomputation time + for i in range(num_nodes): + for s in range(graph.nodes[i]["n_segments"]): + slope = graph.nodes[i]["slopes"][s] + intercept = graph.nodes[i]["intercepts"][s] + prob += rcp[i] >= slope * r[i] + intercept + + # [Constraint] Express recomputation time + # rct_i = (rcp_i * ACT_i) if y_i == 1 else 0 + for i in range(num_nodes): + ACT_i = graph.nodes[i]["sac_runtime"] + prob += rct[i] <= M * y[i] + prob += rct[i] <= ACT_i * rcp[i] + prob += rct[i] >= ACT_i * rcp[i] - M * (1 - y[i]) + + # [Constraint] Peak memory should be below budget + prob += max_m <= memory_budget + + # Set Objeictive + prob += lpSum(rct) + + # Solve + solver = PULP_CBC_CMD(gapRel=0.05, timeLimit=180, msg=0) + status = prob.solve(solver) + + # If solver fails, print status and return empty solution + if status != 1: + logger.error("Solver failed to find a solution: %s", LpStatus[status]) + return {}, 0, -1 + + # Gather and return solution if optimal solution is found + ac_decisions = {} + for i in range(num_nodes): + if round(y[i].varValue) == 1: + ac_decisions[graph.nodes[i]["fqn"]] = round(r[i].varValue, 4) + recomputation_time = round(value(prob.objective), 2) + peak_mem = round(max_m.varValue * MEM_MULTIPLIER) + + return ac_decisions, recomputation_time, peak_mem + + +class SACDecision(IntEnum): + RECOMPUTE = 0 + SAVE = 1 + + +def get_optimal_checkpointing_policy_per_module( + sac_stats: SACStats, memory_budget: float +) -> list[int]: + """ + This is adapted from -- + https://github.com/facebookresearch/xformers/blob/c6c0ac31f1b08542a0bc27278c6ed10f825f6963/xformers/checkpoint.py#L375 + + Given the SACStats of a module, including list of operators, their memory, runtimes, and metadata, + decide via MILP an optimal set of operators to checkpoint under a given ``memory_budget``. + + Args: + sac_stats: the SACStats object of the module + memory_budget: a float between zero and one + + Returns: + List[int]: the decision whether each operator should be saved (1) or recomptued (0). + """ + if not (0 <= memory_budget <= 1): + raise ValueError( + f"`memory_budget` must be a float between 0 and 1. Got {memory_budget}." + ) + num_ops = len(sac_stats.func_names) + + # Create a MILP problem + prob = LpProblem("SAC-per-module", LpMaximize) + + # Create decision variables + # x[i] = 1 means the i-th operator should be saved, otherwise it should be recomputed + x = LpVariable.matrix("x", list(range(num_ops)), 0, 1, LpInteger) + + # Add constraints + # [Constraint] random ops should be saved if ``force_store_random`` is True + # otherwise, random ops should either be all recomputed or all saved + if sac_stats.force_store_random: + for i in sac_stats.rand_ops: + prob += x[i] == SACDecision.SAVE.value + else: + for i1, i2 in zip(sac_stats.rand_ops[:-1], sac_stats.rand_ops[1:]): + prob += x[i1] == x[i2] + + # [Constraint] view-like ops should always be recomputed + for i in sac_stats.view_like_ops: + prob += x[i] == SACDecision.RECOMPUTE.value + + # [Constraint] inplace ops should always be done in conjunction with its parent op + for op, op_parent in sac_stats.inplace_ops: + if op != op_parent: + prob += x[op] == x[op_parent] + else: + prob += x[op] == SACDecision.SAVE.value + + # [Constraint] saved memory should be under the ``memory_budget`` + max_memory = math.ceil(memory_budget * sum(sac_stats.memory)) + prob += lpDot(x, sac_stats.memory) <= max_memory + + # [Objective] minimize recomputation time, note the ILP is a maximization problem + # because x[i] == 1 means the op is saved (not recomputed), and thus recomputation + # time is sum(sac_stats.runtimes) - lpDot(x, sac_stats.runtimes) + prob += lpDot(x, sac_stats.runtimes) + + # Solve + solver = PULP_CBC_CMD(gapRel=0.05, timeLimit=10, msg=0) + status = prob.solve(solver) + + # If solver fails, print status and return empty solution + if status != 1: + logger.error("Solver failed to find a solution: %s", LpStatus[status]) + return [] + + # Gather and return solution if optimal solution is found + return [round(x[i].varValue) for i in range(num_ops)] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..06c814295699405de9a8f8cf7f6a861b07b63a05 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__init__.py @@ -0,0 +1 @@ +from .join import Join, Joinable, JoinHook diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7053dcf6acf759827fe14013b68484d47c6a6cbc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/join.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/join.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..16235e123a00db5b3d46147519c1e9fa6aeab641 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/__pycache__/join.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f0952f51dd0c4a6586410ca76420ba501ff2cccf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/checkpoint_wrapper.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/checkpoint_wrapper.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c78df0b468267895f935580f930d33e640243b33 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/__pycache__/checkpoint_wrapper.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py new file mode 100644 index 0000000000000000000000000000000000000000..8c8f6890f1a917f2336fd9d00d81bf19571b595d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py @@ -0,0 +1,313 @@ +# mypy: allow-untyped-defs +from abc import ABC, abstractmethod +from collections.abc import Callable, Iterator +from enum import auto, Enum +from functools import partial +from typing import Any + +import torch +import torch.nn as nn +from torch.autograd.graph import save_on_cpu +from torch.distributed.utils import _pack_kwargs, _replace_by_prefix, _unpack_kwargs +from torch.utils._typing_utils import copy_method_params +from torch.utils.checkpoint import checkpoint as torch_utils_checkpoint + + +_CHECKPOINT_WRAPPED_MODULE = "_checkpoint_wrapped_module" +_CHECKPOINT_PREFIX = _CHECKPOINT_WRAPPED_MODULE + "." + + +class CheckpointImpl(Enum): + REENTRANT = auto() + NO_REENTRANT = auto() + + +class ActivationWrapper(torch.nn.Module, ABC): + """ + Base class for Activation Checkpoint and Activation Offload. + + Not meant to be instantiated directly. + """ + + def __init__(self, mod): + super().__init__() + self._checkpoint_wrapped_module = mod + # state_dict post hook to remove prefix to allow loading into a + # non-checkpoint wrapped module. + self._register_state_dict_hook(self._post_state_dict_hook) + # load_state_dict pre-hook to allow loading back into + # checkpoint-wrapped module. + self.register_load_state_dict_pre_hook(self._pre_load_state_dict_hook) + + @abstractmethod + def forward(self, *args, **kwargs): + raise ValueError("Subclasses should implement forward().") + + def __getattr__(self, name: str) -> Any: + """Forward missing attributes to wrapped module.""" + try: + return super().__getattr__(name) # defer to nn.Module's logic + except AttributeError: + return getattr(self._checkpoint_wrapped_module, name) + + def __getitem__(self, key: int) -> Any: + """Forward indexing calls in case the module is a nn.Sequential.""" + return self._checkpoint_wrapped_module.__getitem__(key) # type: ignore[operator] + + @copy_method_params(torch.nn.Module.named_parameters) + def named_parameters( + self, + *args, + **kwargs, + ) -> Iterator[tuple[str, torch.nn.Parameter]]: + """ + Override :meth:`named_parameters()` to intercept parameter names. + + remove all occurrences of ``_CHECKPOINT_PREFIX``. + """ + for param_name, param in super().named_parameters(*args, **kwargs): + yield param_name.replace(_CHECKPOINT_PREFIX, ""), param + + @staticmethod + def _post_state_dict_hook( + module: nn.Module, + state_dict: dict[str, Any], + prefix: str, + *args: Any, + ) -> dict[str, Any]: + """ + _post_state_dict_hook() is called after the state_dict() of this FSDP module is executed. + + For ``checkpoint_wrapper``, it will strip checkpoint-wrapped module prefix, + so that this module can be loaded into non-checkpointed modules. + It would still be able to be loaded into checkpoint-wrapped modules as this class, + adds the prefix back before loading the state_dict. + """ + _replace_by_prefix(state_dict, f"{prefix}{_CHECKPOINT_PREFIX}", prefix) + return state_dict + + @staticmethod + def _pre_load_state_dict_hook( + module: nn.Module, + state_dict: dict[str, Any], + prefix: str, + *args: Any, + ) -> None: + """ + ``_pre_state_dict_hook` is called before ``self._load_from_state_dict()`` is called. + + For ``checkpoint_wrapper``, it will add back the module + prefix so that non-checkpointed modules can be loaded into + checkpoint_wrapper modules properly. + """ + _replace_by_prefix(state_dict, prefix, prefix + f"{_CHECKPOINT_PREFIX}") + + +class OffloadWrapper(ActivationWrapper): + def forward(self, *args, **kwargs): + with save_on_cpu(pin_memory=True): + return self._checkpoint_wrapped_module(*args, **kwargs) + + +class CheckpointWrapper(ActivationWrapper): + """ + An ``nn.Module`` that wraps another ``nn.Module`` with checkpointing. + + Note that this module is not meant to be used directly but instead, + it is to be used through the ``checkpoint_wrapper`` function. + """ + + def __init__( + self, + mod: torch.nn.Module, + checkpoint_impl: CheckpointImpl = CheckpointImpl.NO_REENTRANT, + checkpoint_fn=None, + **checkpoint_fn_kwargs, + ): + super().__init__(mod) + self.checkpoint_impl = checkpoint_impl + if checkpoint_fn is None: + # use torch.utils.checkpoint + self.checkpoint_fn = partial( + torch_utils_checkpoint, + use_reentrant=(self.checkpoint_impl == CheckpointImpl.REENTRANT), + **checkpoint_fn_kwargs, + ) + else: + # Construct user-specified checkpoint function. + self.checkpoint_fn = partial( + checkpoint_fn, + **checkpoint_fn_kwargs, + ) + + def forward(self, *args, **kwargs): + # Support keyword arguments for reentrant checkpoint. Note that this + # only works if user has specified self.checkpoint_impl and is not + # using their own custom checkpoint_fn. + if self.checkpoint_impl == CheckpointImpl.REENTRANT and kwargs != {}: + # Pack the args and kwargs + flat_args, kwarg_keys = _pack_kwargs(*args, **kwargs) + + # Function that only takes (packed) args, but can unpack them + # into the original args and kwargs for the checkpointed + # function, and runs that function. + def my_function(*inputs): + # unpack back into args and kwargs + unpacked_args, unpacked_kwargs = _unpack_kwargs(inputs, kwarg_keys) + # run original module + return self._checkpoint_wrapped_module( + *unpacked_args, **unpacked_kwargs + ) + + # Pass the function that only takes packed args into reentrant + # checkpoint API. + return self.checkpoint_fn( # type: ignore[misc] + my_function, + *flat_args, + ) + else: + return self.checkpoint_fn( # type: ignore[misc] + self._checkpoint_wrapped_module, *args, **kwargs + ) + + +def offload_wrapper(module: torch.nn.Module) -> torch.nn.Module: + """ + Wrap a module for activation offloading to CPU. + + Offloads intermediate activations to the CPU for modules wrapped with this function. + Wrappers with activation offload can be composed with ones that do recomputation-based + checkpoint to trade off increased compute versus increased CPU + memory usage and additional H2D transfers. + + Usage:: + offloaded_module = offload_wrapper(module) + outputs = checkpointed_module(inputs) + Args: + module (nn.Module): + The module to be wrapped + Returns: + (nn.Module): + Wrapped module + """ + return OffloadWrapper(module) + + +def checkpoint_wrapper( + module: torch.nn.Module, + checkpoint_impl: CheckpointImpl = CheckpointImpl.NO_REENTRANT, + checkpoint_fn=None, + **checkpoint_fn_kwargs, +) -> torch.nn.Module: + """ + Wrap a module for activation checkpointing. + + If the module is wrapped with this function, all subsequent calls to the module will, + automatically perform checkpointing without the user having to explicitly call ``checkpoint`` function. + + Usage:: + checkpointed_module = checkpoint_wrapper(module) + outputs = checkpointed_module(inputs) + Args: + module (nn.Module): + The module to be wrapped + checkpoint_impl (Optional[CheckpointImpl]): + The checkpointing implementation to use. Note that this will only + be passed into the ``torch.utils.checkpoint.checkpoint`` + implementation, and is ignored if a custom ``checkpoint_fn`` is + specified. Note that for implementations using reentrant checkpoint + from ``torch.utils.checkpoint``, keyword arguments will only be + supported if ``checkpoint_impl`` is passed as ``CheckpointImpl.REENTRANT`. + checkpoint_fn (Optional[Callable]): + Functional checkpoint implementation to use. If this is specified, + it will be used over the default ``torch.utils.checkpoint.checkpoint`` + implementation and the `checkpoint_impl` argument will be ignored. + **checkpoint_fn_kwargs: (Dict[str, Any]): Keyword arguments to pass into `checkpoint_fn`. + + Returns: + (nn.Module): + Wrapped module + """ + return CheckpointWrapper( + module, + checkpoint_impl, + checkpoint_fn, + **checkpoint_fn_kwargs, + ) + + +def apply_activation_checkpointing( + model, + checkpoint_wrapper_fn=checkpoint_wrapper, + check_fn=lambda _: True, + auto_wrap_policy: Callable[[nn.Module, bool, int], bool] | None = None, +): + """ + Apply :func:`checkpoint_wrapper` to modules within `model` based on a user-defined configuration. + + For each module within `model`, the `check_fn` is used to decide + whether `module` should be wrapped with :func:`checkpoint_wrapper` or not. + + Note:: + This function modifies `model` in place and replaces appropriate layers with + their checkpoint-wrapped modules. + Note:: + This function will not wrap the overall root module. If this is needed, please directly use + :func:`checkpoint_wrapper` or :func:`offload_wrapper`. + Usage:: + model = nn.Sequential( + nn.Linear(10, 10), nn.Linear(10, 10), nn.Linear(10, 10) + ) + check_fn = lambda l: isinstance(l, nn.Linear) + # checkpoint activations + apply_activation_checkpointing(model, checkpoint_wrapper_fn=checkpoint_wrapper, check_fn=check_fn) + # Or offload activations to CPU + apply_activation_checkpointing(model, checkpoint_wrapper_fn=offload_wrapper, check_fn=check_fn) + Args: + model (nn.Module): + The model whose submodules should be wrapped with activation checkpointing. + checkpoint_wrapper_fn (Optional[Callable[nn.Module]]) + A ``Callable`` which will wrap modules + check_fn (Optional[Callable[nn.Module, nn.Module]]) + A lambda function which will be passed each child submodule of ``model`` and returns + ``True`` or ``False`` depending on whether the submodule should be wrapped. + auto_wrap_policy (Optional[Callable[[nn.Module, bool, int], bool]]): A policy to wrap model's + submodules with AC. Note that if this is specified, it takes precedence over ``check_fn``. + Returns: None (`model` is modified inplace) + """ + # TODO: Importing inside function to avoid circular import issue between FSDP and + # checkpoint_wrapper. This can be resolved once wrap() APIs are decoupled from FSDP code. + from torch.distributed.fsdp._wrap_utils import _construct_wrap_fn, _post_order_apply + from torch.distributed.fsdp.wrap import ( + _Policy, + _recursive_wrap, + lambda_auto_wrap_policy, + ) + + policy = ( + auto_wrap_policy + if auto_wrap_policy is not None + else partial(lambda_auto_wrap_policy, lambda_fn=check_fn) + ) + if not callable(policy): + if not isinstance(policy, _Policy): + raise ValueError( + f"Expected {policy} to be callable or be a pre-defined wrap policy" + ) + target_module_to_kwargs = policy._run_policy( + model, ignored_modules=set(), root_kwargs={} + ) + wrap_fn = _construct_wrap_fn( + model, target_module_to_kwargs, checkpoint_wrapper_fn + ) + _post_order_apply(model, wrap_fn) + return + + _recursive_wrap( + module=model, + auto_wrap_policy=policy, # type: ignore[arg-type] + wrapper_cls=checkpoint_wrapper_fn, + ignored_modules=set(), + ignored_params=set(), + only_wrap_children=True, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..7b57a075ad729d0ae3004dc15585250b04810f43 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__init__.py @@ -0,0 +1,7 @@ +from . import default_hooks as default + + +LOW_PRECISION_HOOKS = [ + default.fp16_compress_hook, + default.bf16_compress_hook, +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a307a5fae15432b6dbf068206b084fedb4c2fff5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/default_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/default_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e4878324aa4efc16140c9cd83f2483cf01fbb3ad Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/__pycache__/default_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/default_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/default_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..db599158518508471338c7fb47744542996f0d8d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_comm_hooks/default_hooks.py @@ -0,0 +1,191 @@ +# mypy: allow-untyped-defs +import functools + +import torch +import torch.distributed as dist + + +class DefaultState: + r""" + Stores state needed to perform the default communication algorithm within a communication hook. + + Args: + process_group (ProcessGroup): The process group to be used. + """ + + __slots__ = [ + "process_group", + "world_size", + "gradient_predivide_factor", + "gradient_postdivide_factor", + ] + + def __init__(self, process_group: dist.ProcessGroup): + if process_group is None: + raise ValueError(f"Expected to pass in an explicit ProcessGroup to {self}.") + self.process_group = process_group + self.world_size = dist.get_world_size(process_group) + # Setting two factors `self.gradient_predivide_factor` + # and `self.gradient_postdivide_factor` to avoid underflow and overflow + self.gradient_predivide_factor = self._get_gradient_predivide_factor( + self.world_size + ) + self.gradient_postdivide_factor = ( + self.world_size / self.gradient_predivide_factor + ) + + @staticmethod + def _get_gradient_predivide_factor(world_size: int) -> float: + factor: int = 1 + while world_size % factor == 0 and world_size / factor > factor: + factor *= 2 + return float(factor) + + +class LowPrecisionState(DefaultState): + r""" + Stores state needed to perform gradient communication in a lower precision within a communication hook. + + Communication hook will cast gradients back to the original + parameter precision specified by ``parameter_type`` (default: torch.float32). + Builds on top of the :class:`DefaultState`. + + Args: + parameter_type (torch.dtype): The precision of model's parameters. + Required for a hook to cast gradients back to a parameter's precision. + """ + + __slots__ = [ + "parameter_type", + ] + + def __init__( + self, + process_group, + parameter_type=torch.float32, + ): + super().__init__(process_group) + self.parameter_type = parameter_type + + +def _decompress(state: LowPrecisionState, grad: torch.Tensor): + """ + Casts gradients back to full parameter precision so that further computation happens in full precision. + """ + orig_grad_data = grad.data + grad.data = grad.data.to(state.parameter_type) + device_type = "" + try: + if grad.device.type == "privateuse1": + device_type = torch._C._get_privateuse1_backend_name() + else: + device_type = grad.device.type + backend = getattr(torch, device_type) + except AttributeError as e: + raise AttributeError( + f"Device {grad.device} does not have a \ + corresponding backend registered as 'torch.device_type'." + ) from e + + # Don't let this memory get reused until after the transfer. + orig_grad_data.record_stream(backend.current_stream()) # type: ignore[arg-type] + + +def allreduce_hook(state: DefaultState, grad: torch.Tensor): + r""" + Implement the FSDP communication hook for ``all_reduce`` algorithm and a necessary pre- and post-division of gradients. + + Args: + state (DefaultState): State information, configures pre- and post-division factors. + grad (torch.Tensor): A gradient for the local batch that needs to be communicated across ranks. + """ + # Average grad by pre-division factor. Together pre- and post-division factors + # lead to an overall averaging by world_size, required for consistency with PyTorch DDP. + # This is a two-step process to avoid potential underflow and overflow. + if state.gradient_predivide_factor > 1: + grad.div_(state.gradient_predivide_factor) + dist.all_reduce(grad, group=state.process_group) + # Average grad by post-division factor. + if state.gradient_postdivide_factor > 1: + grad.div_(state.gradient_postdivide_factor) + + +def reduce_scatter_hook(state: DefaultState, grad: torch.Tensor, output: torch.Tensor): + r""" + Implement the FSDP communication hook for ``reduce_scatter`` algorithm. + + For sharded FSDP strategies and a necessary pre- and post-division of gradients. + + Args: + state (DefaultState): State information, configures pre- and post-division factors. + grad (torch.Tensor): An unsharded gradient for the local batch that needs to be + communicated across ranks. + output (torch.Tensor): Stores a single shard of the gradient after ``reduce_scatter``. + """ + # Average grad by pre-division factor. + if state.gradient_predivide_factor > 1: + grad.div_(state.gradient_predivide_factor) + dist.reduce_scatter_single(output, grad, group=state.process_group) + # Average grad's shard by post-division factor. + if state.gradient_postdivide_factor > 1: + output.div_(state.gradient_postdivide_factor) + + +def _low_precision_hook( + prec: torch.dtype, + state: LowPrecisionState, + grad: torch.Tensor, + output: torch.Tensor | None, +): + if grad.dtype != prec: + grad.data = grad.data.to(prec) + if output is not None: + if output.dtype != prec: + output.data = output.data.to(prec) + reduce_scatter_hook(state, grad, output) + _decompress(state, output) + else: + allreduce_hook(state, grad) + _decompress(state, grad) + + +def fp16_compress_hook( + state: LowPrecisionState, grad: torch.Tensor, output: torch.Tensor | None = None +): + r""" + Implement FSDP communication hook for a simple gradient compression approach. + Casts ``grad`` to half-precision floating-point format (``torch.float16``). + + It also averages gradients by ``world_size`` in two steps: first it pre-divides gradients by a + ``state.gradient_predivide_factor``, and after a communication step (``all_reduce`` or ``reduce_scatter``) + gradients are averaged by a ``state.gradient_postdivide_factor``. + Once post-division is done, compressed gradients are casted back to parameters' precision. + + Args: + state (LowPrecisionState): State information, configures pre- and post-division factors, parameters' precision. + grad (torch.Tensor): A gradient for the local batch that needs to be communicated across ranks in a lower precision. + output (torch.Tensor): Stores a single shard of the gradient after ``reduce_scatter``. + """ + fp16_hook = functools.partial(_low_precision_hook, torch.float16) + return fp16_hook(state, grad, output) + + +def bf16_compress_hook( + state: LowPrecisionState, grad: torch.Tensor, output: torch.Tensor | None = None +): + r""" + Implement FSDP communication hook for a simple gradient compression approach . + Casts ``grad`` to half-precision floating-point format. + + It also averages gradients by ``world_size`` in two steps: first it pre-divides gradients by a + ``state.gradient_predivide_factor``, and after a communication step (``all_reduce`` or ``reduce_scatter``) + gradients are averaged by a ``state.gradient_postdivide_factor``. + Once post-division is done, compressed gradients are casted back to parameters' precision. + + Args: + state (LowPrecisionState): State information, configures pre- and post-division factors, parameters' precision. + grad (torch.Tensor): A gradient for the local batch that needs to be communicated across ranks in a lower precision. + output (torch.Tensor): Stores a single shard of the gradient after ``reduce_scatter``. + """ + bf16_hook = functools.partial(_low_precision_hook, torch.bfloat16) + return bf16_hook(state, grad, output) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..ba62bfb68f42a136dcfa27bcf378d3892cf6751a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__init__.py @@ -0,0 +1 @@ +from .optimizer_overlap import _as_overlapped_optim diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7cf9a08d8362e91146a7b91bbfda2a11b3a41aea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/optimizer_overlap.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/optimizer_overlap.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..23cec2666d02f5f2fbd059c75d0ffb76980134fd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/__pycache__/optimizer_overlap.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/optimizer_overlap.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/optimizer_overlap.py new file mode 100644 index 0000000000000000000000000000000000000000..569a42ffe7643bb6b6403dfb323a4dfd28493e1b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_optimizer_overlap/optimizer_overlap.py @@ -0,0 +1,96 @@ +# mypy: allow-untyped-defs +import inspect +from abc import ABC, abstractmethod + +from torch.distributed.algorithms.ddp_comm_hooks.default_hooks import allreduce_hook +from torch.distributed.algorithms.ddp_comm_hooks.optimizer_overlap_hooks import ( + _hook_then_optimizer, + _OptimizerHookState, +) +from torch.distributed.fsdp import FullyShardedDataParallel +from torch.distributed.optim import as_functional_optim +from torch.nn.parallel import DistributedDataParallel +from torch.optim import Optimizer + + +# Contains the mappings between the regular and overlapped optimizer types. +_registered_overlapped_optims: dict[type, type] = {} + + +def register_overlapped(optim_cls): + def decorator(target_overlapped_optim_cls): + if target_overlapped_optim_cls in _registered_overlapped_optims: + raise ValueError( + f"{target_overlapped_optim_cls} already registered with optim_cls " + f"{_registered_overlapped_optims[optim_cls]} {optim_cls}, trying to" + f"re-register it for {optim_cls} is not supported." + ) + _registered_overlapped_optims[optim_cls] = target_overlapped_optim_cls + return target_overlapped_optim_cls + + return decorator + + +class OverlappedOptimizer(ABC): + def __init__(self, optim_cls: type) -> None: + """ + Initialize the OverlappedOptimizer. + + Overlappedoptimizer is a base class that child classes can implement to + specify how different optimizers will register themselves with DDP. + """ + self.optim_cls = optim_cls + + @abstractmethod + def register_ddp(self, ddp: DistributedDataParallel) -> None: + """Registers the overlapped optimizer with DDP.""" + raise NotImplementedError( + f"{self.__class__.__name__} does not support overlapped DDP." + ) + + @abstractmethod + def register_fsdp(self, fsdp: FullyShardedDataParallel) -> None: + """Registers the overlapped optimizer with FSDP.""" + raise NotImplementedError( + f"{self.__class__.__name__} does not support overlapped FSDP." + ) + + +@register_overlapped(Optimizer) +class _OverlappedStandardOptimizer(OverlappedOptimizer): + """Overlaps a regular ``Optimizer``.""" + + def __init__(self, optim_cls: type, params, *optim_args, **optim_kwargs) -> None: + super().__init__(optim_cls) + f_optim = as_functional_optim(self.optim_cls, *optim_args, **optim_kwargs) + self._opt_hook_state = _OptimizerHookState(f_optim, params) + + def register_ddp(self, ddp_inst: DistributedDataParallel): + # NOTE: using a custom communication hook and fused optimizer is not + # yet supported. + ddp_inst.register_comm_hook( # type: ignore[operator] + None, # wrapped hook state + _hook_then_optimizer(allreduce_hook, self._opt_hook_state), + ) + + # TODO: register_fsdp once FSDP supports communication hook. + def register_fsdp(self, fsdp: FullyShardedDataParallel) -> None: + """Register the overlapped optimizer with FSDP.""" + raise NotImplementedError( + f"{self.__class__.__name__} does not support overlapped FSDP." + ) + + +def _as_overlapped_optim(optim_cls: type, params, *args, **kwargs): + """Return a new ``OverlappedOptimizer`` instance that supports ``optim_cls``.""" + for clz in inspect.getmro(optim_cls): + try: + return _registered_overlapped_optims[clz]( + optim_cls, params, *args, **kwargs + ) + except KeyError: + pass + + # Fallback to standard overlapped optimizer, which will raise errors if user + # is attempting to use an unsupported optimizer. + return _OverlappedStandardOptimizer(optim_cls, params, *args, **kwargs) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..10b8c0430795dc8340b34f0a9fa007b2c45b76be Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/quantization.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/quantization.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..94f4e3b151ece9c8116bca9e6afdd583e14c0555 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/__pycache__/quantization.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/quantization.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/quantization.py new file mode 100644 index 0000000000000000000000000000000000000000..69d88604561355b344b43129108d276e398e0f9f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/_quantization/quantization.py @@ -0,0 +1,151 @@ +# mypy: allow-untyped-defs +import functools +from enum import Enum + +import torch +import torch.distributed as dist + + +TORCH_HALF_MIN = torch.finfo(torch.float16).min +TORCH_HALF_MAX = torch.finfo(torch.float16).max + + +class DQuantType(Enum): + """ + Different quantization methods for auto_quantize API are identified here. + + auto_quantize API currently supports fp16 and bfp16 methods. + """ + + FP16 = ("fp16",) + BFP16 = "bfp16" + + def __str__(self) -> str: + return self.value + + +def _fp32_to_fp16_with_clamp(tensor: torch.Tensor) -> torch.Tensor: + return torch.clamp(tensor, TORCH_HALF_MIN, TORCH_HALF_MAX).half() + + +def _quantize_tensor(tensor, qtype): + if not isinstance(tensor, torch.Tensor): + raise RuntimeError( + f"_quantize_tensor expecting torch.Tensor as input but found {type(tensor)}" + ) + if qtype == DQuantType.FP16: + return _fp32_to_fp16_with_clamp(tensor) + elif qtype == DQuantType.BFP16: + return torch.ops.quantization._FloatToBfloat16Quantized(tensor) + else: + raise RuntimeError(f"Quantization type {qtype} is not supported") + + +def _quantize_tensor_list(tensor_list, qtype): + if not isinstance(tensor_list, list) or not all( + isinstance(p, torch.Tensor) for p in tensor_list + ): + raise RuntimeError( + f"_quantize_tensor_list expecting list of torch.Tensor as input but found {type(tensor_list)}" + ) + quantized_tensor_list = [_quantize_tensor(t, qtype) for t in tensor_list] + return quantized_tensor_list + + +def _dequantize_tensor(tensor, qtype, quant_loss=None): + if not isinstance(tensor, torch.Tensor): + raise RuntimeError( + f"_dequantize_tensor expecting torch.Tensor as input but found {type(tensor)}" + ) + if qtype == DQuantType.FP16: + if tensor.dtype != torch.float16: + raise RuntimeError( + f"tensor dtype is {tensor.dtype} while expected to be FP16." + ) + elif tensor.dtype == torch.float16 and quant_loss is None: + return tensor.float() + else: + # pyrefly: ignore [unsupported-operation] + return tensor.float() / quant_loss + elif qtype == DQuantType.BFP16: + if tensor.dtype != torch.float16: + raise RuntimeError( + f"tensor dtype is {tensor.dtype} while expected to be FP16." + ) + else: + return torch.ops.quantization._Bfloat16QuantizedToFloat(tensor) + else: + raise RuntimeError(f"Quantization type {qtype} is not supported") + + +def _dequantize_tensor_list(tensor_list, qtype, quant_loss=None): + if not isinstance(tensor_list, list) or not all( + isinstance(p, torch.Tensor) for p in tensor_list + ): + raise RuntimeError( + f"_dequantize_tensor_list expecting list of torch.Tensor as input but found {type(tensor_list)}" + ) + dequantized_tensor_list = [_dequantize_tensor(t, qtype) for t in tensor_list] + return dequantized_tensor_list + + +def auto_quantize(func, qtype, quant_loss=None): + """ + Quantize the input tensors, choose the precision types, and pass other necessary arguments and then dequantizes the output. + + Currently it only supports: + . FP16 and BFP16 quantization method supported for gloo and nccl backends + . all_gather, all_to_all collective ops + Note: BFP16 only supports 2D tensors. + Args: + func (Callable): A function representing collective operations. + qtype (QuantType): Quantization method + quant_loss (float, optional): This can be used to improve accuracy in the dequantization. + Returns: + (Callable): the same collective as func but enables automatic quantization/dequantization. + """ + + @functools.wraps(func) + def wrapper(*args, **kwargs): + group = kwargs.get("group") + async_op = kwargs.get("async_op", False) + if async_op is True: + raise RuntimeError("The async_op=True mode is not supported yet.") + if func is dist.all_gather: + tensors = args[0] + input_tensors = _quantize_tensor(args[1], qtype) + out_tensors = _quantize_tensor_list(tensors, qtype) + dist.all_gather(out_tensors, input_tensors, group=group, async_op=async_op) + for i, t in enumerate( + _dequantize_tensor_list(out_tensors, qtype, quant_loss=quant_loss) + ): + tensors[i] = t + + elif func is dist.all_to_all: + tensors = args[0] + input_tensors = _quantize_tensor_list(args[1], qtype) + out_tensors = _quantize_tensor_list(tensors, qtype) + dist.all_to_all(out_tensors, input_tensors, group=group, async_op=async_op) + for i, t in enumerate( + _dequantize_tensor_list(out_tensors, qtype, quant_loss=quant_loss) + ): + tensors[i] = t + + elif func is dist.all_to_all_single: + tensors = args[0] + out_splits = kwargs.get("out_splits") + in_splits = kwargs.get("in_splits") + # Quantizing the input/output tensor + input_tensors = _quantize_tensor(args[1], qtype) + out_tensors = _quantize_tensor(tensors, qtype) + dist.all_to_all_single( + out_tensors, input_tensors, out_splits, in_splits, group=group + ) + for i, t in enumerate( + _dequantize_tensor(out_tensors, qtype, quant_loss=quant_loss) + ): + tensors[i] = t + else: + raise RuntimeError(f"The collective op {func} is not supported yet") + + return wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..256dc1a600628c5e09950eebaa56d92feb03007a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__init__.py @@ -0,0 +1,141 @@ +# mypy: allow-untyped-defs +import sys +from enum import Enum +from functools import partial + + +# To suppress FutureWarning from partial since 3.13 +if sys.version_info >= (3, 11): + # member was introduced in Python 3.11 + from enum import member + + def _enum_member(x): + return member(x) +else: + + def _enum_member(x): + return x + + +import torch.distributed as dist + +from . import ( + debugging_hooks as debugging, + default_hooks as default, + optimizer_overlap_hooks as optimizer_overlap, + powerSGD_hook as powerSGD, + quantization_hooks as quantization, +) + + +__all__ = ["DDPCommHookType", "register_ddp_comm_hook"] + + +def _ddp_comm_hook_wrapper(comm_hook, model, state): + model.register_comm_hook(state, comm_hook) + + +def _powerSGD_comm_hook_wrapper( + comm_hook, + model, + state, + matrix_approximation_rank, + start_powerSGD_iter=1_000, +): + """ + Wrap PowerSGD communication hook. + + To be consistent with the wrappers of other DDP comm hooks, the input state only needs to be a process group, + which will be wrapped up with other state info. + """ + powerSGD_state = powerSGD.PowerSGDState( + process_group=state, + matrix_approximation_rank=matrix_approximation_rank, + start_powerSGD_iter=start_powerSGD_iter, + ) + model.register_comm_hook(powerSGD_state, comm_hook) + + +class DDPCommHookType(Enum): + """ + Enumerate ``ddp_comm_hooks`` and ``ddp_comm_hook_wrapper`` communication hook types. + + DDPCommHookType enumerates the hooks of ``torch.distributed.algorithms.ddp_comm_hooks`` + as names and ``ddp_comm_hook_wrapper`` partials with hook specified. As an example, + you can register allreduce hook by + ``DDPCommHookType.ALLREDUCE.value(model=model, state=process_group)``. + """ + + ALLREDUCE = _enum_member( + partial(_ddp_comm_hook_wrapper, comm_hook=default.allreduce_hook) + ) + FP16_COMPRESS = _enum_member( + partial(_ddp_comm_hook_wrapper, comm_hook=default.fp16_compress_hook) + ) + BF16_COMPRESS = _enum_member( + partial(_ddp_comm_hook_wrapper, comm_hook=default.bf16_compress_hook) + ) + QUANTIZE_PER_TENSOR = _enum_member( + partial( + _ddp_comm_hook_wrapper, comm_hook=quantization.quantization_pertensor_hook + ) + ) + QUANTIZE_PER_CHANNEL = _enum_member( + partial( + _ddp_comm_hook_wrapper, comm_hook=quantization.quantization_perchannel_hook + ) + ) + POWER_SGD = _enum_member( + partial( + _powerSGD_comm_hook_wrapper, + comm_hook=powerSGD.powerSGD_hook, + matrix_approximation_rank=1, + ) + ) + # Rank-2 PowerSGD can give a higher accuracy than the default rank-1 version, + # but it runs slower and consumes more memory. + POWER_SGD_RANK2 = _enum_member( + partial( + _powerSGD_comm_hook_wrapper, + comm_hook=powerSGD.powerSGD_hook, + matrix_approximation_rank=2, + ) + ) + # Batching can lead to a faster training at the cost of accuracy. + BATCHED_POWER_SGD = _enum_member( + partial( + _powerSGD_comm_hook_wrapper, + comm_hook=powerSGD.batched_powerSGD_hook, + matrix_approximation_rank=1, + ) + ) + BATCHED_POWER_SGD_RANK2 = _enum_member( + partial( + _powerSGD_comm_hook_wrapper, + comm_hook=powerSGD.batched_powerSGD_hook, + matrix_approximation_rank=2, + ) + ) + NOOP = _enum_member( + partial( + _ddp_comm_hook_wrapper, + comm_hook=debugging.noop_hook, + ) + ) + + +def register_ddp_comm_hook(comm_hook_type: DDPCommHookType, model, state=None): + """ + Register ``ddp_comm_hooks`` to DDP model. + + Registers the hooks of ``torch.distributed.algorithms.ddp_comm_hooks`` + to the DDP model. User can specify the type of hook as an enum + ``DDPCommHookType`` type using ``comm_hook_type`` input. State input will + be passed to the model. + Uses Python comm hook implementations. + + Example:: + >>> # xdoctest: +SKIP + >>> register_ddp_comm_hook(DDPCommHookType.FP16_COMPRESS, model, state) + """ + comm_hook_type.value(model=model, state=state) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..aad860afdc2493eb3efd5af652d7031138b01e69 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/ddp_zero_hook.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/ddp_zero_hook.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a50340cb2957446f3b8759ccdc3e7758f3f02010 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/ddp_zero_hook.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/debugging_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/debugging_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b6ce3f2ed502d9314464c914290636b5433122f5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/debugging_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/default_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/default_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e07fa235ce552da5368eee5f0b18dec966209eb5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/default_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/mixed_precision_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/mixed_precision_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cf56769d3530bc55969a7f97f04f8bd41e0a1474 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/mixed_precision_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/optimizer_overlap_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/optimizer_overlap_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9f6319c87122b6f79edba2c01096f0aa4dc1daf8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/optimizer_overlap_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/post_localSGD_hook.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/post_localSGD_hook.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f8ba724b9de54854a3b8eac6bf96b59dbc6308b8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/post_localSGD_hook.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/powerSGD_hook.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/powerSGD_hook.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0d12bda0d32ee944cdfca03546978a00e0e5026e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/powerSGD_hook.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/quantization_hooks.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/quantization_hooks.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..666004b6352a530d402783f1f1f095ebe771a455 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/__pycache__/quantization_hooks.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/ddp_zero_hook.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/ddp_zero_hook.py new file mode 100644 index 0000000000000000000000000000000000000000..9b3e84719ebc3c195fa57129bbf5686606a422bf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/ddp_zero_hook.py @@ -0,0 +1,464 @@ +# mypy: allow-untyped-defs +import weakref +from collections.abc import Callable +from typing import Any + +import torch +import torch.distributed as dist +from torch.distributed.optim import ZeroRedundancyOptimizer +from torch.distributed.optim.zero_redundancy_optimizer import _OverlapStatus +from torch.nn.parallel.distributed import DistributedDataParallel + + +__all__ = ["hook_with_zero_step", "hook_with_zero_step_interleaved"] + +# Functional optimizers require passing a list of gradients to their `step()` +# method, and ZeRO requires a functional optimizer to overlap with DDP +# Passing a `None` instead of an actual gradient indicates to the optimizer +# to not update the corresponding parameter +_NO_PARAM_UPDATE: None = None + + +def _perform_local_step( + bucket: dist.GradBucket, + zero: ZeroRedundancyOptimizer, + rank: int, +): + r""" + Perform a local optimizer step using the gradients provided by ``bucket``. + + Arguments: + bucket (dist.GradBucket): the bucket providing the gradients. + zero (ZeroRedundancyOptimizer): the :class:`ZeroRedundancyOptimizer` + instance to perform the :meth:`_local_step`. + rank (int): the calling process's rank. + + .. warning:: + This function assumes that appropriate synchronization has taken place + so that the bucket's gradients can be used. + """ + overlap_info = zero._overlap_info + bucket_index = bucket.index() + if len(zero.optim.param_groups) != 1: + raise AssertionError( + "Overlapping DDP with ZeRO only supports a single parameter group" + ) + + # Construct the `gradients` input for the local optimizer step, which + # expects `None` in a list position to indicate that the corresponding + # parameter should not be updated + num_local_optim_params = len(zero.optim.param_groups[0]["params"]) + gradients: list[torch.Tensor | None] = [ + _NO_PARAM_UPDATE for _ in range(num_local_optim_params) + ] + if bucket_index not in overlap_info.offsets: + raise AssertionError( + f"Bucket index {bucket_index} was not assigned to rank {rank}" + ) + gradients_offset = overlap_info.offsets[bucket_index] + bucket_assignment = zero._bucket_assignments_per_rank[rank][bucket_index] + bucket_offset = bucket_assignment.offset + length = len(bucket_assignment.parameters) + bucket_gradients = bucket.gradients()[bucket_offset : bucket_offset + length] + for i, grad in enumerate(bucket_gradients): + gradients[gradients_offset + i] = grad + + zero._local_step(gradients) + + +def _broadcast_bucket( + bucket_index: int, + zero: ZeroRedundancyOptimizer, +): + r""" + Broadcasts a bucket's parameters. + + Arguments: + bucket_index (int): the index of the bucket corresponding to the + parameters to broadcast. + zero (ZeroRedundancyOptimizer): the calling process's + :class:`ZeroRedundancyOptimizer` instance. + """ + overlap_info = zero._overlap_info + if len(overlap_info.assigned_ranks_per_bucket) <= bucket_index: + raise AssertionError("`assigned_ranks_per_bucket` is not fully constructed") + # Sort to ensure the same ordering across ranks + assigned_ranks = sorted(overlap_info.assigned_ranks_per_bucket[bucket_index]) + if len(assigned_ranks) <= 0: + raise AssertionError( + f"Bucket {bucket_index} should be assigned to at least one rank" + ) + for assigned_rank in assigned_ranks: + bucket_assignments = zero._bucket_assignments_per_rank[assigned_rank] + if bucket_index in bucket_assignments: + send_tensor = bucket_assignments[bucket_index].tensor + if send_tensor is None: + raise AssertionError + overlap_info.broadcast_handles.append( + dist.broadcast( + send_tensor, + src=dist.get_global_rank(zero.process_group, assigned_rank), + group=zero.process_group, + async_op=True, + ) + ) + + +def _save_ddp_bucket_info( + bucket: dist.GradBucket, + zero: ZeroRedundancyOptimizer, +): + r""" + Save :class:`DistributedDataParallel` gradient bucket information for :class:`ZeroRedundancyOptimizer` instance ``zero``. + + In particular, this function is meant to be called upon seeing each + gradient bucket to use when overlapping, meaning it does not save or compute any global + information. + + Arguments: + bucket (dist.GradBucket): the current gradient bucket. + zero (ZeroRedundancyOptimizer): the calling process's + :class:`ZeroRedundancyOptimizer` instance. + """ + overlap_info = zero._overlap_info + bucket_params = bucket.parameters() + if len(bucket_params) <= 0: + raise AssertionError("Empty bucket") + + # Save the parameters in the bucket + overlap_info.params_per_bucket.append(bucket_params) + if overlap_info.shard_buckets: + # Additionally save the bucket size for the assignment heuristic to use + bucket_size = 0 + for param in bucket_params: + bucket_size += param.numel() + if overlap_info.total_size is None: + raise AssertionError + overlap_info.total_size += bucket_size + + +def _hook_with_zero_step_setup( + ddp_ref: weakref.ReferenceType, + zero: ZeroRedundancyOptimizer, + bucket: dist.GradBucket, +): + r""" + Encapsulate the setup logic for :func:`hook_with_zero_step` and :func:`hook_with_zero_step_interleaved`. + + This means the logic to run in the + hook before the backward pass and optimizer step can actually be + overlapped. This is factored out since it is common to both + :func:`hook_with_zero_step` and :func:`hook_with_zero_step_interleaved`. + + Arguments: + ddp_ref (weakref.ReferenceType): weak reference to the process's + :class:`DistributedDataParallel` instance. + zero (ZeroRedundancyOptimizer): the calling process's + :class:`ZeroRedundancyOptimizer` instance. + bucket (dist.GradBucket): the current gradient bucket. + """ + # Proceed as normal until the DDP buckets have been rebuilt + if not ddp_ref()._has_rebuilt_buckets: # type: ignore[union-attr] + if zero._overlap_info.status != _OverlapStatus.UNINITIALIZED: + raise AssertionError + return + + bucket_index = bucket.index() + overlap_info = zero._overlap_info + if overlap_info.status == _OverlapStatus.UNINITIALIZED: + overlap_info.status = _OverlapStatus.DDP_HAS_REBUILT_BUCKETS + + if overlap_info.status == _OverlapStatus.DDP_HAS_REBUILT_BUCKETS: + if bucket_index == 0 and len(overlap_info.params_per_bucket) > 0: + # This corresponds to the first bucket of the backward pass + # immediately after all information has been saved, so we + # can perform the delayed ZeRO initialization + zero._init_zero_for_overlap() + else: + # Once DDP buckets have been rebuilt but ZeRO has not been + # properly initialized yet, save the information needed + _save_ddp_bucket_info(bucket, zero) + + +def hook_with_zero_step( + hook: Callable[[Any, dist.GradBucket], torch.futures.Future], + ddp: DistributedDataParallel, + zero: ZeroRedundancyOptimizer, + shard_buckets: bool = False, +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + r""" + Modify ``hook`` to overlap :class:`ZeroRedundancyOptimizer` optimizer step with :class:`DistributedDataParallel` backward pass. + + This approach overlaps the optimizer computation and communication with the + backward communication. In particular, the backward computation proceeds + contiguously, and the optimizer computation follows, overlapping with + outstanding backward communication (i.e. all-reduces) and possibly other + optimizer communication (i.e. broadcasts). + The optimizer step computation begins after the last gradient bucket computation has finished. + + This approach may be preferred over :meth:`hook_with_zero_step_interleaved` + if communication is relatively slow compared to computation. + + Arguments: + hook (Callable[[Any, dist.GradBucket], torch.futures.Future]): the hook + to modify. + ddp (DistributedDataParallel): the :class:`DistributedDataParallel` + instance to use. + zero (ZeroRedundancyOptimizer): the :class:`ZeroRedundancyOptimizer` + instance to use. + shard_buckets (bool): if ``True``, then the assignment of each + :class:`DistributedDataParallel` bucket is partitioned across + possibly multiple :class:`ZeroRedundancyOptimizer` instances (i.e. + across possibly multiple ranks) to approximate uniformity; if + ``False``, then each bucket is wholly assigned to a single + :class:`ZeroRedundancyOptimizer` instance (i.e. to a single rank). + + Returns: + The modified hook. + + Raises: + ValueError: if ``zero`` was constructed with ``overlap_with_ddp=False``. + RuntimeError: if using any backend other than NCCL/HCCL since currently + Gloo may hang. + + .. warning:: + Given the way that overlapping :class:`DistributedDataParallel` with + :class:`ZeroRedundancyOptimizer` is currently implemented, the first + two or three training iterations do not perform parameter updates in + the optimizer step, depending on if ``static_graph=False`` or + ``static_graph=True``, respectively. This is because it needs + information about the gradient bucketing strategy used by + :class:`DistributedDataParallel`, which is not finalized until the + second forward pass if ``static_graph=False`` or until the third + forward pass if ``static_graph=True``. + """ + if not zero._overlap_with_ddp: + raise ValueError( + "ZeroRedundancyOptimizer must be constructed with " + "`overlap_with_ddp=True` to use this hook properly" + ) + ddp_ref = weakref.ref(ddp) + + # NOTE: Gloo may hang with this overlapping approach; see https://github.com/pytorch/pytorch/issues/62300 + pg = dist.get_backend(ddp_ref().process_group) # type: ignore[union-attr] + if pg == dist.Backend.GLOO: + raise RuntimeError( + "Gloo backend using Overlapping DDP with ZeRO may meet hangs" + ) + + if shard_buckets: + zero._overlap_info.shard_buckets = True + zero._overlap_info.total_size = 0 + + def hook_with_zero_fn( + state: Any, + bucket: dist.GradBucket, + ) -> torch.futures.Future[torch.Tensor]: + r""" + Return :class:`Future` that runs the optimizer step if this corresponds to the last gradient bucket. + + Perform equivalent of :class:`ZeroRedundancyOptimizer` :meth:`step` if ``bucket`` is last gradient bucket. + The function gives a gradient bucket tensor and + performs additional computation on the iteration that + the :class:`DistributedDataParallel` buckets are rebuilt to collect + information used to implement the modified hook. + + Arguments: + state (Any): any state for the hook. + bucket (dist.GradBucket): the :class:`DistributedDataParallel` + gradient bucket. + """ + fut = hook(state, bucket) + _hook_with_zero_step_setup(ddp_ref, zero, bucket) + if zero._overlap_info.status != _OverlapStatus.INITIALIZED: + return fut + + overlap_info = zero._overlap_info + bucket_index = bucket.index() + rank = zero.global_rank + + if overlap_info.status != _OverlapStatus.INITIALIZED: + raise AssertionError + if len(overlap_info.assigned_ranks_per_bucket) <= bucket_index: + raise AssertionError("`assigned_ranks_per_bucket` is not fully constructed") + assigned_to_bucket = ( + rank in overlap_info.assigned_ranks_per_bucket[bucket_index] + ) + + # Save the bucket reference and all-reduce future for the final bucket + if assigned_to_bucket: + overlap_info.bucket_index_to_bucket[bucket_index] = bucket + overlap_info.bucket_index_to_future[bucket_index] = fut + + # Check that buckets are indexed incrementally starting from 0 in the + # order of their autograd hooks firing + if len(overlap_info.bucket_indices_seen) > 0: + if overlap_info.bucket_indices_seen[-1] != bucket_index - 1: + raise AssertionError("Bucket indices are not in incremental order") + else: + if bucket_index != 0: + raise AssertionError("Bucket indices do not start from 0") + overlap_info.bucket_indices_seen.append(bucket_index) + + # Directly return the future without any optimizer computation if this + # is not the last bucket + num_buckets = len(overlap_info.params_per_bucket) + is_last_bucket = bucket_index == num_buckets - 1 + if not is_last_bucket: + return fut + + # Perform partial optimizer step on all buckets after the final + # bucket has been computed + # NOTE: This should not be chained as a callback to the last bucket's + # all-reduce future since that would add synchronization that delays + # all optimizer computation to wait for that last all-reduce + for bucket_index in range(num_buckets): + assigned_ranks = overlap_info.assigned_ranks_per_bucket[bucket_index] + if rank in assigned_ranks: + # Wait on the bucket's all-reduce future to ensure correct + # gradients + if bucket_index not in overlap_info.bucket_index_to_future: + raise AssertionError( + f"All-reduce future for bucket {bucket_index} not saved " + f"on rank {rank}" + ) + allreduce_future = overlap_info.bucket_index_to_future[bucket_index] + allreduce_future.wait() + + # Perform the partial optimizer step + curr_bucket = overlap_info.bucket_index_to_bucket[bucket_index] + _perform_local_step(curr_bucket, zero, rank) + + _broadcast_bucket(bucket_index, zero) + + # Ensure that all parameter updates are finished before the + # next forward pass + overlap_info.wait_for_broadcasts() + overlap_info.clear_per_iter_info() + + return fut + + return hook_with_zero_fn + + +def hook_with_zero_step_interleaved( + hook: Callable[[Any, dist.GradBucket], torch.futures.Future], + ddp: DistributedDataParallel, + zero: ZeroRedundancyOptimizer, + shard_buckets: bool = False, +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + r""" + Modify ``hook`` to overlap :class:`ZeroRedundancyOptimizer` optimizer step with :class:`DistributedDataParallel` backward pass + + This approach overlaps the optimizer computation and communication with the + backward computation and communication. In particular, once a bucket's + gradients have been computed, the optimizer computation using those + gradients is launched (though the actual computation must wait for the + bucket's all-reduce to complete). This yields an interleaving of all- + reduces and broadcasts in the communication stream. + + This approach may be preferred over :meth:`hook_with_zero_step` if + communication is relatively fast compared to computation. + + Arguments: + hook (Any * dist.GradBucket -> torch.futures.Future): the hook to + modify. + ddp (DistributedDataParallel): the :class:`DistributedDataParallel` + instance to use. + zero (ZeroRedundancyOptimizer): the :class:`ZeroRedundancyOptimizer` + instance to use. + shard_buckets (bool): if ``True``, then the assignment of each + :class:`DistributedDataParallel` bucket is partitioned across + possibly multiple :class:`ZeroRedundancyOptimizer` instances (i.e. + across possibly multiple ranks) to approximate uniformity; if + ``False``, then each bucket is wholly assigned to a single + :class:`ZeroRedundancyOptimizer` instance (i.e. to a single rank). + + Returns: + The modified hook. + + Raises: + ValueError: if ``zero`` was constructed with ``overlap_with_ddp=False``. + RuntimeError: if using any backend other than NCCL since currently + Gloo may hang. + + .. warning:: + Given the way that overlapping :class:`DistributedDataParallel` with + :class:`ZeroRedundancyOptimizer` is currently implemented, the first + two or three training iterations do not perform parameter updates in + the optimizer step, depending on if ``static_graph=False`` or + ``static_graph=True``, respectively. This is because it needs + information about the gradient bucketing strategy used by + :class:`DistributedDataParallel`, which is not finalized until the + second forward pass if ``static_graph=False`` or until the third + forward pass if ``static_graph=True``. + """ + if not zero._overlap_with_ddp: + raise ValueError( + "ZeroRedundancyOptimizer must be constructed with " + "`overlap_with_ddp=True` to use this hook properly" + ) + ddp_ref = weakref.ref(ddp) + + # NOTE: Gloo may hang with this overlapping approach; see https://github.com/pytorch/pytorch/issues/62300 + pg = dist.get_backend(ddp_ref().process_group) # type: ignore[union-attr] + if pg == dist.Backend.GLOO: + raise RuntimeError( + "Gloo backend using Overlapping DDP with ZeRO may meet hangs" + ) + + if shard_buckets: + zero._overlap_info.shard_buckets = True + zero._overlap_info.total_size = 0 + + def hook_with_zero_interleaved_fn( + state, + bucket: dist.GradBucket, + ) -> torch.futures.Future[torch.Tensor]: + r""" + Return :class:`Future` that gives gradient bucket tensor and performs partial :class:`ZeroRedundancyOptimizer` :meth:`step`. + + This function uses the gradients in the given bucket to perform a partial + :class:`ZeroRedundancyOptimizer` :meth:`step` + + Arguments: + state: any state for the hook. + bucket (dist.GradBucket): the :class:`DistributedDataParallel` + gradient bucket. + """ + fut = hook(state, bucket) + _hook_with_zero_step_setup(ddp_ref, zero, bucket) + if zero._overlap_info.status != _OverlapStatus.INITIALIZED: + return fut + + def zero_step(fut: torch.futures.Future) -> torch.Tensor: + r""" + Perform partial :class:`ZeroRedundancyOptimizer` :meth:`step` using gradients in the :class:`DistributedDataParallel`. + + Returns: + A :class:`torch.Tensor` representing the contents of the + gradient bucket. + """ + overlap_info = zero._overlap_info + bucket_index = bucket.index() + rank = zero.global_rank + + assigned_ranks = overlap_info.assigned_ranks_per_bucket[bucket_index] + overlap_info.bucket_indices_seen.append(bucket_index) + if rank in assigned_ranks: + _perform_local_step(bucket, zero, rank) + + _broadcast_bucket(bucket_index, zero) + + num_buckets = len(overlap_info.params_per_bucket) + if len(overlap_info.bucket_indices_seen) == num_buckets: + # Ensure that all parameter updates are finished before the + # next forward pass + overlap_info.wait_for_broadcasts() + overlap_info.clear_per_iter_info() + + return bucket.buffer() + + return fut.then(zero_step) + + return hook_with_zero_interleaved_fn diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/debugging_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/debugging_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..e48242cac67315589dbcea7e52a445ce4570a52e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/debugging_hooks.py @@ -0,0 +1,29 @@ +from typing import Any + +import torch +from torch.distributed import GradBucket + + +__all__ = ["noop_hook"] + + +def noop_hook(_: Any, bucket: GradBucket) -> torch.futures.Future[torch.Tensor]: + """ + Return a future that wraps the input, so it is a no-op that does not incur any communication overheads. + + This hook should **only** be used for headroom analysis of allreduce optimization, + instead of the normal gradient synchronization. + For example, if only less than 10% speedup of training time can be observed after this hook is registered, + it usually implies that allreduce is not a performance bottleneck for this case. + Such instrumentation can be particularly useful + if GPU traces cannot be easily retrieved or the trace analysis is complicated by + some factors such as the overlap between allreduce and computation or the desynchronization across ranks. + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(None, noop_hook) + """ + fut: torch.futures.Future[torch.Tensor] = torch.futures.Future() + fut.set_result(bucket.buffer()) + + return fut diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/default_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/default_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..20a0de7ef318c10f3b5bbdaf98483d9fd19b2691 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/default_hooks.py @@ -0,0 +1,211 @@ +# mypy: allow-untyped-defs +from collections.abc import Callable +from typing import Any, cast + +import torch +import torch.distributed as dist + + +__all__ = [ + "allreduce_hook", + "fp16_compress_hook", + "bf16_compress_hook", + "fp16_compress_wrapper", + "bf16_compress_wrapper", +] + + +def _allreduce_fut( + process_group: dist.ProcessGroup, tensor: torch.Tensor +) -> torch.futures.Future[torch.Tensor]: + """Average the input gradient tensor by allreduce and returns a future.""" + group_to_use = process_group if process_group is not None else dist.group.WORLD + + # Apply the division first to avoid overflow, especially for FP16. + # pyrefly: ignore [missing-attribute] + tensor.div_(group_to_use.size()) + + return ( + dist.all_reduce(tensor, group=group_to_use, async_op=True) + .get_future() + .then(lambda fut: fut.value()[0]) + ) + + +def allreduce_hook( + process_group: dist.ProcessGroup, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + """ + Call ``allreduce`` using ``GradBucket`` tensors. + + Once gradient tensors are aggregated across all workers, its ``then`` + callback takes the mean and returns the result. + + If user registers this DDP communication hook, + DDP results is expected to be same as the case where no hook was registered. + Hence, this won't change behavior of DDP and user can use this as a reference + or modify this hook to log useful information or any other purposes while + unaffecting DDP behavior. + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(process_group, allreduce_hook) + """ + return _allreduce_fut(process_group, bucket.buffer()) + + +def _compress_hook( + dtype: torch.dtype, + process_group: dist.ProcessGroup, + bucket: dist.GradBucket, +) -> torch.futures.Future[torch.Tensor]: + group_to_use = process_group if process_group is not None else dist.group.WORLD + # pyrefly: ignore [missing-attribute] + world_size = group_to_use.size() + + buffer = ( + cast(tuple[torch.Tensor, ...], bucket)[0] + if isinstance(bucket, tuple) + else bucket.buffer() + ) + compressed_tensor = buffer.to(dtype).div_(world_size) + + def decompress(fut): + decompressed_tensor = buffer + # Decompress in place to reduce the peak memory. + # See: https://github.com/pytorch/pytorch/issues/45968 + value = fut if isinstance(fut, torch.Tensor) else fut.value()[0] + decompressed_tensor.copy_(value) + return decompressed_tensor + + if torch.compiler.is_compiling(): + grad = dist._functional_collectives.all_reduce( + compressed_tensor, + "sum", + # pyrefly: ignore [bad-argument-type] + group_to_use, + ) + return decompress(grad) + else: + fut = dist.all_reduce( + compressed_tensor, group=group_to_use, async_op=True + ).get_future() + return fut.then(decompress) + + +def fp16_compress_hook( + process_group: dist.ProcessGroup, + bucket: dist.GradBucket, +) -> torch.futures.Future[torch.Tensor]: + """ + Compress by casting ``GradBucket`` to ``torch.float16`` divided by process group size. + + This DDP communication hook implements a simple gradient compression + approach that casts ``GradBucket`` tensor to half-precision floating-point format (``torch.float16``) + and then divides it by the process group size. + It allreduces those ``float16`` gradient tensors. Once compressed gradient + tensors are allreduced, the chained callback ``decompress`` casts it back to the input data type (such as ``float32``). + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(process_group, fp16_compress_hook) + """ + return _compress_hook(torch.float16, process_group, bucket) + + +def bf16_compress_hook( + process_group: dist.ProcessGroup, + bucket: dist.GradBucket, +) -> torch.futures.Future[torch.Tensor]: + """ + Warning: This API is experimental, and it requires NCCL version later than 2.9.6. + + This DDP communication hook implements a simple gradient compression + approach that casts ``GradBucket`` tensor to half-precision + `Brain floating point format `_ (``torch.bfloat16``) + and then divides it by the process group size. + It allreduces those ``bfloat16`` gradient tensors. Once compressed gradient + tensors are allreduced, the chained callback ``decompress`` casts it back to the input data type (such as ``float32``). + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(process_group, bf16_compress_hook) + """ + return _compress_hook(torch.bfloat16, process_group, bucket) + + +def fp16_compress_wrapper( + hook: Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]], +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + """ + Cast input tensor to ``torch.float16``, cast result of hook back to input dtype. + + This wrapper casts the input gradient tensor of a given DDP communication hook to half-precision + floating point format (``torch.float16``), and casts the resulting tensor of the given hook back to + the input data type, such as ``float32``. + Therefore, ``fp16_compress_hook`` is equivalent to ``fp16_compress_wrapper(allreduce_hook)``. + + Example:: + >>> # xdoctest: +SKIP + >>> state = PowerSGDState(process_group=process_group, matrix_approximation_rank=1, start_powerSGD_iter=10) + >>> ddp_model.register_comm_hook(state, fp16_compress_wrapper(powerSGD_hook)) + """ + + def fp16_compress_wrapper_hook( + hook_state, bucket: dist.GradBucket + ) -> torch.futures.Future[torch.Tensor]: + # Cast bucket tensor to FP16. + bucket.set_buffer(bucket.buffer().to(torch.float16)) + + fut = hook(hook_state, bucket) + + def decompress(fut): + decompressed_tensor = bucket.buffer() + # Decompress in place to reduce the peak memory. + # See: https://github.com/pytorch/pytorch/issues/45968 + decompressed_tensor.copy_(fut.value()) + return decompressed_tensor + + # Decompress after hook has run. + return fut.then(decompress) + + return fp16_compress_wrapper_hook + + +def bf16_compress_wrapper( + hook: Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]], +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + """ + Warning: This API is experimental, and it requires NCCL version later than 2.9.6. + + This wrapper casts the input gradient tensor of a given DDP communication hook to half-precision + `Brain floating point format `_ (``torch.bfloat16``), + and casts the resulting tensor of the given hook back to the input data type, such as ``float32``. + + Therefore, ``bf16_compress_hook`` is equivalent to ``bf16_compress_wrapper(allreduce_hook)``. + + Example:: + >>> # xdoctest: +SKIP + >>> state = PowerSGDState(process_group=process_group, matrix_approximation_rank=1, start_powerSGD_iter=10) + >>> ddp_model.register_comm_hook(state, bf16_compress_wrapper(powerSGD_hook)) + """ + + def bf16_compress_wrapper_hook( + hook_state, bucket: dist.GradBucket + ) -> torch.futures.Future[torch.Tensor]: + # Cast bucket tensor to BF16. + bucket.set_buffer(bucket.buffer().to(torch.bfloat16)) + + fut = hook(hook_state, bucket) + + def decompress(fut): + decompressed_tensor = bucket.buffer() + # Decompress in place to reduce the peak memory. + # See: https://github.com/pytorch/pytorch/issues/45968 + decompressed_tensor.copy_(fut.value()) + return decompressed_tensor + + # Decompress after hook has run. + return fut.then(decompress) + + return bf16_compress_wrapper_hook diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/mixed_precision_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/mixed_precision_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..f1968042e5e21aa1b6714f78356b43896cccdf60 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/mixed_precision_hooks.py @@ -0,0 +1,86 @@ +from dataclasses import dataclass +from typing import Any, no_type_check + +import torch +import torch.distributed as dist +from torch.autograd import Variable +from torch.distributed.utils import _free_storage + + +@dataclass +class _AllreduceUpcastHookState: + """ + State to manage DDP mixed precision in backward / gradient communication. + + This contains a weakref to the DDP module for access to reducer and process + group, and a stream to run parameter and gradient upcasts. + """ + + ddp_weakref: Any + upcast_stream: torch.Stream + wait_for_stream_enqueued: bool = False + + +@no_type_check +def _reducer_allreduce_and_upcast_hook( + hook_state: _AllreduceUpcastHookState, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + """ + Perform allreduce in precision ``reduce_dtype``, upcast to prepare for optimizer. + + Performs allreduce in the reduced precision given by DDP's mixed precision + reduce_dtype, and upcasts parameters and gradients to fp32 in preparation + to run the optimizer. + """ + ddp_weakref = hook_state.ddp_weakref + reducer, process_group = ddp_weakref().reducer, ddp_weakref().process_group + # Cast bucket if different than param_dtype. + if ( + ddp_weakref().mixed_precision.param_dtype + != ddp_weakref().mixed_precision.reduce_dtype + ): + # Cast bucket tensor to reduce_dtype + bucket.set_buffer( + bucket.buffer().to(ddp_weakref().mixed_precision.reduce_dtype) + ) + fut = reducer._run_allreduce_hook(bucket) + ret_fut = torch.futures.Future() + stream = hook_state.upcast_stream + with stream: + fut.wait() + bucket.buffer().div_(process_group.size()) + ret_fut.set_result(bucket.buffer()) + + # Upcast parameters and gradients so optimizer step can run in fp32. + for p in bucket.parameters(): + p.data = p._fp_param + # free storage for mp param as it will be allocated again in next + # forward pass. + _free_storage(p._mp_param) + p.grad.data = p.grad.to(p.data.dtype) + + # enqueue a callback to wait for this stream at end of backward + def wait_for_stream_cb(): + torch.accelerator.current_stream().wait_stream(stream) + # Remove post-backward hooks since they are re-installed in next + # iteration, similar to FSDP. + # Parameters that don't require grad still needed to be casted since + # they may participate in computation. However, they would not be recast + # by hook above as they don't have a grad hook installed, so cast them + # back here. + for _, p in ddp_weakref().module.named_parameters(): + if hasattr(p, "_ddp_mp_hook_state"): + p._ddp_mp_hook_state[1].remove() + delattr(p, "_ddp_mp_hook_state") + if not p.requires_grad and not hasattr(p, "_ddp_ignored"): + p.data = p._fp_param + + # reset for next backward pass + hook_state.wait_for_stream_enqueued = False + + if not hook_state.wait_for_stream_enqueued: + Variable._execution_engine.queue_callback(wait_for_stream_cb) + # mark that the callback is enqueued + hook_state.wait_for_stream_enqueued = True + + return ret_fut diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/optimizer_overlap_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/optimizer_overlap_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..2507bfadb023c766ad5c4e2317bf8a01b7d8cd9a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/optimizer_overlap_hooks.py @@ -0,0 +1,163 @@ +# mypy: allow-untyped-defs +from collections.abc import Callable +from dataclasses import dataclass +from functools import partial +from typing import Any, no_type_check + +import torch +import torch.distributed as dist +from torch.autograd import Variable + + +__all__: list[str] = [] + +_FUNCTIONAL_OPTIM_STEP_METHOD_NAME = "step_param" + + +class _OptimizerHookState: + """ + Holds state for running optimizer in-line after DDP communication hook. + + Currently contains only optimizer class which must have a method `step_param`. + """ + + __slots__ = ["functional_optimizer", "params_to_optimize"] + + def __init__(self, functional_optim, params=None): + self.functional_optimizer = functional_optim + self._check_valid_functional_optim() + self._set_params_to_optimize(params) + + def _set_params_to_optimize(self, params): + if params is not None: + self.params_to_optimize = set(params) + + def _check_valid_functional_optim(self): + if not hasattr(self.functional_optimizer, _FUNCTIONAL_OPTIM_STEP_METHOD_NAME): + raise ValueError( + f"Class {type(self.functional_optimizer)} must implement method " + f"{_FUNCTIONAL_OPTIM_STEP_METHOD_NAME}." + ) + + +@dataclass +class _OptimInBackwardHookState: + optim_stream: torch.Stream + wait_for_optim_stream_enqueued: bool + + +@no_type_check +def _apply_optim_in_backward_hook( + gradient_is_bucket_view: bool, +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + r""" + Register hook to apply the optimizer in backward. + + If torch.distributed.optim._apply_optimizer_in_backward is used to overlap + optimizer with backward pass, DDP will run the below hook to run optimizer + step for parameters after gradient communication has taken place. + """ + optim_in_bwd_state = _OptimInBackwardHookState( + optim_stream=torch.Stream(), + wait_for_optim_stream_enqueued=False, + ) + + def apply_optim_in_backward_hook( + hook_state: Any, + bucket: dist.GradBucket, + optim_stream_state, + ) -> torch.futures.Future[torch.Tensor]: + # Run original hook + ddp_weakref = hook_state + ddp_inst = ddp_weakref() + reducer, process_group = ddp_inst.reducer, ddp_inst.process_group + fut = reducer._run_allreduce_hook(bucket) + optimizer_stream = optim_stream_state.optim_stream + with optimizer_stream: + fut.wait() + # Apply gradient division since C++ side only allreduces and does + # not average. TODO: (rohan-varma) the div factor may be different + # when running with join hook + bucket.buffer().div_(process_group.size()) + model_params = bucket.parameters() + grads = bucket.gradients() + # TODO (rohan-varma): upcast as needed for DDP mixed precision, + # once optimizer in backward + DDP mixed precision is supported. + for p, g in zip(model_params, grads): + if hasattr(p, "_in_backward_optimizers"): + # Note: need to set grad to the bucket's grad, because + # running allreduce results in the bucket's grad being + # reduced, but not grad field. + if not gradient_is_bucket_view: + p.grad = g + for optim in p._in_backward_optimizers: + optim.step() + + # Need to return a Future[Tensor] to obey comm hook API contract. + ret_fut = torch.futures.Future() + ret_fut.set_result(bucket.buffer()) + + # enqueue a callback to wait for this optimizer stream at the end of + # backward and set all DDP managed grads to None. + def wait_for_optim_stream_callback(): + torch.accelerator.current_stream().wait_stream( + optim_stream_state.optim_stream + ) + # Set DDP managed grads to None + for param in ddp_inst._get_data_parallel_params(ddp_inst.module): + if hasattr(param, "_in_backward_optimizers"): + param.grad = None + + # reset for the next backward pass + optim_stream_state.wait_for_optim_stream_enqueued = False + + if not optim_stream_state.wait_for_optim_stream_enqueued: + Variable._execution_engine.queue_callback(wait_for_optim_stream_callback) + # mark that the callback is enqueued + optim_stream_state.wait_for_optim_stream_enqueued = True + + return ret_fut + + comm_hook = partial( + apply_optim_in_backward_hook, optim_stream_state=optim_in_bwd_state + ) + # These are needed for DDP's logging of comm hooks + comm_hook.__name__ = apply_optim_in_backward_hook.__name__ + comm_hook.__qualname__ = apply_optim_in_backward_hook.__qualname__ + + return comm_hook + + +def _hook_then_optimizer( + hook: Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]], + optimizer_state: _OptimizerHookState, +) -> Callable[[Any, dist.GradBucket], torch.futures.Future[torch.Tensor]]: + r"""Run optimizer in a functional fashion after DDP communication hook.""" + has_set_params = ( + hasattr(optimizer_state, "params_to_optimize") + and optimizer_state.params_to_optimize is not None + ) + + def hook_then_optimizer_wrapper( + hook_state, bucket: dist.GradBucket + ) -> torch.futures.Future[torch.Tensor]: + # Run original hook + fut = hook(hook_state, bucket) + + def optimizer_step(fut): + gradient_tensors = bucket.gradients() + model_params = bucket.parameters() + for grad_tensor, model_param in zip(gradient_tensors, model_params): + if ( + not has_set_params + or model_param in optimizer_state.params_to_optimize + ): + optimizer_state.functional_optimizer.step_param( + model_param, + grad_tensor, + ) + return bucket.buffer() + + return fut.then(optimizer_step) + + return hook_then_optimizer_wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/post_localSGD_hook.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/post_localSGD_hook.py new file mode 100644 index 0000000000000000000000000000000000000000..ff513f62183c516b96c62ca89eee51d2b1793e85 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/post_localSGD_hook.py @@ -0,0 +1,124 @@ +# mypy: allow-untyped-defs +import logging + +import torch +import torch.distributed as dist + +from . import default_hooks as default + + +logger = logging.getLogger(__name__) + + +class PostLocalSGDState: + r""" + Store state for all-reducing gradients globally until given step, then locally after. + + Stores the state for all-reducing gradients globally using ``process_group`` until step ``start_localSGD_iter``, + and all-reducing gradients locally using ``subgroup`` afterwards. + + If ``process_group`` is ``None``, the global process group will be used. + If ``subgroup`` is ``None``, the intra-node process group on each machine will be used. + + Additionally, ``post_local_gradient_allreduce`` may be worth tuning, + because both true and false may give a faster convergence. + """ + + __slots__ = [ + "process_group", + "subgroup", + "start_localSGD_iter", + "post_local_gradient_allreduce", + "iter", + ] + + def __init__( + self, + process_group, + subgroup, + start_localSGD_iter, + post_local_gradient_allreduce=True, + ): + """Initialize state object with given parameters and log when localSGD start.""" + logger.info( + "Local SGD will be started after %s iterations", start_localSGD_iter + ) + + # The group used for all-reducing gradients globally. + self.process_group = process_group + # The group used for all-reducing gradients locally. + self.subgroup = subgroup + self.start_localSGD_iter = start_localSGD_iter + # Allreduce gradients locally since iteration `start_localSGD_iter`. + # This may help with the convergence efficiency at the cost of relatively cheap intra-subgroup communication. + self.post_local_gradient_allreduce = post_local_gradient_allreduce + # Iteration/step in the training loop. + self.iter = 0 + + def maybe_increase_iter(self, bucket): + """Track iterations and trigger log message at start of local SGD.""" + # Since bucket 0 is the last bucket to allreduce in an iteration. + # Only increase `iter` when bucket 0 is processed. + if bucket.is_last(): + self.iter += 1 + + if self.iter == self.start_localSGD_iter: + logger.info("Start to apply local SGD after %s iterations.", self.iter) + + +def post_localSGD_hook( + state: PostLocalSGDState, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + """ + Run post-localSGD algorithm. + + This DDP communication hook is used for running post-localSGD algorithm, + by combining with a model averaging component (e.g., + :class:`~torch.distributed.algorithms.model_averaging.averagers.PeriodicModelAverager`) + that runs after the optimizer step. + + Args: + state (PostLocalSGDState): State information to run post-localSGD. + Users mainly need to tune ``start_localSGD_iter`` to determine when to start local SGD. + bucket (dist.GradBucket): Bucket that stores a 1D flattened gradient tensor that batches multiple per-variable tensors. + Note that since DDP comm hook only supports single process single device mode, + only exactly one tensor is stored in this bucket. + + Returns: + Future handler of the communication, which updates the gradients in place. + + Example:: + >>> # xdoctest: +SKIP + >>> state = PostLocalSGDState(process_group=process_group, subgroup=subgroup, + start_localSGD_iter=10) + >>> ddp_model.register_comm_hook(state, post_localSGD_hook) + >>> # Also need to establish a model averaging module and run model averaging after ``optimizer.step()``. + >>> # Please refer to the examples in ``torch.distributed.algorithms.model_averaging.averagers`` module. + """ + global_group_to_use = ( + state.process_group if state.process_group is not None else dist.group.WORLD + ) + + # The input tensor is a flattened 1D tensor. + input_tensor = bucket.buffer() + + # Run allreduce using `global_group_to_use` in the first `start_localSGD_iter` iterations. + if state.iter < state.start_localSGD_iter: + state.maybe_increase_iter(bucket) + return default._allreduce_fut(global_group_to_use, input_tensor) # type: ignore[arg-type] + + # If `post_local_gradient_allreduce` is not set, + # then no gradient synchronization after the first `start_localSGD_iter` iterations. + if not state.post_local_gradient_allreduce: + fut: torch.futures.Future[torch.Tensor] = torch.futures.Future() + fut.set_result(input_tensor) + return fut + + # Run allreduce using `subgroup` after the first `start_localSGD_iter` iterations. + # Note that by default, a separate subgroup for each node is created which + # causes an intra-node allreduce to be done at each training step. + # From this moment, model averaging should run after the optimizer step, + # to globally allreduce all the parameters. + if state.subgroup is None: + state.subgroup, _ = dist.new_subgroups() + return default._allreduce_fut(state.subgroup, input_tensor) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/powerSGD_hook.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/powerSGD_hook.py new file mode 100644 index 0000000000000000000000000000000000000000..544cbd4c36cf387008f3316fe1bfa8d411e893b7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/powerSGD_hook.py @@ -0,0 +1,865 @@ +# mypy: allow-untyped-defs +import logging +import math +from collections import defaultdict + +import torch +import torch.distributed as dist +from torch.distributed import distributed_c10d +from torch.utils._typing_utils import not_none + +from . import default_hooks as default + + +__all__ = ["PowerSGDState", "powerSGD_hook", "batched_powerSGD_hook"] + +logger = logging.getLogger(__name__) + + +def _orthogonalize(matrices, epsilon=0): + """ + Decide between Gram-Schmidt or QR factorization to orthogonalize a batch of matrices. + + QR factorization doesn't work with half-precision, but it is usually faster with a rank > 2. + """ + if not (len(matrices.shape) == 3 and matrices.shape[2] <= matrices.shape[1]): + raise AssertionError + + num_matrices = matrices.shape[0] + rank = matrices.shape[2] + dtype = matrices.dtype + if rank <= 2 or dtype in [torch.float16, torch.bfloat16]: + _orthogonalize_gram_schmidt(matrices, epsilon=epsilon) + else: + torch.linalg.qr( + matrices, + out=( + matrices, + torch.empty( + num_matrices, rank, rank, device=matrices.device, dtype=dtype + ), + ), + ) + + +def _orthogonalize_gram_schmidt(matrices, epsilon=0): + """ + Apply Gram-Schmidt procedure to orthogonalize a batch of matrices. + + If epsilon is 0, this is equivalent to `torch.linalg.qr(matrices, out=(matrices, _))`, + """ + num_cols = matrices.shape[2] + for i in range(num_cols): + # Normalize the i'th column. + col = matrices[:, :, i : i + 1] + # If no epsilon is added here, division by zero may be caused by vanishing gradients. + # This epsilon is not needed if the input batch of matrices covers the gradients of at least one entire layer + # in the neural network. + if epsilon == 0: + # Note that col ** 2 can underflow/overflow if we use FP16. + # May need to consider multiplying a scaling factor and dividing it later, or using bfloat16 instead. + try: + col /= torch.linalg.vector_norm(col, dim=1, keepdim=True) + except ZeroDivisionError: + logger.error( + "The matrices to be orthogonalized has at least a column of all 0s. Please set a small value such as 1e-8 " + "as `orthogonalization_epsilon` in PowerSGD state." + ) + # Recover the values from NaNs to 0s. + col.fill_(0.0) + else: + col /= torch.linalg.vector_norm(col, dim=1, keepdim=True) + epsilon + # Project it on the rest and remove it. + if i + 1 < num_cols: + rest = matrices[:, :, i + 1 :] + rest -= torch.sum(col * rest, dim=1, keepdim=True) * col + + +def _should_compress( + num_rows, num_cols, matrix_approximation_rank, min_compression_rate +): + """ + Recommend if tensor given is worth compressing. + + Returns a recommendation as to whether the 2D tensor described by the arguments is worth compressing, + including statistics describing the expected savings from compression. We consider a tensor worth + compressing when ``min_compression_rate`` < uncompressed size / compressed size, where + uncompressed size = ``num_rows`` * ``num_cols``, + and compressed size = (``num_rows`` + ``num_cols``) * ``matrix_approximation_rank``. + + The result of this function is a tuple of the form (compression_recommendation, uncompressed_el_count, compressed_el_count), where: + + compression_recommendation is true if the tensor is worth compressing, and false otherwise (see above); + + uncompressed_el_count is the uncompressed element count, i.e. ``num_rows`` * ``num_cols``; and, + + compress_el_count is the element count after compression, i.e. (``num_rows`` + ``num_cols``) * ``matrix_approximation_rank``. + """ + uncompressed_size = num_rows * num_cols + compressed_size = (num_rows + num_cols) * matrix_approximation_rank + return ( + compressed_size * min_compression_rate < uncompressed_size, + uncompressed_size, + compressed_size, + ) + + +def _report_compression_stats(bucket, state): + """Report compression stats at frequency of ``compression_stats_logging_frequency`` specified in PowerSGD state.""" + if bucket.is_last() and state.iter >= state.next_stats_report: + stats = state.compression_stats() + logger.info( + "Compression stats: iter %s, total before compression %s, total after compression %s, " + "rate %s", + state.iter, + stats[1], + stats[2], + stats[0], + ) + state.next_stats_report = state.iter + state.compression_stats_logging_frequency + + +class PowerSGDState: + r""" + Store both the algorithm's hyperparameters and internal state for all gradients during training. + + Particularly, ``matrix_approximation_rank`` and ``start_powerSGD_iter`` are the main hyperparameters that should be tuned by the user. + For performance, we suggest to keep binary hyperparameters ``use_error_feedback`` and ``warm_start`` on. + + 1. ``matrix_approximation_rank`` controls the size of compressed low-rank tensors, which determines the compression rate. The lower the rank, the stronger the compression. + + 1.1. If ``matrix_approximation_rank`` is too low, the full model quality will need more training steps to reach or will never reach and yield loss in accuracy. + + 1.2. The increase of ``matrix_approximation_rank`` can substantially increase the computation costs of the compression, and the accuracy may not be further improved beyond a certain ``matrix_approximation_rank`` threshold. + + To tune ``matrix_approximation_rank``, we suggest to start from 1 and increase by factors of 2 (like an exponential grid search, 1, 2, 4, ...), until a satisfactory accuracy is reached. Typically only a small value 1-4 is used. For some NLP tasks (as shown in Appendix D of the original paper), this value has been increased to 32. + + 2. ``start_powerSGD_iter`` defers PowerSGD compression until step ``start_powerSGD_iter``, and vanilla allreduce runs prior to step ``start_powerSGD_iter``. This hybrid scheme of **vanilla allreduce + PowerSGD** can effectively improve the accuracy, even a relatively small ``matrix_approximation_rank`` is used. This is because that, the beginning of training phase is usually very sensitive to inaccurate gradients, and compressing gradients too early may make the training quickly take a suboptimal trajectory, which can result in an irrecoverable impact on the accuracy. + + To tune ``start_powerSGD_iter``, we suggest to start with 10% of total training steps, and increase it until a satisfactory accuracy is reached. If there is a warm-up stage in the training, ``start_powerSGD_iter`` typically should be no less than the number of warm-up steps. + + 3. ``min_compression_rate`` is the minimum compression rate required when a layer is compressed. Due to the computation overheads incurred by the compression, a tensor is worth compressing only if there can be sufficient saving in bandwidth, where ``(num_rows + num_cols) * matrix_approximation_rank * min_compression_rate < num_rows * num_cols``. If the specified compression rate threshold cannot be satisfied, the tensor will be directly allreduced without compression. + + Compression statistics are logged every ``compression_stats_logging_frequency`` iterations once PowerSGD compression starts. + + 4. ``orthogonalization_epsilon`` can be a very small value (e.g., 1e-8) added to every normalized matrix column in orthogonalization step, to prevent div-by-zero error if any column has all 0s. If this can already be prevented (e.g., by batch normalization), an epsilon of 0 is recommended for accuracy. + + 5. ``batch_tensors_with_same_shape`` controls whether to compress and decompress tensors with same shape in a batched operation to achieve higher parallelism. Note that you should also increase the bucket size (i.e., ``bucket_cap_mb`` arg in DDP constructor) to make more same-shaped tensors appear in the same bucket, however this may reduce the overlap between computation and communication, and increase the memory footprint due to stacking the tensors of the same shape. Set to ``True`` if the compression / decompression computation is a bottleneck. + + .. warning :: + If error feedback or warm-up is enabled, the minimum value of ``start_powerSGD_iter`` allowed in DDP is 2. + This is because there is another internal optimization that rebuilds buckets at iteration 1 in DDP, + and this can conflict with any tensor memorized before the rebuild process. + """ + + __slots__ = [ + "process_group", + # The fields below are the hyperparameters that often need to be tuned by the user. + "matrix_approximation_rank", + "start_powerSGD_iter", + # The fields below are the hyperparameters that seldom need be tuned by the user. + "min_compression_rate", + "orthogonalization_epsilon", + # The fields below are the binary hyperparameters recommended to be turned on for performance and accuracy. + "use_error_feedback", + "warm_start", + "batch_tensors_with_same_shape", + # The fields below are internal state. + "rng", + "error_dict", + "p_memory_dict", + "q_memory_dict", + "iter", + # The fields below are for recording compression stats. + "total_numel_before_compression", + "total_numel_after_compression", + "compression_stats_logging_frequency", + "next_stats_report", + ] + + def __init__( + self, + process_group, + matrix_approximation_rank=1, + start_powerSGD_iter=1_000, + min_compression_rate=2, + use_error_feedback=True, + warm_start=True, + orthogonalization_epsilon=0, + random_seed=0, + compression_stats_logging_frequency=10_000, + batch_tensors_with_same_shape: bool = False, + ): + logger.info( + "PowerSGD config: matrix_approximation_rank = %s; start_powerSGD_iter = %s; " + "min_compression_rate = %s; orthogonalization_epsilon = %s; use_error_feedback = %s; warm_start = %s; " + "random_seed = %s; compression_stats_logging_frequency = %s; batch_tensors_with_same_shape = %s", + matrix_approximation_rank, + start_powerSGD_iter, + min_compression_rate, + orthogonalization_epsilon, + use_error_feedback, + warm_start, + random_seed, + compression_stats_logging_frequency, + batch_tensors_with_same_shape, + ) + + self.process_group = process_group + self.matrix_approximation_rank = matrix_approximation_rank + # Deferring PowerSGD compression until step 'start_powerSGD_iter' can have two advantages: + # 1) It turns out that PowerSGD may lead to a non-trivial accuracy loss, + # even if the matrix approximation rank is increased to a large value. + # To mitigate the accuracy loss, a simple yet effective way is mixing vanilla allreduce + # (or a more conservative compression such as FP16 compression) with PowerSGD. + # 2) There is an internal optimization of rebuilding buckets process in DDP, + # in order to save the memory space. + # This step takes place after the first iteration. + # However, this means that the shape of input bucketized tensors is subject to change, + # which will complicate the implementations of error feedback and warm-up. + # Running vanilla allreduce in the first few iterations can avoid this complexity. + if (use_error_feedback or warm_start) and start_powerSGD_iter <= 1: + raise ValueError( + "Expect `start_powerSGD_iter` > 1 if `use_error_feedback` or `warm_start` is enabled, " + "because PowerSGD can only be applied after the first two iterations in DDP." + ) + self.start_powerSGD_iter = start_powerSGD_iter + self.min_compression_rate = min_compression_rate + # Error feedback is usually crucial for both for convergence and generalization, + # because PowerSGD is a biased compressor, + # i.e., compressing and decompressing a random gradient does not yield the original in expectation. + # This mechanism requires a temporary copy of the input gradients, + # so it increases the peak memory consumption by the size of the gradient tensor. + # However, if the target matrices are known to be exactly low-ranked (instead of just low stable rank), + # sometimes it is possible to converge to the optima without error feedback. + # See: http://proceedings.mlr.press/v54/yurtsever17a/yurtsever17a.pdf + self.use_error_feedback = use_error_feedback + # Warm-start reuses P(s) and Q(s) from the previous iteration. + # This can improve the approximation quality and hence improve the accuracy. + # Additionally, by avoiding the initialization of these low-rank tensors at every step, + # this can also accelerate training. + # However, this is at the cost of extra memory. + self.warm_start = warm_start + # Can use a very small value to prevent div-by-zero error caused by orthogonalization of vanishing gradients. + self.orthogonalization_epsilon = orthogonalization_epsilon + # The purpose of this RNG is to generate different random seeds for initializing Q across iterations, + # but in the same order for all the DDP replicas. + # Different random seeds across iterations indicate different 'projections' of the gradients at different SGD steps. + # If the same random projection is used, + # there will be differences between the gradients that are never synchronized. + import numpy as np + + self.rng = np.random.RandomState(random_seed) + # Since there is only a single state instance for all the input buckets, + # need to maintain a dictionary that maps each bucket index to the local error. + self.error_dict: dict[int, torch.Tensor] = {} + self.p_memory_dict: dict[int, torch.Tensor] = {} + self.q_memory_dict: dict[int, torch.Tensor] = {} + # Iteration/step in the training loop. + self.iter = 0 + # Compression stats accumulators + self.total_numel_before_compression = 0 + self.total_numel_after_compression = 0 + # We'll report compression stats every 'compression_stats_logging_frequency' iterations + # Note that we always report compression stats at least once. + self.compression_stats_logging_frequency = max( + 1, compression_stats_logging_frequency + ) + self.next_stats_report = 0 + # Batching tensors with same shape can increase parallelism in compression / decompression computation. + # This requires a larger bucket size to make more same-shaped tensor to appear in one bucket, however + # this may reduce the overlap between computation and communication, and increase the memory footprint + # due to stacking tensors. + # Turn on if compression / decompression computation is a bottleneck. + self.batch_tensors_with_same_shape = batch_tensors_with_same_shape + + def __getstate__(self): + r""" + Return a ``Dict[str, Any]`` which will be pickled and saved. + + ``process_group`` is not serializable and excluded from + a returned state. + """ + logger.warning( + "NOTE: Process group is not serializable and excluded from a saved state." + ) + return { + slot: getattr(self, slot) + for slot in self.__slots__ + if slot != "process_group" + } + + def __setstate__(self, state): + r""" + Take a provided ``state`` and set to this ``PowerSGDState`` instance. + + ``process_group`` is set to default. + """ + self.process_group = distributed_c10d._get_default_group() + logger.warning( + "NOTE: Process group will be set to a default group (i.e. the world size).\ + If a different group is desired, please set `self.process_group` after PowerSGD state is loaded." + ) + for slot, value in state.items(): + setattr(self, slot, value) + + def maybe_increase_iter(self, bucket): + """Track iterations and trigger log message at start of local SGD.""" + # Since bucket 0 is the last bucket to allreduce in an iteration. + # Only increase `iter` when bucket 0 is processed. + if bucket.is_last(): + self.iter += 1 + + if self.iter == self.start_powerSGD_iter: + logger.info("Start to apply PowerSGD after %s iterations.", self.iter) + + def compression_stats(self): + r""" + Return latest compression statistics as tuple. + + Returns tuple of form (compress_rate, numel_before_compression, numel_after_compression) where: + + compress_rate is the effective compression rate i.e. (number of elements before compression) / (number of elements after compression); + + numel_before_compression is the total number of elements before compression was applied; and, + + numel_after_compression is the total number of elements after compression was applied. + """ + compress_rate = ( + self.total_numel_before_compression / self.total_numel_after_compression + if self.total_numel_after_compression > 0 + else 0 + ) + return ( + compress_rate, + self.total_numel_before_compression, + self.total_numel_after_compression, + ) + + +def powerSGD_hook( + state: PowerSGDState, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + r""" + Implement PowerSGD algorithm. + + This DDP communication hook implements PowerSGD gradient compression + algorithm described in the `paper `_. + Once gradient tensors are aggregated across all workers, this hook applies + compression as follows: + + 1. Views the input flattened 1D gradient tensor as a list of per-parameter tensors, and divides all the tensors into two groups: + + 1.1 The tensors that should be compressed before allreduce, because the compression can give enough saving in bandwidth. + + 1.2 Rest of the tensors will be directly allreduced without compression, including all the vector tensors (for biases). + + 2. Handles uncompressed tensors: + + 2.1. Allocate contiguous memory for those uncompressed tensors, and allreduces all the uncompressed tensors as a batch, without compression; + + 2.2. Copies the individual uncompressed tensors from the contiguous memory back to the input tensor. + + 3. Handles the tensors that should be compressed by PowerSGD compression: + + 3.1. For each tensor M, creates two low-rank tensors P and Q for decomposing M, + such that M = PQ^T, where Q is initialized from a standard normal distribution and orthogonalized; + + 3.2. Computes each P in Ps, which is equal to MQ; + + 3.3. Allreduces Ps as a batch; + + 3.4. Orthogonalizes each P in Ps; + + 3.5. Computes each Q in Qs, which is approximately equal to M^TP; + + 3.6. Allreduces Qs as a batch; + + 3.7. Computes each M among all the compressed tensors, which is approximately equal to PQ^T. + + Note that this communication hook enforces vanilla allreduce for the first ``state.start_powerSGD_iter`` iterations. + This not only gives the user more control over the tradeoff between speedup and accuracy, + but also helps abstract away some complexity of the internal optimization of DDP for future communication hook developers. + + Args: + state (PowerSGDState): State information to configure the compression rate and support error feedback, warm start, etc. + To tune the compression configs, mainly need to tune ``matrix_approximation_rank``, ``start_powerSGD_iter`` + and ``min_compression_rate``. + bucket (dist.GradBucket): Bucket that stores a 1D flattened gradient tensor that batches multiple per-variable tensors. + Note that since DDP comm hook only supports single process single device mode, + only exactly one tensor is stored in this bucket. + + Returns: + Future handler of the communication, which updates the gradients in place. + + Example:: + >>> # xdoctest: +SKIP + >>> state = PowerSGDState(process_group=process_group, matrix_approximation_rank=1, + start_powerSGD_iter=10, min_compression_rate=0.5) + >>> ddp_model.register_comm_hook(state, powerSGD_hook) + """ + process_group = state.process_group + group_to_use = ( + process_group if process_group is not None else not_none(dist.group.WORLD) + ) + world_size = group_to_use.size() + + # The input tensor is a flattened 1D tensor. + input_tensor = bucket.buffer() + + # Run vanilla allreduce in the first `start_powerSGD_iter` iterations. + if state.iter < state.start_powerSGD_iter: + state.maybe_increase_iter(bucket) + return default._allreduce_fut(group_to_use, input_tensor) + + # Apply PowerSGD after `start_powerSGD_iter` iterations. + device = input_tensor.device + dtype = input_tensor.dtype + + # Incorporate the error from the previous state into the gradients. + bucket_index = bucket.index() + input_tensor_cp = None + total_length = input_tensor.shape[0] + if state.use_error_feedback: + if bucket_index in state.error_dict: + input_tensor.add_(state.error_dict[bucket_index]) + else: + logger.info( + "A zero tensor of length %s that represents local error is created.", + total_length, + ) + state.error_dict[bucket_index] = torch.zeros( + total_length, device=device, dtype=dtype + ) + + # Keep a copy of the input tensor, + # so that we can compute the local error caused by compression later, + # by comparing this copy and the input tensor updated after decompression. + input_tensor_cp = input_tensor.detach().clone() + + # Unflatten the input tensor into per-parameter tensors, for layer-wise compression. + tensors = bucket.gradients() + + # Step I: Divide all the tensors into two groups, + # one will be compressed before allreduce and the other will be directly allreduced without compression. + tensors_to_compress, uncompressed_tensors = [], [] + total_Ps_size = 0 + total_Qs_size = 0 + for tensor in tensors: + matrix = tensor.view(tensor.shape[0], -1) + n, m = matrix.shape + matrix_approximation_rank = min(n, m, state.matrix_approximation_rank) + compress_test = _should_compress( + n, m, matrix_approximation_rank, state.min_compression_rate + ) + state.total_numel_before_compression += compress_test[1] + if compress_test[0]: + tensors_to_compress.append(matrix) + total_Ps_size += n * matrix_approximation_rank + total_Qs_size += m * matrix_approximation_rank + state.total_numel_after_compression += compress_test[2] + else: + uncompressed_tensors.append(tensor) + state.total_numel_after_compression += compress_test[1] + + _report_compression_stats(bucket, state) + + # Step II: Handle uncompressed tensors. + # Allocate contiguous memory for these tensors to allreduce efficiently. + uncompressed_tensors_memory = ( + torch.cat([tensor.view(-1) for tensor in uncompressed_tensors]) + if uncompressed_tensors + else torch.tensor([], device=device, dtype=dtype) + ) + + # Step III: Handle the tensors that should be compressed. + # Allocate contiguous memory for Ps and Qs to allreduce efficiently. + # If warm-start is enabled, reuse Ps and Qs from the previous iteration if possible. + # The memory spaces of Ps and Qs need to be allocated in the first iteration when PowerSGD is applied. + need_randomize_qs = False + if not state.warm_start or bucket_index not in state.p_memory_dict: + need_randomize_qs = True + # If warm-start is disabled, low-rank tensors will be initialized at every step. + # Only log this if warm-start to avoid spamming. + if state.warm_start: + logger.info( + "Allocating contiguous memory of length %s for Ps, and of length %s for Qs, respectively.", + total_Ps_size, + total_Qs_size, + ) + state.p_memory_dict[bucket_index] = torch.empty( + total_Ps_size, device=device, dtype=dtype + ) + state.q_memory_dict[bucket_index] = torch.empty( + total_Qs_size, device=device, dtype=dtype + ) + + # Batch tensors to compress by shape. + shape_to_tensors = defaultdict(list) + for tensor in tensors_to_compress: + shape_to_tensors[tensor.shape].append(tensor) + + # This function decides whether to batch tensors with same shape or not according to the argument, + # so the following process could share the same code. + def maybe_batched_tensors_to_compress(): + for tensors in shape_to_tensors.values(): + if state.batch_tensors_with_same_shape: + batch_size = len(tensors) + if batch_size == 1: + # Use the original tensor to avoid copy. + yield tensors[0].unsqueeze(0) + else: + yield torch.stack(tensors) + else: + for tensor in tensors: + yield tensor.unsqueeze(0) + + # Create Ps and Qs that point to the allocated memory. + tensors_to_compress = [] + ps = [] + qs = [] + p_idx = 0 + q_idx = 0 + for tensor in maybe_batched_tensors_to_compress(): + batch_size, n, m = tensor.shape + matrix_approximation_rank = min(n, m, state.matrix_approximation_rank) + tensors_to_compress.append(tensor) + ps.append( + state.p_memory_dict[bucket_index][ + p_idx : p_idx + batch_size * n * matrix_approximation_rank + ].view(batch_size, n, matrix_approximation_rank) + ) + qs.append( + state.q_memory_dict[bucket_index][ + q_idx : q_idx + batch_size * m * matrix_approximation_rank + ].view(batch_size, m, matrix_approximation_rank) + ) + p_idx += batch_size * n * matrix_approximation_rank + q_idx += batch_size * m * matrix_approximation_rank + + # If warm-start is enabled, reuse Qs from the previous iteration if possible and skip filling random values. + # The exception is the first iteration when PowerSGD is applied. + if not need_randomize_qs: + for q in qs: + _orthogonalize(q, state.orthogonalization_epsilon) + else: + with torch.random.fork_rng(devices=[]): + # Fork this RNG to avoid changing the seed globally and affecting the random sampling anywhere else in the training. + # The seed makes sure that the initial random values are the same across all the DDP replicas. + # This seed should differ at every step. + # Since it is very slow to fork RNG state across all the CUDA devices, + # only fork on CPU and then move the generated tensor to the CUDA device (by overwriting q). + torch.manual_seed(state.rng.randint(1_000_000_000)) + for q in qs: + q.copy_( + torch.randn( + *q.shape, + device="cpu", + dtype=dtype, + ) + ) + _orthogonalize(q, state.orthogonalization_epsilon) + + # Compute Ps. + for tensor, q, p in zip(tensors_to_compress, qs, ps): + torch.bmm(tensor, q, out=p) + + # This allreduce is only applied to uncompressed tensors, + # so it should have been kicked off before the above computation on the compressed tensors to hide more communication costs. + # However, this somehow requires a separate future chain at this time. + allreduce_contiguous_uncompressed_tensors_fut = dist.all_reduce( + uncompressed_tensors_memory, group=group_to_use, async_op=True + ).get_future() + + def unpack_uncompressed_tensors_and_allreduce_ps(fut): + uncompressed_tensors_memory = fut.value()[0].div_(world_size) + idx = 0 + for tensor in uncompressed_tensors: + tensor.copy_( + uncompressed_tensors_memory[idx : idx + tensor.numel()].view_as(tensor) + ) + idx += tensor.numel() + + # Since these Ps will be orthogonalized later, no need to divide them by world size. + return ( + dist.all_reduce( + state.p_memory_dict[bucket_index], group=group_to_use, async_op=True + ) + .get_future() + .wait()[0] + ) + + def compute_qs(fut): + state.p_memory_dict[bucket_index] = fut.value() + for p in ps: + _orthogonalize(p, state.orthogonalization_epsilon) + + # Compute Qs. + for tensor, p, q in zip(tensors_to_compress, ps, qs): + torch.bmm(tensor.transpose(1, 2), p, out=q) + + # TODO: The above procedure does two matmul+allreduce steps per iteration -- + # one left multiplication and one right multiplication. + # For warm-start, can take one such step at a time, and alternate between them. + + # Allreduce Qs. + return ( + dist.all_reduce( + state.q_memory_dict[bucket_index], group=group_to_use, async_op=True + ) + .get_future() + .wait()[0] + ) + + def decompress(fut): + state.q_memory_dict[bucket_index] = fut.value().div_(world_size) + + for p, q, tensor in zip(ps, qs, tensors_to_compress): + torch.bmm(p, q.transpose(1, 2), out=tensor) + + # Copy batched tensors back to original buffer. + if state.batch_tensors_with_same_shape: + for tensor in tensors_to_compress: + if tensor.shape[0] == 1: + # Skip tensor with batch_size == 1 since itself is the original tensor. + continue + original_tensors = shape_to_tensors[tensor.shape[1:]] + for i, original_tensor in enumerate(original_tensors): + original_tensor.copy_(tensor[i]) + + if torch.cuda.is_available(): + torch.cuda.synchronize(device) + + if state.use_error_feedback: + # Memorize the local errors. + if input_tensor_cp is None: + raise AssertionError + state.error_dict[bucket_index] = input_tensor_cp - input_tensor + if not state.warm_start: + state.p_memory_dict.clear() + state.q_memory_dict.clear() + + state.maybe_increase_iter(bucket) + + return input_tensor + + return ( + allreduce_contiguous_uncompressed_tensors_fut.then( + unpack_uncompressed_tensors_and_allreduce_ps + ) + .then(compute_qs) + .then(decompress) + ) + + +def batched_powerSGD_hook( + state: PowerSGDState, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + r""" + Implement simplified PowerSGD algorithm. + + This DDP communication hook implements a simplified PowerSGD gradient compression + algorithm described in the `paper `_. + This variant does not compress the gradients layer by layer, + but instead compresses the flattened input tensor that batches all the gradients. + Therefore, it is **faster** than :meth:`powerSGD_hook`, + but usually results in a **much lower accuracy**, unless ``matrix_approximation_rank`` is 1. + + .. warning :: + Increasing ``matrix_approximation_rank`` here may not necessarily increase the accuracy, + because batching per-parameter tensors without column/row alignment can destroy low-rank structure. + Therefore, the user should always consider :meth:`powerSGD_hook` first, + and only consider this variant when a satisfactory accuracy can be achieved when ``matrix_approximation_rank`` is 1. + + Once gradient tensors are aggregated across all workers, this hook applies + compression as follows: + + 1. Views the input flattened 1D gradient tensor as a square-shaped tensor M with 0 paddings; + + 2. Creates two low-rank tensors P and Q for decomposing M, such that M = PQ^T, where Q is initialized from a standard normal distribution and orthogonalized; + + 3. Computes P, which is equal to MQ; + + 4. Allreduces P; + + 5. Orthogonalizes P; + + 6. Computes Q, which is approximately equal to M^TP; + + 7. Allreduces Q; + + 8. Computes M, which is approximately equal to PQ^T. + + 9. Truncates the input tensor to the original length. + + Note that this communication hook enforces vanilla allreduce for the first ``state.start_powerSGD_iter`` iterations. + This not only gives the user more control over the tradeoff between speedup and accuracy, + but also helps abstract away some complexity of the internal optimization of DDP for future communication hook developers. + + Args: + state (PowerSGDState): State information to configure the compression rate and support error feedback, warm start, etc. + To tune the compression configs, mainly need to tune ``matrix_approximation_rank`` and ``start_powerSGD_iter``. + bucket (dist.GradBucket): Bucket that stores a 1D flattened gradient tensor that batches multiple per-variable tensors. + Note that since DDP comm hook only supports single process single device mode, + only exactly one tensor is stored in this bucket. + + Returns: + Future handler of the communication, which updates the gradients in place. + + Example:: + >>> # xdoctest: +SKIP + >>> state = PowerSGDState(process_group=process_group, matrix_approximation_rank=1) + >>> ddp_model.register_comm_hook(state, batched_powerSGD_hook) + """ + process_group = state.process_group + group_to_use = ( + process_group if process_group is not None else not_none(dist.group.WORLD) + ) + world_size = group_to_use.size() + + # The input tensor is a flattened 1D tensor. + input_tensor = bucket.buffer() + + # Run vanilla allreduce in the first `start_powerSGD_iter` iterations. + if state.iter < state.start_powerSGD_iter: + state.maybe_increase_iter(bucket) + return default._allreduce_fut(group_to_use, input_tensor) + + # Apply PowerSGD after `start_powerSGD_iter` iterations. + device = input_tensor.device + total_length = input_tensor.shape[0] + state.total_numel_before_compression += total_length + + # View the input tensor as a 2D square-shape tensor, and pad 0s if necessary. + square_side_length = math.ceil(math.sqrt(total_length)) + state.total_numel_after_compression += ( + square_side_length * state.matrix_approximation_rank * 2 + ) + padded_total_length = square_side_length**2 + input_tensor.resize_(padded_total_length) + input_tensor[total_length:padded_total_length].fill_(0) + + _report_compression_stats(bucket, state) + + # Incorporate the error from the previous state into the gradients. + bucket_index = bucket.index() + input_tensor_cp = None + if state.use_error_feedback: + if bucket_index in state.error_dict: + input_tensor.add_(state.error_dict[bucket_index]) + else: + logger.info( + "A zero tensor of length %s that represents local error is created.", + padded_total_length, + ) + state.error_dict[bucket_index] = torch.zeros( + padded_total_length, device=device, dtype=input_tensor.dtype + ) + + # Keep a copy of the input tensor, + # so that we can compute the local error caused by compression later, + # by comparing this copy and the input tensor updated after decompression. + input_tensor_cp = input_tensor.detach().clone() + matrix = input_tensor.view(square_side_length, square_side_length) + + # Reuse P and Q from the previous iteration if possible. + # The memory spaces of P and Q need to be allocated in the first iteration when PowerSGD is applied. + if not state.warm_start or bucket_index not in state.p_memory_dict: + # If warm-start is disabled, low-rank tensors will be initialized at every step. + # Only log this if warm-start to avoid spamming. + if state.warm_start: + logger.info( + "Initializing low-rank tensors P and Q, each of which has a shape of %s x %s.", + square_side_length, + state.matrix_approximation_rank, + ) + + def create_low_rank_tensor(fill_random_values, rng): + """Return a low-rank 2D tensor of square_side_length * matrix_approximation_rank.""" + if fill_random_values: + with torch.random.fork_rng(devices=[]): + # Fork this RNG to avoid changing the seed globally and affecting the random sampling + # anywhere else in the training. + # The seed makes sure that the initial random values are the same across all the DDP replicas. + # This seed should differ at every step. + # Since it is very slow to fork RNG state across all the CUDA devices, + # only fork on CPU and then move the generated tensor to the CUDA device. + torch.manual_seed(rng.randint(1_000_000_000)) + return torch.randn( + square_side_length, + state.matrix_approximation_rank, + device="cpu", + dtype=input_tensor.dtype, + ).to(device) + else: + return torch.empty( + square_side_length, + state.matrix_approximation_rank, + device=device, + dtype=input_tensor.dtype, + ) + + state.p_memory_dict[bucket_index] = create_low_rank_tensor( + fill_random_values=False, rng=state.rng + ) + state.q_memory_dict[bucket_index] = create_low_rank_tensor( + fill_random_values=True, rng=state.rng + ) + _orthogonalize(state.q_memory_dict[bucket_index]) + + torch.matmul( + matrix, state.q_memory_dict[bucket_index], out=state.p_memory_dict[bucket_index] + ) + allreduce_p_fut = dist.all_reduce( + state.p_memory_dict[bucket_index], group=group_to_use, async_op=True + ).get_future() + + def compute_q(fut): + state.p_memory_dict[bucket_index] = fut.value()[0] + _orthogonalize(state.p_memory_dict[bucket_index]) + + torch.matmul( + matrix.t(), + state.p_memory_dict[bucket_index], + out=state.q_memory_dict[bucket_index], + ) + + # TODO: The above procedure does two matmul+allreduce steps per iteration -- + # one left multiplication and one right multiplication. + # For warm-start, can take one such step at a time, and alternate between them. + + return ( + dist.all_reduce( + state.q_memory_dict[bucket_index], group=group_to_use, async_op=True + ) + .get_future() + .wait()[0] + ) + + def decompress(fut): + state.q_memory_dict[bucket_index] = fut.value().div_(world_size) + torch.matmul( + state.p_memory_dict[bucket_index], + state.q_memory_dict[bucket_index].t(), + out=matrix, + ) + + if state.use_error_feedback: + # Memorize the local errors. + if input_tensor_cp is None: + raise AssertionError + state.error_dict[bucket_index] = input_tensor_cp - input_tensor + # Removing this seemingly unnecessary sync somehow may cause failures. + # See: https://github.com/pytorch/pytorch/pull/54838 + if torch.cuda.is_available(): + torch.cuda.synchronize(device) + if not state.warm_start: + state.p_memory_dict.clear() + state.q_memory_dict.clear() + ret = input_tensor.resize_(total_length) + + state.maybe_increase_iter(bucket) + + return ret + + return allreduce_p_fut.then(compute_q).then(decompress) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/quantization_hooks.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/quantization_hooks.py new file mode 100644 index 0000000000000000000000000000000000000000..886155908e1a702972184a550082c33c677eacdc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/ddp_comm_hooks/quantization_hooks.py @@ -0,0 +1,220 @@ +# mypy: allow-untyped-defs +import torch +import torch.distributed as dist +from torch import nn + + +def _quantize_per_tensor_backend(x, scale, zero_point): + y = torch.round(x / scale) + zero_point + y = torch.clamp(y, 0, 255).to(torch.uint8) + return y + + +def _dequantize_per_tensor_backend(y, scale, zero_point): + x = scale * (y.to(torch.float32) - zero_point) + return x + + +def _quantize_per_channel_backend(x, scale, zero_point): + y = torch.zeros(x.size(), device=x.device) + for i in range(x.size()[0]): + y[i, :] = torch.round(x[i, :] / scale[i]) + zero_point[i] + y = torch.clamp(y, 0, 255).to(torch.uint8) + return y + + +def _dequantize_per_channel_backend(y, scale, zero_point): + y = y.to(torch.float32).to(y.device) + x = torch.zeros_like(y, device=y.device) + for i in range(x.size()[0]): + x[i, :] = scale[i] * (y[i, :] - zero_point[i]) + return x + + +def _get_allgather_out_list(all_gather_in_list, world_size): + out_list = [ + torch.zeros_like( + all_gather_in_list, + device=all_gather_in_list.device, + dtype=all_gather_in_list.dtype, + ) + for _ in range(world_size) + ] + return out_list + + +def quantization_pertensor_hook( + process_group: dist.ProcessGroup, bucket: dist.GradBucket +) -> torch.futures.Future[torch.Tensor]: + """ + Apply ``torch.quantize_per_tensor`` logic to DDP using ``allgather`` protocol. + + Workers first allgather the scale and zero point of their own + ``GradBucket`` prior to the quantization. After all workers have that information, + the first ``then`` callback called ``quantize_and_allgather`` quantizes worker's + own gradient tensor, and uses ``allgather`` to communicate these across all workers. + The final ``then`` callback called ``dequantize_and_aggregate``, dequantizes and + aggregates each quantized gradient tensor locally and returns the mean. + + .. warning :: + This is experimental, and uses ``allgather`` protocol which is considerably slower than + ``allreduce`` protocol. It works only with flattened grads. + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(process_group, quantization_pertensor_hook) + """ + group_to_use = process_group if process_group is not None else dist.group.WORLD + rank = process_group.rank() if process_group is not None else dist.get_rank() + # pyrefly: ignore [missing-attribute] + world_size = group_to_use.size() + + tensor = bucket.buffer() + + myObserver = torch.ao.quantization.MinMaxObserver().to(tensor.device) + myObserver(tensor) + + s, z = myObserver.calculate_qparams() + s_and_z = torch.FloatTensor([s, z]).to(tensor.device) + + all_ranks_s_and_z = _get_allgather_out_list(s_and_z, world_size) + + # First, allgather scale and zeros. + fut = dist.all_gather( + all_ranks_s_and_z, s_and_z, group=group_to_use, async_op=True + ).get_future() + + def quantize_and_allgather(fut): + # Store scale and zeros across all workers. + all_ranks_s_and_z = fut.wait()[0] + # All workers quantize their own ``GradBucket`` tensors. + quantized_tensor = _quantize_per_tensor_backend( + tensor, all_ranks_s_and_z[rank][0], all_ranks_s_and_z[rank][1] + ) + # Allgather quantized tensors. + fut = dist.all_gather( + _get_allgather_out_list(quantized_tensor, world_size), + quantized_tensor, + group=group_to_use, + async_op=True, + ).get_future() + + return fut.wait() + + def dequantize_and_aggregate(fut): + all_ranks_quantized_tensor = fut.wait()[0] + + aggregated_dequantized_tensor = torch.zeros_like( + all_ranks_quantized_tensor[0], device=tensor.device, dtype=torch.float32 + ) + # Using previously allgathered scales and zeros, dequantize gradient tensors + # locally and then aggregate them. + for r, quantized_tensor in enumerate(all_ranks_quantized_tensor): + aggregated_dequantized_tensor += _dequantize_per_tensor_backend( + quantized_tensor, all_ranks_s_and_z[r][0], all_ranks_s_and_z[r][1] + ) + + return aggregated_dequantized_tensor / world_size + + return fut.then(quantize_and_allgather).then(dequantize_and_aggregate) + + +def quantization_perchannel_hook( + process_group: dist.ProcessGroup, bucket: dist.GradBucket, bucket_size=512 +) -> torch.futures.Future[torch.Tensor]: + """ + Apply``torch.quantize_per_channel`` logic to DDP using ``allgather`` protocol. + + Compared to per-tensor, the main motivation of per-channel is + for considerably large tensors such as a tensor that contains 6 million + elements quantizing per a bucket size of 512 (or 128) elements may significantly + increase the resolution. + + It first splits ``GradBucket`` tensor into multiple chunks (channels) of ``bucket_size`` + elements. Then, workers allgather the scales and zero points of their own + ``GradBucket`` prior to the quantization. After all workers have that information, + the first ``then`` callback called ``quantize_and_allgather`` quantizes worker's + own gradient tensor, and uses ``allgather`` to communicate these across all workers. + The final ``then`` callback called ``dequantize_and_aggregate``, dequantizes, flattens, and + aggregates each quantized gradient tensor locally and returns the mean. + + .. warning :: + This is experimental, and uses ``allgather`` protocol which is considerably slower than + ``allreduce`` protocol. It works only with flattened grads. + + Example:: + >>> # xdoctest: +SKIP + >>> ddp_model.register_comm_hook(process_group, quantization_perchannel_hook) + """ + group_to_use = process_group if process_group is not None else dist.group.WORLD + rank = process_group.rank() if process_group is not None else dist.get_rank() + # pyrefly: ignore [missing-attribute] + world_size = group_to_use.size() + + tensor = bucket.buffer() + + tensor_in_channels = ( + nn.functional.pad( + input=tensor, + pad=(0, bucket_size - len(tensor) % bucket_size), + mode="constant", + value=0, + ) + .view(-1, bucket_size) + .to(tensor.device) + ) + + myPerChannelObserver = torch.ao.quantization.PerChannelMinMaxObserver().to( + tensor.device + ) + myPerChannelObserver(tensor_in_channels) + + s_ch, z_ch = myPerChannelObserver.calculate_qparams() + s_and_z = torch.stack((s_ch, z_ch)).to(tensor.device) + + all_ranks_s_and_z = _get_allgather_out_list(s_and_z, world_size) + # First, allgather scale and zeros. + fut = dist.all_gather( + all_ranks_s_and_z, s_and_z, group=group_to_use, async_op=True + ).get_future() + + def quantize_and_allgather(fut): + # Store scale and zeros across all workers. + all_ranks_s_and_z = fut.wait()[0] + # All workers quantize their corresponding ``GradBucket`` tensors. + quantized_tensor = _quantize_per_channel_backend( + tensor_in_channels, + all_ranks_s_and_z[rank, 0, :], + all_ranks_s_and_z[rank, 1, :], + ) + # Allgather quantized tensors. + fut = dist.all_gather( + _get_allgather_out_list(quantized_tensor, world_size), + quantized_tensor, + group=group_to_use, + async_op=True, + ).get_future() + + return fut.wait() + + def dequantize_and_aggregate(fut): + all_ranks_quantized_tensor = fut.wait()[0] + + aggregated_dequantized_tensor = torch.zeros_like( + all_ranks_quantized_tensor[0], device=tensor.device, dtype=torch.float32 + ) + # Using previously allgathered scales and zeros, dequantize gradient tensors + # locally and then aggregate them. + for r, quantized_tensor in enumerate(all_ranks_quantized_tensor): + aggregated_dequantized_tensor += _dequantize_per_channel_backend( + quantized_tensor, all_ranks_s_and_z[r][0], all_ranks_s_and_z[r][1] + ) + + return ( + torch.flatten(aggregated_dequantized_tensor).to(tensor.device)[ + : tensor.size()[0] + ] + / world_size + ) + + return fut.then(quantize_and_allgather).then(dequantize_and_aggregate) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/join.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/join.py new file mode 100644 index 0000000000000000000000000000000000000000..2c56b1fdd96870d0a5a882f4f27cd2609bfb84e2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/join.py @@ -0,0 +1,352 @@ +# mypy: allow-untyped-defs +import warnings +from abc import ABC, abstractmethod +from types import TracebackType +from typing import Any, NamedTuple + +import torch +import torch.distributed as dist + + +__all__ = ["JoinHook", "Joinable", "Join"] + + +class JoinHook: + r""" + This defines a join hook, which provides two entry points in the join context manager. + + Entry points : a main hook, which is called repeatedly while there exists a non-joined + process, and a post-hook, which is called once all processes have joined. + + To implement a join hook for the generic join context manager, define a + class that inherits from :class:`JoinHook` and override ``main_hook()`` and + ``post_hook()`` as appropriate. + """ + + def main_hook(self) -> None: + r"""Call this hook while there exists a non-joined process to shadow collective communications in a training iteration. + + Training iteration i.e., in one forward pass, backward pass, and optimizer step. + """ + + def post_hook(self, is_last_joiner: bool) -> None: + r""" + Call hook after all processes have joined. + + It is passed an additional ``bool`` argument ``is_last_joiner``, which indicates if the rank is one of the last to join. + + Arguments: + is_last_joiner (bool): ``True`` if the rank is one of the last to + join; ``False`` otherwise. + """ + + +class Joinable(ABC): + r""" + This defines an abstract base class for joinable classes. + + A joinable class + (inheriting from :class:`Joinable`) should implement :meth:`join_hook`, + which returns a :class:`JoinHook` instance, in addition to + :meth:`join_device` and :meth:`join_process_group` that return device and + process group information, respectively. + """ + + @abstractmethod + def __init__(self) -> None: + super().__init__() + self._join_config = _JoinConfig.construct_disabled_join_config() + + @abstractmethod + def join_hook(self, **kwargs) -> JoinHook: + r""" + Return a :class:`JoinHook` instance for the given :class:`Joinable`. + + Arguments: + kwargs (dict): a :class:`dict` containing any keyword arguments + to modify the behavior of the join hook at run time; all + :class:`Joinable` instances sharing the same join context + manager are forwarded the same value for ``kwargs``. + """ + ... + + @property + @abstractmethod + def join_device(self) -> torch.device: + r"""Return the device from which to perform collective communications needed by the join context manager.""" + ... + + @property + @abstractmethod + def join_process_group(self) -> Any: + r"""Returns the process group for the collective communications needed by the join context manager itself.""" + ... + + +class _JoinConfig(NamedTuple): + r"""This includes all fields needed from a :class:`Joinable` instance for the join context manager side.""" + + enable: bool + throw_on_early_termination: bool + is_first_joinable: bool + + @staticmethod + def construct_disabled_join_config(): + r"""Return a :class:`_JoinConfig` instance indicating that join-related logic should be disabled. + + e.g. if the caller is not in a join context manager. + """ + return _JoinConfig( + enable=False, throw_on_early_termination=False, is_first_joinable=False + ) + + +class Join: + r""" + This class defines the generic join context manager, which allows custom hooks to be called after a process joins. + + These hooks should shadow the + collective communications of non-joined processes to prevent hanging and + erroring and to ensure algorithmic correctness. Refer to :class:`JoinHook` + for details about the hook definition. + + .. warning:: + The context manager requires each participating :class:`Joinable` to + call the method :meth:`notify_join_context()` before its own per- + iteration collective communications to ensure correctness. + + .. warning:: + The context manager requires that all ``process_group`` attributes in + the :class:`JoinHook` objects are the same. If there are multiple + :class:`JoinHook` objects, then the ``device`` of the first is used. + The process group and device information is used for checking for non- + joined processes and for notifying processes to throw an exception if + ``throw_on_early_termination`` is enabled, both of which using an all- + reduce. + + Arguments: + joinables (List[Joinable]): a list of the participating + :class:`Joinable` s; their hooks are iterated over in the given + order. + + enable (bool): a flag enabling uneven input detection; setting to + ``False`` disables the context manager's functionality and should + only be set when the user knows the inputs will not be uneven + (default: ``True``). + + throw_on_early_termination (bool): a flag controlling whether to throw an + exception upon detecting uneven inputs (default: ``False``). + + Example:: + + >>> import os + >>> import torch + >>> import torch.distributed as dist + >>> import torch.multiprocessing as mp + >>> # xdoctest: +SKIP + >>> import torch.nn.parallel.DistributedDataParallel as DDP + >>> import torch.distributed.optim.ZeroRedundancyOptimizer as ZeRO + >>> from torch.distributed.algorithms.join import Join + >>> + >>> # On each spawned worker + >>> def worker(rank): + >>> dist.init_process_group("nccl", rank=rank, world_size=2) + >>> model = DDP(torch.nn.Linear(1, 1).to(rank), device_ids=[rank]) + >>> optim = ZeRO(model.parameters(), torch.optim.Adam, lr=0.01) + >>> # Rank 1 gets one more input than rank 0 + >>> inputs = [torch.tensor([1.]).to(rank) for _ in range(10 + rank)] + >>> with Join([model, optim]): + >>> for input in inputs: + >>> loss = model(input).sum() + >>> loss.backward() + >>> optim.step() + >>> # All ranks reach here without hanging/erroring + """ + + def __init__( + self, + joinables: list[Joinable], + enable: bool = True, + throw_on_early_termination: bool = False, + **kwargs, + ): + if len(joinables) == 0: + raise ValueError("The join context manager requires at least one joinable") + self._joinables = joinables + self._join_hooks = [ + joinable.join_hook(**kwargs) for joinable in self._joinables + ] + self._enable = enable + self._throw_on_early_termination = throw_on_early_termination + self._set_joinable_configs() + self._extract_dist_info() + + def _set_joinable_configs(self) -> None: + r"""Set the :class:`_JoinConfig` of each participating :class:`Joinable`.""" + if len(self._joinables) <= 0: + raise AssertionError + is_first_joinable = True + for joinable in self._joinables: + joinable._join_config = _JoinConfig( + enable=self._enable, + throw_on_early_termination=self._throw_on_early_termination, + is_first_joinable=is_first_joinable, + ) + is_first_joinable = False + + def _extract_dist_info(self) -> None: + r""" + Extract the process group and device information from the joinables. + + If there are multiple joinables, then the context manager uses the + first specified device. + + Preconditions: + ``self._joinables`` is not ``None`` and is non-empty. + + Raises: + ValueError + If there are multiple conflicting ``process_group`` attributes + among the ``Joinable`` objects. + """ + process_group = None + device = None + # pyrefly: ignore [bad-assignment] + for joinable in self._joinables: + if process_group is None: + process_group = joinable.join_process_group + elif process_group != joinable.join_process_group: + raise ValueError( + "Using join context manager with multiple process groups" + ) + if device is None: + device = joinable.join_device + self._process_group = process_group + self._rank = dist.get_rank(self._process_group) + self._device = device + + def __enter__(self): ... + + def __exit__( + self, + type: type[BaseException] | None, + value: BaseException | None, + traceback: TracebackType | None, + ): + r""" + Repeatedly runs the main hooks until all processes join; then, runs the post-hooks. + + Raises: + RuntimeError + If ``throw_on_early_termination=True``. + """ + if not self._enable or type: + return # propagate the exception directly if one was raised + + all_procs_joined = False + is_last_joiner = True + + i = 0 + WARN_THRESHOLD = 1000 + warnings.simplefilter("once") + + while not all_procs_joined: + if i > WARN_THRESHOLD: + warnings.warn( + "Detected uneven input skew of greater than " + f"{WARN_THRESHOLD}. This means that rank " + f"{self._rank} has at least {WARN_THRESHOLD} " + f"fewer inputs than other currently-active ranks. " + "This level of skew could lead to performance " + "degradation during training.", + stacklevel=2, + ) + # Shadow the all-reduce in non-joined processes + num_nonjoined_procs = self._get_num_nonjoined_procs() + if num_nonjoined_procs == 0: + all_procs_joined = True + else: + if self._throw_on_early_termination: + self._notify_procs_to_terminate() + + # Run main hooks + for join_hook in self._join_hooks: + join_hook.main_hook() + + is_last_joiner = False + i += 1 + + # Run post-hooks + for join_hook in self._join_hooks: + join_hook.post_hook(is_last_joiner) + + def _get_num_nonjoined_procs(self): + r"""Return the number of non-joined processes by shadowing an all-reduce in the non-joined processes.""" + num_nonjoined_procs = torch.zeros(1, device=self._device) + dist.all_reduce(num_nonjoined_procs, group=self._process_group) + return num_nonjoined_procs.item() + + def _notify_procs_to_terminate(self): + r"""Schedule an all-reduce to notify non-joined processes to terminate. + + Also raise a ``RuntimeError`` indicating that the current process has exhausted its inputs. + """ + ones = torch.ones(1, device=self._device) + dist.all_reduce(ones, group=self._process_group) + raise RuntimeError(f"Rank {self._rank} exhausted all inputs.") + + @staticmethod + def notify_join_context(joinable: Joinable): + r""" + Notifies the join context manager that the calling process has not yet joined. + + Then, if ``throw_on_early_termination=True``, checks if uneven inputs have been detected + (i.e. if one process has already joined) and throws an exception if so. + + This method should be called from a :class:`Joinable` object before + its per-iteration collective communications. For example, this should + be called at the beginning of the forward pass in + :class:`DistributedDataParallel`. + + Only the first :class:`Joinable` object passed into the context + manager performs the collective communications in this method, and + for the others, this method is vacuous. + + Arguments: + joinable (Joinable): the :class:`Joinable` object calling this + method. + + Returns: + An async work handle for the all-reduce meant to notify the context + manager that the process has not yet joined if ``joinable`` is the + first one passed into the context manager; ``None`` otherwise. + """ + if not hasattr(joinable, "_join_config"): + raise AssertionError( + f"Check that the {type(joinable)} constructor calls the " + "``Joinable`` constructor" + ) + + join_config = joinable._join_config + # First joinable is responsible for the collective communications + if not join_config.is_first_joinable or not join_config.enable: + return None + + device = joinable.join_device + process_group = joinable.join_process_group + + # Schedule an all-reduce to indicate that the caller has not yet joined + ones = torch.ones(1, device=device) + work = dist.all_reduce(ones, group=process_group, async_op=True) + + if join_config.throw_on_early_termination: + # Check if uneven inputs have been detected + zeros = torch.zeros(1, device=device) + dist.all_reduce(zeros, group=process_group) + should_throw = zeros.item() + if should_throw: + raise RuntimeError( + "Detected at least one rank that exhausted inputs. " + "Throwing across all ranks." + ) + return work diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..41fbc4cc1f21e2f5fb30fe33471d8e248397f4e3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/averagers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/averagers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2cce32df3016fea734ae36b2439696f8248f7093 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/averagers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/hierarchical_model_averager.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/hierarchical_model_averager.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5b476fdb31070c5dd786955303a6ede0f925b3b1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/hierarchical_model_averager.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..75ce00f0ede0873c8b746cc7cbdd4198a67c1936 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/averagers.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/averagers.py new file mode 100644 index 0000000000000000000000000000000000000000..5d669d4ea592250556ed5188b21ae265bb3b2c9c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/averagers.py @@ -0,0 +1,128 @@ +# mypy: allow-untyped-defs +import warnings +from abc import ABC, abstractmethod +from collections.abc import Iterable + +import torch +import torch.distributed as dist +import torch.distributed.algorithms.model_averaging.utils as utils +from torch.utils._typing_utils import not_none as _not_none + + +__all__ = ["ModelAverager", "PeriodicModelAverager"] + + +class ModelAverager(ABC): + r"""Base class for all model averagers. + + Args: + process_group: The process group to be used for all-reduce. + If ``None``, the default process group, which + is created by :func:`torch.distributed.init_process_group`, + will be used. (default: ``None``) + """ + + def __init__(self, process_group: dist.ProcessGroup | None = None): + self.process_group = ( + process_group if process_group is not None else _not_none(dist.group.WORLD) + ) + self.step = 0 + + @abstractmethod + def average_parameters(self, params): + raise NotImplementedError + + +class PeriodicModelAverager(ModelAverager): + r""" + Averages parameters periodically after the warm-up stage. + + This can be used for running `post-local SGD `_, + by running :class:`~torch.nn.DistributedDataParallel` (DDP) + using the subgroups created by :meth:`~torch.distributed.new_subgroups`. + + Args: + period (int): The number of steps per model averaging. + Usually the period should be greater than ``1`` to reduce the communication cost. + Otherwise, only DDP needs to be used. + warmup_steps (int): The number of warm-up steps. During this stage, + model averaging is skipped. + process_group: The process group to be used for all-reduce. + If ``None``, the default process group, which + is created by :func:`torch.distributed.init_process_group`, + will be used. (default: ``None``) + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> import torch + >>> import torch.distributed as dist + >>> import torch.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook as post_localSGD + >>> import torch.distributed.algorithms.model_averaging.averagers as averagers + >>> import torch.nn as nn + >>> + >>> dist.init_process_group("nccl", rank=rank, world_size=16) + >>> torch.cuda.set_device(rank) + >>> module = nn.Linear(1, 1, bias=False).cuda() + >>> model = nn.parallel.DistributedDataParallel( + >>> module, device_ids=[rank], output_device=rank + >>> ) + >>> # Register a post-localSGD communication hook. + >>> state = PostLocalSGDState(process_group=None, subgroup=None, start_localSGD_iter=100) + >>> model.register_comm_hook(state, post_localSGD_hook) + >>> + >>> # In the first 100 steps, run global gradient averaging like normal DDP at every step. + >>> # After 100 steps, run model averaging every 4 steps. + >>> # Note that ``warmup_steps`` must be the same as ``start_localSGD_iter`` used in ``PostLocalSGDState``. + >>> averager = averagers.PeriodicModelAverager(period=4, warmup_steps=100) + >>> for step in range(0, 200): + >>> optimizer.zero_grad() + >>> loss = loss_fn(output, labels) + >>> loss.backward() + >>> optimizer.step() + >>> # Will average model parameters globally every 4 steps. Thus, + >>> # inter-node communication only occurs every 4 iterations after + >>> # the initial ``warmup_steps`` period. + >>> averager.average_parameters(model.parameters()) + """ + + def __init__( + self, period, warmup_steps=0, process_group: dist.ProcessGroup | None = None + ): + super().__init__(process_group) + if warmup_steps < 0: + raise ValueError("Arg ``warmup_steps`` must be a non-negative number.") + self.warmup_steps = warmup_steps + if period < 1: + raise ValueError("Arg ``period`` must be a positive value.") + elif period == 1: + warnings.warn( + "When period is 1, no need to use model averaging because the communication cost " + "of all-reducing parameters will be no less than the cost of all-reducing gradients " + "by DistributedDataParallel in the backward pass. Therefore, only " + "DistributedDataParallel should be used for this case.", + stacklevel=2, + ) + self.period = period + + def average_parameters( + self, + params: Iterable[torch.nn.Parameter] | Iterable[dict[str, torch.nn.Parameter]], + ): + """ + Averages parameters or parameter groups of an optimizer if ``step`` is no less than ``warmup_steps``. + + Can be divided by ``period``, where ``step`` is increased by 1 + at each iteration in the training loop. + Args: + params: The parameters of a model or parameter groups of an optimizer. + + """ + if ( + self.step >= self.warmup_steps + and (self.step - self.warmup_steps) % self.period == 0 + ): + utils.average_parameters_or_parameter_groups( + params, _not_none(self.process_group) + ) + self.step += 1 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/hierarchical_model_averager.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/hierarchical_model_averager.py new file mode 100644 index 0000000000000000000000000000000000000000..584c3a89131e70b1487099c4f695d1968595a9d5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/hierarchical_model_averager.py @@ -0,0 +1,179 @@ +# mypy: allow-untyped-defs +# Copyright 2022 Cruise LLC +import logging +import warnings +from collections import OrderedDict +from collections.abc import Iterable + +import torch +import torch.distributed as dist +import torch.distributed.algorithms.model_averaging.averagers as averagers +import torch.distributed.algorithms.model_averaging.utils as utils + + +logger = logging.getLogger(__name__) + + +class HierarchicalModelAverager(averagers.ModelAverager): + r""" + Runs hierarchical model averaging (`hierarchical SGD `_). + + Process groups of different sizes are organized in a hierarchy, and they average parameters + by using different periods concurrently after the warm-up stage. + This is an extension of :class:`~torch.distributed.algorithms.model_averaging.averagers.PeriodicModelAverager` + that supports `post-local SGD `_, which essentially only supports + a two-level hierarchy: the intra-machine level and the global level, where the intra-machine + level is usually embedded in :meth:`~torch.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook`. + Similarly, the process groups within this class do not have such an intra-machine process + subgroup, which should be embedded by the post-local SGD communication hook instead. + + Args: + period_group_size_dict: An ordered dict mapping keys of model averaging period to + process group size, used for initializing process groups of + different sizes in a hierarchy to average parameters concurrently. + Particularly, at each iteration, there will be at most a single + process group that runs averaging -- the period of such group should + have the largest period which the current step can be divided by. + For example, if the dict has three keys: 2, 4, and 8, + then this means totally three process groups will be created to + average parameters every 2, 4, and 8 iterations, respectively. + At the 4th iteration, only the second process group will run + averaging, because the first process group should be a + subset of the second process group, and no need to execute the first + process group redundantly. + On the other hand, the third process group can only be triggered + every 8 iterations, so it will not be triggered at the 4th iteration. + warmup_steps (int): The number of warm-up steps. During this stage, model averaging is skipped. + process_group (ProcessGroup, optional): The overall process group containing all the processes that runs model averaging. + If ``None``, the default process group, which is created + by :func:`torch.distributed.init_process_group`, will be used. + (default: ``None``) + + Example:: + >>> # xdoctest: +SKIP('undefined rank') + >>> from collections import OrderedDict + >>> import torch + >>> import torch.distributed as dist + >>> from torch.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook import ( + >>> PostLocalSGDState, + >>> post_localSGD_hook, + >>> ) + >>> import torch.distributed.algorithms.model_averaging.hierarchical_model_averager as hierarchicalSGD + >>> import torch.nn as nn + >>> + >>> dist.init_process_group("nccl", rank=rank, world_size=16) + >>> torch.cuda.set_device(rank) + >>> module = nn.Linear(1, 1, bias=False).to(rank) + >>> model = nn.parallel.DistributedDataParallel( + >>> module, device_ids=[rank], output_device=rank + >>> ) + >>> # Register a post-localSGD communication hook. + >>> # Assume that each machine has 4 GPUs, then each intra-machine subgroup has a size of 4. + >>> subgroup, _ = dist.new_subgroups() + >>> state = PostLocalSGDState(process_group=None, subgroup=subgroup, start_localSGD_iter=100) + >>> model.register_comm_hook(state, post_localSGD_hook) + >>> + >>> # Average parameters among each group of 8 processes every 4 iterations, and among all + >>> # the 16 processes every 16 iterations. + >>> averager = hierarchicalSGD.HierarchicalModelAverager( + >>> period_group_size_dict=OrderedDict([(4, 8), (16, 16)]), warmup_steps=100) + >>> # Note that ``warmup_steps`` must be the same as ``start_localSGD_iter`` used in ``PostLocalSGDState``. + >>> # In the first 100 steps, run global gradient averaging like normal DDP at every step. + >>> # After 100 steps, run model averaging at two levels. + >>> for step in range(0, 200): + >>> optimizer.zero_grad() + >>> loss = loss_fn(output, labels) + >>> loss.backward() + >>> optimizer.step() + >>> # Average parameters after ``optimizer.step()``. + >>> # Thus, the inter-node communication only occurs periodically after ``warmup_steps``. + >>> averager.average_parameters(model.parameters()) + + .. warning :: + The last group size in the dict must be the size of the provided ``process_group``, + which indicates model averaging at the highest level of the hierarchy. + If ``process_group`` is not provided, then the last group size should be equal to the world size. + + .. warning :: + `HierarchicalModelAverager` is experimental and subject to change. + """ + + def __init__(self, period_group_size_dict=None, warmup_steps=0, process_group=None): + super().__init__(process_group) + if not period_group_size_dict: + raise ValueError("Arg ``period_group_size_dict`` must not be empty.") + self._periods = list(period_group_size_dict.keys()) + if self._periods[0] <= 0: + raise ValueError( + "The minimum period in arg ``period_group_size_dict`` must be a positive value." + ) + elif self._periods[-1] == 1: + warnings.warn( + "When the maximum period in arg ``period_group_size_dict`` is 1, " + "no need to use model averaging because the communication cost " + "of all-reducing parameters will be no less than the cost of all-reducing gradients " + "by DistributedDataParallel in the backward pass. Therefore, only " + "DistributedDataParallel should be used for this case.", + stacklevel=2, + ) + overall_group_size = dist.get_world_size(group=self.process_group) + if list(period_group_size_dict.values())[-1] != overall_group_size: + raise ValueError( + f"The last value in arg ``period_group_size_dict`` {list(period_group_size_dict.values())[-1]} " + f"must be equal to the size of arg ``process_group`` {overall_group_size}." + ) + + self.period_process_group_dict = OrderedDict() + logger.info("Model averaging hierarchy:") + for period, group_size in period_group_size_dict.items(): + logger.info( + "\tEach group that has %s processes average parameters every %s iterations, " + "if no higher-level averaging.", + group_size, + period, + ) + if group_size != overall_group_size: + self.period_process_group_dict[period], _ = dist.new_subgroups( + group_size=group_size, group=self.process_group + ) + else: + self.period_process_group_dict[period] = self.process_group + + if warmup_steps < 0: + raise ValueError("Arg ``warmup_steps`` must be a non-negative number.") + self.warmup_steps = warmup_steps + + def _find_process_group(self): + """ + Return a process group as the value of a ``period_process_group_dict`` entry. + + If ``step`` can be divided by multiple periods in the keys of ``period_process_group_dict``, + then the returned process group is the one corresponding to the largest period, + since this process group will be used for averaging parameters at this ``step``. + Returns ``None`` if not found. + """ + for period in reversed(self._periods): + if self.step % period == 0: + return self.period_process_group_dict[period] + return None + + def average_parameters( + self, + params: Iterable[torch.nn.Parameter] | Iterable[dict[str, torch.nn.Parameter]], + ): + """ + Averages parameters or parameter groups of an optimizer. + + Averaging only occurs if ``step`` is no less than ``warmup_steps`` + and it can be divided by a period in the keys of ``period_process_group_dict``, + where ``step`` is increased by 1 at each iteration in the training loop. + If ``step`` can be divided by multiple periods in the keys of ``period_process_group_dict``, + only the largest period is used, and the corresponding process group is used for averaging parameters. + Args: + params: The parameters of a model or parameter groups of an optimizer. + """ + if self.step >= self.warmup_steps: + group = self._find_process_group() + if group is not None: + utils.average_parameters_or_parameter_groups(params, group) + self.step += 1 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..6a61c036913edd6cf7fbcde6b77bc6ee5970065e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/algorithms/model_averaging/utils.py @@ -0,0 +1,86 @@ +# mypy: allow-untyped-defs +import itertools +from collections.abc import Iterable, Iterator + +import torch +import torch.distributed as dist + +# The two imports below are not always available depending on the +# USE_DISTRIBUTED compile flag. Make sure they raise import error +# if we're trying to use them. +from torch.distributed import group, ProcessGroup + + +__all__ = [ + "average_parameters", + "get_params_to_average", + "average_parameters_or_parameter_groups", +] + + +def average_parameters( + params: Iterator[torch.nn.Parameter], process_group: ProcessGroup +): + """ + Averages all the given parameters. + + For allreduce efficiency, all the parameters are flattened into a contiguous buffer. + Thus, it requires extra memory of the same size as the given parameters. + """ + group_to_use = process_group if process_group is not None else group.WORLD + # Do not update any parameter if not in the process group. + if dist._rank_not_in_group(group_to_use): + return + + params_it1, params_it2 = itertools.tee(params) + # If the input parameters have different data types, + # packing these parameters will trigger an implicit type up-casting. + # The original parameter data types will be restored during the subsequent unpacking. + flat_params = torch.cat([p.data.reshape(-1) for p in params_it1]) + flat_params /= dist.get_world_size(group_to_use) + # Make sure the allreduce will not conflict with any other ongoing process group. + if torch.accelerator.is_available(): + torch.accelerator.synchronize() + dist.all_reduce(flat_params, group=group_to_use) + + offset = 0 + for p in params_it2: + p.data = flat_params[offset : offset + p.numel()].view_as(p).type_as(p) + offset += p.numel() + + +def get_params_to_average( + params: Iterable[torch.nn.Parameter] | Iterable[dict[str, torch.nn.Parameter]], +): + """ + Return a list of parameters that need to average. + + This filters out the parameters that do not contain any gradients. + Args: + params: The parameters of a model or parameter groups of an optimizer. + """ + filtered_params = [] + for param in params: + if isinstance(param, torch.nn.Parameter): + # model.parameters() input + param_data = param + if param_data.grad is not None: + filtered_params.append(param_data) + elif isinstance(param, dict): + # optimizer.param_groups input + for param_data in param["params"]: + if param_data.grad is not None: + filtered_params.append(param_data) + else: + raise NotImplementedError( + f"Parameter input of type {type(param)} is not supported" + ) + return filtered_params + + +def average_parameters_or_parameter_groups( + params: Iterable[torch.nn.Parameter] | Iterable[dict[str, torch.nn.Parameter]], + process_group: ProcessGroup, +): + """Averages parameters of a model or parameter groups of an optimizer.""" + average_parameters(iter(get_params_to_average(params)), process_group) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/argparse_util.py b/.venv/lib/python3.12/site-packages/torch/distributed/argparse_util.py new file mode 100644 index 0000000000000000000000000000000000000000..c475eebf21273abb53ab99e3edcbdef18e9f0c8f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/argparse_util.py @@ -0,0 +1,104 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import os +from argparse import Action + + +class env(Action): + """ + Get argument values from ``PET_{dest}`` before defaulting to the given ``default`` value. + + For flags (e.g. ``--standalone``) + use ``check_env`` instead. + + .. note:: when multiple option strings are specified, ``dest`` is + the longest option string (e.g. for ``"-f", "--foo"`` + the env var to set is ``PET_FOO`` not ``PET_F``) + + Example: + :: + + parser.add_argument("-f", "--foo", action=env, default="bar") + + ./program -> args.foo="bar" + ./program -f baz -> args.foo="baz" + ./program --foo baz -> args.foo="baz" + PET_FOO="env_bar" ./program -f baz -> args.foo="baz" + PET_FOO="env_bar" ./program --foo baz -> args.foo="baz" + PET_FOO="env_bar" ./program -> args.foo="env_bar" + + parser.add_argument("-f", "--foo", action=env, required=True) + + ./program -> fails + ./program -f baz -> args.foo="baz" + PET_FOO="env_bar" ./program -> args.foo="env_bar" + PET_FOO="env_bar" ./program -f baz -> args.foo="baz" + """ + + def __init__(self, dest, default=None, required=False, **kwargs) -> None: + env_name = f"PET_{dest.upper()}" + default = os.environ.get(env_name, default) + + # ``required`` means that it NEEDS to be present in the command-line args + # rather than "this option requires a value (either set explicitly or default" + # so if we found default then we don't "require" it to be in the command-line + # so set it to False + if default: + required = False + + super().__init__(dest=dest, default=default, required=required, **kwargs) + + def __call__(self, parser, namespace, values, option_string=None): + setattr(namespace, self.dest, values) + + +class check_env(Action): + """ + Check whether the env var ``PET_{dest}`` exists before defaulting to the given ``default`` value. + + Equivalent to + ``store_true`` argparse built-in action except that the argument can + be omitted from the commandline if the env var is present and has a + non-zero value. + + .. note:: it is redundant to pass ``default=True`` for arguments + that use this action because a flag should be ``True`` + when present and ``False`` otherwise. + + Example: + :: + + parser.add_argument("--verbose", action=check_env) + + ./program -> args.verbose=False + ./program --verbose -> args.verbose=True + PET_VERBOSE=1 ./program -> args.verbose=True + PET_VERBOSE=0 ./program -> args.verbose=False + PET_VERBOSE=0 ./program --verbose -> args.verbose=True + + Anti-pattern (don't do this): + + :: + + parser.add_argument("--verbose", action=check_env, default=True) + + ./program -> args.verbose=True + ./program --verbose -> args.verbose=True + PET_VERBOSE=1 ./program -> args.verbose=True + PET_VERBOSE=0 ./program -> args.verbose=False + + """ + + def __init__(self, dest, default=False, **kwargs) -> None: + env_name = f"PET_{dest.upper()}" + default = bool(int(os.environ.get(env_name, "1" if default else "0"))) + super().__init__(dest=dest, const=True, default=default, nargs=0, **kwargs) + + def __call__(self, parser, namespace, values, option_string=None): + setattr(namespace, self.dest, self.const) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..6a52c36942e48e389a7e344abeb929febdb62c6c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__init__.py @@ -0,0 +1,66 @@ +from __future__ import annotations + +from typing import Any, TYPE_CHECKING + +import torch + + +if TYPE_CHECKING: + from types import TracebackType + + +def is_available() -> bool: + return hasattr(torch._C, "_dist_autograd_init") + + +if is_available() and not torch._C._dist_autograd_init(): + raise RuntimeError("Failed to initialize torch.distributed.autograd") + +if is_available(): + from torch._C._distributed_autograd import ( + _current_context, + _get_debug_info, + _get_max_id, + _init, + _is_valid_context, + _new_context, + _release_context, + _retrieve_context, + backward, + DistAutogradContext, + get_gradients, + ) + +__all__ = ["context", "is_available"] + + +class context: + """ + Context object to wrap forward and backward passes when using + distributed autograd. The ``context_id`` generated in the ``with`` + statement is required to uniquely identify a distributed backward pass + on all workers. Each worker stores metadata associated with this + ``context_id``, which is required to correctly execute a distributed + autograd pass. + + Example:: + >>> # xdoctest: +SKIP + >>> import torch.distributed.autograd as dist_autograd + >>> with dist_autograd.context() as context_id: + >>> t1 = torch.rand((3, 3), requires_grad=True) + >>> t2 = torch.rand((3, 3), requires_grad=True) + >>> loss = rpc.rpc_sync("worker1", torch.add, args=(t1, t2)).sum() + >>> dist_autograd.backward(context_id, [loss]) + """ + + def __enter__(self) -> int: + self.autograd_context = _new_context() + return self.autograd_context._context_id() + + def __exit__( + self, + exc_type: type[BaseException] | None, + exc_value: BaseException | None, + traceback: TracebackType | None, + ) -> None: + _release_context(self.autograd_context._context_id()) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ec668b443d2972838077cfba657ba2f4767accfb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/autograd/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/c10d_logger.py b/.venv/lib/python3.12/site-packages/torch/distributed/c10d_logger.py new file mode 100644 index 0000000000000000000000000000000000000000..1dfae5b92962f44f4dea3a3393cbcb6ae752999b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/c10d_logger.py @@ -0,0 +1,100 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import functools +import logging +from collections.abc import Callable +from typing import Any, TypeVar +from typing_extensions import ParamSpec + +import torch +import torch.distributed as dist +from torch.distributed.logging_handlers import _log_handlers +from torch.monitor import _WaitCounter + + +__all__: list[str] = [] + +_DEFAULT_DESTINATION = "default" + + +def _get_or_create_logger(destination: str = _DEFAULT_DESTINATION) -> logging.Logger: + logging_handler, log_handler_name = _get_logging_handler(destination) + logger = logging.getLogger(f"c10d-{log_handler_name}") + logger.setLevel(logging.DEBUG) + formatter = logging.Formatter( + "%(asctime)s %(filename)s:%(lineno)s %(levelname)s p:%(processName)s t:%(threadName)s: %(message)s" + ) + logging_handler.setFormatter(formatter) + logger.propagate = False + logger.addHandler(logging_handler) + return logger + + +def _get_logging_handler( + destination: str = _DEFAULT_DESTINATION, +) -> tuple[logging.Handler, str]: + log_handler = _log_handlers[destination] + log_handler_name = f"{type(log_handler).__name__}-{destination}" + return (log_handler, log_handler_name) + + +# pyrefly: ignore [unknown-name] +global _c10d_logger +_c10d_logger = _get_or_create_logger() + + +def _get_msg_dict(func_name, *args, **kwargs) -> dict[str, Any]: + if dist.is_initialized(): + group = kwargs.get("group") or kwargs.get("process_group") + msg_dict = { + "func_name": f"{func_name}", + "pg_name": f"{dist._get_process_group_name(kwargs.get('pg'))}", # type: ignore[arg-type] + "backend": f"{dist.get_backend(group)}", + "world_size": f"{dist.get_world_size()}", + "group_size": f"{dist.get_world_size(group)}", + "global_rank": f"{dist.get_rank()}", + "local_rank": f"{dist.get_rank(group)}", + } + if msg_dict["backend"] == "nccl": + nccl_version = torch.cuda.nccl.version() + msg_dict["nccl_version"] = ".".join(str(v) for v in nccl_version) + else: + msg_dict = { + "func_name": f"{func_name}", + } + return msg_dict + + +_T = TypeVar("_T") +_P = ParamSpec("_P") + + +def _exception_logger(func: Callable[_P, _T]) -> Callable[_P, _T]: + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _T: + try: + return func(*args, **kwargs) + except Exception as error: + msg_dict = _get_msg_dict(func.__name__, *args, **kwargs) + msg_dict["error"] = f"{error}" + _c10d_logger.debug(msg_dict) + raise + + return wrapper + + +def _time_logger(func: Callable[_P, _T]) -> Callable[_P, _T]: + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _T: + with _WaitCounter(f"pytorch.wait_counter.c10d.{func.__name__}").guard(): + func_return = func(*args, **kwargs) + return func_return + + return wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..06caa8adcb80f08ed1f4248a7cac62ddf9099c39 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__init__.py @@ -0,0 +1,22 @@ +from . import _extension +from .api import CheckpointException +from .default_planner import DefaultLoadPlanner, DefaultSavePlanner +from .filesystem import FileSystemReader, FileSystemWriter +from .hf_storage import HuggingFaceStorageReader, HuggingFaceStorageWriter +from .metadata import ( + BytesStorageMetadata, + ChunkStorageMetadata, + Metadata, + TensorStorageMetadata, +) +from .optimizer import load_sharded_optimizer_state_dict +from .planner import LoadPlan, LoadPlanner, ReadItem, SavePlan, SavePlanner, WriteItem +from .protocol import CheckpointableTensor +from .quantized_hf_storage import QuantizedHuggingFaceStorageReader + +# pyrefly: ignore [deprecated] +from .state_dict_loader import load, load_state_dict + +# pyrefly: ignore [deprecated] +from .state_dict_saver import async_save, save, save_state_dict +from .storage import StorageReader, StorageWriter diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0aa4328b6b1b2e74d892fcbd539bbf5e39e8283a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_executor.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_executor.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f828a1a9d9c7daedfe04c9c0c760dc4e454b7aae Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_executor.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_process_executor.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_process_executor.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6556d7ad3c780a234e1c14d581941c947a1c9975 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_process_executor.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_thread_executor.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_thread_executor.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..734fb4c0a4316e8344c7998574d0f157fac0fd5c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_async_thread_executor.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_checkpointer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_checkpointer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..67b446523a85b5d054135f1e880a921b6b48a13d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_checkpointer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_consolidate_hf_safetensors.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_consolidate_hf_safetensors.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..412bbf14b3c79190f3cfb3944e963c780a5a9e4f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_consolidate_hf_safetensors.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_save_plans.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_save_plans.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6073311bc41f970859e02903295fb507d2f36d4a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_save_plans.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_tensors.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_tensors.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2e9bc9290083795d29333500ddc7b7000f4ce1dc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_dedup_tensors.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_extension.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_extension.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8fa559132a4645160b92bb1c829d5c275acb77f7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_extension.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_fsspec_filesystem.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_fsspec_filesystem.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0a99b40ad583bd02ae5574ddafd5ad3baf10c2b6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_fsspec_filesystem.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_hf_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_hf_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d43c820621adcb785b49c8773c815a74e570eb7d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_hf_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_nested_dict.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_nested_dict.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..48979deaab225f7e1ca6df0aca7c4603a3bf614c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_nested_dict.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_pg_transport.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_pg_transport.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d1011a4babf0625f35b029ea6fe21e59ddf747c5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_pg_transport.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_sharded_tensor_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_sharded_tensor_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6dfe3e08a2c03dd77b6547ee5131a794a32c40c6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_sharded_tensor_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_state_dict_stager.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_state_dict_stager.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..19f69519754e44ee78ae93db9b789baea23c7948 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_state_dict_stager.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_storage_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_storage_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8f80c380caaf988398cc996d74983a037be890ae Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_storage_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_traverse.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_traverse.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..99fa3eb0daa19f79baef50ffde1834fe16d699bb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_traverse.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_version.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_version.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc2492d22a51e5abcd12cd722d0794b6943a5c6d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/_version.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e2c3766666dcc91f0452e4648463e76ffee8c691 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/default_planner.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/default_planner.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..947f7f21589b9379c8b5422dd4432c9b4b9753d0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/default_planner.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/filesystem.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/filesystem.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0a5b9ba7f66228bf879c2691e9253191d7f42301 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/filesystem.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/format_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/format_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fc4451bcd7d81b0e7d3310e42ab5a8d551d8bdbc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/format_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/hf_storage.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/hf_storage.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1bf44c777fd0b5158a88358e73b0a47f514b7222 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/hf_storage.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logger.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logger.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..40709694526e8be7d1ab4514e84fd7b32a58a1bd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logger.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logging_handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logging_handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a4d978c62afb7ccbb8a9f8178c6a70eda854eaae Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/logging_handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/metadata.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/metadata.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..01953278da5ae1c85bcb9c4d0025635448381c28 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/metadata.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/optimizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/optimizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bc89d867b16a224b06c86fed3ac0d5427ca816c9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/optimizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8cd45b3f18681699c29decb2f6e679bc4c202d91 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner_helpers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner_helpers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c50a18711a89f7a6855f230c7304d1447202e7d3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/planner_helpers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/protocol.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/protocol.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..11c687f0c44d5fc8507c867290adde4584cd5b8a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/protocol.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/quantized_hf_storage.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/quantized_hf_storage.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..62e9bb7343ef55de402cec04c4592530fcddb63b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/quantized_hf_storage.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/resharding.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/resharding.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8addb437dee9e471bfe1d0ed7c8ac62b0b730417 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/resharding.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/staging.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/staging.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f118cd703299eb5d8765fe76169138c2bf9a8479 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/staging.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fbb794f7b2556383876354d378cbc4638d9a6e38 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_loader.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_loader.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..256d74b53938c76c5d21bc9bd3eacc4c1c840812 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_loader.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_saver.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_saver.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b202e1d354cd6249655f5e2b7bf100b9aaac5d50 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/state_dict_saver.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/stateful.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/stateful.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e4045d52509bde92c71e19fd40a511177f2c49a8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/stateful.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/storage.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/storage.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ac5749ff4bc7def7a0c8b68a9ddc866224912c34 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/storage.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6e49bbe450a4309694c645a401a76c8f8e7e3253 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_executor.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_executor.py new file mode 100644 index 0000000000000000000000000000000000000000..75122b32a9e24e9500449d42e7ef0fcbb642651d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_executor.py @@ -0,0 +1,33 @@ +# pyre-strict +# mypy: allow-untyped-defs +import abc +import os +from concurrent.futures import Future + +import torch.distributed as dist +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE +from torch.distributed.checkpoint.planner import SavePlanner +from torch.distributed.checkpoint.storage import StorageWriter + + +class _AsyncCheckpointExecutor(abc.ABC): + @abc.abstractmethod + def execute_save( + self, + staging_future_or_state_dict: STATE_DICT_TYPE | Future[STATE_DICT_TYPE], + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Future: + """ + Execute the checkpoint save request asynchronously. + + This method is intended to be used as an abstraction for + implementing async checkpointing. The actual checkpoint save + operation is executed in a separate thread or process depending + on the implementation of this interface. + """ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_process_executor.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_process_executor.py new file mode 100644 index 0000000000000000000000000000000000000000..fd90399d8a5f1738747f2e2c2c533a8912c400e8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_process_executor.py @@ -0,0 +1,453 @@ +# pyre-strict +# mypy: allow-untyped-defs +import gc +import logging +import os +from concurrent.futures import Future, ThreadPoolExecutor +from dataclasses import dataclass +from enum import Enum +from typing import Any +from uuid import uuid4 + +import torch.distributed as dist +import torch.multiprocessing as mp +from torch.distributed import PrefixStore, TCPStore +from torch.distributed.checkpoint._async_executor import _AsyncCheckpointExecutor +from torch.distributed.checkpoint.logger import _dcp_method_logger, _init_logger +from torch.distributed.checkpoint.metadata import Metadata, STATE_DICT_TYPE +from torch.distributed.checkpoint.planner import SavePlanner +from torch.distributed.checkpoint.storage import StorageWriter +from torch.distributed.checkpoint.utils import _DistWrapper +from torch.distributed.elastic.agent.server.api import _get_fq_hostname +from torch.distributed.elastic.utils.distributed import get_free_port + + +logger = logging.getLogger() + + +class _CheckpointSaveProcessControlOpts(Enum): + INIT_COMPLETE = "init_complete" + TERMINATE = "terminate" + + +@dataclass(init=False, unsafe_hash=True) +class _CheckpointRequestIdentifier: + checkpoint_id: str | os.PathLike | None + uuid: str + + def __init__(self, checkpoint_id: str | os.PathLike | None): + self.checkpoint_id = checkpoint_id + self.uuid = str(uuid4()) + + +@dataclass +class _AsyncCheckpointRequest: + staged_state_dict: STATE_DICT_TYPE + checkpoint_request_id: _CheckpointRequestIdentifier + storage_writer: StorageWriter | None = None + planner: SavePlanner | None = None + no_dist: bool = False + use_collectives: bool = True + + +@dataclass(init=False) +class _ProcessGroupInitInfo: + local_rank: int + global_rank: int + world_size: int + tcp_store_master_addr: str + tcp_store_master_port: int + use_prefix_store: bool + disable_automatic_gc: bool + disable_manual_gc: bool + + def __init__(self, process_group: dist.ProcessGroup | None = None): + self.local_rank = dist.get_node_local_rank(fallback_rank=0) + self.global_rank = dist.get_rank(process_group) + self.world_size = dist.get_world_size(process_group) + self.use_prefix_store = os.environ.get("DCP_USE_PREFIX_STORE", "0") == "1" + self.disable_automatic_gc = ( + os.environ.get("DCP_DISABLE_AUTOMATIC_GC", "0") == "1" + ) + self.disable_manual_gc = os.environ.get("DCP_DISABLE_MANUAL_GC", "0") == "1" + + # Let coordinator rank find a port on the localhost. + # Broadcast the (master_addr, port) to all ranks; each rank in the + # checkpoint daemon process will use TCPStore (master_addr, port) + # for collective communication. + dist_wrapper: _DistWrapper = _DistWrapper( + group=process_group, + use_dist=True, + coordinator_rank=0, + ) + + def get_master_addr_and_port() -> tuple[str, int]: + if self.use_prefix_store: + master_addr = os.environ.get("MASTER_ADDR") + master_port = os.environ.get("MASTER_PORT") + if master_addr is None: + raise AssertionError("DCP needs MASTER_ADDR to use prefix store") + if master_port is None: + raise AssertionError("DCP needs MASTER_PORT to use prefix store") + master_port = int(master_port) + else: + master_addr = os.environ.get("MASTER_ADDR") + if master_addr is None: + master_addr = _get_fq_hostname() + master_port = get_free_port() + + return master_addr, master_port + + self.tcp_store_master_addr, self.tcp_store_master_port = dist_wrapper.broadcast( + step="get_master_addr_and_port", + map_fun=get_master_addr_and_port, + ) + + +class _AsyncCheckpointProcess: + def __init__( + self, + pg_init_info: _ProcessGroupInitInfo, + ): + self.ctx = mp.get_context("spawn") + self._process_pipe, child_end = self.ctx.Pipe() + + self._save_process = self.ctx.Process( + target=self._checkpointing_subprocess, + args=( + pg_init_info, + child_end, + ), + daemon=True, + ) + + self._save_process.start() + + # Close the parent's copy of child end after we pass it into the child, + # so the recv()s on it will fail-fast if the child process dies. + child_end.close() + + # Wait for the checkpoint background process to initialize. + # Using default GLOO init timeout. + response = self._wait_for_response(timeout=1800) + if not response == _CheckpointSaveProcessControlOpts.INIT_COMPLETE: + raise AssertionError(f"Expected INIT_COMPLETE response, got {response}") + + def __del__(self) -> None: + if self._save_process.is_alive(): + try: + logger.info("Terminating the checkpoint background process.") + self._send(_CheckpointSaveProcessControlOpts.TERMINATE) + self._save_process.join(timeout=5) + finally: + if self._save_process.is_alive(): + logger.warning( + "Checkpoint background process is still alive after termination request. Sending SIGTERM." + ) + self._save_process.terminate() + + def _send(self, data: Any) -> None: + self._process_pipe.send(data) + + def _wait_for_response(self, timeout: float | None = None) -> Any: + if not self._save_process.is_alive(): + logger.info("Checkpoint background process is dead calling join()...") + self._save_process.join() + raise RuntimeError( + f"Checkpoint background process is dead. Exit code: {self._save_process.exitcode}" + ) + + if timeout is not None and not self._process_pipe.poll(timeout=timeout): + raise RuntimeError( + f"Timed out after {timeout}s while waiting for response from checkpointer process pid: {self._save_process.pid}" + ) + + try: + response = self._process_pipe.recv() + except EOFError: + raise RuntimeError( # noqa: B904 + f"Checkpoint background process is dead. Exit code: {self._save_process.exitcode}" + ) + + if isinstance(response, BaseException): + raise response + + return response + + def save( + self, + staged_state_dict: STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Metadata: + # Create a unique identifier to locate requests/responses + # from the checkpoint daemon process. + checkpoint_request_id = _CheckpointRequestIdentifier(checkpoint_id) + async_cp_request = _AsyncCheckpointRequest( + staged_state_dict=staged_state_dict, + checkpoint_request_id=checkpoint_request_id, + storage_writer=storage_writer, + planner=planner, + no_dist=no_dist, + use_collectives=use_collectives, + ) + self._send(async_cp_request) + result = self._wait_for_response() + if not isinstance(result, Metadata): + raise AssertionError(f"Expected Metadata response, got {type(result)}") + return result + + @staticmethod + def _execute_save( + state_dict: STATE_DICT_TYPE, + *, + checkpoint_request_id: _CheckpointRequestIdentifier, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Metadata: + from torch.distributed.checkpoint.state_dict_saver import save + + metadata = save( + state_dict, + checkpoint_id=checkpoint_request_id.checkpoint_id, + storage_writer=storage_writer, + planner=planner, + no_dist=no_dist, + use_collectives=use_collectives, + ) + return metadata + + @staticmethod + def _checkpointing_subprocess( + pg_init_info: _ProcessGroupInitInfo, + parent_conn, + ) -> None: + # Phase 1: Process Group Initialization + # Only needs to execute once during the lifetime of the checkpoint background process. + try: + _init_logger(pg_init_info.global_rank) + + # Setup environment variables for process group initialization. + os.environ["TORCHELASTIC_USE_AGENT_STORE"] = "False" + os.environ["MASTER_ADDR"] = pg_init_info.tcp_store_master_addr + os.environ["MASTER_PORT"] = str(pg_init_info.tcp_store_master_port) + os.environ["LOCAL_RANK"] = str(pg_init_info.local_rank) + os.environ["RANK"] = str(pg_init_info.global_rank) + os.environ["WORLD_SIZE"] = str(pg_init_info.world_size) + + logger.info( + "Initializing dist.ProcessGroup in checkpoint background process on port %s", + pg_init_info.tcp_store_master_port, + ) + # NOTE: GLOO backend is enforced here. + if pg_init_info.use_prefix_store: + logger.info( + "Initializing dist.ProcessGroup in checkpoint background process with prefix store" + ) + store = PrefixStore( + "AsyncCheckpointProcess/", + TCPStore( + pg_init_info.tcp_store_master_addr, + pg_init_info.tcp_store_master_port, + ), + ) + dist.init_process_group( + backend=dist.Backend.GLOO, + store=store, + world_size=pg_init_info.world_size, + rank=pg_init_info.global_rank, + ) + else: + dist.init_process_group(backend=dist.Backend.GLOO) + dist.barrier() + + logger.info("Checkpoint background process is running...") + parent_conn.send(_CheckpointSaveProcessControlOpts.INIT_COMPLETE) + + if pg_init_info.disable_automatic_gc: + # Disable automatic garbage collection + # GC can optionally be called manually after each checkpoint + gc.disable() + logger.info("Disabled automatic garbage collection") + except BaseException as e: + logger.error( + f"Checkpoint background process failed during initialization: {e}" # noqa: G004 + ) + parent_conn.send(e) + return + + # Phase 2: Serving Loop + try: + first_request = True + while True: + logger.info("Waiting for checkpoint save request...") + obj = parent_conn.recv() + if ( + isinstance(obj, _CheckpointSaveProcessControlOpts) + and obj == _CheckpointSaveProcessControlOpts.TERMINATE + ): + logger.info("Terminating the checkpoint background process.") + return + if not isinstance(obj, _AsyncCheckpointRequest): + raise AssertionError( + f"Expected _AsyncCheckpointRequest, got {type(obj)}" + ) + logger.info( + f"Received async checkpoint request with id={obj.checkpoint_request_id.checkpoint_id}" # noqa: G004 + ) + + try: + response = _AsyncCheckpointProcess._execute_save( + obj.staged_state_dict, + checkpoint_request_id=obj.checkpoint_request_id, + storage_writer=obj.storage_writer, + planner=obj.planner, + no_dist=obj.no_dist, + use_collectives=obj.use_collectives, + ) + parent_conn.send(response) + logger.info( + f"Completed checkpoint save request for checkpoint_id={obj.checkpoint_request_id}" # noqa: G004 + ) + + # in theory this manual gc should not be needed as we shouldn't be leaking anything from checkpointing process + if ( + pg_init_info.disable_automatic_gc + and not pg_init_info.disable_manual_gc + ): + del obj + + collected_objects = gc.collect() + + logger.info( + f"Manual garbage collection completed - collected {collected_objects} objects." # noqa: G004 + ) + if first_request: + # Freeze GC to not check GC for large checkpoint save plans + # After freezing, subsequent gc.collect() calls will only scan + # NEW objects created after this point, not the frozen save plan + logger.info( + "First checkpoint request completed - freezing gc" + ) + gc.freeze() + first_request = False + except BaseException as e: + logger.error( + f"Checkpoint save failed for checkpoint_id={obj.checkpoint_request_id.checkpoint_id}: {e}" # noqa: G004 + ) + parent_conn.send(e) + # Continue serving loop - don't exit process + finally: + logger.info("Checkpoint background process is shutting down...") + dist.destroy_process_group() + parent_conn.close() + + +_CHECKPOINT_PROCESS: _AsyncCheckpointProcess | None = None + + +class _ProcessBasedAsyncCheckpointExecutor(_AsyncCheckpointExecutor): + def __init__(self) -> None: + self._executor = ThreadPoolExecutor(max_workers=1) + + @staticmethod + def _execute_save_impl( + *, + pg_init_info: _ProcessGroupInitInfo | None, + staging_future_or_state_dict: Future[STATE_DICT_TYPE] | STATE_DICT_TYPE, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Metadata: + global _CHECKPOINT_PROCESS + if _CHECKPOINT_PROCESS is None: + if pg_init_info is None: + raise AssertionError( + "pg_init_info must not be None when _CHECKPOINT_PROCESS is None" + ) + ckpt_kwargs = {} + if (ckpt_id := getattr(storage_writer, "checkpoint_id", None)) is not None: + ckpt_kwargs["checkpoint_id"] = ckpt_id + ckpt_kwargs["process_group"] = process_group + + @_dcp_method_logger(**ckpt_kwargs) + def create_checkpoint_daemon_process() -> None: + global _CHECKPOINT_PROCESS + # pyrefly: ignore [bad-argument-type] + _CHECKPOINT_PROCESS = _AsyncCheckpointProcess(pg_init_info=pg_init_info) + + create_checkpoint_daemon_process() + + if _CHECKPOINT_PROCESS is None: + raise AssertionError( + "_CHECKPOINT_PROCESS must not be None after initialization" + ) + staged_state_dict = ( + staging_future_or_state_dict.result() + if isinstance(staging_future_or_state_dict, Future) + else staging_future_or_state_dict + ) + return _CHECKPOINT_PROCESS.save( + staged_state_dict=staged_state_dict, + checkpoint_id=checkpoint_id, + storage_writer=storage_writer, + planner=planner, + no_dist=no_dist, + use_collectives=use_collectives, + ) + + def execute_save( + self, + staging_future_or_state_dict: Future[STATE_DICT_TYPE] | STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Future: + """ + NOTE: + + - Checkpoint process is implemented as a daemon process. + The AsyncCheckpointProcess' lifetime is tied to the lifetime of the + main process (e.g. trainer process). + + - The first call to execute_save_in_process() will initialize the checkpoint + daemon process. Subsequent async checkpoint requests will not need process + initialization. Therefore, the first async checkpoint request will take longer to complete. + + - Process initialization can have significant overhead, dominated by latency for all ranks to spawn + a background process + process group initialization in the background process. + """ + + global _CHECKPOINT_PROCESS + pg_init_info: _ProcessGroupInitInfo | None = None + if _CHECKPOINT_PROCESS is None: + # Find a port on coordinator rank and broadcast + # to all ranks. + pg_init_info = _ProcessGroupInitInfo(process_group) + + f: Future = self._executor.submit( + self._execute_save_impl, + pg_init_info=pg_init_info, + staging_future_or_state_dict=staging_future_or_state_dict, + checkpoint_id=checkpoint_id, + storage_writer=storage_writer, + planner=planner, + no_dist=no_dist, + use_collectives=use_collectives, + ) + f.add_done_callback(lambda f: self._executor.shutdown(wait=False)) + + return f diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_thread_executor.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_thread_executor.py new file mode 100644 index 0000000000000000000000000000000000000000..5039f4bac20541a898c64d7b81e666f9be4d63ec --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_async_thread_executor.py @@ -0,0 +1,70 @@ +# pyre-strict +# mypy: allow-untyped-defs +import os +from concurrent.futures import Future, ThreadPoolExecutor + +import torch.distributed as dist +from torch.distributed.checkpoint._async_executor import _AsyncCheckpointExecutor +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE +from torch.distributed.checkpoint.planner import SavePlanner +from torch.distributed.checkpoint.storage import StorageWriter + + +def save_wrapper( + staging_future_or_state_dict: Future[STATE_DICT_TYPE] | STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, +) -> Future: + from torch.distributed.checkpoint.state_dict_saver import save + + staged_dict = ( + staging_future_or_state_dict.result() + if isinstance(staging_future_or_state_dict, Future) + else staging_future_or_state_dict + ) + return save( + staged_dict, + checkpoint_id=checkpoint_id, + storage_writer=storage_writer, + planner=planner, + process_group=process_group, + no_dist=no_dist, + use_collectives=use_collectives, + ) + + +class _ThreadBasedAsyncCheckpointExecutor(_AsyncCheckpointExecutor): + def __init__(self) -> None: + self._executor = ThreadPoolExecutor( + max_workers=1, thread_name_prefix="AsyncCheckpointExecutor" + ) + + def execute_save( + self, + staging_future_or_state_dict: Future[STATE_DICT_TYPE] | STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, + ) -> Future: + f: Future = self._executor.submit( + save_wrapper, + staging_future_or_state_dict=staging_future_or_state_dict, + checkpoint_id=checkpoint_id, + storage_writer=storage_writer, + planner=planner, + process_group=process_group, + no_dist=no_dist, + use_collectives=use_collectives, + ) + f.add_done_callback(lambda f: self._executor.shutdown(wait=False)) + + return f diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_checkpointer.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_checkpointer.py new file mode 100644 index 0000000000000000000000000000000000000000..87ecb043db4f99f44d340710b2a1c7075294fcc0 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_checkpointer.py @@ -0,0 +1,103 @@ +from concurrent.futures import Future +from typing import Any + +import torch.distributed as dist +import torch.distributed.checkpoint.state_dict_loader as loader +import torch.distributed.checkpoint.state_dict_saver as saver +from torch.distributed.checkpoint.metadata import Metadata, STATE_DICT_TYPE +from torch.distributed.checkpoint.storage import ( + LoadPlanner, + SavePlanner, + StorageReader, + StorageWriter, +) + + +__all__: list[str] = [] + + +class _Checkpointer: + """This base class specifies a high level API for saving and loading + distributed `state_dict` 's. It provides an abstraction over the low-level APIs + provided by :py:mod:`torch.distributed.checkpoint.storage`, essentially calling + :py:meth: `torch.distributed.state_dict_saver.save` and + :py:meth: `torch.distributed.state_dict_loader.load` with the provided storage + readers and writers. + + .. warning:: + This feature is experimental and subject to removal/change. + + """ + + def __init__( + self, + storage_writer: StorageWriter, + storage_reader: StorageReader, + *, + process_group: dist.ProcessGroup | None = None, + coordinator_rank: int = 0, + no_dist: bool = False, + load_planner: LoadPlanner | None = None, + save_planner: SavePlanner | None = None, + ): + """Initializes the Checkpointer instance. + + Args: + storage_writer: Instance of StorageWrite used to perform writes. + storage_reader: StorageReader used to load data from. + process_group: ProcessGroup to be used for cross-rank synchronization. + coordinator_rank: Rank to use to coordinate the checkpoint. rank0 is used by default. + no_dist: If ``True``, distributed checkpoint will not load in SPMD style. (Default: ``False``) + loader_planner: Instance of LoadPlanner to use when loading. + save_planner: Instance of SavePlanner to use when saving. + """ + self.storage_writer = storage_writer + self.storage_reader = storage_reader + self.process_group = process_group + self.coordinator_rank = coordinator_rank + self.no_dist = no_dist + self.load_planner = load_planner + self.save_planner = save_planner + + def save( + self, + state_dict: STATE_DICT_TYPE, + ) -> Metadata: + """Calls :py:meth: `torch.distributed.state_dict_saver.save`. Utilizing values passed during initialization.""" + return saver.save( + state_dict, + self.storage_writer, + process_group=self.process_group, + coordinator_rank=self.coordinator_rank, + no_dist=self.no_dist, + planner=self.save_planner, + ) + + def async_save( + self, + state_dict: STATE_DICT_TYPE, + ) -> Future: + """ + Calls :py:meth: `torch.distributed.state_dict_saver._async_save`. Utilizing values passed during initialization. + + Returns: + Future: A future holding the resultant Metadata object from `save`. + """ + response = saver.async_save( + state_dict, + storage_writer=self.storage_writer, + process_group=self.process_group, + planner=self.save_planner, + ) + if not isinstance(response, Future): + raise AssertionError("response should be a Future instance") + return response + + def load(self, state_dict: dict[str, Any]) -> None: + """Calls :py:meth: `torch.distributed.state_dict_loader.load`. Utilizing values passed during initialization.""" + loader.load( + state_dict, + storage_reader=self.storage_reader, + process_group=self.process_group, + planner=self.load_planner, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_consolidate_hf_safetensors.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_consolidate_hf_safetensors.py new file mode 100644 index 0000000000000000000000000000000000000000..cb06faa962add4005a9ba94097d9eff9130c3ba1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_consolidate_hf_safetensors.py @@ -0,0 +1,749 @@ +# pyre-strict + +import concurrent.futures +import glob +import json +import logging +import math +import os +import struct +import time +from dataclasses import dataclass, field +from typing import Any + +import torch +from torch import distributed as dist +from torch.distributed.checkpoint._hf_utils import ( + _gen_file_name, + _get_dcp_custom_metadata, + _get_safetensors_file_metadata, + _metadata_fn, + DATA_OFFSETS_KEY, + DEFAULT_EXTRA_METADATA_KEY, + DTYPE_KEY, + SAVED_OFFSETS_KEY, + SHAPE_KEY, + SUFFIX, +) + + +logger: logging.Logger = logging.getLogger(__name__) + + +@dataclass +class _FqnData: + """ + Dataclass to store information about a tensor (identified by its fully qualified name). + + Attributes: + offset_in_file: Byte offset where this tensor's data begins in the output file + shape_in_file: Shape of the tensor in the output file + dtype_size: Size of the tensor's data type in bytes + dtype_str: String representation of the tensor's data type + """ + + offset_in_file: int = 0 + shape_in_file: list[int] = field(default_factory=list) + dtype_size: int = 0 + dtype_str: str = "" + + +@dataclass +class _OutputFileData: + """ + Dataclass to store information about an output safetensors file. + + Attributes: + metadata_size: Size of the metadata section in bytes + fqn_data: Dictionary mapping tensor names to their metadata + """ + + metadata_size: int = 0 + fqn_data: dict[str, _FqnData] = field(default_factory=dict) + + +@dataclass +class _InputFileData: + """ + Dataclass to store information about an input safetensors file. + + Attributes: + metadata_size: Size of the metadata section in bytes + metadata: Json metadata from the safetensors file + """ + + metadata_size: int = 0 + metadata: Any = None + + +def _parse_input_metadata( + input_files_data: dict[str, _InputFileData], + output_files_data: dict[str, _OutputFileData], +) -> None: + """ + Parse metadata from input safetensors files to determine the full tensor shapes and types. + + This function analyzes the metadata from all input files to determine the complete shape + of each tensor after consolidation. It updates the output_files_data with this information. + + Args: + input_files_data: dict of metadata from input safetensors files + output_files_data: Dictionary mapping output file paths to their metadata + + Raises: + ValueError: If no DCP custom metadata is found in a safetensors file + """ + + from safetensors.torch import _getdtype # type: ignore[import] + + # Dictionary to track the full size of each tensor across all shards + fqn_to_size_mapping: dict[str, tuple[list[int], str]] = {} + + for file_data in input_files_data.values(): + safetensors_metadata = file_data.metadata + dcp_sharding_info = _get_dcp_custom_metadata(safetensors_metadata) + if not dcp_sharding_info: + raise ValueError( + "No DCP custom metadata found in safetensors file. The file must be saved with DCP to be consolidated." + ) + + for key, val in safetensors_metadata.items(): + if key == DEFAULT_EXTRA_METADATA_KEY: + continue + + # Get the shape of this tensor shard and its offset in the full tensor + sizes = val[SHAPE_KEY] + offsets = dcp_sharding_info[key][SAVED_OFFSETS_KEY] + + if key not in fqn_to_size_mapping: + # First time seeing this tensor - calculate its full size by adding offsets to dimensions + cur_size = [size + offset for size, offset in zip(sizes, offsets)] + fqn_to_size_mapping[key] = (cur_size, val[DTYPE_KEY]) + else: + # We've seen this tensor before - update its size if this shard extends beyond current known dimensions + cur_size = fqn_to_size_mapping[key][0] + for i in range(len(sizes)): + cur_size[i] = max(cur_size[i], sizes[i] + offsets[i]) + + # Now that we know the full size of each tensor, populate the output file data + for fqn, tensor_info in fqn_to_size_mapping.items(): + tensor_size = tensor_info[0] + dtype_str = tensor_info[1] + for output_data in output_files_data.values(): + # Add this tensor to the output file if it's already assigned there + if fqn in output_data.fqn_data: + output_data.fqn_data[fqn] = _FqnData( + shape_in_file=tensor_size, + dtype_size=torch.finfo(_getdtype(dtype_str)).bits + // 8, # Convert bits to bytes + dtype_str=dtype_str, + ) + + +def _write_metadata( + output_files_data: dict[str, _OutputFileData], +) -> None: + """ + Write metadata to the beginning of each output safetensors file. + + This function writes the metadata section to each output file, including information + about tensor shapes, data types, and offsets. It also updates the offset_in_file + field for each tensor in the output_files_data. + + Args: + output_files_data: Dictionary mapping output file paths to their metadata + """ + # Process each output file + for file_path, output_data in output_files_data.items(): + with open(file_path, "wb") as f: + metadata = {} + curr_offset = 0 + + # Calculate offsets for each tensor in the file + for fqn, fqn_data in output_data.fqn_data.items(): + # Calculate the end offset by multiplying all dimensions and the data type size + end_offset = ( + curr_offset + + math.prod(fqn_data.shape_in_file) * fqn_data.dtype_size + ) + + # Store metadata for this tensor + metadata[fqn] = { + SHAPE_KEY: fqn_data.shape_in_file, + DTYPE_KEY: fqn_data.dtype_str, + DATA_OFFSETS_KEY: [ + curr_offset, + end_offset, + ], # Start and end byte offsets + } + # Store the offset for later use when writing the actual tensor data + fqn_data.offset_in_file = curr_offset + + # Update current offset for the next tensor + curr_offset = end_offset + + # Convert metadata to JSON and encode as bytes + json_metadata = json.dumps(metadata) + json_bytes = json_metadata.encode("utf-8") + + # Write the metadata size as an 8-byte unsigned integer (little-endian) + size_in_bytes = len(json_bytes) + header_len = struct.pack(" bytes: + """ + Read a specific byte range of tensor data from an open safetensors file. + + Args: + f: An open file object (handle) for the safetensors file + start_offset: Start offset of tensor data within the data section + end_offset: End offset of tensor data within the data section + metadata_size: Size of the metadata header + + Returns: + Raw tensor data as bytes + """ + absolute_start = metadata_size + start_offset + length = end_offset - start_offset + + f.seek(absolute_start) + return f.read(length) + + +def _process_output_file( + output_file: str, + output_data: _OutputFileData, + input_files_data: dict[str, _InputFileData], +) -> None: + """ + Process a single output file by writing tensor data from input files using direct reads. + + This function is designed to be run in parallel for different output files. + + Args: + output_file: Path to the output file + output_data: Metadata for the output file + input_files_data: Dictionary mapping input file paths to their metadata + """ + + sorted_tensors = sorted( + output_data.fqn_data.items(), key=lambda x: x[1].offset_in_file + ) + + file_handles = {} + dcp_metadata = {} + for safetensors_file, file_data in input_files_data.items(): + dcp_metadata[safetensors_file] = _get_dcp_custom_metadata(file_data.metadata) + + try: + # Open all input files for reading + for safetensors_file in input_files_data: + file_handles[safetensors_file] = open(safetensors_file, "rb") # noqa: SIM115 + + with open(output_file, "r+b") as output_stream: + output_stream.seek(0, os.SEEK_END) + # Process each tensor in sequential output order + for tensor_fqn, tensor_fqn_data in sorted_tensors: + full_tensor_mv = memoryview( + bytearray( + math.prod(tensor_fqn_data.shape_in_file) + * tensor_fqn_data.dtype_size + ) + ) + + # Process each input safetensors file + for safetensors_file in input_files_data: + file_metadata = input_files_data[safetensors_file].metadata + input_metadata_size = input_files_data[ + safetensors_file + ].metadata_size + + if tensor_fqn not in file_metadata: + continue + + metadata = file_metadata[tensor_fqn] + + data_offsets = metadata[DATA_OFFSETS_KEY] + + # Use explicit reads to fetch tensor data efficiently + data_to_write = _read_tensor_data( + file_handles[safetensors_file], + data_offsets[0], + data_offsets[1], + input_metadata_size, + ) + + # Get the offsets of this tensor shard within the full tensor + # pyrefly: ignore [unsupported-operation] + fqn_custom_metadata = dcp_metadata[safetensors_file][tensor_fqn] # type: ignore[index] + offsets_of_tensor_being_read = fqn_custom_metadata[ + SAVED_OFFSETS_KEY + ] # type: ignore[index] + + # Write this tensor shard to the appropriate position in the output file + _write_sub_tensor_to_file_optimized( + full_tensor_mv, + data_to_write, + tensor_fqn_data.dtype_size, # Size of each element in bytes + tensor_fqn_data.shape_in_file, # Full tensor shape + offsets_of_tensor_being_read, # Where this shard belongs in the full tensor + metadata[SHAPE_KEY], # Shape of this shard + ) + + output_stream.write(full_tensor_mv) + + finally: + for f in file_handles.values(): + f.close() + + +def _write_data( + input_files_data: dict[str, _InputFileData], + output_files_data: dict[str, _OutputFileData], + num_threads: int = 1, +) -> None: + """ + Write tensor data from input files to the output files using memory mapping. + + This function reads tensor data from each input file and writes it to the appropriate + position in the output files based on the tensor's offsets. When num_threads > 1, + the work is split across threads with each thread handling a different output file. + + Args: + input_files_data: Dictionary mapping input file paths to their metadata + output_files_data: Dictionary mapping output file paths to their metadata + num_threads: Number of threads to use for parallel processing + """ + if num_threads <= 1 or len(output_files_data) <= 1: + # Sequential processing + for output_file, output_data in output_files_data.items(): + _process_output_file(output_file, output_data, input_files_data) + else: + # Parallel processing with ThreadPoolExecutor + with concurrent.futures.ThreadPoolExecutor( + max_workers=min(num_threads, len(output_files_data)) + ) as executor: + futures = [] + for output_file, output_data in output_files_data.items(): + futures.append( + executor.submit( + _process_output_file, + output_file, + output_data, + input_files_data, + ) + ) + + # Wait for all futures to complete + for future in concurrent.futures.as_completed(futures): + # Handle any exceptions that might have occurred + try: + future.result() + except Exception as e: + print(f"Error processing output file: {e}") + raise + + +def _write_sub_tensor_to_file_optimized( + full_tensor_mv: memoryview, + sub_tensor_bytes: bytes, + element_size: int, + tensor_shape: list[int], + sub_tensor_offsets: list[int], + sub_tensor_shape: list[int], +) -> None: + """ + Optimized version that writes the maximum number of contiguous bytes possible. + + Uses a unified algorithm that calculates the maximum contiguous bytes that can be + written in each iteration and continues until the entire subtensor is written. + Handles all sharding patterns efficiently: + - Full sub-tensor at once for row-wise sharding + - Row-by-row for column-wise sharding + - Optimized chunks for other patterns + + Args: + full_tensor_mv: Buffer to write the full tensor to + sub_tensor_bytes: Raw tensor data as bytes + element_size: Size of each element in bytes + tensor_shape: Shape of the full tensor + sub_tensor_offsets: Starting offsets of the sub-tensor within the full tensor + sub_tensor_shape: Shape of the sub-tensor + """ + # Handle empty tensors + if not tensor_shape or not sub_tensor_shape: + return + + # Calculate tensor strides for efficient indexing + tensor_strides = [1] + for i in range(len(tensor_shape) - 1, 0, -1): + tensor_strides.insert(0, tensor_strides[0] * tensor_shape[i]) + + sub_tensor_strides = [1] + for i in range(len(sub_tensor_shape) - 1, 0, -1): + sub_tensor_strides.insert(0, sub_tensor_strides[0] * sub_tensor_shape[i]) + + total_elements = math.prod(sub_tensor_shape) + + elements_written = 0 + while elements_written < total_elements: + # Convert linear index to multi-dimensional indices + temp_idx = elements_written + indices = [] + for dim_size in reversed(sub_tensor_shape): + indices.append(temp_idx % dim_size) + temp_idx //= dim_size + indices.reverse() + + # Calculate maximum contiguous elements we can write from this position + max_contiguous = _calculate_max_contiguous_elements( + indices, sub_tensor_shape, tensor_shape + ) + + # Calculate source position in bytes + src_pos = sum(idx * stride for idx, stride in zip(indices, sub_tensor_strides)) + src_byte_offset = src_pos * element_size + + # Calculate destination position in bytes + dest_indices = [ + idx + offset for idx, offset in zip(indices, sub_tensor_offsets) + ] + dest_pos = sum( + idx * stride for idx, stride in zip(dest_indices, tensor_strides) + ) + dest_byte_offset = dest_pos * element_size + + # Write the contiguous chunk + bytes_to_write = max_contiguous * element_size + chunk_data = sub_tensor_bytes[ + src_byte_offset : src_byte_offset + bytes_to_write + ] + full_tensor_mv[dest_byte_offset : dest_byte_offset + bytes_to_write] = ( + chunk_data + ) + + elements_written += max_contiguous + + +def _calculate_max_contiguous_elements( + indices: list[int], + sub_tensor_shape: list[int], + tensor_shape: list[int], +) -> int: + """ + Calculate the maximum number of contiguous elements that can be written from current position. + + This determines the largest chunk by checking how elements are laid out in memory + and finding natural boundaries where contiguity breaks. + + Args: + indices: Current position indices in the sub-tensor + sub_tensor_shape: Shape of the sub-tensor being written + tensor_shape: Shape of the full tensor + + Raises: + ValueError: If input lists are empty, have mismatched lengths, or contain invalid values + """ + # Validate input lists are not empty + if not indices or not sub_tensor_shape or not tensor_shape: + raise ValueError("Input lists cannot be empty") + + # Validate all lists have the same length (same number of dimensions) + if not (len(indices) == len(sub_tensor_shape) == len(tensor_shape)): + raise ValueError( + f"All input lists must have the same length. Got indices: {len(indices)}, " + f"sub_tensor_shape: {len(sub_tensor_shape)}, tensor_shape: {len(tensor_shape)}" + ) + + # Validate indices are within bounds of sub_tensor_shape + for i, (idx, sub_dim) in enumerate(zip(indices, sub_tensor_shape)): + if idx >= sub_dim: + raise ValueError( + f"Index {idx} at dimension {i} is out of bounds for sub-tensor shape {sub_tensor_shape}" + ) + + # Validate sub_tensor dimensions don't exceed tensor dimensions + for i, (sub_dim, tensor_dim) in enumerate(zip(sub_tensor_shape, tensor_shape)): + if sub_dim > tensor_dim: + raise ValueError( + f"Sub-tensor dimension {sub_dim} at position {i} exceeds tensor dimension {tensor_dim}" + ) + + # Start with elements remaining in the last dimension + max_contiguous = sub_tensor_shape[-1] - indices[-1] + + # Check if we can extend across multiple dimensions + # We can write across dimension boundaries if we're writing complete "rows" + # and the layout in destination tensor maintains contiguity + + # For 2D case: check if we can write multiple complete rows + if len(sub_tensor_shape) >= 2: + # If we're at the start of a row and can write complete rows + if indices[-1] == 0: # At start of last dimension (column) + rows_remaining = sub_tensor_shape[-2] - indices[-2] # Rows left to write + + # Check if writing complete rows maintains contiguity in destination + # This is true for row-wise sharding or when sub-tensor spans full width + if sub_tensor_shape[-1] == tensor_shape[-1]: # Full width + max_contiguous = rows_remaining * sub_tensor_shape[-1] + + # For higher dimensions, check if we can extend further + if len(sub_tensor_shape) >= 3 and indices[-2] == 0: + # Check if we can write complete 2D slices + remaining_in_dim = sub_tensor_shape[-3] - indices[-3] + if ( + sub_tensor_shape[-1] == tensor_shape[-1] + and sub_tensor_shape[-2] == tensor_shape[-2] + ): + max_contiguous = ( + remaining_in_dim * sub_tensor_shape[-2] * sub_tensor_shape[-1] + ) + + return max_contiguous + + +def _write_overall_metadata_file( + output_dir: str, + output_files_data: dict[str, _OutputFileData], +) -> None: + """ + Write the overall metadata file that maps tensor names to their file locations. + + This creates a model.safetensors.index.json file that HuggingFace models use + to locate tensors across multiple files. + + Args: + output_dir: Directory where the metadata file will be written + output_files_data: Dictionary mapping output file paths to their metadata + """ + total_size = 0 + weight_map = {} + for output_path, value in output_files_data.items(): + for fqn, fqn_data in value.fqn_data.items(): + total_size += math.prod(fqn_data.shape_in_file) * fqn_data.dtype_size + weight_map[fqn] = os.path.basename(output_path) + + metadata_to_write: dict[str, Any] = {} + metadata_to_write["metadata"] = {"total_size": total_size} + metadata_to_write["weight_map"] = weight_map + + metadata_path = os.path.join(output_dir, f"{_metadata_fn}") + with open(metadata_path, "w") as metadata_file: + json.dump(metadata_to_write, metadata_file, indent=2) + + +def _consolidate_safetensors_files( + input_dir: str, + output_dir: str, + fqn_to_file_mapping: dict[str, str], + num_threads: int, +) -> dict[str, _OutputFileData]: + output_files_data: dict[str, _OutputFileData] = {} + # Create multiple output files based on the provided mapping + for fqn, filename in fqn_to_file_mapping.items(): + output_path = os.path.join(output_dir, filename) + + if output_path not in output_files_data: + output_files_data[output_path] = _OutputFileData(fqn_data={fqn: _FqnData()}) + else: + output_files_data[output_path].fqn_data[fqn] = _FqnData() + + # Find all safetensors files in the input directory + safetensors_files = glob.glob(os.path.join(input_dir, f"*{SUFFIX}")) + + # Read metadata from all input files + input_files_data: dict[str, _InputFileData] = {} + for safetensor_file in safetensors_files: + with open(safetensor_file, "rb") as f: + metadata, size = _get_safetensors_file_metadata(f) + input_files_data[safetensor_file] = _InputFileData( + metadata_size=size, metadata=metadata + ) + # Step 1: Parse metadata to determine tensor shapes and types + _parse_input_metadata(input_files_data, output_files_data) + + # Step 2: Write metadata headers to output files + _write_metadata(output_files_data) + # Step 3: Write actual tensor data from input files to output files + _write_data(input_files_data, output_files_data, num_threads) + + return output_files_data + + +def consolidate_safetensors_files( + input_dir: str, + output_dir: str, + fqn_to_index_mapping: dict[str, int], + num_threads: int = 1, +) -> None: + """ + Main function to consolidate sharded safetensors files into one or more output files. + + This function orchestrates the entire consolidation process: + 1. Sets up the output file structure based on the fqn_to_index_mapping + 2. Finds all safetensors files in the input directory + 3. Parses metadata from all input files + 4. Writes metadata to the output files + 5. Writes tensor data from input files to output files + 6. Writes overall model.index.safetensors.json file with weight map + + Args: + input_dir: Directory containing sharded safetensors files + output_dir: Directory where consolidated files will be written + fqn_to_index_mapping: Optional mapping of tensor names to output file indices. + If None, all tensors will be consolidated into a single file. + num_threads: Number of threads to use for parallel processing of saving data to output files. + """ + start_time = time.time() + logger.info( + "Consolidating safetensors files from %s to %s. Beginning at time %f", + input_dir, + output_dir, + start_time, + ) + + max_index = max(fqn_to_index_mapping.values()) + fqn_to_file_mapping = { + fqn: _gen_file_name(idx, max_index) for fqn, idx in fqn_to_index_mapping.items() + } + + output_files_data = _consolidate_safetensors_files( + input_dir, output_dir, fqn_to_file_mapping, num_threads + ) + + # Step 4: Write overall model.index.safetensors.json file with weight map + _write_overall_metadata_file(output_dir, output_files_data) + + logger.info("Done consolidating. Took %.2f secs.", time.time() - start_time) + + +def consolidate_safetensors_files_on_every_rank( + input_dir: str, + output_dir: str, + fqn_to_index_mapping: dict[str, int], + num_threads: int = 1, + process_group: dist.ProcessGroup | None = None, +) -> None: + """ + Consolidate sharded safetensors files across multiple ranks, with each rank handling a subset of output files. + + This function distributes the consolidation work by assigning output files to different ranks. + All tensors with the same index in fqn_to_index_mapping are processed by the same rank, + as they belong to the same output file. + + If process_group is provided, rank and world_size will be derived from it. Otherwise, + they will be automatically detected from the distributed environment if available. + + Args: + input_dir: Directory containing sharded safetensors files + output_dir: Directory where consolidated files will be written + fqn_to_index_mapping: Mapping of tensor names to output file indices + num_threads: Number of threads to use for parallel processing on each rank + process_group: PyTorch distributed process group (default: None, will use default group) + """ + + start_time = time.time() + # Derive rank and world_size from process_group or default distributed environment + if dist.is_available() and dist.is_initialized(): + rank = dist.get_rank(group=process_group) + world_size = dist.get_world_size(group=process_group) + else: + # Default to single process mode if distributed is not initialized + rank = 0 + world_size = 1 + logger.warning( + "Distributed environment not initialized. Running in single process mode." + ) + logger.info( + "Rank %d/%d: Consolidating safetensors files from %s to %s", + rank, + world_size, + input_dir, + output_dir, + ) + + # Find all unique indices in the mapping + unique_indices = set(fqn_to_index_mapping.values()) + + # Distribute indices across ranks + indices_for_this_rank = [] + for idx in unique_indices: + # Simple distribution: index % world_size == rank + if idx % world_size == rank: + indices_for_this_rank.append(idx) + + logger.info( + "Rank %d: Assigned %d output files out of %d total files", + rank, + len(indices_for_this_rank), + len(unique_indices), + ) + + # Filter the fqn_to_index_mapping to only include tensors for this rank + filtered_mapping = { + fqn: idx + for fqn, idx in fqn_to_index_mapping.items() + if idx in indices_for_this_rank + } + + output_files_data: dict[str, _OutputFileData] = {} + if filtered_mapping: + # Convert index mapping to filename mapping + max_index = max(unique_indices) + filtered_filename_mapping = {} + for fqn, idx in filtered_mapping.items(): + filename = _gen_file_name(idx, max_index) + filtered_filename_mapping[fqn] = filename + + # Call the existing consolidation function with the filtered mapping + output_files_data = _consolidate_safetensors_files( + input_dir=input_dir, + output_dir=output_dir, + fqn_to_file_mapping=filtered_filename_mapping, + num_threads=num_threads, + ) + + logger.info( + "Rank %d: Done consolidating. Processed %d unique indices in %.2f secs.", + rank, + len(indices_for_this_rank), + time.time() - start_time, + ) + + # Wait for all ranks to complete and gather output_files_data on rank 0 + if dist.is_available() and dist.is_initialized(): + gathered_output_files_data: list[dict[str, _OutputFileData]] | None = ( + [{} for _ in range(world_size)] if rank == 0 else None + ) + dist.gather_object( + output_files_data, + gathered_output_files_data, + dst=0, + group=process_group, + ) + + if rank == 0: + # Merge all output_files_data from all ranks + all_output_files_data: dict[str, _OutputFileData] = {} + if gathered_output_files_data is None: + raise AssertionError + for rank_data in gathered_output_files_data: + all_output_files_data.update(rank_data) + + _write_overall_metadata_file(output_dir, all_output_files_data) + logger.info("Rank 0: Wrote overall metadata file.") + logger.info("Total time taken: %.2f secs.", time.time() - start_time) + dist.barrier(group=process_group) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_save_plans.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_save_plans.py new file mode 100644 index 0000000000000000000000000000000000000000..acb81c41862852320cdc1d412ddaffdd48e73841 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_save_plans.py @@ -0,0 +1,65 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import dataclasses +from collections import defaultdict +from typing import TYPE_CHECKING + +from torch.distributed.checkpoint.planner import SavePlan, WriteItem + + +if TYPE_CHECKING: + from torch.distributed.checkpoint.metadata import MetadataIndex + +__all__ = ["dedup_save_plans"] + + +def dedup_save_plans( + all_plans: list[SavePlan], + save_to_lowest_rank: bool = False, +) -> list[SavePlan]: + """ + Removes duplicate entries from appearing on multiple SavePlans. For each duplicate across + a set of SavePlans, only the smallest SavePlan in terms of planned storage keeps the entry. + + Please note that this function does not modify the original SavePlans, but rather returns + """ + + # Map to query the plan indices that a write item is duplicated in + write_item_to_plan_indices: dict[MetadataIndex, set[int]] = defaultdict(set) + # Map to query the write item from its index + write_item_idx_to_write_item: dict[MetadataIndex, WriteItem] = {} + # Set of write item indices that are present in each plan + # After deduplication, this will be the set of write item indices that are present in the final plans + plan_to_item_indices: list[set[MetadataIndex]] = [ + {item.index for item in plan.items} for plan in all_plans + ] + + for plan_idx, plan in enumerate(all_plans): + for write_item in plan.items: + # map each write item to its plan + write_item_to_plan_indices[write_item.index].add(plan_idx) + write_item_idx_to_write_item[write_item.index] = write_item + plan_to_size = [0] * len(all_plans) + for write_item_idx, plan_indices in write_item_to_plan_indices.items(): + if save_to_lowest_rank: + select_plan_idx = min(plan_indices) + else: + select_plan_idx = min( + plan_indices, key=lambda plan_idx: plan_to_size[plan_idx] + ) + + write_item = write_item_idx_to_write_item[write_item_idx] + # Ignore the storage size of anything that is not a tensor, since + # we don't know how much storage they represent + plan_to_size[select_plan_idx] += write_item.tensor_storage_size() or 1 + for plan_idx in plan_indices - {select_plan_idx}: + plan_to_item_indices[plan_idx].discard(write_item_idx) + # Sanity check + if len(all_plans) != len(plan_to_item_indices): + raise AssertionError("len(all_plans) != len(plan_to_item_indices)") + # Create new plans with the updated write items post deduplication + return [ + dataclasses.replace( + plan, items=[item for item in plan.items if item.index in item_indexes] + ) + for plan, item_indexes in zip(all_plans, plan_to_item_indices) + ] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_tensors.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_tensors.py new file mode 100644 index 0000000000000000000000000000000000000000..c57b2e149106abbac66522aa571d1a462db4157d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_dedup_tensors.py @@ -0,0 +1,62 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import dataclasses +import logging +from typing import TYPE_CHECKING + +from torch.distributed.checkpoint.planner import SavePlan + + +if TYPE_CHECKING: + from torch.distributed.checkpoint.metadata import MetadataIndex + +__all__ = ["dedup_tensors"] + + +def init_logger() -> logging.Logger: + logger = logging.getLogger(__name__) + level = logging.INFO + logger.setLevel(level) + console = logging.StreamHandler() + formatter = logging.Formatter( + "%(asctime)s %(filename)s:%(lineno)s %(levelname)s p:%(processName)s t:%(threadName)s: %(message)s" + ) + console.setFormatter(formatter) + console.setLevel(level) + logger.addHandler(console) + logger.propagate = False + return logger + + +logger = init_logger() + + +# TODO add docstring for dedup_tensors +def dedup_tensors(all_plans: list[SavePlan]) -> list[SavePlan]: + all_plans = list(all_plans) + key_to_plan: dict[MetadataIndex, list[int]] = {} + for plan_idx, plan in enumerate(all_plans): + for write_item in plan.items: + key_to_plan.setdefault(write_item.index, []).append(plan_idx) + + replicated_items = {k: v for k, v in key_to_plan.items() if len(v) > 1} + + # Remove duplicates by always keeping the first entry. + # Compute the per-rank remove set. + plan_to_keys: dict[int, list[MetadataIndex]] = {} + for key, plans in replicated_items.items(): + for plan_idx in plans[1:]: + plan_to_keys.setdefault(plan_idx, []).append(key) + if len(plan_to_keys) > 0: + logger.info("Duplicate keys to remove: %s", plan_to_keys) + + for plan_idx, keys in plan_to_keys.items(): + key_set = set(keys) + # rewrite items and remove elements + new_items = [ + write_item + for write_item in all_plans[plan_idx].items + if write_item.index not in key_set + ] + all_plans[plan_idx] = dataclasses.replace(all_plans[plan_idx], items=new_items) + + return all_plans diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..8361362eb3a5ed5abae10d39c1db54c3e8739b46 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__init__.py @@ -0,0 +1,53 @@ +""" +Checkpoint functionality for machine learning models. + +This module provides classes for saving and loading model checkpoints in a distributed +training environment. It includes functionality for coordinating checkpoint operations +across multiple processes and customizing the checkpoint process through hooks. + +Key components: +- Checkpointer: Main class for orchestrating checkpoint operations (save, load) +- CheckpointWriter: Handles writing state dictionaries to storage +- CheckpointReader: Handles reading state dictionaries from storage read +- Barrier: Synchronization mechanism for distributed checkpointing +- RankInfo: Information about the current rank in a distributed environment +""" + +from .barriers import ( + Barrier, + BarrierConfig, + create_barrier_from_config, + TCPStoreBarrier, +) +from .builder import make_async_checkpointer, make_sync_checkpointer +from .checkpoint_reader import CheckpointReader +from .checkpoint_writer import CheckpointWriter, CheckpointWriterConfig, WriterHook +from .checkpointer import AsyncCheckpointer, Checkpointer, SyncCheckpointer +from .config import CheckpointerConfig +from .staging import CheckpointStager, CheckpointStagerConfig, DefaultStager +from .types import RankInfo, STATE_DICT +from .utils import wrap_future + + +__all__ = [ + "Barrier", + "TCPStoreBarrier", + "CheckpointReader", + "CheckpointWriter", + "CheckpointWriterConfig", + "WriterHook", + "Checkpointer", + "SyncCheckpointer", + "AsyncCheckpointer", + "CheckpointerConfig", + "BarrierConfig", + "create_barrier_from_config", + "CheckpointStager", + "CheckpointStagerConfig", + "DefaultStager", + "RankInfo", + "STATE_DICT", + "wrap_future", + "make_sync_checkpointer", + "make_async_checkpointer", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ebf0f758e5d1b16d340c11180548b1a168cee531 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/barriers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/barriers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eda22c6e926bdb32b17d6f750908e9ff236d47a8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/barriers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/builder.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/builder.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d2b76b702b28c497b5cbb9028edbd43851366704 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/builder.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_process.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_process.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..14e0692a8f87d1ed830fdcf7ab08bb9174db4ace Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_process.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_reader.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_reader.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9b1661bc9e59e0dcb59646503329246ec5064f19 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_reader.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_writer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_writer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..36b0df28228692959a0abea2a5796d86705fb1ab Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpoint_writer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpointer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpointer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc562b60799127e247f69cad0ca7e38904c6856f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/checkpointer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/config.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/config.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4a6fcb92408f45c5d2e167e81fff94bc302bf699 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/config.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/staging.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/staging.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dfa9e9a6699f6262233c1304e542bf83d49c9051 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/staging.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/types.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/types.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c79958ab4e525fa1a9b83be8cfe6952bae12c192 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/types.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9ba40550bdfc23ded28431e48b53622aa5000774 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/barriers.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/barriers.py new file mode 100644 index 0000000000000000000000000000000000000000..38a416c8029c96188d220ca01882143c93622320 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/barriers.py @@ -0,0 +1,267 @@ +""" +Barrier implementations for synchronizing distributed checkpoint operations. + +This module provides abstract and concrete barrier implementations that ensure +all ranks in a distributed training environment complete their checkpoint operations +before proceeding, which is essential for data consistency. +""" + +import abc +import logging +from collections import Counter +from dataclasses import dataclass, field +from datetime import timedelta +from typing import Any, Optional + +import torch.distributed as dist +import torch.distributed.elastic.utils.store as store_util + + +logger = logging.getLogger() + + +# Registry of barrier types +BARRIER_REGISTRY: dict[str, type] = {} + + +def register_barrier(barrier_class: type) -> type: + """Register a barrier class in the global registry.""" + if hasattr(barrier_class, "barrier_type"): + BARRIER_REGISTRY[barrier_class.barrier_type] = barrier_class + return barrier_class + + +@dataclass +class BarrierConfig: + """ + Configuration for barrier construction. + + This class provides a flexible way to configure different barrier implementations + with their specific constructor arguments. The barrier type will be looked up + from a registry and instantiated with rank_info and barrier_args. + + Attributes: + barrier_type: A string identifying the barrier type (e.g., "tcp_store"). + If None, no barrier will be used. + barrier_args: Dictionary of arguments to pass to the barrier constructor. + rank_info will be automatically injected as the first argument. + + Examples: + # No barrier + BarrierConfig() + + # TCPStore barrier + BarrierConfig( + barrier_type="tcp_store", + barrier_args={ + 'timeout_barrier_init_secs': 30, + 'barrier_prefix_list': ['checkpoint'], + 'use_checkpoint_barrier_tcpstore_libuv': False, + 'tcpstore_port': 12345, + 'master_address': 'localhost' + } + ) + """ + + barrier_type: str | None = None + barrier_args: dict[str, Any] = field(default_factory=dict) + + +def create_barrier_from_config( + barrier_config: BarrierConfig, +) -> Optional["Barrier"]: + """ + Create a barrier instance from BarrierConfig. + + Args: + barrier_config: Configuration for barrier construction. + + Returns: + Barrier instance or None if no barrier type is configured. + + Raises: + ValueError: If the barrier_type is not found in the registry. + """ + if barrier_config.barrier_type is None: + return None + + if barrier_config.barrier_type not in BARRIER_REGISTRY: + raise ValueError( + f"Unknown barrier type: {barrier_config.barrier_type}. " + f"Available types: {list(BARRIER_REGISTRY.keys())}" + ) + + barrier_class = BARRIER_REGISTRY[barrier_config.barrier_type] + return barrier_class(**barrier_config.barrier_args) + + +class Barrier(abc.ABC): + """ + Abstract base class for synchronization barriers. + + A barrier ensures that all ranks in a distributed environment reach a certain + point in execution before any rank proceeds further, which is essential for + coordinating operations like checkpointing across multiple processes. + """ + + @abc.abstractmethod + def __init__(self, **kwargs: dict[str, Any]): + """ + Initialize a barrier. + + Args: + **kwargs: Keyword arguments for specific barrier implementations. + Common arguments may include rank information, barrier prefixes, + timeout settings, and other barrier-specific configuration. + """ + # No implementation needed in the abstract base class + + @abc.abstractmethod + def execute_barrier(self) -> None: + """ + Execute a synchronization barrier. + + This method uses the barrier_prefix provided during initialization to + coordinate synchronization across processes. + """ + + +@register_barrier +class DistBarrier(Barrier): + """ + A barrier implementation using PyTorch's distributed barrier for synchronization. + + This barrier uses the built-in torch.distributed.barrier() function to coordinate + synchronization across multiple processes. It's simpler than TCPStoreBarrier but + requires an initialized process group. + """ + + barrier_type = "dist_barrier" + + def __init__( + self, + ) -> None: + """ + Initialize a DistBarrier. + + This barrier requires an initialized PyTorch distributed process group. + No additional arguments are needed as it uses the current process group. + + Raises: + AssertionError: If the distributed process group is not initialized. + """ + if not dist.is_initialized(): + raise AssertionError("DistBarrier requires an initialized process group.") + + def execute_barrier(self) -> None: + """ + Execute a synchronization barrier using the prefix provided during initialization. + """ + # Note: dist.barrier() doesn't support explicit timeouts + # The timeout is handled by the underlying implementation + dist.barrier() + + +@register_barrier +class TCPStoreBarrier(Barrier): + """ + A barrier implementation using PyTorch's TCPStore for synchronization. + + This barrier uses a TCP-based distributed key-value store to coordinate + synchronization across multiple processes. It uses a single TCP store + for all barrier operations, with different prefixes to distinguish between + different barrier types. + """ + + barrier_type = "tcp_store" + + def __init__( + self, + global_rank: int, + global_world_size: int, + barrier_prefix: str, + timeout_barrier_init_secs: int, + use_checkpoint_barrier_tcpstore_libuv: bool, + tcpstore_port: int, + master_address: str, + timeout_secs: int, + ): + """ + Initialize a TCPStoreBarrier. + + Args: + global_rank: The rank of the current process in the distributed environment. + global_world_size: The total number of processes in the distributed environment. + barrier_prefix: A string prefix to identify this specific barrier. + timeout_barrier_init_secs: Timeout in seconds for initializing the TCPStore. + use_checkpoint_barrier_tcpstore_libuv: Whether to use libuv for the TCPStore. + tcpstore_port: Port number for the TCPStore. + master_address: Address of the master node for the TCPStore. + timeout_secs: Maximum time in seconds to wait for all ranks to reach the barrier. + """ + logger.info( + "Initializing TCPStore master_address=%s tcpstore_port=%s rank=%s " + "world_size=%s barrier_prefix=%s timeout_barrier_init_secs=%s " + "use_checkpoint_barrier_tcpstore_libuv=%s timeout_secs=%s", + master_address, + tcpstore_port, + global_rank, + global_world_size, + barrier_prefix, + timeout_barrier_init_secs, + use_checkpoint_barrier_tcpstore_libuv, + timeout_secs, + ) + + # Counter collection to track barrier seq on a per barrier prefix basis. + self._tcp_store_barrier_seq: Counter = Counter() + self._barrier_prefix = barrier_prefix + + # Store rank and world size for barrier operations + self._global_rank = global_rank + self._global_world_size = global_world_size + self._timeout_secs = timeout_secs + + # Create a single TCP store for all barrier operations + self._tcp_store = dist.TCPStore( + master_address, + int(tcpstore_port), + world_size=self._global_world_size, + timeout=timedelta(seconds=timeout_barrier_init_secs), + is_master=(self._global_rank == 0), + ) + + def execute_barrier(self) -> None: + """ + Execute a synchronization barrier using the prefix provided during initialization. + + The implementation uses a sequence number that is incremented every time + a barrier is reached. The sequence number is per barrier prefix to allow + different barriers to operate concurrently. + """ + barrier_prefix = self._barrier_prefix + + logger.info( + "Executing barrier barrier_prefix=%s timeout_secs=%s", + barrier_prefix, + self._timeout_secs, + ) + + def _rank_key(rank: int) -> str: + return f"rank{rank}" + + # Track which barrier sequence this rank is joining. + self._tcp_store.set( + _rank_key(self._global_rank), + str(self._tcp_store_barrier_seq[barrier_prefix]), + ) + + # Execute barrier for that sequence number (for the specific prefix). + store_util.barrier( + store=self._tcp_store, + world_size=self._global_world_size, + key_prefix=( + barrier_prefix + str(self._tcp_store_barrier_seq[barrier_prefix]) + ), + ) + self._tcp_store_barrier_seq[barrier_prefix] += 1 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/builder.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/builder.py new file mode 100644 index 0000000000000000000000000000000000000000..2d351f6d30be9b6b2880505fcce52ce73698c7c6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/builder.py @@ -0,0 +1,174 @@ +""" +Factory functions for creating checkpointer instances with sensible defaults. + +This module provides high-level factory functions that simplify the creation +of checkpointer instances by automatically handling component initialization +and configuration with reasonable defaults. +""" + +from collections.abc import Callable +from typing import Any + +import torch.distributed as dist + +from .barriers import create_barrier_from_config +from .checkpoint_process import CheckpointProcess +from .checkpoint_reader import CheckpointReader +from .checkpoint_writer import CheckpointWriter, CheckpointWriterConfig, WriterHook +from .checkpointer import AsyncCheckpointer, SyncCheckpointer +from .config import CheckpointerConfig +from .staging import DefaultStager +from .types import RankInfo + + +def _get_default_rank_info() -> RankInfo: + """ + Get default rank information from the current distributed environment. + + Returns: + RankInfo: Rank information from the default process group if initialized, + otherwise single-rank fallback. + """ + if dist.is_initialized(): + return RankInfo( + global_world_size=dist.get_world_size(), + global_rank=dist.get_rank(), + ) + else: + # Single-rank fallback + return RankInfo(global_world_size=1, global_rank=0) + + +def default_subprocess_init_fn(*_: Any) -> None: + """Default subprocess initialization function (no-op).""" + + +def default_writer_init_fn(rank_info: RankInfo) -> CheckpointWriter: + """Default checkpoint writer initialization function.""" + return CheckpointWriter( + config=CheckpointWriterConfig(), + rank_info=rank_info, + ) + + +def make_sync_checkpointer( + config: CheckpointerConfig = CheckpointerConfig(), + rank_info: RankInfo | None = None, + commit_hook: WriterHook | None = None, +) -> SyncCheckpointer: + """ + Factory function to create a SyncCheckpointer instance with sensible defaults. + + This function creates a synchronous checkpointer with default components, automatically + detecting rank information from the default process group if available, and using the + provided component configurations. + + Args: + config: CheckpointerConfig containing component-specific configurations + (writer_config, staging_config, process_config). Defaults to CheckpointerConfig(). + rank_info: RankInfo for distributed training. Defaults to auto-detection from + the default PyTorch distributed process group if initialized, otherwise + falls back to single-rank (world_size=1, rank=0). + commit_hook: Optional hook for custom actions before and after checkpoint commits. + + Returns: + SyncCheckpointer: A configured synchronous checkpointer instance. + + Examples: + # Simplest usage - auto-detect rank, default config + checkpointer = make_sync_checkpointer() + + # Explicit rank configuration + checkpointer = make_sync_checkpointer( + rank_info=RankInfo(global_world_size=4, global_rank=0) + ) + + # Disable barrier + from .barriers import BarrierConfig + config = CheckpointerConfig(barrier_config=BarrierConfig(barrier_type=None)) + checkpointer = make_sync_checkpointer(config=config) + """ + if rank_info is None: + rank_info = _get_default_rank_info() + + reader = CheckpointReader( + rank_info=rank_info, + ) + + barrier = create_barrier_from_config(config.barrier_config) + + writer = CheckpointWriter( + config=config.writer_config, + rank_info=rank_info, + barrier=barrier, + commit_hook=commit_hook, + ) + + return SyncCheckpointer( + writer=writer, + reader=reader, + ) + + +def make_async_checkpointer( + config: CheckpointerConfig = CheckpointerConfig(), + rank_info: RankInfo | None = None, + subprocess_init_fn: Callable[..., None] = default_subprocess_init_fn, + subprocess_init_args: tuple[Any, ...] = (), + checkpoint_writer_init_fn: Callable[..., CheckpointWriter] = default_writer_init_fn, + checkpoint_writer_init_args: dict[str, Any] | None = None, +) -> AsyncCheckpointer: + """ + Factory function to create an AsyncCheckpointer instance with sensible defaults. + + This function creates an asynchronous checkpointer using the provided configuration, + automatically detecting rank information if not provided. + + Args: + config: CheckpointerConfig containing component-specific configurations. + rank_info: RankInfo for distributed training. Defaults to auto-detection. + subprocess_init_fn: Function to initialize the subprocess. Defaults to no-op. + subprocess_init_args: Arguments to pass to subprocess_init_fn. + checkpoint_writer_init_fn: Function to create CheckpointWriter instance. + checkpoint_writer_init_args: Arguments to pass to checkpoint_writer_init_fn. + + Returns: + AsyncCheckpointer: A configured asynchronous checkpointer instance. + + Examples: + # Create with default config + checkpointer = make_async_checkpointer() + + # Create with custom init functions + checkpointer = make_async_checkpointer( + subprocess_init_fn=my_subprocess_init_fn, + checkpoint_writer_init_fn=my_writer_init_fn + ) + """ + if rank_info is None: + rank_info = _get_default_rank_info() + + reader = CheckpointReader( + rank_info=rank_info, + ) + + checkpoint_stager = DefaultStager( + config=config.staging_config, + ) + + checkpoint_writer_init_args = checkpoint_writer_init_args or {} + + checkpoint_process = CheckpointProcess( + rank_info=rank_info, + config=config.process_config, + subprocess_init_fn=subprocess_init_fn, + subprocess_init_args=subprocess_init_args, + checkpoint_writer_init_fn=checkpoint_writer_init_fn, + checkpoint_writer_init_args=checkpoint_writer_init_args, + ) + + return AsyncCheckpointer( + checkpoint_stager=checkpoint_stager, + checkpoint_process=checkpoint_process, + reader=reader, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_process.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_process.py new file mode 100644 index 0000000000000000000000000000000000000000..c3b4b92165ce4d3bbc0eaa47680f7bf81eeec701 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_process.py @@ -0,0 +1,361 @@ +import logging +import os +import traceback +from collections.abc import Callable +from concurrent.futures import Future, ThreadPoolExecutor +from dataclasses import dataclass +from enum import Enum +from multiprocessing.connection import Connection +from typing import Any + +import torch.multiprocessing as mp +from torch.multiprocessing.spawn import ProcessExitedException + +from .checkpoint_writer import CheckpointWriter +from .types import RankInfo, STATE_DICT + + +logger = logging.getLogger(__name__) + + +@dataclass +class CheckpointProcessConfig: + """ + Configuration options for the CheckpointProcess. + + This class provides configuration options for the checkpoint process, + including initialization functions, timeouts, and writer configuration. + + Attributes: + subprocess_init_timeout_secs: Maximum time in seconds to wait for subprocess initialization. + subprocess_shutdown_timeout_secs: Maximum time in seconds to wait for subprocess shutdown. + """ + + subprocess_init_timeout_secs: int = 30 + subprocess_shutdown_timeout_secs: int = 60 + + +class RequestType(Enum): + PING = "ping" + WRITE_CHECKPOINT = "write_checkpoint" + TERMINATE_PROCESS = "exit" + + +@dataclass +class WorkerRequest: + """ + A dataclass for storing the command to be sent to the worker process. + Note: This relies on pickling to send the command to the worker process. Handle + backward compatibility accordingly. + """ + + request_type: RequestType + payload: dict[str, Any] + + +@dataclass +class WorkerResponse: + request_type: RequestType + success: bool + error_msg: str | None = None + payload: dict[str, Any] | None = None + + +class CheckpointProcess: + """ + A checkpoint writer that writes checkpoints to a remote process. + """ + + def __init__( + self, + rank_info: RankInfo, + config: CheckpointProcessConfig, + subprocess_init_fn: Callable[[Any], None], + subprocess_init_args: tuple[Any, ...], + checkpoint_writer_init_fn: Callable[..., CheckpointWriter], + checkpoint_writer_init_args: dict[str, Any], + ): + self._executor = ThreadPoolExecutor(max_workers=1) + self._rank_info = rank_info + self._config = config + self._subprocess_init_fn = subprocess_init_fn + self._subprocess_init_args = subprocess_init_args + self._checkpoint_writer_init_fn = checkpoint_writer_init_fn + self._checkpoint_writer_init_args = checkpoint_writer_init_args + self.process = None + self._parent_end: Connection | None = None + self._child_end: Connection | None = None + + self.process_creation_future = self._executor.submit( + self._create_subprocess, + config, + ) + + def _create_subprocess( + self, + config: CheckpointProcessConfig, + ) -> None: + logger.info( + "Creating checkpoint subprocess for rank %d", self._rank_info.global_rank + ) + + spawn_context = mp.get_context("spawn") + self._parent_end, child_end = spawn_context.Pipe() + + # Known workaround for https://github.com/pytorch/pytorch/issues/37377 + os.environ["MKL_SERVICE_FORCE_INTEL"] = "GNU" + + logger.debug("Spawning subprocess for rank_info=%s", self._rank_info) + self.process = mp.spawn( + fn=CheckpointProcess._subprocess, + args=( + self._rank_info, + child_end, + self._subprocess_init_fn, + self._subprocess_init_args, + self._checkpoint_writer_init_fn, + self._checkpoint_writer_init_args, + ), + nprocs=1, + join=False, + daemon=True, + ) + + # close the child end of the pipe so recv on it will fail + # fast when the child process is terminated unexpectedly. + child_end.close() + self._send( + request_type=RequestType.PING, + payload={}, + ) + + logger.debug( + "Waiting for checkpoint subprocess to initialize (timeout: %ds)", + config.subprocess_init_timeout_secs, + ) + + # wait for the timeout or a response from subprocess + if self._parent_end is None: + raise AssertionError("Parent end of pipe should be initialized") + if not self._parent_end.poll(timeout=config.subprocess_init_timeout_secs): + msg = f"Timed out after {config.subprocess_init_timeout_secs}s waiting for checkpoint subprocess to initialize" + logger.error(msg) + raise TimeoutError(msg) + + self._recv() + logger.info("Checkpoint subprocess initialized successfully") + + @staticmethod + def _subprocess( + sub_rank: int, + rank_info: RankInfo, + parent_pipe: Connection, + subprocess_init_fn: Callable[[Any], None], + subprocess_init_args: tuple[Any, ...], + checkpoint_writer_init_fn: Callable[..., CheckpointWriter], + checkpoint_writer_init_args: dict[str, Any], + ) -> None: + logger.debug( + "Checkpoint subprocess started for rank %d/%d (PID: %d)", + rank_info.global_rank, + rank_info.global_world_size, + os.getpid(), + ) + + if sub_rank != 0: + raise AssertionError("We need only one checkpointer per parent training") + request = WorkerRequest(request_type=RequestType.PING, payload={}) + + try: + # Calling initialize callback, so we can perform app-specific initialization of the subprocess. + subprocess_init_fn(*subprocess_init_args) + + # Initialize checkpoint writer - automatically include rank_info in init_args + writer_init_args = dict(checkpoint_writer_init_args) + if "rank_info" not in writer_init_args: + writer_init_args["rank_info"] = rank_info + checkpoint_writer = checkpoint_writer_init_fn(**writer_init_args) + + while True: + request = parent_pipe.recv() + + if request.request_type == RequestType.PING: + parent_pipe.send( + WorkerResponse(request_type=RequestType.PING, success=True) + ) + elif request.request_type == RequestType.WRITE_CHECKPOINT: + path = request.payload["path"] + logger.info("Writing checkpoint to %s", path) + + checkpoint_writer.write( + path=path, + state_dict=request.payload["state_dict"], + **request.payload["kwargs"], + ) + + logger.info("Checkpoint written successfully to %s", path) + parent_pipe.send( + WorkerResponse(RequestType.WRITE_CHECKPOINT, success=True) + ) + elif request.request_type == RequestType.TERMINATE_PROCESS: + logger.debug("Received termination request.") + parent_pipe.send( + WorkerResponse(RequestType.TERMINATE_PROCESS, success=True) + ) + logger.info("Subprocess terminated gracefully") + break + else: + error_msg = f"Unknown request type: {request.request_type}" + logger.error(error_msg) + raise ValueError(error_msg) + + except Exception as e: + error_text = traceback.format_exc() + logger.error( + "Exception in subprocess (%s): %s", type(e).__name__, error_text + ) + + # Communicating exception via the queue to the main process + parent_pipe.send( + WorkerResponse( + request_type=request.request_type, + success=False, + error_msg=error_text, + ) + ) + parent_pipe.close() + logger.exception("Subprocess terminated due to exception") + + def _send(self, request_type: RequestType, payload: dict[str, Any]) -> None: + try: + if self._parent_end is None: + raise AssertionError("Parent end of pipe should be initialized") + self._parent_end.send( + WorkerRequest( + request_type=request_type, + payload=payload, + ) + ) + except OSError as e: + error_msg = "Child process terminated unexpectedly" + logger.exception( + "Communication failed during %s request", request_type.value + ) + raise RuntimeError(error_msg) from e + + def _recv(self) -> dict[str, Any] | None: + try: + if self._parent_end is None: + raise AssertionError("Parent end of pipe should be initialized") + response = self._parent_end.recv() + if response.success is False: + error_msg = ( + f"Unexpected response from worker process: {response.error_msg}" + ) + logger.error(error_msg) + raise RuntimeError(error_msg) + return response.payload + except (EOFError, BrokenPipeError, ConnectionResetError) as e: + error_msg = f"Child process terminated unexpectedly: {e}" + logger.error(error_msg) + raise RuntimeError(error_msg) from e + + def write( + self, + state_dict: STATE_DICT | Future[STATE_DICT], + path: str, + **kwargs: Any, + ) -> Future[None] | None: + logger.debug("Waiting for subprocess initialization to complete") + + # wait until the process is started + self.process_creation_future.result() + + return self._executor.submit( + self._write, + state_dict, + path, + **kwargs, + ) + + def _write( + self, + state_dict: STATE_DICT | Future[STATE_DICT], + path: str, + **kwargs: Any, + ) -> None: + logger.debug("Starting checkpoint write to %s", path) + + # wait for staging state_dict to be available + if isinstance(state_dict, Future): + logger.debug("Waiting for state_dict Future to resolve") + sd = state_dict.result() + else: + sd = state_dict + + # Log state_dict info only if debug logging is enabled (performance-conscious) + if logger.isEnabledFor(logging.DEBUG): + if hasattr(sd, "keys"): + logger.debug("State_dict contains %d keys", len(sd.keys())) + + self._send( + request_type=RequestType.WRITE_CHECKPOINT, + payload={ + "state_dict": sd, + "path": path, + "kwargs": kwargs, + }, + ) + + logger.debug("Waiting for write completion response") + # wait for response + self._recv() + logger.debug("Checkpoint write to %s completed successfully", path) + + def close(self) -> None: + logger.debug( + "Closing CheckpointProcess for rank %d", self._rank_info.global_rank + ) + self._executor.shutdown(wait=True, cancel_futures=True) + + if self.process and self.process.processes[0].is_alive(): + subprocess_pid = self.process.processes[0].pid + # send graceful termination to sub process + try: + # pyrefly: ignore [missing-attribute] + self._parent_end.send( + WorkerRequest( + request_type=RequestType.TERMINATE_PROCESS, + payload={}, + ) + ) + except BrokenPipeError: + logger.warning( + "BrokenPipeError when sending termination request - subprocess (PID: %d) may have already terminated", + subprocess_pid, + ) + # subprocess terminated unexpectedly and below code will raise a + # ProcessExitedException. + + logger.debug( + "Waiting for subprocess to terminate gracefully (timeout: %ds)", + self._config.subprocess_shutdown_timeout_secs, + ) + + try: + if not self.process.join( + timeout=self._config.subprocess_shutdown_timeout_secs + ): + # graceful shutdown failed, kill the process. + logger.warning( + "Subprocess (PID: %d) did not terminate gracefully within %ds, killing it", + subprocess_pid, + self._config.subprocess_shutdown_timeout_secs, + ) + self.process.processes[0].kill() + logger.info("Subprocess killed forcefully") + except ProcessExitedException: + logger.exception("ProcessExitedException during subprocess termination") + raise + + logger.debug("CheckpointProcess closed successfully") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_reader.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_reader.py new file mode 100644 index 0000000000000000000000000000000000000000..aaf705352cf172a8f0fa81d5e8c97f1dfde8bb8d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_reader.py @@ -0,0 +1,223 @@ +""" +Checkpoint reader functionality for machine learning models. + +This module provides classes for reading checkpoints from storage, including +determining checkpoint layout and configuring the reader. +""" + +import logging +import os +from itertools import zip_longest +from pathlib import Path +from typing import Any + +import torch +from torch._subclasses.fake_tensor import FakeTensorMode + +from .types import RankInfo, STATE_DICT + + +logger = logging.getLogger(__name__) + + +class CheckpointReader: + """ + Handles reading state dictionaries from storage. + + This class is responsible for reading model state dictionaries from storage according + to the specified checkpoint layout. It supports synchronization barriers to ensure + all ranks in a distributed setting complete their checkpoint operations. + """ + + def __init__( + self, + rank_info: RankInfo, + ): + """ + Initialize a CheckpointReader. + + Args: + rank_info: Information about the current rank in a distributed setting. + """ + + self._rank_info = rank_info + + def read( + self, + path: str, + state_dict: STATE_DICT | None = None, + *, + map_location: Any = None, + **kwargs: dict[str, Any], + ) -> tuple[STATE_DICT, list[str]]: + """ + Reads a state dictionary from storage. + + Args: + path (str): The path from which to read the checkpoint. + map_location (Any): Device mapping function or device name for relocating tensors. + **kwargs: Additional keyword arguments passed to torch.load. + + Returns: + STATE_DICT: The loaded state dictionary. + list[str]: List of missing keys. + """ + logger.debug( + "Reading checkpoint from %s for rank %s", + path, + self._rank_info.global_rank, + ) + + dir_path = Path(path) + file_path = dir_path / f"checkpoint_{self._rank_info.global_rank}.pt" + + # Check if the file exists + if not os.path.exists(file_path): + logger.error("Checkpoint file not found at %s", file_path) + raise FileNotFoundError(f"Checkpoint file not found at {file_path}") + + if state_dict is None: + result: tuple[STATE_DICT, list[str]] = ( + torch.load(file_path, map_location=map_location), + [], + ) + else: + result = self._partial_read( + file_path, state_dict, map_location=map_location, **kwargs + ) + logger.debug("Successfully read checkpoint file from %s", file_path) + return result + + def _partial_read( + self, + file_path: Path, + state_dict: STATE_DICT, + *, + map_location: Any = None, + **kwargs: dict[str, Any], + ) -> tuple[STATE_DICT, list[str]]: + """ + Reads only the keys present in state_dict from the checkpoint file. + + This method optimizes checkpoint loading by only loading the tensors that + are actually needed, based on the keys present in the input state_dict. + This can significantly reduce memory usage and loading time for large checkpoints + when only a subset of the model needs to be loaded. + + Args: + file_path (str): The path to the checkpoint file. + state_dict (STATE_DICT): The state dictionary containing keys to load. + map_location (Any): Device mapping function or device name for relocating tensors. + **kwargs: Additional keyword arguments passed to torch.load. + + Returns: + tuple[STATE_DICT, list[str]]: The updated state dictionary with loaded values and a list of missing keys. + """ + + with FakeTensorMode(): + metadata_dict = torch.load(file_path, map_location=map_location) + + missing_keys = [] + + with open(file_path, "rb") as file: + # Helper function to load tensor data from file + def load_tensor( + target: torch.Tensor | None, source: torch.Tensor, full_key: str + ) -> torch.Tensor: + if target is not None and ( + target.size() != source.size() or target.dtype != source.dtype + ): + raise RuntimeError( + f"Target tensor size={target.size()} dtype={target.dtype} does not match " + f"source tensor size={source.size()} dtype={source.dtype} for key {full_key}" + ) + + tensor_offset = source.untyped_storage()._checkpoint_offset + + if tensor_offset is None: + raise AssertionError( + "checkpoint_offset for tensor in torch serialized file is not set. This could " + "happen if the checkpoint was saved with an older version of PyTorch. " + "Please make sure that the checkpoint was saved with Pytorch 2.7 or later." + ) + + tensor_len = source.nelement() * source.element_size() + file.seek( + tensor_offset + source.element_size() * int(source.storage_offset()) + ) + if target is None: + target = torch.empty( + source.size(), dtype=source.dtype, device=source.device + ) + + buffer = file.read(tensor_len) + cpu_tensor = torch.frombuffer(buffer, dtype=source.dtype) + tensor = cpu_tensor.view(source.size()) + target.copy_(tensor) + return target + + # Helper function to recursively process nested structures + def process_value( + target_value: Any, source_value: Any, key_path: str + ) -> Any: + source_type = type(source_value) + if source_type is torch._subclasses.fake_tensor.FakeTensor: + source_type = torch.Tensor + if target_value is not None and not isinstance( + target_value, source_type + ): + raise RuntimeError( + f"Target value {key_path} is set to {type(target_value)}, but source value is {type(source_value)}" + ) + if isinstance(source_value, torch.Tensor): + return load_tensor(target_value, source_value, key_path) + elif isinstance(source_value, dict): + if target_value is None: + # create a new map with all the keys present in source_value + target_value = dict.fromkeys(source_value.keys()) + + # pyrefly: ignore [missing-attribute] + for key in list(target_value.keys()): + current_path = f"{key_path}.{key}" if key_path else key + if key in source_value: + target_value[key] = process_value( + target_value[key], source_value[key], current_path + ) + else: + missing_keys.append(current_path) + + return target_value + elif isinstance(source_value, list): + if target_value is None: + target_value = [None] * len(source_value) + result = [] + for i, (target_item, source_item) in enumerate( + zip_longest(target_value, source_value, fillvalue=None) + ): + current_path = f"{key_path}[{i}]" if key_path else f"[{i}]" + result.append( + process_value(target_item, source_item, current_path) + ) + return result + else: + return source_value + + # Start recursive processing from the root of the state dictionary + updated_state_dict = process_value(state_dict, metadata_dict, "") + + if missing_keys: + if len(missing_keys) > 10: + logger.warning( + "Missing %s keys from checkpoint: %s... (and %s more)", + len(missing_keys), + missing_keys[:10], + len(missing_keys) - 10, + ) + else: + logger.warning( + "Missing %s keys from checkpoint: %s", + len(missing_keys), + missing_keys, + ) + + return updated_state_dict, missing_keys diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_writer.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_writer.py new file mode 100644 index 0000000000000000000000000000000000000000..3c2cc9044d82333b7bb18fcf042c49a627b86895 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpoint_writer.py @@ -0,0 +1,163 @@ +""" +Checkpoint writer functionality for machine learning models. + +This module provides classes for writing checkpoints to storage, including +determining checkpoint layout, configuring the writer, and defining hooks +for custom actions during the checkpoint writing process. +""" + +import abc +import logging +import os +from concurrent.futures import Future +from dataclasses import dataclass +from pathlib import Path +from typing import Any + +import torch + +from .barriers import Barrier +from .types import RankInfo, STATE_DICT + + +logger = logging.getLogger(__name__) + + +class WriterHook(abc.ABC): + """ + Abstract base class for checkpoint commit hooks. + + A commit hook provides callbacks that are executed before and after a checkpoint + is committed to storage. This allows for custom actions to be performed at specific + points in the checkpoint writing process, such as metadata updates, cleanup operations, + or notifications. + """ + + @abc.abstractmethod + def pre_commit(self, path: str, **kwargs: dict[str, Any]) -> None: + """ + Performs actions before committing the checkpoint. + """ + + @abc.abstractmethod + def post_commit(self, path: str, **kwargs: dict[str, Any]) -> None: + """ + Performs actions after committing the checkpoint. + """ + + +@dataclass +class CheckpointWriterConfig: + """ + Configuration options for the CheckpointWriter. + + Attributes: + write_barrier_timeout_secs: Maximum time in seconds to wait for all ranks + to reach the checkpoint barrier before timing out. Default is 600 seconds. + """ + + write_barrier_timeout_secs: int = 600 + + +class CheckpointWriter: + """ + Handles writing state dictionaries to storage. + + This class is responsible for writing model state dictionaries to storage according + to the specified checkpoint layout. It supports synchronization barriers to ensure + all ranks in a distributed setting complete their checkpoint operations. + """ + + def __init__( + self, + config: CheckpointWriterConfig, + rank_info: RankInfo, + barrier: Barrier | None = None, + commit_hook: WriterHook | None = None, + ): + """ + Initialize a CheckpointWriter. + + Args: + config: Configuration options for the checkpoint writer. + rank_info: Information about the current rank in a distributed setting. + barrier: Optional synchronization barrier for distributed checkpointing. + Note: The barrier should be initialized with the appropriate barrier_prefix + and timeout_secs parameters. + commit_hook: Optional hook for custom actions before and after checkpoint commits. + """ + + self._config = config + self._rank_info = rank_info + self._commit_hook = commit_hook + self._barrier = barrier + + def write( + self, + path: str, + state_dict: STATE_DICT, + **kwargs: dict[str, Any], + ) -> Future[None] | None: + """ + Writes the state_dict to storage. + + Args: + path (str): The path to write the checkpoint to. + state_dict (STATE_DICT): The state_dict to write. + **kwargs: Additional keyword arguments passed to hooks. + + Returns: + Optional[Future[None]]: A future for tracking the write operation, if applicable. + """ + logger.debug( + "Writing checkpoint to %s for rank %s", + path, + self._rank_info.global_rank, + ) + dir_path = Path(path) + full_path = dir_path / f"checkpoint_{self._rank_info.global_rank}.pt" + os.makedirs( + os.path.dirname(full_path), + exist_ok=True, + ) + torch.save(state_dict, full_path) + logger.debug("Successfully saved checkpoint file to %s", full_path) + + # Execute pre-commit hook if available + commit_hook = self._commit_hook + if commit_hook is not None: + logger.debug("Executing pre-commit hook for %s", path) + commit_hook.pre_commit(path, **kwargs) + + # Wait for all ranks to finish writing if barrier is available + barrier = self._barrier + if barrier is not None: + logger.info( + "Waiting for all ranks at barrier with timeout %ss", + self._config.write_barrier_timeout_secs, + ) + barrier.execute_barrier() + logger.info("All ranks passed barrier") + else: + logger.info("No barrier configured, skipping synchronization") + + # Execute commit hook if available + if commit_hook is not None: + logger.debug("Executing commit hook for %s", path) + commit_hook.post_commit(path, **kwargs) + + logger.info( + "Successfully wrote checkpoint to %s for rank %s", + path, + self._rank_info.global_rank, + ) + return None + + def close(self) -> None: + """ + Close the writer and release any resources. + + This is a no-op for the base CheckpointWriter but may be overridden + by subclasses that need to perform cleanup. + """ + logger.debug("Closing checkpoint writer") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpointer.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpointer.py new file mode 100644 index 0000000000000000000000000000000000000000..5f79a431e16d5cf3eefddd70b9e300336ff8d6b8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/checkpointer.py @@ -0,0 +1,341 @@ +import abc +import logging +from concurrent.futures import Future +from typing import Any, TypeVar + +from .checkpoint_process import CheckpointProcess +from .checkpoint_reader import CheckpointReader +from .checkpoint_writer import CheckpointWriter +from .staging import CheckpointStager +from .types import STATE_DICT +from .utils import wrap_future + + +logger = logging.getLogger(__name__) + +LOG_INTERVAL = 60 +T = TypeVar("T") + + +class Checkpointer(abc.ABC): + """ + WARNING: This class is experimental, and is created to validate certain ideas, + and is subjected to change or deprecation and we strong discourage any usages at + this time. + + Abstract base class that defines the API for checkpointing. + + This class defines the interface for coordinating the writing and loading of model + state dictionaries to and from storage. It provides abstract methods to save and load model states + with support for both synchronous and asynchronous operations. + + Concrete implementations of this class must implement all the abstract methods. + """ + + @abc.abstractmethod + def save( + self, + path: str, + state_dict: STATE_DICT, + **kwargs: dict[str, Any], + ) -> tuple[Future, Future] | None: + """ + Save a state dictionary to storage. + + Args: + path: The path where the checkpoint should be saved. + state_dict: The state dictionary to save. + **kwargs: Additional keyword arguments to pass to the writer. + + Returns: + For synchronous implementations: None + For asynchronous implementations: tuple of (stage_future, write_future) + representing the staging and writing operations. + """ + + @abc.abstractmethod + def load( + self, + path: str, + state_dict: STATE_DICT | None = None, + *, + default_map_location: Any = None, + strict: bool = False, + **kwargs: dict[str, Any], + ) -> STATE_DICT: + """ + Load a state dictionary from storage. + + Args: + path: The path from which to load the checkpoint. + state_dict: Optional state dictionary to update with loaded values. + If provided, only keys in this dictionary will be loaded. + default_map_location: Device mapping function or device name for relocating tensors. + strict: If True, raises an error when there are missing keys in the checkpoint. + **kwargs: Additional keyword arguments to pass to the reader. + + Returns: + The loaded state dictionary. + """ + + @abc.abstractmethod + def close(self) -> None: + """ + Close the checkpointer and release any resources. + + This method should be called when the checkpointer is no longer needed to ensure + proper cleanup of resources. + """ + + +class SyncCheckpointer(Checkpointer): + """ + Synchronous implementation of Checkpointer. + + This class coordinates the writing and loading of model state dictionaries to and from storage + using only synchronous operations. It provides a simple, efficient interface for checkpoint + operations without async overhead. + + Attributes: + _writer: CheckpointWriter for writing state dictionaries to storage. + _reader: CheckpointReader for reading state dictionaries from storage. + + Example: + checkpointer = SyncCheckpointer(writer=writer, reader=reader) + checkpointer.save(state_dict, path) + loaded_state_dict = checkpointer.load(path) + """ + + def __init__( + self, + writer: CheckpointWriter, + reader: CheckpointReader, + ): + """ + Initialize a synchronous checkpointer. + + Args: + writer: CheckpointWriter for writing checkpoints to storage. + reader: CheckpointReader for reading checkpoints from storage. + """ + self._writer = writer + self._reader = reader + + def save( + self, + path: str, + state_dict: STATE_DICT, + **kwargs: dict[str, Any], + ) -> tuple[Future, Future] | None: + """ + Save a state dictionary to storage synchronously. + + Args: + path: The path where the checkpoint should be saved. + state_dict: The state dictionary to save. + **kwargs: Additional keyword arguments to pass to the writer. + + Returns: + Always returns None as operations are synchronous. + + Example: + checkpointer.save("/path/to/checkpoint", state_dict) + """ + logger.debug("Saving checkpoint synchronously to %s", path) + self._writer.write(path, state_dict, **kwargs) + return None + + def load( + self, + path: str, + state_dict: STATE_DICT | None = None, + *, + default_map_location: Any = None, + strict: bool = False, + **kwargs: dict[str, Any], + ) -> STATE_DICT: + """ + Load a state dictionary from storage. + + Args: + path: The path from which to load the checkpoint. + state_dict: Optional state dictionary to update with loaded values. + If provided, only keys in this dictionary will be loaded. + default_map_location: Device mapping function or device name for relocating tensors. + strict: If True, raises an error when there are missing keys in the checkpoint. + **kwargs: Additional keyword arguments to pass to the reader. + + Returns: + The loaded state dictionary. + + Raises: + RuntimeError: If strict=True and there are missing keys in the checkpoint. + FileNotFoundError: If the checkpoint file is not found. + """ + logger.info("Loading checkpoint from %s", path) + + loaded_state_dict, missing_keys = self._reader.read( + path=path, + state_dict=state_dict, + map_location=default_map_location, + **kwargs, + ) + if strict and missing_keys is not None and missing_keys != []: + raise RuntimeError(f"Checkpoint at {path} is missing keys: {missing_keys}") + return loaded_state_dict + + def close(self) -> None: + """ + Close the checkpointer and release any resources. + + This method should be called when the checkpointer is no longer needed to ensure + proper cleanup of resources. + """ + self._writer.close() + logger.info("SyncCheckpointer closed") + + +class AsyncCheckpointer(Checkpointer): + """ + Asynchronous implementation of Checkpointer. + + This class coordinates the writing and loading of model state dictionaries to and from storage + using asynchronous operations for saving. It provides efficient async checkpoint operations + with staging and background writing capabilities. + + Attributes: + _reader: CheckpointReader for reading state dictionaries from storage. + _checkpoint_stager: Stager for async operations. + _checkpoint_process: Process for async operations. + _write_future: Future representing the ongoing async write operation. + + Example: + checkpointer = AsyncCheckpointer( + reader=reader, + checkpoint_stager=stager, + checkpoint_process=process + ) + stage_future, write_future = checkpointer.save(state_dict, path) + # ... do other work ... + write_future.result() # Wait for completion + """ + + def __init__( + self, + checkpoint_stager: CheckpointStager, + checkpoint_process: CheckpointProcess, + reader: CheckpointReader, + ): + """ + Initialize an asynchronous checkpointer. + + Args: + checkpoint_stager: Stager for async operations. + checkpoint_process: Process for async operations. + reader: CheckpointReader for reading checkpoints from storage. + """ + self._reader = reader + self._checkpoint_stager = checkpoint_stager + self._checkpoint_process = checkpoint_process + self._write_future: Future[Any] | None = None + + def save( + self, + path: str, + state_dict: STATE_DICT, + **kwargs: Any, + ) -> tuple[Future, Future] | None: + """ + Save a state dictionary to storage asynchronously. + + Args: + path: The path where the checkpoint should be saved. + state_dict: The state dictionary to save. + **kwargs: Additional keyword arguments to pass to the stager and writer. + + Returns: + A tuple of (stage_future, write_future) representing the staging and writing operations. + + Example: + stage_future, write_future = checkpointer.save("/path/to/checkpoint", state_dict) + # ... do other work ... + write_future.result() # Wait for completion + """ + logger.info( + "Initiating checkpoint save to %s. Will wait for prev checkpoints to complete.", + path, + ) + # Wait for previous checkpoint ops to finish and verify they are successful + if self._write_future is not None: + self._write_future.result() + + logger.debug("Starting state dictionary staging") + staging_result = self._checkpoint_stager.stage( + state_dict=state_dict, + **kwargs, + ) + + logger.debug("Starting checkpoint write to %s", path) + self._write_future = self._checkpoint_process.write( + staging_result, path, **kwargs + ) + logger.info("Checkpoint save to %s initiated", path) + + # Return futures for the staging and writing operations + if self._write_future is not None: + return wrap_future(staging_result), self._write_future + else: + # This should not happen since we just assigned _write_future above + raise RuntimeError("Write future is unexpectedly None") + + def load( + self, + path: str, + state_dict: STATE_DICT | None = None, + *, + default_map_location: Any = None, + strict: bool = False, + **kwargs: Any, + ) -> STATE_DICT: + """ + Load a state dictionary from storage. + + Loading is always performed synchronously, even in AsyncCheckpointer. + + Args: + path: The path from which to load the checkpoint. + state_dict: Optional state dictionary to update with loaded values. + If provided, only keys in this dictionary will be loaded. + default_map_location: Device mapping function or device name for relocating tensors. + strict: If True, raises an error when there are missing keys in the checkpoint. + **kwargs: Additional keyword arguments to pass to the reader. + + Returns: + The loaded state dictionary. + + Raises: + RuntimeError: If strict=True and there are missing keys in the checkpoint. + FileNotFoundError: If the checkpoint file is not found. + """ + logger.info("Loading checkpoint from %s", path) + + loaded_state_dict, missing_keys = self._reader.read( + path=path, + state_dict=state_dict, + map_location=default_map_location, + **kwargs, + ) + if strict and missing_keys is not None and missing_keys != []: + raise RuntimeError(f"Checkpoint at {path} is missing keys: {missing_keys}") + return loaded_state_dict + + def close(self) -> None: + """ + Close the checkpointer and release any resources. + + This method should be called when the checkpointer is no longer needed to ensure + proper cleanup of async resources. + """ + self._checkpoint_stager.close() + self._checkpoint_process.close() + logger.info("AsyncCheckpointer closed") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/config.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/config.py new file mode 100644 index 0000000000000000000000000000000000000000..a81156e3929cac9edb13135b925c8096dc4e702a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/config.py @@ -0,0 +1,44 @@ +""" +Configuration classes for checkpointer construction. + +This module provides configuration dataclasses that consolidate all +configuration options needed to construct checkpointers. +""" + +from dataclasses import dataclass, field + +from .barriers import BarrierConfig +from .checkpoint_process import CheckpointProcessConfig +from .checkpoint_writer import CheckpointWriterConfig +from .staging import CheckpointStagerConfig + + +@dataclass +class CheckpointerConfig: + """ + Configuration class for checkpointer construction. + + This class consolidates the core component configuration options needed to construct + a checkpointer, providing a clean separation of concerns where each component + manages its own configuration. + + Attributes: + writer_config: Configuration options for the checkpoint writer component. + barrier_config: Configuration for barrier construction and arguments. + staging_config: Configuration options for the async staging component. + process_config: Configuration options for the async checkpoint process component. + + """ + + writer_config: CheckpointWriterConfig = field( + default_factory=CheckpointWriterConfig + ) + barrier_config: BarrierConfig = field(default_factory=BarrierConfig) + + # Below configs are used for async checkpointing + staging_config: CheckpointStagerConfig = field( + default_factory=CheckpointStagerConfig + ) + process_config: CheckpointProcessConfig = field( + default_factory=CheckpointProcessConfig + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/staging.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/staging.py new file mode 100644 index 0000000000000000000000000000000000000000..076bc113b054a171d57ac7814773d550e0ea376f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/staging.py @@ -0,0 +1,216 @@ +""" +Experimental staging module for PyTorch Distributed Checkpointing. + +This module provides advanced staging capabilities for checkpoints including: +- Asynchronous staging using ThreadPoolExecutor +- Pinned memory allocation for faster CPU-GPU transfers +- Shared memory support for multi-process scenarios +- Non-blocking CUDA operations with stream synchronization +- Caching of frequently used storages for efficient memory management +- Automatic resource cleanup and memory management + +Classes: + CheckpointStager: Abstract base class defining the staging interface + StagingOptions: Configuration dataclass for staging behavior + DefaultStager: Default implementation with comprehensive staging features +""" + +import abc +from concurrent.futures import Future, ThreadPoolExecutor +from dataclasses import dataclass +from typing import Any, TypeVar + +import torch +from torch.distributed.checkpoint._state_dict_stager import StateDictStager + +from .types import STATE_DICT + + +T = TypeVar("T") + + +class CheckpointStager(abc.ABC): + """ + Abstract base class for checkpoint staging implementations. + + CheckpointStager defines the interface that all staging implementations + must follow. Staging is the process of offloading state dictionaries + for async checkpointing. + """ + + @abc.abstractmethod + def stage( + self, + state_dict: STATE_DICT, + **kwargs: Any, + ) -> STATE_DICT | Future[STATE_DICT]: + """ + Stage a state dictionary for checkpointing. + + Args: + state_dict: The state dictionary to stage + **kwargs: Additional staging parameters + + Returns: + Either a staged state dictionary (synchronous) or a Future + that will resolve to the staged state dictionary (asynchronous) + """ + + @abc.abstractmethod + def close(self) -> None: + """ + Clean up all resources used by the stager. + """ + + +@dataclass +class CheckpointStagerConfig: + """ + Configuration options for checkpoint staging behavior. + + Attributes: + use_pinned_memory (bool): Enable pinned memory allocation for faster + CPU-GPU transfers. Requires CUDA to be available. Default: True + use_shared_memory (bool): Enable shared memory for multi-process + scenarios. Useful when multiple processes need access to the + same staged data. Default: True + use_async_staging (bool): Enable asynchronous staging using a + background thread pool. Allows overlapping computation with + staging operations. Requires CUDA. Default: True + use_non_blocking_copy (bool): Use non-blocking device memory + copies with stream synchronization. Improves performance by + allowing CPU work to continue during GPU transfers. Default: True + + Note: + CUDA-dependent features will raise exception if CUDA is not available. + """ + + use_pinned_memory: bool = True + use_shared_memory: bool = True + use_async_staging: bool = True + use_non_blocking_copy: bool = True + + +class DefaultStager(CheckpointStager): + """ + DefaultStager provides a full-featured staging implementation that combines + multiple optimization techniques for efficient checkpoint preparation. + + The staging process works as follows: + 1. State dictionary is submitted for staging (sync or async) + 2. Tensors are copied from GPU to optimized CPU storage + 3. CUDA operations are synchronized if non-blocking copies are used + 4. Staged state dictionary is returned or made available via Future + + NOTE: state_dict should be deep-copyable object as staging will create a + copy of it. + + Usage Patterns: + # Synchronous staging + stager = DefaultStager(CheckpointStagerConfig(use_async_staging=False)) + staged_dict = stager.stage(state_dict) + stager.close() + + # Asynchronous staging + stager = DefaultStager(CheckpointStagerConfig(use_async_staging=True)) + future = stager.stage(state_dict) + # ... do other work ... + staged_dict = future.result() + stager.close() + + # Context manager pattern (recommended) + with DefaultStager(config) as stager: + result = stager.stage(state_dict) + # Automatic cleanup on exit + + Performance Considerations: + - Async staging provides best performance when model computation + can overlap with staging operations + - Pinned memory improves CPU-GPU transfer speeds but uses more memory + - Shared memory allows efficient IPC to checkpoint process + - Non-blocking copies reduce GPU idle time during memory transfers + + Thread Safety: + DefaultStager is not thread-safe. Each thread should use its own + instance, or external synchronization should be provided. + """ + + def __init__( + self, + config: CheckpointStagerConfig = CheckpointStagerConfig(), + ): + self._config = config + self._state_dict_stager = StateDictStager( + pin_memory=config.use_pinned_memory, share_memory=config.use_shared_memory + ) + self._staging_executor = None + self._staging_stream = None + + if self._config.use_async_staging: + self._staging_executor = ThreadPoolExecutor(max_workers=1) + if torch.accelerator.is_available(): + # Note: stream needs to be initialized on the main thread after default cuda + # stream is setup/used to avoid the risk of accidentally reusing the main + # compute stream or in other cases kernels actually launching from the + # main thread. + self._staging_stream = torch.Stream() + + if self._config.use_non_blocking_copy: + if not torch.accelerator.is_available(): + raise AssertionError( + "Non-blocking copy requires that the current accelerator is available." + ) + + def stage( + self, + state_dict: STATE_DICT, + **kwargs: Any, + ) -> STATE_DICT | Future[STATE_DICT]: + if self._config.use_async_staging: + if self._staging_executor is None: + raise AssertionError( + "Staging executor should be initialized for async staging" + ) + return self._staging_executor.submit( + self._stage, + state_dict, + **kwargs, + ) + else: + return self._stage(state_dict, **kwargs) + + def _stage(self, state_dict: STATE_DICT, **kwargs: Any) -> STATE_DICT: + state_dict = self._state_dict_stager.stage( + state_dict, non_blocking=self._config.use_non_blocking_copy, **kwargs + ) + + if self._config.use_non_blocking_copy: + if not (self._staging_stream or not self._config.use_async_staging): + raise AssertionError( + "Non-blocking copy in a background thread for async staging needs staging_stream to be initialized." + ) + + # waits for the enqueued copy operations to finish. + self._staging_stream.synchronize() if self._staging_stream else torch.accelerator.synchronize() + + return state_dict + + def close(self) -> None: + """ + Clean up all resources used by the DefaultStager. Shuts down the ThreadPoolExecutor + used for async staging operations and cleans up the underlying StateDictStager's + cached storages. Should be called when the stager is no longer needed to prevent + resource leaks, especially in long-running applications. After calling close(), + the stager should not be used for further staging operations. + + state_dict should be deep-copyable object. + + Example: + stager = DefaultStager(CheckpointStagerConfig(use_async_staging=True)) + # ... do staging operations ... + stager.close() # Clean up all resources + """ + if self._staging_executor: + self._staging_executor.shutdown(wait=True) + + self._state_dict_stager.close() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/types.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/types.py new file mode 100644 index 0000000000000000000000000000000000000000..61268fd5b14a89e6ee8dc21fdac300a50a2dd21f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/types.py @@ -0,0 +1,28 @@ +""" +Type definitions for distributed training and checkpointing. + +This module provides type definitions and classes for managing rank information +in distributed training environments, which is essential for proper checkpoint +saving and loading. +""" + +from dataclasses import dataclass +from typing import Any, TypeAlias + + +# Type alias for state dictionaries used in checkpointing +STATE_DICT: TypeAlias = dict[str, Any] + + +@dataclass +class RankInfo: + """ + Information about the current rank in a distributed training environment. + + Attributes: + global_rank: The global rank ID of the current process. + global_world_size: The total number of processes in the distributed environment. + """ + + global_rank: int + global_world_size: int diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..271e9aa112f682c8b56393c13db9eeefdeb37aa7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_experimental/utils.py @@ -0,0 +1,42 @@ +""" +Utility functions for the experimental checkpoint module. + +This module contains helper functions and utilities used across the experimental +checkpoint functionality. +""" + +from concurrent.futures import Future +from typing import Any + + +def wrap_future(original_result: Any) -> Future[None]: + """ + Wraps a result (Future or not) to return a Future with None result. + + If the input is a Future, returns a new Future that completes with None when + the original Future completes successfully, or propagates any exception. + If the input is not a Future, returns a completed Future with None result. + + Args: + original_result: The result to wrap (Future or any other value). + + Returns: + A Future that completes with None on success or propagates exceptions. + """ + masked_future: Future[None] = Future() + + if isinstance(original_result, Future): + + def on_complete(_: Future[Any]) -> None: + try: + original_result.result() + masked_future.set_result(None) + except Exception as e: + masked_future.set_exception(e) + + original_result.add_done_callback(on_complete) + else: + # Return a completed future with None result + masked_future.set_result(None) + + return masked_future diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_extension.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_extension.py new file mode 100644 index 0000000000000000000000000000000000000000..686e1affb176f76d25419163b946730bd0733d98 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_extension.py @@ -0,0 +1,222 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import abc +import io +from collections.abc import Sequence +from typing import cast, IO + +# introduced as collections.abc.Buffer in Python 3.12 +from typing_extensions import Buffer + +from torch._utils import try_import + + +# NOTE: everything in this file is experimental, and subject to +# change. Feedback and bug fixes are always welcome. + +pyzstd_module_name = "pyzstd" +pyzstd = try_import(pyzstd_module_name) +zstandard_module_name = "zstandard" +zstandard = try_import(zstandard_module_name) + + +__all__ = [ + "Extension", + "StreamTransformExtension", + "ZStandard", + "ExtensionRegistry", +] + + +class Extension(abc.ABC): + """ + Extensions provide modular additions to functionality within distributed checkpointing, + which affect the layout or format of the written artifacts. Extensions may be + built into pytorch, or provided externally. + + When writing, the caller provides a list of extension instances of the appropriate + type. Each extension can output a descriptor which is used to reconstitute the + extension at read-time. + """ + + @staticmethod + @abc.abstractmethod + def registry_name() -> str: + """ + See ExtensionRegistry.from_descriptor_list + """ + + @staticmethod + @abc.abstractmethod + def from_descriptor(version: str) -> "Extension": + """ + See ExtensionRegistry.from_descriptor_list + """ + + @abc.abstractmethod + def get_descriptor(self) -> str: + """ + Return descriptor name to be included in metadata. The form should be + "extension_name[@local-domain][/version]". + """ + + +class StreamTransformExtension(Extension): + """ + An extension which performs transformation on a byte stream, such as compression + or encryption. + + Implementations should try to be memory friendly and performant. For example, don't + read the whole input, then transform it, and write it back. If at all possible, do it in + chunks. But, don't read/transform/write one byte at a time, either. + """ + + @abc.abstractmethod + def transform_to(self, output: IO[bytes]) -> IO[bytes]: + """ + Takes a writeable output stream, and generates a new stream which implements the + output transform. Input data written to the returned stream will be transformed + and written to the `output` argument stream. + """ + + @abc.abstractmethod + def transform_from(self, input: IO[bytes]) -> IO[bytes]: + """ + Takes a readable input stream, and generates a new stream which implements the + input transform. When the returned stream is read, data will be read from the + 'input' stream, transformed, and returned. + """ + + +class ZStandard(StreamTransformExtension): + @staticmethod + def is_available() -> bool: + return zstandard is not None or pyzstd is not None + + @staticmethod + def from_descriptor(version: str) -> "ZStandard": + if version.partition(".")[0] != "1": + raise ValueError(f"Unknown extension {version=}") + if not ZStandard.is_available(): + raise ValueError( + f"Stream with ZStandard compression cannot be processed because " + f"no module named '{zstandard_module_name}' or '{pyzstd_module_name}'" + ) + return ZStandard() + + @staticmethod + def registry_name() -> str: + return "stream.zstd" + + def __init__(self) -> None: + super().__init__() + if not ZStandard.is_available(): + raise ValueError( + f"ZStandard extension is unavailable because no module named '{zstandard_module_name}' or '{pyzstd_module_name}'" + ) + + def get_descriptor(self) -> str: + return f"{self.registry_name()}/1" + + def transform_to(self, output: IO[bytes]) -> IO[bytes]: + if zstandard is not None: + compressor = zstandard.ZstdCompressor() # type: ignore[union-attr] + return compressor.stream_writer(output) + + class Writer(io.RawIOBase): + def __init__(self, output: IO[bytes]) -> None: + self.output = output + self.compressor = pyzstd.ZstdCompressor() # type: ignore[union-attr] + + def writeable(self) -> bool: + return True + + def write(self, b: Buffer) -> int | None: + outdata = self.compressor.compress(b) + if outdata: + self.output.write(outdata) + return len(memoryview(b)) + + def flush(self) -> None: + outdata = self.compressor.flush() + if outdata: + self.output.write(outdata) + self.output.flush() + + return cast(IO[bytes], Writer(output)) + + def transform_from(self, input: IO[bytes]) -> IO[bytes]: + if zstandard is not None: + decompressor = zstandard.ZstdDecompressor() # type: ignore[union-attr] + return decompressor.stream_reader(input) + + class Reader(io.RawIOBase): + def __init__(self, input: IO[bytes]) -> None: + self.input = input + self.decompressor = pyzstd.EndlessZstdDecompressor() # type: ignore[union-attr] + + def readable(self) -> bool: + return True + + def readinto(self, b: Buffer) -> int | None: + # This needs to read enough so it can decompress + # something so the output doesn't look like EOF. This + # means reading at least one block. The max block + # size is 128KB, so we read that plus some + # overhead to be sure. + + if self.decompressor.needs_input: + indata = self.input.read((128 + 6) * 1024) + else: + indata = b"" + + bview = memoryview(b) + blen = len(bview) + outdata = self.decompressor.decompress(indata, blen) + if outdata is None: + return None + + count = len(outdata) + bview[:count] = outdata + return count + + def seekable(self) -> bool: + return False + + return cast(IO[bytes], Reader(input)) + + +class ExtensionRegistry: + def __init__(self) -> None: + # Populate default registry contents + self.extensions: dict[str, type[Extension]] = { + cls.registry_name(): cls for cls in (ZStandard,) + } + + def register(self, cls: type[Extension]) -> None: + self.extensions[cls.registry_name()] = cls + + def from_descriptor_list(self, descriptors: Sequence[str]) -> Sequence[Extension]: + """ + Given a sequence of descriptor strings as returned by + Extension.get_descriptor at save time, creates a sequence of + Extension instances. The name[@local-domain] preceding the + version number is used to look up an implementation class in + the registry, and the version is passed to the class's + from_descriptor static method. If the registry contains no + match, this will throw ValueError. If the from_descriptor + method raises an exception, that will pass through to the + caller. + """ + + def from_descriptor(desc: str) -> Extension: + name, _, version = desc.partition("/") + if version is None: + version = 0 + ext = self.extensions.get(name) + if not ext: + raise ValueError(f"Unknown extension {name=}") + # pyrefly: ignore [bad-argument-type] + return ext.from_descriptor(version) + + return [from_descriptor(desc) for desc in descriptors] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_fsspec_filesystem.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_fsspec_filesystem.py new file mode 100644 index 0000000000000000000000000000000000000000..0d2738d7198eb285f03fa4988f72a02c4b4b0882 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_fsspec_filesystem.py @@ -0,0 +1,162 @@ +# Mypy will not try inferring the types of any 3rd party libraries installed. +# mypy: ignore-errors + +import io +import os +from collections.abc import Generator, Sequence +from contextlib import contextmanager +from pathlib import Path +from typing import TYPE_CHECKING + +from fsspec.core import url_to_fs + +from torch.distributed.checkpoint._extension import StreamTransformExtension +from torch.distributed.checkpoint.filesystem import ( + FileSystemBase, + FileSystemReader, + FileSystemWriter, + SerializationFormat, +) + + +if TYPE_CHECKING: + from fsspec import AbstractFileSystem + + +__all__ = [ + "FsspecWriter", + "FsspecReader", +] + + +class FileSystem(FileSystemBase): + def __init__(self) -> None: + self.fs: AbstractFileSystem | None = None + + @contextmanager + def create_stream( + self, path: str | os.PathLike, mode: str + ) -> Generator[io.IOBase, None, None]: + if self.fs is None: + raise AssertionError("fs should not be None") + path = os.fspath(path) + + # fsspec does not support concurrent transactions, and not all + # AbstractFileSystem have working rollback implementations, so + # just manually delete the file if necessary on errors. + with self.fs.open(path, mode) as stream: + try: + yield stream + except: + if any(ch in mode for ch in "w+a"): # cleanup file if not read-only + try: + self.rm_file(path) + except: # noqa: E722 + pass + raise + + def concat_path(self, path: str | os.PathLike, suffix: str) -> str | os.PathLike: + return os.path.join(path, suffix) + + def init_path(self, path: str | os.PathLike, **kwargs) -> str | os.PathLike: + self.fs, _ = url_to_fs(path, **kwargs) + return path + + def rename(self, path: str | os.PathLike, new_path: str | os.PathLike) -> None: + self.fs.rename(path, new_path) + + def mkdir(self, path: str | os.PathLike) -> None: + self.fs.makedirs(path, exist_ok=True) + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + if isinstance(checkpoint_id, Path): + return False + + try: + url_to_fs(checkpoint_id) + except ValueError: + return False + + return True + + def exists(self, path: str | os.PathLike) -> bool: + return self.fs.exists(path) + + def rm_file(self, path: str | os.PathLike) -> None: + self.fs.rm(path) + + def ls(self, path: str | os.PathLike) -> list[str]: + # setting detail to False explicitly to keep the list[str] return type, + # instead of the list[Dict] return type when detail=True + return self.fs.ls(path, detail=False) + + +# TODO: add the dcp.async_save mixin +class FsspecWriter(FileSystemWriter): + """ + Basic implementation of StorageWriter using fsspec. + + This implementation makes the following assumptions and simplifications: + + * The checkpoint path is an empty or non-existing directory. + * File creation is atomic + + The checkpoint consist of one file per write request plus + a `.metadata` file with the serialized metadata. + + """ + + def __init__( + self, + path: str | os.PathLike, + single_file_per_rank: bool = True, + sync_files: bool = True, + thread_count: int = 1, + per_thread_copy_ahead: int = 10_000_000, + overwrite: bool = True, + _extensions: Sequence[StreamTransformExtension] | None = None, + serialization_format: SerializationFormat = SerializationFormat.TORCH_SAVE, + **kwargs, + ) -> None: + """ + Initialize the writer pointing to `path`. + + Args: + path: directory where the checkpoint will be written to. + single_file_per_rank: Produce one file per rank instead of one file per tensor/blob. Default to True. + sync_files : force files to be synced to permanent storage. Default to True. + thread_count: Number of IO threads to use to write. Default to 1. + per_thread_copy_ahead: How many bytes to copy from the GPU ahead of saving them. Default 10Mb. + overwrite: Whether to allow overwriting existing checkpoints. Defaults to True. + _extensions: Extensions to apply to output streams (EXPERIMENTAL) + + N. B. If sync_files is disabled, there's no guarantee that the checkpoint will be consistent in the case of a failure. + """ + super().__init__( + path, + single_file_per_rank, + sync_files, + thread_count, + per_thread_copy_ahead, + overwrite=overwrite, + _extensions=_extensions, + serialization_format=serialization_format, + ) + self.fs = FileSystem() + self.path = self.fs.init_path(path, **kwargs) + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + return FileSystem.validate_checkpoint_id(checkpoint_id) + + +class FsspecReader(FileSystemReader): + def __init__(self, path: str | os.PathLike, **kwargs) -> None: + super().__init__(path) + self.fs = FileSystem() + self.path = self.fs.init_path(path, **kwargs) + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + return FileSystem.validate_checkpoint_id(checkpoint_id) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_hf_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_hf_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..0e95bdfb78dfa31c5e5589fa0f8c0058ca14ac24 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_hf_utils.py @@ -0,0 +1,106 @@ +import io +import json +import struct +from dataclasses import dataclass +from typing import Any + +import torch + + +_metadata_fn: str = "model.safetensors.index.json" + +FILE_NAME = "model-{cpt_idx}-of-{num_files}" +SHARDED_FILE_NAME = "shard-{shard_idx}-model-{cpt_idx}-of-{num_files}" +SUFFIX = ".safetensors" + +# metadata keys +CUSTOM_METADATA_KEY = "DCP_SHARDING_INFO" +DEFAULT_EXTRA_METADATA_KEY = "__metadata__" +SAVED_OFFSETS_KEY = "saved_offsets" +SHAPE_KEY = "shape" +DATA_KEY = "data" +DTYPE_KEY = "dtype" +DATA_OFFSETS_KEY = "data_offsets" + +DTYPE_MAP = { + "F16": torch.float16, + "F32": torch.float32, + "F64": torch.float64, + "I8": torch.int8, + "U8": torch.uint8, + "I16": torch.int16, + "I32": torch.int32, + "I64": torch.int64, + "BF16": torch.bfloat16, +} + +HF_DCP_VERSION: float = 1.0 +DCP_VERSION_KEY = "DCP_VERSION" +DCP_SHARDING_INFO_KEY = "DCP_SHARDING_INFO" + +FORMAT_KEY = "format" +FORMAT_VALUE = "pt" + +NUM_BYTES_FOR_HEADER_LEN = 8 + +SHARDED_DIR_NAME = "sharded" + + +@dataclass +class _HFStorageInfo: + """This is the per entry storage info.""" + + relative_path: str + shape: torch.Size + dtype: torch.dtype + + +def _gen_file_name( + index: int, largest_index: int, shard_index: int | None = None +) -> str: + if shard_index is not None: + return ( + SHARDED_FILE_NAME.format( + shard_idx=f"{shard_index}".zfill(5), + cpt_idx=f"{index}".zfill(5), + num_files=f"{largest_index}".zfill(5), + ) + + SUFFIX + ) + else: + return ( + FILE_NAME.format( + cpt_idx=f"{index}".zfill(5), num_files=f"{largest_index}".zfill(5) + ) + + SUFFIX + ) + + +def _get_safetensors_file_metadata(file_bytes: io.IOBase) -> tuple[Any, int]: + # this uses the same logic that's done in HF code base + # https://github.com/2404589803/huggingface_hub/blob/main/src/huggingface_hub/hf_api.py#L5308 + # and follows their documentation on how their files are serialized + # https://huggingface.co/docs/safetensors/index#format + + header_len_bytes = file_bytes.read(NUM_BYTES_FOR_HEADER_LEN) + header_len = struct.unpack(" torch.dtype: + try: + dtype = DTYPE_MAP[dtype_str] + except KeyError: + dtype = torch.get_default_dtype() + + return dtype + + +def _get_dcp_custom_metadata(metadata: Any) -> Any | None: + if DEFAULT_EXTRA_METADATA_KEY in metadata: + custom_metadata = metadata[DEFAULT_EXTRA_METADATA_KEY] + if CUSTOM_METADATA_KEY in custom_metadata: + return json.loads(custom_metadata[CUSTOM_METADATA_KEY]) + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_nested_dict.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_nested_dict.py new file mode 100644 index 0000000000000000000000000000000000000000..eb26058370f766fbb96e4a5f1530577234eed62a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_nested_dict.py @@ -0,0 +1,69 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE + +from . import _version +from ._traverse import ( + OBJ_PATH, + set_element, + STATE_DICT_ITEM, + traverse_state_dict, + traverse_state_dict_v_2_3, +) + + +""" +TODO: +Need to add ability to handle tuple, OrderedDict, NamedTuple. +Update mappings from dict to a class. +Change set_element to recreate the right type for tuple, OrderedDict, and NamedTuple. +""" + + +FLATTEN_MAPPING = dict[str, OBJ_PATH] + + +# TODO: Update Docstring for nested_dict.py +def flatten_state_dict( + state_dict: STATE_DICT_TYPE, +) -> tuple[STATE_DICT_TYPE, FLATTEN_MAPPING]: + """ + Flatten ``state_dict`` made of nested dicts and lists into a top level dictionary. + + Use ``unflatten_state_dict`` to revert this process. + Returns: + A tuple with the flatten state_dict and a mapping from original to new state_dict. + N.B. The new keys are derived from the object paths, joined by dot. + For example: ``{ 'a': {'b':...}}`` results in the key `a.b`. + """ + flattened: STATE_DICT_TYPE = {} + mappings: FLATTEN_MAPPING = {} + + def flat_copy(path: OBJ_PATH, value: STATE_DICT_ITEM) -> None: + new_fqn = ".".join(map(str, path)) + if new_fqn in flattened: + raise ValueError(f"duplicated flatten key {new_fqn}") + flattened[new_fqn] = value + mappings[new_fqn] = path + + # We started to flatten dictionary since v2.4. But in order to not break + # the checkpoints that were saved before v2.4, we need to keep the old + # traversal so that we can reconstruct those checkpoints. + use_v_2_3 = ( + _version._derived_version is not None and _version._derived_version == "2_3" + ) + if use_v_2_3: + traverse_state_dict_v_2_3(state_dict, flat_copy) + else: + traverse_state_dict(state_dict, flat_copy) + return flattened, mappings + + +def unflatten_state_dict( + state_dict: STATE_DICT_TYPE, mapping: FLATTEN_MAPPING +) -> STATE_DICT_TYPE: + """Restore the original nested state_dict according to ``mapping`` and the flattened ``state_dict``.""" + nested: STATE_DICT_TYPE = {} + for key, value in state_dict.items(): + set_element(nested, mapping[key], value) + return nested diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_pg_transport.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_pg_transport.py new file mode 100644 index 0000000000000000000000000000000000000000..46267c40eddc7a23c6bf50e29339bc3e54cf2ad4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_pg_transport.py @@ -0,0 +1,384 @@ +import logging +import pickle +import time +from collections.abc import Callable, Generator +from contextlib import contextmanager +from dataclasses import dataclass +from datetime import timedelta +from typing import cast, TypeVar + +import torch +from torch.distributed import ProcessGroup, Work +from torch.distributed._shard.sharded_tensor import ( + Shard as ShardedTensorShard, + ShardedTensor, + ShardMetadata, +) +from torch.distributed._shard.sharded_tensor.metadata import ShardedTensorMetadata +from torch.distributed.tensor import _DTensorSpec, DTensor +from torch.utils._pytree import ( + KeyPath, + tree_flatten_with_path, + tree_unflatten, + TreeSpec, +) + + +logger: logging.Logger = logging.getLogger(__name__) + +T = TypeVar("T") + + +@dataclass +class _TensorMeta: + """ + This is the metadata for a tensor that is used to transfer checkpoints. + It contains the shape, the dtype, the storage offset and the stride of the + tensor. + + This must be pickleable so that it can be sent over the wire. + """ + + shape: torch.Size + dtype: torch.dtype + storage_offset: int + stride: tuple[int, ...] + nbytes: int + + +@dataclass +class _DTensorMeta: + """ + This is the metadata for a DTensor that is used to transfer checkpoints. + It contains the metadata for the local tensor and the spec of the DTensor. + + This must be pickleable so that it can be sent over the wire. + """ + + local: _TensorMeta + spec: _DTensorSpec + + +@dataclass +class _ShardedTensorMeta: + """ + This is the metadata for a ShardedTensor that is used to transfer checkpoints. + It contains the metadata for all local shards and the global tensor metadata. + + This must be pickleable so that it can be sent over the wire. + """ + + local_shards_meta: list[_TensorMeta] + local_shards_shard_metadata: list[ + ShardMetadata + ] # Original shard metadata for each local shard + sharded_tensor_metadata: ShardedTensorMetadata + + +@dataclass +class _StateDictMeta: + """ + This is the metadata for a state dict that is used to transfer checkpoints. + It contains the step, the pytree spec of the state dict and the metadata for + each tensor in the state dict. + + This must be pickleable so that it can be sent over the wire. + + Args: + step: the step of the checkpoint to verify consistency + treespec: the pytree spec of the state dict + paths: the path of each leaf in the state dict + non_tensor_leaves: the metadata for each tensor in the state dict and any + non-tensor leaves in the state dict + """ + + treespec: TreeSpec + paths: list[KeyPath] + non_tensor_leaves: list[object | _TensorMeta | _DTensorMeta | _ShardedTensorMeta] + + +@contextmanager +def _timeit(name: str) -> Generator[None, None, None]: + start = time.perf_counter() + yield + dur = time.perf_counter() - start + logger.info("%s took %ss", name, dur) + + +def _prepare_tensor(tensor: torch.Tensor) -> tuple[torch.Tensor, _TensorMeta]: + return ( + _cast_tensor(tensor, torch.uint8), + _TensorMeta( + shape=tensor.shape, + dtype=tensor.dtype, + storage_offset=cast(int, tensor.storage_offset()), + stride=tensor.stride(), + nbytes=tensor.untyped_storage().nbytes(), + ), + ) + + +def _prepare_state_dict( + state_dict: object, + device: torch.device, +) -> tuple[_StateDictMeta, list[torch.Tensor]]: + leaves: list[tuple[KeyPath, object]] + leaves, treespec = tree_flatten_with_path(state_dict) + + paths: list[KeyPath] = [] + non_tensor_leaves: list[ + object | _TensorMeta | _DTensorMeta | _ShardedTensorMeta + ] = [] + tensors: list[torch.Tensor] = [] + for key_path, v in leaves: + paths.append(key_path) + + if isinstance(v, DTensor): + tensor, tensor_meta = _prepare_tensor(v._local_tensor) + + tensors.append(tensor) + + non_tensor_leaves.append( + _DTensorMeta( + local=tensor_meta, + spec=v._spec, + ) + ) + elif isinstance(v, ShardedTensor): + # Handle ShardedTensor by extracting all local shards + local_shards = v.local_shards() + + # Prepare metadata for all local shards + local_shards_meta = [] + local_shards_shard_metadata = [] + for shard in local_shards: + tensor, tensor_meta = _prepare_tensor(shard.tensor) + tensors.append(tensor) + local_shards_meta.append(tensor_meta) + local_shards_shard_metadata.append(shard.metadata) + + non_tensor_leaves.append( + _ShardedTensorMeta( + local_shards_meta=local_shards_meta, + local_shards_shard_metadata=local_shards_shard_metadata, + sharded_tensor_metadata=v.metadata(), # Complete metadata + ) + ) + elif isinstance(v, torch.Tensor): + tensor, tensor_meta = _prepare_tensor(v) + tensors.append(tensor) + non_tensor_leaves.append(tensor_meta) + else: + non_tensor_leaves.append(v) + + return ( + _StateDictMeta( + treespec=treespec, + paths=paths, + non_tensor_leaves=non_tensor_leaves, + ), + tensors, + ) + + +def _cast_tensor(tensor: torch.Tensor, dtype: torch.dtype) -> torch.Tensor: + """ + Casts the underlying storage to a tensor of the given dtype. + + The returned tensor will be of size ``storage.nbytes``. + + This works for all datatypes and supports strided/offset tensors with the + caveat that the cast tensor may be larger than the original tensor due to + the differences in striding. + """ + if type(tensor) is not torch.Tensor: + raise AssertionError(f"can only cast standard tensors not {type(tensor)}") + storage = tensor.untyped_storage() + ret = torch.tensor(storage, dtype=dtype, device=tensor.device) + if ret.untyped_storage() is not storage: + raise AssertionError("storage should be the same") + return ret + + +class PGTransport: + """ + This is a checkpoint transport that uses the process group to transfer checkpoints. + This allows for fast recovery of workers by fetching the current weights + from an existing worker. + + Args: + pg: the process group to use for communication + timeout: the timeout for communication + device: the device to use for tensors + state_dict: if specified this function will be called to do an inplace + receive into the returned state_dict. This is much faster than + having to allocate new tensors and transferring them to the CPU. + """ + + def __init__( + self, + pg: ProcessGroup, + timeout: timedelta, + device: torch.device, + state_dict: Callable[[], object] | None = None, + ) -> None: + self._work: list[Work] = [] + self._pg = pg + self._timeout = timeout + self._device = device + self._state_dict = state_dict + + def send_checkpoint(self, dst_ranks: list[int], state_dict: object) -> None: + """ + Send a checkpoint to multiple destination ranks. + + The process: + 1. Prepares the state dict by converting tensors to a serializable format + 2. Sends metadata as pickled data + 3. Sends each tensor sequentially to all destination ranks + + Args: + dst_ranks: List of destination ranks to send the checkpoint to + state_dict: The state dictionary containing model parameters + """ + with _timeit("preparing state_dict"): + meta, tensors = _prepare_state_dict(state_dict, device=self._device) + + work = [] + + with _timeit("send meta"): + buf = pickle.dumps(meta) + len_t = torch.tensor([len(buf)], dtype=torch.int64, device=self._device) + buf_t = torch.frombuffer(buf, dtype=torch.uint8).to(self._device) + for dst_rank in dst_ranks: + work.append(self._pg.send([len_t], dst_rank, tag=1)) + work.append(self._pg.send([buf_t], dst_rank, tag=2)) + + with _timeit("send tensors"): + for i, t in enumerate(tensors): + original_device = t.device + t = t.to(self._device) + for dst_rank in dst_ranks: + work.append(self._pg.send([t], dst_rank, tag=3 + i)) + + # if we did a copy we should wait for the work to complete so we + # can free the memory to avoid OOMs + if original_device == torch.device("cpu"): + for w in work: + w.wait() + work = [] + + for w in work: + w.wait() + + def recv_checkpoint(self, src_rank: int) -> object: + """ + Receive a checkpoint from a source rank. + + The process: + 1. Receives metadata about the checkpoint structure + 2. Receives each tensor, potentially reusing existing tensors for in-place updates + 3. Reconstructs the original state dict structure + + Args: + src_rank: The source rank to receive the checkpoint from + + Returns: + The reconstructed state dictionary with model parameters + """ + state_dict = self._state_dict() if self._state_dict else {} + state_dict_leaves, _ = tree_flatten_with_path(state_dict) + + dst_tensors: dict[KeyPath, object] = dict(state_dict_leaves) + + len_t = torch.zeros(1, dtype=torch.int64, device=self._device) + self._pg.recv([len_t], src_rank, tag=1).wait() + length = cast(int, len_t.item()) + + buf = torch.empty(length, dtype=torch.uint8, device=self._device) + self._pg.recv([buf], src_rank, tag=2).wait() + + meta: _StateDictMeta = pickle.loads(buf.cpu().numpy().tobytes()) + + i: int = 0 + works: list[Work] = [] + + def recv(path: KeyPath, v: _TensorMeta) -> torch.Tensor: + nonlocal i + + inplace = dst_tensors.get(path) + if ( + isinstance(inplace, torch.Tensor) + and inplace.device.type == self._device.type + ): + if isinstance(inplace, DTensor): + inplace = inplace._local_tensor + t = _cast_tensor(inplace, torch.uint8) + if t.nbytes != v.nbytes: + raise AssertionError("inplace tensor storage must be the same size") + else: + t = torch.empty(v.nbytes, dtype=torch.uint8, device=self._device) + + work = self._pg.recv([t], src_rank, tag=3 + i) + i += 1 + + if inplace is None: + # if not inplace we need to copy it to CPU to avoid OOMing + work.wait() + t = t.cpu() + else: + works.append(work) + + return torch.as_strided( + t.view(v.dtype), + size=v.shape, + stride=v.stride, + storage_offset=v.storage_offset, + ) + + values: list[object] = [] + for path, v in zip(meta.paths, meta.non_tensor_leaves): + if isinstance(v, _TensorMeta): + values.append(recv(path, v)) + elif isinstance(v, _DTensorMeta): + tensor = recv(path, v.local) + # pyrefly: ignore [bad-argument-type, bad-argument-count, unexpected-keyword] + values.append(DTensor(tensor, v.spec, requires_grad=False)) + elif isinstance(v, _ShardedTensorMeta): + # Receive all local shards that were sent to us + local_shards = [] + current_rank = self._pg.rank() + + # Receive tensors for each local shard that was sent + for j, shard_meta in enumerate(v.local_shards_meta): + tensor = recv(path, shard_meta) + + # Use the original shard metadata that was stored during preparation + # but update the placement to reflect the current rank/device + original_shard_metadata = v.local_shards_shard_metadata[j] + updated_shard_metadata = ShardMetadata( + shard_offsets=original_shard_metadata.shard_offsets, + shard_sizes=original_shard_metadata.shard_sizes, + placement=f"rank:{current_rank}/{tensor.device.type}", + ) + + local_shard = ShardedTensorShard( + tensor=tensor, metadata=updated_shard_metadata + ) + local_shards.append(local_shard) + + # Use complete metadata to reconstruct ShardedTensor + sharded_tensor = ( + ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards=local_shards, + sharded_tensor_metadata=v.sharded_tensor_metadata, + ) + ) + values.append(sharded_tensor) + else: + values.append(v) + + for work in works: + work.wait() + + return tree_unflatten(values, meta.treespec) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_sharded_tensor_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_sharded_tensor_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..a68bcddeb7f9d9ffe6f89056dfe1ccc30cc12eb5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_sharded_tensor_utils.py @@ -0,0 +1,107 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import copy +from typing import TYPE_CHECKING + +import torch.distributed as dist +from torch.distributed._shard.sharded_tensor import Shard, ShardedTensor, ShardMetadata +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE +from torch.distributed.remote_device import _remote_device + +from ._traverse import OBJ_PATH, set_element, STATE_DICT_ITEM, traverse_state_dict +from .utils import _element_wise_add, _normalize_device_info + + +if TYPE_CHECKING: + from torch.distributed._shard.sharded_tensor.metadata import ShardedTensorMetadata + + +# TODO: We need to refactor this code. +def _flatten_sharded_tensors(state_dict: STATE_DICT_TYPE) -> STATE_DICT_TYPE: + r""" + Transform ``state_dict`` by flattening all nested ShardedTensor instances found. + + The resulting ShardedTensor instances are only correct regarding the local shard and + MUST not be used for any other purpose but checkpointing, as no operator will work with them. + + This function should be used in conjunction with a state_dict produced by FSDP's + StateDictType.SHARDED_STATE_DICT methods. + """ + new_state_dict: STATE_DICT_TYPE = {} + + def rewrite_dict(path: OBJ_PATH, value: STATE_DICT_ITEM) -> None: + if not isinstance(value, ShardedTensor): + set_element(new_state_dict, path, value) + return + shards = value.local_shards() + + if len(shards) == 0: + return + if len(shards) != 1: + set_element(new_state_dict, path, value) + return + + outer_shard = shards[0] + + inner_st = outer_shard.tensor + if not isinstance(inner_st, ShardedTensor): + set_element(new_state_dict, path, value) + return + + if len(inner_st.local_shards()) != 1: + raise ValueError("Cannot handle inner tensor with more than 1 shard") + inner_shard = inner_st.local_shards()[0] + + local_shards = [ + Shard( + tensor=inner_shard.tensor, + metadata=ShardMetadata( + shard_offsets=_element_wise_add( + outer_shard.metadata.shard_offsets, + inner_shard.metadata.shard_offsets, + ), + shard_sizes=inner_shard.metadata.shard_sizes, + placement=f"rank:{dist.get_rank()}/{inner_shard.tensor.device}", + ), + ) + ] + + st_meta: ShardedTensorMetadata = copy.deepcopy(value.metadata()) + other_rank = 0 if dist.get_rank() > 0 else 1 + device_info = _normalize_device_info(inner_shard.tensor.device.type, 0) + + # Remove the outer ST shard the inner ST covers + for i, shard_md in enumerate(st_meta.shards_metadata): + if shard_md.shard_offsets == outer_shard.metadata.shard_offsets: + st_meta.shards_metadata.pop(i) + break + + # Attribute other rank for the other shards + for shard_md in st_meta.shards_metadata: + shard_md.placement = _remote_device(f"rank:{other_rank}/{device_info}") + + # Add other inner shards from the inner tensor + for inner_md in inner_st.metadata().shards_metadata: + if inner_md.shard_offsets != inner_shard.metadata.shard_offsets: + st_meta.shards_metadata.append( + ShardMetadata( + shard_offsets=_element_wise_add( + outer_shard.metadata.shard_offsets, + inner_md.shard_offsets, + ), + shard_sizes=inner_md.shard_sizes, + placement=f"rank:{other_rank}/{device_info}", + ) + ) + + # Finally add this shard + st_meta.shards_metadata.append(local_shards[0].metadata) + + st = ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards=local_shards, + sharded_tensor_metadata=st_meta, + ) + set_element(new_state_dict, path, st) + + traverse_state_dict(state_dict, rewrite_dict) + return new_state_dict diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_state_dict_stager.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_state_dict_stager.py new file mode 100644 index 0000000000000000000000000000000000000000..deb43f1aa8a20416e4410aad9727db21670a5602 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_state_dict_stager.py @@ -0,0 +1,473 @@ +# mypy: allow-untyped-defs +import types +import warnings +import weakref +from copyreg import dispatch_table +from typing import Any + +import torch +import torch.cuda._pin_memory_utils as pin_memory_utils +from torch.storage import UntypedStorage +from torch.utils.weak import WeakIdKeyDictionary + + +class StateDictStager: + """ + A class for optimizing storage objects during staging for async checkpointing. + + StateDictStager stages the state_dict to CPU DRAM while applying optimizations + like memory sharing and pinning to improve performance. It caches storage objects + to avoid redundant copies and can be configured to automatically share memory + (for multi-process usage) and pin memory (for faster CPU-GPU transfers). + + Attributes: + pin_memory (bool): Whether to pin CPU memory for faster CPU-GPU transfers + share_memory (bool): Whether to share memory across processes + pin_memory_min_bytes (int): Minimum tensor size in bytes to pin memory (default: 5) + _cached_storage_mapping (WeakIdKeyDictionary): Maps storage objects to optimized CPU storages using weak references + """ + + def __init__( + self, + pin_memory: bool = False, + share_memory: bool = False, + pin_memory_min_bytes: int = 5, + ): + if pin_memory and not torch.cuda.is_available(): + warnings.warn( + "Ignoring pin_memory flag for checkpoint staging as pinning memory" + "requires CUDA, but CUDA is not available. ", + stacklevel=2, + ) + self.pin_memory = False + else: + self.pin_memory = pin_memory + self.share_memory = share_memory + # Mapping from original storage objects to CPU storages using weak references + self._cached_storage_mapping = WeakIdKeyDictionary() + self.pin_memory_min_bytes = pin_memory_min_bytes + + def _deepcopy_atomic(x, _): + return x + + def _deepcopy_list(x, memo, non_blocking=False): + y: list = [] + memo[id(x)] = y + append = y.append + for a in x: + append( + self.deepcopy_with_tensor_offload( + a, memo, non_blocking=non_blocking + ) + ) + return y + + def _deepcopy_tuple(x, memo, non_blocking=False): + y = [ + self.deepcopy_with_tensor_offload(a, memo, non_blocking=non_blocking) + for a in x + ] + # We're not going to put the tuple in the memo, but it's still important we + # check for it, in case the tuple contains recursive mutable structures. + try: + return memo[id(x)] + except KeyError: + pass + + # Check if any elements changed during deepcopy + for k, j in zip(x, y): + if k is not j: + # At least one element changed, create new tuple + return tuple(y) + + # No elements changed, return original tuple + return x + + def _deepcopy_dict(x, memo, non_blocking=False): + y: dict = {} + memo[id(x)] = y + for key, value in x.items(): + y[ + self.deepcopy_with_tensor_offload( + key, memo, non_blocking=non_blocking + ) + ] = self.deepcopy_with_tensor_offload( + value, memo, non_blocking=non_blocking + ) + return y + + def _deepcopy_method(x, memo, non_blocking=False): # Copy instance methods + return type(x)( + x.__func__, + self.deepcopy_with_tensor_offload( + x.__self__, memo, non_blocking=non_blocking + ), + ) + + d: dict[Any, Any] = {} + self._deepcopy_dispatch = d + d[type(None)] = _deepcopy_atomic + d[int] = _deepcopy_atomic + d[float] = _deepcopy_atomic + d[bool] = _deepcopy_atomic + d[complex] = _deepcopy_atomic + d[bytes] = _deepcopy_atomic + d[str] = _deepcopy_atomic + d[types.CodeType] = _deepcopy_atomic + d[type] = _deepcopy_atomic + d[range] = _deepcopy_atomic + d[types.BuiltinFunctionType] = _deepcopy_atomic + d[types.FunctionType] = _deepcopy_atomic + d[weakref.ref] = _deepcopy_atomic + d[property] = _deepcopy_atomic + d[types.MethodType] = _deepcopy_method + d[dict] = _deepcopy_dict + d[tuple] = _deepcopy_tuple + d[list] = _deepcopy_list + + def _stage_untyped_storage( + self, + storage: UntypedStorage, + non_blocking: bool = False, + ): + """ + Called from the hooked storage_deepcopy function in torch.Tensor.__deepcopy__. + + This method handles the storage optimization logic for the StagingStateDict class. + It checks if the storage has already been cached, and if so, reuses it. + Otherwise, it creates a new CPU storage and applies memory optimizations. + + Args: + storage: The storage to optimize + + Returns: + The optimized storage + """ + # Check if we've already cached this storage + if storage in self._cached_storage_mapping: + cached_storage = self._cached_storage_mapping[storage] + if cached_storage.size() != storage.size(): + raise AssertionError( + "For async checkpointing, We cache storages in DRAM and reuse them." + "Cached storage size does not match original storage size." + "This should never happen as we track the original storage weakref " + "and clean up the cache storage. Please report this to PyTorch Distributed Checkpointing." + ) + # Reuse cached storage but update with new data + cached_storage.copy_(storage, non_blocking=non_blocking) + return cached_storage + + # Create new CPU storage + if self.share_memory: + new_storage = type(storage)._new_shared(storage.size(), device="cpu") + else: + new_storage = type(storage)(storage.size(), device="cpu") + + # Skip pinning for tensors below the minimum size threshold + # Small tensors (e.g., optimizer step counters, scalars) have negligible + # transfer time improvement from pinning, but pinning overhead is significant + if self.pin_memory and new_storage.nbytes() >= self.pin_memory_min_bytes: + pin_memory_utils.pin_memory(new_storage.data_ptr(), new_storage.nbytes()) + # Set up a weak reference to unpin when cpu storage is garbage collected + f = weakref.finalize( + new_storage, pin_memory_utils.unpin_memory, new_storage.data_ptr() + ) + # This makes sure that the finalizer is not called after + # cuda context is destroyed. + f.atexit = False + + new_storage.copy_(storage, non_blocking=non_blocking) + + # Cache the storage - WeakIdKeyDictionary will automatically clean up when storage is garbage collected + self._cached_storage_mapping[storage] = new_storage + return new_storage + + @torch.no_grad() + def stage( + self, + state_dict: Any, + non_blocking: bool = False, + ) -> Any: + return self.deepcopy_with_tensor_offload(state_dict, None, [], non_blocking) + + def _offload_tensor(self, x, memo, non_blocking=False): + """ + Deep copy a PyTorch tensor with optimized storage handling. + + This method creates a CPU copy of a tensor while applying memory optimizations + like sharing and pinning based on the StateDictStager configuration. + + Args: + x: The tensor to copy + memo: Memo dictionary for tracking already copied objects + non_blocking: Whether to perform non-blocking copies where possible + + Returns: + A CPU copy of the tensor with optimized storage + """ + # if data_ptr is not 0, we allocate a new storage below. so we can skip + # memory allocation by using [] for size. + y = x.new_empty([] if x.data_ptr() != 0 else x.size(), device="cpu") + + # Store in memo dict early to handle recursive references + d = id(x) + memo[d] = y + + if type(x) is torch.Tensor or x.data_ptr() != 0: + # Get the untyped storage + untyped_storage = x.untyped_storage() + storage_id = id(untyped_storage) + + # Check if this storage has already been staged in this deepcopy operation + # This handles the case where different tensors share the same storage + # (e.g., FSDP state_dict where norm.weight and norm_weight reference same storage) + # PyTorch caches untyped_storage() calls, so same storage -> same id + if storage_id in memo: + copied_storage = memo[storage_id] + else: + # Storage not seen before in this operation, stage it + copied_storage = self._stage_untyped_storage( + untyped_storage, non_blocking=non_blocking + ) + # Add to memo to avoid re-staging if we see this storage again + memo[storage_id] = copied_storage + + # Set the tensor data using the staged storage + y.set_(copied_storage, x.storage_offset(), x.size(), x.stride()) + + # Copy any attributes the tensor might have + if hasattr(x, "__dict__"): + for attr_name, attr_value in x.__dict__.items(): + setattr( + y, + attr_name, + self.deepcopy_with_tensor_offload( + attr_value, memo, non_blocking=non_blocking + ), + ) + + if hasattr(x, "__slots__"): + for slot in x.__slots__: + if hasattr(x, slot): + setattr( + y, + slot, + self.deepcopy_with_tensor_offload( + getattr(x, slot), memo, non_blocking=non_blocking + ), + ) + + return y + + def close(self): + """ + Clean up all cached storages and release associated resources. + + This method clears the internal storage cache, allowing garbage collection + of cached CPU storages. Any pinned memory associated with cached storages + will be automatically unpinned through weak reference finalizers. + + It also clears the _deepcopy_dispatch dict to break the reference cycle + created by closures that capture self. Without this, it may + cause memory leaks. + """ + self._cached_storage_mapping.clear() + self._deepcopy_dispatch.clear() + + @torch.no_grad() + def deepcopy_with_tensor_offload(self, x, memo=None, _nil=[], non_blocking=False): # noqa: B006 + """Deep copy operation on arbitrary Python objects with special handling for PyTorch tensors. + + This implementation extends the standard deepcopy functionality to handle PyTorch tensors + and their storages in a way that optimizes memory usage and performance, similar to the + stage method. It applies memory sharing and pinning optimizations based on the StateDictStager + configuration. + + Args: + x: The object to deep copy + memo: Memo dictionary for tracking already copied objects + _nil: Sentinel value for memo dictionary + non_blocking: Whether to perform non-blocking copies where possible + + Returns: + A deep copy of the input object with optimized tensor storage handling + """ + if memo is None: + memo = {} + + d = id(x) + y = memo.get(d, _nil) + if y is not _nil: + return y + + cls = type(x) + + # tensors and subclasses of tensors are handled separately + if isinstance(x, torch.Tensor): + y = self._offload_tensor(x, memo, non_blocking=non_blocking) + else: + # Use the dispatch table for standard types + copier = self._deepcopy_dispatch.get(cls) + if copier is not None: + # Check if this is an atomic copier (only accepts x and memo) + if copier.__name__ == "_deepcopy_atomic": + y = copier(x, memo) + else: + y = copier(x, memo, non_blocking=non_blocking) + else: + if issubclass(cls, type): + # type copier is also atomic + y = self._deepcopy_dispatch[type](x, memo) + else: + copier = getattr(x, "__deepcopy__", None) + if copier is not None: + y = copier(memo) + else: + reductor = dispatch_table.get(cls) + if reductor: + rv = reductor(x) + else: + reductor = getattr(x, "__reduce_ex__", None) + if reductor is not None: + rv = reductor(4) + else: + reductor = getattr(x, "__reduce__", None) + if reductor: + rv = reductor() + else: + raise RuntimeError( + f"un(deep)copyable object of type {cls}" + ) + if isinstance(rv, str): + y = x + else: + # Unpack rv tuple elements (up to 5 from pickle protocol) + # and explicitly pass non_blocking as keyword arg + if len(rv) == 2: + func, args = rv + y = self._reconstruct( + x, memo, func, args, non_blocking=non_blocking + ) + elif len(rv) == 3: + func, args, state = rv + y = self._reconstruct( + x, + memo, + func, + args, + state, + non_blocking=non_blocking, + ) + elif len(rv) == 4: + func, args, state, listiter = rv + y = self._reconstruct( + x, + memo, + func, + args, + state, + listiter, + non_blocking=non_blocking, + ) + elif len(rv) == 5: + func, args, state, listiter, dictiter = rv + y = self._reconstruct( + x, + memo, + func, + args, + state, + listiter, + dictiter, + non_blocking=non_blocking, + ) + else: + raise RuntimeError( + f"Unexpected pickle protocol return value length: {len(rv)}" + ) + + # If it is its own copy, don't memoize. + if y is not x: + memo[d] = y + self._keep_alive(x, memo) # Make sure x lives at least as long as d + return y + + def _keep_alive(self, x, memo): + """Keeps a reference to the object x in the memo. + + Because we remember objects by their id, we have + to assure that possibly temporary objects are kept + alive by referencing them. + We store a reference at the id of the memo, which should + normally not be used unless someone tries to deepcopy + the memo itself... + """ + try: + memo[id(memo)].append(x) + except KeyError: + # aha, this is the first one :-) + memo[id(memo)] = [x] + + def _reconstruct( + self, + x, + memo, + func, + args, + state=None, + listiter=None, + dictiter=None, + non_blocking=False, + ): + deep = memo is not None + if deep and args: + args = tuple( + self.deepcopy_with_tensor_offload(arg, memo, non_blocking=non_blocking) + for arg in args + ) + y = func(*args) + if deep: + memo[id(x)] = y + + if state is not None: + if deep: + state = self.deepcopy_with_tensor_offload( + state, memo, non_blocking=non_blocking + ) + if hasattr(y, "__setstate__"): + y.__setstate__(state) + else: + if isinstance(state, tuple) and len(state) == 2: + state, slotstate = state + else: + slotstate = None + if state is not None: + y.__dict__.update(state) + if slotstate is not None: + for key, value in slotstate.items(): + setattr(y, key, value) + + if listiter is not None: + if deep: + for item in listiter: + item = self.deepcopy_with_tensor_offload( + item, memo, non_blocking=non_blocking + ) + y.append(item) + else: + for item in listiter: + y.append(item) + if dictiter is not None: + if deep: + for key, value in dictiter: + key = self.deepcopy_with_tensor_offload( + key, memo, non_blocking=non_blocking + ) + value = self.deepcopy_with_tensor_offload( + value, memo, non_blocking=non_blocking + ) + y[key] = value + else: + for key, value in dictiter: + y[key] = value + return y diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_storage_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_storage_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..261dab2a314a56cac2a1f01f04ca642f9c0f500b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_storage_utils.py @@ -0,0 +1,48 @@ +import os + +from .filesystem import FileSystemReader, FileSystemWriter +from .storage import StorageReader, StorageWriter + + +def _storage_setup( + storage: StorageReader | StorageWriter | None, + checkpoint_id: str | os.PathLike | None, + reader: bool = False, +) -> StorageReader | StorageWriter | None: + if storage: + if checkpoint_id is not None: + storage.reset(checkpoint_id) + return storage + + if not checkpoint_id: + raise RuntimeError( + "`checkpoint_id` must be specified if " + "storage_reader/storage_writer is None." + ) + + targets: list[type[StorageReader | StorageWriter]] = [] + if reader: + targets = [ + FileSystemReader, + ] + else: + targets = [ + FileSystemWriter, + ] + try: + from ._fsspec_filesystem import FsspecReader, FsspecWriter + + targets.append(FsspecReader if reader else FsspecWriter) + except Exception: + pass + + for target in targets: + if target.validate_checkpoint_id(checkpoint_id): + storage = target(checkpoint_id) # type: ignore[call-arg] + storage.reset(checkpoint_id) + return storage + + raise RuntimeError( + "Cannot detect which StorageReader or StorageWriter to use. " + "Please specify the storage_reader/storage_writer." + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_traverse.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_traverse.py new file mode 100644 index 0000000000000000000000000000000000000000..5c3ac34bbe8720f8302b008b79d2ae04d1907b20 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_traverse.py @@ -0,0 +1,208 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +# ruff: noqa: F821 +# flake8: noqa: F821 +from collections.abc import Callable, Collection, Mapping, MutableMapping +from typing import cast, TypeVar +from typing_extensions import TypeIs + +import torch +from torch.distributed._shard.sharded_tensor.api import ShardedTensor +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE +from torch.distributed.tensor import DTensor + + +PATH_ITEM = str | int +OBJ_PATH = tuple[PATH_ITEM, ...] +T = TypeVar("T") + +STATE_DICT_ITEM = object +CONTAINER_TYPE = MutableMapping[PATH_ITEM, STATE_DICT_ITEM] + +__all__ = ["traverse_state_dict", "set_element", "get_element", "print_tensor"] + + +def _keep_visiting_tensors(value: STATE_DICT_ITEM) -> TypeIs[torch.Tensor]: + return isinstance(value, torch.Tensor) + + +# TODO: update docstring for traverse.py +def traverse_state_dict( + state_dict: STATE_DICT_TYPE, + visitor: Callable[[OBJ_PATH, STATE_DICT_ITEM], None], + keep_traversing: Callable[[STATE_DICT_ITEM], bool] = _keep_visiting_tensors, +) -> None: + """ + Invoke ``visitor`` for each value recursively in ``state_dict``. + Mapping will be traversed and ``visitor`` will be applied to the leaf elements. + ``visitor`` will only be applied to elements in a list or a tuple, if the + container contains tensors or mappings. + """ + + def _is_terminal(value: STATE_DICT_ITEM) -> bool: + values: Collection[STATE_DICT_ITEM] + if isinstance(value, Mapping): + return False + elif isinstance(value, list): + values = value + else: + return True + + for entry in values: + if isinstance(entry, (Mapping, list)) and not _is_terminal(entry): + return False + if keep_traversing is not None and keep_traversing(entry): + return False + return True + + def _traverse_obj(path: OBJ_PATH, value: STATE_DICT_ITEM) -> None: + if isinstance(value, Mapping): + for k, v in value.items(): + _traverse_obj(path + (str(k),), v) + elif _is_terminal(value): + visitor(path, value) + elif isinstance(value, (list, tuple)): + for i, v in enumerate(value): + _traverse_obj(path + (i,), v) + + for key, value in state_dict.items(): + _traverse_obj((str(key),), value) + + # release reference cycle to prevent memory leaks in async_save + del _traverse_obj, _is_terminal + + +def traverse_state_dict_v_2_3( + state_dict: STATE_DICT_TYPE, + visitor: Callable[[OBJ_PATH, STATE_DICT_ITEM], None], + keep_traversing: Callable[[STATE_DICT_ITEM], bool] = _keep_visiting_tensors, +) -> None: + """ + Traversal is short-circuited when it finds a collection for which ``keep_visiting_tensors`` evaluates + to false for all elements. + By default, all collections with at least one ``torch.Tensor`` element are traversed. + Visitor takes a path argument that is a tuple of the keys used to reach it. + """ + + # a value is terminal if it has no other containers values inside it + def _is_terminal(value: STATE_DICT_ITEM) -> bool: + values: Collection[STATE_DICT_ITEM] + if isinstance(value, Mapping): + values = value.values() + elif isinstance(value, list): + values = value + else: + return True + + for entry in values: + if isinstance(entry, (Mapping, list)) and not _is_terminal(entry): + return False + if keep_traversing is not None and keep_traversing(entry): + return False + return True + + def _traverse_obj(path: OBJ_PATH, value: STATE_DICT_ITEM) -> None: + if _is_terminal(value): + visitor(path, value) + elif isinstance(value, Mapping): + for k, v in value.items(): + _traverse_obj(path + (str(k),), v) + elif isinstance(value, list): + for i, v in enumerate(value): + _traverse_obj(path + (i,), v) + + for key, value in state_dict.items(): + _traverse_obj((str(key),), value) + + # release reference cycle to prevent memory leaks in async_save + del _traverse_obj, _is_terminal + + +def set_element( + root_dict: STATE_DICT_TYPE, path: OBJ_PATH, value: STATE_DICT_ITEM +) -> None: + """Set ``value`` in ``root_dict`` along the ``path`` object path.""" + cur_container = cast(CONTAINER_TYPE, root_dict) + + def extend_list(lst: list[STATE_DICT_ITEM], idx: int) -> None: + while len(lst) <= idx: + lst.append(None) + + for i in range(1, len(path)): + prev_key = path[i - 1] + key = path[i] + def_val = cast(STATE_DICT_ITEM, {} if type(key) is str else []) + + if isinstance(cur_container, Mapping): + cur_container = cast( + CONTAINER_TYPE, cur_container.setdefault(prev_key, def_val) + ) + else: + # pyrefly: ignore [bad-argument-type] + extend_list(cur_container, prev_key) + if cur_container[prev_key] is None: + cur_container[prev_key] = def_val + cur_container = cur_container[prev_key] + + key = path[-1] + if type(key) is int: + extend_list(cast(list[STATE_DICT_ITEM], cur_container), key) + + cur_container[key] = value + + +def get_element( + root_dict: STATE_DICT_TYPE, + path: OBJ_PATH, + default_value: T | None = None, +) -> T | None: + """Retrieve the value at ``path``from ``root_dict``, returning ``default_value`` if not found.""" + cur_value = cast(CONTAINER_TYPE, root_dict) + for part in path: + if type(part) is int: + if not isinstance(cur_value, list) or len(cur_value) < part: + return default_value + elif not isinstance(cur_value, Mapping) or part not in cur_value: + return default_value + + cur_value = cast(CONTAINER_TYPE, cur_value[part]) + return cast(T | None, cur_value) + + +def _print_nested( + value: STATE_DICT_ITEM, + prefix: str = "", + print_fun: Callable[[str], None] = print, +) -> None: + if type(value) is ShardedTensor: + print_fun(f"{prefix} ShardedTensor size: {value.size()}") + for shard in value.local_shards(): + _print_nested( + shard.tensor, + f"{shard.metadata.shard_offsets} ", + print_fun=print_fun, + ) + elif type(value) is (DTensor): + print_fun(f"{prefix} DistributedTensor size: {value.size()}") + # TODO: add local offset for _local_tensor in print_nested. + _print_nested( + value._local_tensor, + print_fun=print_fun, + ) + elif isinstance(value, torch.Tensor): + print_fun(f"{prefix} Tensor size: {value.size()}") + else: + print_fun(f"{prefix} Type: {type(value)}") + + +def print_tensor( + path: OBJ_PATH, + value: STATE_DICT_ITEM, + print_fun: Callable[[str], None] = print, +) -> None: + """ + Use this callback with traverse_state_dict to print its content. + + By default the content is printed using the builtin ``print`` but this can + be changed by passing a different ``print_fun`` callable. + """ + _print_nested(value, prefix=str(path), print_fun=print_fun) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_version.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_version.py new file mode 100644 index 0000000000000000000000000000000000000000..07b99b2c8b582bea6aba25b276804721874251a2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/_version.py @@ -0,0 +1,4 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + + +_derived_version: str | None = None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/api.py new file mode 100644 index 0000000000000000000000000000000000000000..4d9817f235ca4863cf56dd668218b15cb09f6db9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/api.py @@ -0,0 +1,50 @@ +import traceback as tb +from typing import Any + + +WRAPPED_EXCEPTION = tuple[BaseException, tb.StackSummary] + +__all__ = ["CheckpointException"] + + +def _wrap_exception(exc: BaseException) -> WRAPPED_EXCEPTION: + summary = tb.extract_tb(exc.__traceback__) + # Python 3.13+ stores bytecode objects in FrameSummary._code, + # which cannot be pickled. Clear them so gather_object succeeds + # and the real exception is reported instead of a misleading + # "cannot pickle code objects" TypeError. + for frame in summary: + if hasattr(frame, "_code"): + object.__setattr__(frame, "_code", None) + return (exc, summary) + + +def _is_wrapped_exception(obj: Any) -> bool: + if not isinstance(obj, tuple): + return False + if len(obj) != 2: + return False + return isinstance(obj[0], BaseException) and isinstance(obj[1], tb.StackSummary) + + +class CheckpointException(BaseException): + """Exception raised if failure was detected as part of a checkpoint load or save.""" + + def __init__(self, msg: str, failures: dict[int, WRAPPED_EXCEPTION]): + super().__init__(msg, failures) + self._failures = failures + + @property + def failures(self) -> dict[int, WRAPPED_EXCEPTION]: + """Return a dictionary mapping node ranks to their associated exceptions in case of failure.""" + return self._failures + + def __str__(self) -> str: + str = f"CheckpointException ranks:{self._failures.keys()}\n" + for rank, exc_pair in self._failures.items(): + exc, trace = exc_pair + str += f"Traceback (most recent call last): (RANK {rank})\n" + if trace is not None: + str += "".join(tb.format_list(trace)) + str += "".join(tb.format_exception_only(type(exc), value=exc)) + return str diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/default_planner.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/default_planner.py new file mode 100644 index 0000000000000000000000000000000000000000..8cfe78e2e9e5dc052933e01fdc2ba8983ec897ce --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/default_planner.py @@ -0,0 +1,717 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates + +import dataclasses +import io +import logging +import math +import sys +from bisect import bisect_right, insort +from collections import ChainMap +from typing import Any, cast + +import torch +from torch.distributed._shard._utils import narrow_tensor_by_index +from torch.distributed.checkpoint._dedup_save_plans import dedup_save_plans +from torch.distributed.checkpoint._nested_dict import ( + FLATTEN_MAPPING, + flatten_state_dict, +) +from torch.distributed.checkpoint._sharded_tensor_utils import _flatten_sharded_tensors +from torch.distributed.checkpoint._traverse import set_element +from torch.distributed.checkpoint.metadata import ( + BytesStorageMetadata, + ChunkStorageMetadata, + Metadata, + MetadataIndex, + STATE_DICT_TYPE, + STORAGE_TYPES, + StorageMeta, + TensorStorageMetadata, +) +from torch.distributed.checkpoint.planner import ( + LoadPlan, + LoadPlanner, + ReadItem, + SavePlan, + SavePlanner, + WriteItem, + WriteItemType, +) +from torch.distributed.checkpoint.planner_helpers import ( + _compare_save_plans, + _contains_usable_plan, + _create_default_metadata_only_plan, + _create_read_items, + _create_write_items, + _init_state_dict, + _merge_delta_local_plans, +) +from torch.distributed.checkpoint.protocol import _is_checkpointable_tensor +from torch.distributed.checkpoint.utils import find_state_dict_object +from torch.distributed.tensor import DTensor + +from . import _version + + +logger: logging.Logger = logging.getLogger(__name__) + + +__all__ = [ + "DefaultSavePlanner", + "DefaultLoadPlanner", + "create_default_local_load_plan", + "create_default_global_load_plan", + "create_default_local_save_plan", + "create_default_global_save_plan", +] + + +# TODO: Update docstrings for default_planner.py +class DefaultSavePlanner(SavePlanner): + mappings: FLATTEN_MAPPING + + def __init__( + self, + flatten_state_dict: bool = True, + flatten_sharded_tensors: bool = True, + dedup_replicated_tensors: bool | None = None, + dedup_save_to_lowest_rank: bool = False, + enable_plan_caching: bool = False, + ) -> None: + self.flatten_state_dict = flatten_state_dict + self.flatten_sharded_tensors = flatten_sharded_tensors + self.mappings = {} + self.dedup_save_to_lowest_rank = dedup_save_to_lowest_rank + if dedup_replicated_tensors is not None: + logger.warning( + "DefaultSavePlanner's `dedup_replicated_tensors` argument is being " + "deprecated, and no longer has any effect. Please remove this argument " + "from your call." + ) + self._cached_plans_key: str = self.__class__.__name__ + self._enable_plan_caching = enable_plan_caching + + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + storage_meta: StorageMeta | None = None, + is_coordinator: bool = False, + ) -> None: + if self.flatten_state_dict: + state_dict, self.mappings = flatten_state_dict(state_dict) + if self.flatten_sharded_tensors: + state_dict = _flatten_sharded_tensors(state_dict) + self.state_dict = state_dict + self.is_coordinator = is_coordinator + + def create_local_plan(self) -> SavePlan: + plan = create_default_local_save_plan(self.state_dict, self.is_coordinator) + if self.flatten_state_dict: + plan = dataclasses.replace(plan, planner_data=self.mappings) + self.plan = plan + + if self._enable_plan_caching: + # If plans are equal, we can skip sending the plan to the coordinator. + if ( + self._cached_plans_key in SavePlanner._cached_save_plan + and _compare_save_plans( + plan, SavePlanner._cached_save_plan[self._cached_plans_key] + ) + ): + logger.info( + "No change in the local plan. Skipping sending the plan to the coordinator" + ) + return SavePlan([], usable=False) + else: + # Store the plan as pending. It will be promoted to the + # class-level cache in finish_plan after the global plan + # has succeeded. This avoids a stale local cache when + # the global plan fails (e.g. validation error) but the + # local cache was already populated. + self._pending_local_plan = plan + + return self.plan + + def _dedup_save_plans(self, all_plans: list[SavePlan]) -> list[SavePlan]: + return dedup_save_plans(all_plans, self.dedup_save_to_lowest_rank) + + def _create_global_plan( + self, all_plans: list[SavePlan] + ) -> tuple[list[SavePlan], Metadata]: + deduped_plans = self._dedup_save_plans(all_plans) + + global_plan, metadata = create_default_global_save_plan(deduped_plans) + + if self.flatten_state_dict: + # | does not work for Python 3.8 or older version. + # merged_mappings = reduce( + # lambda x, y: x | y, (p.planner_data for p in global_plan) + # ) + planner_data_dict = [p.planner_data for p in global_plan] + merged_mappings = dict(ChainMap(*planner_data_dict)) + metadata = dataclasses.replace(metadata, planner_data=merged_mappings) + + validation_errors = _validate_global_plan(global_plan, metadata) + if validation_errors: + error_summary = "; ".join(validation_errors) + if len(error_summary) > 500: + error_summary = error_summary[:500] + "... (truncated)" + raise ValueError(f"Failed to validate global plan: {error_summary}") + + return global_plan, metadata + + def _create_global_plan_with_caching( + self, all_plans: list[SavePlan] + ) -> tuple[list[SavePlan], list[SavePlan], Metadata]: + """ + Create global plan with caching. + Returns a tuple of global_plan_delta, global_plan, metadata. + """ + global_plan_delta: list[SavePlan] = [] + + if self._cached_plans_key not in SavePlanner._cached_all_plans: + # Case 1: If the plans are not cached, the cache will be hydrated with the + # all_plans, global_plans (Deduped), and metadata. + + # First create and validate the global plan. Only cache everything + # after success to avoid partial cache state + global_plan, metadata = self._create_global_plan(all_plans) + + # Cache all plans atomically after successful validation + SavePlanner._cached_all_plans[self._cached_plans_key] = all_plans + SavePlanner._cached_global_plan[self._cached_plans_key] = global_plan + SavePlanner._cached_metadata[self._cached_plans_key] = metadata + # If plans are not cached, global_plan delta will be the same as global plan. + return global_plan, global_plan, metadata + + # Case 2: Plans are cached + if not _contains_usable_plan(all_plans): + # Case 2.1: Plans are cached and the local plans have NOT changed (No usable plans). + # Global plan delta will be empty plans to avoid the collective overhead. + # We can reuse the deduped global plan and metadata from the cache directly. + global_plan_delta = [SavePlan([], usable=False)] * len(all_plans) + global_plan = SavePlanner._cached_global_plan[self._cached_plans_key] + metadata = SavePlanner._cached_metadata[self._cached_plans_key] + else: + # Case 2.2: Plans are cached but the local plans have changed. + # We will merge the changed local plans with the cached local plans. + # Updated plans will overwrite the cached plans. New global plan and metadata will be created and cached. + # Global plan delta will be created by comparing the new global plan with the cached global plan. + # Only the global plan delta (updated ones) will be sent to the coordinator to avoid the collective overhead. + merged_plans = _merge_delta_local_plans( + SavePlanner._cached_all_plans[self._cached_plans_key], all_plans + ) + # Cache the updated local plans + SavePlanner._cached_all_plans[self._cached_plans_key] = merged_plans + global_plan, metadata = self._create_global_plan(merged_plans) + + if self._cached_plans_key in self._cached_global_plan: + for cached_plan, new_plan in zip( + SavePlanner._cached_global_plan[self._cached_plans_key], global_plan + ): + if _compare_save_plans(cached_plan, new_plan): + global_plan_delta.append(SavePlan([], usable=False)) + else: + global_plan_delta.append(new_plan) + + # Cache the new global plan and the metadata + SavePlanner._cached_global_plan[self._cached_plans_key] = global_plan + SavePlanner._cached_metadata[self._cached_plans_key] = metadata + + return global_plan_delta, global_plan, metadata + + def create_global_plan( + self, all_plans: list[SavePlan] + ) -> tuple[list[SavePlan], Metadata]: + global_plan_delta: list[SavePlan] = [] + if self._enable_plan_caching: + # If the plans are cached, we only need to send the global plan delta to be scattered + # across ranks. Ranks will use the cached final plans instead. + ( + global_plan_delta, + global_plan, + metadata, + ) = self._create_global_plan_with_caching(all_plans) + else: + global_plan, metadata = self._create_global_plan(all_plans) + # If the caching is not enabled, global delta plan will always be same as the new global plan. + global_plan_delta = global_plan + + self.global_plan = global_plan + self.metadata = metadata + + return global_plan_delta, self.metadata + + def _finish_plan_with_caching(self, new_plan: SavePlan) -> SavePlan: + finished_plan: SavePlan = new_plan + + if not new_plan.usable: + finished_plan = SavePlanner._cached_final_save_plan[self._cached_plans_key] + else: + finished_plan = new_plan + SavePlanner._cached_final_save_plan[self._cached_plans_key] = new_plan + return finished_plan + + def finish_plan(self, new_plan: SavePlan) -> SavePlan: + finished_plan: SavePlan = new_plan + + if self._enable_plan_caching: + finished_plan = self._finish_plan_with_caching(new_plan) + + # Promote the pending local plan to the class-level cache now + # that the global plan has succeeded and we are finalizing. + # This ensures the local cache is only populated after a + # successful end-to-end checkpoint plan creation. + if hasattr(self, "_pending_local_plan"): + SavePlanner._cached_save_plan[self._cached_plans_key] = ( + self._pending_local_plan + ) + del self._pending_local_plan + + self.plan = finished_plan + return self.plan + + def resolve_data(self, write_item: WriteItem) -> torch.Tensor | io.BytesIO: + object = self.lookup_object(write_item.index) + return self.transform_object(write_item, object) + + def lookup_object(self, index: MetadataIndex) -> Any: + """Extension from the planner interface to make it easy to extend the default planner.""" + return find_state_dict_object(self.state_dict, index) + + def transform_object(self, write_item: WriteItem, object: Any): + """Extension from the planner interface to make it easy to extend the default planner.""" + if write_item.type == WriteItemType.BYTE_IO: + bytes = io.BytesIO() + torch.save(object, bytes) + object = bytes + return object + + +class DefaultLoadPlanner(LoadPlanner): + """ + DefaultLoadPlanner that adds multiple features on top of LoadPlanner. + + In particular it adds the following: + + flatten_state_dict: Handle state_dict with nested dicts + flatten_sharded_tensors: For FSDP in 2D parallel mode + allow_partial_load: If False, will raise a runtime error if a key is present in state_dict, but not in the checkpoint. + """ + + original_state_dict: STATE_DICT_TYPE + mappings: FLATTEN_MAPPING + + def __init__( + self, + flatten_state_dict: bool = True, + flatten_sharded_tensors: bool = True, + allow_partial_load: bool = False, + ) -> None: + self.flatten_state_dict = flatten_state_dict + self.flatten_sharded_tensors = flatten_sharded_tensors + self.original_state_dict = {} + self.mappings = {} + self.allow_partial_load = allow_partial_load + + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + metadata: Metadata | None = None, + is_coordinator: bool = False, + ) -> None: + _init_state_dict(state_dict) + self.original_state_dict = state_dict + + if self.flatten_sharded_tensors: + state_dict = _flatten_sharded_tensors(state_dict) + + if self.flatten_state_dict: + state_dict, self.mappings = flatten_state_dict(state_dict) + + self.state_dict = state_dict + self.metadata = metadata + self.is_coordinator = is_coordinator + + def create_local_plan(self) -> LoadPlan: + if self.metadata is None: + raise AssertionError("self.metadata is not None") + if self.flatten_state_dict: + # To support checkpoints that are saved before v2.4, we have to + # differentiate if the missing keys are due to old checkpoints. + # The contracts are: + # 1. There are 3 cases when we found a missing key. + # 1.1 Actual missing key, but allow_partial_load is False + # 1.2 Actual missing key, but allow_partial load is True + # 1.3 Old checkpoint, but allow_partial_load is False + # 1.4 Old checkpoint, but allow_partial_load is True + # 2. If we found a missing key, we first convert the keys back to + # the key format of v2.3 + # 3. If the previous missing keys are in the v2.3 keys, we assume + # this is an old checkpoint. + # 4. Pass the state_dict to `create_default_local_load_plan()`, + # which has the logic to check missing for allow_partial_load. + # So for 1.2 and 1.4 cases, we delegate allow_partial_load check to + # `create_default_local_load_plan()`. The logic here is to determine + # whether the checkpoint belong to 2.3 (or before) or 2.4 (or after). + current_keys = set(self.state_dict.keys()) + load_keys = set(self.metadata.state_dict_metadata.keys()) + missing_keys = load_keys - current_keys + if missing_keys: + _version._derived_version = "2_3" + old_state_dict, old_mappings = flatten_state_dict( + self.original_state_dict + ) + old_keys = set(old_state_dict.keys()) + if old_keys & missing_keys: + self.state_dict, self.mappings = old_state_dict, old_mappings + # _derived_version is only used by flatten_state_dict now. + # Set it back to None so that later we can save to a new version. + _version._derived_version = None + + return create_default_local_load_plan( + self.state_dict, self.metadata, not self.allow_partial_load + ) + + def create_global_plan(self, global_plan: list[LoadPlan]) -> list[LoadPlan]: + return create_default_global_load_plan(global_plan) + + def finish_plan(self, new_plan: LoadPlan) -> LoadPlan: + return new_plan + + def load_bytes(self, read_item: ReadItem, value: io.BytesIO) -> None: + if self.flatten_state_dict: + set_element( + self.original_state_dict, + self.mappings[read_item.dest_index.fqn], + torch.load(value, weights_only=False), + ) + else: + self.state_dict[read_item.dest_index.fqn] = torch.load( + value, weights_only=False + ) + + def resolve_tensor(self, read_item: ReadItem): + tensor = self.lookup_tensor(read_item.dest_index) + return self.transform_tensor(read_item, tensor) + + def commit_tensor(self, read_item: ReadItem, tensor: torch.Tensor) -> None: + pass + + def lookup_tensor(self, index: MetadataIndex) -> torch.Tensor: + """Extension from the planner interface to make it easy to extend the default planner.""" + return find_state_dict_object(self.state_dict, index) + + def transform_tensor(self, read_item: ReadItem, tensor: torch.Tensor): + """Extension from the planner interface to make it easy to extend the default planner.""" + return narrow_tensor_by_index(tensor, read_item.dest_offsets, read_item.lengths) + + +class _EmptyStateDictLoadPlanner(DefaultLoadPlanner): + """ + Extension of DefaultLoadPlanner, which rebuilds state_dict from the saved metadata. + Useful for loading in state_dict without first initializing a model, such as + when converting a DCP checkpoint into a Torch save file. + + . N.B. `state_dict` must be an empty dictionary when used with this LoadPlanner + + .. warning:: + Because the entire state dict is initialized, It's recommended to only utilize + this LoadPlanner on a single rank or process to avoid OOM. + + """ + + def __init__(self, keys=None, *args, **kwargs): + self.keys = keys + super().__init__(*args, **kwargs) + + def _should_include_key(self, key: str, metadata: Metadata) -> bool: + if self.keys is None: + return True + + if key in self.keys: + return True + + unflattened_keys: list[str] = [] + planner_data = metadata.planner_data.get(key) + for unflattened_key in planner_data: + if unflattened_keys: + unflattened_keys.append( + ".".join([unflattened_keys[-1], str(unflattened_key)]) + ) + + else: + unflattened_keys.append(unflattened_key) + + if any(unflattened_key in self.keys for unflattened_key in unflattened_keys): + return True + + return False + + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + metadata: Metadata | None = None, + is_coordinator: bool = False, + ) -> None: + if state_dict: + raise AssertionError("not state_dict") + if metadata is None: + raise AssertionError("metadata is not None") + + # rebuild the state dict from the metadata + for k, v in metadata.state_dict_metadata.items(): + if not self._should_include_key(k, metadata): + continue + + if isinstance(v, TensorStorageMetadata): + v = torch.empty(v.size, dtype=v.properties.dtype) # type: ignore[assignment] + if metadata.planner_data is not None and k in metadata.planner_data: + set_element(state_dict, metadata.planner_data[k], v) + else: + state_dict[k] = v + + super().set_up_planner(state_dict, metadata, is_coordinator) + + +def create_default_local_load_plan( + state_dict: dict[str, Any], metadata: Metadata, strict: bool = True +) -> LoadPlan: + requests = [] + """ + Create the ``LoadPlan`` used by DefaultLoadPlanner. + + It produces one read item per value in ``state_dict`` using the metadata in ``metadata``. + + The default behavior is to match key exactly between state_dict and metadata. + It handles resharding by issuing multiple read requests against storage in order to match + load requirements. + """ + + for fqn, obj in state_dict.items(): + # ignore state_dict keys which do not exist in `state_dict` if strict=False + if fqn not in metadata.state_dict_metadata: + if strict: + raise RuntimeError(f"Missing key in checkpoint state_dict: {fqn}.") + else: + continue + + md = metadata.state_dict_metadata[fqn] + if ( + isinstance(md, TensorStorageMetadata) + and getattr(obj, "size", None) is not None + ): + obj_size = ( + torch.Size(obj.global_shape) + if _is_checkpointable_tensor(obj) + else obj.size() + ) + if md.size != obj_size: + raise ValueError( + f"Size mismatch between saved {md.size} and current: {obj_size} for {fqn}", + ) + # Since DTensor supports submesh, adding extra check to ensure _create_read_items() + # gets called only when the current rank is part of the mesh for the corresponding DTensor. + if isinstance(obj, DTensor): + if obj.device_mesh.get_coordinate() is not None: + requests += _create_read_items(fqn, md, obj) + else: + requests += _create_read_items(fqn, md, obj) + + return LoadPlan(requests) + + +def create_default_global_load_plan( + all_plans: list[LoadPlan], +) -> list[LoadPlan]: + """ + Create global load plan used by DefaultLoadPlanner. + + The default load behavior involved no global coordination and this function + currently doesn't change the local plans. + """ + return all_plans + + +def create_default_local_save_plan( + state_dict: dict[str, Any], is_coordinator: bool +) -> SavePlan: + """ + Create the ``SavePlan`` used by DefaultSavePlanner. + + On non-coordinator ranks, this function ignores tensors and non-tensor objects, + only producing writes for ShardedTensor objects. + + On the coordinator rank, produce writes for all values. + """ + requests = [] + for fqn, obj in state_dict.items(): + # Since DTensor supports submesh, adding extra check to ensure _create_write_items() + # gets called only when the current rank is part of the mesh for the corresponding DTensor. + if isinstance(obj, DTensor): + if obj.device_mesh.get_coordinate() is not None: + requests += _create_write_items(fqn, obj) + else: + # For the plain tensor and non-tensor values, add the request for all + # the ranks. Coordinator will decide whether to deduplicate the + # values based on the keys. + requests += _create_write_items(fqn, obj) + + return SavePlan(requests) + + +def create_default_global_save_plan( + all_plans: list[SavePlan], + rewrite_index_hints: bool = True, +) -> tuple[list[SavePlan], Metadata]: + """ + Create the global plan and metadata used by DefaultSavePlanner. + + Metadata is produced by concatenating the metadata of all ``WriteItem`` from the supplied plans. + + The only global planning change is to update index hints in all ``MetadataIndex`` objects if + ``rewrite_index_hints`` is True. + """ + md: dict[str, STORAGE_TYPES] = {} + new_plans = [] + for plan in all_plans: + new_items = [] + for item in plan.items: + if item.type != WriteItemType.SHARD: + if item.index.fqn in md: + raise AssertionError("item.index.fqn not in md") + + if item.type == WriteItemType.BYTE_IO: + md[item.index.fqn] = BytesStorageMetadata() + new_items.append(item) + else: + if item.tensor_data is None: + raise AssertionError("item.tensor_data is not None") + tensor_md = cast( + TensorStorageMetadata, + md.setdefault( + item.index.fqn, + TensorStorageMetadata( + properties=item.tensor_data.properties, + size=item.tensor_data.size, + chunks=[], + ), + ), + ) + new_item = item + if rewrite_index_hints: + new_index = dataclasses.replace( + item.index, index=len(tensor_md.chunks) + ) + new_item = dataclasses.replace(item, index=new_index) + new_items.append(new_item) + + if item.tensor_data.chunk is None: + raise AssertionError(f""" + Cannot create MD for tensor without bounds. + FQN: {item.index.fqn} + """) + tensor_md.chunks.append(item.tensor_data.chunk) + new_plans.append(dataclasses.replace(plan, items=new_items)) + return (new_plans, Metadata(md)) + + +def _create_default_local_metadata(state_dict: STATE_DICT_TYPE) -> Metadata: + """Return the ``Metadata`` if DefaultSavePlanner was used to checkpoint ``state_dict``.""" + plan = _create_default_metadata_only_plan(state_dict) + _, md = create_default_global_save_plan([plan]) + return md + + +def _check_box_overlap(box0: ChunkStorageMetadata, box1: ChunkStorageMetadata) -> bool: + """Check if two boxes overlap. Tuples are (offset, lengths).""" + # For each dim of each shard, check if one shard resides on the other + # end of second shard with respect to that dim. As an example for a 2D + # shard, we would check if one shard is above or on the left of the + # other shard. + ndims = len(box0.offsets) + for i in range(ndims): + if box0.offsets[i] >= box1.offsets[i] + box1.sizes[i]: + return False + if box1.offsets[i] >= box0.offsets[i] + box0.sizes[i]: + return False + + return True + + +def _check_box_bounds( + outer_box_size: torch.Size, inner_box: ChunkStorageMetadata +) -> bool: + for i in range(len(outer_box_size)): + if inner_box.offsets[i] < 0: + return False + if inner_box.sizes[i] < 0: + return False + if inner_box.offsets[i] + inner_box.sizes[i] > outer_box_size[i]: + return False + + return True + + +def _validate_global_plan(global_plan: list[SavePlan], metadata: Metadata) -> list[str]: + """Validate the global plan and return a list of error messages (empty if valid).""" + errors: list[str] = [] + for key, value in metadata.state_dict_metadata.items(): + if isinstance(value, BytesStorageMetadata): + continue + if len(value.size) == 0: + continue + chunks = value.chunks + chunks_volume = 0 + for chunk in chunks: + # Compute the volume + if not _check_box_bounds(value.size, chunk): + msg = ( + f"key:{key} has out of bounds chunk: " + f"tensor-size:{value.size} chunk: {chunk}" + ) + logger.warning(msg) + errors.append(msg) + chunks_volume += math.prod(chunk.sizes) + + if len(chunks) > 1: + dims = len(value.size) + sweep_dim = max(range(dims), default=0, key=lambda d: value.size[d]) + sorted_indices = sorted( + range(len(chunks)), + key=lambda idx: ( + chunks[idx].offsets[sweep_dim], + *(chunks[idx].offsets[d] for d in range(dims)), + ), + ) + active: list[tuple[int, int]] = [] + for idx in sorted_indices: + current = chunks[idx] + start = current.offsets[sweep_dim] + end = start + current.sizes[sweep_dim] + + cutoff = bisect_right(active, (start, sys.maxsize)) + if cutoff: + del active[:cutoff] + + for _, other_idx in active: + other = chunks[other_idx] + if _check_box_overlap(current, other): + msg = f"key:{key} has overlapping chunks: {current} {other}" + logger.warning(msg) + errors.append(msg) + + insort(active, (end, idx)) + + # Check whether combined chunk cover the whole tensor + tensor_volume = math.prod(value.size) + if len(global_plan) > 1 and chunks_volume != tensor_volume: + msg = ( + f"key:{key} invalid fill tensor-volume: " + f"{tensor_volume} chunks-volume: {chunks_volume}" + ) + logger.warning(msg) + errors.append(msg) + + return errors diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/filesystem.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/filesystem.py new file mode 100644 index 0000000000000000000000000000000000000000..43f85eb52025ababb9b185e9e305f8ba8c20cf79 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/filesystem.py @@ -0,0 +1,1038 @@ +# mypy: allow-untyped-defs +import collections +import dataclasses +import io +import json +import operator +import os +import pickle +import queue +import threading +import uuid +import warnings +from abc import ABC, abstractmethod +from collections.abc import Callable, Generator, Iterable, Iterator, Sequence +from contextlib import contextmanager +from dataclasses import dataclass +from enum import Enum +from io import UnsupportedOperation +from pathlib import Path +from typing import Any, cast, Final, IO + +# introduced as collections.abc.Buffer in Python 3.12 +from typing_extensions import Buffer + +import torch +from torch import Tensor +from torch._utils import _get_available_device_type, _get_device_module +from torch.distributed._shard._utils import narrow_tensor_by_index +from torch.distributed.checkpoint._extension import ( + ExtensionRegistry, + StreamTransformExtension, +) +from torch.distributed.checkpoint._hf_utils import ( + CUSTOM_METADATA_KEY, + DCP_VERSION_KEY, + FORMAT_KEY, + FORMAT_VALUE, + HF_DCP_VERSION, +) +from torch.distributed.checkpoint.metadata import Metadata, STATE_DICT_TYPE, StorageMeta +from torch.distributed.checkpoint.planner import ( + LoadItemType, + LoadPlan, + LoadPlanner, + ReadItem, + SavePlan, + SavePlanner, + WriteItem, + WriteItemType, +) +from torch.distributed.checkpoint.staging import BlockingAsyncStager +from torch.distributed.checkpoint.storage import ( + StorageReader, + StorageWriter, + WriteResult, +) +from torch.distributed.checkpoint.utils import _create_file_view +from torch.futures import Future + + +__all__ = [ + "FileSystemWriter", + "FileSystemReader", + "FileSystem", + "FileSystemBase", + "SerializationFormat", +] + +_metadata_fn: str = ".metadata" + +CURRENT_DCP_VERSION: Final[str] = "1.0.0" + + +@dataclass +class _StorageInfo: + """This is the per entry storage info.""" + + relative_path: str + offset: int + length: int + transform_descriptors: Sequence[str] | None = None + + def __getstate__(self): + return {k: v for k, v in self.__dict__.items() if v is not None} + + +@dataclass +class _StoragePrefix: + prefix: str + + +class SerializationFormat(Enum): + TORCH_SAVE = "torch_save" + SAFETENSORS = "safetensors" + + +DEFAULT_SUFFIX = ".distcp" + + +def _generate_uuid() -> str: + return str(uuid.uuid4()) + + +class _TensorLoader(ABC): + @abstractmethod + def add(self, size: int, obj: object) -> None: + pass + + @abstractmethod + def start_loading(self) -> None: + pass + + @abstractmethod + def values(self) -> Iterator[tuple[torch.Tensor, object]]: + pass + + +class _SerialCpuLoader(_TensorLoader): + def __init__(self, resolve_fun: Callable) -> None: + self.resolve_fun = resolve_fun + self.items: list[tuple[int, object]] = [] + + def add(self, size: int, obj: object) -> None: + self.items.append((size, obj)) + + def start_loading(self) -> None: + pass + + def values(self) -> Iterator[tuple[torch.Tensor, object]]: + for _, obj in self.items: + tensor = self.resolve_fun(obj).detach() + tensor = tensor.cpu() + if tensor.untyped_storage().size() != tensor.nbytes: + # creates a new tensor with minimal storage while preserving memory format. + tensor = tensor.clone() + yield ( + tensor, + obj, + ) + + +class _OverlappingCpuLoader(_TensorLoader): + def __init__( + self, + resolve_fun: Callable, + stream: torch.Stream | None = None, + inflight_threshhold: int = 1_000_000, + ) -> None: + self.resolve_fun = resolve_fun + self.items: list[tuple[int, object]] = [] + self.inflight_threshhold = inflight_threshhold + self.in_flight_data = 0 + self.current_items: collections.deque = collections.deque() + self.idx = 0 + self.started = False + self.device_type = ( + stream.device_type if stream else _get_available_device_type() + ) + self.device_module = _get_device_module(self.device_type) + self.stream = cast( + torch.cuda.Stream, stream or self.device_module.current_stream() + ) + if self.stream != self.device_module.current_stream(): + self.stream.wait_stream(self.device_module.current_stream()) + + @property + def _done(self) -> bool: + return self.idx >= len(self.items) + + def _drain(self) -> list[tuple[torch.Tensor, object]]: + drained = [] + if self.in_flight_data >= self.inflight_threshhold: + self.stream.synchronize() + while self.in_flight_data >= self.inflight_threshhold: + val = self.current_items.popleft() + self.in_flight_data -= val[0].numel() * val[0].element_size() + drained.append(val) + return drained + + def _refill(self) -> None: + with self.device_module.stream(self.stream): + while not self._done and self.in_flight_data < self.inflight_threshhold: + _, obj = self.items[self.idx] + self.idx += 1 + tensor = self.resolve_fun(obj).detach() + if tensor.device.type == self.device_type: + tensor = tensor.to(device="cpu", non_blocking=True) + elif tensor.device == torch.device("cpu"): + if ( + tensor.untyped_storage().size() + != tensor.numel() * tensor.itemsize + ): + # creates a new tensor with minimal storage while preserving memory format. + tensor = tensor.clone() + + self.current_items.append( + ( + tensor, + obj, + ) + ) + self.in_flight_data += tensor.numel() * tensor.element_size() + + def _finish(self) -> Iterable[tuple[torch.Tensor, object]]: + if not self._done: + raise AssertionError("_finish called before all items were processed") + if len(self.current_items) > 0: + self.stream.synchronize() + return self.current_items + + def add(self, size: int, obj: object) -> None: + if self.started: + raise RuntimeError("cannot add items after loading started") + self.items.append((size, obj)) + + def start_loading(self) -> None: + if self.started: + return + self.started = True + self.items.sort(key=operator.itemgetter(0)) + self._refill() + + def values(self) -> Iterator[tuple[torch.Tensor, object]]: + self.start_loading() + while not self._done: + drained = self._drain() + self._refill() + yield from drained + + yield from self._finish() + + +class _StorageWriterTransforms: + """ + This is experimental, and will likely move elsewhere in the + future. It lives here to minimize changes while we are still + learning and gathering feedback. + """ + + def __init__( + self, extensions: Sequence[StreamTransformExtension] | None = None + ) -> None: + """ + If the extensions arg is None, this means the implementation + should provide whatever defaults it chooses. An empty + sequence indicates no extensions should be used. At this + time, the default extensions sequence is empty. + """ + self.extensions = () if extensions is None else extensions + + def transform_save_stream( + self, write_item: WriteItem, raw_stream: io.IOBase + ) -> tuple[IO[bytes], list[str]]: + # In order to avoid leaking fds, transformers' close must + # cascade to wrapped streams, but since this function can + # append to the raw stream, we can't close the actual stream. + # So, we use this to put a wrapper around the raw stream's + # close() to make it a noop, and it gets closed once all files + # are appended. + + class NoCloseWriter(io.IOBase): + def __init__(self, raw: io.IOBase): + self.raw = raw + + def writeable(self) -> bool: + return True + + def write(self, b: Buffer) -> int: + return self.raw.write(b) + + def close(self): + self.flush() + self.raw.flush() + # but not close. + + transform_to = cast(IO[bytes], NoCloseWriter(raw_stream)) + + for ex in self.extensions: + transform_to = ex.transform_to(transform_to) + + return (transform_to, [ex.get_descriptor() for ex in reversed(self.extensions)]) + + +def _item_size(item: WriteItem) -> int: + size = 1 + if item.tensor_data is None: + raise AssertionError("WriteItem tensor_data must not be None") + # can't use math.prod as PT needs to support older python + for s in item.tensor_data.size: + size *= s + + dtype = item.tensor_data.properties.dtype + return size * torch._utils._element_size(dtype) + + +def _split_by_size_and_type(bins: int, items: list[WriteItem]) -> list[list[WriteItem]]: + if bins == 1: + return [items] + + bytes_w = [wi for wi in items if wi.type == WriteItemType.BYTE_IO] + tensor_w = [wi for wi in items if wi.type != WriteItemType.BYTE_IO] + + buckets: list[list[WriteItem]] = [[] for _ in range(bins)] + bucket_sizes = [0 for _ in range(bins)] + + tensor_w.sort(key=_item_size, reverse=True) + + for i, wi in enumerate(bytes_w): + buckets[i % bins].append(wi) + + for wi in tensor_w: + # TODO replace with heapq + idx = min(enumerate(bucket_sizes), key=operator.itemgetter(1))[0] + buckets[idx].append(wi) + bucket_sizes[idx] += _item_size(wi) + + return buckets + + +def _write_item( + transforms: _StorageWriterTransforms, + stream: io.IOBase, + data: io.BytesIO | torch.Tensor, + write_item: WriteItem, + storage_key: str, + serialization_format: SerializationFormat, +) -> WriteResult: + offset = stream.tell() + + (transform_to, transform_descriptors) = transforms.transform_save_stream( + write_item, stream + ) + + if write_item.type == WriteItemType.BYTE_IO: + if not isinstance(data, io.BytesIO): + raise AssertionError("Data must be io.BytesIO for BYTE_IO write items") + transform_to.write(data.getbuffer()) + else: + if not isinstance(data, torch.Tensor): + raise AssertionError( + "Data must be torch.Tensor for non-BYTE_IO write items" + ) + if data.device != torch.device("cpu"): + raise AssertionError("Tensor must be on CPU device") + if serialization_format == SerializationFormat.TORCH_SAVE: + torch.save(data, transform_to) + + transform_to.close() + + if serialization_format == SerializationFormat.TORCH_SAVE or isinstance( + data, io.BytesIO + ): + length = stream.tell() - offset + else: + length = data.numel() * data.element_size() + + # For consistency with earlier versions, leave this field out of the + # metadata if there are no extensions. + info_transform_descriptors = ( + None if len(transform_descriptors) == 0 else transform_descriptors + ) + + return WriteResult( + index=write_item.index, + size_in_bytes=length, + storage_data=_StorageInfo( + storage_key, + offset, + length, + transform_descriptors=info_transform_descriptors, + ), + ) + + +def _write_files_from_queue( + create_stream: Callable, + file_queue: queue.Queue, + result_queue: queue.Queue, + planner: SavePlanner, + transforms: _StorageWriterTransforms, + inflight_threshhold: int, + use_fsync: bool, + thread_count: int, + serialization_format: SerializationFormat, +) -> None: + try: + while True: + file_name, storage_key, write_items = file_queue.get_nowait() + loader: _TensorLoader + + custom_backend_name = torch._C._get_privateuse1_backend_name() + custom_device_mod = getattr(torch, custom_backend_name, None) + + # TODO: Using the OverlappingCpuLoader with multiple threads creates significant + # performance degradation, observed as being related to cuda stream syncs. We + # should try to fix this and use _OverlappingCpuLoader for all threaded cases + if ( + thread_count == 1 + and ( + torch.cuda.is_available() + or (custom_device_mod and custom_device_mod.is_available()) + ) + and inflight_threshhold > 0 + ): + loader = _OverlappingCpuLoader( + planner.resolve_data, + inflight_threshhold=inflight_threshhold, + ) + else: + loader = _SerialCpuLoader( + planner.resolve_data, + ) + + tensor_w = [wi for wi in write_items if wi.type != WriteItemType.BYTE_IO] + for write_item in tensor_w: + loader.add(_item_size(write_item), write_item) + loader.start_loading() + + bytes_w = [wi for wi in write_items if wi.type == WriteItemType.BYTE_IO] + write_results = [] + + with create_stream(file_name, "wb") as stream: + for write_item in bytes_w: + data = planner.resolve_data(write_item) + write_results.append( + _write_item( + transforms, + stream, + data, + write_item, + storage_key, + serialization_format, + ) + ) + + tensor_dict = {} + metadata_dict = {} + for tensor, write_item in loader.values(): + if not tensor.is_cpu: + raise AssertionError("Tensor must be on CPU") + write_results.append( + _write_item( + transforms, + stream, + tensor, + write_item, # type: ignore[arg-type] + storage_key, + serialization_format, + ) + ) + tensor_dict[write_item.index.fqn] = tensor # type: ignore[attr-defined] + metadata_dict[write_item.index.fqn] = { # type: ignore[attr-defined] + "saved_offsets": write_item.tensor_data.chunk.offsets # type: ignore[attr-defined] + } + + if serialization_format == SerializationFormat.SAFETENSORS: + from safetensors.torch import save # type: ignore[import-not-found] + + stream.write( + save( + tensor_dict, + metadata={ + CUSTOM_METADATA_KEY: json.dumps(metadata_dict), + DCP_VERSION_KEY: str(HF_DCP_VERSION), + FORMAT_KEY: FORMAT_VALUE, + }, + ) + ) + + if use_fsync: + # Flush Python-level buffers (OS for local files, network for cloud storage) before fsync. + stream.flush() + try: + os.fsync(stream.fileno()) + except (AttributeError, UnsupportedOperation) as e: + warnings.warn( + f"fsync not supported for this stream, relying on flush(): {e}" + ) + stream.close() + result_queue.put(write_results) + except queue.Empty: + pass + + +class FileSystemBase(ABC): + @contextmanager + @abstractmethod + def create_stream( + self, path: str | os.PathLike, mode: str + ) -> Generator[io.IOBase, None, None]: ... + + @abstractmethod + def concat_path( + self, path: str | os.PathLike, suffix: str + ) -> str | os.PathLike: ... + + @abstractmethod + def rename(self, path: str | os.PathLike, new_path: str | os.PathLike) -> None: ... + + @abstractmethod + def init_path(self, path: str | os.PathLike) -> str | os.PathLike: ... + + @abstractmethod + def mkdir(self, path: str | os.PathLike) -> None: ... + + @classmethod + @abstractmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: ... + + @abstractmethod + def exists(self, path: str | os.PathLike) -> bool: ... + + @abstractmethod + def rm_file(self, path: str | os.PathLike) -> None: ... + + +class FileSystem(FileSystemBase): + @contextmanager + def create_stream( + self, path: str | os.PathLike, mode: str + ) -> Generator[io.IOBase, None, None]: + if not isinstance(path, Path): + path = Path(path) + with path.open(mode) as stream: + yield cast(io.IOBase, stream) + + def concat_path(self, path: str | os.PathLike, suffix: str) -> str | os.PathLike: + if not isinstance(path, Path): + path = Path(path) + return path / suffix + + def init_path(self, path: str | os.PathLike) -> str | os.PathLike: + if not isinstance(path, Path): + path = Path(path) + return path + + def rename(self, path: str | os.PathLike, new_path: str | os.PathLike) -> None: + if not isinstance(path, Path): + path = Path(path) + + path.rename(cast(Path, new_path)) + + def mkdir(self, path: str | os.PathLike) -> None: + if not isinstance(path, Path): + path = Path(path) + path.mkdir(parents=True, exist_ok=True) + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + if isinstance(checkpoint_id, Path): + return True + + if "://" in str(checkpoint_id): + return False + + for p in Path(checkpoint_id).parents: + if p.exists() and os.access(str(p), os.W_OK): + return True + + return False + + def exists(self, path: str | os.PathLike) -> bool: + if not isinstance(path, Path): + path = Path(path) + return path.exists() + + def rm_file(self, path: str | os.PathLike) -> None: + if not isinstance(path, Path): + path = Path(path) + path.unlink() + + def ls(self, path: str | os.PathLike) -> list[str]: + if not isinstance(path, Path): + path = Path(path) + return [str(p) for p in path.iterdir()] + + +class _FileSystemWriter(StorageWriter): + """ + Basic implementation of StorageWriter using file IO. + + This implementation makes the following assumptions and simplifications: + + * The checkpoint path is an empty or non-existing directory. + * File creation is atomic + + The checkpoint consist of one file per write request plus + a `.metadata` file with the serialized metadata. + + """ + + def __init__( + self, + path: str | os.PathLike, + single_file_per_rank: bool = True, + sync_files: bool = True, + thread_count: int = 1, + per_thread_copy_ahead: int = 10_000_000, + overwrite: bool = True, + _extensions: Sequence[StreamTransformExtension] | None = None, + serialization_format: SerializationFormat = SerializationFormat.TORCH_SAVE, + *args: Any, + **kwargs: Any, + ) -> None: + """ + Initialize the writer pointing to `path`. + + Args: + path: directory where the checkpoint will be written to. + single_file_per_rank: Produce one file per rank instead of one file per tensor/blob. Default to True. + sync_files : force files to be synced to permanent storage. Default to True. + thread_count: Number of IO threads to use to write. Default to 1. + per_thread_copy_ahead: How many bytes to copy from the GPU ahead of saving them. Default 10Mb. + overwrite: Whether to allow overwriting existing checkpoints. Defaults to True. + _extensions: Extensions to apply to output streams (EXPERIMENTAL) + + N. B. If sync_files is disabled, there's no guarantee that the checkpoint will be consistent in the case of a failure. + """ + super().__init__() + self.fs = FileSystem() + self.path = self.fs.init_path(path) + self.single_file_per_rank = single_file_per_rank + self.sync_files = sync_files + self.thread_count = thread_count + self.per_thread_copy_ahead = per_thread_copy_ahead + self.save_id = _generate_uuid() + self.overwrite = overwrite + self.transforms = _StorageWriterTransforms(_extensions) + self.serialization_format = serialization_format + self.rank: int | None = None + self.use_collectives: bool = True + + def reset(self, checkpoint_id: str | os.PathLike | None = None) -> None: + if checkpoint_id: + self.path = self.fs.init_path(checkpoint_id) + self.save_id = _generate_uuid() + + def set_up_storage_writer( + self, is_coordinator: bool, *args: Any, **kwargs: Any + ) -> None: + self.rank = kwargs.get("rank") + self.use_collectives = kwargs.get("use_collectives", True) + + def _metadata_exists(self) -> bool: + if self.use_collectives: + # A global checkpoint metadata file + metadata_path = self._get_metadata_path(rank=None) + else: + # A rank 0 specific metadata file if every rank has written its own metadata + # Just looking for lowest rank metadata file is sufficient + metadata_path = self._get_metadata_path(rank=0) + + return self.fs.exists(metadata_path) + + def prepare_local_plan(self, plan: SavePlan) -> SavePlan: + self.fs.mkdir(self.path) + if self._metadata_exists(): + if self.overwrite: + warnings.warn( + f"Detected an existing checkpoint in {self.path}, overwriting since {self.overwrite=}." + " Past version 2.5 of PyTorch, `overwrite` will default to False. Set this variable to True to" + " maintain this functionality or False to raise when an existing checkpoint is found.", + stacklevel=2, + ) + else: + raise RuntimeError(f"Checkpoint already exists and {self.overwrite=}.") + + if self.rank is not None and not self.use_collectives: + plan = dataclasses.replace( + plan, storage_data=_StoragePrefix(f"__{self.rank}_") + ) + + return plan + + def prepare_global_plan(self, plans: list[SavePlan]) -> list[SavePlan]: + new_plans = [ + dataclasses.replace(plan, storage_data=_StoragePrefix(f"__{i}_")) + if plan.storage_data is None + else plan + for i, plan in enumerate(plans) + ] + return new_plans + + def write_data( + self, + plan: SavePlan, + planner: SavePlanner, + ) -> Future[list[WriteResult]]: + storage_plan: _StoragePrefix = plan.storage_data + file_count = 0 + + def gen_file(): + nonlocal file_count + file_name = f"{storage_plan.prefix}{file_count}{DEFAULT_SUFFIX}" + file_count += 1 + return file_name + + file_queue: queue.Queue = queue.Queue() + if self.single_file_per_rank: + for bucket in _split_by_size_and_type(self.thread_count, plan.items): + file_name = gen_file() + path = self.fs.concat_path(self.path, file_name) + file_queue.put((path, file_name, bucket)) + else: + for item in plan.items: + file_name = gen_file() + path = self.fs.concat_path(self.path, file_name) + file_queue.put((path, file_name, [item])) + + return self._write_data(planner, file_queue) + + def _write_data( + self, + planner: SavePlanner, + file_queue: queue.Queue, + ) -> Future[list[WriteResult]]: + result_queue: queue.Queue = queue.Queue() + + threads = [] + for _ in range(1, self.thread_count): + t = threading.Thread( + target=_write_files_from_queue, + args=( + self.fs.create_stream, + file_queue, + result_queue, + planner, + self.transforms, + self.per_thread_copy_ahead, + self.sync_files, + self.thread_count, + self.serialization_format, + ), + ) + t.start() + threads.append(t) + + _write_files_from_queue( + create_stream=self.fs.create_stream, + file_queue=file_queue, + result_queue=result_queue, + planner=planner, + transforms=self.transforms, + inflight_threshhold=self.per_thread_copy_ahead, + use_fsync=self.sync_files, + thread_count=self.thread_count, + serialization_format=self.serialization_format, + ) + + for t in threads: + t.join() + + res = [] + try: + while True: + res += result_queue.get_nowait() + except queue.Empty: + fut: Future[list[WriteResult]] = Future() + fut.set_result(res) + return fut + + def finish(self, metadata: Metadata, results: list[list[WriteResult]]) -> None: + metadata.version = CURRENT_DCP_VERSION + + storage_md = {} + for wr_list in results: + storage_md.update({wr.index: wr.storage_data for wr in wr_list}) + metadata.storage_data = storage_md + + metadata.storage_meta = self.storage_meta() + tmp_filename = ( + f"__{self.rank}{_metadata_fn}.tmp" + if not self.use_collectives and self.rank is not None + else f"{_metadata_fn}.tmp" + ) + tmp_path = cast(Path, self.fs.concat_path(self.path, tmp_filename)) + with self.fs.create_stream(tmp_path, "wb") as metadata_file: + pickle.dump(metadata, metadata_file) + if self.sync_files: + # Flush Python-level buffers (OS for local files, network for cloud storage) before fsync. + metadata_file.flush() + try: + os.fsync(metadata_file.fileno()) + except (AttributeError, UnsupportedOperation) as e: + warnings.warn( + f"fsync not supported for this stream, relying on flush(): {e}" + ) + + # delete in-case other checkpoints were present. + if not self.use_collectives and self.rank is not None: + metadata_path = self._get_metadata_path(self.rank) + else: + metadata_path = self._get_metadata_path() + + if self.fs.exists(metadata_path): + self.fs.rm_file(metadata_path) + + self.fs.rename(tmp_path, metadata_path) + + def storage_meta(self) -> StorageMeta | None: + return StorageMeta(checkpoint_id=self.checkpoint_id, save_id=self.save_id) + + def _get_metadata_path(self, rank: int | None = None) -> os.PathLike: + filename = f"{_metadata_fn}" if rank is None else f"__{rank}{_metadata_fn}" + return cast(Path, self.fs.concat_path(self.path, filename)) + + @property + def checkpoint_id(self) -> str | os.PathLike: + """ + return the checkpoint_id that will be used to save the checkpoint. + """ + return self.path + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + return FileSystem.validate_checkpoint_id(checkpoint_id) + + +class _StorageReaderTransforms: + """ + This is experimental, and will likely move elsewhere in the + future. It lives here to minimize changes while we are still + learning and gathering feedback. + """ + + def __init__(self, extension_registry: ExtensionRegistry | None = None) -> None: + self.extension_registry = ( + ExtensionRegistry() if extension_registry is None else extension_registry + ) + + def transform_load_stream( + self, + read_item: ReadItem, + transform_descriptors: Sequence[str], + raw_stream: IO[bytes], + ) -> IO[bytes]: + extensions = self.extension_registry.from_descriptor_list(transform_descriptors) + transform_from = raw_stream + for ex in extensions: + if isinstance(ex, StreamTransformExtension): + transform_from = ex.transform_from(transform_from) + return transform_from + + +class FileSystemReader(StorageReader): + def __init__( + self, + path: str | os.PathLike, + _extension_registry: ExtensionRegistry | None = None, # EXPERIMENTAL + ) -> None: + super().__init__() + self.fs = FileSystem() + self.path = self.fs.init_path(path) + self.storage_data: dict[Any, Any] = {} + self.load_id = _generate_uuid() + self.transforms = _StorageReaderTransforms(_extension_registry) + self.rank = None + self.use_collectives = True + + def _slice_file(self, file, sinfo: _StorageInfo) -> IO[bytes]: + return cast(IO[bytes], _create_file_view(file, sinfo.offset, sinfo.length)) + + def reset(self, checkpoint_id: str | os.PathLike | None = None) -> None: + self.storage_data = {} + if checkpoint_id: + self.path = self.fs.init_path(checkpoint_id) + self.load_id = _generate_uuid() + + def read_data(self, plan: LoadPlan, planner: LoadPlanner) -> Future[None]: + # group requests by file + per_file: dict[str, list[ReadItem]] = {} + for read_item in plan.items: + item_md: _StorageInfo = self.storage_data[read_item.storage_index] + path = item_md.relative_path + per_file.setdefault(path, []).append(read_item) + + for relative_path, reqs in per_file.items(): + new_path = self.fs.concat_path(self.path, relative_path) + with self.fs.create_stream(new_path, "rb") as stream: + # TODO sort by offset and cache the reading + for req in reqs: + item_md = self.storage_data[req.storage_index] + file_slice = self._slice_file(stream, item_md) + transform_from = self.transforms.transform_load_stream( + req, + # This field wasn't present in older + # implementations so provide a fallback. + item_md.transform_descriptors or (), + file_slice, + ) + + if req.type == LoadItemType.BYTE_IO: + read_bytes = io.BytesIO(transform_from.read(-1)) + read_bytes.seek(0) + planner.load_bytes(req, read_bytes) + else: + if transform_from.seekable(): + seekable = transform_from + else: + # torch.load requires a seekable input, so read the transform + # stream now and store the output if needed + seekable = io.BytesIO(transform_from.read(-1)) + seekable.seek(0) + + tensor = cast( + Tensor, + torch.load( + seekable, + map_location="cpu", + weights_only=True, + ), + ) + tensor = narrow_tensor_by_index( + tensor, req.storage_offsets, req.lengths + ) + target_tensor = planner.resolve_tensor(req).detach() + + if target_tensor.size() != tensor.size(): + raise AssertionError( + f"req {req.storage_index} mismatch sizes {target_tensor.size()} vs {tensor.size()}" + ) + target_tensor.copy_(tensor) + planner.commit_tensor(req, target_tensor) + + fut: Future = Future() + fut.set_result(None) + return fut + + def _get_metadata_path(self, rank: int | None = None) -> os.PathLike: + filename = f"{_metadata_fn}" if rank is None else f"__{rank}{_metadata_fn}" + return cast(Path, self.fs.concat_path(self.path, filename)) + + # Implementing the abstract function in StorageReader + def read_metadata(self, *args: Any, **kwargs: Any) -> Metadata: + rank = kwargs.get("rank") + path = self._get_metadata_path(rank) + with self.fs.create_stream(path, "rb") as metadata_file: + metadata = pickle.load(metadata_file) + + if getattr(metadata, "storage_meta", None) is None: + metadata.storage_meta = StorageMeta() + metadata.storage_meta.load_id = self.load_id + + return metadata + + def set_up_storage_reader( + self, metadata: Metadata, is_coordinator: bool, *args: Any, **kwargs: Any + ) -> None: + self.storage_data = metadata.storage_data + self.rank = kwargs.get("rank") + self.use_collectives = kwargs.get("use_collectives", True) + if self.storage_data is None: + raise AssertionError("storage_data must not be None in metadata") + + def prepare_local_plan(self, plan: LoadPlan) -> LoadPlan: + return plan + + def prepare_global_plan(self, plans: list[LoadPlan]) -> list[LoadPlan]: + return plans + + @property + def checkpoint_id(self) -> str | os.PathLike: + """ + return the checkpoint_id that will be used to load the checkpoint. + """ + return self.path + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + return FileSystem.validate_checkpoint_id(checkpoint_id) + + +class FileSystemWriter(_FileSystemWriter, BlockingAsyncStager): + """ + Basic implementation of StorageWriter using file IO. + + This implementation makes the following assumptions and simplifications: + + * The checkpoint path is an empty or non-existing directory. + * File creation is atomic + + The checkpoint consist of one file per write request plus + a global `.metadata` file with the serialized metadata if rank coordination is enabled. + a rank local `__{rank}.metadata` file with the serialized metadata if rank coordination is NOT enabled. + + """ + + def __init__( + self, + path: str | os.PathLike, + single_file_per_rank: bool = True, + sync_files: bool = True, + thread_count: int = 1, + per_thread_copy_ahead: int = 10_000_000, + cache_staged_state_dict: bool = False, + overwrite: bool = True, + _extensions: Sequence[StreamTransformExtension] | None = None, + serialization_format: SerializationFormat = SerializationFormat.TORCH_SAVE, + ) -> None: + """ + Initialize the writer pointing to `path`. + + Args: + path: directory where the checkpoint will be written to. + single_file_per_rank: Produce one file per rank instead of one file per tensor/blob. Default to True. + sync_files : force files to be synced to permanent storage. Default to True. + thread_count: Number of IO threads to use to write. Default to 1. + per_thread_copy_ahead: How many bytes to copy from the GPU ahead of saving them. Default 10Mb. + cache_staged_state_dict: Whether to cache the staged state_dict. This option decreases staging latency + at the cost of increased memory usage. Additionally, if this parameter is set to True, it's the expectation + that the stager is maintained and reused for multiple dcp.async_save calls. Default to False. + overwrite: Whether to allow overwriting existing checkpoints. Defaults to True. + _extensions: Extensions to apply to output streams (EXPERIMENTAL) + + N. B. If sync_files is disabled, there's no guarantee that the checkpoint will be consistent in the case of a failure. + """ + _FileSystemWriter.__init__( + self, + path=path, + single_file_per_rank=single_file_per_rank, + sync_files=sync_files, + thread_count=thread_count, + per_thread_copy_ahead=per_thread_copy_ahead, + overwrite=overwrite, + _extensions=_extensions, + serialization_format=serialization_format, + ) + BlockingAsyncStager.__init__( + self, + cache_staged_state_dict=cache_staged_state_dict, + ) + + def stage(self, state_dict: STATE_DICT_TYPE) -> STATE_DICT_TYPE: + """Override of AsyncStager.stage""" + # in the async case, the state dict is already on CPU, so maintaining this + # buffer makes no sense + self.per_thread_copy_ahead = 0 + return super().stage(state_dict) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/format_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/format_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..a2b19edc2420cbf51294486f44209e3bda2ab44e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/format_utils.py @@ -0,0 +1,292 @@ +# mypy: allow-untyped-defs +import argparse +import os +from enum import Enum +from typing import cast + +import torch +import torch.distributed as dist +from torch.distributed._shard._utils import narrow_tensor_by_index +from torch.distributed.checkpoint import FileSystemReader, FileSystemWriter +from torch.distributed.checkpoint._nested_dict import flatten_state_dict +from torch.distributed.checkpoint.default_planner import ( + _EmptyStateDictLoadPlanner, + DefaultLoadPlanner, +) +from torch.distributed.checkpoint.metadata import ( + Metadata, + STATE_DICT_TYPE, + STORAGE_TYPES, + TensorProperties, + TensorStorageMetadata, +) +from torch.distributed.checkpoint.planner import LoadItemType, LoadPlan, LoadPlanner +from torch.distributed.checkpoint.planner_helpers import _create_chunk_list +from torch.distributed.checkpoint.state_dict_loader import _load_state_dict +from torch.distributed.checkpoint.state_dict_saver import _save_state_dict +from torch.distributed.checkpoint.storage import StorageReader +from torch.futures import Future + + +__all__ = [ + "dcp_to_torch_save", + "torch_save_to_dcp", + "BroadcastingTorchSaveReader", + "DynamicMetaLoadPlanner", +] + + +class BroadcastingTorchSaveReader(StorageReader): + """ + StorageReader for reading a Torch Save file. This reader will read the entire checkpoint + on the coordinator rank, and then broadcast and shard each tensor to all ranks. + + . N.B. Intended to be used with DynamicMetaLoadPlanner + + .. warning:: + Current implementation only supports loading Tensors. + + >>> # xdoctest: +SKIP("undefined vars") + >>> sd = {"mode": model} + >>> dcp.load( + >>> sd, + >>> storage_reader=BroadcastingTorchSaveReader(), + >>> planner=DynamicMetaLoadPlanner(), + >>> checkpoint_id="path_to_model.pt" + >>> ) + """ + + def __init__( + self, + checkpoint_id: str | os.PathLike | None = None, + coordinator_rank: int = 0, + ) -> None: + self.checkpoint_id = checkpoint_id + self.coordinator_rank = coordinator_rank + + # pyrefly: ignore [bad-override] + def read_metadata(self) -> Metadata: + """Extends the default StorageReader to support building the metadata file""" + # Metadata is built in planner.set_up_planner, since we are not actually reading metadata from + # the disk + return Metadata(state_dict_metadata={}) + + def read_data(self, plan: LoadPlan, planner: LoadPlanner) -> Future[None]: + """ + Reads torch save data on the coordinator rank, and broadcast afterwards + this incurrs a communication cost, but avoids having to load + the entire checkpoint on each rank, hopefully preventing OOM issues + """ + planner = cast(DefaultLoadPlanner, planner) + + # data is read in on the coordinator rank, and broadcast afterwards + # this incurs a communication cost, but it avoids having to load + # the entire checkpoint on each rank, hopefully preventing OOM issues + # TODO: read on each host, instead of only the coordinator + if self.is_coordinator: + if self.checkpoint_id is None: + raise AssertionError("checkpoint_id must be set before reading data") + torch_state_dict = torch.load( + self.checkpoint_id, map_location="cpu", weights_only=False + ) + if planner.flatten_state_dict: + torch_state_dict, _ = flatten_state_dict(torch_state_dict) + else: + torch_state_dict = None + + for req in plan.items: + if req.type == LoadItemType.BYTE_IO: + raise RuntimeError( + f"Non-tensor value identified at {req.storage_index.fqn}. " + f"At this time {type(self).__name__} only supports loading Tensors." + ) + + # Broadcast the tensor from the coordinator rank + if self.is_coordinator: + pg_device = dist.distributed_c10d._get_pg_default_device() + # pyrefly: ignore [unsupported-operation] + tensor = torch_state_dict[req.storage_index.fqn].to(pg_device) + else: + tensor = torch.empty_like(planner.state_dict[req.storage_index.fqn]) + + dist.broadcast(tensor, src=self.coordinator_rank, async_op=False) + + tensor = narrow_tensor_by_index(tensor, req.storage_offsets, req.lengths) + target_tensor = planner.resolve_tensor(req).detach() + if not target_tensor.size() == tensor.size(): + raise AssertionError( + f"req {req.storage_index} mismatch sizes, " + f"{target_tensor.size()} vs {tensor.size()}" + ) + target_tensor.copy_(tensor) + planner.commit_tensor(req, target_tensor) + + fut: Future = Future() + fut.set_result(None) + return fut + + # pyrefly: ignore [bad-override] + def set_up_storage_reader(self, metadata: Metadata, is_coordinator: bool) -> None: + """Implementation of the StorageReader method""" + self.is_coordinator = is_coordinator + if self.is_coordinator: + if not dist.get_rank() == self.coordinator_rank: + raise AssertionError( + f"Coordinator rank mismatch: expected {self.coordinator_rank}, " + f"got {dist.get_rank()}" + ) + + if self.checkpoint_id is None: + raise AssertionError( + "checkpoint_id must be set before setting up storage reader" + ) + + def prepare_local_plan(self, plan: LoadPlan) -> LoadPlan: + """Implementation of the StorageReader method""" + return plan + + def prepare_global_plan(self, global_plan: list[LoadPlan]) -> list[LoadPlan]: + """Implementation of the StorageReader method""" + return global_plan + + def reset(self, checkpoint_id: str | os.PathLike | None = None) -> None: + """Implementation of the StorageReader method""" + self.checkpoint_id = checkpoint_id + + @classmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + """Implementation of the StorageReader method""" + return os.path.isfile(checkpoint_id) + + +class DynamicMetaLoadPlanner(DefaultLoadPlanner): + """ + Extension of DefaultLoadPlanner, which creates a new Metadata object based on the passed in state dict, + avoiding the need to read metadata from disk. This is useful when reading formats which don't have a + metadata file, like Torch Save files. + + . N.B. Intended to be used with BroadcastingTorchSaveReader + + .. warning:: + Current implementation only supports loading Tensors. + + >>> # xdoctest: +SKIP("undefined vars") + >>> sd = {"mode": model} + >>> dcp.load( + >>> sd, + >>> storage_reader=BroadcastingTorchSaveReader(), + >>> planner=DynamicMetaLoadPlanner(), + >>> checkpoint_id="path_to_model.pt" + >>> ) + """ + + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + metadata: Metadata | None = None, + is_coordinator: bool = False, + ) -> None: + """Setups of the planner, extnding default behavior by creating the Metadata object from the state dict""" + super().set_up_planner(state_dict, metadata, is_coordinator) + + state_dict_metadata: dict[str, STORAGE_TYPES] = {} + for key, tensor in self.state_dict.items(): + if not torch.is_tensor(tensor): + raise RuntimeError( + f"Non-tensor value identified at {key}. " + f"At this time {type(self).__name__} only supports loading Tensors." + ) + + state_dict_metadata[key] = TensorStorageMetadata( + TensorProperties(dtype=tensor.dtype), + tensor.size(), + _create_chunk_list(tensor), + ) + self.metadata = Metadata(state_dict_metadata=state_dict_metadata) + + +def dcp_to_torch_save( + dcp_checkpoint_dir: str | os.PathLike, + torch_save_path: str | os.PathLike, +): + """ + Given a directory containing a DCP checkpoint, this function will convert it into a + Torch save file. + + Args: + dcp_checkpoint_dir: Directory containing the DCP checkpoint. + torch_save_path: Filename to store the converted Torch save file. + + .. warning:: + To avoid OOM, it's recommended to only run this function on a single rank. + """ + sd: STATE_DICT_TYPE = {} + _load_state_dict( + sd, + storage_reader=FileSystemReader(dcp_checkpoint_dir), + planner=_EmptyStateDictLoadPlanner(), + no_dist=True, + ) + torch.save(sd, torch_save_path) + + +def torch_save_to_dcp( + torch_save_path: str | os.PathLike, + dcp_checkpoint_dir: str | os.PathLike, +): + """ + Given the location of a torch save file, converts it into a DCP checkpoint. + + Args: + torch_save_path: Filename of the Torch save file. + dcp_checkpoint_dir: Directory to store the DCP checkpoint. + + .. warning:: + To avoid OOM, it's recommended to only run this function on a single rank. + """ + + state_dict = torch.load(torch_save_path, weights_only=False) + # we don't need stateful behavior here because the expectation is anything loaded by + # torch.load would not contain stateful objects. + _save_state_dict( + state_dict, storage_writer=FileSystemWriter(dcp_checkpoint_dir), no_dist=True + ) + + +if __name__ == "__main__": + + class FormatMode(Enum): + TORCH_TO_DCP = "torch_to_dcp" + DCP_TO_TORCH = "dcp_to_torch" + + # Parse command-line arguments + parser = argparse.ArgumentParser() + parser.add_argument( + "mode", + type=str, + help="Conversion mode", + choices=[m.value for m in FormatMode], + default=FormatMode.TORCH_TO_DCP, + ) + parser.add_argument("src", type=str, help="Path to the source model") + parser.add_argument("dst", type=str, help="Path to the destination model") + args = parser.parse_args() + + print( + f"Converting checkpoint from {args.src} to {args.dst} using method: '{args.mode}'" + ) + checkpoint_missing_warning = ( + f"No checkpoint found at {args.src}. Skipping conversion." + ) + if args.mode == FormatMode.TORCH_TO_DCP.value: + if os.path.isfile(args.src): + torch_save_to_dcp(args.src, args.dst) + else: + print(checkpoint_missing_warning) + elif args.mode == FormatMode.DCP_TO_TORCH.value: + if os.path.isdir(args.src): + dcp_to_torch_save(args.src, args.dst) + else: + print(checkpoint_missing_warning) + else: + raise ValueError(f"Unknown conversion mode: {args.mode}") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/hf_storage.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/hf_storage.py new file mode 100644 index 0000000000000000000000000000000000000000..248d075681e1477880dafe055aa676e29ce5d0e1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/hf_storage.py @@ -0,0 +1,391 @@ +# mypy: allow-untyped-defs +import dataclasses +import json +import logging +import queue +import threading +from typing import Any + +import torch +from torch.distributed.checkpoint import FileSystemReader, FileSystemWriter +from torch.distributed.checkpoint._consolidate_hf_safetensors import ( + consolidate_safetensors_files, +) +from torch.distributed.checkpoint._hf_utils import ( + _gen_file_name, + _HFStorageInfo, + _metadata_fn, + CUSTOM_METADATA_KEY, + SAVED_OFFSETS_KEY, + SHARDED_DIR_NAME, + SUFFIX, +) +from torch.distributed.checkpoint.filesystem import SerializationFormat +from torch.distributed.checkpoint.metadata import ( + ChunkStorageMetadata, + Metadata, + MetadataIndex, + StorageMeta, + TensorProperties, + TensorStorageMetadata, +) +from torch.distributed.checkpoint.planner import ( + LoadPlan, + LoadPlanner, + ReadItem, + SavePlan, + SavePlanner, + WriteItem, +) +from torch.distributed.checkpoint.storage import WriteResult +from torch.futures import Future + + +logger: logging.Logger = logging.getLogger(__name__) + +__all__ = ["HuggingFaceStorageWriter", "HuggingFaceStorageReader"] + + +class HuggingFaceStorageWriter(FileSystemWriter): + """ + A writer that writes to storage in the huggingface safetensors format. + """ + + def __init__( + self, + path: str, + fqn_to_index_mapping: dict[str, int] | None = None, + thread_count: int = 1, + save_distributed: bool = False, + enable_consolidation: bool = False, + thread_count_consolidation: int = 1, + ) -> None: + """ + Initialize the huggingface writer pointing to path. + + Args: + path: directory where the checkpoint will be read from. + fqn_to_index_mapping: A mapping from tensor FQN to the index of the file that the tensor should be written to. + Indices are from 1 to N, where N is the number of files. If not provided, + the tensors will be written to a single file. If none, then all the tensors on the + same rank will be written to the same file. + thread_count: Number of threads to use to write distributed checkpoint. Default to 1. + save_distributed: If True, save the checkpoint using distributed APIs where every rank saves its own shard. + Default is False which assumes rank-0 checkpointing of the full state_dict. + enable_consolidation: If True, consolidate the sharded checkpoint after saving. The sharded tensors will be + saved to path/sharded and the full tensors will be saved to path. Default to False. + thread_count_consolidation: Number of threads to use for parallel processing of saving data + to consolidated output files. Default to 1. + """ + + super().__init__( + path=path, + serialization_format=SerializationFormat.SAFETENSORS, + thread_count=thread_count, + ) + self.fqn_to_index_mapping: dict[str, int] | None = fqn_to_index_mapping + self.save_distributed: bool = save_distributed + self.enable_consolidation: bool = enable_consolidation + self.consolidated_output_path: str | None = None + if self.enable_consolidation: + self.consolidated_output_path = str(self.path) + self.path = self.fs.concat_path(self.path, SHARDED_DIR_NAME) + self.thread_count_consolidation = thread_count_consolidation + + def prepare_global_plan(self, plans: list[SavePlan]) -> list[SavePlan]: + new_plans = [] + for i, plan in enumerate(plans, start=1): + storage_data: dict[str, Any] = {} + if self.fqn_to_index_mapping is not None: + storage_data["fqn_to_index_mapping"] = self.fqn_to_index_mapping + if self.save_distributed: + storage_data["shard_index"] = i + + new_plans.append(dataclasses.replace(plan, storage_data=storage_data)) + + return new_plans + + def write_data( + self, + plan: SavePlan, + planner: SavePlanner, + ) -> Future[list[WriteResult]]: + if len(plan.items) == 0: + fut: Future = Future() + fut.set_result([]) + return fut + + # storage_plan is a map from key to file index + storage_data: dict[str, Any] = plan.storage_data + storage_plan: dict[str, int] | None = None + shard_index: int | None = None + if "fqn_to_index_mapping" in storage_data: + storage_plan = storage_data["fqn_to_index_mapping"] + if "shard_index" in storage_data: + shard_index = storage_data["shard_index"] + + buckets = self._split_by_storage_plan(storage_plan, plan.items) + highest_index = max(storage_plan.values()) if storage_plan is not None else 1 + + file_queue: queue.Queue = queue.Queue() + for file_index, write_items in buckets.items(): + file_name = _gen_file_name(file_index, highest_index, shard_index) + file_queue.put( + (self.fs.concat_path(self.path, file_name), file_name, write_items) + ) + + return super()._write_data(planner, file_queue) + + def finish(self, metadata: Metadata, results: list[list[WriteResult]]) -> None: + if self.save_distributed and not self.enable_consolidation: + # if we are saving distributed, without consolidating, + # then we have no metadata to write because a metadata + # file with fqn to file mapping doesn't make sense + # in this case, because fqns will be in multiple files + logger.info("Not consolidating sharded checkpoint in finish step.") + return + if self.save_distributed: + fqn_to_index_mapping: dict[str, int] = ( + self.fqn_to_index_mapping + if self.fqn_to_index_mapping is not None + else dict.fromkeys(metadata.state_dict_metadata.keys(), 1) + ) + + return consolidate_safetensors_files( + input_dir=str(self.path), + output_dir=self.consolidated_output_path, # type: ignore[arg-type] + num_threads=self.thread_count_consolidation, + fqn_to_index_mapping=fqn_to_index_mapping, + ) + + # writing a model.index.safetensors.json file with fqn to file mapping + # for the rank-0 checkpointing case + metadata_to_write = {} + storage_md = {} + total_size = 0 + for wr_list in results: + storage_md.update( + {wr.index.fqn: wr.storage_data.relative_path for wr in wr_list} + ) + total_size += sum([wr.storage_data.length for wr in wr_list]) + metadata_to_write["metadata"] = {"total_size": total_size} + metadata_to_write["weight_map"] = storage_md + + metadata_path = self.fs.concat_path(self.path, f"{_metadata_fn}") + with self.fs.create_stream(metadata_path, "w") as metadata_file: + json.dump(metadata_to_write, metadata_file, indent=2) + + def _split_by_storage_plan( + self, storage_plan: dict[str, int] | None, items: list[WriteItem] + ) -> dict[int, list[WriteItem]]: + # storage_plan is a map from key to index + if storage_plan is None: + return {1: items} + + buckets = {} + for item in items: + key = item.index.fqn + + idx = storage_plan[key] + if idx not in buckets: + buckets[idx] = [item] + else: + buckets[idx].append(item) + + return buckets + + @property + def metadata_path(self) -> str: + return _metadata_fn + + +class HuggingFaceStorageReader(FileSystemReader): + """ + A reader that reads a checkpoint in the huggingface safetensors format. + """ + + def __init__(self, path: str, thread_count: int = 1) -> None: + """ + Initialize the huggingface reader pointing to path. + + Args: + path: directory where the checkpoint will be read from. + thread_count: Number of threads to use to read distributed checkpoint. Default to 1. + """ + + super().__init__(path=path) + self.thread_count = thread_count + + def _process_read_request(self, f, req: ReadItem, planner: LoadPlanner) -> None: + """Helper function to process a single read request.""" + # Create slices for each dimension based on offsets and lengths + slices = tuple( + slice(offset, offset + length) + for offset, length in zip(req.storage_offsets, req.lengths) + ) + tensor = f.get_slice(req.storage_index.fqn)[slices] + target_tensor = planner.resolve_tensor(req).detach() + + if target_tensor.size() != tensor.size(): + raise AssertionError( + f"req {req.storage_index} mismatch sizes {target_tensor.size()} vs {tensor.size()}" + ) + + target_tensor.copy_(tensor) + planner.commit_tensor(req, target_tensor) + + def _read_files_from_queue( + self, + file_queue: queue.Queue, + result_queue: queue.Queue, + planner: LoadPlanner, + ) -> None: + from safetensors import safe_open # type: ignore[import] + + try: + while True: + file_name, reqs = file_queue.get_nowait() + with safe_open(filename=file_name, framework="pt") as f: + for req in reqs: + self._process_read_request(f, req, planner) + result_queue.put(True) # Signal that this file has been processed + except queue.Empty: + pass + + def read_data(self, plan: LoadPlan, planner: LoadPlanner) -> Future[None]: + from safetensors import safe_open # type: ignore[import] + + per_file: dict[str, list[ReadItem]] = {} + + for read_item in plan.items: + item_md: _HFStorageInfo = self.storage_data[read_item.storage_index] + file_name = item_md.relative_path + per_file.setdefault(file_name, []).append(read_item) + + if self.thread_count <= 1 or len(per_file) <= 1: + for file_name, reqs in per_file.items(): + with safe_open(filename=file_name, framework="pt") as f: + for req in reqs: + self._process_read_request(f, req, planner) + else: + # Use parallel implementation with thread pool + file_queue: queue.Queue = queue.Queue() + result_queue: queue.Queue = queue.Queue() + + # Fill the queue with files to process + for file_name, reqs in per_file.items(): + file_queue.put((file_name, reqs)) + + # Create and start worker threads + threads = [] + num_threads = min(self.thread_count, len(per_file)) + for _ in range(num_threads): + t = threading.Thread( + target=self._read_files_from_queue, + args=(file_queue, result_queue, planner), + ) + t.start() + threads.append(t) + + # Wait for all threads to complete + for t in threads: + t.join() + + # Check if all files were processed + processed_count = 0 + try: + while True: + result_queue.get_nowait() + processed_count += 1 + except queue.Empty: + pass + + if processed_count != len(per_file): + raise AssertionError( + f"Not all files were processed: {processed_count} out of {len(per_file)}" + ) + + fut: Future = Future() + fut.set_result(None) + return fut + + # pyrefly: ignore [bad-override] + def read_metadata(self) -> Metadata: + from safetensors import safe_open # type: ignore[import] + from safetensors.torch import _getdtype # type: ignore[import] + + state_dict_metadata: dict[str, TensorStorageMetadata] = {} + storage_data: dict[MetadataIndex, _HFStorageInfo] = {} + + safetensors_files = [] + for file in self.fs.ls(self.path): + if file.endswith(SUFFIX): + safetensors_files.append(file) + + for safetensor_file in safetensors_files: + with safe_open(safetensor_file, framework="pt") as f: + keys = f.keys() + extra_metadata = f.metadata() + + dcp_sharding_info = None + if extra_metadata and extra_metadata.get(CUSTOM_METADATA_KEY): + dcp_sharding_info = json.loads( + extra_metadata.get(CUSTOM_METADATA_KEY) + ) + + for key in keys: + shape = f.get_slice(key).get_shape() + dtype = f.get_slice(key).get_dtype() + # construct state_dict_metadata + if dcp_sharding_info is not None: + offset = dcp_sharding_info[key][SAVED_OFFSETS_KEY] + else: + offset = [0] * len(shape) + + if key not in state_dict_metadata: + state_dict_metadata[key] = TensorStorageMetadata( + properties=TensorProperties(dtype=_getdtype(dtype)), + size=torch.Size( + [saved + offset for saved, offset in zip(shape, offset)] + ), + chunks=[ + ChunkStorageMetadata( + offsets=torch.Size(offset), + sizes=torch.Size(shape), + ) + ], + ) + else: + state_dict_metadata[key].chunks.append( + ChunkStorageMetadata( + torch.Size(offset), sizes=torch.Size(shape) + ) + ) + size = list(state_dict_metadata[key].size) + for i in range(len(size)): + size[i] = max(size[i], shape[i] + offset[i]) + state_dict_metadata[key].size = torch.Size(size) + + # construct storage data + if dcp_sharding_info is not None: + metadata_index = MetadataIndex( + fqn=key, offset=dcp_sharding_info[key][SAVED_OFFSETS_KEY] + ) + else: + metadata_index = MetadataIndex(fqn=key, offset=[0] * len(shape)) + storage_data[metadata_index] = _HFStorageInfo( + relative_path=safetensor_file, + shape=torch.Size(shape), + dtype=_getdtype(dtype), + ) + + metadata = Metadata( + state_dict_metadata=state_dict_metadata, # type: ignore[arg-type] + storage_data=storage_data, + ) + + if getattr(metadata, "storage_meta", None) is None: + metadata.storage_meta = StorageMeta() + metadata.storage_meta.load_id = self.load_id # type: ignore[union-attr] + + return metadata diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logger.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logger.py new file mode 100644 index 0000000000000000000000000000000000000000..677cac0339cb9fab60c77f75da04bc7ef06504f3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logger.py @@ -0,0 +1,121 @@ +# mypy: allow-untyped-defs +import functools +import logging +import time +from collections.abc import Callable +from typing import Any, TypeVar +from typing_extensions import ParamSpec +from uuid import uuid4 + +import torch.distributed.c10d_logger as c10d_logger +from torch.distributed.checkpoint.logging_handlers import DCP_LOGGER_NAME + + +logger = logging.getLogger() + + +__all__: list[str] = [] + +# pyrefly: ignore [unknown-name] +global _dcp_logger +_dcp_logger = c10d_logger._get_or_create_logger(DCP_LOGGER_NAME) + +_T = TypeVar("_T") +_P = ParamSpec("_P") + + +def _msg_dict_from_dcp_method_args(*args, **kwargs) -> dict[str, Any]: + """ + Extracts log data from dcp method args + """ + msg_dict = {} + + # checkpoint ID can be passed in through the serializer or through the checkpoint id directly + storage_writer = kwargs.get("storage_writer") + storage_reader = kwargs.get("storage_reader") + planner = kwargs.get("planner") + + checkpoint_id = kwargs.get("checkpoint_id") + if not checkpoint_id and (serializer := storage_writer or storage_reader): + checkpoint_id = getattr(serializer, "checkpoint_id", None) + + msg_dict["checkpoint_id"] = ( + # pyrefly: ignore [unsupported-operation] + str(checkpoint_id) if checkpoint_id is not None else checkpoint_id + ) + + # Uniquely identify a _dcp_method_logger wrapped function call. + msg_dict["uuid"] = str(uuid4().int) + + if storage_writer: + msg_dict["storage_writer"] = storage_writer.__class__.__name__ + + if storage_reader: + msg_dict["storage_reader"] = storage_reader.__class__.__name__ + + if planner: + msg_dict["planner"] = planner.__class__.__name__ + + return msg_dict + + +def _get_msg_dict(func_name, *args, **kwargs) -> dict[str, Any]: + msg_dict = _msg_dict_from_dcp_method_args(*args, **kwargs) + msg_dict.update(c10d_logger._get_msg_dict(func_name, *args, **kwargs)) + + return msg_dict + + +def _dcp_method_logger( + log_exceptions: bool = False, **wrapper_kwargs: Any +) -> Callable[[Callable[_P, _T]], Callable[_P, _T]]: # pyre-ignore + """This method decorator logs the start, end, and exception of wrapped events.""" + + def decorator(func: Callable[_P, _T]): + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _T: + msg_dict = _get_msg_dict( + func.__name__, *args, **{**wrapper_kwargs, **kwargs} + ) + + # log start event + msg_dict["event"] = "start" + t0 = time.time_ns() + msg_dict["time"] = t0 + msg_dict["log_exceptions"] = log_exceptions + _dcp_logger.debug(msg_dict) + + # exceptions + try: + result = func(*args, **kwargs) + except BaseException as error: + if log_exceptions: + msg_dict["event"] = "exception" + msg_dict["error"] = f"{error}" + msg_dict["time"] = time.time_ns() + _dcp_logger.error(msg_dict) + raise + + # end event + msg_dict["event"] = "end" + t1 = time.time_ns() + msg_dict["time"] = time.time_ns() + msg_dict["times_spent"] = t1 - t0 + _dcp_logger.debug(msg_dict) + + return result + + return wrapper + + return decorator + + +def _init_logger(rank: int): + logger.setLevel(logging.INFO) + ch = logging.StreamHandler() + ch.setLevel(logging.INFO) + formatter = logging.Formatter( + f"[{rank}] %(asctime)s - %(name)s - %(levelname)s - %(message)s" + ) + ch.setFormatter(formatter) + logger.addHandler(ch) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logging_handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logging_handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..99c3ee4156ce340e37a2723106df5ea64b19170d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/logging_handlers.py @@ -0,0 +1,14 @@ +import logging + +from torch.distributed.logging_handlers import _log_handlers + + +__all__: list[str] = [] + +DCP_LOGGER_NAME = "dcp_logger" + +_log_handlers.update( + { + DCP_LOGGER_NAME: logging.NullHandler(), + } +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/metadata.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/metadata.py new file mode 100644 index 0000000000000000000000000000000000000000..7855b94df33ce85efd115d163fbdf1dff599bbbc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/metadata.py @@ -0,0 +1,185 @@ +# mypy: allow-untyped-defs +import os +from collections.abc import Sequence +from dataclasses import dataclass, field +from enum import Enum +from typing import Any + +import torch +from torch.distributed.checkpoint.stateful import StatefulT + + +__all__ = [ + "ChunkStorageMetadata", + "TensorStorageMetadata", + "BytesStorageMetadata", + "Metadata", + "MetadataIndex", + "TensorProperties", + "StorageMeta", +] + + +@dataclass +class ChunkStorageMetadata: + """ + Each chunk is expected to have the same properties of the TensorStorageMetadata + that includes it. + """ + + offsets: torch.Size + sizes: torch.Size + + +class _MEM_FORMAT_ENCODING(Enum): + """Describe the memory format of a tensor.""" + + TORCH_CONTIGUOUS_FORMAT = 0 + TORCH_CHANNELS_LAST = 1 + TORCH_PRESERVE_FORMAT = 2 + + +@dataclass +class TensorProperties: + """Properties used to create :class:`Tensor`""" + + # Regular tensor fields + dtype: torch.dtype = field(default_factory=torch.get_default_dtype) + # This field is deprecated. + layout: torch.layout = field(default=torch.strided) + # This field is deprecated. + requires_grad: bool = False + # This field is deprecated. + memory_format: torch.memory_format = field(default=torch.contiguous_format) + # This field is deprecated. + pin_memory: bool = False + + def __getstate__(self): + # Since torch.memory_format cannot be pickled! + memory_format = self.memory_format + if memory_format == torch.contiguous_format: + mem_format_encoding = _MEM_FORMAT_ENCODING.TORCH_CONTIGUOUS_FORMAT + elif memory_format == torch.channels_last: + mem_format_encoding = _MEM_FORMAT_ENCODING.TORCH_CHANNELS_LAST + elif memory_format == torch.preserve_format: + mem_format_encoding = _MEM_FORMAT_ENCODING.TORCH_PRESERVE_FORMAT + else: + raise RuntimeError(f"Invalid torch.memory_format: {memory_format}") + + return ( + self.dtype, + self.layout, + self.requires_grad, + mem_format_encoding, + self.pin_memory, + ) + + def __setstate__( + self, + state, + ): + ( + self.dtype, + self.layout, + self.requires_grad, + mem_format_encoding, + self.pin_memory, + ) = state + + if mem_format_encoding == _MEM_FORMAT_ENCODING.TORCH_CONTIGUOUS_FORMAT: + memory_format = torch.contiguous_format + elif mem_format_encoding == _MEM_FORMAT_ENCODING.TORCH_CHANNELS_LAST: + memory_format = torch.channels_last + elif mem_format_encoding == _MEM_FORMAT_ENCODING.TORCH_PRESERVE_FORMAT: + memory_format = torch.preserve_format + else: + raise RuntimeError( + f"Invalid torch.memory_format encoding: {mem_format_encoding}" + ) + + self.memory_format = memory_format + + @staticmethod + def create_from_tensor(tensor: torch.Tensor) -> "TensorProperties": + return TensorProperties( + dtype=tensor.dtype, + layout=tensor.layout, + requires_grad=tensor.requires_grad, + memory_format=torch.contiguous_format, + pin_memory=tensor.is_pinned(), + ) + + +@dataclass +class TensorStorageMetadata: + properties: TensorProperties + size: torch.Size + chunks: list[ChunkStorageMetadata] + + +@dataclass +class BytesStorageMetadata: + pass + + +STORAGE_TYPES = TensorStorageMetadata | BytesStorageMetadata +STATE_DICT_TYPE = dict[str, StatefulT | Any] + + +@dataclass +class StorageMeta: + checkpoint_id: str | os.PathLike | None = None + save_id: str | None = None + load_id: str | None = None + modules: list[str] = field(default_factory=list) + + +@dataclass +class Metadata: + """This class represents the metadata of the checkpoint.""" + + # Keys are the same from the `state_dict` used. + state_dict_metadata: dict[str, STORAGE_TYPES] + # It is the responsibility of the planner and storage plugins to ensure + # backward compatibility of the planner_data and storage_data. DCP will + # also ensure the backward compatibility of the metadata in this file and + # the metadata of the built-in planner and storage plugins. + planner_data: Any = None + storage_data: Any = None + storage_meta: StorageMeta | None = None + version: str | None = None + + +@dataclass(frozen=True) +class MetadataIndex: + """This class represents a lookup key for items in a state dict or Metadata.""" + + fqn: str + """Fully Qualified Name of the object""" + + offset: torch.Size | None = None + """If the object is a tensor, offset into the tensor we're looking for""" + + index: int | None = field(hash=False, compare=False, default=None) + """ + Index hint when searching for tensor chunk to speedup lookups (optional) + + A common representation of a sharded tensor is as a list of chunks so to + find the index in such a list you need to linear search it. + + When constructing an instance of MetadataIndex that points to that list, + one can provide the index as a hint and it will be probed first before + the linear search and thus making it significantly faster. + """ + + def __init__( + self, + fqn: str, + offset: Sequence[int] | None = None, + index: int | None = None, + ): + # We must use object.__setattr__ due to frozen=True + object.__setattr__(self, "fqn", fqn) + object.__setattr__(self, "index", index) + if offset is not None: + object.__setattr__(self, "offset", torch.Size(offset)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/optimizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/optimizer.py new file mode 100644 index 0000000000000000000000000000000000000000..485a7ad9f43a15d2bf0926bfc6db0e89a467ad9c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/optimizer.py @@ -0,0 +1,363 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import dataclasses +from collections.abc import Sequence +from typing import cast + +import torch +import torch.distributed as dist +from torch._utils import _get_device_module +from torch.distributed._shard.sharded_tensor.api import ShardedTensor +from torch.distributed._shard.sharded_tensor.metadata import ( + TensorProperties as ShardTensorProperties, +) +from torch.distributed._shard.sharded_tensor.shard import Shard +from torch.distributed._shard.sharding_spec.chunk_sharding_spec import ChunkShardingSpec +from torch.distributed.checkpoint._nested_dict import unflatten_state_dict +from torch.distributed.checkpoint.default_planner import DefaultLoadPlanner +from torch.distributed.checkpoint.metadata import ( + BytesStorageMetadata, + ChunkStorageMetadata, + Metadata, + MetadataIndex, + STATE_DICT_TYPE, + TensorProperties, + TensorStorageMetadata, +) +from torch.distributed.checkpoint.planner import LoadPlan, LoadPlanner +from torch.distributed.checkpoint.planner_helpers import ( + _create_read_items, + create_read_items_for_chunk_list, +) + +# pyrefly: ignore [deprecated] +from torch.distributed.checkpoint.state_dict_loader import load_state_dict +from torch.distributed.checkpoint.storage import StorageReader +from torch.distributed.checkpoint.utils import ( + _element_wise_add, + _element_wise_sub, + _normalize_device_info, +) +from torch.distributed.distributed_c10d import _get_default_group +from torch.distributed.fsdp._shard_utils import _create_chunk_sharded_tensor +from torch.distributed.remote_device import _remote_device +from torch.distributed.tensor import DTensor + + +STATE_DICT_2D_LAYOUT = dict[str, tuple[Sequence[int] | None, Sequence[int]]] + + +# TODO: Update docstrings for optimizer.py +__all__ = [ + "load_sharded_optimizer_state_dict", +] + + +def _gen_rank_device(global_rank: int, device_type: str = "cuda") -> str: + if device_type == "cpu": + return "cpu" + device_module = _get_device_module(device_type) + if device_module.is_available(): + return _normalize_device_info( + device_type, global_rank % device_module.device_count() + ) + return "cpu" + + +def _create_colwise_spec( + pg: dist.ProcessGroup | None = None, +) -> ChunkShardingSpec: + pg_device_type = dist.distributed_c10d._get_pg_default_device(pg).type + if pg is None: + placements = [ + f"rank:{idx}/{_gen_rank_device(idx, pg_device_type)}" + for idx in range(dist.get_world_size()) + ] + else: + placements = [ + f"rank:{idx}/{_gen_rank_device(dist.get_global_rank(pg, idx), pg_device_type)}" + for idx in range(pg.size()) + ] + return ChunkShardingSpec( + dim=0, + placements=cast(list[_remote_device | str], placements), + ) + + +def _is_nested_tensor(val: torch.Tensor) -> bool: + if type(val) is ShardedTensor: + if len(val.local_shards()) == 0: + return False + if type(val.local_shards()[0].tensor) is ShardedTensor: + return True + if type(val.local_shards()[0].tensor) is DTensor: + raise ValueError("Cannot handle DTensor nested inside ShardedTensor") + elif type(val) is DTensor and ( + type(val._local_tensor) is DTensor or type(val._local_tensor) is ShardedTensor + ): + raise ValueError("Cannot handle nested DTensor") + return False + + +def _alloc_tensor( + props: TensorProperties, size: Sequence[int], device_type: str = "cuda" +) -> torch.Tensor: + if device_type == "cpu": + device = cast(torch.device, _get_device_module(device_type).current_device()) + else: + device = torch.device( + device_type, _get_device_module(device_type).current_device() + ) + + return torch.empty( + size=size, + dtype=props.dtype, + layout=props.layout, + requires_grad=props.requires_grad, + pin_memory=props.pin_memory, + device=device, + ) + + +def _get_state_dict_2d_layout( + state_dict: STATE_DICT_TYPE, +) -> tuple[STATE_DICT_2D_LAYOUT, dist.ProcessGroup | None]: + """ + Load the right TP slice of the optimizer state. + + This is not easy since the per-tensor slicing can't be inferred from checkpoint metadata. + We take advantage of the model state_dict producing a sliced ST to figure out what we need to load. + This is pretty fragile and it might be easier for FSDP to compute this info for us. + Returns a dictionary where keys are the same of the state_dict and the value is a tuple of + (offset, size) for the current rank TP slice. + N.B. The state_dict *MUST* come from FSDP.sharded_state_dict. + """ + specs: STATE_DICT_2D_LAYOUT = {} + dp_pg: dist.ProcessGroup | None = None + for key, value in state_dict.items(): + specs[key] = (None, value.size()) + if _is_nested_tensor(value): + if not len(value.local_shards()) == 1: + raise AssertionError("Cannot handle ST with multiple shards") + if not isinstance(value, ShardedTensor): + raise AssertionError("Can only handle nested ShardedTensor") + shard = value.local_shards()[0] + specs[key] = ( + shard.metadata.shard_offsets, + shard.metadata.shard_sizes, + ) + dp_pg = shard.tensor._process_group # type: ignore[attr-defined] + + return ( + specs, + dp_pg, + ) + + +class _ReaderWithOffset(DefaultLoadPlanner): + translation: dict[MetadataIndex, MetadataIndex] + state_dict: STATE_DICT_TYPE + # pyrefly: ignore [bad-override] + metadata: Metadata + + def __init__(self, fqn_to_offset: dict[str, Sequence[int]]) -> None: + super().__init__() + self.fqn_to_offset = fqn_to_offset + self.metadata = Metadata({}) + self.state_dict = {} + self.translation = {} + + def create_local_plan(self) -> LoadPlan: + requests = [] + self.translation = {} + for fqn, obj in self.state_dict.items(): + md = self.metadata.state_dict_metadata[fqn] + if not isinstance(obj, ShardedTensor): + requests += _create_read_items(fqn, md, obj) + continue + + if fqn not in self.fqn_to_offset: + requests += _create_read_items(fqn, md, obj) + continue + + offset = self.fqn_to_offset[fqn] + + if not len(obj.local_shards()) == 1: + raise AssertionError("Expected exactly one local shard") + original_shard = obj.local_shards()[0] + local_chunks = [ + ChunkStorageMetadata( + offsets=torch.Size( + _element_wise_add(original_shard.metadata.shard_offsets, offset) + ), + sizes=torch.Size(original_shard.metadata.shard_sizes), + ) + ] + + reqs = create_read_items_for_chunk_list( + fqn, cast(TensorStorageMetadata, md), local_chunks + ) + # TODO: The ReadItems will have a displaced MetadataIndex, fix it. + # TODO: we should change _create_sharded_read_items to have more ergonomic API + for ri in reqs: + if ri.dest_index.offset is None: + raise AssertionError("dest_index.offset must not be None") + original_offset = _element_wise_sub(ri.dest_index.offset, offset) + original_index = dataclasses.replace( + ri.dest_index, offset=torch.Size(original_offset) + ) + self.translation[ri.dest_index] = original_index + + requests += reqs + return LoadPlan(requests) + + def lookup_tensor(self, index: MetadataIndex) -> torch.Tensor: + return super().lookup_tensor(self.translation.get(index, index)) + + +def load_sharded_optimizer_state_dict( + model_state_dict: STATE_DICT_TYPE, + optimizer_key: str, + storage_reader: StorageReader, + planner: LoadPlanner | None = None, +) -> STATE_DICT_TYPE: + """ + Load a state_dict in conjunction with FSDP sharded optimizer state. + + This is the current recommended way to checkpoint FSDP. + + Examples:: + + >>> # xdoctest: +SKIP + >>> import torch.distributed.checkpoint as dist_cp + >>> # Save + >>> model: torch.nn.Model + >>> optim_params = model.parameters() + >>> optim = torch.optim.SGD(optim_params, lr=0.01) + >>> # Save + >>> with FSDP.state_dict_type(model, StateDictType.SHARDED_STATE_DICT): + >>> state_dict = { + >>> "optimizer": FSDP.optim_state_dict(model, optim), + >>> "model": model.state_dict() + >>> } + >>> dist_cp.save_state_dict( + >>> state_dict=optim_state, + >>> storage_writer=dist_cp.FileSystemWriter("checkpoint"), + >>> planner=dist_cp.DefaultSavePlanner(), + >>> ) + >>> + >>> # Load + >>> with FSDP.state_dict_type(model_tp, StateDictType.SHARDED_STATE_DICT): + >>> model_state_dict = model_tp.state_dict() + >>> checkpoint = { + >>> "model": model_state_dict + >>> } + >>> dist_cp.load_state_dict( + >>> state_dict=checkpoint, + >>> storage_reader=dist_cp.FileSystemReader(checkpoint_file), + >>> planner=dist_cp.DefaultLoadPlanner(), + >>> ) + >>> model.load_state_dict(checkpoint["model_state"]) + >>> + >>> optim_state = dist_cp.load_sharded_optimizer_state_dict( + >>> model_state_dict, + >>> optimizer_key="optimizer", + >>> storage_reader=dist_cp.FileSystemReader("checkpoint"), + >>> ) + >>> + >>> flattened_osd = FSDP.optim_state_dict_to_load( + >>> model, optim, optim_state["optimizer"] + >>> ) + >>> + >>> optim.load_state_dict(flattened_osd) + """ + metadata = storage_reader.read_metadata() + + layout_specs, dp_pg = _get_state_dict_2d_layout(model_state_dict) + dp_pg_device_type = dist.distributed_c10d._get_pg_default_device(dp_pg).type + device_module = _get_device_module(dp_pg_device_type) + + if dp_pg is None: + placements = [] + for i in range(dist.get_world_size()): + device_info = _normalize_device_info( + dp_pg_device_type, i % device_module.device_count() + ) + placements.append(f"rank:{i}/{device_info}") + sharding_spec = ChunkShardingSpec(dim=0, placements=placements) # type: ignore[arg-type] + else: + sharding_spec = _create_colwise_spec(dp_pg) + + # Create a state_dict for optimizer state + state_dict: STATE_DICT_TYPE = {} + + fqn_to_offset: dict[str, Sequence[int]] = {} + for key, value in metadata.state_dict_metadata.items(): + key_path = metadata.planner_data[key] + if key_path[0] != optimizer_key: + continue + + if isinstance(value, BytesStorageMetadata): + state_dict[key] = "" + continue + + # value: TensorStorageMetadata + if value.size.numel() == 1: + state_dict[key] = _alloc_tensor( + value.properties, value.size, dp_pg_device_type + ) + elif dp_pg is None: + state_dict[key] = _create_chunk_sharded_tensor( + _alloc_tensor(value.properties, value.size, dp_pg_device_type), + rank=dist.get_rank(), + world_size=dist.get_world_size(), + num_devices_per_node=device_module.device_count(), + pg=_get_default_group(), + ) + else: + spec_key = key_path[2] + alloc_size = layout_specs.get(spec_key, (None, value.size))[1] + + properties = ShardTensorProperties( + dtype=value.properties.dtype, + layout=value.properties.layout, + requires_grad=value.properties.requires_grad, + memory_format=value.properties.memory_format, + pin_memory=value.properties.pin_memory, + ) + + st_md = sharding_spec.build_metadata(torch.Size(alloc_size), properties) + local_shards = [] + current_rank = dist.get_rank(dp_pg) + for shard_md in st_md.shards_metadata: + if cast(_remote_device, shard_md.placement).rank() != current_rank: + continue + local_shards.append( + Shard( + tensor=_alloc_tensor( + value.properties, shard_md.shard_sizes, dp_pg_device_type + ), + metadata=shard_md, + ) + ) + + st = ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards, st_md, process_group=dp_pg + ) + + if spec_key in layout_specs and layout_specs[spec_key][0] is not None: + fqn_to_offset[key] = cast(Sequence[int], layout_specs[spec_key][0]) + + state_dict[key] = st + + # Whether we unflatten before or after doesn't matter + load_state_dict( + state_dict=state_dict, + storage_reader=storage_reader, + # FIXME the type of planner is wrong in load_state_dict + planner=_ReaderWithOffset(fqn_to_offset) if dp_pg is not None else planner, + ) + + state_dict = unflatten_state_dict(state_dict, metadata.planner_data) + + return state_dict diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner.py new file mode 100644 index 0000000000000000000000000000000000000000..1558592534aec9a63f59b6d254652027f89ce4e8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner.py @@ -0,0 +1,450 @@ +import abc +import io +import operator +from dataclasses import dataclass +from enum import auto, Enum +from functools import reduce +from typing import Any + +import torch +from torch.distributed.checkpoint.metadata import ( + ChunkStorageMetadata, + Metadata, + MetadataIndex, + STATE_DICT_TYPE, + StorageMeta, + TensorProperties, +) + + +__all__ = [ + "WriteItemType", + "LoadItemType", + "BytesIOWriteData", + "TensorWriteData", + "WriteItem", + "ReadItem", + "SavePlan", + "LoadPlan", + "SavePlanner", + "LoadPlanner", +] + + +class WriteItemType(Enum): + TENSOR = auto() + SHARD = auto() + BYTE_IO = auto() + + +class LoadItemType(Enum): + TENSOR = auto() + BYTE_IO = auto() + + +@dataclass(frozen=True) +class BytesIOWriteData: + nbytes: int + + +@dataclass(frozen=True) +class TensorWriteData: + chunk: ChunkStorageMetadata + properties: TensorProperties + size: torch.Size + + +@dataclass(frozen=True) +class WriteItem: + """Dataclass which holds information about what needs to be written to storage.""" + + index: MetadataIndex + type: WriteItemType + + # Size of bytesIO data to be written. + bytes_io_data: BytesIOWriteData | None = None + + # Value present if it's a tensor write + tensor_data: TensorWriteData | None = None + + def tensor_storage_size(self) -> int | None: + """ + Calculates the storage size of the underlying tensor, or None if this is not a tensor write. + + Returns: + Optional[int] storage size, in bytes of underlying tensor if any. + """ + if self.tensor_data is None: + return None + + numels = reduce(operator.mul, self.tensor_data.size, 1) + dtype_size = torch._utils._element_size(self.tensor_data.properties.dtype) + return numels * dtype_size + + +@dataclass(frozen=True) +class ReadItem: + # Read Item + type: LoadItemType + + # Index into the state_dict + dest_index: MetadataIndex + # Offsets into destination tensor + dest_offsets: torch.Size + + # Index into the checkpoint + storage_index: MetadataIndex + # Offset into the checkpoint data + storage_offsets: torch.Size + + # Size of the hypercube to copy + lengths: torch.Size + + +@dataclass(frozen=True) +class SavePlan: + items: list[WriteItem] + storage_data: Any = None + planner_data: Any = None + # This is used to indicate that the ranks should + # use the cached plans to write data instead. + usable: bool = True + + +@dataclass +class LoadPlan: + items: list[ReadItem] + storage_data: Any = None + planner_data: Any = None + + +class SavePlanner(abc.ABC): + """ + Abstract class defining the protocol used by save_state_dict to plan the save process. + + SavePlanners are stateful objects that can be used to customize the whole save process. + + SavePlanner acts as an access proxy to the state_dict, so any transformation done to it + will be visible to the whole process. + + A planner subclass can expect the following sequence of calls during save_state_dict: + + 1) set_up_planner - called on all ranks. + Signals the start of a checkpoint save. + + 2) create_local_plan - called on all ranks. + Process the state_dict and produces a `SavePlan` that will be sent for global planning. + + 3) create_global_plan - called on the coordinator rank only. + Takes the SavePlan from all ranks and make any global decision. + + 4) finish_plan - called on all ranks. + This gives each rank a chance to adjust to global planning decisions. + + 5) resolve_data - called multiple times on each rank + Lookups a value on the `state_dict` for the storage layer to write. + + Users are recommended to extend DefaultSavePlanner instead of this interface directly as + most changes can be expressed by changes in a single method. + + There are 3 usual patterns of extension: + + Rewriting state_dict. This is the simplest way to extend the save process as it + doesn't require understanding the intricacies of how SavePlan works: + + >>> # xdoctest: +SKIP("undefined vars") + >>> class RenamePlanner(DefaultSavePlanner): + >>> def set_up_planner( + >>> self, + >>> state_dict: STATE_DICT_TYPE, + >>> storage_meta: Optional[StorageMeta], + >>> is_coordinator: bool, + >>> ) -> None: + >>> # prefix all keys with `foo_`` + >>> super().set_up_planner({"foo_" + k: v for k, v in state_dict.items()}, storage_meta, is_coordinator) + + Modifying local plan and lookup in tandem. This is useful when fine control of how data is persisted + + >>> # xdoctest: +SKIP("undefined vars") + >>> class FP16Planner(DefaultSavePlanner): + >>> def create_local_plan(self): + >>> plan = super().create_local_plan() + >>> for p in plan: + >>> if p.tensor_data is not None: + >>> p.tensor_data.properties.dtype = torch.float16 + >>> return plan + >>> + >>> def resolve_data(self, write_item): + >>> item = super().resolve_data(write_item) + >>> return item if write_item.type == WriteItemType.BYTE_IO else item.to(torch.float16) + + Using the global planning step to make central decisions that can't be made individually by each rank + + >>> # xdoctest: +SKIP("undefined vars") + >>> from itertools import zip_longest + >>> from dataclasses import replace + >>> class DDPLoadBalancingPlanner(DefaultSavePlanner): + >>> # This uses the default local plan behavior of having all non-sharded writes in rank 0 + >>> # This sample doesn't handle ShardedTensors + >>> def create_global_plan(self, all_plans): + >>> iters = [iter(all_plans[0].items)] * len(all_plans) + >>> items_per_rank = [ + >>> [item for item in items if item is not None] + >>> for items in zip(*zip_longest(*iters), strict=True) + >>> ] + >>> all_plans = [ + >>> replace(plan, items=items) + >>> for plan, items in zip(all_plans, items_per_rank, strict=True) + >>> ] + >>> return super().create_global_plan(all_plans) + + Finally, some planners need to save additional metadata in the checkpoint, this is + accomplished by having each rank contribute their data items in the local plan and + the global planner aggregate them: + + >>> # xdoctest: +SKIP("undefined vars") + >>> class SaveExtraDataPlanner(DefaultSavePlanner): + >>> def create_local_plan(self) -> SavePlan: + >>> plan = super().create_local_plan() + >>> return replace(plan, planner_data="per-rank-data") + >>> + >>> def create_global_plan(self, all_plans: List[SavePlan]) -> Tuple[List[SavePlan], Metadata]: + >>> global_plan, metadata = super().create_global_plan(all_plans) + >>> merged_data = [p.planner_data for p in global_plan] + >>> metadata = replace(metadata, planner_data=merged_data) + >>> return global_plan, metadata + """ + + # Save plan for the current rank as computed by `create_local_plan` API + # Cached on the local rank. + _cached_save_plan: dict[str, SavePlan] = {} + # Final save plan for the current rank. + # This is created by merging the plan created by `create_local_plan` API + # and the result of `create_global_plan` for the given rank. + # This is the final plan computed by the `finish_plan` API that gets + # sent to the `write_data`. + # Cached on the local rank. + _cached_final_save_plan: dict[str, SavePlan] = {} + # Collection of all the local plans from all the ranks. + # This is the input to the `create_global_plan` API. + # Cached on the coordinator rank. + _cached_all_plans: dict[str, list[SavePlan]] = {} + # Global checkpoint plan as computed by `create_global_plan` API. + # Cached on the coordinator rank. + _cached_global_plan: dict[str, list[SavePlan]] = {} + # Metadata for the global checkpoint plan as computed by `create_global_plan` API. + # Cached on the coordinator rank. + _cached_metadata: dict[str, Metadata] = {} + + @abc.abstractmethod + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + storage_meta: StorageMeta | None = None, + is_coordinator: bool = False, + ) -> None: + """ + Initialize this planner to save ``state_dict``. + + Implementations should save those values as they won't be provided later in the save process. + + This is called on all ranks. + """ + + @abc.abstractmethod + def create_local_plan(self) -> SavePlan: + """ + Compute the save plan for the current rank. + + This will be aggregated and passed to create_global_plan. + Planner specific data can be passed through SavePlan::planner_data. + + This is called on all ranks. + """ + + @abc.abstractmethod + def create_global_plan( + self, all_plans: list[SavePlan] + ) -> tuple[list[SavePlan], Metadata]: + """ + Compute the global checkpoint plan and return the local plan of each rank. + + This is called on the coordinator rank only. + """ + + @abc.abstractmethod + def finish_plan(self, new_plan: SavePlan) -> SavePlan: + """ + Merge the plan created by `create_local_plan` and the result of `create_global_plan`. + + This is called on all ranks. + """ + + @abc.abstractmethod + def resolve_data(self, write_item: WriteItem) -> torch.Tensor | io.BytesIO: + """ + Transform and prepare ``write_item`` from ``state_dict`` for storage, ensuring idempotency and thread-safety. + + Lookup the object associated with ``write_item`` in ``state_dict`` and apply any + transformation (such as serialization) prior to the storage layer consuming it. + + Called on each rank multiple times, at least once per WriteItem in the final SavePlan. + + This method should be idempotent and thread-save. StorageWriter implementations + are free to call it as frequently as they need. + + Any transformation that allocates memory should be lazily done when his method + is called in order to reduce peak memory required by checkpointing. + + When returning tensors, they can be on any device or format, they can be views too. + It's the storage layer responsibility to figure out how to save them. + """ + + +class LoadPlanner: + """ + Abstract class defining the protocol used by load_state_dict to plan the load process. + + LoadPlanner are stateful objects that can be used to customize the whole load process. + + LoadPlanner acts as an access proxy to the state_dict, so any transformation done to it + will be visible to the whole process. + + A planner subclass can expect the following sequence of calls during load_state_dict: + + 1) set_up_planner - called on all ranks. + Signals the start of loading a checkpoint. + + 2) create_local_plan - called on all ranks. + Process the state_dict and produces a `LoadPlan` that will be sent for global planning. + + 3) create_global_plan - called on the coordinator rank only. + Takes the LoadPlan from all ranks and make any global decision. + + 4) load_bytes - called multiple times on each rank + This is called once per non-tensor value in state_dict. + + 5) resolve_tensor and commit_tensor - called multiple times on each rank + They are called in pair for each Tensor value in state_dict. + + Users are recommended to extend DefaultLoadPlanner instead of this interface directly as + most changes can be expressed by changes in a single method. + + There are two usual patterns of extension: + + Rewriting state_dict. This is the simplest way to extend the load process as it + doesn't require understanding the intricacies of how LoadPlan works. We need + to keep a reference to the original state_dict as load happens in place so + we need to be able to perform it in place + + >>> # xdoctest: +SKIP("undefined vars") + >>> class RenamePlanner(DefaultLoadPlanner): + >>> def set_up_planner( + >>> self, + >>> state_dict: STATE_DICT_TYPE, + >>> metadata: Metadata, + >>> is_coordinator: bool, + >>> ) -> None: + >>> self.original_state_dict = state_dict + >>> state_dict = {"foo_" + k: v for k, v in state_dict.items()} + >>> + >>> if self.flatten_sharded_tensors: + >>> state_dict = _flatten_sharded_tensors(state_dict) + >>> + >>> if self.flatten_state_dict: + >>> state_dict, self.mappings = flatten_state_dict(state_dict) + >>> + >>> self.state_dict = state_dict + >>> self.metadata = metadata + >>> self.is_coordinator = is_coordinator + >>> + >>> def load_bytes(self, read_item, value): + >>> # Remove the "foo_" prefix + >>> self.original_state_dict[read_item.dest_index.fqn[4:]] = torch.load(value, weights_only=False) + + + Modifying resolve_tensor and commit_tensor to handle load time transformation. + + >>> # xdoctest: +SKIP("undefined vars") + >>> class MetaModelMaterialize(DefaultSavePlanner): + >>> def resolve_tensor(self, read_item): + >>> tensor = super().resolve_tensor(read_item) + >>> return torch.empty_like(tensor, device="cpu") + >>> + >>> def commit_tensor(self, read_item, tensor): + >>> self.state_dict[read_item.dest_index.fqn] = tensor + """ + + @abc.abstractmethod + def set_up_planner( + self, + state_dict: STATE_DICT_TYPE, + metadata: Metadata | None = None, + is_coordinator: bool = False, + ) -> None: + """ + Initialize this instance to load data into ``state_dict``. + + . N.B. This is called on every rank. + """ + + @abc.abstractmethod + def create_local_plan(self) -> LoadPlan: + """ + Create a LoadPlan based on state_dict and metadata provided by set_up_planner. + + . N.B. This is called on every rank. + """ + + @abc.abstractmethod + def create_global_plan(self, global_plan: list[LoadPlan]) -> list[LoadPlan]: + """ + Compute the global load plan and return plans for each rank. + + . N.B. This is called on the coordinator rank only + """ + + @abc.abstractmethod + def finish_plan(self, central_plan: LoadPlan) -> LoadPlan: + """Accept the plan from coordinator and return final LoadPlan.""" + + @abc.abstractmethod + def load_bytes(self, read_item: ReadItem, value: io.BytesIO) -> None: + """ + Load the item described by ``read_item``and ``value``. + + This method is expected to modify in-place the underlying state_dict. + + The contents of ``value`` are defined by the SavePlanner used to produce + the checkpoint being loaded. + """ + + def resolve_bytes(self, read_item: ReadItem) -> io.BytesIO: + """ + Return the BytesIO to be used by the StorageReader to load `read_item`. + + The BytesIO should alias with one on the underlying state_dict as StorageReader will replace its contents. + """ + raise NotImplementedError("LoadPlanner.resolve_bytes is not implemented") + + @abc.abstractmethod + def resolve_tensor(self, read_item: ReadItem) -> torch.Tensor: + """ + Return the tensor described by ``read_item`` to be used by the StorageReader to load `read_item`. + + The tensor should alias with one on the underlying state_dict as StorageReader will replace its contents. + If, for any reason, that's not possible, the planner can use the ``commit_tensor`` method to copy the data + back to the one in state_dict. + """ + + @abc.abstractmethod + def commit_tensor(self, read_item: ReadItem, tensor: torch.Tensor) -> None: + """ + Call once the StorageReader finished loading data into ``tensor``. + + The provided tensor is the same one returned by the call to ``resolve_tensor``. + This method is only needed if this LoadPlanner needs to post process ``tensor`` prior to + copying it back to the one in the state_dict. + + The contents of tensor will follow its device synchronization model. + """ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner_helpers.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner_helpers.py new file mode 100644 index 0000000000000000000000000000000000000000..ecdc7a143abd5165cfb98fb36f38672819b9b938 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/planner_helpers.py @@ -0,0 +1,597 @@ +# mypy: allow-untyped-defs +import io +import itertools +from bisect import bisect_right, insort +from collections.abc import Callable +from typing import Any, cast + +import torch +import torch.distributed as dist +from torch._utils import _get_device_module +from torch.distributed._shard.metadata import ShardMetadata +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed.tensor import DTensor +from torch.distributed.tensor._utils import compute_local_shape_and_global_offset + +from .metadata import ( + BytesStorageMetadata, + ChunkStorageMetadata, + MetadataIndex, + STATE_DICT_TYPE, + STORAGE_TYPES, + TensorProperties, + TensorStorageMetadata, +) +from .planner import ( + LoadItemType, + ReadItem, + SavePlan, + TensorWriteData, + WriteItem, + WriteItemType, +) +from .protocol import ( + _copy_checkpointable_tensor_metadata, + _get_checkpointable_tensor_chunks, + _is_checkpointable_tensor, + CheckpointableTensor, +) +from .resharding import ( + _check_shard_metadata_pair_overlap, + _shards_get_overlap_region_wrt_saved_tensor, +) + + +__all__: list[str] = ["create_read_items_for_chunk_list"] + + +def _compare_save_plans(plan: SavePlan, other_plan: SavePlan) -> bool: + """ + Compare the two Save plans and return True if they are equal. + + Args: + plan (SavePlan): First SavePlan to compare. + other_plan (SavePlan): Second SavePlan to compare. + + Returns: + True if the two plans are equal, False otherwise. + """ + if plan.usable != other_plan.usable: + return False + + # Both the plans should have the same number of items + if len(plan.items) != len(other_plan.items): + return False + + # Both the plans should have the same write items. + for plan_item, other_plan_item in zip(plan.items, other_plan.items): + # Write item type should be same + if plan_item.type != other_plan_item.type: + return False + + plan_metadata_index = plan_item.index + other_plan_metadata_index = other_plan_item.index + + # Write item metadata_index should be same + if ( + plan_metadata_index.fqn != other_plan_metadata_index.fqn + or plan_metadata_index.offset != other_plan_metadata_index.offset + or plan_metadata_index.index != other_plan_metadata_index.index + ): + return False + + # Write item tensor_data should be present in both the write items plans, if it exists in either of them. + tensor_data = plan_item.tensor_data + other_tensor_data = other_plan_item.tensor_data + if (tensor_data and not other_tensor_data) or ( + not tensor_data and other_tensor_data + ): + return False + + if tensor_data and other_tensor_data: + # Write item tensor_data size should be same + if tensor_data.size != other_tensor_data.size: + return False + + # Write item tensor_data chunk should be present in both the write items, if it exists in either of them. + chunk = tensor_data.chunk + other_chunk = other_tensor_data.chunk + if (chunk and not other_chunk) or (not chunk and other_chunk): + return False + + # Write item tensor_data chunk offsets and sizes should be same + if chunk and other_chunk: + if ( + chunk.offsets != other_chunk.offsets + or chunk.sizes != other_chunk.sizes + ): + return False + + return True + + +def _contains_usable_plan(delta_plans: list[SavePlan]) -> bool: + """ + Check if any delta plan is usable, indicating the plan has changed. + + Args: + delta_plans (List[SavePlan]): A list of delta plans to check. + Returns: + True if any delta plan is usable, False otherwise. + """ + return any(delta_plan and delta_plan.usable for delta_plan in delta_plans) + + +def _merge_delta_local_plans( + cached_plans: list[SavePlan], + delta_plans: list[SavePlan], +) -> list[SavePlan]: + """ + Merge a list of delta plans into a single plan. + + Args: + cached_plans (List[SavePlan]): A list of cached plans. + delta_plans (List[SavePlan]): A list of delta plans to merge. It can contain empty plans + + Returns: + A single merged plan. If a delta plan is not usable, use the cached plan. Otherwise, use the delta plan. + """ + merged_plans = [] + + for cached_plan, delta_plan in zip(cached_plans, delta_plans): + if delta_plan and not delta_plan.usable: + merged_plans.append(cached_plan) + else: + merged_plans.append(delta_plan) + + return merged_plans + + +def _create_chunk_from_tensor(tensor: torch.Tensor) -> ChunkStorageMetadata: + return ChunkStorageMetadata( + offsets=torch.Size([0] * len(tensor.size())), sizes=tensor.size() + ) + + +def _chunk_for_shard(shard_md: ShardMetadata) -> ChunkStorageMetadata: + return ChunkStorageMetadata( + offsets=torch.Size(shard_md.shard_offsets), + sizes=torch.Size(shard_md.shard_sizes), + ) + + +def _sharded_tensor_metadata( + sharded_tensor: ShardedTensor, shard_md: ShardMetadata +) -> TensorWriteData: + shard_properties = sharded_tensor.metadata().tensor_properties + + properties = TensorProperties( + dtype=shard_properties.dtype, + layout=shard_properties.layout, + requires_grad=shard_properties.requires_grad, + memory_format=shard_properties.memory_format, + pin_memory=shard_properties.pin_memory, + ) + + return TensorWriteData( + chunk=_chunk_for_shard(shard_md), + properties=properties, + size=sharded_tensor.metadata().size, + ) + + +def _create_write_items_for_dtensor(fqn: str, tensor: DTensor) -> WriteItem: + sizes, offsets = compute_local_shape_and_global_offset( + tensor.shape, tensor.device_mesh, tensor.placements + ) + sizes, offsets = torch.Size(sizes), torch.Size(offsets) + + return WriteItem( + index=MetadataIndex(fqn, offsets), + type=WriteItemType.SHARD, + tensor_data=TensorWriteData( + chunk=ChunkStorageMetadata( + offsets=offsets, + sizes=sizes, + ), + properties=TensorProperties.create_from_tensor(tensor.to_local()), + size=tensor.size(), + ), + ) + + +def _create_write_item_for_shard( + fqn: str, sharded_tensor: ShardedTensor, shard_md: ShardMetadata +) -> WriteItem: + offsets = torch.Size(shard_md.shard_offsets) + return WriteItem( + index=MetadataIndex(fqn, offsets), + type=WriteItemType.SHARD, + tensor_data=_sharded_tensor_metadata(sharded_tensor, shard_md), + ) + + +def _create_write_item_for_tensor(fqn: str, tensor: torch.Tensor) -> WriteItem: + offsets = torch.Size([0] * len(tensor.size())) + return WriteItem( + index=MetadataIndex(fqn, offsets), + type=WriteItemType.TENSOR, + tensor_data=TensorWriteData( + chunk=ChunkStorageMetadata(offsets=offsets, sizes=tensor.size()), + properties=TensorProperties.create_from_tensor(tensor), + size=tensor.size(), + ), + ) + + +def _get_checkpointable_tensor_write_items( + fqn: str, tensor: CheckpointableTensor +) -> list[WriteItem]: + if not isinstance(tensor, torch.Tensor): + raise TypeError("CheckpointableTensor must also be a torch.Tensor") + properties = TensorProperties.create_from_tensor(tensor) + return [ + WriteItem( + index=MetadataIndex(fqn, chunk.offsets, idx), + type=WriteItemType.SHARD, + tensor_data=TensorWriteData( + chunk=chunk, + properties=properties, + size=torch.Size(tensor.global_shape), + ), + ) + for idx, chunk in enumerate(_get_checkpointable_tensor_chunks(tensor)) + ] + + +def _create_write_item_for_bytesio(fqn: str, bytes: Any): + return WriteItem( + index=MetadataIndex(fqn), + type=WriteItemType.BYTE_IO, + ) + + +def _create_read_item_for_byteio( + dest_index, dest_offset, storage_index, storage_offset, length +): + return ReadItem( + type=LoadItemType.BYTE_IO, + dest_index=dest_index, + dest_offsets=torch.Size((dest_offset,)), + storage_index=storage_index, + storage_offsets=torch.Size((storage_offset,)), + lengths=torch.Size((length,)), + ) + + +def _create_read_item_for_tensor( + dest_index, dest_offsets, storage_index, storage_offsets, lengths +): + return ReadItem( + type=LoadItemType.TENSOR, + dest_index=dest_index, + dest_offsets=torch.Size(dest_offsets), + storage_index=storage_index, + storage_offsets=torch.Size(storage_offsets), + lengths=torch.Size(lengths), + ) + + +def create_read_items_for_chunk_list( + fqn: str, + checkpoint_md: TensorStorageMetadata, + local_chunks: list[ChunkStorageMetadata], +) -> list[ReadItem]: + """ + Create a list of ``ReadItem`` based on the checkpoint and local chunks. + + This applies the resharding algorithm and computes the reads needed + to satisfy ``local_chunks`` with a checkpoint described by ``checkpoint_md``. + + Args: + fqn (str) : The state_dict FQN to pass to ``ReadItem``. + checkpoint_md (TensorStorageMetadata): metadata for a given tensor + from a checkpoint. + local_chunks (List[ChunkStorageMetadata]): Local chunks that needs to be + loaded. + + Returns: + A list of ``ReadItem`` that will satisfy all input chunks. + """ + read_items: list[ReadItem] = [] + saved_chunks = checkpoint_md.chunks + + if not local_chunks or not saved_chunks: + return read_items + + num_dims = len(local_chunks[0].offsets) + + # Find sweep dimension (dimension with largest extent for better pruning) + sweep_dim = 0 + if num_dims > 1: + max_size = 0 + for dim in range(num_dims): + dim_size = max( + chunk.offsets[dim] + chunk.sizes[dim] + for chunk in itertools.chain(local_chunks, saved_chunks) + ) + if dim_size > max_size: + max_size = dim_size + sweep_dim = dim + + # Pre-compute bounds: (start, end) for each chunk in sweep dimension + # For 0-d tensors, use (0, 1) so all chunks overlap in the sweep line + if num_dims == 0: + saved_bounds = [(0, 1)] * len(saved_chunks) + local_bounds = [(0, 1)] * len(local_chunks) + else: + saved_bounds = [ + (c.offsets[sweep_dim], c.offsets[sweep_dim] + c.sizes[sweep_dim]) + for c in saved_chunks + ] + local_bounds = [ + (c.offsets[sweep_dim], c.offsets[sweep_dim] + c.sizes[sweep_dim]) + for c in local_chunks + ] + + saved_sorted_indices = sorted( + range(len(saved_chunks)), + key=lambda idx: saved_bounds[idx][0], + ) + local_sorted_indices = sorted( + range(len(local_chunks)), + key=lambda idx: local_bounds[idx][0], + ) + + active_saved: list[tuple[int, int]] = [] + saved_ptr = 0 + num_saved = len(saved_sorted_indices) + + for local_idx in local_sorted_indices: + local_chunk = local_chunks[local_idx] + local_start, local_end = local_bounds[local_idx] + + cutoff = bisect_right(active_saved, (local_start, -1)) + if cutoff: + del active_saved[:cutoff] + + while saved_ptr < num_saved: + storage_idx = saved_sorted_indices[saved_ptr] + storage_chunk = saved_chunks[storage_idx] + saved_start, saved_end = saved_bounds[storage_idx] + + if saved_start >= local_end: + break + + insort(active_saved, (saved_end, storage_idx)) + saved_ptr += 1 + + for _, storage_idx in active_saved: + storage_chunk = saved_chunks[storage_idx] + if not _check_shard_metadata_pair_overlap(local_chunk, storage_chunk): + continue + + storage_offsets = [] + dest_offsets = [] + lengths = [] + for ( + _dim, + offset_for_saved_tensor, + offset_for_current_tensor, + length, + ) in _shards_get_overlap_region_wrt_saved_tensor( + saved_shard=storage_chunk, current_shard=local_chunk + ): + storage_offsets.append(offset_for_saved_tensor) + dest_offsets.append(offset_for_current_tensor) + lengths.append(length) + + read_items.append( + _create_read_item_for_tensor( + dest_index=MetadataIndex(fqn, local_chunk.offsets, local_idx), + dest_offsets=dest_offsets, + storage_index=MetadataIndex( + fqn, storage_chunk.offsets, storage_idx + ), + storage_offsets=storage_offsets, + lengths=lengths, + ) + ) + return read_items + + +def _create_default_metadata_only_plan(state_dict: STATE_DICT_TYPE) -> SavePlan: + requests = [] + for fqn, obj in state_dict.items(): + if isinstance(obj, DTensor): + requests.append(_create_write_items_for_dtensor(fqn, obj)) + elif isinstance(obj, ShardedTensor): + requests.extend( + _create_write_item_for_shard(fqn, obj, shard_md) + for shard_md in obj.metadata().shards_metadata + ) + elif _is_checkpointable_tensor(obj): + # Avoid treating a local shard as a plain Tensor: the WriteItems + # carry both local chunk metadata and the global tensor size. + requests.extend(_get_checkpointable_tensor_write_items(fqn, obj)) + elif isinstance(obj, torch.Tensor): + requests.append(_create_write_item_for_tensor(fqn, obj)) + else: + requests.append(_create_write_item_for_bytesio(fqn, obj)) + return SavePlan(requests) + + +def _create_write_items(fqn: str, object: Any) -> list[WriteItem]: + if hasattr(object, "__create_write_items__"): + # DTensor implements _Checkpointable + return object.__create_write_items__(fqn, object) # type: ignore[attr-defined] + elif isinstance(object, ShardedTensor): + return [ + _create_write_item_for_shard(fqn, object, shard.metadata) + for shard in object.local_shards() + ] + elif _is_checkpointable_tensor(object): + return _get_checkpointable_tensor_write_items(fqn, object) + elif isinstance(object, torch.Tensor): + return [_create_write_item_for_tensor(fqn, object)] + else: + return [_create_write_item_for_bytesio(fqn, object)] + + +def _create_chunk_from_dtensor(tensor: DTensor) -> ChunkStorageMetadata: + sizes, offsets = compute_local_shape_and_global_offset( + tensor.shape, tensor.device_mesh, tensor.placements + ) + sizes, offsets = torch.Size(sizes), torch.Size(offsets) + return ChunkStorageMetadata( + offsets=offsets, + sizes=sizes, + ) + + +def _create_chunk_list(tensor: torch.Tensor) -> list[ChunkStorageMetadata]: + if hasattr(tensor, "__create_chunk_list__"): + # DTensor implements _Checkpointable + local_chunks = tensor.__create_chunk_list__() # type: ignore[attr-defined] + elif _is_checkpointable_tensor(tensor): + local_chunks = _get_checkpointable_tensor_chunks(tensor) + elif isinstance(tensor, ShardedTensor): + local_chunks = [ + _chunk_for_shard(shard.metadata) for shard in tensor.local_shards() + ] + elif isinstance(tensor, torch.Tensor): + local_chunks = [_create_chunk_from_tensor(tensor)] + else: + raise ValueError( + "Unsupported Type, expecting one of [Tensor, DTensor, ShardedTensor] " + f",but got {type(tensor)}" + ) + + return local_chunks + + +def _create_read_items(fqn: str, md: STORAGE_TYPES, obj: Any) -> list[ReadItem]: + if not isinstance(md, BytesStorageMetadata): + try: + local_chunks = _create_chunk_list(obj) + except ValueError as ex: + raise ValueError( + f"Invalid checkpoint metadata for {fqn}, " + + f"expected BytesStorageMetadata but found {type(md)}", + ) from ex + + return create_read_items_for_chunk_list(fqn, md, local_chunks) + else: + return [ + _create_read_item_for_byteio( + dest_index=MetadataIndex(fqn), + dest_offset=0, + storage_index=MetadataIndex(fqn), + storage_offset=0, + length=0, + ) + ] + + +def _init_state_dict(state_dict: dict[str, Any]) -> Any: + """ + Initializes meta tensor if the meta tensor is DTensor or torch.Tensor. + """ + + def dtensor_func(value: DTensor): + device = getattr(value, "device", None) + if device == torch.device("meta"): + device_type = dist.distributed_c10d._get_pg_default_device().type + device = cast( + torch.device, _get_device_module(device_type).current_device() + ) + new_local_tensor = torch.empty_like(value.to_local(), device=device) + # We need to pass shape and stride explicitly, since DTensor might be + # sharded unevenly. + dtensor = DTensor.from_local( + new_local_tensor, + device_mesh=value.device_mesh, + placements=value.placements, + shape=value.size(), + stride=value.stride(), + ) + return dtensor + else: + return value + + def sharded_tensor_func(value: Any): + device = getattr(value, "device", None) + if device == torch.device("meta"): + raise RuntimeError( + f"Found unsupported type {type(value)} for meta device loading." + ) + else: + return value + + def tensor_func(value: torch.Tensor): + device = getattr(value, "device", None) + if device == torch.device("meta"): + device_type = dist.distributed_c10d._get_pg_default_device().type + device = cast( + torch.device, _get_device_module(device_type).current_device() + ) + tensor = torch.empty_like(value, device=device) + if _is_checkpointable_tensor(value): + _copy_checkpointable_tensor_metadata(value, tensor) + return tensor + else: + return value + + _iterate_state_dict( + state_dict, + dtensor_func, + sharded_tensor_func, + tensor_func, + ) + + +def _iterate_state_dict( + iter_object: Any, + dtensor_func: Callable, + sharded_tensor_func: Callable, + tensor_func: Callable, +): + """ + Iterate through the state dict, applying the given functions to each tensor type + and update the state dict in place. + + Args: + iter_object (Any): the target state_dict. + sharded_tensor_func (Callable): the function to apply to ShardedTensor + dtensor_func (Callable): the function to apply to DTensor + tensor_func (Callable): the function to apply to Tensor + + # TODO: let state_dict_util._iterate_state_dict() to support in place option + so we don't need to have two versions of _iterate_state_dict. + """ + + if isinstance(iter_object, DTensor): + return dtensor_func(iter_object) + elif isinstance(iter_object, ShardedTensor): + return sharded_tensor_func(iter_object) + elif isinstance(iter_object, torch.Tensor): + return tensor_func(iter_object) + elif ( + isinstance(iter_object, (int, float, str, bytes, io.BytesIO)) + or iter_object is None + ): + return iter_object + elif isinstance(iter_object, dict): + for key, value in iter_object.items(): + iter_object[key] = _iterate_state_dict( + value, dtensor_func, sharded_tensor_func, tensor_func + ) + return iter_object + elif isinstance(iter_object, (list, tuple)): + ret = [ + _iterate_state_dict(v, dtensor_func, sharded_tensor_func, tensor_func) + for v in iter_object + ] + if isinstance(iter_object, tuple): + ret = tuple(ret) # type: ignore[assignment] + return ret diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/protocol.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/protocol.py new file mode 100644 index 0000000000000000000000000000000000000000..4992109127e2b8175b320ccde7dd82fb4b5bcde3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/protocol.py @@ -0,0 +1,167 @@ +from __future__ import annotations + +from typing import Protocol, runtime_checkable, TypeGuard + +import torch + +from .metadata import ChunkStorageMetadata, MetadataIndex + + +__all__ = ["CheckpointableTensor"] + + +@runtime_checkable +class CheckpointableTensor(Protocol): + """Protocol fields for checkpointing a local tensor as global tensor shards. + + A tensor does not need to be wrapped or subclassed for DCP to checkpoint it + as a shard. It can stay a regular local ``torch.Tensor``; implementing + these fields is enough for DCP to map one or more slices of that tensor + into a logical global tensor. + + Attributes: + global_shape: Full logical tensor shape to write into checkpoint + metadata; needed because ``tensor.size()`` is only the local buffer + size. + global_offsets: Global start coordinate for each local shard; needed to + name checkpoint chunks and match load requests by global offset. + local_offsets: Start coordinate for each local shard inside the local + tensor; needed when one tensor stores multiple shards or includes + padding. + local_sizes: Shape of each local shard; needed to build checkpoint + chunks and slice the local tensor during load. + """ + + global_shape: tuple[int, ...] + global_offsets: tuple[tuple[int, ...], ...] + local_offsets: tuple[tuple[int, ...], ...] + local_sizes: tuple[tuple[int, ...], ...] + + +def _is_checkpointable_tensor(obj: object) -> TypeGuard[CheckpointableTensor]: + return isinstance(obj, torch.Tensor) and isinstance(obj, CheckpointableTensor) + + +def _copy_checkpointable_tensor_metadata( + src: CheckpointableTensor, dst: torch.Tensor +) -> None: + setattr(dst, "global_shape", src.global_shape) # noqa: B010 + setattr(dst, "global_offsets", src.global_offsets) # noqa: B010 + setattr(dst, "local_offsets", src.local_offsets) # noqa: B010 + setattr(dst, "local_sizes", src.local_sizes) # noqa: B010 + + +def _get_checkpointable_tensor_chunks( + tensor: CheckpointableTensor, +) -> list[ChunkStorageMetadata]: + _validate_checkpointable_tensor_metadata(tensor) + return [ + ChunkStorageMetadata( + offsets=torch.Size(global_offset), + sizes=torch.Size(local_size), + ) + for global_offset, local_size in zip( + tensor.global_offsets, + tensor.local_sizes, + strict=True, + ) + ] + + +def _get_checkpointable_tensor_shard( + tensor: CheckpointableTensor, + index: MetadataIndex, +) -> torch.Tensor: + _validate_checkpointable_tensor_metadata(tensor) + + if index.offset is None: + if len(tensor.global_offsets) == 1: + shard_idx = 0 + else: + raise ValueError( + f"Cannot lookup {index.fqn} with multiple checkpointable shards and no offset" + ) + elif ( + index.index is not None + and index.index < len(tensor.global_offsets) + and torch.Size(tensor.global_offsets[index.index]) == index.offset + ): + shard_idx = index.index + else: + shard_idx = -1 + for idx, global_offset in enumerate(tensor.global_offsets): + if torch.Size(global_offset) == index.offset: + shard_idx = idx + break + if shard_idx < 0: + raise ValueError( + f"Could not find checkpointable tensor shard at '{index.offset}' " + f"for FQN: '{index.fqn}'" + ) + + local_offset = tensor.local_offsets[shard_idx] + local_size = tensor.local_sizes[shard_idx] + if not isinstance(tensor, torch.Tensor): + raise TypeError("CheckpointableTensor must also be a torch.Tensor") + local_tensor = tensor + if not local_offset: + return local_tensor + return local_tensor[ + tuple( + slice(offset, offset + size) + for offset, size in zip(local_offset, local_size, strict=True) + ) + ] + + +def _validate_checkpointable_tensor_metadata(tensor: CheckpointableTensor) -> None: + num_shards = len(tensor.global_offsets) + if len(tensor.local_offsets) != num_shards: + raise ValueError("global_offsets and local_offsets must have the same length") + if len(tensor.local_sizes) != num_shards: + raise ValueError("global_offsets and local_sizes must have the same length") + + global_shape = tensor.global_shape + if not isinstance(tensor, torch.Tensor): + raise TypeError("CheckpointableTensor must also be a torch.Tensor") + tensor_shape = tuple(tensor.size()) + for idx, (global_offset, local_offset, local_size) in enumerate( + zip( + tensor.global_offsets, + tensor.local_offsets, + tensor.local_sizes, + strict=True, + ) + ): + if len(global_offset) != len(global_shape): + raise ValueError( + f"global_offsets[{idx}] must have {len(global_shape)} dimensions" + ) + if len(local_offset) != len(tensor_shape): + raise ValueError( + f"local_offsets[{idx}] must have {len(tensor_shape)} dimensions" + ) + if len(local_size) != len(global_shape): + raise ValueError( + f"local_sizes[{idx}] must have {len(global_shape)} dimensions" + ) + if len(local_size) != len(tensor_shape): + raise ValueError( + f"local_sizes[{idx}] must have {len(tensor_shape)} local dimensions" + ) + + for dim, (offset, size, global_dim) in enumerate( + zip(global_offset, local_size, global_shape, strict=True) + ): + if offset < 0 or size < 0 or offset + size > global_dim: + raise ValueError( + f"global shard {idx} dimension {dim} is outside global_shape" + ) + + for dim, (offset, size, local_dim) in enumerate( + zip(local_offset, local_size, tensor_shape, strict=True) + ): + if offset < 0 or size < 0 or offset + size > local_dim: + raise ValueError( + f"local shard {idx} dimension {dim} is outside tensor shape" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/quantized_hf_storage.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/quantized_hf_storage.py new file mode 100644 index 0000000000000000000000000000000000000000..29c5e9b158cc3438efc84ace6904f4d5600dab47 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/quantized_hf_storage.py @@ -0,0 +1,506 @@ +# mypy: allow-untyped-defs +import json +import logging +import math +from pathlib import Path +from typing import Any + +import torch +from torch.distributed.checkpoint._hf_utils import _metadata_fn +from torch.distributed.checkpoint.metadata import TensorStorageMetadata +from torch.distributed.checkpoint.planner import LoadPlanner, ReadItem + +from .hf_storage import HuggingFaceStorageReader + + +logger: logging.Logger = logging.getLogger(__name__) + +__all__ = ["QuantizedHuggingFaceStorageReader"] + + +class QuantizedHuggingFaceStorageReader(HuggingFaceStorageReader): + """ + Extension of HuggingFaceStorageReader that handles quantized tensors. + Checkpoint should have the full tensor in a SafeTensor file. The quantized + tensor should not be sharded across multiple files. + + This reader handles the dequantization of tensors during the read process, + converting them from quantized blocks to full dequantized tensors before + copying to the target tensor. + """ + + def __init__( + self, + path: str, + thread_count: int = 1, + target_dtype: torch.dtype = torch.float32, + block_size: int = 128, + ): + """ + Initialize the HuggingFace storage reader to load quantized checkpoints + + Args: + path: directory where the checkpoint will be read from. + thread_count: Number of threads to use to read distributed checkpoint. Defaults to 1. + target_dtype: Target dtype for dequantized tensor. Defaults to torch.float32. + block_size: Fixed block size for dequantization. Defaults to 128. + """ + super().__init__(path=path, thread_count=thread_count) + + self.target_dtype: torch.dtype = target_dtype + self.block_size: int = block_size + self._weight_scale_mapping: dict[str, str] = {} + # Track which file contains each tensor + self._weight_map: dict[str, str] = {} + # Cache for full tensor shapes (fqn -> shape) + self._tensor_full_shapes: dict[str, torch.Size] = {} + + def read_metadata(self) -> Any: + metadata = super().read_metadata() + + # Load quantization metadata first. + self._load_quantization_metadata() + + # Build a cache of FQN -> full tensor shape, correcting for quantized tensors. + for fqn, tensor_metadata in metadata.state_dict_metadata.items(): + # Only process TensorStorageMetadata which has size attribute. + if isinstance(tensor_metadata, TensorStorageMetadata): + # Check if this is a MXFP4 quantized tensor that needs shape correction. + if fqn.endswith("_blocks"): + # Save the quantized tensor shapes for lookup when dequantization. + self._tensor_full_shapes[fqn + "_quantized"] = tensor_metadata.size + *prefix_shape, G, B = tensor_metadata.size + dequantized_size = torch.Size([*prefix_shape, G * B * 2]) + + # Update the metadata with the size after dequantization. + # Metadata used by planner to slice state dict. + tensor_metadata.size = dequantized_size + self._tensor_full_shapes[fqn] = dequantized_size + else: + self._tensor_full_shapes[fqn] = tensor_metadata.size + + return metadata + + def _load_quantization_metadata(self): + """Load quantization metadata from the checkpoint.""" + checkpoint_path = Path(self.path) + # Load weight mapping from index file + index_file = checkpoint_path / _metadata_fn + + with open(index_file) as f: + index_data = json.load(f) + weight_map = index_data.get("weight_map", {}) + self._build_weight_scale_mapping(weight_map) + + def _build_weight_scale_mapping(self, weight_map: dict[str, str]): + """Analyze and build weight-scale tensor pairs from weight mapping.""" + # Store the complete weight map for file location lookups. + self._weight_map = weight_map + + for tensor_name in weight_map: + if tensor_name.endswith(".weight_scale_inv"): + weight_name = tensor_name.replace(".weight_scale_inv", ".weight") + if weight_name in weight_map: + self._weight_scale_mapping[weight_name] = tensor_name + # Handle MXFP4 format: _blocks and _scales. + elif tensor_name.endswith("_scales"): + blocks_name = tensor_name.replace("_scales", "_blocks") + if blocks_name in weight_map: + self._weight_scale_mapping[blocks_name] = tensor_name + + def _process_read_request( + self, f: Any, req: ReadItem, planner: LoadPlanner + ) -> None: + """Override the Helper function that processes a single read request.""" + tensor_fqn = req.storage_index.fqn + + # Check if this is a quantized tensor that needs dequantization + if self._is_tensor_quantized(tensor_fqn): + tensor = self._read_quantized_tensor_with_block_alignment(req, f) + else: + # Standard tensor reading + slices = tuple( + slice(offset, offset + length) + for offset, length in zip(req.storage_offsets, req.lengths) + ) + tensor = f.get_slice(tensor_fqn)[slices] + + target_tensor = planner.resolve_tensor(req).detach() + + if target_tensor.size() != tensor.size(): + raise AssertionError( + f"req {req.storage_index} mismatch sizes {target_tensor.size()} vs {tensor.size()}" + ) + + target_tensor.copy_(tensor) + planner.commit_tensor(req, target_tensor) + + def _get_slice_to_block_mapping( + self, req: ReadItem + ) -> tuple[tuple[int, int], tuple[int, int], slice, slice]: + """ + Calculate which blocks correspond to the requested slice. + + Args: + req: Read request containing tensor info and required slices + + Returns: + Tuple of (row_block_range, col_block_range, row_slice, col_slice) + """ + # Get the slice information + row_slice = slice( + req.storage_offsets[0], req.storage_offsets[0] + req.lengths[0] + ) + col_slice = slice( + req.storage_offsets[1], req.storage_offsets[1] + req.lengths[1] + ) + + # Calculate which blocks this slice spans + row_start_block = row_slice.start // self.block_size + row_end_block = (row_slice.stop - 1) // self.block_size + 1 # Inclusive end + + col_start_block = col_slice.start // self.block_size + col_end_block = (col_slice.stop - 1) // self.block_size + 1 # Inclusive end + + return ( + (row_start_block, row_end_block), + (col_start_block, col_end_block), + row_slice, + col_slice, + ) + + def _dequantize_tensor_mxfp4( + self, + blocks: torch.Tensor, + scales: torch.Tensor, + req: ReadItem, + group_start: int, + offset_in_first_group: int, + ) -> torch.Tensor: + """ + Dequantize a 4D tensor using MXFP4 format. + Adapted from openai's implementation: + https://github.com/openai/gpt-oss/blob/8890e95919f975a490fc0ba09ffb10890ec7319d/gpt_oss/torch/weights.py#L68 + + Args: + blocks: Sliced quantized weight tensor of shape [a_slice, b_slice, groups_slice, B] in uint8 + scales: FULL scale tensor of shape [a, b, c] in uint8 (will be converted to exponents) + req: Read request containing slice information + group_start: The starting group index in the checkpoint + offset_in_first_group: Offset in values within the first group + + Returns: + Dequantized tensor matching the requested shape + """ + # FP4 lookup table + FP4_VALUES = [ + +0.0, + +0.5, + +1.0, + +1.5, + +2.0, + +3.0, + +4.0, + +6.0, + -0.0, + -0.5, + -1.0, + -1.5, + -2.0, + -3.0, + -4.0, + -6.0, + ] + + # blocks: [a_slice, b_slice, groups_slice, B] uint8. + # Read slightly more groups than needed, and slice at the end. + + # Slice the scales to match the blocks dimensions. + # [a_full, b_full, c_full] -> [a_slice, b_slice, groups_slice] + dim0_start = req.storage_offsets[0] + dim0_end = dim0_start + req.lengths[0] + dim1_start = req.storage_offsets[1] + dim1_end = dim1_start + req.lengths[1] + num_groups = blocks.shape[2] + scales = scales[ + dim0_start:dim0_end, + dim1_start:dim1_end, + group_start : group_start + num_groups, + ] + + scales = scales.to(torch.int32) - 127 + + if blocks.shape[:-1] != scales.shape: + raise AssertionError(f"{blocks.shape=} does not match {scales.shape=}") + + lut = torch.tensor(FP4_VALUES, dtype=self.target_dtype, device=blocks.device) + + *prefix_shape, G, B = blocks.shape + rows_total = math.prod(prefix_shape) * G + + blocks = blocks.reshape(rows_total, B) + scales = scales.reshape(rows_total, 1) + + out = torch.empty( + rows_total, B * 2, dtype=self.target_dtype, device=blocks.device + ) + + rows_per_chunk = 16384 * 512 + + for r0 in range(0, rows_total, rows_per_chunk): + r1 = min(r0 + rows_per_chunk, rows_total) + + blk = blocks[r0:r1] + exp = scales[r0:r1] + + # nibble indices -> int64 + idx_lo = (blk & 0x0F).to(torch.long) + idx_hi = (blk >> 4).to(torch.long) + + sub = out[r0:r1] + sub[:, 0::2] = lut[idx_lo] + sub[:, 1::2] = lut[idx_hi] + + torch.ldexp(sub, exp, out=sub) + + del idx_lo, idx_hi, blk, exp + + result = out.reshape(*prefix_shape, G, B * 2).view(*prefix_shape, G * B * 2) + + # Slice the last dimension to match the requested range. + if offset_in_first_group > 0 or result.shape[-1] > req.lengths[2]: + end_offset = offset_in_first_group + req.lengths[2] + result = result[..., offset_in_first_group:end_offset] + + return result + + def _dequantize_tensor( + self, + weight: torch.Tensor, + scale_inv: torch.Tensor, + full_tensor_shape: torch.Size, + slice_info: tuple[tuple[int, int], tuple[int, int], slice, slice], + ) -> torch.Tensor: + """ + Dequantize a sliced tensor using the appropriate portion of the scale tensor. + + Args: + weight: Sliced quantized weight tensor + scale_inv: Full scale inverse tensor for dequantization + full_tensor_shape: Shape of the original full tensor + slice_info: Block mapping information from _get_slice_to_block_mapping + + Returns: + Dequantized tensor + """ + (row_block_range, col_block_range, row_slice, col_slice) = slice_info + + # Convert to float32 for computation + # Certain quantized dtypes like Float8_e4m3fn + # don't support multiplication on CPU yet in PyTorch. + upcasted_weight = weight.to(torch.float32) + + # Create output tensor in target dtype + dequantized = weight.detach().to(dtype=self.target_dtype, copy=True) + + # Get the actual slice boundaries + row_start_global = row_slice.start + row_end_global = row_slice.stop + col_start_global = col_slice.start + col_end_global = col_slice.stop + + # Apply scaling factors to each block that intersects with our slice + for block_i in range(row_block_range[0], row_block_range[1]): + for block_j in range(col_block_range[0], col_block_range[1]): + # Calculate the block boundaries in global coordinates + block_row_start_global = block_i * self.block_size + block_row_end_global = min( + block_row_start_global + self.block_size, full_tensor_shape[0] + ) + block_col_start_global = block_j * self.block_size + block_col_end_global = min( + block_col_start_global + self.block_size, full_tensor_shape[1] + ) + + # Find the intersection of the block with our slice + intersect_row_start = max(block_row_start_global, row_start_global) + intersect_row_end = min(block_row_end_global, row_end_global) + intersect_col_start = max(block_col_start_global, col_start_global) + intersect_col_end = min(block_col_end_global, col_end_global) + + # Skip if no intersection + if ( + intersect_row_start >= intersect_row_end + or intersect_col_start >= intersect_col_end + ): + continue + + # Convert global coordinates to local coordinates in the sliced tensor + local_row_start = intersect_row_start - row_start_global + local_row_end = intersect_row_end - row_start_global + local_col_start = intersect_col_start - col_start_global + local_col_end = intersect_col_end - col_start_global + + # Get the block from the sliced tensor + block = upcasted_weight[ + local_row_start:local_row_end, local_col_start:local_col_end + ] + + # Apply the scale factor + scale = scale_inv[block_i, block_j] + block = block * scale + + # Convert block to target dtype and store + block_converted = block.to(dtype=self.target_dtype) + dequantized[ + local_row_start:local_row_end, local_col_start:local_col_end + ] = block_converted + + return dequantized + + def _is_tensor_quantized(self, tensor_fqn: str) -> bool: + """ + Check if a tensor is a quantized. + + Args: + tensor_fqn: Fully qualified name of the tensor + + Returns: + True if tensor is quantized and has a corresponding scale tensor, + False otherwise + """ + # Skip scale tensors themselves + if tensor_fqn.endswith((".weight_scale_inv", "_scales")): + return False + + # Check if this weight tensor has a corresponding scale tensor + if tensor_fqn not in self._weight_scale_mapping: + return False + + return True + + def _read_quantized_tensor_with_block_alignment( + self, req: ReadItem, safetensor_file: Any + ) -> torch.Tensor: + """ + Read a quantized tensor with block alignment. + + Args: + req: Read request containing tensor info and required slices + safetensor_file: Open safetensors file handle + + Returns: + Dequantized tensor ready for use + """ + tensor_fqn = req.storage_index.fqn + scale_fqn = self._weight_scale_mapping[tensor_fqn] + + try: + group_start = 0 + offset_in_first_group = 0 + if tensor_fqn.endswith("_blocks"): + # Full tensor is a 4D MXFP4 quantized tensor: [..., G, B]. + # Each group G produces B * 2 dequantized values. + # Checkpoint [..., G, B] -> dequantized [..., G*B*2]. + + # The planner gives 3D requests based on the dequantized shape. + # Need to figure out which groups (dimension 2 in checkpoint) to read. + + # Use the quantized checkpoint shape to get the correct B. + *prefix_shape, B = self._tensor_full_shapes[tensor_fqn + "_quantized"] + values_per_group = B * 2 # Each byte has 2 nibbles (4-bit values). + + # Calculate which groups we need based on the requested range in dim 2. + # Ensure the request is in 3D. + if len(req.storage_offsets) != 3: + raise AssertionError + + # Positions in dequantized space. + dim2_start_deq = req.storage_offsets[2] + dim2_length_deq = req.lengths[2] + dim2_end_deq = dim2_start_deq + dim2_length_deq + + # Convert to group indices. + group_start = dim2_start_deq // values_per_group + group_end = (dim2_end_deq + values_per_group - 1) // values_per_group + + # Read only the necessary groups from checkpoint. + weight_slices_4d = ( + slice( + req.storage_offsets[0], req.storage_offsets[0] + req.lengths[0] + ), + slice( + req.storage_offsets[1], req.storage_offsets[1] + req.lengths[1] + ), + slice(group_start, group_end), + slice(None), # Read all B values for each group. + ) + quantized_tensor = safetensor_file.get_slice(tensor_fqn)[ + weight_slices_4d + ] + + # Also track the offset within the first group + offset_in_first_group = dim2_start_deq - ( + group_start * values_per_group + ) + else: + # 2D quantized tensor, use 2d block partition. + weight_slices = tuple( + slice(offset, offset + length) + for offset, length in zip(req.storage_offsets, req.lengths) + ) + quantized_tensor = safetensor_file.get_slice(tensor_fqn)[weight_slices] + + # Load the corresponding scale inverse tensor (full tensor) + scale_file_name = self._weight_map.get(scale_fqn) + if scale_file_name is None: + raise ValueError(f"Scale tensor {scale_fqn} not found in weight_map") + + # Check if scale tensor is in the same file as the weight tensor + weight_file_name = self._weight_map.get(tensor_fqn) + + if scale_file_name == weight_file_name: + # Scale tensor is in the same file, use current handle + scale_inv = safetensor_file.get_tensor(scale_fqn) + else: + # Scale tensor is in a different file, need to open it + from safetensors import safe_open # type: ignore[import] + + scale_file_path = Path(self.path) / scale_file_name + with safe_open( + scale_file_path, framework="pt", device="cpu" + ) as scale_file: + scale_inv = scale_file.get_tensor(scale_fqn) + + # Get the full tensor shape from our O(1) lookup cache + full_tensor_shape = self._tensor_full_shapes.get(tensor_fqn) + if full_tensor_shape is None: + raise ValueError(f"Could not find full tensor shape for {tensor_fqn}") + + # Determine which dequantization function to use. + if len(full_tensor_shape) == 2: + # 2D block-wise quantization, e.g., used in deepseek v3.1 + slice_info = self._get_slice_to_block_mapping(req) + dequantized_tensor = self._dequantize_tensor( + weight=quantized_tensor, + scale_inv=scale_inv, + full_tensor_shape=full_tensor_shape, + slice_info=slice_info, + ) + elif tensor_fqn.endswith("_blocks"): + # 4D with blocks along dimension 2, used in MXFP4, e.g. gpt-oss + dequantized_tensor = self._dequantize_tensor_mxfp4( + blocks=quantized_tensor, + scales=scale_inv, + req=req, + group_start=group_start, + offset_in_first_group=offset_in_first_group, + ) + else: + raise ValueError("Unsupported quantization types") + + return dequantized_tensor + + except Exception as e: + logger.error("Failed to read the quantized tensor!!") + raise e diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/resharding.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/resharding.py new file mode 100644 index 0000000000000000000000000000000000000000..b49aad1928d9491d0032f3628d30455be822d660 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/resharding.py @@ -0,0 +1,69 @@ +from torch.distributed.checkpoint.metadata import ChunkStorageMetadata + + +__all__: list[str] = [] + + +def _check_shard_metadata_pair_overlap( + shard1: ChunkStorageMetadata, shard2: ChunkStorageMetadata +) -> bool: + """Check if two shards overlap.""" + # For each dim of each shard, check if one shard resides on the other + # end of second shard with respect to that dim. As an example for a 2D + # shard, we would check if one shard is above or on the left of the + # other shard. + ndims = len(shard1.offsets) + for i in range(ndims): + if shard1.offsets[i] >= shard2.offsets[i] + shard2.sizes[i]: + return False + if shard2.offsets[i] >= shard1.offsets[i] + shard1.sizes[i]: + return False + + return True + + +def _shards_get_overlap_region_wrt_saved_tensor( + saved_shard: ChunkStorageMetadata, current_shard: ChunkStorageMetadata +) -> list[tuple[int, int, int, int]]: + """ + Return the overlapping region between saved_shard and current_shard. + + The returned list has the same number of elements as the tensor's dimension. + For each element, we produce a tuple with the following contents: + (dimension, `saved_shard` offset, `current_shard` offset, length) + + Offsets are relative to each shard. + """ + narrows = [] + for dim, ( + saved_shard_offset, + current_shard_offset, + saved_shard_size, + current_shard_size, + ) in enumerate( + zip( + saved_shard.offsets, + current_shard.offsets, + saved_shard.sizes, + current_shard.sizes, + ) + ): + min_range_end = min( + saved_shard_offset + saved_shard_size, + current_shard_offset + current_shard_size, + ) + + length = min_range_end - max(current_shard_offset, saved_shard_offset) + + if saved_shard_offset > current_shard_offset: + offset_for_saved_tensor = 0 + offset_for_current_tensor = saved_shard_offset - current_shard_offset + else: + offset_for_saved_tensor = current_shard_offset - saved_shard_offset + offset_for_current_tensor = 0 + + narrows.append( + (dim, offset_for_saved_tensor, offset_for_current_tensor, length) + ) + + return narrows diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/staging.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/staging.py new file mode 100644 index 0000000000000000000000000000000000000000..86df4fdfb787a0ec79956cb06d7db6cf1c917c8c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/staging.py @@ -0,0 +1,468 @@ +import os +import tempfile +from concurrent.futures import Future, ThreadPoolExecutor +from contextlib import nullcontext +from dataclasses import dataclass +from datetime import timedelta +from typing import Any, cast +from typing_extensions import deprecated, Protocol, runtime_checkable + +import torch +import torch.distributed as dist +from torch.distributed import ProcessGroup +from torch.distributed._state_dict_utils import _copy_state_dict, _create_cpu_state_dict +from torch.distributed.checkpoint._pg_transport import PGTransport +from torch.distributed.checkpoint._state_dict_stager import StateDictStager +from torch.distributed.checkpoint.metadata import STATE_DICT_TYPE + + +__all__ = ["AsyncStager", "BlockingAsyncStager", "DefaultStager", "StagingOptions"] + +""" +Experimental staging module for PyTorch Distributed Checkpointing. +This module provides advanced staging capabilities for checkpoints including: +- Asynchronous staging using ThreadPoolExecutor +- Pinned memory allocation for faster CPU-GPU transfers +- Shared memory support for multi-process scenarios +- Non-blocking CUDA operations with stream synchronization +- Caching of frequently used storages for efficient memory management +- Automatic resource cleanup and memory management +Classes: + AsyncStager: Protocol defining the staging interface + StagingOptions: Configuration dataclass for staging behavior + DefaultStager: Default implementation with comprehensive staging features + BlockingAsyncStager: Implementation of AsyncStager which stages the state_dict + on CPU RAM and blocks until the copy is complete. Please use DefaultStager instead. +""" + + +@runtime_checkable +class AsyncStager(Protocol): + """ + This protocol is meant to provide customization and extensibility for dcp.async_save, allowing users + to customize how data is staged previous to executing the usual dcp.save path in parallel. + The expected order of operations (concretely defined in `torch.distributed.state_dict_saver.async_save`) + is the following: + + 1. AsyncStager.stage_data(state_dict): + This call gives the AsyncStager the opportunity to 'stage' + the state_dict. The expectation and purpose of staging in this context is to create a "training-safe" + representation of the state dict, meaning that any updates to module data after staging is complete + should not be reflected in the state dict returned from this method. For example, in the default + case a copy of the entire state dict is created on CPU RAM and returned here, allowing users + to continue training without risking changes to data which is being serialized. + + 2. dcp.save is called on the state_dict returned from stage in parallel. This call is responsible + for serializing the state_dict and writing it to storage. + + 3. If AsyncStager.should_synchronize_after_execute is True, this method will be called immediately after + the serialization thread starts and before returning from dcp.async_save. If this is set to False, + the assumption is the user has defined a custom synchronization point for the purpose of further + optimizing save latency in the training loop (for example, by overlapping staging with the + forward/backward pass), and it is the respondsibility of the user to call `AsyncStager.synchronize_staging` + at the appropriate time. + + """ + + # default to True since the common case is to stage synchronously + _synchronize_after_execute: bool = True + + @property + def should_synchronize_after_execute(self) -> bool: + """ + Whether to synchronize after executing the stage. + """ + return self._synchronize_after_execute + + def stage( + self, state_dict: STATE_DICT_TYPE + ) -> Future[STATE_DICT_TYPE] | STATE_DICT_TYPE: + """ + Returns a "staged" copy of `state_dict`. The expectation of the staged copy is that it is + inoculated from any updates incurred after the stage call is complete. + """ + raise NotImplementedError( + f"{self.__class__.__name__} must implement stage method" + ) + + @deprecated( + "`synchronize_staging` is deprecated and will be removed in future versions." + "Please use staging_future from AsyncSaveResponse instead.", + category=FutureWarning, + ) + def synchronize_staging(self) -> None: + """ + In the case `stage` is async in some way, this method should be called to ensure staging + is complete and it is safe to begin modifying the original `state_dict` + """ + + def close(self) -> None: + """ + Clean up all resources used by the stager. + """ + + +@dataclass +class StagingOptions: + """ + Configuration options for checkpoint staging behavior. + + Attributes: + use_pinned_memory (bool): Enable pinned memory allocation for faster + CPU-GPU transfers. Requires CUDA to be available. Default: True + use_shared_memory (bool): Enable shared memory for multi-process + scenarios. Useful when multiple processes need access to the + same staged data. Default: True + use_async_staging (bool): Enable asynchronous staging using a + background thread pool. Allows overlapping computation with + staging operations. Requires CUDA. Default: True + use_non_blocking_copy (bool): Use non-blocking device memory + copies with stream synchronization. Improves performance by + allowing CPU work to continue during GPU transfers. Default: True + + Note: + CUDA-dependent features will raise exception if CUDA is not available. + """ + + use_pinned_memory: bool = True + use_shared_memory: bool = True + use_async_staging: bool = True + use_non_blocking_copy: bool = True + + +class DefaultStager(AsyncStager): + """ + DefaultStager provides a full-featured staging implementation that combines + multiple optimization techniques for efficient checkpoint preparation. + + The staging process works as follows: + 1. State dictionary is submitted for staging (sync or async) + 2. Tensors are copied from GPU to optimized CPU storage + 3. CUDA operations are synchronized if non-blocking copies are used + 4. Staged state dictionary is returned or made available via Future + + Usage Patterns: + # Synchronous staging + stager = DefaultStager(StagingOptions(use_async_staging=False)) + staged_dict = stager.stage(state_dict) + stager.close() + + # Asynchronous staging + stager = DefaultStager(StagingOptions(use_async_staging=True)) + future = stager.stage(state_dict) + # ... do other work ... + staged_dict = future.result() + stager.close() + + Performance Considerations: + - Async staging provides best performance when model computation + can overlap with staging operations + - Pinned memory improves CPU-GPU transfer speeds but uses more memory + - Shared memory allows efficient IPC to checkpoint process + - Non-blocking copies reduce GPU idle time during memory transfers + + Thread Safety: + DefaultStager is not thread-safe. Each thread should use its own + instance, or external synchronization should be provided. + """ + + def __init__( + self, + config: StagingOptions = StagingOptions(), + ): + self._config = config + self._state_dict_stager = StateDictStager( + pin_memory=config.use_pinned_memory, share_memory=config.use_shared_memory + ) + self._staging_executor = None + self._staging_stream = None + if self._config.use_async_staging: + self._staging_executor = ThreadPoolExecutor(max_workers=1) + if torch.accelerator.is_available(): + # Note: stream needs to be initialized on the main thread after default cuda + # stream is setup/used to avoid the risk of accidentally reusing the main + # compute stream or in other cases kernels actually launching from the + # main thread. + self._staging_stream = torch.Stream() + + if self._config.use_non_blocking_copy: + if not torch.accelerator.is_available(): + raise AssertionError( + "Non-blocking copy requires that the current accelerator is available." + ) + + self._staging_future: Future[STATE_DICT_TYPE] | None = None + + def stage( + self, + state_dict: STATE_DICT_TYPE, + **kwargs: Any, + ) -> STATE_DICT_TYPE | Future[STATE_DICT_TYPE]: + """ + This function is responsible for staging staging the state_dict. + See class docstring for more details on staging. + If use_async_staging is True, it will return a Future object that will be + fulfilled when staging is complete. + If use_async_staging is False, it will return the fully staged state_dict. + + Args: + state_dict (STATE_DICT_TYPE): The state_dict to be staged. + """ + if self._config.use_async_staging: + if self._staging_executor is None: + raise AssertionError( + "staging_executor should not be None for async staging" + ) + self._staging_future = self._staging_executor.submit( + self._stage, + state_dict, + **kwargs, + ) + return self._staging_future + else: + return self._stage(state_dict, **kwargs) + + def _stage(self, state_dict: STATE_DICT_TYPE, **kwargs: Any) -> STATE_DICT_TYPE: + if self._config.use_non_blocking_copy: + if not (self._staging_stream or not self._config.use_async_staging): + raise AssertionError( + "Non-blocking copy in a background thread for async staging needs staging_stream to be initialized." + ) + with ( + self._staging_stream + if self._staging_stream is not None + else nullcontext() + ): + state_dict = self._state_dict_stager.stage( + state_dict, non_blocking=self._config.use_non_blocking_copy + ) + # waits for the enqued copy operations to finish. + self._staging_stream.synchronize() if self._staging_stream else torch.accelerator.synchronize() + else: + state_dict = self._state_dict_stager.stage(state_dict, non_blocking=False) + + return state_dict + + def close(self) -> None: + """ + Clean up all resources used by the DefaultStager. Shuts down the ThreadPoolExecutor + used for async staging operations and cleans up the underlying StateDictStager's + cached storages. Should be called when the stager is no longer needed to prevent + resource leaks, especially in long-running applications. After calling close(), + the stager should not be used for further staging operations. + + Example Usage: + stager = DefaultStager(StagingOptions(use_async_staging=True)) + future = stager.stage(state_dict) + result = future.result() + stager.close() # Clean up all resources + """ + if self._staging_executor: + self._staging_executor.shutdown(wait=True) + self._state_dict_stager.close() + + def synchronize_staging(self) -> None: + """ + When use_async_staging is True, this method will wait until staging is complete. + If use_async_staging is False, this method is a no-op. + """ + if self._staging_future is not None: + self._staging_future.result() + + +class BlockingAsyncStager(AsyncStager): + """ + An implementation of AsyncStager which stages the state_dict on CPU RAM and blocks until the copy is complete. + This implementation also provides an option to optimize stage latency using pinned memory. + + N.B. synchronize_staging is a no-op in this case. + + + """ + + # default to True since the common case is to stage synchronously + _synchronize_after_execute: bool = False + + def __init__( + self, + cache_staged_state_dict: bool = False, + type_check: bool = False, + ): + """ + Initializes the BlockingAsyncStager. + + Args: + cache_staged_state_dict: Whether to cache the staged state_dict. This option decreases staging latency + at the cost of increased memory usage. Additionally, if this parameter is set to True, it's the expectation + that the stager is maintained and reused for multiple dcp.async_save calls. Default to False. + type_check: Whether to perform a type check during cpu_offload. Defaults to False. + + """ + self.cache_staged_state_dict = cache_staged_state_dict + self.type_check = type_check + self.state_dict_cache: STATE_DICT_TYPE | None = None + + def stage(self, state_dict: STATE_DICT_TYPE) -> STATE_DICT_TYPE: + """ + Returns a copy of `state_dict` on the CPU. + """ + + if not self.cache_staged_state_dict: + staged_state_dict = _create_cpu_state_dict(state_dict) + _copy_state_dict(state_dict, staged_state_dict, type_check=self.type_check) + return staged_state_dict + + if self.state_dict_cache is None: + self.state_dict_cache = _create_cpu_state_dict(state_dict, pin_memory=True) + return _copy_state_dict(state_dict, self.state_dict_cache) + + def synchronize_staging(self) -> None: + """ + No-op function, since staging is blocking. + """ + + def close(self) -> None: + pass + + +class _ReplicationStager(AsyncStager): + """ + An AsyncStager implementation that replicates state_dict across training ranks + using PGTransport. + + Args: + pg: ProcessGroup for distributed communication + timeout: Timeout for communication operations + device: Device to use for tensor operations + storage_dir: Directory to store persisted state_dicts + + Warning: This is experimental and subject to change. + """ + + _synchronize_after_execute: bool = False + + def __init__( + self, + pg: ProcessGroup, + timeout: timedelta = timedelta(minutes=30), + device: torch.device = torch.device("cpu"), + storage_dir: str | None = None, + ): + self._pg = pg + self._timeout = timeout + self._device = device + self._transport = PGTransport(pg, timeout, device, None) + + # Set up storage directory for persisting exchanged state_dicts + if storage_dir is None: + self._storage_dir = tempfile.mkdtemp(prefix="replication_stager_") + else: + self._storage_dir = storage_dir + os.makedirs(self._storage_dir, exist_ok=True) + + def stage( + self, state_dict: STATE_DICT_TYPE + ) -> Future[STATE_DICT_TYPE] | STATE_DICT_TYPE: + """ + Stage the state_dict by replicating it across ranks. Returns a state_dict representing + the received replica. + + Perform the actual replication logic. Creates bidirectional pairs where each rank exchanges + state_dict with its partner at (rank + world_size//2) % world_size. + Uses simple rank-based ordering to prevent deadlocks. + + Assumes world_size is always even. + """ + if not dist.is_initialized(): + return state_dict + + world_size = dist.get_world_size() + + current_rank = dist.get_rank() + + # Calculate partner rank using half-world offset + # creates bidirectional pairs for replication. + offset = world_size // 2 + partner_rank = (current_rank + offset) % world_size + + # Use simple rank-based ordering to prevent deadlocks. + # Lower-numbered rank sends first, higher-numbered rank receives first. + if current_rank < partner_rank: + # Send first, then receive + self._transport.send_checkpoint([partner_rank], state_dict) + received_state_dict = self._transport.recv_checkpoint(partner_rank) + else: + # Receive first, then send + received_state_dict = self._transport.recv_checkpoint(partner_rank) + self._transport.send_checkpoint([partner_rank], state_dict) + + # Persist the received state_dict for future discoverability + received_state_dict = cast(STATE_DICT_TYPE, received_state_dict) + self._persist_state_dict(received_state_dict, current_rank, partner_rank) + + return received_state_dict + + def _persist_state_dict( + self, state_dict: STATE_DICT_TYPE, current_rank: int, partner_rank: int + ) -> None: + """ + Persist the received state_dict to disk for future discoverability. + Only keeps one replica per rank, overwriting any previous replica. + Uses atomic write pattern (temp file + rename). + + Args: + state_dict: The state_dict received from partner rank + current_rank: Current rank that received the state_dict + partner_rank: Rank that sent the state_dict + """ + final_path = self._get_persisted_path(current_rank, partner_rank) + temp_path = final_path + ".tmp" + + try: + # Ensure parent directory exists and is writable + os.makedirs(os.path.dirname(final_path), exist_ok=True) + + # Write to temporary file with explicit flushing + with open(temp_path, "wb") as f: + torch.save(state_dict, f) + # Flush application buffers to OS buffers + f.flush() + # Force OS buffers to disk for durability + os.fsync(f.fileno()) + + # Atomic rename to final location + os.rename(temp_path, final_path) + except Exception as e: + # Clean up temp file if it exists + try: + if os.path.exists(temp_path): + os.remove(temp_path) + except Exception: + pass # Ignore cleanup errors + # Re-raise the original exception with more context + raise RuntimeError( + f"Failed to persist state_dict from rank {partner_rank} to rank {current_rank}: {e}" + ) from e + + def _get_persisted_path(self, current_rank: int, partner_rank: int) -> str: + """ + Get the file path where a state_dict would be persisted. + + Args: + current_rank: Current rank + + Returns: + File path for the persisted state_dict + """ + filename = f"rank_{current_rank}_replica_partner_{partner_rank}.pt" + return os.path.join(self._storage_dir, filename) + + def synchronize_staging(self) -> None: + """ + No-op function, since staging is blocking. + """ + + def close(self) -> None: + """ + Clean up resources. Persisted files are intentionally left for future discovery. + """ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict.py new file mode 100644 index 0000000000000000000000000000000000000000..4341900b2de9f7b18ed78318e805f1c46f0c9d2b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict.py @@ -0,0 +1,1662 @@ +# mypy: allow-untyped-defs +import contextlib +import functools +import gc +import warnings +from collections.abc import Callable, Generator, Iterable +from dataclasses import asdict, dataclass, field +from itertools import chain +from typing import Any, cast, no_type_check + +import torch +import torch.distributed as dist +import torch.nn as nn +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed._state_dict_utils import ( + _broadcast_state_dict, + _distribute_state_dict, + _flatten_state_dict, + _gather_state_dict, + _offload_state_dict_to_cpu, + _unflatten_state_dict, +) +from torch.distributed.algorithms._checkpoint.checkpoint_wrapper import ( + _CHECKPOINT_PREFIX, +) +from torch.distributed.fsdp import ( + FullOptimStateDictConfig, + FullStateDictConfig, + FullyShardedDataParallel as FSDP, + OptimStateDictConfig, + ShardedOptimStateDictConfig, + ShardedStateDictConfig, + StateDictConfig, + StateDictType, +) +from torch.distributed.fsdp._common_utils import ( + _get_module_fsdp_state_if_fully_sharded_module, + FSDP_WRAPPED_MODULE, +) +from torch.distributed.fsdp._init_utils import HYBRID_SHARDING_STRATEGIES +from torch.distributed.tensor import DTensor +from torch.nn.modules.module import _IncompatibleKeys +from torch.nn.parallel import DistributedDataParallel as DDP +from torch.utils._pytree import tree_map_only + + +__all__ = [ + "FQNS_T", + "PrimitiveType", + "ValueType", + "DictValueType", + "ListDictValueType", + "OptimizerStateType", + "StateDictOptions", + "get_model_state_dict", + "get_optimizer_state_dict", + "get_state_dict", + "set_model_state_dict", + "set_optimizer_state_dict", + "set_state_dict", +] + + +_FLAT_PARAM = "_flat_param" +_PG = "param_groups" +_PARAMS = "params" +_STATE = "state" + +FQNS_T = set[str] +PrimitiveType = DTensor | ShardedTensor | torch.Tensor | int | float | str +ValueType = ( + PrimitiveType | list[PrimitiveType] | tuple[PrimitiveType] | dict[str, "ValueType"] +) +DictValueType = dict[str, ValueType] +ListDictValueType = list[DictValueType] +OptimizerStateType = dict[str, DictValueType | ListDictValueType] + + +_patched_state_dict: set[Callable] = set() + + +@contextlib.contextmanager +def _gc_context(): + is_enabled = gc.isenabled() + gc.disable() + try: + yield + finally: + if is_enabled: + gc.enable() + + +@dataclass +class StateDictOptions: + """ + This dataclass specifies how get_state_dict/set_state_dict will work. + + - ``full_state_dict``: if this is set to True, all the tensors in the + returned state_dict will be gathered. No ShardedTensor and DTensor + will be in the returned state_dict. + + - ``cpu_offload``: offload all the tensors to cpu. To prevent CPU OOM, if + ``full_state_dict`` is also true, then only the rank0 will get the + state_dict and all other ranks will get empty state_dict. + + - ``ignore_frozen_params``: if the value is True, the returned state_dict + won't contain any frozen parameters -- the ``requires_grad`` is False. + The default value is False. + + - ``keep_submodule_prefixes`` (deprecated): when ``submodules`` is not None, this option + indicates whether to keep the submodule prefixes from the state_dict keys. + or example, if the submodule is ``module.pretrain`` and the full FQN of + the parameter is ``pretrain.layer1.weight`` of the param. When this option + is True, the parameter's key in the returned state_dict will be + ``pretrain.layer1.weight``. If the options is False, the key will be + ``layer1.weight``. + Note that if ``keep_submodule_prefixes`` is False, there may be conflicted + FQNs, hence there should be only one submodule in ``submodules``. + + - ``strict``: the ``strict`` option when ``set_state_dict`` calls + model.load_state_dict(). + + - ``broadcast_from_rank0``: when the option is True, rank0 should receive a + full state_dict and will broadcast the tensors in the state_dict/ + optim_state_dict one by one to other ranks. Other ranks will receive + the tensors and shard according to the local shards in the model and + optimizer. ``full_state_dict`` must be set to True when using this option. + This option currently only supports DTensor, not the legacy ShardedTensor. + """ + + full_state_dict: bool = False + cpu_offload: bool = False + ignore_frozen_params: bool = False + keep_submodule_prefixes: bool = True + strict: bool = True + broadcast_from_rank0: bool = False + flatten_optimizer_state_dict: bool = False + dsd_fqn_modifiers: str = "_fqn_modifiers" + + +@dataclass +class _StateDictInfo(StateDictOptions): + fqn_param_mapping: dict[ + str | torch.Tensor, + FQNS_T | torch.Tensor, + ] = field(default_factory=dict) + shared_params_mapping: dict[ + str | torch.Tensor, + FQNS_T | torch.Tensor, + ] = field(default_factory=dict) + submodule_prefixes: set[str] = field(default_factory=set) + handle_model: bool = True + handle_optim: bool = True + fsdp_context: Callable = contextlib.nullcontext + fsdp_modules: list[nn.Module] = field(default_factory=list) + + +def _get_fqns( + model: nn.Module, + name: str, + dsd_fqn_modifiers: str = "_fqn_modifiers", + skip_ddp_prefix: bool = True, + skip_compiler_prefix: bool = True, +) -> FQNS_T: + """ + This API is used to convert the name of a parameter to the FQNs. For FSDP + without `use_orig_params`, the name of FlatParameter can be mapped to + multiple original parameters. As a result, the return type of this function + is `set[str]`. + + Args: + module (nn.Module): the root model. + name (str): the name + skip_ddp_prefix (bool): whether to skip DDP's `module` prefix + + Returns: + The canonical FQNs based on the model traversal. + """ + + # Remove the checkpoint prefix, if it exists. + name = name.replace(_CHECKPOINT_PREFIX, "") + if "." not in name: + return {name} + + obj_names = name.split(".") + fqn_obj_names = [] + curr_obj = model + for i, curr_obj_name in enumerate(obj_names): + if isinstance(curr_obj, DDP): + if curr_obj_name != "module": + raise AssertionError(f"Expected 'module', got '{curr_obj_name}'") + curr_obj = curr_obj.module + if not skip_ddp_prefix: + fqn_obj_names.append(curr_obj_name) + elif isinstance(curr_obj, FSDP): + if i < len(obj_names) - 1 and obj_names[i + 1] == _FLAT_PARAM: + prefix = ".".join(fqn_obj_names) + flat_param = getattr(curr_obj, _FLAT_PARAM) + if prefix: + prefix = f"{prefix}." + return {f"{prefix}{fqn}" for fqn in flat_param._fqns} + curr_obj = getattr(curr_obj, FSDP_WRAPPED_MODULE) + if curr_obj_name != FSDP_WRAPPED_MODULE: + fqn_obj_names.append(curr_obj_name) + curr_obj = getattr(curr_obj, curr_obj_name) + elif isinstance(curr_obj, torch._dynamo.eval_frame.OptimizedModule): + if curr_obj_name != "_orig_mod": + raise AssertionError(f"Expected '_orig_mod', got '{curr_obj_name}'") + curr_obj = curr_obj._orig_mod + if not skip_compiler_prefix: + fqn_obj_names.append(curr_obj_name) + else: + # In some modules, _fqn_modifiers would not shown in the state_dict keys, + # skip them in the fqn to ensure load stat dict successfully for them. + if hasattr(curr_obj, dsd_fqn_modifiers): + if removed_fqn := getattr(curr_obj, dsd_fqn_modifiers)().get( + curr_obj_name + ): + if hasattr(curr_obj, removed_fqn): + curr_obj = getattr(curr_obj, removed_fqn) + fqn_obj_names.append(curr_obj_name) + if curr_obj_name == nn.modules.module._EXTRA_STATE_KEY_SUFFIX: + if i != len(obj_names) - 1: + raise RuntimeError("Expect `_extra_state` to be the last obj name") + else: + curr_obj = getattr(curr_obj, curr_obj_name) + + return {".".join(fqn_obj_names).replace(_CHECKPOINT_PREFIX, "")} + + +class _EXTRA_STATE: + pass + + +def _iterate_valid_model_state(model, dsd_fqn_modifiers="_fqn_modifiers"): + visited_modules: set[nn.Module] = set() + + def recurse(module: nn.Module, curr_fqn: str) -> Generator: + visited_modules.add(module) + + curr_fqn = f"{curr_fqn}." if curr_fqn else "" + for name, submodule in module.named_children(): + if submodule in visited_modules: + continue + # if user have state_dict_hooks in their model, they can add the state_dict key changes + # at dsd_fqn_modifiers in input to align with the function of state_dict_hook + if ( + hasattr(module, dsd_fqn_modifiers) + and name in getattr(module, dsd_fqn_modifiers)().values() + ): + # skip _fqn_modifiers here thus remove the last `.` added + new_fqn = curr_fqn[:-1] + else: + new_fqn = f"{curr_fqn}{name}" + yield from recurse(submodule, new_fqn) + + for name, obj in chain( + module.named_buffers(recurse=False), module.named_parameters(recurse=False) + ): + if name in module._non_persistent_buffers_set: + continue + new_fqn = f"{curr_fqn}{name}" + yield new_fqn, obj + + if ( + getattr(module.__class__, "get_extra_state", nn.Module.get_extra_state) + != nn.Module.get_extra_state + ): + new_fqn = f"{curr_fqn}{nn.modules.module._EXTRA_STATE_KEY_SUFFIX}" + yield new_fqn, _EXTRA_STATE() + + yield from recurse(model, "") + + +def _verify_options( + model: nn.Module, + optims: tuple[torch.optim.Optimizer, ...], + optim_only: bool, + *, + submodules: set[nn.Module] | None = None, + options: StateDictOptions | None = None, +) -> _StateDictInfo: + """ + Verify the model and options passed by the user and generates _StateDictInfo. + """ + if submodules: + warnings.warn( + "Getting submodules only model/optim state_dict is deprecated and " + "will be removed in 2.5. This feature can be achieved by manually " + "filtering out the state_dict returned from get_state_dict.", + FutureWarning, + stacklevel=2, + ) + if optim_only and not optims: + raise RuntimeError( + "Optimizers are not passed in but optim_only is set to True." + ) + + options = options or StateDictOptions() + + fqn_param_mapping: dict[str | torch.Tensor, set[str] | torch.Tensor] = {} + shared_params_mapping: dict[str | torch.Tensor, set[str] | torch.Tensor] = {} + for name, param in _iterate_valid_model_state(model): + if isinstance(param, _EXTRA_STATE): + continue + + fqns = _get_fqns(model, name) + fqn = fqn_param_mapping.get(param) + if fqn is not None: + cast(set[str], fqn_param_mapping[param]).update(fqns) + shared_params_mapping[param] = fqn_param_mapping[param] + else: + # We need to do copy as _get_fqns is lru_cached + fqn_param_mapping[param] = fqns.copy() + for fqn in fqns: + if not isinstance(param, _EXTRA_STATE): + fqn_param_mapping[fqn] = param + + for param_, fqns_ in list(shared_params_mapping.items()): + for fqn in fqns_: + shared_params_mapping[fqn] = cast(torch.Tensor, param_) + + submodule_prefixes: set[str] = set() + if submodules: + submodules = set(submodules) + for name, module in model.named_modules(): + if module not in submodules: + continue + fqns = _get_fqns(model, name) + if len(fqns) != 1: + raise AssertionError("Submodule FQN should only have 1 instance") + submodule_prefixes.update(f"{fqn}." for fqn in fqns) + + if options.broadcast_from_rank0 and not options.full_state_dict: + raise ValueError( + "full_state_dict must be True when broadcast_from_rank0 is True." + ) + fsdp_modules = FSDP.fsdp_modules(model) + state_dict_config: StateDictConfig + optim_state_dict_config: OptimStateDictConfig + fsdp_context: Callable + if fsdp_modules: + # FSDP API only work if at least one FSDP instance exists. + if options.full_state_dict: + state_dict_config = FullStateDictConfig( + offload_to_cpu=options.cpu_offload, rank0_only=options.cpu_offload + ) + optim_state_dict_config = FullOptimStateDictConfig( + offload_to_cpu=options.cpu_offload, + rank0_only=(options.cpu_offload or options.broadcast_from_rank0), + ) + state_dict_type = StateDictType.FULL_STATE_DICT + else: + state_dict_config = ShardedStateDictConfig( + offload_to_cpu=options.cpu_offload, + ) + optim_state_dict_config = ShardedOptimStateDictConfig( + offload_to_cpu=options.cpu_offload, + ) + state_dict_type = StateDictType.SHARDED_STATE_DICT + + @contextlib.contextmanager + def fsdp_state_dict_type_without_warning( + module, + state_dict_type, + state_dict_config, + optim_state_dict_config, + ): + with warnings.catch_warnings(): + warnings.filterwarnings( + "ignore", message="FSDP.state_dict_type", category=FutureWarning + ) + with FSDP.state_dict_type( + module=module, + state_dict_type=state_dict_type, + state_dict_config=state_dict_config, + optim_state_dict_config=optim_state_dict_config, + ): + yield + + fsdp_context = functools.partial( + fsdp_state_dict_type_without_warning, + module=model, + state_dict_type=state_dict_type, + state_dict_config=state_dict_config, + optim_state_dict_config=optim_state_dict_config, + ) + else: + fsdp_context = contextlib.nullcontext + + return _StateDictInfo( + **asdict(options), + fqn_param_mapping=fqn_param_mapping, + shared_params_mapping=shared_params_mapping, + submodule_prefixes=submodule_prefixes, + fsdp_context=fsdp_context, + fsdp_modules=cast(list[nn.Module], fsdp_modules), + handle_model=not optim_only, + handle_optim=(len(optims) > 0), + ) + + +def _verify_state_dict( + model_state_dict: dict[str, ValueType], + optim_state_dict: OptimizerStateType, + info: _StateDictInfo, +) -> None: + for module in info.fsdp_modules: + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state is None: + raise AssertionError("Expected a fsdp_state with a fsdp module.") + + # Verify if the model_state_dict and optim_state_dict are valid. This API + # should give the users an explicit error message to debug or report. + if ( + info.handle_model + and not model_state_dict + and not info.submodule_prefixes + and not info.ignore_frozen_params + and not (info.cpu_offload and info.full_state_dict) + and info.strict + and not info.broadcast_from_rank0 + ): + raise RuntimeError( + "The option indicates that model state_dict is required to save " + "or load, but model state_dict is empty." + f"rank = {dist.get_rank()=}." + ) + + if info.handle_optim: + if ( + not optim_state_dict + and not (info.cpu_offload and info.full_state_dict) + and (not info.broadcast_from_rank0) + ): + raise RuntimeError( + "The option indicates that model state_dict is required to save, " + f"or load but optim state_dict is empty. {optim_state_dict}" + ) + + for key in model_state_dict: + if _FLAT_PARAM in key: + raise RuntimeError( + f"{key} contains {_FLAT_PARAM}. This can happen if the model " + "is not the root module." + ) + + +def _state_dict_fn(obj: nn.Module | torch.optim.Optimizer, api: str) -> Callable: + call = getattr(obj, api) + if call in _patched_state_dict: + call = functools.partial(getattr(obj.__class__, api), self=obj) + return call + + +def _get_fsdp_process_group( + model: nn.Module, info: _StateDictInfo +) -> dist.ProcessGroup | None: + if isinstance(model, FSDP): + fsdp_module = model + elif info.fsdp_modules: + fsdp_module = cast(FSDP, info.fsdp_modules[0]) + else: + return None + + if fsdp_module.sharding_strategy in HYBRID_SHARDING_STRATEGIES: + return None + + process_group = fsdp_module.process_group + if isinstance(process_group, tuple): + return None + return process_group + + +def _maybe_full_or_cpu_state_dict( + state_dict: dict[str, Any], info: _StateDictInfo +) -> dict[str, Any]: + if info.full_state_dict: + ranks_only = ( + () + if (not info.cpu_offload or not torch.distributed.is_initialized()) + else (0,) + ) + return _gather_state_dict( + state_dict, cpu_offload=info.cpu_offload, ranks_only=ranks_only + ) + elif info.cpu_offload: + return _offload_state_dict_to_cpu(state_dict) + else: + return state_dict + + +@torch.no_grad() +def _get_model_state_dict( + model: nn.Module, info: _StateDictInfo +) -> dict[str, ValueType]: + if not info.handle_model: + return {} + + with info.fsdp_context(): + state_dict = _state_dict_fn(model, "state_dict")() + + for key in list(state_dict.keys()): + fqns = _get_fqns(model, key) + if len(fqns) != 1: + raise AssertionError( + f"Expected 1 FQN for key '{key}', got {len(fqns)}: {fqns}" + ) + fqn = next(iter(fqns)) + if fqn != key: + # As we only support FSDP, DDP, and TP, the only cases are + # wrapper-based DDP and compiler. Verify if the assumption + # is correct. + def verify(key, fqn) -> bool: + if len(fqn) >= len(key): + return False + fqn_split = fqn.split(".") + key_split = key.split(".") + fqn_idx = 0 + for key_idx, key_name in enumerate(key_split): + if key_name == fqn_split[fqn_idx]: + fqn_idx += 1 + if fqn_idx == len(fqn_split): + return key_idx == len(key_split) - 1 + elif key_name in ("module", "_orig_mod"): + continue + else: + return False + return True + + if not verify(key, fqn): + raise RuntimeError(f"An unexpected key, {key}, exists. FQN is {fqn}") + state_dict[fqn] = state_dict.pop(key) + + if info.submodule_prefixes: + new_state_dict: dict[str, ValueType] = {} + # TODO: make this faster. + for fqn in state_dict: + for prefix in info.submodule_prefixes: + if not fqn.startswith(prefix): + continue + if info.keep_submodule_prefixes: + new_state_dict[fqn] = state_dict[fqn] + else: + new_fqn = fqn[len(prefix) :] + new_state_dict[new_fqn] = state_dict[fqn] + state_dict = new_state_dict + + if info.ignore_frozen_params: + for key, param in model.named_parameters(): + if param.requires_grad: + continue + fqns = _get_fqns(model, key) + for fqn in fqns: + state_dict.pop(fqn) + + return _maybe_full_or_cpu_state_dict(state_dict, info) + + +@torch.no_grad() +def _load_model_state_dict( + model: nn.Module, + state_dict: dict[str, ValueType], + info: _StateDictInfo, +) -> _IncompatibleKeys: + if not info.handle_model or (not state_dict and not info.broadcast_from_rank0): + return _IncompatibleKeys({}, {}) + + local_state_dict = {} + for key, value in _iterate_valid_model_state(model, info.dsd_fqn_modifiers): + fqns = _get_fqns(model, key, info.dsd_fqn_modifiers) + fqns_with_prefix = _get_fqns( + model, + key, + info.dsd_fqn_modifiers, + skip_ddp_prefix=False, + skip_compiler_prefix=False, + ) + + for fqn, fqn_with_prefix in zip(fqns, fqns_with_prefix): + if ( + not info.broadcast_from_rank0 or dist.get_rank() == 0 + ) and fqn != fqn_with_prefix: + load_value = state_dict.pop(fqn, None) + if load_value is None: + if info.strict: + raise RuntimeError(f"Missing key: {fqn}.") + else: + state_dict[fqn_with_prefix] = load_value + local_state_dict[fqn_with_prefix] = value + + assign = False + if info.broadcast_from_rank0 or info.full_state_dict: + devices = set() + for value in local_state_dict.values(): + if torch.is_tensor(value) and value.dim() > 0: + devices.add(value.device) + # In lora state_dict, there could be multiple devices, with meta device inside. + # Take the other device in the broadcast/distribute, and set assign to True + if torch.device("meta") in devices: + devices.remove(torch.device("meta")) + assign = True + if len(devices) == 0: + devices.add(dist.distributed_c10d._get_pg_default_device()) + elif len(devices) > 1: + raise ValueError("Multiple devices found") + + if info.broadcast_from_rank0: + _broadcast_state_dict( + state_dict, + local_state_dict, + device=devices.pop(), + strict=info.strict, + cpu_offload=info.cpu_offload, + ) + elif info.full_state_dict: + _distribute_state_dict(state_dict, local_state_dict, device=devices.pop()) + state_dict.update(local_state_dict) + + with info.fsdp_context(): + return cast( + _IncompatibleKeys, + _state_dict_fn(model, "load_state_dict")( + state_dict=state_dict, strict=info.strict, assign=assign + ), + ) + + +def _init_optim_state(optim: torch.optim.Optimizer) -> None: + """ + Initialize optim states by calling the step() with zero grads. + """ + if optim.state: + # The optimizer state is initialized. + return + + # There are some stateless optimizers like SGD. These optimizer will + # not return in the above condition. So if gradients exist, we should also + # return. If gradients do not exist, the following initialization should + # not disturb SGD because the gradients and lr are both zero. + for param_group in optim.param_groups: + for param in param_group[_PARAMS]: + if param.grad is not None: + return + + for param_group in optim.param_groups: + for param in param_group[_PARAMS]: + if param.requires_grad: + param.grad = torch.zeros_like(param) + + # Some optimizers will update parameters regardless of grads due to lr, so + # make lr to zero when calling `step()`. + lrs = [] + for param_group in optim.param_groups: + if "lr" in param_group: + lrs.append(param_group["lr"]) + param_group["lr"] = ( + torch.tensor(0.0) + if isinstance(param_group["lr"], torch.Tensor) + else 0.0 + ) + optim.step(closure=None) + # Whether to recover the "lr" should not matter too much as we will + # restore checkpointing later. + for param_group in optim.param_groups: + if "lr" in param_group: + param_group["lr"] = lrs.pop(0) + optim.zero_grad(set_to_none=True) + + +def _flatten_optim_state_dict(state_dict: OptimizerStateType) -> dict[str, ValueType]: + """ + This API flattens the optimizer state_dict to support optimizer resharding for + MPMD, e.g., pipeline parallelism. + + Without the API, the original optimizer state_dict looks like: + { + "state": { + "layer1.weight": { + "step": 10, "exp_avg": SomeTensor, "exp_avg_sq": SomeTensor + }, + "layer2.weight": { + "step": 10, "exp_avg": SomeTensor, "exp_avg_sq": SomeTensor + }, + }, + "param_groups": [ + { + "lr": 0.0, + "betas": (0.9, 0.95), ..., + "params": ["layer1.weight", "layer2.weight"] + } + ] + } + + With this API, the optimizer state_dict looks like: + { + "state.layer1.weight.step": 10, + "state.layer2.weight.step": 10, + "state.layer1.weight.exp_avg": SomeTensor, + "state.layer2.weight.exp_avg": SomeTensor, + "state.layer1.weight.exp_avg_sq": SomeTensor, + "state.layer2.weight.exp_avg_sq": SomeTensor, + "param_groups.layer1.weight.lr": 0.1, + "param_groups.layer2.weight.lr": 0.1, + "param_groups.layer1.weight.betas": (0.9, 0.95), + "param_groups.layer2.weight.betas": (0.9, 0.95), + } + + The "state" section supports arbitrary levels of nesting for optimizers like Shampoo. + """ + + def _flatten_state_nested_dict( + nested_dict: dict[str, Any], prefix: str + ) -> dict[str, ValueType]: + """ + Recursively flatten a nested dictionary with dot-separated keys. + + Args: + nested_dict: The dictionary to flatten + prefix: The prefix to prepend to all keys + + Returns: + Flattened dictionary with dot-separated keys + """ + flattened: dict[str, ValueType] = {} + + for key, value in nested_dict.items(): + # Convert all keys to strings for flattening + str_key = str(key) + full_key = f"{prefix}.{str_key}" if prefix else str_key + + if isinstance(value, dict): + # Recursively flatten nested dictionaries + flattened.update(_flatten_state_nested_dict(value, full_key)) + else: + # Base case: store the value with the flattened key + _raise_if_type_not_supported(value) + flattened[full_key] = value + + return flattened + + def _raise_if_type_not_supported(v): + if not isinstance(v, (torch.Tensor, int, float, dict)): + raise NotImplementedError( + "Flattening optimizer state_dict only supports " + "tensor, int, float, dict states now. " + f"Type is {type(v)}." + ) + + ret: dict[str, ValueType] = {} + + # Handle the "state" section with recursive flattening + for fqn, state in cast(DictValueType, state_dict[_STATE]).items(): + state_prefix = f"{_STATE}.{fqn}" + ret.update( + _flatten_state_nested_dict(cast(dict[str, Any], state), state_prefix) + ) + + # Handle the "param_groups" section with two-level flattening + for param_group in cast(ListDictValueType, state_dict[_PG]): + fqns = param_group.pop(_PARAMS) + for fqn in cast(list[str], fqns): + for k, v in param_group.items(): + ret[f"{_PG}.{fqn}.{k}"] = v + + return ret + + +def _unflatten_optim_state_dict( + optim: torch.optim.Optimizer, + state_dict: dict[str, ValueType], + info: _StateDictInfo, +) -> OptimizerStateType: + """ + This API unflattens the state_dict generated by _flatten_optim_state_dict(). + Supports arbitrary levels of nesting in the state section through recursive reconstruction. + + See the docstring of _flatten_optim_state_dict() for more detail. + """ + + def _reconstruct_nested_dict( + flattened_key: str, flattened_dict: dict[str, ValueType] + ) -> dict[str, ValueType]: + """ + Reconstructs a potentially nested value from flattened keys. + For non-nested values, returns the value directly. + For nested values, reconstructs the nested structure with string keys. + """ + + # Create the prefix to search for nested keys + # e.g., if flattened_key is "state.layer1.weight", prefix becomes "state.layer1.weight." + prefix = f"{flattened_key}." + # Initialize an empty dictionary to build our nested structure + nested_dict: dict[str, Any] = {} + + # Iterate through all keys in the flattened dictionary + for key, value in flattened_dict.items(): + # Check if this key is nested under our target key + # e.g., "state.layer1.weight.exp_avg" starts with "state.layer1.weight." + if not key.startswith(prefix): + # Skip keys that don't belong to this nested structure + continue + + # Remove the prefix to get just the nested part + # e.g., "state.layer1.weight.exp_avg" -> "exp_avg" + remaining_key = key[len(prefix) :] + # Split the remaining key into parts to build the nested structure + # e.g., "step" -> ["step"] or "momentum_buffer" -> ["momentum_buffer"] + parts = remaining_key.split(".") + # Start at the root of our new nested dictionary + current = nested_dict + + # Navigate through or create the nested dictionary structure + # For each part except the last one (which will hold the value) + for part in parts[:-1]: + # Create the nested dictionary if it doesn't exist yet + if part not in current: + current[part] = {} + # Move deeper into the nested structure + if not isinstance(current[part], dict): + raise AssertionError + current = current[part] + + # Set the value at the final level using the last part as the key + # e.g., current["exp_avg"] = tensor(...) + current[parts[-1]] = value + + # Return the reconstructed nested dictionary (empty dict if no keys matched at all) + return nested_dict + + state: DictValueType = {} + pg_state: ListDictValueType = [] + return_osd: OptimizerStateType = {_STATE: state, _PG: pg_state} + + for param_group in optim.param_groups: + pg_state.append({_PARAMS: []}) + for param in param_group[_PARAMS]: + for fqn in info.fqn_param_mapping[param]: + # If a parameter is shared, only one of the FQN will be used. + # So we need to verify which if this fqn is actually used in + # the state_dict. + if fqn in info.shared_params_mapping: + in_params = False + for k in param_group: + if k == _PARAMS: + continue + flatten_key = f"{_PG}.{fqn}.{k}" + if flatten_key in state_dict: + in_params = True + break + else: + in_params = True + + if not in_params: + continue + + params = pg_state[-1][_PARAMS] + if not isinstance(params, list): + raise AssertionError(f"Expected list, got {type(params)}") + params.append(fqn) + + # Only add state if param requires grad + if not param.requires_grad: + continue + + # Reconstruct state for this parameter + # pyrefly: ignore [unsupported-operation] + state[fqn] = {} + for state_name in optim.state[param]: + flattened_state_key = f"{_STATE}.{fqn}.{state_name}" + + if flattened_state_key not in state_dict: + # Try to reconstruct the value + reconstructed_value = _reconstruct_nested_dict( + flattened_state_key, state_dict + ) + # pyrefly: ignore [bad-index] + cast(DictValueType, state[fqn])[state_name] = ( + reconstructed_value + ) + else: + # Existing keys mean no nesting, directly use the value. + # pyrefly: ignore [bad-index] + cast(DictValueType, state[fqn])[state_name] = state_dict[ + flattened_state_key + ] + + first_param_fqn = cast(list[str], pg_state[-1][_PARAMS])[0] + for k in param_group: + if k == _PARAMS: + continue + value = state_dict[f"{_PG}.{first_param_fqn}.{k}"] + if k not in pg_state[-1]: + pg_state[-1][k] = value + elif pg_state[-1][k] != value: + raise RuntimeError( + "All the parameters in the same parameter group should have " + f"the same saved param_group value. But {first_param_fqn}.{k} " + f"is {value} while other(s) is {pg_state[-1][k]}." + ) + + return return_osd + + +@torch.no_grad() +def _get_optim_state_dict( + model: nn.Module, + optimizers: tuple[torch.optim.Optimizer, ...], + info: _StateDictInfo, +) -> OptimizerStateType: + if not info.handle_optim: + return {} + + optim_state_dict: OptimizerStateType = {_STATE: {}, _PG: []} + for optim in optimizers: + _init_optim_state(optim) + osd = _state_dict_fn(optim, "state_dict")() + if info.fsdp_modules: + with info.fsdp_context(): + osd = FSDP.optim_state_dict( + model, + optim, + osd, + group=_get_fsdp_process_group(model, info), + ) + + # We need to specially handle FlatParameter FSDP as + # FlatParameter FSDP converts the FQNs. + # There are no easy ways to do this conversion systematically. + # We can only use a string replacement without correctness check. + if not osd: + continue + for k in list(osd[_STATE].keys()): + if "_orig_mod" in k: + osd[_STATE][k.replace("_orig_mod.", "")] = osd[_STATE].pop(k) + for g in osd[_PG]: + params = [k.replace("_orig_mod.", "") for k in g[_PARAMS]] + g[_PARAMS] = params + else: + params = list(chain.from_iterable(g[_PARAMS] for g in optim.param_groups)) + param_pid_mapping = dict(zip(params, range(len(params)))) + fqn_pid_mapping = {} + for key, param in model.named_parameters(): + fqns = _get_fqns(model, key) + if len(fqns) != 1: + raise AssertionError( + f"Expected 1 FQN for key '{key}', got {len(fqns)}" + ) + fqn = next(iter(fqns)) + if param not in param_pid_mapping: + continue + # pyrefly: ignore [bad-index] + pid = param_pid_mapping[param] + fqn_pid_mapping[fqn] = pid + # pyrefly: ignore [unsupported-operation] + fqn_pid_mapping[pid] = fqn + + # Only convert top-level parameter IDs to FQNs, preserve nested key types + for key in list(osd[_STATE].keys()): + fqn = fqn_pid_mapping[key] + # Move the entire state dict value (which may contain nested integer keys) + # without modifying its internal structure + osd[_STATE][fqn] = osd[_STATE].pop(key) + + for group in osd[_PG]: + group[_PARAMS] = [fqn_pid_mapping[pid] for pid in group[_PARAMS]] + + if not osd: + continue + + cast(DictValueType, optim_state_dict[_STATE]).update(osd[_STATE]) + cast(ListDictValueType, optim_state_dict[_PG]).extend(osd[_PG]) + + if info.flatten_optimizer_state_dict: + optim_state_dict = cast( + OptimizerStateType, _flatten_optim_state_dict(optim_state_dict) + ) + + return _maybe_full_or_cpu_state_dict(optim_state_dict, info) + + +def _split_optim_state_dict( + model: nn.Module, + optim: torch.optim.Optimizer, + optim_state_dict: OptimizerStateType, + info: _StateDictInfo, +) -> OptimizerStateType: + """ + Extract the corresponding optim state_dict from ``optim_state_dict`` for + ``optim`` and return the result optim state_dict. + + Args: + model (nn.Module): the root model. + optim (torch.optim.Optimizer): the optimizer. + optim_state_dict (Dict[str, ValueType]): the superset optim state_dict that + contains the optim state_dict of ``optim``. + info (_StateDictInfo): state dict information. + + Returns: + The optim state_dict of ``optim``. + """ + + state: DictValueType = {} + pg_state: ListDictValueType = [] + return_osd: OptimizerStateType = {_STATE: state, _PG: pg_state} + pg_mapping: dict[int, int] = {} + + if all(isinstance(k, int) for k in cast(DictValueType, optim_state_dict[_STATE])): + return optim_state_dict + + for param_group in optim.param_groups: + pg_state.append({_PARAMS: []}) + for param in param_group[_PARAMS]: + for fqn in info.fqn_param_mapping[param]: + if fqn in info.shared_params_mapping: + in_params = False + for loaded_param_group in cast( + ListDictValueType, optim_state_dict[_PG] + ): + if fqn in cast(list[str], loaded_param_group[_PARAMS]): + in_params = True + break + else: + in_params = True + if not in_params: + continue + + params = pg_state[-1][_PARAMS] + if not isinstance(params, list): + raise AssertionError(f"Expected list, got {type(params)}") + params.append(fqn) + if param.requires_grad: + if fqn in cast(DictValueType, optim_state_dict[_STATE]): + state[fqn] = cast(DictValueType, optim_state_dict[_STATE])[fqn] + elif info.strict: + raise RuntimeError( + f"Missing optimizer state for parameter '{fqn}' in checkpoint. " + "The parameter requires gradients but has no saved optimizer state. " + "To load anyway, use StateDictOptions(strict=False)." + ) + for loaded_param_group in cast( + ListDictValueType, optim_state_dict[_PG] + ): + if fqn in cast(list[str], loaded_param_group[_PARAMS]): + pg_mapping[id(loaded_param_group)] = len(return_osd[_PG]) - 1 + + if len(param_group[_PARAMS]) == 0: + # Param_group with empty params. + ret = [] + for loaded_param_group in cast(ListDictValueType, optim_state_dict[_PG]): + if len(cast(list[str], loaded_param_group[_PARAMS])) == 0: + ret.append(loaded_param_group) + if len(ret) != 1: + raise ValueError( + "There are param groups that have zero parameters. " + "In such a case, DSD only support exactly one param group " + "with zero parameters." + "But the loaded state_dict has zero or more than one param groups " + "that have zero parameters." + ) + if len(optim_state_dict[_PG]) != len(optim.param_groups): + raise ValueError( + "When there is a parameter group that has zero parameters, " + "multiple optimizers are not supported." + ) + pg_mapping[id(loaded_param_group)] = len(return_osd[_PG]) - 1 + + for param_group in cast(ListDictValueType, optim_state_dict[_PG]): + pg_idx = pg_mapping.get(id(param_group), -1) + if pg_idx == -1: + continue + + for key, value in param_group.items(): + if key == _PARAMS: + continue + # TODO: check if value is the same if exists. + pg_state[pg_idx][key] = value + + return return_osd + + +@torch.no_grad() +def _load_optim_state_dict( + model: nn.Module, + optimizers: tuple[torch.optim.Optimizer, ...], + state_dict: OptimizerStateType, + info: _StateDictInfo, +) -> None: + if not info.handle_optim: + return + + for optim in optimizers: + _init_optim_state(optim) + if state_dict: + if _STATE in state_dict: + optim_state_dict = _split_optim_state_dict( + model, optim, state_dict, info + ) + else: + optim_state_dict = _unflatten_optim_state_dict( + optim, cast(dict[str, ValueType], state_dict), info + ) + else: + optim_state_dict = {} + if info.fsdp_modules: + # We need to specially handle FlatParameter FSDP as + # FlatParameter FSDP converts the FQNs. + for original_fqn, _ in model.named_parameters(): + fqns = _get_fqns(model, original_fqn) + fqns_with_compiler = _get_fqns( + model, original_fqn, skip_compiler_prefix=False + ) + if fqns == fqns_with_compiler: + continue + + if len(fqns) != 1: + raise AssertionError( + f"Expected 1 FQN for '{original_fqn}', got {len(fqns)}" + ) + fqn = fqns.pop() + fqn_with_compiler = fqns_with_compiler.pop() + for g in optim_state_dict[_PG]: + val = cast(dict[str, Any], g) + params = [ + key.replace(fqn, fqn_with_compiler) for key in val[_PARAMS] + ] + val[_PARAMS] = params + osd_state = cast(DictValueType, optim_state_dict[_STATE]) + for k in list(osd_state.keys()): + if fqn in k: + osd_state[k.replace(fqn, fqn_with_compiler)] = osd_state.pop(k) + + with info.fsdp_context(): + optim_state_dict = FSDP.optim_state_dict_to_load( + model, + optim, + optim_state_dict, + group=_get_fsdp_process_group(model, info), + ) + elif info.full_state_dict: + info.full_state_dict = False + local_state_dict = _get_optim_state_dict(model, (optim,), info) + info.full_state_dict = True + device = None + + def _device(t): + if t.dim() > 0: + nonlocal device + if device is None: + device = t.device + elif device != t.device: + raise ValueError("Device mismatch") + return t + + _ = tree_map_only(torch.Tensor, _device, local_state_dict) + if device is None: + raise AssertionError("Expected device to be set") + flatten_osd, osd_mapping = _flatten_state_dict(optim_state_dict) + flatten_local_osd, local_osd_mapping = _flatten_state_dict(local_state_dict) + if info.broadcast_from_rank0: + _broadcast_state_dict(flatten_osd, flatten_local_osd, device=device) + else: + _distribute_state_dict(flatten_osd, flatten_local_osd, device=device) + # The modifications listed seek to address the problem where optim might possess + # dissimilar parameters in comparison to optim_state_dict. This is achieved by + # incorporating differential parameters within local, which may result in optim + # having additional parameters ultimately. + for optim_key in flatten_osd: + if optim_key not in flatten_local_osd: + if optim_key not in osd_mapping: + raise AssertionError( + f"Expected key '{optim_key}' in osd_mapping" + ) + flatten_local_osd[optim_key] = flatten_osd[optim_key] + local_osd_mapping[optim_key] = osd_mapping[optim_key] + optim_state_dict = _unflatten_state_dict( + flatten_local_osd, local_osd_mapping + ) + for pg in optim_state_dict[_PG]: + if _PARAMS not in pg: + cast(dict[str, ValueType], pg)[_PARAMS] = [] + + # Note that we do not have to convert the FQN back to param id here if + # order in optim.param_groups[idx][_PARAMS] is the same as the one in + # optim_state_dict[_PG][idx][_PARAMS]. + _state_dict_fn(optim, "load_state_dict")(state_dict=optim_state_dict) + + +def get_model_state_dict( + model: nn.Module, + *, + submodules: set[nn.Module] | None = None, + options: StateDictOptions | None = None, +) -> dict[str, ValueType]: + """ + Return the model state_dict of ``model``. + + See ``get_state_dict`` for the detail usage. + + Args: + model (nn.Module): the nn.Module to the model. + submodules (deprecated): Optional[set[nn.Module]]: only return the model parameters + that belong to the submodules. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be returned. See + `StateDictOptions` for the details. + + Returns: + The state_dict for ``model``. + + :rtype: typing.Dict[str, ValueType] + """ + with _gc_context(): + info = _verify_options( + model, + (), + optim_only=False, + submodules=submodules, + options=options, + ) + model_state_dict = _get_model_state_dict(model, info) + _verify_state_dict(model_state_dict, {}, info) + return model_state_dict + + +def get_optimizer_state_dict( + model: nn.Module, + optimizers: torch.optim.Optimizer | Iterable[torch.optim.Optimizer], + *, + submodules: set[nn.Module] | None = None, + options: StateDictOptions | None = None, +) -> OptimizerStateType: + """ + Return the combined state_dict for optimizers. + + See ``get_state_dict`` for the detail usage. + + Args: + model (nn.Module): the nn.Module to the model. + optimizers (Union[None, Optimizer, Iterable[Optimizer]]): + The optimizers that are used to optimize ``model``. + submodules (deprecated): Optional[set[nn.Module]]: only return the model parameters + that belong to the submodules. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be returned. See + `StateDictOptions` for the details. + + Returns: + The state_dict for ``optimizers``. + + :rtype: OptimizerStateType + """ + with _gc_context(): + optimizers = ( + (optimizers,) + if isinstance(optimizers, torch.optim.Optimizer) + else tuple(optimizers) + ) + info = _verify_options( + model, + optimizers, + optim_only=True, + submodules=submodules, + options=options, + ) + optim_state_dict = _get_optim_state_dict(model, optimizers, info) + _verify_state_dict({}, optim_state_dict, info) + return optim_state_dict + + +def get_state_dict( + model: nn.Module, + optimizers: torch.optim.Optimizer | Iterable[torch.optim.Optimizer], + *, + submodules: set[nn.Module] | None = None, + options: StateDictOptions | None = None, +) -> tuple[dict[str, ValueType], OptimizerStateType]: + """ + Return the model state_dict and optimizers state_dict. + + ``get_state_dict`` can process any module that is parallelized by PyTorch + FSDP/fully_shard, DDP/replicate, tensor_parallel/parallelize_module, and any + combination of these parallelisms. The main functions of ``get_state_dict`` + are: 1.) returning a model and optimizer state_dict that can be resharded + with a different number of trainers and/or different parallelisms. + 2.) hiding the parallelism-specific state_dict APIs. Users don't have to call + these APIs. + 3.) sanity checking the result state_dict. + + The keys of the result state dictionary are the canonical FQNs (Fully + Qualified Names). A canonical FQN refers to the FQN based on a parameter's + position in an nn.Module hierarchy. More specifically, a canonical FQN to a + parameter is the FQN returned by ``module.named_parameters()`` or + ``module.named_buffers()`` when the module is not distributed by any + parallelisms. Since the optimizer internally uses parameter IDs to represent + a parameter, there will be a conversion from the parameter IDs to the + canonical FQNs when calling this API. + + ``get_state_dict`` can also process a module that is not parallelized. In + such a case, ``get_state_dict`` only performs one function -- converting the + optimizer parameter IDs to the canonical FQNs. + + Example: + >>> # xdoctest: +SKIP + >>> import torch + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> from torch.nn.parallel import DistributedDataParallel as DDP + >>> from torch.distributed.checkpoint.state_dict import get_state_dict + + >>> fsdp_model = FSDP(copy.deepcopy(model)) + >>> fsdp_optim = torch.optim.Adam(model.parameters(), lr=1e-3) + >>> ddp_model = DDP(copy.deepcopy(model)) + >>> ddp_optim = torch.optim.Adam(model.parameters(), lr=1e-3) + + + >>> ddp_state_dict, ddp_optim_state_dict = get_state_dict(ddp_model, ddp_optim) + >>> fsdp_state_dict, fsdp_optim_state_dict = get_state_dict( + ... fsdp_model, fsdp_optim + ... ) + + >>> # if we simply call ddp_model.state_dict() and fsdp_model.state_dict(), + >>> # the asserts will fail. + >>> assert ddp_state_dict == fsdp_state_dict + >>> assert ddp_optim_state == fsdp_optim_state_dict + + + Args: + model (nn.Module): the nn.Module to the model. + optimizers (Union[None, Optimizer, Iterable[Optimizer]]): + The optimizers that are used to optimize ``model``. + submodules (deprecated): Optional[set[nn.Module]]: only return the model parameters + that belong to the submodules. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be returned. See + `StateDictOptions` for the details. + + Returns: + ``Tuple`` that contain model state_dict and optimizer state_dict. + + :rtype: typing.Tuple[typing.Dict[str, ValueType], OptimizerStateType] + """ + + with _gc_context(): + optimizers = ( + (optimizers,) + if isinstance(optimizers, torch.optim.Optimizer) + else tuple(optimizers) + ) + info = _verify_options( + model, + optimizers, + optim_only=False, + submodules=submodules, + options=options, + ) + model_state_dict = _get_model_state_dict(model, info) + optim_state_dict = _get_optim_state_dict(model, optimizers, info) + _verify_state_dict(model_state_dict, optim_state_dict, info) + return model_state_dict, optim_state_dict + + +def _unflatten_model_state_dict( + model: nn.Module, + state_dict: dict[nn.Module, dict[str, ValueType]] | dict[str, ValueType], +) -> dict[str, ValueType]: + if not state_dict: + return {} + + if isinstance(next(iter(state_dict.keys())), nn.Module): + warnings.warn( + "Passing model_state_dict as a ``Dict[nn.Module, Dict[str, Any]]``" + "is deprecated and will be removed in 2.5. If you need this " + "feature, please preprocess the model_state_dict to achieve the " + "same functionality.", + FutureWarning, + stacklevel=2, + ) + cast_state_dict = cast(dict[nn.Module, dict[str, ValueType]], state_dict) + new_state_dict: dict[str, ValueType] = {} + for submodule, sub_state_dict in cast_state_dict.items(): + for name, m in model.named_modules(): + if m != submodule: + continue + + fqns = _get_fqns(model, name) + if len(fqns) != 1: + raise AssertionError( + "FQNs for a submodule should only have 1 element" + ) + prefix = f"{next(iter(fqns))}." + new_state_dict.update( + {prefix + subfqn: value for subfqn, value in sub_state_dict.items()} + ) + return new_state_dict + else: + return cast(dict[str, ValueType], state_dict) + + +def set_model_state_dict( + model: nn.Module, + model_state_dict: dict[str, ValueType], + *, + options: StateDictOptions | None = None, +) -> _IncompatibleKeys: + """Load the model state_dict. + + The counterpart of ``get_model_state_dict`` to set the state_dict to the + model. See ``set_state_dict`` for the detail usage. + + Args: + model (nn.Module): the nn.Module to the model. + model_state_dict: (Dict[str, ValueType]): + the model state_dict to load. If the key of the ``model_state_dict`` + is nn.Module, the key is a submodule of ``model`` and the value should + be the state_dict of the submodule. When loading the state_dict, + the prefix of the submodule will be append to the state_dict. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be loaded. See + `StateDictOptions` for the details. + + Returns: + ``NamedTuple`` with ``missing_keys`` and ``unexpected_keys`` fields: + * **missing_keys** is a list of str containing the missing keys + * **unexpected_keys** is a list of str containing the unexpected keys + + :type model_state_dict: typing.Dict[str, ValueType] + """ + model_state_dict: dict[str, ValueType] = _unflatten_model_state_dict( + model, model_state_dict + ) + with _gc_context(): + info = _verify_options(model, (), optim_only=False, options=options) + + _verify_state_dict(model_state_dict, {}, info) + return _load_model_state_dict(model, model_state_dict, info) + + +def set_optimizer_state_dict( + model: nn.Module, + optimizers: torch.optim.Optimizer | Iterable[torch.optim.Optimizer], + optim_state_dict: OptimizerStateType, + *, + options: StateDictOptions | None = None, +) -> None: + """Load the optimizers state_dict. + + The counterpart of ``get_optimizer_state_dict`` to set the state_dict to the + optimizers. See ``set_state_dict`` for the detail usage. + + WARN: ``set_optimizer_state_dict`` can only be called before ``backward()`` or after + ``step()`` is called on the optimizers. Otherwise, the optimizer states won't be + initialized correctly. + + Args: + model (nn.Module): the nn.Module to the model. + optimizers (Union[Optimizer, Iterable[Optimizer]]): + The optimizers that are used to optimize ``model``. + optim_state_dict: OptimizerStateType: + the optimizer state_dict to load. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be loaded. See + `StateDictOptions` for the details. + + Returns: + None + + :type optim_state_dict: typing.OptimizerStateType + """ + with _gc_context(): + optimizers = ( + (optimizers,) + if isinstance(optimizers, torch.optim.Optimizer) + else tuple(optimizers) + ) + info = _verify_options(model, optimizers, optim_only=True, options=options) + + _verify_state_dict({}, optim_state_dict, info) + _load_optim_state_dict(model, optimizers, optim_state_dict, info) + + +def set_state_dict( + model: nn.Module, + optimizers: torch.optim.Optimizer | Iterable[torch.optim.Optimizer], + *, + model_state_dict: dict[str, ValueType], + optim_state_dict: OptimizerStateType, + options: StateDictOptions | None = None, +) -> _IncompatibleKeys: + """Load the model state_dict and optimizers state_dict. + + The counterpart of ``get_state_dict`` to set the state_dict to the model and + optimizers. The given ``model_state_dict`` and ``optim_state_dict`` do not + have to be returned by ``get_state_dict`` but must meet the following + requirements: 1) all FQNs are canonical FQNs as defined in ``get_state_dict``, + 2) if a tensor is sharded, it must be either a ShardedTensor or DTensor, + 3) optimizer state_dict cannot contain the parameter IDs; the keys should be + the canonical FQNs. + + WARN: ``set_state_dict`` can only be called before ``backward()`` or after ``step()`` + is called on the optimizers. Otherwise, the optimizer states won't be initialized + correctly. + + Args: + model (nn.Module): the nn.Module to the model. + optimizers (Union[Optimizer, Iterable[Optimizer]]): + The optimizers that are used to optimize ``model``. + model_state_dict: (Union[Dict[nn.Module, Dict[str, ValueType]], Dict[str, ValueType]]): + the model state_dict to load. If the key of the ``model_state_dict`` + is nn.Module, the key is a submodule of ``model`` and the value should + be the state_dict of the submodule. When loading the state_dict, + the prefix of the submodule will be append to the state_dict. + optim_state_dict: OptimizerStateType: + the optimizer state_dict to load. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be loaded. See + `StateDictOptions` for the details. + + Returns: + ``NamedTuple`` with ``missing_keys`` and ``unexpected_keys`` fields: + * **missing_keys** is a list of str containing the missing keys of the model state_dict. + * **unexpected_keys** is a list of str containing the unexpected keys of the model state_dict. + + :type model_state_dict: typing.Dict[str, ValueType] + :type optim_state_dict: typing.OptimizerStateType + """ + + model_state_dict: dict[str, ValueType] = _unflatten_model_state_dict( + model, model_state_dict + ) + with _gc_context(): + optimizers = ( + (optimizers,) + if isinstance(optimizers, torch.optim.Optimizer) + else tuple(optimizers) + ) + info = _verify_options( + model, optimizers, optim_only=not model_state_dict, options=options + ) + + _verify_state_dict(model_state_dict, optim_state_dict, info) + _load_optim_state_dict(model, optimizers, optim_state_dict, info) + return _load_model_state_dict(model, model_state_dict, info) + + +# TODO: correct the state_dict function signature. +# TODO: this API is not yet fully tested. Make it private +@no_type_check +def _patch_model_state_dict( + model: nn.Module, + *, + options: StateDictOptions | None = None, +) -> None: + """Patch the ``state_dict`` and ``load_state_dict`` attributes of ``model``. + + Patch the ``state_dict`` and ``load_state_dict`` attributes of ``model`` to + be a partial function to call ``get_state_dict`` and ``set_state_dict``. + + Example: + from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + from torch.distributed.checkpoint.state_dict import patch_model_state_dict + + model = fsdp(model) + patch_model_state_dict(model) + + Args: + model (nn.Module): the nn.Module to the model. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be loaded. See + `StateDictOptions` for the details. + Returns: + None + """ + + _state_dict_call = functools.partial( + get_model_state_dict, + model=model, + options=options, + ) + + def state_dict_call(): + return _state_dict_call() + + model.state_dict = state_dict_call + + _load_state_dict_call = functools.partial( + set_model_state_dict, + model=model, + options=options, + ) + + def load_state_dict_call(state_dict: dict[str, Any]): + _load_state_dict_call(model_state_dict=state_dict) + + model.load_state_dict = load_state_dict_call + + _patched_state_dict.add(state_dict_call) + _patched_state_dict.add(load_state_dict_call) + + +# TODO: correct the load_state_dict function signature. +# TODO: this API is not yet fully tested. Make it private +@no_type_check +def _patch_optimizer_state_dict( + model: nn.Module, + *, + optimizers: tuple[torch.optim.Optimizer, ...], + options: StateDictOptions | None = None, +) -> None: + """Patch the ``state_dict`` and ``load_state_dict`` attributes of ``optimizers``. + + Patch the ``state_dict`` and ``load_state_dict`` attributes of ``optimizers`` to + be a partial function to call ``get_state_dict`` and ``set_state_dict``. + + Note that if there are multiple optimizers, all of the optimizers will be patched. + So users only need to call one of the state_dict() to get the full result. + + Example: + from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + from torch.distributed.checkpoint.state_dict import patch_model_state_dict + + model = fsdp(model) + patch_model_state_dict(model) + + Args: + model (nn.Module): the nn.Module to the model. + options (StateDictOptions): the options to control how + model state_dict and optimizer state_dict should be loaded. See + `StateDictOptions` for the details. + Returns: + None + """ + + _state_dict_call = functools.partial( + get_optimizer_state_dict, + model=model, + optimizers=optimizers, + options=options, + ) + + def state_dict_call(): + return _state_dict_call() + + _load_state_dict_call = functools.partial( + set_optimizer_state_dict, + model=model, + optimizers=optimizers, + options=options, + ) + + def load_state_dict_call(state_dict: dict[str, Any]): + _load_state_dict_call(optim_state_dict=state_dict) + + _patched_state_dict.add(state_dict_call) + _patched_state_dict.add(load_state_dict_call) + optimizers = ( + (optimizers,) + if isinstance(optimizers, torch.optim.Optimizer) + else tuple(optimizers) + ) + for optim in optimizers: + optim.state_dict = state_dict_call + optim.load_state_dict = load_state_dict_call diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_loader.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_loader.py new file mode 100644 index 0000000000000000000000000000000000000000..1ea9eca98befb579093336166fdcafd81347175b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_loader.py @@ -0,0 +1,401 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs +import inspect +import logging +import os +import warnings +from typing import Any, cast, TYPE_CHECKING +from typing_extensions import deprecated + +import torch +import torch.distributed as dist +from torch.distributed.checkpoint.default_planner import _EmptyStateDictLoadPlanner +from torch.distributed.checkpoint.logger import _dcp_method_logger +from torch.distributed.checkpoint.stateful import Stateful + +from ._storage_utils import _storage_setup +from .default_planner import DefaultLoadPlanner +from .planner import LoadPlan, LoadPlanner +from .storage import StorageReader +from .utils import _api_bc_check, _DistWrapper, _profile + + +if TYPE_CHECKING: + from torch.distributed.checkpoint.metadata import Metadata + +__all__ = ["load_state_dict", "load"] + +logger = logging.getLogger() + + +@deprecated( + "`load_state_dict` is deprecated and will be removed in future versions. " + "Please use `load` instead.", + category=FutureWarning, +) +def load_state_dict( + state_dict: dict[str, Any], + storage_reader: StorageReader, + process_group: dist.ProcessGroup | None = None, + coordinator_rank: int = 0, + no_dist: bool = False, + planner: LoadPlanner | None = None, +) -> None: + """This method is deprecated. Please switch to 'load'.""" + storage_reader.reset() + with _profile(): + # TODO: test returning `load` here instead. + return _load_state_dict( + state_dict, + storage_reader, + process_group, + coordinator_rank, + no_dist, + planner, + ) + + +@_dcp_method_logger(log_exceptions=True) +@_api_bc_check +def load( + state_dict: dict[str, Any], + *, + checkpoint_id: str | os.PathLike | None = None, + storage_reader: StorageReader | None = None, + planner: LoadPlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, +) -> None: + """ + Load a checkpoint into a distributed state dict in SPMD style. + + Each rank must have the same keys in their ``state_dict`` provided to this + API. Mismatched keys may result in hangs or errors. If unsure, you can use + the ``utils._assert_same_keys`` API to check (but may incur communication + costs). + + Each rank will try to read the least amount of data necessary + to fulfill the requested `state_dict`. When loading :class:`ShardedTensor` + or :class:`DTensor` instances, each rank only reads data for their local shards. + + For each ``Stateful`` object (having both a ``state_dict`` and a ``load_state_dict``), + load will first call ``state_dict`` before attempting deserialization, followed by + ``load_state_dict`` once the deserialization is complete. + For each non-``Stateful`` object, load will deserialize the object, and then replace + it in the ``state_dict`` with the deserialized object. + + .. warning:: + All tensors in ``state_dict`` must be allocated on their + destination device *prior to* calling this function. + + All non-tensor data is loaded using `torch.load()` and modified in place + on state_dict. + + .. warning:: + Users must call `load_state_dict` on the root module to ensure load + post-processing and non-tensor data properly propagates. + + .. note: + If no process group is initialized, this function will assume the intent + is to load a checkpoint into the local process. This can be useful in the + case of local inference, and when using regular Tensors (as opposed to DTensor + or ShardedTensor) + + .. note: + Rank 0 is assumed to be the coordinator rank. + + Args: + state_dict (Dict[str, Any]): The state_dict to load the checkpoint into. + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is a key-value store. + (Default: ``None``) + storage_reader (Optional[StorageReader]): + Instance of StorageWriter used to perform reads. If this is not + specified, DCP will automatically infer the reader based on the + checkpoint_id. If checkpoint_id is also None, an exception will + be raised. (Default: ``None``) + planner (Optional[LoadPlanner]): + Instance of LoadPlanner. If this is not specified, the default + planner will be used. (Default: ``None``) + process_group (Optional[ProcessGroup]): + ProcessGroup to be used for cross-rank synchronization. + (Default: ``None``) + no_dist (bool): If ``True``, this function will assume the intent is to load + a checkpoint without using cross-rank synchronization. (Default: ``False``) + Returns: + None. + + Examples + >>> # xdoctest: +SKIP + >>> my_model = MyModule() + >>> optimizer = Adagrad(my_model.parameters()) + >>> model_state_dict = my_model.state_dict() + >>> fs_storage_reader = torch.distributed.checkpoint.FileSystemReader( + ... "/checkpoint/1" + ... ) + + >>> torch.distributed.checkpoint.load_state_dict( + >>> state_dict=model_state_dict, + >>> storage_reader=fs_storage_reader, + >>> ) + + >>> # module.load_state_dict() function might have customized steps + >>> # to flush the state_dict, must call it to + >>> # ensure correct behavior. + >>> my_model.load_state_dict(model_state_dict) + + .. note:: + load_state_dict uses collectives to coordinate reads across ranks. + For NCCL-based process groups, internal tensor representations of + objects must be moved to the GPU device before communication takes place. + In this case, the device used is given by ``torch.cuda.current_device()`` + and it is the user's responsibility to ensure that this is set so that each + rank has an individual GPU, via ``torch.cuda.set_device()``. + """ + + no_dist = no_dist or (not dist.is_available()) or (not dist.is_initialized()) + if no_dist: + warnings.warn( + "torch.distributed is disabled, unavailable or uninitialized, assuming the intent is to load in a single process.", + stacklevel=2, + ) + + with _profile(): + storage_reader = cast( + StorageReader, _storage_setup(storage_reader, checkpoint_id, reader=True) + ) + + # All ranks must have the same keys in their `state_dict` provided to + # this API. See documentation for more details. + # Here we simply sort the keys to ensure that all ranks load values in + # the same order. + keys = sorted(state_dict.keys()) + + stateful_sd = {} + for key in keys: + if key not in state_dict: + continue + elem = state_dict[key] + stateful_sd[key] = elem.state_dict() if isinstance(elem, Stateful) else elem + + _load_state_dict( + state_dict=stateful_sd, + storage_reader=storage_reader, + process_group=process_group, + no_dist=no_dist, + planner=planner, + ) + for key in keys: + if key not in state_dict: + continue + elem = state_dict[key] + if isinstance(elem, Stateful): + # If the state_dict is a Stateful object, + # DCP does an in-place load in the original state dict. + elem.load_state_dict(stateful_sd[key]) + else: + # Otherwise, replace the state_dict with the loaded state_dict. + state_dict[key] = stateful_sd[key] + + +def _load_state_dict( + state_dict: dict[str, Any], + storage_reader: StorageReader, + process_group: dist.ProcessGroup | None = None, + coordinator_rank: int = 0, + no_dist: bool = False, + planner: LoadPlanner | None = None, +) -> None: + torch._C._log_api_usage_once("torch.distributed.checkpoint.load_state_dict") + + distW = _DistWrapper(process_group, not no_dist, coordinator_rank) + if planner is None: + planner = DefaultLoadPlanner() + + ckpt_kwargs = {} + if (ckpt_id := getattr(storage_reader, "checkpoint_id", None)) is not None: + ckpt_kwargs["checkpoint_id"] = ckpt_id + ckpt_kwargs["process_group"] = distW.group + + use_collectives = True + metadata: Metadata | None = None + + @_dcp_method_logger(**ckpt_kwargs) + def local_step(): + nonlocal use_collectives + nonlocal metadata + + # Use global metadata if available, otherwise fallback to rank local metadata + global_metadata_exc: Exception | None = None + rank_metadata_exc: Exception | None = None + try: + metadata = storage_reader.read_metadata() + except Exception as e: + global_metadata_exc = e + logger.warning( + "Global metadata is not found. Falling back to rank local metadata.", + exc_info=True, + ) + + if ( + not metadata + and "kwargs" in inspect.signature(storage_reader.read_metadata).parameters + ): + try: + metadata = storage_reader.read_metadata(rank=distW.rank) + use_collectives = False + except Exception as e: + rank_metadata_exc = e + logger.warning("Rank local metadata is not found.", exc_info=True) + + if planner is None: + raise AssertionError("planner is None") + if metadata is None: + error_parts = ["metadata is None"] + if global_metadata_exc is not None: + error_parts.append( + f"global metadata read failed: {global_metadata_exc}" + ) + if rank_metadata_exc is not None: + error_parts.append( + f"rank local metadata read failed: {rank_metadata_exc}" + ) + raise AssertionError("; ".join(error_parts)) + planner.set_up_planner(state_dict, metadata, distW.is_coordinator) + + if ( + "kwargs" + in inspect.signature(storage_reader.set_up_storage_reader).parameters + ): + storage_reader.set_up_storage_reader( + metadata, + distW.is_coordinator, + rank=distW.rank, + use_collectives=use_collectives, + ) + else: + storage_reader.set_up_storage_reader(metadata, distW.is_coordinator) + + local_plan = planner.create_local_plan() + local_plan = storage_reader.prepare_local_plan(local_plan) + return local_plan + + @_dcp_method_logger(**ckpt_kwargs) + def global_step(all_local_plans): + if planner is None: + raise AssertionError("planner is None") + all_local_plans = planner.create_global_plan(all_local_plans) + all_local_plans = storage_reader.prepare_global_plan(all_local_plans) + return all_local_plans + + central_plan: LoadPlan | None = None + if use_collectives: + central_plan = distW.reduce_scatter("plan", local_step, global_step) + else: + local_plan: LoadPlan = local_step() + global_plan: list[LoadPlan] = global_step([local_plan]) + central_plan = global_plan[0] + + @_dcp_method_logger(**ckpt_kwargs) + def read_data(): + if planner is None: + raise AssertionError("planner is None") + if central_plan is None: + raise AssertionError("central_plan is None") + final_local_plan = planner.finish_plan(central_plan) + all_reads = storage_reader.read_data(final_local_plan, planner) + + all_reads.wait() + return None + + if use_collectives: + _ = distW.all_gather("read", read_data) + else: + read_data() + distW.barrier() + + +def _load_state_dict_from_keys( + keys: set[str] | str | None = None, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_reader: StorageReader | None = None, + process_group: dist.ProcessGroup | None = None, +) -> dict[str, Any]: + """ + Load only the specified keys from the checkpoint, if no keys are specified, the entire + checkpoint will be loaded. Note, this method completely loads the checkpoint into the + current process and is not distributed. + + .. warning:: + + + .. warning:: + + All non-tensor data is loaded using `torch.load()` + + .. note: + As opposed to the usual pattern, this function does not take a state dict as input + and does not load inplace. Instead, a new state dict is directly initialized and read + from file. + + .. note: + If no process group is initialized, this function will assume the intent + is to load a checkpoint into the local process. This can be useful in the + case of local inference, and when using regular Tensors (as opposed to DTensor + or ShardedTensor) + + .. note: + Rank 0 is assumed to be the coordinator rank. + + Args: + keys (Optional[Union[set[str], str]]): + Loads any key specified in this set. If no keys are specified, the entire checkpoint + is loaded. + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is a key-value store. + (Default: ``None``) + storage_reader (Optional[StorageReader]): + Instance of StorageWriter used to perform reads. If this is not + specified, DCP will automatically infer the reader based on the + checkpoint_id. If checkpoint_id is also None, an exception will + be raised. (Default: ``None``) + process_group (Optional[ProcessGroup]): + ProcessGroup to be used for cross-rank synchronization. + (Default: ``None``) + + Returns: + State dict from specified keys + """ + torch._C._log_api_usage_once( + "torch.distributed.checkpoint._load_state_dict_from_keys" + ) + + no_dist = not (dist.is_available() and dist.is_initialized()) + if no_dist: + warnings.warn( + "torch.distributed is unavailable or uninitialized, assuming the intent is to load in a single process.", + stacklevel=2, + ) + + storage_reader = cast( + StorageReader, _storage_setup(storage_reader, checkpoint_id, reader=True) + ) + + if isinstance(keys, str): + keys = {keys} + + sd: dict[str, Any] = {} + _load_state_dict( + state_dict=sd, + storage_reader=storage_reader, + process_group=process_group, + no_dist=no_dist, + planner=_EmptyStateDictLoadPlanner(keys=keys), + ) + + return sd diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_saver.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_saver.py new file mode 100644 index 0000000000000000000000000000000000000000..ed5443862e77ff25a9da6cc4502dfaae5e5202a6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/state_dict_saver.py @@ -0,0 +1,519 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs +import contextlib +import inspect +import os +import warnings +from concurrent.futures import Future +from dataclasses import dataclass +from enum import Enum +from typing import cast, TYPE_CHECKING +from typing_extensions import deprecated + +import torch +import torch.distributed as dist +from torch.distributed._state_dict_utils import STATE_DICT_TYPE +from torch.distributed.checkpoint._async_process_executor import ( + _ProcessBasedAsyncCheckpointExecutor, +) +from torch.distributed.checkpoint._async_thread_executor import ( + _ThreadBasedAsyncCheckpointExecutor, +) +from torch.distributed.checkpoint._storage_utils import _storage_setup +from torch.distributed.checkpoint.default_planner import DefaultSavePlanner +from torch.distributed.checkpoint.logger import _dcp_method_logger +from torch.distributed.checkpoint.metadata import Metadata +from torch.distributed.checkpoint.planner import SavePlan, SavePlanner +from torch.distributed.checkpoint.staging import ( + AsyncStager, + DefaultStager, + StagingOptions, +) +from torch.distributed.checkpoint.stateful import Stateful +from torch.distributed.checkpoint.storage import StorageWriter, WriteResult +from torch.distributed.distributed_c10d import _get_default_group + +from .utils import _api_bc_check, _DistWrapper, _profile + + +if TYPE_CHECKING: + from torch.distributed.checkpoint._async_executor import _AsyncCheckpointExecutor + + +__all__ = [ + "save_state_dict", + "save", + "async_save", + "AsyncCheckpointerType", + "AsyncSaveResponse", +] + + +class AsyncCheckpointerType(Enum): + """Enum for async checkpointer type.""" + + THREAD = "thread" + PROCESS = "process" + + +@deprecated( + "`save_state_dict` is deprecated and will be removed in future versions." + "Please use `save` instead.", + category=FutureWarning, +) +def save_state_dict( + state_dict: STATE_DICT_TYPE, + storage_writer: StorageWriter, + process_group: dist.ProcessGroup | None = None, + coordinator_rank: int = 0, + no_dist: bool = False, + planner: SavePlanner | None = None, +) -> Metadata: + """This method is deprecated. Please switch to 'save'.""" + storage_writer.reset() + + # TODO: test returning `save` here instead. + with _profile(): + return _save_state_dict( + state_dict, + storage_writer, + process_group, + coordinator_rank, + no_dist, + planner, + ) + + +@_dcp_method_logger(log_exceptions=True) # type: ignore[arg-type] +@_api_bc_check +def save( + state_dict: STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + no_dist: bool = False, + use_collectives: bool = True, +) -> Metadata: + """ + Save a distributed model in SPMD style. + + This function is different from ``torch.save()`` as it handles + ``ShardedTensor`` , and ``DTensor`` by having each rank only save their local shards. + + For each ``Stateful`` object (having both a ``state_dict`` and a ``load_state_dict``), + save will call ``state_dict`` before serialization. + + .. warning:: + There are no guarantees of Backwards Compatibility across PyTorch versions + for saved state_dicts. + + .. warning:: + If using the `process_group` argument, make sure that only its ranks + call `save_state_dict` and that all data in state_dict belong to it. + + .. note:: + When saving checkpoint for FSDP's `ShardingStrategy.HYBRID_SHARD`, only one of + the shard_group should be calling `save_state_dict` and the corresponding process + group needs to be passed in. + + .. note:: + If no process group is available, this function assumes the intention is to save the + state_dict in the local process. + + .. note: + Rank 0 is assumed to be the coordinator rank. + + + Args: + state_dict (Dict[str, Any]): The state_dict to save. + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is a key-value store. + (Default: ``None``) + storage_writer (Optional[StorageWriter]): + Instance of StorageWriter used to perform writes. If this is not + specified, DCP will automatically infer the writer based on the + checkpoint_id. If checkpoint_id is also None, an exception will + be raised. (Default: ``None``) + planner (Optional[SavePlanner]): + Instance of SavePlanner. If this is not specified, the default + planner will be used. (Default: ``None``) + process_group (Optional[ProcessGroup]): + ProcessGroup to be used for cross-rank synchronization. + (Default: ``None``) + no_dist (bool): + If ``True``, this function will assume the intent is to load + a checkpoint on a single rank/process. + (Default: ``False``) + use_collectives (bool): If ``False``, this function will assume the intent is to save + a checkpoint without using cross-rank synchronization. + (Default: ``True``) + This configuration is experimental and should be used with caution. + It will change the format of the saved checkpoint and may not be backward compatible. + + Returns: + Metadata: Metadata object for the saved checkpoint. + + Example: + >>> # xdoctest: +SKIP + >>> my_model = MyModule() + + >>> state_dict = {"model": my_model} + + >>> fs_storage_writer = torch.distributed.checkpoint.FileSystemWriter( + ... "/checkpoint/1" + ... ) + >>> torch.distributed.checkpoint.save( + >>> state_dict=state_dict, + >>> storage_writer=fs_storage_writer, + >>> ) + + .. note:: + save_state_dict uses collectives to coordinate writes across ranks. + For NCCL-based process groups, internal tensor representations of + objects must be moved to the GPU device before communication takes place. + In this case, the device used is given by ``torch.cuda.current_device()`` + and it is the user's responsibility to ensure that this is set so that + each rank has an individual GPU, via ``torch.cuda.set_device()``. + """ + torch._C._log_api_usage_once("torch.distributed.checkpoint.save") + + no_dist = no_dist or (not dist.is_available()) or (not dist.is_initialized()) + if no_dist: + warnings.warn( + "torch.distributed is disabled, unavailable or uninitialized, assuming the intent is to save in a single process.", + stacklevel=2, + ) + + with _profile(): + storage_writer = cast( + StorageWriter, _storage_setup(storage_writer, checkpoint_id, reader=False) + ) + + return _save_state_dict( + state_dict=_stateful_to_state_dict(state_dict), + storage_writer=storage_writer, + process_group=process_group, + no_dist=no_dist, + planner=planner, + use_collectives=use_collectives, + ) + + +@dataclass +class AsyncSaveResponse: + """This class contains futures for staging and upload completion. + It is returned by async_save(). + staging_completion is a future that indicates when local copy + of state_dict is complete. + upload_completion is a future that indicates when a checkpoint + completed saving. + """ + + staging_completion: Future[None] + upload_completion: Future[None] + + +@_dcp_method_logger(log_exceptions=True) +def async_save( + state_dict: STATE_DICT_TYPE, + *, + checkpoint_id: str | os.PathLike | None = None, + storage_writer: StorageWriter | None = None, + planner: SavePlanner | None = None, + process_group: dist.ProcessGroup | None = None, + async_checkpointer_type: AsyncCheckpointerType = AsyncCheckpointerType.THREAD, + async_stager: AsyncStager | None = None, + no_dist: bool = False, + use_collectives: bool = True, +) -> Future | AsyncSaveResponse: + """Asynchronous version of ``save``. This code first de-stages the state_dict on to the + staging storage (defaults to CPU memory), and then calls the `save` in a separate thread. + + .. warning:: + This feature is experimental and subject to change. + If you provide an ``async_stager``, call ``close()`` after the last + checkpoint is saved. Internally created default stagers are closed + automatically. + + Args: + state_dict (Dict[str, Any]): The state_dict to save. + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is a key-value store. + (Default: ``None``) + storage_writer (Optional[StorageWriter]): + Instance of StorageWriter used to perform 'stage' and 'save'. If + this is not specified, DCP will automatically infer the writer based on the + checkpoint_id. If checkpoint_id is also None, an exception will + be raised. (Default: ``None``) + planner (Optional[SavePlanner]): + Instance of SavePlanner. If this is not specified, the default + planner will be used. (Default: ``None``) + process_group (Optional[ProcessGroup]): + ProcessGroup to be used for cross-rank synchronization. + (Default: ``None``) + async_checkpointer_type (AsyncCheckpointerType): + whether to do checkpoint in separate thread or process + (Default: ``AsyncCheckpointerType.THREAD``) + async_stager (AsyncStager): + provides staging implementation. If ``storage_writer`` implements + ``AsyncStager`` and ``async_stager`` is not provided, ``storage_writer`` + will be used for staging. User-provided stagers remain owned by the + caller and should be closed after the last checkpoint is saved. + no_dist (bool): + If ``True``, this function will assume the intent is to save + a checkpoint on a single rank/process. + (Default: ``False``) + use_collectives: If False, Save the checkpoint without rank coordination. (Default: ``True``) + This configuration is experimental and should be used with caution. + It will change the format of the saved checkpoint and may not be backward compatible. + + Returns: + Future: A future holding the resultant Metadata object from `save`. + + Example: + >>> # xdoctest: +SKIP + >>> my_model = MyModule() + + >>> state_dict = {"model": my_model} + + >>> fs_storage_writer = torch.distributed.checkpoint.FileSystemWriter( + ... "/checkpoint/1" + ... ) + >>> checkpoint_future = torch.distributed.checkpoint.async_save( + >>> state_dict=state_dict, + >>> storage_writer=fs_storage_writer, + >>> ) + >>> + >>> # ... do some work ... + >>> + >>> checkpoint_future.result() + + """ + torch._C._log_api_usage_once("torch.distributed.checkpoint.async_save") + + if dist.is_available() and dist.is_initialized(): + pg = process_group or _get_default_group() + if torch.device("cpu") not in pg._device_types: + raise AssertionError( + "A CPU backend must be enabled for async save; try initializing process group with 'cpu:gloo,cuda:nccl'" + ) + + owned_async_stager: AsyncStager | None = None + if async_stager is None: + if storage_writer is not None and isinstance(storage_writer, AsyncStager): + # bwc with old storage_writers + async_stager = storage_writer + else: + async_stager = DefaultStager( + StagingOptions( + False, + False, + False, + False, + ) + ) + owned_async_stager = async_stager + + state_dict = _stateful_to_state_dict(state_dict) + + owned_async_stager_closed = False + + def maybe_close_owned_async_stager() -> None: + nonlocal owned_async_stager_closed + if owned_async_stager is not None and not owned_async_stager_closed: + owned_async_stager_closed = True + owned_async_stager.close() + + @_dcp_method_logger(log_exceptions=True) + def stage_state_dict() -> Future[STATE_DICT_TYPE] | STATE_DICT_TYPE: + return async_stager.stage(state_dict) + + with contextlib.ExitStack() as stack: + if owned_async_stager is not None: + stack.callback(maybe_close_owned_async_stager) + + staging_future_or_state_dict = stage_state_dict() + + upload_executor: _AsyncCheckpointExecutor = ( + _ProcessBasedAsyncCheckpointExecutor() + if async_checkpointer_type == AsyncCheckpointerType.PROCESS + else _ThreadBasedAsyncCheckpointExecutor() + ) + + upload_future: Future = upload_executor.execute_save( + staging_future_or_state_dict, + checkpoint_id=checkpoint_id, + storage_writer=storage_writer, + planner=planner, + process_group=process_group, + no_dist=no_dist, + use_collectives=use_collectives, + ) + + if owned_async_stager is not None: + upload_future.add_done_callback(lambda _: maybe_close_owned_async_stager()) + # in the success path transfer cleanup ownership to the upload future + stack.pop_all() + + if isinstance(staging_future_or_state_dict, Future): + staging_future = staging_future_or_state_dict + return_staging_future: Future[None] = Future() + + def callback( + original_staging_future: Future[STATE_DICT_TYPE], + return_staging_future: Future[None] = return_staging_future, + ): + try: + original_staging_future.result() + return_staging_future.set_result(None) + except Exception as e: + return_staging_future.set_exception(e) + + if not staging_future.done(): + staging_future.add_done_callback(callback) + else: + return_staging_future.set_result(None) + + # return new AsyncSaveResponse for users using new ZOC implementation + return AsyncSaveResponse( + staging_completion=return_staging_future, upload_completion=upload_future + ) + else: + + @_dcp_method_logger(log_exceptions=True) + def maybe_synchronize_staging(): + if async_stager.should_synchronize_after_execute: + async_stager.synchronize_staging() + + maybe_synchronize_staging() + return upload_future + + +@_dcp_method_logger(log_exceptions=True) +def _stateful_to_state_dict(state_dict: STATE_DICT_TYPE) -> STATE_DICT_TYPE: + """Creates a shallow copy of `state_dict` where `state_dict` is called for each Stateful object.""" + stateful_state_dict = {} + for key, elem in state_dict.items(): + # Apply _dcp_method_logger to each state_dict() call + def _elem_to_state_dict(elem): + return elem.state_dict() if isinstance(elem, Stateful) else elem + + _elem_to_state_dict.__name__ = f"_stateful_to_state_dict.{key}" + + stateful_state_dict[key] = _dcp_method_logger(log_exceptions=True)( + _elem_to_state_dict + )(elem) + return stateful_state_dict + + +def _save_state_dict( + state_dict: STATE_DICT_TYPE, + storage_writer: StorageWriter, + process_group: dist.ProcessGroup | None = None, + coordinator_rank: int = 0, + no_dist: bool = False, + planner: SavePlanner | None = None, + use_collectives: bool = True, +) -> Metadata: + torch._C._log_api_usage_once("torch.distributed.checkpoint.save_state_dict") + + distW = _DistWrapper(process_group, not no_dist, coordinator_rank) + if planner is None: + planner = DefaultSavePlanner() + if planner is None: + raise AssertionError("planner is None") + + global_metadata = None + + ckpt_kwargs = {} + if (ckpt_id := getattr(storage_writer, "checkpoint_id", None)) is not None: + ckpt_kwargs["checkpoint_id"] = ckpt_id + ckpt_kwargs["process_group"] = distW.group + + @_dcp_method_logger(**ckpt_kwargs) + def local_step(): + if planner is None: + raise AssertionError("planner is None") + storage_meta = storage_writer.storage_meta() + if "storage_meta" not in inspect.signature(planner.set_up_planner).parameters: + warnings.warn( + "The function definition for SavePlanner.set_up_planner has been updated" + " to include the storage_meta argument. Please update your implementation" + " to include this parameter.", + stacklevel=2, + ) + planner.set_up_planner(state_dict, distW.is_coordinator) # type: ignore[call-arg, arg-type] + else: + planner.set_up_planner( + state_dict=state_dict, + storage_meta=storage_meta, + is_coordinator=distW.is_coordinator, + ) + + if ( + "kwargs" + in inspect.signature(storage_writer.set_up_storage_writer).parameters + ): + storage_writer.set_up_storage_writer( + distW.is_coordinator, + rank=distW.rank, + use_collectives=use_collectives, + ) + else: + storage_writer.set_up_storage_writer(distW.is_coordinator) + + local_plan = planner.create_local_plan() + local_plan = storage_writer.prepare_local_plan(local_plan) + return local_plan + + @_dcp_method_logger(**ckpt_kwargs) + def global_step(all_local_plans): + nonlocal global_metadata + + if planner is None: + raise AssertionError("planner is None") + all_local_plans, global_metadata = planner.create_global_plan(all_local_plans) + all_local_plans = storage_writer.prepare_global_plan(all_local_plans) + return all_local_plans + + central_plan: SavePlan | None = None + if use_collectives: + central_plan = distW.reduce_scatter("plan", local_step, global_step) + else: + local_plan: SavePlan = local_step() + global_plan: list[SavePlan] = global_step([local_plan]) + central_plan = global_plan[0] + + @_dcp_method_logger(**ckpt_kwargs) + def write_data(): + if planner is None: + raise AssertionError("planner is None") + if central_plan is None: + raise AssertionError("central_plan is None") + final_local_plan = planner.finish_plan(central_plan) + all_writes = storage_writer.write_data(final_local_plan, planner) + + all_writes.wait() + return all_writes.value() + + @_dcp_method_logger(**ckpt_kwargs) + def finish_checkpoint(all_results): + if global_metadata is None: + raise AssertionError("global_metadata is None") + storage_writer.finish(metadata=global_metadata, results=all_results) + return global_metadata + + if use_collectives: + metadata = distW.all_reduce("write", write_data, finish_checkpoint) + else: + write_results: list[WriteResult] = write_data() + metadata = finish_checkpoint([write_results]) + distW.barrier() + + return metadata diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/stateful.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/stateful.py new file mode 100644 index 0000000000000000000000000000000000000000..15e227d92fb5d29631b0316b3971c435120ad15b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/stateful.py @@ -0,0 +1,42 @@ +from typing import Any, TypeVar +from typing_extensions import Protocol, runtime_checkable + + +__all__ = ["Stateful", "StatefulT"] + + +@runtime_checkable +class Stateful(Protocol): + """ + Stateful protocol for objects that can be checkpointed and restored. + """ + + def state_dict(self) -> dict[str, Any]: + """ + Objects should return their state_dict representation as a dictionary. + The output of this function will be checkpointed, and later restored in + `load_state_dict()`. + + .. warning:: + Because of the inplace nature of restoring a checkpoint, this function + is also called during `torch.distributed.checkpoint.load`. + + + Returns: + Dict: The objects state dict + """ + + ... + + def load_state_dict(self, state_dict: dict[str, Any]) -> None: + """ + Restore the object's state from the provided state_dict. + + Args: + state_dict: The state dict to restore from + """ + + ... + + +StatefulT = TypeVar("StatefulT", bound=Stateful) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/storage.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/storage.py new file mode 100644 index 0000000000000000000000000000000000000000..0e0649f632b6e77e7da8665d99d4291b36a54a75 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/storage.py @@ -0,0 +1,288 @@ +import abc +import os +from dataclasses import dataclass +from typing import Any + +from torch.distributed.checkpoint.metadata import Metadata, MetadataIndex, StorageMeta +from torch.distributed.checkpoint.planner import ( + LoadPlan, + LoadPlanner, + SavePlan, + SavePlanner, +) +from torch.futures import Future + + +__all__ = ["WriteResult", "StorageWriter", "StorageReader"] + + +@dataclass(frozen=True) +class WriteResult: + index: MetadataIndex + + size_in_bytes: int + storage_data: Any + + +class StorageWriter(abc.ABC): + """ + Interface used by ``save_state_dict`` to write to storage. + + One StorageWriter instance acts as both the coordinator and the follower + in a distributed checkpoint. As part of initialization, each instance + is told its role. + + A subclass should expect the following sequence of calls. + + 0) (all ranks) set checkpoint_id if users pass a valid checkpoint_id. + 1) (all ranks) set_up_storage_writer() + 2) (all ranks) prepare_local_plan() + 3) (coordinator) prepare_global_plan() + 4) (all ranks) write_data() + 5) (coordinator) finish() + """ + + @abc.abstractmethod + def reset(self, checkpoint_id: str | os.PathLike | None = None) -> None: + """ + Calls to indicates a brand new checkpoint write is going to happen. + A checkpoint_id may be present if users set the checkpoint_id for + this checkpoint write. The meaning of the checkpoint_id is + storage-dependent. It can be a path to a folder/file or a key for + a key-value storage. + + Args: + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is a key-value store. + (Default: ``None``) + """ + ... + + @abc.abstractmethod + def set_up_storage_writer( + self, is_coordinator: bool, *args: Any, **kwargs: Any + ) -> None: + """ + Initialize this instance. + + Args: + is_coordinator (bool): Whether this instance is responsible for coordinating + the checkpoint. + """ + + @abc.abstractmethod + def prepare_local_plan(self, plan: SavePlan) -> SavePlan: + """ + Perform storage-specific local planning. + + While this method can produce a completely different plan, the recommended + way is to store storage specific data in SavePlan::storage_data. + + Args: + plan (SavePlan): The local plan from the ``SavePlanner`` in use. + + Returns: + A transformed ``SavePlan`` after storage local planning + """ + + @abc.abstractmethod + def prepare_global_plan(self, plans: list[SavePlan]) -> list[SavePlan]: + """ + Perform centralized planning of storage. + + This method is only called on the coordinator instance. + + While this method can produce a completely different plan, the preferred + way is to store storage specific data in SavePlan::storage_data. + + Args: + plans: A list of ``SavePlan`` instances, one for each rank. + + Returns: + A list of transformed ``SavePlan`` after storage global planning + """ + + @abc.abstractmethod + def write_data( + self, plan: SavePlan, planner: SavePlanner + ) -> Future[list[WriteResult]]: + """ + Write all items from ``plan`` using ``planner`` to resolve the data. + + A subclass should call ``SavePlanner::resolve_data`` on each item + from the plan to get access to the underlying object to write. + + Subclasses should lazily call `resolve_data` as it can allocate memory. + In case of tensors, make following assumptions: + + - They might be on any device, including not matching the one on ``WriteItem::tensor_data`` + - They might be views or not contiguous. Only the projection needs to be saved. + + Args: + plan (SavePlan): The save plan to execute. + planner (SavePlanner): Planner object to be used to resolve items to data. + + Returns: + A future that completes to a list of WriteResult + """ + + @abc.abstractmethod + def finish(self, metadata: Metadata, results: list[list[WriteResult]]) -> None: + """ + Write the metadata and marks the current checkpoint as successful. + + The actual format/schema used for serializing `metadata` is an + implementation detail. The only requirement is that it's recoverable + in to the same object graph. + + Args: + metadata (Metadata): metadata for the new checkpoint + results: A list of WriteResults from all ranks. + + Returns: + None + """ + + @classmethod + @abc.abstractmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + """ + Check if the given checkpoint_id is supported by the storage. This allows + us to enable automatic storage selection. + """ + ... + + def storage_meta(self) -> StorageMeta | None: + """ + Return the storage-specific metadata. This is used to store additional information + in a checkpoint that can be useful for providing request-level observability. StorageMeta + is passed to the ``SavePlanner`` during save calls. Returns None by default. + + TODO: provide an example + """ + return None + + +class StorageReader(abc.ABC): + """ + Interface used by ``load_state_dict`` to read from storage. + + One StorageReader instance acts as both the coordinator and the follower + in a distributed checkpoint. As part of initialization, each instance + is told its role. + + A subclass should expect the following sequence of calls by ``load_state_dict``: + + 0) (all ranks) set checkpoint_id if users pass a valid checkpoint_id. + 1) (all ranks) read_metadata() + 2) (all ranks) set_up_storage_reader() + 3) (all ranks) prepare_local_plan() + 4) (coordinator) prepare_global_plan() + 5) (all ranks) read_data() + """ + + @abc.abstractmethod + def reset(self, checkpoint_id: str | os.PathLike | None = None) -> None: + """ + Calls to indicates a brand new checkpoint read is going to happen. + A checkpoint_id may be present if users set the checkpoint_id for + this checkpoint read. The meaning of the checkpoint_id is + storage-dependent. It can be a path to a folder/file or a key for + a key-value storage. + + Args: + checkpoint_id (Union[str, os.PathLike, None]): + The ID of this checkpoint instance. The meaning of the checkpoint_id + depends on the storage. It can be a path to a folder or to a file. + It can also be a key if the storage is more like a key-value store. + (Default: ``None``) + """ + ... + + @abc.abstractmethod + def read_metadata(self, *args: Any, **kwargs: Any) -> Metadata: + """ + Read the checkpoint metadata. + + Returns: + The metadata object associated with the checkpoint being loaded. + + """ + + @abc.abstractmethod + def set_up_storage_reader( + self, metadata: Metadata, is_coordinator: bool, *args: Any, **kwargs: Any + ) -> None: + """ + Initialize this instance. + + Args: + metadata (Metadata): The metadata schema to use. + is_coordinator (bool): Whether this instance is responsible for coordinating + the checkpoint. + """ + + @abc.abstractmethod + def prepare_local_plan(self, plan: LoadPlan) -> LoadPlan: + """ + Perform storage-specific local planning. + + While this method can produce a completely different plan, the recommended + way is to store storage specific data in LoadPlan::storage_data. + + Args: + plan (LoadPlan): The local plan from the ``LoadPlan`` in use. + + Returns: + A transformed ``LoadPlan`` after storage local planning + """ + + @abc.abstractmethod + def prepare_global_plan(self, plans: list[LoadPlan]) -> list[LoadPlan]: + """ + Perform centralized planning of storage loading. + + This method is only called on the coordinator instance. + + While this method can produce a completely different plan, the preferred + way is to store storage specific data in LoadPlan::storage_data. + + Args: + plans: A list of ``LoadPlan`` instances, one for each rank. + + Returns: + A list of transformed ``LoadPlan`` after storage global planning + """ + + @abc.abstractmethod + def read_data(self, plan: LoadPlan, planner: LoadPlanner) -> Future[None]: + """ + Read all items from ``plan`` using ``planner`` to resolve the data. + + A subclass should call ``LoadPlanner::load_bytes`` to deserialize a BytesIO + object into the right place. + + A subclass should call ``LoadPlanner::resolve_tensor`` to get access to the + tensors that in should load data into. + + It's the StorageLayer responsibility to properly schedule any cross device copies + required. + + Args: + plan (LoadPlan): The local plan to execute on + planner (LoadPlanner): The planner object to use to resolve items. + + Returns: + A future that completes once all reads are finished. + """ + + @classmethod + @abc.abstractmethod + def validate_checkpoint_id(cls, checkpoint_id: str | os.PathLike) -> bool: + """ + Check if the given checkpoint_id is supported by the storage. This allows + us to enable automatic storage selection. + """ + ... diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..8ca9f43bdcc0f73ba6b12307e365e892ad20382f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/checkpoint/utils.py @@ -0,0 +1,490 @@ +# mypy: allow-untyped-defs +import cProfile +import inspect +import io +import itertools +import os +import warnings +from collections.abc import Callable, Sequence +from contextlib import contextmanager +from functools import wraps +from pstats import Stats +from typing import Any, cast, TypeVar + +import torch +import torch.distributed as dist +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed._shard.sharded_tensor.shard import Shard + +from .api import ( + _is_wrapped_exception, + _wrap_exception, + CheckpointException, + WRAPPED_EXCEPTION, +) +from .metadata import MetadataIndex, STATE_DICT_TYPE +from .protocol import _get_checkpointable_tensor_shard, _is_checkpointable_tensor + + +__all__ = ["find_tensor_shard", "find_state_dict_object"] + +T = TypeVar("T") +R = TypeVar("R") + + +def _get_failure_dict( + results: list[T | WRAPPED_EXCEPTION], +) -> dict[int, WRAPPED_EXCEPTION]: + return cast( + dict[int, WRAPPED_EXCEPTION], + {i: err for i, err in enumerate(results) if _is_wrapped_exception(err)}, + ) + + +def _all_gather_keys( + local_dict: dict[str, Any], group: dist.ProcessGroup | None = None +) -> set[str]: + """Gathers all keys, and returns them sorted.""" + keys = list(local_dict.keys()) + gathered_keys: list[list[str]] = [None] * dist.get_world_size(group) # type: ignore[list-item] + + dist.all_gather_object(gathered_keys, keys, group=group) + return set(itertools.chain.from_iterable(gathered_keys)) + + +def _assert_same_keys( + state_dict: dict[str, Any], process_group: dist.ProcessGroup | None = None +) -> None: + """ + Asserts that all ranks have the same keys in their state dict. + This is a collective call which requires all ranks in ``process_group`` to + join. It will also induce cross-rank communication and block CPU. + """ + + if dist.get_world_size(process_group) == 1: + return + + all_keys = _all_gather_keys(state_dict, process_group) + my_keys = set(state_dict.keys()) + diff = all_keys - my_keys + if len(diff) > 0: + raise AssertionError( + f"Key(s) present in other ranks but not this one, difference: {diff}" + ) + + +class _DistWrapper: + """ + This is a wrapper around PG that provides a series of features around object collectives. + + It works without distributed initialized, where most collectives turn into nops. + + All variants that take functions are exception robust, meaning that if one or more + ranks raise errors, all ranks will observe those. + """ + + def __init__( + self, + group: dist.ProcessGroup | None, + use_dist: bool, + coordinator_rank: int, + ): + self.group = group + self.use_dist = use_dist + self.coordinator_rank = coordinator_rank + if self.use_dist: + self.global_coordinator_rank = ( + dist.get_global_rank(group, coordinator_rank) + if group is not None + else coordinator_rank + ) + self.rank = dist.get_rank(group) + self.is_coordinator = self.rank == coordinator_rank + else: + self.global_coordinator_rank = 0 + self.rank = 0 + self.is_coordinator = True + + def get_rank(self) -> int: + return self.rank + + def get_world_size(self) -> int: + if self.use_dist: + return dist.get_world_size(self.group) + return 1 + + def broadcast_object(self, object: T | None) -> T: + """Implement functionality similar to c10d::broadcast_object_list but without distributed enabled.""" + object_list = [object] + if self.use_dist: + dist.broadcast_object_list( + object_list=object_list, + group=self.group, + src=self.global_coordinator_rank, + ) + return cast(T, object_list[0]) + + def gather_object(self, object: T) -> list[T] | None: + """Implement functionality similar to c10d::gather_object but without distributed enabled.""" + if self.use_dist: + gather_objs = ( + cast(list[T], [None] * dist.get_world_size(self.group)) + if self.is_coordinator + else None + ) + + dist.gather_object( + obj=object, + object_gather_list=gather_objs if self.is_coordinator else None, + dst=self.global_coordinator_rank, + group=self.group, + ) + result = gather_objs + else: + result = [object] + return result + + def all_gather_object(self, object: T) -> list[T]: + """Implement functionality similar to c10d::all_gather_object but without distributed enabled.""" + if self.use_dist: + gather_objs = cast(list[T], [None] * dist.get_world_size(self.group)) + + dist.all_gather_object( + object_list=gather_objs, obj=object, group=self.group + ) + else: + gather_objs = [object] + return gather_objs + + def scatter_object(self, object_list: list[T] | None) -> T: + """Implement functionality similar to c10d::scatter_object but without distributed enabled.""" + if self.use_dist: + gather_result = cast(list[T], [None]) + dist.scatter_object_list( + scatter_object_output_list=gather_result, + scatter_object_input_list=object_list if self.is_coordinator else None, + src=self.global_coordinator_rank, + group=self.group, + ) + + local_reply = gather_result[0] + else: + if object_list is None: + raise AssertionError("object_list is None") + local_reply = object_list[0] + return local_reply + + def reduce_scatter( + self, + step: str, + map_fun: Callable[[], T], + reduce_fun: Callable[[list[T]], list[R]], + ) -> R: + """ + Compute a value on each rank, then do centralized reduce on a single rank, followed by a scatter. + + This method operates in the following way: + Run ``map_fun`` on all ranks + Gather results on rank 0 + Call ``reduce_fun`` on all those values + Scatter to each rank part of the result. + """ + local_data: WRAPPED_EXCEPTION | T + try: + local_data = map_fun() + except BaseException as e: + local_data = _wrap_exception(e) + + all_data = self.gather_object(local_data) + all_results: list[R | CheckpointException] | None = None + if self.is_coordinator: + if all_data is None: + raise AssertionError("all_data is None") + node_failures = _get_failure_dict(all_data) + + if len(node_failures) == 0: + try: + # N.B. why can't mypy cast List[R] to List[Union[R, WRAPPED_EXCEPTION]]? + all_results = cast( + list[R | CheckpointException], + reduce_fun(cast(list[T], all_data)), + ) + except BaseException as e: + node_failures[self.rank] = _wrap_exception(e) + + if len(node_failures) > 0: + all_results = [ + CheckpointException(step, node_failures) + ] * self.get_world_size() + + result = self.scatter_object(all_results) + if isinstance(result, CheckpointException): + raise result + return result + + def all_reduce( + self, + step: str, + map_fun: Callable[[], T], + reduce_fun: Callable[[list[T]], R], + ) -> R: + """ + Compute a value on each rank, then do centralized reduce on a single rank, followed by a broadcast. + + This method operates in the following way: + Run ``map_fun`` on all ranks + Gather results on rank 0 + Call ``reduce_fun`` on all those values + Broadcast the reduced value to all ranks. + """ + local_data: T | WRAPPED_EXCEPTION + try: + local_data = map_fun() + except BaseException as e: + local_data = _wrap_exception(e) + + all_data = self.gather_object(local_data) + result: R | CheckpointException | None = None + if self.is_coordinator: + if all_data is None: + raise AssertionError("all_data is None") + node_failures = _get_failure_dict(all_data) + if len(node_failures) == 0: + try: + result = reduce_fun(cast(list[T], all_data)) + except BaseException as e: + node_failures[self.rank] = _wrap_exception(e) + + if len(node_failures) > 0: + result = CheckpointException(step, node_failures) + + # pyrefly: ignore [bad-argument-type] + final_result = self.broadcast_object(result) + if isinstance(final_result, CheckpointException): + raise final_result + # pyrefly: ignore [redundant-cast] + return cast(R, final_result) + + def all_gather( + self, + step: str, + map_fun: Callable[[], T], + ) -> list[T]: + """ + Compute a value on each rank, then all_gather them. + + This method operates in the following way: + Run ``map_cp`` on all ranks + all_gather the values to all ranks + """ + result: T | WRAPPED_EXCEPTION + try: + result = map_fun() + except BaseException as e: + result = _wrap_exception(e) + + all_results = self.all_gather_object(result) + + node_failures = _get_failure_dict(all_results) + if len(node_failures) > 0: + raise CheckpointException(step, node_failures) + return cast(list[T], all_results) + + def broadcast( + self, + step: str, + map_fun: Callable[[], T], + ) -> T: + """ + Compute a value on rank 0 and broadcast it. + + This method operates in the following way: + Run ``map_cp`` on rank 0 + broadcast the value + """ + result: T | CheckpointException | None = None + if self.is_coordinator: + try: + result = map_fun() + except BaseException as e: + result = CheckpointException(step, {self.rank: _wrap_exception(e)}) + # pyrefly: ignore [bad-argument-type] + final_result = self.broadcast_object(result) + if isinstance(final_result, CheckpointException): + raise final_result + # pyrefly: ignore [redundant-cast] + return cast(T, final_result) + + def barrier(self) -> None: + """ + Add a synchronization point across all processes when using distributed. + If torch.distributed is initialized, this function will invoke a barrier across the global process group. + If torch.distributed is not initialized, this function is a no-op. + """ + if not self.use_dist: + return + dist.barrier(group=self.group) + + +def _find_shard(tensor: ShardedTensor, index: MetadataIndex) -> Shard: + if index.offset is None: + raise ValueError( + f"Cannot lookup {index.fqn} since its a ShardedTensor and no offset was provided" + ) + + shards = tensor.local_shards() + # index fast path + if index.index is not None: + if ( + len(shards) > index.index + and torch.Size(shards[index.index].metadata.shard_offsets) == index.offset + ): + return shards[index.index] + + for shard in shards: + if torch.Size(shard.metadata.shard_offsets) == index.offset: + return shard + raise ValueError(f"Could not find shard at '{index.offset}' for FQN: '{index.fqn}'") + + +def find_tensor_shard(tensor: torch.Tensor, index: MetadataIndex) -> torch.Tensor: + if hasattr(tensor, "__get_tensor_shard__"): + # DTensor implements _Checkpointable + return tensor.__get_tensor_shard__(index) # type: ignore[attr-defined] + if _is_checkpointable_tensor(tensor): + return _get_checkpointable_tensor_shard(tensor, index) + if isinstance(tensor, ShardedTensor): + return _find_shard(tensor, index).tensor + if index.offset is not None: + # special case looking up a tensor by origin + if index.offset == torch.Size([0] * len(tensor.size())): + return tensor + raise ValueError( + f"FQN: '{index.fqn}' is not a ShardedTensor, can't find by offset: '{index.offset}'" + ) + return tensor + + +def find_state_dict_object(state_dict: STATE_DICT_TYPE, index: MetadataIndex) -> Any: + if index.fqn not in state_dict: + raise ValueError(f"Could not find FQN: '{index.fqn}'") + obj = state_dict[index.fqn] + + if isinstance(obj, torch.Tensor): + return find_tensor_shard(obj, index) + elif index.offset is not None: + raise ValueError( + f"FQN: '{index.fqn}' is not a ShardedTensor, can't find by offset: '{index.offset}'" + ) + return obj + + +def _element_wise_add(a: Sequence[int], b: Sequence[int]) -> list[int]: + return [i_a + i_b for i_a, i_b in zip(a, b)] + + +def _element_wise_sub(a: Sequence[int], b: Sequence[int]) -> list[int]: + return [i_a - i_b for i_a, i_b in zip(a, b)] + + +class _ReaderView(io.IOBase): + def __init__(self, base_stream: io.IOBase, offset: int, len: int): + super().__init__() + self.offset = offset + self.len = len + self.base_stream = base_stream + self.seek(0) + + def seek(self, offset: int, whence: int = os.SEEK_SET, /) -> int: + if whence == os.SEEK_SET: + offset = self.offset + offset + elif whence == os.SEEK_END: + whence = os.SEEK_SET + offset = (self.offset + self.len) - offset + return self.base_stream.seek(offset, whence) + + def tell(self) -> int: + return self.base_stream.tell() - self.offset + + def readable(self) -> bool: + return self.base_stream.readable() + + def seekable(self) -> bool: + return self.base_stream.seekable() + + def readinto(self, b): + max_size = self.len - self.tell() + if max_size == 0: + return 0 + if len(b) > max_size: + b = memoryview(b)[:max_size] + return self.base_stream.readinto(b) # type: ignore[attr-defined] + + def read(self, size=-1): + max_size = self.len - self.tell() + if size == -1 or size > max_size: + size = max_size + return self.base_stream.read(size) + + +def _create_file_view(file: io.IOBase, offset: int, length: int) -> io.IOBase: + # FIXME (kumpera) torch.load fails if we wrap with io.BufferedReader + return _ReaderView(file, offset, length) + + +def _normalize_device_info(device_type: str, device_id: int) -> str: + """Device info normalization.""" + if device_type == "cpu": + return "cpu" + return f"{device_type}:{device_id}" + + +# TODO: integrate with distributed logging flag +ENABLE_PROFILE = False + + +@contextmanager +def _profile(): + # Only log the profiling when it is enable and is on rank0 or dist is not + # available. + if ENABLE_PROFILE and (not dist.is_available() or dist.get_rank() == 0): + profiler = cProfile.Profile() + profiler.enable() + try: + yield + finally: + profiler.disable() + stats = Stats(profiler) + stats.sort_stats("time").print_stats(10) + else: + yield + + +def _api_bc_check(func): + @wraps(func) + def inner_func(*args, **kwargs) -> Any: + if len(args) == 2: + warnings.warn( + f"The argument order of {func.__name__} has been changed. " + "Please check the document to avoid future breakages.", + stacklevel=2, + ) + sig = inspect.signature(func) + kwonlyargs = [ + p.name for p in sig.parameters.values() if p.kind == p.KEYWORD_ONLY + ] + if "storage_writer" in kwonlyargs: + if "storage_writer" in kwargs: + raise AssertionError(f"storage_writer in kwargs: {(args, kwargs)}") + kwargs["storage_writer"] = args[1] + elif "storage_reader" in kwonlyargs: + if "storage_reader" in kwargs: + raise AssertionError(f"storage_reader in kwargs: {(args, kwargs)}") + kwargs["storage_reader"] = args[1] + else: + raise RuntimeError(f"Unexpected kwonlyargs = {kwonlyargs}") + return func(args[0], **kwargs) + else: + return func(*args, **kwargs) + + return inner_func diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/collective_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/collective_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..07d63d7f62e2a198ee8da0cd3f89910f18d7a499 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/collective_utils.py @@ -0,0 +1,349 @@ +#!/usr/bin/env python3 + + +""" +A set of primitive functions for performing collective ops. + +Each should also handle single rank scenario. +""" + +from __future__ import annotations + +import importlib +import logging +from collections import defaultdict +from dataclasses import dataclass +from typing import Any, cast, Generic, TYPE_CHECKING, TypeVar + + +if TYPE_CHECKING: + from collections.abc import Callable, Iterable + +import torch +import torch.distributed as dist + + +__all__: list[str] = [ + "SyncPayload", + "broadcast", + "all_gather", + "all_gather_object_enforce_type", +] + +logger = logging.getLogger(__name__) + +T = TypeVar("T") + + +@dataclass +class SyncPayload(Generic[T]): + stage_name: str | None + success: bool + payload: T + exception: Exception | None = None + + +def broadcast( + data_or_fn: T | Callable[[], T], + *, + success: bool = True, + stage_name: str | None = None, + rank: int = 0, + pg: dist.ProcessGroup | None = None, +) -> T: + """ + Broadcasts the data payload from rank 0 to all other ranks. + Or if a function is passed, execute it in rank 0 and broadcast result to all other ranks. + + Can be used to broadcast a failure signal to stop all ranks. + + If the function raises an exception, all ranks will raise. + + Args: + data_or_fn: the data to broadcast or function to execute and broadcast result. + success: False to stop all ranks. + stage_name: the name of the logical stage for synchronization and debugging + rank: rank to broadcast data or execute function and broadcast results. + pg: the process group for sync + Throws: + RuntimeError from original exception trace + Returns: + the value after synchronization + + Example usage: + >> id = broadcast(data_or_fn=allocate_id, rank=0, pg=ext_pg.my_pg) + """ + + if not success and data_or_fn is not None: + raise AssertionError( + "Data or Function is expected to be None if not successful" + ) + + payload: T | None = None + exception: Exception | None = None + # if no pg is passed then execute if rank is 0 + if (pg is None and rank == 0) or (pg is not None and pg.rank() == rank): + # determine if it is an executable function or data payload only + if callable(data_or_fn): + try: + payload = data_or_fn() + except Exception as e: + success = False + exception = e + else: + payload = data_or_fn + + # broadcast the exception type if any to all ranks for failure categorization + sync_obj = SyncPayload( + stage_name=stage_name, + success=success, + payload=payload, + exception=exception, + ) + + if pg is not None: + broadcast_list = [sync_obj] + dist.broadcast_object_list(broadcast_list, src=rank, group=pg) + if len(broadcast_list) != 1: + raise AssertionError( + f"Expected broadcast_list to have exactly 1 element, got {len(broadcast_list)}" + ) + sync_obj = broadcast_list[0] + + # failure in any rank will trigger a throw in every rank. + if not sync_obj.success: + error_msg = f"Rank {rank} failed" + if stage_name is not None: + error_msg += f": stage {sync_obj.stage_name}" + if sync_obj.exception is not None: + error_msg += f": exception {sync_obj.exception}" + + raise RuntimeError(error_msg) from sync_obj.exception + + return cast(T, sync_obj.payload) + + +def all_gather( + data_or_fn: T | Callable[[], T], + stage_name: str | None = None, + pg: dist.ProcessGroup | None = None, +) -> list[T]: + """ + A simple all_gather primitive with basic synchronization guard logic, + by checking payload from all ranks has the same stage name. + + Args: + data_or_fn: the data to be all gathered across ranks or function to be executed + stage_name: the sync stage name for out-of-sync protection + pg: the process group for sync + Throws: + RuntimeError from original exception trace + Returns: + a list of synced data from all ranks + + Example usage: + >> all_ids = all_gather(data_or_fn=allocate_id, pg=ext_pg.my_pg) + """ + payload: T | None = None + exception: Exception | None = None + success = True + # determine if it is an executable function or data payload only + if callable(data_or_fn): + try: + payload = data_or_fn() + except Exception as e: + success = False + exception = e + else: + payload = data_or_fn + + sync_obj = SyncPayload( + stage_name=stage_name, + success=success, + payload=payload, + exception=exception, + ) + + if pg is not None: + # List of success/failure across all ranks. + total_list = [None] * dist.get_world_size(pg) + all_gather_object_enforce_type(pg, total_list, sync_obj) + # Each rank will throw RuntimeError in case of failure on any rank. + stage_name = cast(SyncPayload[T], total_list[0]).stage_name + exception_list: list[tuple[int, Exception]] = [] + ret_list: list[T] = [] + error_msg: str = "" + + for i, sp in enumerate(cast(list[SyncPayload[T]], total_list)): + if sp.stage_name != stage_name: + error_msg += ( + f"Unexpected stage name received from rank {i}: {sp.stage_name} " + ) + continue + if not sp.success and sp.exception is not None: + exception_list.append((i, sp.exception)) + continue + ret_list.append(sp.payload) + + if len(exception_list) > 0: + raise RuntimeError( # type: ignore[misc] + error_msg, + exception_list, + ) from exception_list[0] # pyrefly: ignore [bad-raise] + return ret_list + else: + if not sync_obj.success: + raise RuntimeError( + f"all_gather failed with exception {sync_obj.exception}", + ) from sync_obj.exception + return [sync_obj.payload] # type: ignore[list-item] + + +# Note: use Any for typing for now so users can pass in +# either a list of None or target type placeholders +# otherwise pyre would complain +def all_gather_object_enforce_type( + pg: dist.ProcessGroup, + # pyre-fixme[2]: Parameter must have a type that does not contain `Any` + object_list: list[Any], + # pyre-fixme[2]: Parameter must have a type other than `Any` + obj: Any, + # pyre-fixme[2]: Parameter must have a type that does not contain `Any` + type_checker: Callable[[Any, Any], bool] = lambda x, y: type(x) is type(y), +) -> None: + """ + Similar to plain all_gather_object but with additional type checking + AFTER gather is done to ensure basic consistency. + If check does not pass, all ranks will fail with exception. + + This is generally to prevent conditional logic leading to + unexpected messages being received. This is considered fatal code error, + but due to logic stacks this might happen implicitly in practice. + + The default check does not check sub type (considered different) + or covariance (considered same) but users can pass in custom checker + if more complicated check is needed. + """ + dist.all_gather_object(object_list, obj, group=pg) + + # conservative check + list_len = len(object_list) + if list_len == 0: + return + first_obj = object_list[0] + for i in range(1, list_len): + if not type_checker(first_obj, object_list[i]): + raise TypeError( + f"Object type at index {i} is {type(object_list[i])}, " + f"while first object type is {type(first_obj)}" + ) + + +def _summarize_ranks(ranks: Iterable[int]) -> str: + ranks = sorted(ranks) + if min(ranks) < 0: + raise AssertionError("ranks should all be positive") + if len(set(ranks)) != len(ranks): + raise AssertionError("ranks should not contain duplicates") + curr: int | range | None = None + ranges = [] + while ranks: + x = ranks.pop(0) + if curr is None: + curr = x + elif isinstance(curr, int): + if x == curr + 1: + curr = range(curr, x + 1, 1) + else: + step = x - curr + curr = range(curr, x + step, step) + else: + if not isinstance(curr, range): + raise AssertionError("curr must be an instance of range") + if x == curr.stop: + curr = range(curr.start, curr.stop + curr.step, curr.step) + else: + ranges.append(curr) + curr = x + + if isinstance(curr, int): + ranges.append(range(curr, curr + 1, 1)) + elif isinstance(curr, range): + ranges.append(curr) + + result = [] + for r in ranges: + if len(r) == 1: + result.append(f"{r.start}") + elif r.step == 1: + result.append(f"{r.start}:{r.stop}") + else: + result.append(f"{r.start}:{r.stop}:{r.step}") + return ",".join(result) + + +def _check_philox_rng_sync( + generator: torch.Generator, group: dist.ProcessGroup +) -> tuple[dict[Any, set], str]: + local_state = generator.get_state() + all_states = [torch.empty_like(local_state) for _ in range(group.size())] + torch.distributed.all_gather(all_states, local_state) + seeds_offsets = [ + (state[:8].view(torch.uint64).item(), state[8:].view(torch.uint64).item()) + for state in all_states + ] + seed_offset_ranks = defaultdict(set) + for rank, (seed, offset) in enumerate(seeds_offsets): + seed_offset_ranks[(seed, offset)].add(rank) + return seed_offset_ranks, "(Seed, Offset)" + + +def _check_cpu_rng_sync( + generator: torch.Generator, group: dist.ProcessGroup +) -> tuple[dict[Any, set], str]: + # seed is returned as uint64_t from C impl, so may not fit in torch int64 tensor directly. + state_tensor = generator.get_state() + all_state_tensors = [torch.empty_like(state_tensor) for _ in range(group.size())] + torch.distributed.all_gather(all_state_tensors, state_tensor) + state_ranks = defaultdict(set) + for rank, state_tensor in enumerate(all_state_tensors): + # Summarize the state vector of the CPU rng. + # The properties that matter most are (1) its different if there is a state difference, (2) its printable + # (see desync table- not viable to print whole state vector of size 5k) + state_ranks[torch.hash_tensor(state_tensor).item()].add(rank) + return state_ranks, "Generator state hash" + + +def _check_rng_sync_internal( + generator: torch.Generator, group: dist.ProcessGroup +) -> tuple[dict[Any, set], str]: + if generator.device.type in {"cuda", "xpu"}: + return _check_philox_rng_sync(generator, group) + elif generator.device.type == "cpu": + return _check_cpu_rng_sync(generator, group) + else: + raise NotImplementedError( + f"Unsupported generator device: {generator.device.type}" + ) + + +def _desync_table_str(tag: str, value_ranks: dict[Any, set[int]]) -> str: + headers = ["Ranks", f"{tag} values"] + rank_values = [ + [_summarize_ranks(ranks), str(value)] for value, ranks in value_ranks.items() + ] + if importlib.util.find_spec("tabulate"): + from tabulate import tabulate + + return tabulate(rank_values, headers=headers) + row_str = "\n".join([str(row) for row in rank_values]) + return str(f"{headers}\n{row_str}") + + +def _check_rng_sync(generator: torch.Generator, group: dist.ProcessGroup) -> str | None: + value_ranks, value_header = _check_rng_sync_internal(generator, group) + log_str = None + if len(value_ranks) > 1: + log_str = f"Generator desync detected:\n{_desync_table_str(value_header, value_ranks)}" + logger.error(log_str) + return log_str diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/config.py b/.venv/lib/python3.12/site-packages/torch/distributed/config.py new file mode 100644 index 0000000000000000000000000000000000000000..ace990966688a4a182f9210ee2ebb1a52732a489 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/config.py @@ -0,0 +1,54 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +""" +Global configuration flags for torch.distributed +""" + +import sys +from typing import TYPE_CHECKING + +from torch.utils._config_module import Config, install_config_module + + +__all__ = ["compile_on_one_rank", "use_torchcomms", "pipeline_per_direction_p2p"] + +# Deprecated alias. The canonical flag now lives in torch.compiler.config -- it is read +# across the compiler stack (make_fx, inductor) not just by distributed. Kept here for +# back-compat (reads, writes, and .patch forward to the canonical flag). +compile_on_one_rank: bool = Config( + alias="torch.compiler.config.compile_on_one_rank", + deprecated=True, + deprecation_message="use torch.compiler.config.compile_on_one_rank instead", +) + +# When enabled, uses TorchComms for communication backend instead of the +# traditional ProcessGroup backends (NCCL, Gloo, etc.). +use_torchcomms: bool = Config( + default=False, + env_name_default="TORCH_DISTRIBUTED_USE_TORCHCOMMS", +) + +# When enabled, pipeline stages carry downstream (r -> r+1, forward activations) +# and upstream (r -> r-1, backward gradients) P2P on two separate communicators +# instead of sharing one. A single PP communicator serializes all send/recv in one +# FIFO: coalescing makes a single mixed batch deadlock-free, but across batches +# (pipeline skew, looped / V schedules, skip connections) the shared FIFO can +# still form a dependency cycle and deadlock. Splitting by direction removes that +# hazard and restores full-duplex bandwidth. Requires a device-bound default +# process group. +# +# This flag force-enables the behavior; it is auto-enabled when TorchComms is in +# use regardless of this flag (see PipelineStage), so it mainly matters for the +# non-TorchComms backends. +pipeline_per_direction_p2p: bool = Config( + default=False, + env_name_default="TORCH_DISTRIBUTED_PIPELINE_PER_DIRECTION_P2P", +) + + +if TYPE_CHECKING: + from torch.utils._config_typing import * # noqa: F403 + + +# adds patch, save_config, invalid config checks, etc +install_config_module(sys.modules[__name__]) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/constants.py b/.venv/lib/python3.12/site-packages/torch/distributed/constants.py new file mode 100644 index 0000000000000000000000000000000000000000..0a077bd6d4e5e5b614c3651e16286c41a814d983 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/constants.py @@ -0,0 +1,25 @@ +from datetime import timedelta + +from torch._C._distributed_c10d import _DEFAULT_PG_TIMEOUT + + +__all__ = ["default_pg_timeout", "default_pg_nccl_timeout"] + +# Default process group wide timeout, if applicable. +# This only applies to the non-nccl backends +# To make an attempt at backwards compatibility with THD, we use an +# extraordinarily high default timeout, given that THD did not have timeouts. +default_pg_timeout: timedelta = _DEFAULT_PG_TIMEOUT +# Separate timeout for PGNCCL mainly because it's always been that way in the C++ layer, but until recently +# there was one default that applied across all backends in the python layer. +# Later, we could consider merging them back together at the c++ layer if we can align on a same value. +# (only if TORCH_NCCL_BLOCKING_WAIT or TORCH_NCCL_ASYNC_ERROR_HANDLING is set to 1). + +try: + from torch._C._distributed_c10d import _DEFAULT_PG_NCCL_TIMEOUT + + default_pg_nccl_timeout: timedelta | None = _DEFAULT_PG_NCCL_TIMEOUT +except ImportError: + # if C++ NCCL support is not compiled, we don't have access to the default nccl value. + # if anyone is actually trying to use nccl in this state, it should error. + default_pg_nccl_timeout = None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..5d84089e1d87a449cfda775b33583b3cf6f0d4f6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__init__.py @@ -0,0 +1,148 @@ +import logging +import multiprocessing +import socket +from typing import Literal, TYPE_CHECKING + +# import for registration side effect +import torch.distributed.debug._handlers +from torch._C._distributed_c10d import _WorkerServer +from torch.distributed.debug._store import get_rank, tcpstore_client + + +if TYPE_CHECKING: + from torch.distributed.debug._frontend import DebugHandler + + +__all__ = [ + "start_debug_server", + "stop_debug_server", +] + +logger: logging.Logger = logging.getLogger(__name__) + +_WORKER_SERVER: _WorkerServer | None = None +_DEBUG_SERVER_PROC: multiprocessing.Process | None = None + + +def start_debug_server( + port: int = 25999, + worker_port: int = 0, + start_method: Literal["fork", "spawn", "forkserver"] | None = None, + dump_dir: str | None = None, + dump_interval: float = 60.0, + enabled_dumps: set[str] | None = None, + handlers: list["DebugHandler"] | None = None, + fetch_timeout: float = 60.0, +) -> None: + """ + Start the debug server stack on all workers. The frontend debug server is + only started on rank0 while the per rank worker servers are started on all + ranks. + + This server provides an HTTP frontend that allows for debugging slow and + deadlocked distributed jobs across all ranks simultaneously. This collects + data such as stack traces, FlightRecorder events, and performance profiles. + + This depends on dependencies which are not installed by default. + + Dependencies: + - Jinja2 + - aiohttp + + WARNING: This is intended to only be used in trusted network environments. + The debug server is not designed to be secure and should not be exposed to + the public internet. See SECURITY.md for more details. + + WARNING: This is an experimental feature and may change at any time. + + Args: + port (int): The port to start the frontend debug server on. + worker_port (int): The port to start the worker server on. Defaults to 0, which + will cause the worker server to bind to an ephemeral port. + start_method (str | None): The multiprocessing start method to use for the + frontend server process. One of "fork", "spawn", or "forkserver". + If None, uses the default start method. Using "spawn" is recommended + when using CUDA or when fork safety is a concern. + dump_dir (str | None): Directory to write periodic debug dumps to. If None, + periodic dumping is disabled. + dump_interval (float): Seconds between periodic dumps. Defaults to 60. + enabled_dumps (set[str] | None): Set of handler dump filenames to enable + (e.g. {"stacks", "fr_trace", "tcpstore"}). If None, all handlers that + implement dump() are enabled. + handlers (list[DebugHandler] | None): List of debug handlers to use. If None, + uses the default handlers. See torch.distributed.debug._handlers for + the default handlers. + fetch_timeout (float): Timeout in seconds for fetching data from individual + workers. Defaults to 60. Workers that don't respond within this time + will be reported as unavailable. + """ + global _WORKER_SERVER, _DEBUG_SERVER_PROC + + if _WORKER_SERVER is not None: + raise AssertionError("debug server already started") + if _DEBUG_SERVER_PROC is not None: + raise AssertionError("debug server already started") + + logger.info("Starting debug server on port %d", port) + + store = tcpstore_client() + + _WORKER_SERVER = _WorkerServer("::", worker_port) + + RANK = get_rank() + store.set(f"rank{RANK}", f"http://{socket.gethostname()}:{_WORKER_SERVER.port}") + + if RANK == 0: + from torch.distributed.debug._debug_handlers import default_handlers + from torch.distributed.debug._frontend import main + + if handlers is None: + handlers = default_handlers() + if enabled_dumps is None: + enabled_dumps = { + "stacks", + "fr_trace", + "torchcomms_health_check", + } + + main_kwargs = { + "port": port, + "dump_dir": dump_dir, + "dump_interval": dump_interval, + "enabled_dumps": enabled_dumps, + "handlers": handlers, + "fetch_timeout": fetch_timeout, + } + + if start_method is not None: + ctx = multiprocessing.get_context(start_method) + # pyre-ignore[16]: BaseContext has Process attribute at runtime + _DEBUG_SERVER_PROC = ctx.Process( + target=main, kwargs=main_kwargs, daemon=True + ) + else: + _DEBUG_SERVER_PROC = multiprocessing.Process( + target=main, kwargs=main_kwargs, daemon=True + ) + _DEBUG_SERVER_PROC.start() + + +def stop_debug_server() -> None: + """ + Shutdown the debug server and stop the frontend debug server process. + """ + global _WORKER_SERVER, _DEBUG_SERVER_PROC + + if _DEBUG_SERVER_PROC is None: + raise AssertionError + if _WORKER_SERVER is None: + raise AssertionError + + logger.info("Stopping debug server") + + _DEBUG_SERVER_PROC.terminate() + _WORKER_SERVER.shutdown() + _DEBUG_SERVER_PROC.join() + + _WORKER_SERVER = None + _DEBUG_SERVER_PROC = None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d6b2a4164cb52864dfcf512d44752c1e388fe935 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_debug_handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_debug_handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..43932e79fe9b3dd1d38386cad70d27dd3f214836 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_debug_handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_frontend.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_frontend.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3d1cbb482a5e79dade6d435cf3b833d86bc2d71d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_frontend.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f5a193b4bf801f5f421e7d6d2572305697e8dd3f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_store.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_store.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..96069197e90e441ab03b5dc85c4c6e0097fd635d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/debug/__pycache__/_store.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/_debug_handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_debug_handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..5b33fb1c822e8b61c970dde50d8d7045c5476458 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_debug_handlers.py @@ -0,0 +1,714 @@ +from __future__ import annotations + +import logging +from typing import TYPE_CHECKING + +from tabulate import tabulate + +from torch.distributed.debug._frontend import ( + DebugHandler, + fetch_all, + format_fetch_summary, + format_json, + NavLink, + Response, + Route, +) +from torch.distributed.debug._store import tcpstore_client +from torch.distributed.flight_recorder.components.builder import build_db +from torch.distributed.flight_recorder.components.config_manager import JobConfig +from torch.distributed.flight_recorder.components.types import ( + Collective, + Database, + Group, + Membership, + NCCLCall, +) + + +if TYPE_CHECKING: + from torch.distributed.debug._frontend import FrontendServer, HTTPRequestHandler + + +logger = logging.getLogger(__name__) + + +# --------------------------------------------------------------------------- +# Handler-specific templates +# --------------------------------------------------------------------------- + +INDEX_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}Index{% endblock %}

+{% endblock %} +{% block content %} +Hi +{% endblock %} + """ + +PYSPY_DUMP_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}py-spy Stack Traces{% endblock %}

+{% endblock %} +{% block content %} +
+ + + + + +
+ + {% for i, (addr, resp) in enumerate(zip(addrs, resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to fetch: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + 
{{ resp.text }}
+ {% endif %} + {% endfor %} +{% endblock %} + """ + +FR_TRACE_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}{{ title }}{% endblock %}

+{% endblock %} +{% block content %} + {% if fetch_summary %}
{{ fetch_summary }}
{% endif %} +

Groups

+ {{ groups | safe }} +

Memberships

+ {{ memberships | safe }} +

Collectives

+ {{ collectives | safe }} +

NCCL Calls

+ {{ ncclcalls | safe }} +{% endblock %} + """ + +PROFILE_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}torch.profiler{% endblock %}

+{% endblock %} + +{% block content %} +
+ + + +
+ + + + {% for i, (addr, resp) in enumerate(zip(addrs, resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to fetch: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + + + + {% endif %} + {% endfor %} +{% endblock %} + """ + +TCPSTORE_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}TCPStore Keys{% endblock %}

+{% endblock %} +{% block content %} + 
+    {% for k, v in zip(keys, values) -%}
+{{ k }}: {{ v | truncate(100) }}
+    {% endfor %}
+
+{% endblock %} + """ + + +# --------------------------------------------------------------------------- +# Handler classes +# --------------------------------------------------------------------------- + + +class IndexHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/", "Home")] + + def templates(self) -> dict[str, str]: + return {"index.html": INDEX_TEMPLATE} + + def _handle(self, req: HTTPRequestHandler) -> bytes: + return req.frontend.render_template("index.html") + + +class StacksHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/stacks", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/stacks", "Python Stack Traces")] + + def _handle(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all("dump_traceback", timeout=self.fetch_timeout) + return req.frontend.render_template( + "raw_resp.html", title="Stacks", addrs=addrs, resps=resps + ) + + def dump(self) -> str | None: + addrs, resps = fetch_all("dump_traceback", timeout=self.fetch_timeout) + parts: list[str] = [] + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + for i, (addr, resp) in enumerate(zip(addrs, resps)): + parts.append(f"=== Rank {i}: {addr} ===") + parts.append( + resp.text if resp.status_code == 200 else f"Error: {resp.status_code}" + ) + return "\n".join(parts) + + def dump_filename(self) -> str: + return "stacks" + + +class PySpyHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/pyspy_dump", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/pyspy_dump", "py-spy Stacks")] + + def templates(self) -> dict[str, str]: + return {"pyspy_dump.html": PYSPY_DUMP_TEMPLATE} + + def _handle(self, req: HTTPRequestHandler) -> bytes: + query = req.get_raw_query() + if "nonblocking" not in query: + query = f"nonblocking=1&{query}" if query else "nonblocking=1" + addrs, resps = fetch_all("pyspy_dump", query, timeout=self.fetch_timeout) + return req.frontend.render_template( + "pyspy_dump.html", + addrs=addrs, + resps=resps, + ) + + def dump(self) -> str | None: + addrs, resps = fetch_all( + "pyspy_dump", "nonblocking=1", timeout=self.fetch_timeout + ) + parts: list[str] = [] + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + for i, (addr, resp) in enumerate(zip(addrs, resps)): + parts.append(f"=== Rank {i}: {addr} ===") + parts.append( + resp.text if resp.status_code == 200 else f"Error: {resp.status_code}" + ) + return "\n".join(parts) + + def dump_filename(self) -> str: + return "pyspy_dump" + + +class FlightRecorderHandler(DebugHandler): + def routes(self) -> list[Route]: + return [ + Route("/fr_trace", self._handle_fr_trace), + Route("/fr_trace_json", self._handle_fr_trace_json), + Route("/fr_trace_nccl", self._handle_fr_trace_nccl), + Route("/fr_trace_nccl_json", self._handle_fr_trace_nccl_json), + ] + + def nav_links(self) -> list[NavLink]: + return [ + NavLink("/fr_trace", "FlightRecorder CPU"), + NavLink("/fr_trace_json", "(JSON)"), + NavLink("/fr_trace_nccl", "FlightRecorder NCCL"), + NavLink("/fr_trace_nccl_json", "(JSON)"), + ] + + def templates(self) -> dict[str, str]: + return {"fr_trace.html": FR_TRACE_TEMPLATE} + + @staticmethod + def _build_db(addrs: list[str], resps: list[Response]) -> Database: + config = JobConfig() + args = config.parse_args(args=[]) + args.allow_incomplete_ranks = True + args.verbose = True + + details = {} + for rank, resp in enumerate(resps): + if resp.status_code != 200: + continue + dump = { + "rank": rank, + "host_name": addrs[rank], + **resp.json(), + } + if "entries" not in dump: + dump["entries"] = [] + details[f"rank{rank}.json"] = dump + + if not details: + raise RuntimeError("All workers failed to respond") + + version = next(iter(details.values()))["version"] + # pyrefly: ignore [bad-argument-type] + return build_db(details, args, version) + + def _render_tables( + self, server: FrontendServer, addrs: list[str], resps: list[Response] + ) -> bytes: + db = self._build_db(addrs, resps) + return server.render_template( + "fr_trace.html", + title="FlightRecorder", + fetch_summary=format_fetch_summary(addrs, resps), + groups=tabulate(db.groups, headers=Group._fields, tablefmt="html"), + memberships=tabulate( + db.memberships, headers=Membership._fields, tablefmt="html" + ), + collectives=tabulate( + db.collectives, headers=Collective._fields, tablefmt="html" + ), + ncclcalls=tabulate(db.ncclcalls, headers=NCCLCall._fields, tablefmt="html"), + ) + + def _handle_fr_trace(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all("fr_trace_json", timeout=self.fetch_timeout) + return self._render_tables(req.frontend, addrs, list(resps)) + + def _handle_fr_trace_json(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all("fr_trace_json", timeout=self.fetch_timeout) + return req.frontend.render_template( + "json_resp.html", + title="FlightRecorder", + addrs=addrs, + resps=resps, + ) + + def _handle_fr_trace_nccl(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all( + "dump_nccl_trace_json", "onlyactive=true", timeout=self.fetch_timeout + ) + return self._render_tables(req.frontend, addrs, list(resps)) + + def _handle_fr_trace_nccl_json(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all( + "dump_nccl_trace_json", "onlyactive=true", timeout=self.fetch_timeout + ) + return req.frontend.render_template( + "json_resp.html", + title="FlightRecorder NCCL", + addrs=addrs, + resps=resps, + ) + + def dump(self) -> str | None: + parts = [] + + addrs, resps = fetch_all("fr_trace_json", timeout=self.fetch_timeout) + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + db = self._build_db(addrs, resps) + parts.extend( + [ + "=== FR Trace ===", + "--- Groups ---", + tabulate(db.groups, headers=Group._fields, tablefmt="plain"), + "--- Memberships ---", + tabulate(db.memberships, headers=Membership._fields, tablefmt="plain"), + "--- Collectives ---", + tabulate(db.collectives, headers=Collective._fields, tablefmt="plain"), + "--- NCCL Calls ---", + tabulate(db.ncclcalls, headers=NCCLCall._fields, tablefmt="plain"), + ] + ) + + try: + nccl_addrs, nccl_resps = fetch_all( + "dump_nccl_trace_json", "onlyactive=true", timeout=self.fetch_timeout + ) + nccl_db = self._build_db(nccl_addrs, nccl_resps) + parts.extend( + [ + "", + "=== FR Trace NCCL ===", + "--- Groups ---", + tabulate(nccl_db.groups, headers=Group._fields, tablefmt="plain"), + "--- Memberships ---", + tabulate( + nccl_db.memberships, + headers=Membership._fields, + tablefmt="plain", + ), + "--- Collectives ---", + tabulate( + nccl_db.collectives, + headers=Collective._fields, + tablefmt="plain", + ), + "--- NCCL Calls ---", + tabulate( + nccl_db.ncclcalls, + headers=NCCLCall._fields, + tablefmt="plain", + ), + ] + ) + except Exception: + parts.append("\n=== FR Trace NCCL ===\nFailed to fetch NCCL trace") + + return "\n".join(parts) + + def dump_filename(self) -> str: + return "fr_trace" + + +class ProfilerHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/profile", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/profile", "torch profiler")] + + def templates(self) -> dict[str, str]: + return {"profile.html": PROFILE_TEMPLATE} + + def _handle(self, req: HTTPRequestHandler) -> bytes: + duration = req.get_query_arg("duration", default=1.0, type=float) + addrs, resps = fetch_all( + "torch_profile", f"duration={duration}", timeout=self.fetch_timeout + ) + return req.frontend.render_template("profile.html", addrs=addrs, resps=resps) + + +class WaitCountersHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/wait_counters", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/wait_counters", "Wait Counters")] + + def _handle(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all("wait_counter_values", timeout=self.fetch_timeout) + return req.frontend.render_template( + "json_resp.html", title="Wait Counters", addrs=addrs, resps=resps + ) + + def dump(self) -> str | None: + addrs, resps = fetch_all("wait_counter_values", timeout=self.fetch_timeout) + parts: list[str] = [] + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + for i, (addr, resp) in enumerate(zip(addrs, resps)): + parts.append(f"=== Rank {i}: {addr} ===") + if resp.status_code == 200: + parts.append(format_json(resp.text)) + else: + parts.append(f"Error: {resp.status_code}") + return "\n".join(parts) + + def dump_filename(self) -> str: + return "wait_counters" + + +class TCPStoreHandler(DebugHandler): + def routes(self) -> list[Route]: + return [Route("/tcpstore", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/tcpstore", "TCPStore")] + + def templates(self) -> dict[str, str]: + return {"tcpstore.html": TCPSTORE_TEMPLATE} + + def _handle(self, req: HTTPRequestHandler) -> bytes: + store = tcpstore_client(prefix="") + keys = store.list_keys() + keys.sort() + values = [repr(v) for v in store.multi_get(keys)] + return req.frontend.render_template("tcpstore.html", keys=keys, values=values) + + def dump(self) -> str | None: + store = tcpstore_client(prefix="") + keys = store.list_keys() + keys.sort() + values = [repr(v) for v in store.multi_get(keys)] + parts = [f"{k}: {v}" for k, v in zip(keys, values)] + return "\n".join(parts) + + def dump_filename(self) -> str: + return "tcpstore" + + +class TorchCommsFlightRecorderHandler(DebugHandler): + """Handler for TorchComms FlightRecorder trace data.""" + + def routes(self) -> list[Route]: + return [ + Route("/torchcomms_fr_trace", self._handle_torchcomms_fr_trace), + Route("/torchcomms_fr_trace_json", self._handle_torchcomms_fr_trace_json), + ] + + def nav_links(self) -> list[NavLink]: + return [ + NavLink("/torchcomms_fr_trace", "TorchComms FR"), + NavLink("/torchcomms_fr_trace_json", "(JSON)"), + ] + + def templates(self) -> dict[str, str]: + return {"fr_trace.html": FR_TRACE_TEMPLATE} + + def _handle_torchcomms_fr_trace(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all( + "torchcomms_fr_trace_json", "onlyactive=true", timeout=self.fetch_timeout + ) + return self._render_tables(req.frontend, addrs, list(resps)) + + def _handle_torchcomms_fr_trace_json(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all( + "torchcomms_fr_trace_json", "onlyactive=true", timeout=self.fetch_timeout + ) + return req.frontend.render_template( + "json_resp.html", + title="TorchComms FlightRecorder", + addrs=addrs, + resps=resps, + ) + + def _render_tables( + self, server: FrontendServer, addrs: list[str], resps: list[Response] + ) -> bytes: + db = FlightRecorderHandler._build_db(addrs, resps) + return server.render_template( + "fr_trace.html", + title="TorchComms FlightRecorder", + fetch_summary=format_fetch_summary(addrs, resps), + groups=tabulate(db.groups, headers=Group._fields, tablefmt="html"), + memberships=tabulate( + db.memberships, headers=Membership._fields, tablefmt="html" + ), + collectives=tabulate( + db.collectives, headers=Collective._fields, tablefmt="html" + ), + ncclcalls=tabulate(db.ncclcalls, headers=NCCLCall._fields, tablefmt="html"), + ) + + def dump(self) -> str | None: + addrs, resps = fetch_all("torchcomms_fr_trace_json", timeout=self.fetch_timeout) + parts: list[str] = [] + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + db = FlightRecorderHandler._build_db(addrs, resps) + parts.extend( + [ + "=== TorchComms FR Trace ===", + "--- Groups ---", + tabulate(db.groups, headers=Group._fields, tablefmt="plain"), + "--- Memberships ---", + tabulate(db.memberships, headers=Membership._fields, tablefmt="plain"), + "--- Collectives ---", + tabulate(db.collectives, headers=Collective._fields, tablefmt="plain"), + "--- NCCL Calls ---", + tabulate(db.ncclcalls, headers=NCCLCall._fields, tablefmt="plain"), + ] + ) + return "\n".join(parts) + + def dump_filename(self) -> str: + return "torchcomms_fr_trace" + + +HEALTH_CHECK_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}TorchComms Health Check{% endblock %}

+{% endblock %} +{% block content %} +

Health Status

+ {% if fetch_summary %}
{{ fetch_summary }}
{% endif %} + {% for i, (addr, resp) in enumerate(zip(addrs, resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to fetch: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + 
{{ resp.text }}
+ {% endif %} + {% endfor %} + {% if dump_triggered %} +

Flight Recorder Dump (triggered by unhealthy rank)

+ {% for i, (addr, resp) in enumerate(zip(dump_addrs, dump_resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to dump: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + 
{{ resp.text }}
+ {% endif %} + {% endfor %} + {% endif %} +{% endblock %} + """ + + +class TorchCommsHealthCheckHandler(DebugHandler): + """Health check that detects watchdog timeouts and triggers FR dumps.""" + + def routes(self) -> list[Route]: + return [Route("/torchcomms_health_check", self._handle)] + + def nav_links(self) -> list[NavLink]: + return [NavLink("/torchcomms_health_check", "TorchComms Health")] + + def templates(self) -> dict[str, str]: + return {"torchcomms_health_check.html": HEALTH_CHECK_TEMPLATE} + + @staticmethod + def _any_unhealthy(resps: list[Response]) -> bool: + """Return True if any rank explicitly reports itself as unhealthy. + + Only responses with a 200 status code are inspected. Non-200 + responses (connection errors, timeouts, etc.) are intentionally + ignored so that a single unreachable rank does not trigger an + expensive Flight Recorder dump across the entire job. A dump is + only triggered when a rank is reachable *and* positively reports + ``{"healthy": false}``. + """ + for resp in resps: + if resp.status_code == 200: + try: + data = resp.json() + except Exception: + logger.exception( + "failed to parse health check response as JSON; " + "treating rank as healthy" + ) + continue + if not data.get("healthy", True): + return True + return False + + def _handle(self, req: HTTPRequestHandler) -> bytes: + addrs, resps = fetch_all("torchcomms_health_check", timeout=self.fetch_timeout) + dump_triggered = False + dump_addrs: list[str] = [] + dump_resps: list[Response] = [] + if self._any_unhealthy(resps): + dump_addrs, dump_resps = fetch_all( + "torchcomms_fr_dump_file", timeout=self.fetch_timeout + ) + dump_triggered = True + return req.frontend.render_template( + "torchcomms_health_check.html", + fetch_summary=format_fetch_summary(addrs, resps), + addrs=addrs, + resps=resps, + dump_triggered=dump_triggered, + dump_addrs=dump_addrs, + dump_resps=dump_resps, + ) + + def dump(self) -> str | None: + addrs, resps = fetch_all("torchcomms_health_check", timeout=self.fetch_timeout) + parts: list[str] = [] + summary = format_fetch_summary(addrs, resps) + if summary: + parts.append(summary) + parts.append("") + for i, (addr, resp) in enumerate(zip(addrs, resps)): + parts.append(f"=== Rank {i}: {addr} ===") + parts.append( + resp.text if resp.status_code == 200 else f"Error: {resp.status_code}" + ) + if self._any_unhealthy(resps): + parts.append("") + parts.append("=== Unhealthy rank detected, triggering FR dump ===") + dump_addrs, dump_resps = fetch_all( + "torchcomms_fr_dump_file", timeout=self.fetch_timeout + ) + for i, (addr, resp) in enumerate(zip(dump_addrs, dump_resps)): + parts.append(f"--- Rank {i}: {addr} ---") + parts.append( + resp.text + if resp.status_code == 200 + else f"Error: {resp.status_code}" + ) + return "\n".join(parts) + + def dump_filename(self) -> str: + return "torchcomms_health_check" + + +def default_handlers() -> list[DebugHandler]: + return [ + IndexHandler(), + StacksHandler(), + PySpyHandler(), + FlightRecorderHandler(), + TorchCommsFlightRecorderHandler(), + TorchCommsHealthCheckHandler(), + ProfilerHandler(), + WaitCountersHandler(), + TCPStoreHandler(), + ] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/_frontend.py b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_frontend.py new file mode 100644 index 0000000000000000000000000000000000000000..5a8779ef7424ea01aafc57fe5b8017d306040fa7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_frontend.py @@ -0,0 +1,497 @@ +import asyncio +import json +import logging +import os +import socket +import threading +import time +from abc import ABC, abstractmethod +from collections.abc import Callable +from concurrent.futures import ThreadPoolExecutor +from dataclasses import dataclass +from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer +from urllib.parse import parse_qs, urlparse + +from jinja2 import DictLoader, Environment + +from torch.distributed.debug._store import get_world_size, tcpstore_client + + +logger: logging.Logger = logging.getLogger(__name__) + +_DEFAULT_FETCH_TIMEOUT: float = 60.0 + + +# --------------------------------------------------------------------------- +# Base types +# --------------------------------------------------------------------------- + + +@dataclass(slots=True) +class Response: + status_code: int + text: str + + def raise_for_status(self): + if self.status_code != 200: + raise RuntimeError(f"HTTP {self.status_code}: {self.text}") + + def json(self): + return json.loads(self.text) + + +@dataclass(slots=True) +class NavLink: + path: str + label: str + + +@dataclass(slots=True) +class Route: + path: str + handler: Callable[["HTTPRequestHandler"], bytes] + + +class DebugHandler(ABC): + fetch_timeout: float = _DEFAULT_FETCH_TIMEOUT + + @abstractmethod + def routes(self) -> list[Route]: ... + + @abstractmethod + def nav_links(self) -> list[NavLink]: ... + + def templates(self) -> dict[str, str]: + return {} + + def dump(self) -> str | None: + return None + + def dump_filename(self) -> str: + return type(self).__name__.lower() + + +# --------------------------------------------------------------------------- +# Network helpers +# --------------------------------------------------------------------------- + + +def fetch_thread_pool(urls: list[str], timeout: float) -> list[Response]: + # late import for optional dependency + import requests + + max_workers = 20 + + def get(url: str) -> Response: + try: + resp = requests.post(url, timeout=timeout) + return Response(resp.status_code, resp.text) + except requests.exceptions.Timeout as e: + return Response(408, f"Timeout: {e}") + except requests.exceptions.ConnectionError as e: + return Response(503, f"ConnectionError: {e}") + except Exception as e: + return Response(502, f"{type(e).__name__}: {e}") + + with ThreadPoolExecutor(max_workers=max_workers) as executor: + resps = list(executor.map(get, urls)) + + return resps + + +def fetch_aiohttp(urls: list[str], timeout: float) -> list[Response]: + # late import for optional dependency + # pyrefly: ignore [missing-import] + import aiohttp + + async def fetch(session: aiohttp.ClientSession, url: str) -> Response: + try: + async with session.post(url) as resp: + text = await resp.text() + return Response(resp.status, text) + except asyncio.TimeoutError as e: + return Response(408, f"TimeoutError: {e}") + except aiohttp.ClientError as e: + return Response(503, f"{type(e).__name__}: {e}") + except Exception as e: + return Response(502, f"{type(e).__name__}: {e}") + + async def gather(urls: list[str]) -> list[Response]: + client_timeout = aiohttp.ClientTimeout(total=timeout) + async with aiohttp.ClientSession(timeout=client_timeout) as session: + return list(await asyncio.gather(*[fetch(session, url) for url in urls])) + + return asyncio.run(gather(urls)) + + +def fetch_all( + endpoint: str, args: str = "", *, timeout: float +) -> tuple[list[str], list[Response]]: + store = tcpstore_client() + keys = [f"rank{r}" for r in range(get_world_size())] + addrs = store.multi_get(keys) + addrs = [f"{addr.decode()}/handler/{endpoint}?{args}" for addr in addrs] + + try: + resps = fetch_aiohttp(addrs, timeout=timeout) + except ImportError: + resps = fetch_thread_pool(addrs, timeout=timeout) + + return addrs, resps + + +def format_json(blob: str): + parsed = json.loads(blob) + return json.dumps(parsed, indent=2) + + +def format_fetch_summary(addrs: list[str], resps: list[Response]) -> str | None: + """Return a summary string if any workers failed, or None if all succeeded.""" + failed = [(i, r) for i, r in enumerate(resps) if r.status_code != 200] + if not failed: + return None + total = len(addrs) + ok = total - len(failed) + lines = [f"PARTIAL DATA: {ok}/{total} workers responded"] + for rank, resp in failed: + lines.append(f" Rank {rank}: {resp.text}") + return "\n".join(lines) + + +# --------------------------------------------------------------------------- +# Template constants +# --------------------------------------------------------------------------- + + +BASE_TEMPLATE = """ + + + {% block title %}{% endblock %} - PyTorch Distributed + + + + + + + +
+ {% block header %}{% endblock %} + {% block content %}{% endblock %} +
+ """ + +RAW_RESP_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}{{title}}{% endblock %}

+{% endblock %} +{% block content %} + {% for i, (addr, resp) in enumerate(zip(addrs, resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to fetch: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + 
{{ resp.text }}
+ {% endif %} + {% endfor %} +{% endblock %} + """ + +JSON_RESP_TEMPLATE = """ +{% extends "base.html" %} +{% block header %} +

{% block title %}{{ title }}{% endblock %}

+{% endblock %} +{% block content %} + {% for i, (addr, resp) in enumerate(zip(addrs, resps)) %} +

Rank {{ i }}: {{ addr }}

+ {% if resp.status_code != 200 %} +

Failed to fetch: status={{ resp.status_code }}

+ 
{{ resp.text }}
+ {% else %} + 
{{ format_json(resp.text) }}
+ {% endif %} + {% endfor %} +{% endblock %} + """ + + +# --------------------------------------------------------------------------- +# PeriodicDumper +# --------------------------------------------------------------------------- + + +class PeriodicDumper: + def __init__( + self, + handlers: list[DebugHandler], + output_dir: str, + interval_seconds: float = 60.0, + ) -> None: + self._handlers = handlers + self._output_dir = output_dir + self._interval_seconds = interval_seconds + self._stop_event = threading.Event() + self._thread: threading.Thread | None = None + + def start(self) -> None: + os.makedirs(self._output_dir, exist_ok=True) + self._thread = threading.Thread( + target=self._run, + daemon=True, + name="distributed.debug.PeriodicDumper", + ) + self._thread.start() + + def stop(self) -> None: + self._stop_event.set() + if self._thread is not None: + self._thread.join() + + def _run(self) -> None: + while not self._stop_event.is_set(): + for handler in self._handlers: + try: + content = handler.dump() + except Exception: + logger.exception("Failed to dump %s", handler.dump_filename()) + continue + if content is None: + continue + timestamp = time.strftime("%Y%m%d_%H%M%S") + filename = f"{handler.dump_filename()}_{timestamp}.txt" + path = os.path.join(self._output_dir, filename) + try: + with open(path, "w") as f: + f.write(content) + except Exception: + logger.exception("Failed to write dump to %s", path) + self._stop_event.wait(self._interval_seconds) + + +# --------------------------------------------------------------------------- +# HTTP server +# --------------------------------------------------------------------------- + + +class _IPv6HTTPServer(ThreadingHTTPServer): + address_family: socket.AddressFamily = socket.AF_INET6 # pyre-ignore + request_queue_size: int = 1024 + + +class HTTPRequestHandler(BaseHTTPRequestHandler): + frontend: "FrontendServer" + + def log_message(self, format, *args): + logger.info( + "%s %s", + self.client_address[0], + format % args, + ) + + def do_GET(self): + self.frontend._handle_request(self) + + def get_path(self) -> str: + return urlparse(self.path).path + + def get_query(self) -> dict[str, list[str]]: + return parse_qs(self.get_raw_query()) + + def get_raw_query(self) -> str: + return urlparse(self.path).query + + def get_query_arg( + self, name: str, default: object = None, type: type = str + ) -> object: + query = self.get_query() + if name not in query: + return default + return type(query[name][0]) + + +class FrontendServer: + def __init__( + self, + port: int, + handlers: list[DebugHandler] | None = None, + ): + if handlers is None: + from torch.distributed.debug._debug_handlers import default_handlers + + handlers = default_handlers() + + # Build nav HTML from handlers + nav_html = "\n".join( + f' {link.label} ' + for handler in handlers + for link in handler.nav_links() + ) + + # Merge all handler templates + shared templates + all_templates: dict[str, str] = { + "base.html": BASE_TEMPLATE, + "raw_resp.html": RAW_RESP_TEMPLATE, + "json_resp.html": JSON_RESP_TEMPLATE, + } + for handler in handlers: + all_templates.update(handler.templates()) + + loader = DictLoader(all_templates) + self._jinja_env = Environment(loader=loader, enable_async=True) + self._jinja_env.globals.update( + zip=zip, + format_json=format_json, + enumerate=enumerate, + nav_links=nav_html, + ) + + # Build route table from handlers + self._routes: dict[str, Callable[[HTTPRequestHandler], bytes]] = {} + for handler in handlers: + for route in handler.routes(): + self._routes[route.path] = route.handler + + self._handlers = handlers + + # Create HTTP server + RequestHandlerClass = type( + "HTTPRequestHandler", + (HTTPRequestHandler,), + {"frontend": self}, + ) + + server_address = ("", port) + self._server = _IPv6HTTPServer(server_address, RequestHandlerClass) + + self._thread = threading.Thread( + target=self._serve, + args=(), + daemon=True, + name="distributed.debug.FrontendServer", + ) + self._thread.start() + + def _serve(self) -> None: + try: + self._server.serve_forever() + except Exception: + logger.exception("got exception in frontend server") + + def join(self) -> None: + self._thread.join() + + def _handle_request(self, req: HTTPRequestHandler) -> None: + path = req.get_path() + if path not in self._routes: + req.send_error(404, f"Handler not found: {path}") + return + + handler = self._routes[path] + try: + resp = handler(req) + # Catch SystemExit to not crash when FlightRecorder errors. + except (Exception, SystemExit) as e: + logger.exception( + "Exception in frontend server when handling %s", + path, + ) + req.send_error(500, f"Exception: {repr(e)}") + return + + req.send_response(200) + req.send_header("Content-type", "text/html") + req.end_headers() + req.wfile.write(resp) + + def render_template(self, template: str, **kwargs: object) -> bytes: + return self._jinja_env.get_template(template).render(**kwargs).encode() + + +def main( + port: int, + dump_dir: str | None, + dump_interval: float, + handlers: list[DebugHandler], + enabled_dumps: set[str], + fetch_timeout: float = 60.0, +) -> None: + for handler in handlers: + handler.fetch_timeout = fetch_timeout + + logger.setLevel(logging.INFO) + + server = FrontendServer(port=port, handlers=handlers) + logger.info("Frontend server started on port %d", server._server.server_port) + + dumper: PeriodicDumper | None = None + if dump_dir is not None: + dumper = PeriodicDumper( + [ + handler + for handler in handlers + if handler.dump_filename() in enabled_dumps + ], + dump_dir, + dump_interval, + ) + dumper.start() + logger.info( + "Periodic dumper started, writing to %s every %.0fs", + dump_dir, + dump_interval, + ) + + try: + server.join() + finally: + if dumper is not None: + dumper.stop() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/_handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..b8095c5b34bea5d2408ed87b21b541bf8966f4ad --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_handlers.py @@ -0,0 +1,23 @@ +import pathlib +import tempfile +import time + +from torch._C._distributed_c10d import _register_handler, _Request, _Response +from torch.profiler import _ExperimentalConfig, profile + + +def _torch_profile(req: _Request, resp: _Response) -> None: + experimental_config = _ExperimentalConfig( + profile_all_threads=True, + ) + duration = float(req.get_param("duration")) + with profile(record_shapes=True, experimental_config=experimental_config) as prof: + time.sleep(duration) + + with tempfile.NamedTemporaryFile(prefix="torch_debug", suffix=".json") as f: + prof.export_chrome_trace(f.name) + resp.set_content(pathlib.Path(f.name).read_bytes(), "application/json") + resp.set_status(200) + + +_register_handler("torch_profile", _torch_profile) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/debug/_store.py b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_store.py new file mode 100644 index 0000000000000000000000000000000000000000..487dd30abd6aff96d676ee3cf10d98490613e2a1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/debug/_store.py @@ -0,0 +1,25 @@ +import os + +import torch.distributed as dist + + +def get_rank() -> int: + return int(os.environ["RANK"]) + + +def get_world_size() -> int: + return int(os.environ["WORLD_SIZE"]) + + +def tcpstore_client(prefix: str = "debug_server") -> dist.Store: + MASTER_ADDR = os.environ["MASTER_ADDR"] + MASTER_PORT = int(os.environ["MASTER_PORT"]) + + store = dist.TCPStore( + host_name=MASTER_ADDR, + port=MASTER_PORT, + is_master=False, + ) + if prefix: + store = dist.PrefixStore(prefix, store) + return store diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/device_mesh.py b/.venv/lib/python3.12/site-packages/torch/distributed/device_mesh.py new file mode 100644 index 0000000000000000000000000000000000000000..e0bb4f2ab1977e8a9b7d6873b7c6c829e2adb9dd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/device_mesh.py @@ -0,0 +1,1776 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import logging +import os +import threading +import warnings +from collections.abc import Callable, Iterator +from itertools import zip_longest +from typing import Any, TYPE_CHECKING + +import torch +from torch._custom_class_base import CustomClassBase +from torch.distributed import is_available +from torch.distributed._mesh_layout import _FlatLayout, _MeshLayout +from torch.types import IntLikeType +from torch.utils._typing_utils import not_none + + +__all__ = ["init_device_mesh", "DeviceMesh"] + + +if not is_available(): + import sys + + # We need to create the stubs when distributed is not available. + # Otherwise, we would fail the doc tests (```./.ci/pytorch/docs-test.sh```), + # since it would try to import ``torch.distributed.device_mesh`` or + # ``torch.distributed.init_device_mesh`` but cannot find them. + + class _DeviceMeshStub: + pass + + def _init_device_mesh_stub(): + pass + + sys.modules["torch.distributed.device_mesh"].DeviceMesh = _DeviceMeshStub # type: ignore[attr-defined] + # pyrefly: ignore [missing-attribute] + sys.modules[ + "torch.distributed.device_mesh" + ].init_device_mesh = _init_device_mesh_stub # type: ignore[attr-defined] + + +else: + from torch._C._distributed_c10d import Backend as C10dBackend + from torch.distributed import config as dist_config + from torch.distributed.distributed_c10d import ( + _get_default_group, + _register_process_group_opaque_type, + _resolve_process_group, + get_backend, + get_process_group_ranks, + get_rank, + get_world_size, + GroupMember, + GroupName, + init_process_group, + is_initialized, + new_group, + ProcessGroup, + split_group, + ) + + logger = logging.getLogger(__name__) + + # only import numpy typing when type checking + if TYPE_CHECKING: + try: + from numpy.typing import ArrayLike + except ImportError: + logger.warning( + "DeviceMesh requires numpy >= 1.21 to be installed for type checking" + ) + + BackendConfig = tuple[str | None, C10dBackend.Options | None] + torch.serialization.add_safe_globals([_FlatLayout, _MeshLayout]) + + def _get_pg_from_name(mesh: "DeviceMesh", name: str) -> ProcessGroup: + """ + This method allows us to torch.compile through DeviceMesh and lift its + PGs as inputs to the graph since all PGs will have a source from the + DeviceMesh through the `_pg_registry`. + This will be moved to the DeviceMesh backend object once we separate + DeviceMesh into the frontend and backend. + """ + if torch.compiler.is_compiling(): + pg = mesh._pg_registry.get(name, None) + if pg is None: + raise RuntimeError( + f"PG {name} was not found while torch.compile tracing " + "This is probably because we pickle/unpickled a device mesh " + "before the PGs were created." + ) + return pg + else: + return _resolve_process_group(name) # pyrefly: ignore[bad-argument-type] + + def _maybe_traced_group(mesh: "DeviceMesh", mesh_dim: int) -> ProcessGroup | None: + """Resolve the group via an in-graph op when tracing under CooR. + + Under compile_on_one_rank and an active make_fx/proxy trace, return the + group as the output of _dtensor.mesh_get_process_group so the + ProcessGroup is extracted from the (graph-input) mesh at runtime instead + of being baked into the graph as an unserializable torchbind constant. + Returns None to take the eager path otherwise. Mirrors + _functional_collectives._resolve_group, but also covers eager collectives + (dist.all_reduce -> c10d.allreduce_) that bind the ProcessGroup directly. + + This is a module-level helper rather than a DeviceMesh method so that + Dynamo does not reject it as an unregistered opaque-object member. + """ + if not torch.compiler.config.compile_on_one_rank: + return None + # Under torch.compile, Dynamo lifts the ProcessGroup from the mesh's + # _pg_registry itself (see _get_pg_from_name); the op-emitting path here + # is only for make_fx tracing. is_compiling() folds to a constant under + # Dynamo, so this returns cleanly without tracing the op below. + if torch.compiler.is_compiling(): + return None + from torch.fx.experimental.proxy_tensor import get_proxy_mode + + # get_proxy_mode() is None in eager and inside the op's own fake/eager + # impl (the proxy mode is popped while dispatching), so this does not + # recurse. + if get_proxy_mode() is None: + return None + # Ensure the op is registered without a module-level circular import. + from torch.distributed.tensor import _collective_utils # noqa: F401 + + return torch.ops._dtensor.mesh_get_process_group(mesh, mesh_dim) + + class _MeshEnv(threading.local): + def __init__(self) -> None: + self.mesh_stack: list[DeviceMesh] = [] + + def get_current_mesh(self) -> "DeviceMesh": + if len(self.mesh_stack) == 0: + raise RuntimeError("No device mesh is currently active!") + return self.mesh_stack[-1] + + # TODO: to remove it once we move all use cases into new API. + def get_root_mesh(self, device_mesh: "DeviceMesh") -> "DeviceMesh": + # If a mesh could not be found in the child_to_root_mapping, it is a root mesh itself. + # A root mesh is not created through slicing. + # We considers the root mesh of a root mesh is itself. + # We keep this function for backward compatibility. + warnings.warn( + "This get_root_mesh API will be deprecated soon." + "Please use `get_root_mesh` inside DeviceMesh instead.", + stacklevel=2, + ) + if not device_mesh: + return device_mesh + return device_mesh._get_root_mesh() + + @staticmethod + def num_devices_per_host(device_type: str) -> int: + return _get_device_handle(device_type).device_count() + + @staticmethod + def num_hosts(device_type: str) -> int: + # ProcessGroup can't tell us this info so we have to infer it, assume + # homogeneous hardware for now + return get_world_size() // _MeshEnv.num_devices_per_host(device_type) + + # TODO: to remove it once we move all use cases into new API. + # We keep this API for backward compatibility. + def _get_all_submeshes( + self, device_mesh: "DeviceMesh", mesh_dim_name: str + ) -> list["DeviceMesh"]: + warnings.warn( + "This _get_all_submeshes API will be deprecated soon." + "Please use `_get_all_submeshes` inside DeviceMesh instead.", + stacklevel=2, + ) + return device_mesh._get_all_submeshes(mesh_dim_name) + + _mesh_resources: _MeshEnv = _MeshEnv() + + def _get_device_handle(device_type: str = "cuda"): + """ + Get the module corresponding to the device_type which is cuda or cuda-like device. + For example, when the device_type is cuda, the module `torch.cuda` is returned. + Return None when there is no corresponding module for device_type, otherwise + return the corresponding module. + """ + return getattr(torch, device_type, None) + + class DeviceMesh(CustomClassBase): + """ + DeviceMesh represents a mesh of devices, where layout of devices could be + represented as a n-d dimension array, and each value of the n-d dimensional + array is the global id of the default process group ranks. + + DeviceMesh could be used to setup the N dimensional device connections across the cluster, + and manage the ProcessGroups for N dimensional parallelisms. Communications could happen on + each dimension of the DeviceMesh separately. DeviceMesh respects the device that user selects + already (i.e. if user call `torch.cuda.set_device` before the DeviceMesh initialization), + and will select/set the device for the current process if user does not set the device + beforehand. Note that manual device selection should happen BEFORE the DeviceMesh initialization. + + DeviceMesh can also be used as a context manager when using together with DTensor APIs. + + .. note:: + DeviceMesh follows SPMD programming model, which means the same PyTorch Python program + is running on all processes/ranks in the cluster. Therefore, users need to make sure the + `mesh` array (which describes the layout of devices) should be identical across all ranks. + Inconsistent `mesh` will lead to silent hang. + + Args: + device_type (str): The device type of the mesh. Currently supports: "cpu", "cuda/cuda-like". + mesh (ndarray): A multi-dimensional array or an integer tensor describing the layout + of devices, where the IDs are global IDs of the default process group. + _rank (int): (experimental/internal) + The global rank of the current process. If not provided, it will + be inferred from the default process group. + + Returns: + DeviceMesh: A :class:`DeviceMesh` object representing the device layout. + + The following program runs on each process/rank in an SPMD manner. In this example, we have 2 + hosts with 4 GPUs each. + A reduction over the first dimension of mesh will reduce across + columns (0, 4), .. and (3, 7), a reduction over the second dimension + of mesh reduces across rows (0, 1, 2, 3) and (4, 5, 6, 7). + + Example:: + + >>> # xdoctest: +SKIP("no rank") + >>> from torch.distributed.device_mesh import DeviceMesh + >>> + >>> # Initialize device mesh as (2, 4) to represent the topology + >>> # of cross-host(dim 0), and within-host (dim 1). + >>> mesh = DeviceMesh(device_type="cuda", mesh=[[0, 1, 2, 3],[4, 5, 6, 7]]) + """ + + _rank: int + _device_type: str + _rank_map: torch.Tensor + _mesh_dim_names: tuple[str, ...] | None + _layout: _MeshLayout + _root_mesh: "DeviceMesh | None" = None + _thread_id: int | None + # Record flatten mesh name to its flattened mesh in root mesh. + _flatten_mapping: dict[str, "DeviceMesh"] + # Registry mapping group names to ProcessGroup objects (to avoid C++ lookup) + _pg_registry: dict[str, ProcessGroup] + + def __init__( + self, + device_type: str, + mesh: "torch.Tensor | ArrayLike | None" = None, + *, + mesh_dim_names: tuple[str, ...] | None = None, + backend_override: tuple[BackendConfig, ...] | None = None, + _init_backend: bool = True, + _rank: int | None = None, + _layout: _MeshLayout | None = None, + _rank_map: torch.Tensor | None = None, + _root_mesh: "DeviceMesh | None" = None, + ) -> None: + # no-op in OSS, logs API usage metrics in meta-internal runs + torch._C._log_api_usage_once( + "torch.distributed.device_mesh.DeviceMesh.__init__" + ) + if mesh is not None: + if _layout is not None or _rank_map is not None: + raise TypeError( + "Cannot provide _layout and/or _rank_map if passing explicit mesh" + ) + if isinstance(mesh, torch.Tensor) and mesh.device.type != "cpu": + mesh = mesh.to("cpu") + mesh_tensor = ( + mesh.detach().to(dtype=torch.int).contiguous() + if isinstance(mesh, torch.Tensor) + else torch.tensor(mesh, device="cpu", dtype=torch.int) + ) + _layout = _MeshLayout.from_sizes_strides( + tuple(mesh_tensor.size()), tuple(mesh_tensor.stride()) + ) + _rank_map = mesh_tensor.flatten() + else: + if _layout is None or _rank_map is None: + raise TypeError( + "The mesh argument is required except for PRIVATE USAGE ONLY!" + ) + + if not _layout.collapse().check_orthogonal(): + raise AssertionError( + "Please use a non-overlapping layout when creating a DeviceMesh." + ) + if _rank_map.ndim != 1: + raise AssertionError("The rank map must be 1-dimensional") + if not _rank_map.is_contiguous(): + raise AssertionError("The rank map must be contiguous") + if _rank_map.numel() < _layout.cosize(): + raise AssertionError( + f"The rank map contains {_rank_map.numel()} element, " + f"which isn't large enough for layout {_layout}" + ) + + self._device_type = device_type + self._layout = _layout + self._rank_map = _rank_map + self._mesh_dim_names = tuple(mesh_dim_names) if mesh_dim_names else None + self._root_mesh = _root_mesh + + if backend_override is None: + backend_override = ((None, None),) * len(self._layout) + elif len(backend_override) != len(self._layout): + raise ValueError( + f"backend_override should have the same length as the number of mesh dimensions, " + f"but got {len(backend_override)} and {len(self._layout)}." + ) + # Internal bookkeeping for the device mesh. + self._layout = ( + _layout + if _layout + else _MeshLayout.from_sizes_strides( + tuple(self.mesh.size()), tuple(self.mesh.stride()) + ) + ) + if not self._layout.collapse().check_orthogonal(): + raise AssertionError( + "Please use a non-overlapping layout when creating a DeviceMesh." + ) + # Because we still need to support slicing of flattened dim from root mesh, so we don't check stride here. + if self._layout.numel() != self.mesh.numel(): + raise AssertionError( + "Please use a valid layout when creating a DeviceMesh." + f"The layout {self._layout} is not consistent with the mesh size {self.mesh.size()}." + ) + + # private field to pre-generate DeviceMesh's hash + self._flatten_rank_map = tuple(self._rank_map.tolist()) + self._thread_id = None + # Initialize instance-specific flatten mapping + self._flatten_mapping = {} + # Initialize process group registry + self._pg_registry = {} + + # Skip process group initialization if xla device or init backend is False + # TODO(yeounoh) implement DeviceMesh backend and register XLA backend. + if device_type != "xla": + # always try to create default (world) pg, even if it is not initialized + # already. The world pg is used for device mesh identity (rank) on each + # process (we need to know if the current global rank is in the mesh or not). + if _init_backend: + self._setup_world_group_and_device() + self._dim_group_names = self._init_process_groups( + self._layout, + self._rank_map, + self._mesh_dim_names, + backend_override, + ) + # Populate the process group registry + # If we have a root mesh, add to root's registry for lookups + target_registry = ( + self._root_mesh._pg_registry + if self._root_mesh is not None + else self._pg_registry + ) + for name in self._dim_group_names: + pg = _resolve_process_group(name) + if pg is not None: + target_registry[name] = pg + + if is_initialized() and get_backend() == "threaded": + self._thread_id = threading.get_ident() + + # Now that the process group is initialized, we can get the rank + if _rank is None: + self._rank = get_rank() + else: + self._rank = _rank + + self._coordinate_on_dim = self._compute_coordinate_on_dim() + + self._hash: int | None = None + + @staticmethod + def _compute_coordinates_from_mesh( + mesh_tensor: torch.Tensor, + rank: int, + ) -> tuple[int, ...] | None: + """ + Compute the coordinates of a rank within a mesh tensor. + + Args: + mesh_tensor: The mesh tensor to search in + rank: The rank to find coordinates for + + Returns: + A tuple of coordinates if the rank is found in the mesh, None otherwise + + Raises: + AssertionError: If the rank appears more than once in the mesh + """ + rank_coords = (mesh_tensor == rank).nonzero() + if rank_coords.size(0) not in (0, 1): + raise AssertionError( + f"rank_coords.size(0) must be 0 or 1, got {rank_coords.size(0)}" + ) + + if rank_coords.size(0) == 0: + return None + + coords = rank_coords[0].tolist() + return tuple(coords) + + def _compute_coordinate_on_dim(self) -> tuple[int, ...] | None: + # calculate the coordinates of the current global rank on the mesh + return self._compute_coordinates_from_mesh(self.mesh, self._rank) + + def __getstate__(self) -> dict: + # Exclude _pg_registry from pickle since ProcessGroup objects can't be pickled + state = self.__dict__.copy() + state.pop("_pg_registry", None) + return state + + def __setstate__(self, state: dict) -> None: + self.__dict__.update(state) + # Reconstruct _pg_registry from _dim_group_names + self._pg_registry = {} + if hasattr(self, "_dim_group_names"): + for name in self._dim_group_names: + try: + pg = _resolve_process_group(name) + if pg is not None: + self._pg_registry[name] = pg + except RuntimeError: + # Note: process groups may not exist if loading in a different process + logger.warning( + "It seems like pickling/unpickling of the DeviceMesh " + "occurred before the PGs were created. This will cause PG " + "lookup to fail when torch.compile is enabled" + ) + + @property + def device_type(self) -> str: + """Returns the device type of the mesh.""" + return self._device_type + + @staticmethod + def _get_mesh_tensor_from_full_mesh( + full_mesh: torch.Tensor, + current_rank: int | None = None, + ) -> torch.Tensor: + if full_mesh.size(0) == 1: + return full_mesh[0] + + if current_rank is None: + current_rank = get_rank() + + my_coords = (full_mesh == current_rank).nonzero() + if my_coords.size(0) > 0: + return full_mesh[my_coords[0, 0]] + raise RuntimeError( + "In order to get the mesh Tensor of a DeviceMesh it needs to " + "either have all its original dimensions (e.g., no slicing) " + "or it needs to contain the local rank" + ) + + @property + def mesh(self) -> torch.Tensor: + """Returns the tensor representing the layout of devices.""" + full_mesh = self._layout.remap_to_tensor(self._rank_map) + return self._get_mesh_tensor_from_full_mesh(full_mesh) + + @property + def mesh_dim_names(self) -> tuple[str, ...] | None: + """Returns the names of mesh dimensions.""" + return self._mesh_dim_names + + def _setup_world_group_and_device(self): + default_initialized = is_initialized() + if not default_initialized: + init_process_group() + + world_size = get_world_size() + if self._layout.numel() > world_size: + raise RuntimeError( + f"Mesh should not be bigger than default world size {world_size}, but found {self._layout.numel()} ranks!" + ) + + # Skip device setup for fake backend (cross-compilation mode). + # The fake backend is used to simulate distributed training on a + # single process without actual devices, enabling compilation of + # GPU programs on CPU-only machines. + backend = get_backend() + if backend == "fake": + return _get_default_group() + + # ONLY set the device if the current device is not initialized, if user already + # set the device before DeviceMesh init, we respect the user's choice. + device_handle = _get_device_handle(self._device_type) + if device_handle and not device_handle.is_initialized(): + # auto set the cuda/cuda-like device only if user has not set it, if there's LOCAL_RANK + # env variable from launchers, we use it to set the device. + if "LOCAL_RANK" in os.environ: + local_rank = int(os.environ["LOCAL_RANK"]) + logger.info( + "Setting default device for the current process based on LOCAL_RANK=%s", + local_rank, + ) + device_handle.set_device(local_rank) + else: + # heuristic to set the current cuda/cuda-like device base on num of gpu devices available in each host + # NOTE: This device selection would only work for homogeneous hardware. + num_devices_per_host = device_handle.device_count() + # Skip device setup if no devices are available (cross-compilation mode) + if num_devices_per_host == 0: + return _get_default_group() + warnings.warn( + "It seems like you did not set/select the default device for the current process before the DeviceMesh " + "initialization or use a launcher (i.e. torchrun) which populates `LOCAL_RANK` environment variable. " + "It is recommended to set the current device for the process BEFORE the DeviceMesh initialization so that " + "the underlying communicator (i.e. NCCL) can be initialized properly. " + "Given that the current process has no default device selected, DeviceMesh will use a heuristic to set the " + "device_id via `global_rank % num_devices_per_host`, assuming homogeneous hardware cluster. ", + stacklevel=2, + ) + if ( + world_size > num_devices_per_host + and world_size % num_devices_per_host != 0 + ): + raise RuntimeError( + f"DeviceMesh only support homogeneous hardware, but found " + f"{world_size} ranks and {num_devices_per_host} {self._device_type} devices!" + ) + device_handle.set_device(get_rank() % num_devices_per_host) + + return _get_default_group() + + @staticmethod + def _init_one_process_group( + sub_layout: _FlatLayout, + rank_map: torch.Tensor, + dim_name: str, + backend_override: BackendConfig, + ) -> GroupName | None: + # Generate a 2D global mesh tensor for the current dim for PG creation. + pg_ranks_by_dim = _MeshLayout([sub_layout]).remap_to_tensor(rank_map) + backend, pg_options = backend_override + # We need to explicitly pass in timeout when specified in option, otherwise + # the default timeout will be used to override the timeout set in option. + # TODO: remove this once we have fixed inside c10d level. + timeout = pg_options._timeout if pg_options else None + + # If we have a 2D mesh with mesh_dim_names ("dp", "tp"), the group description + # of the subgroups would be `mesh_dim_dp` and `mesh_name_tp`. + # If the mesh doesn't have a mesh_dim_names, then the group description of the + # subgroup would be `mesh_dim_0` and `mesh_dim_1`. + group_desc = f"mesh_{dim_name}" + + dim_group = None + default_group = _get_default_group() + + # Early return if there is only one sub_layout in the mesh layout. + if sub_layout.numel() == get_world_size() and backend_override == ( + None, + None, + ): + # Append the default pg to the first dim groups only if the default pg is compatible with `self._device_type`. + # Otherwise, create new pg. + ranks = list(range(get_world_size())) + dim_group = ( + new_group( + backend=backend, + ranks=ranks, + group_desc="mesh_default", + ) + if torch.cuda.is_available() + and get_backend(default_group) == "gloo" + else default_group + ) + return dim_group.group_name # type: ignore[union-attr] + + # If bound_device_id exists, it means the nccl communicator has been eagerly initialized + # so that we can use `split_group` to create subgroups through `ncclCommSplit`. + # In this case, we only need to make one API call (`split_group``) for the subgroup creation + # for each mesh dimension. In a 2 * 4 mesh, we only need to make two API calls per ranks to create + # all the subgroups. + # Otherwise, we need to make more than one API call (`new_group`) for subgroup creations. The + # numbers of API calls are equal to the number of subgroups for each mesh dimension. In a 2 * 4 + # mesh, we need to make two API calls per ranks to create all the subgroups. + if ( + ( + getattr(default_group, "bound_device_id", None) is not None + or dist_config.use_torchcomms + ) + and torch.accelerator.is_available() + and ( + backend is None + or default_group._get_backend( + torch.accelerator.current_accelerator() # pyrefly: ignore[bad-argument-type] + ).name() + == backend + ) + ): + dim_group = split_group( + parent_pg=default_group, + timeout=timeout, + pg_options=pg_options, + split_ranks=pg_ranks_by_dim.tolist(), + group_desc=group_desc, + ) + if dim_group == GroupMember.NON_GROUP_MEMBER: + return None + return dim_group.group_name + + # If the subgroup has been already created through `split_group`, we simply loop over `pg_ranks_by_dim` + # and append the `group_name` to the `dim_group_names` list when the current rank is in the subgroup. + # Otherwise, we use `new_group` instead of `split_group` to create subgroups by looping over `pg_ranks_by_dim` + # along with appending information to the `dim_group_names` list whenever necessary. + # When torchcomms is enabled with a fake backend (e.g. disabled mesh dimensions), use hashed PG names so they + # stay consistent with the hash-based names produced by split_group for real backends. Sequential integer names + # from new_group are not resolvable from compiled code when mixed with split_group hash names. + use_hashed = dist_config.use_torchcomms and backend == "fake" + pg_name = None + for dim_mesh in pg_ranks_by_dim: + subgroup_ranks = dim_mesh.tolist() + dim_group = new_group( + ranks=subgroup_ranks, + timeout=timeout, + backend=backend, + pg_options=pg_options, + group_desc=group_desc, + use_local_synchronization=use_hashed, + ) + + # only add to dim_groups if the current rank in the subgroup + if get_rank() in subgroup_ranks: + if dim_group == GroupMember.NON_GROUP_MEMBER: + raise AssertionError( + f"Rank {get_rank()} was not included in process group {subgroup_ranks}" + ) + if pg_name is not None: + raise RuntimeError( + f"Each device mesh dimension should get only one process group, but got {get_rank()} " + f"in {subgroup_ranks}!" + ) + pg_name = dim_group.group_name + return pg_name + + @staticmethod + def _init_process_groups( + layout: _MeshLayout, + rank_map: torch.Tensor, + mesh_dim_names: tuple[str, ...] | None, + backend_override: tuple[BackendConfig, ...], + ) -> list[GroupName]: + # group_name associated with each mesh dimension, each + # mesh dimension should have one sub-group per rank + dim_group_names: list[GroupName | None] = [] + # create sub pgs base on the mesh argument specified + for dim in range(len(layout)): + dim_name = mesh_dim_names[dim] if mesh_dim_names else f"dim_{dim}" + dim_group_names.append( + DeviceMesh._init_one_process_group( + layout[dim], + rank_map, + dim_name, + backend_override[dim], + ) + ) + # Filter out None values. If any are None then they should all be None. + dim_non_none_group_names = [n for n in dim_group_names if n is not None] + if dim_non_none_group_names and len(dim_non_none_group_names) != len( + dim_group_names + ): + raise AssertionError + return dim_non_none_group_names + + def _get_root_mesh(self) -> "DeviceMesh": + return self._root_mesh if self._root_mesh else self + + def __enter__(self) -> "DeviceMesh": + # set this mesh as the current mesh in mesh env + _mesh_resources.mesh_stack.append(self) + return self + + # pyre-fixme[2]: Parameter must be annotated. + def __exit__(self, exc_type, exc_value, exc_traceback) -> None: + # pop this mesh from mesh env + _mesh_resources.mesh_stack.pop() + + def __repr__(self) -> str: + device_mesh_repr = ( + f"({', '.join(f'{k}={v}' for k, v in zip(self._mesh_dim_names, self._layout.top_level_sizes))})" + if self._mesh_dim_names + else f"{self._layout.top_level_sizes}" + ) + stride = tuple( + axis.stride[0] if len(axis.stride) == 1 else axis.stride + for axis in self._layout + ) + device_mesh_repr = ( + f"DeviceMesh({device_mesh_repr}, '{self.device_type}', stride={stride}" + ) + # We only print the mesh tensor if the debug mode is turned on. + if os.environ.get("TORCH_DISTRIBUTED_DEBUG", "") == "DETAIL": + device_mesh_repr += f", Mesh: {self.mesh.tolist()}" + return f"{device_mesh_repr})" + + def _hash_key(self) -> tuple[Any, ...]: + """Return the tuple used for hashing. Used by both __hash__ and _stable_hash.""" + return ( + self._flatten_rank_map, + self._layout, + self._device_type, + self._mesh_dim_names, + self._thread_id, + ) + + def __hash__(self): + # lazily compute hash + self._hash = getattr(self, "_hash", None) + if not self._hash: + self._hash = hash(self._hash_key()) + return self._hash + + def __eq__(self, other: object) -> bool: + if self is other: + return True + if not isinstance(other, DeviceMesh): + return False + return ( + self._flatten_rank_map == other._flatten_rank_map + and self._layout == other._layout + and self._device_type == other._device_type + and self._mesh_dim_names == other._mesh_dim_names + and self._thread_id == other._thread_id + ) + + def _stable_hash(self) -> str: + """ + Return a stable hash for AOT autograd caching. + [See note: Tensor subclass stable hashing for AOT autograd cache] + """ + import hashlib + + return hashlib.blake2b( + repr(self._hash_key()).encode(), digest_size=16 + ).hexdigest() + + def __getitem__(self, mesh_dim_names: str | tuple[str, ...]) -> "DeviceMesh": + """ + Slice the current DeviceMesh based on the mesh_dim_names given to create a submesh. + The submesh created consists of the dimensions and the communicators indicated by + ``mesh_dim_names`` + + Args: + mesh_dim_names (Union[str, tuple[str, ...]]): the name or the tuple of names of the + mesh dimension of the DeviceMesh to create the submesh for. + Returns: + A :class:`DeviceMesh` object + + The following program runs on each process/rank in an SPMD manner in a world size of 8. + In the first example: + Calling mesh_2d["tp"] on rank 0, 1, 2, 3 returns a 1D submesh of DeviceMesh:([0, 1, 2, 3]). + Calling mesh_2d["tp"] on rank 4, 5, 6, 7 returns a 1D submesh of DeviceMesh:([4, 5, 6, 7]). + Calling mesh_2d["dp"] on rank 0, 4 returns a 1D submesh of DeviceMesh:([0, 4]). + Calling mesh_2d["dp"] on rank 1, 5 returns a 1D submesh of DeviceMesh:([1, 5]). + Calling mesh_2d["dp"] on rank 2, 6 returns a 1D submesh of DeviceMesh:([2, 6]). + Calling mesh_2d["dp"] on rank 3, 7 returns a 1D submesh of DeviceMesh:([3, 7]). + + In the second example: + Calling mesh_3d["dp", "cp"] on rank 0, 1, 4, 5 returns a 2D submesh of DeviceMesh:([[0, 1], [4, 5]]). + Calling mesh_3d["dp", "cp"] on rank 2, 3, 6, 7 returns a 2D submesh of DeviceMesh:([[2, 3], [6, 7]]). + Calling mesh_3d["cp", "dp"] on rank 0, 1, 4, 5 returns a 2D submesh of DeviceMesh:([[0, 4], [1, 5]]). + Calling mesh_3d["cp", "dp"] on rank 2, 3, 6, 7 returns a 2D submesh of DeviceMesh:([[2, 6], [3, 7]]). + + Example:: + + >>> # xdoctest: +SKIP("no rank") + >>> from torch.distributed.device_mesh import DeviceMesh + >>> + >>> # Initialize a 2D device mesh as (2, 4) to represent the topology + >>> # of cross-host(dim 0), and within-host (dim 1). + >>> mesh_2d = init_device_mesh(device_type="cuda", (2,4), mesh_dim_names=("dp", "tp")) + >>> tp_mesh = mesh_2d["tp"] + >>> dp_mesh = mesh_2d["dp"] + >>> + >>> # Initialize a 3D mesh. + >>> mesh_3d = init_device_mesh(device_type="cuda", (2,2,2), mesh_dim_names=("dp", "pp", "cp")) + >>> # The order of the mesh_dim_names provided determines the order of dimensions in the submesh. + >>> dp_cp_mesh = mesh_3d["dp", "cp"] + >>> cp_dp_mesh = mesh_3d["cp", "dp"] + """ + if not self._mesh_dim_names: + raise RuntimeError("Cannot slice a DeviceMesh without mesh_dim_names!") + + mesh_dim_names = ( + (mesh_dim_names,) if isinstance(mesh_dim_names, str) else mesh_dim_names + ) + + if mesh_dim_names == self._mesh_dim_names: + return self + else: + sliced_mesh_layout = self._get_slice_mesh_layout(mesh_dim_names) + # When using FakeTensorMode to trace the model, `_create_sub_mesh()` will + # fail as it will require a real tensor to manipulate. + # `unset_fake_temporarily()` and `disable_proxy_modes_tracing()` + # will allow us to materialize the tensors within + # `_create_sub_mesh`, which should not affect modeling. + # + # Note that this should be orthogonal to torch.compile(). But whether + # we can compile device_mesh `slicing` (no graph break) is not verified + # yet and need a follow-up, + # TODO: compiler + device_mesh slicing. + from torch.fx.experimental.proxy_tensor import ( + disable_proxy_modes_tracing, + ) + + with ( + torch._subclasses.fake_tensor.unset_fake_temporarily(), + disable_proxy_modes_tracing(), + ): + submesh = self._create_sub_mesh(sliced_mesh_layout, mesh_dim_names) + return submesh + + def get_group(self, mesh_dim: int | str | None = None) -> ProcessGroup: + """ + Returns the single ProcessGroup specified by mesh_dim, or, if mesh_dim is not specified and the + DeviceMesh is 1-dimensional, returns the only ProcessGroup in the mesh. + + Args: + mesh_dim (str/int, optional): it can be the name of the mesh dimension or the index + of the mesh dimension. Default is None. + + Returns: + A :class:`ProcessGroup` object. + """ + if not hasattr(self, "_dim_group_names"): + raise RuntimeError("DeviceMesh process groups not initialized!") + + if len(self._layout) > 1 and mesh_dim is None: + raise RuntimeError( + f"Found the DeviceMesh have {len(self._layout)} dimensions", + "Optional kwarg `mesh_dim` needs to be specified when device_mesh.ndim > 1.", + "If you want to get the list of all the ProcessGroups in the DeviceMesh," + "please use `get_all_groups()` instead.", + ) + + root_mesh = self._get_root_mesh() + + # Quick return if the current device_mesh is a 1D mesh. + if len(self._layout) == 1 and mesh_dim is None: + traced = _maybe_traced_group(self, 0) + if traced is not None: + return traced + return not_none(_get_pg_from_name(root_mesh, self._dim_group_names[0])) + + root_to_flatten_mapping = root_mesh._flatten_mapping + if root_to_flatten_mapping and mesh_dim in root_to_flatten_mapping: + dim_group_name = root_to_flatten_mapping[ + mesh_dim # type: ignore[index] + ]._dim_group_names[0] + return not_none(_get_pg_from_name(root_mesh, dim_group_name)) + else: + mesh_dim = ( + self._get_mesh_dim_by_name(mesh_dim) + if isinstance(mesh_dim, str) + else mesh_dim + ) + if not isinstance(mesh_dim, int): + raise AssertionError( + f"mesh_dim must be an int, got {type(mesh_dim)}" + ) + traced = _maybe_traced_group(self, mesh_dim) + if traced is not None: + return traced + return not_none( + _get_pg_from_name(root_mesh, self._dim_group_names[mesh_dim]) + ) + + def get_all_groups(self) -> list[ProcessGroup]: + """ + Returns a list of ProcessGroups for all mesh dimensions. + + Returns: + A list of :class:`ProcessGroup` object. + """ + return [self.get_group(i) for i in range(len(self._layout))] + + def _create_sub_mesh( + self, + layout: _MeshLayout, + submesh_dim_names: tuple[str, ...], + ) -> "DeviceMesh": + with torch._dynamo.disable_nested_graph_breaks(): + root_mesh = self._get_root_mesh() + slice_dim_group_name = [] + dim_names = not_none(self._mesh_dim_names) + if len(self._dim_group_names) > 0: + if len(self._dim_group_names) != len(dim_names): + raise AssertionError( + "The number of dim_group_names and mesh_dim_names " + "should have the same length if the rank is in the mesh." + ) + for name in submesh_dim_names: + if name in dim_names: + slice_dim_group_name.append( + self._dim_group_names[dim_names.index(name)] + ) + else: + # If device_mesh is not root_mesh, we already throw error in _get_slice_mesh_layout + # Since we will deprecate the slicing of flattened dim_name from root mesh soon, + # we don't want to optimize the code furthermore. + flatten_mesh = self._flatten_mapping[name] + slice_dim_group_name.append( + flatten_mesh._dim_group_names[ + not_none(flatten_mesh._mesh_dim_names).index(name) + ] + ) + res_submesh = DeviceMesh( + self._device_type, + _layout=layout, + _rank_map=root_mesh._rank_map, + mesh_dim_names=submesh_dim_names, + _root_mesh=root_mesh, + _init_backend=False, + ) + res_submesh._dim_group_names = slice_dim_group_name + return res_submesh + + def _create_flatten_mesh( + self, + mesh_dim_name: str, + backend_override: BackendConfig = (None, None), + ) -> None: + """ + Creates a flattened mesh and stores it in the root mesh's _flatten_mapping. + This is a helper method for _flatten() that performs the actual mesh creation. + """ + root_mesh = self._get_root_mesh() + + # Check whether the mesh_dim_name for flattened mesh is valid. + invalid_dim_names = not_none(root_mesh._mesh_dim_names) + if mesh_dim_name in invalid_dim_names: + raise ValueError( + f"{mesh_dim_name} already exists for submesh of the {root_mesh}. ", + f"The mesh_dim_names of submesh and flattened mesh are {invalid_dim_names}. " + f"Please specify another valid mesh_dim_name.", + ) + + flattened_mesh_layout = _MeshLayout([self._layout.collapse()]) + + # Check if the flatten mesh has been created before. + if mesh_dim_name in root_mesh._flatten_mapping: + if ( + flattened_mesh_layout + == root_mesh._flatten_mapping[mesh_dim_name]._layout + ): + # Already exists with same layout, nothing to do + return + else: + raise ValueError( + f"Flatten mesh with mesh_dim_name {mesh_dim_name} has been created before, " + f"Please specify another valid mesh_dim_name." + ) + + res_flattened_mesh = DeviceMesh( + root_mesh._device_type, + _layout=flattened_mesh_layout, + _rank_map=root_mesh._rank_map, + mesh_dim_names=(mesh_dim_name,), + _root_mesh=root_mesh, + backend_override=(backend_override,), + ) + root_mesh._flatten_mapping[mesh_dim_name] = res_flattened_mesh + + def _get_root_mesh_dim(self) -> int | None: + """ + Returns the index of the mesh dim in the root mesh. + The device_mesh passed in needs to be sliced out from the root mesh + or submesh of the root mesh. + """ + root_mesh = self._get_root_mesh() + child_mesh_dim_names = self._mesh_dim_names + if root_mesh and child_mesh_dim_names: + if len(child_mesh_dim_names) != 1: + raise AssertionError("The submesh can only be a 1D mesh.") + child_mesh_dim_name = child_mesh_dim_names[0] + return root_mesh._get_mesh_dim_by_name(child_mesh_dim_name) + return None + + def _get_mesh_dim_by_name(self, mesh_dim_name: str) -> int: + if self._mesh_dim_names is None or len(self._mesh_dim_names) == 0: + raise KeyError( + "No `mesh_dim_names` found.", + ) + if mesh_dim_name not in self._mesh_dim_names: + raise KeyError( + f"Mesh dimension '{mesh_dim_name}' does not exist.", + f"Available mesh dimensions are: mesh_dim_names={self._mesh_dim_names}", + ) + return not_none(self._mesh_dim_names.index(mesh_dim_name)) + + def _get_slice_mesh_layout( + self, mesh_dim_names: tuple[str, ...] + ) -> _MeshLayout: + """ + Validate whether the mesh_dim_names is valid for slicing the given device_mesh. + If valid, return dim indexes of the slice mesh in the device mesh. + """ + slice_from_root = True + if self != self._get_root_mesh(): + slice_from_root = False + + # The slice mesh_dim_names should consist either the current device_mesh's mesh_dim_names + # or its flattened mesh's mesh_dim_names if it's root_mesh. + flatten_name_to_root_layout: dict[str, _FlatLayout] = ( + { + key: mesh._layout[0] # Extract the single axis from flattened mesh + for key, mesh in self._get_root_mesh()._flatten_mapping.items() + } + if slice_from_root + else {} + ) + valid_mesh_dim_names = [ + *not_none(self._mesh_dim_names), + *flatten_name_to_root_layout, + ] + + if not all( + mesh_dim_name in valid_mesh_dim_names + for mesh_dim_name in mesh_dim_names + ): + raise KeyError( + f"Invalid mesh_dim_names {mesh_dim_names} specified. " + f"Valid mesh_dim_names are {valid_mesh_dim_names}." + ) + + layout_sliced: list[_FlatLayout] = [] + for name in mesh_dim_names: + if name in not_none(self._mesh_dim_names): + layout_sliced.append( + self._layout[not_none(self._mesh_dim_names).index(name)] + ) + elif name in flatten_name_to_root_layout: + warnings.warn( + "Slicing a flattened dim from root mesh will be deprecated in PT 2.11. " + "Users need to bookkeep the flattened mesh directly. ", + stacklevel=2, + ) + layout_sliced.append(flatten_name_to_root_layout[name]) + result_layout = _MeshLayout(layout_sliced) + + # The check below is from DeviceMesh's implementation before adopting CuTe layout for internal + # bookkeeping and it can be removed but we need to define what is the expected behavior. + # TODO: Remove the below check and define the expected behavior. + # Validate the order of the slice mesh dim indices. + # This needs to be in ascending order. + if not result_layout.collapse().check_sorted(): + raise KeyError( + f"Invalid mesh_dim_names {mesh_dim_names} specified. " + "Mesh dim indices should be in ascending order." + ) + + # When users sliced dim_names outside from current mesh, we will check whether + # there is layout overlap. + # TODO: Eventually we will just directly throw error here because + # we will deprecate the slicing of flattened dim_name from root mesh. + if not result_layout.collapse().check_orthogonal(): + raise RuntimeError( + f"Slicing overlapping dim_names {mesh_dim_names} is not allowed." + ) + + return result_layout + + # TODO: to make this use case by other components public API in the future. + def _get_all_submeshes(self, mesh_dim_name: str) -> list["DeviceMesh"]: + """ + Return all the submeshes of a given mesh dimension of the device mesh. + """ + mesh_dim = self._get_mesh_dim_by_name(mesh_dim_name) + layout = self._layout[mesh_dim] + pg_ranks_by_dim = _MeshLayout([layout]).remap_to_tensor(self._rank_map) + cur_rank = self.get_rank() + res_submeshes = [] + for mesh_1d in pg_ranks_by_dim: + submesh = DeviceMesh( + self._device_type, + mesh_1d, + mesh_dim_names=(mesh_dim_name,), + _init_backend=False, + ) + submesh._dim_group_names = ( # type: ignore[has-type] + [self._dim_group_names[mesh_dim]] # type: ignore[has-type] + if cur_rank in mesh_1d + else [] + ) + res_submeshes.append(submesh) + + return res_submeshes + + @staticmethod + def from_group( + group: ProcessGroup | list[ProcessGroup], + device_type: str, + mesh: "torch.Tensor | ArrayLike | None" = None, + *, + mesh_dim_names: tuple[str, ...] | None = None, + ) -> "DeviceMesh": + """ + Constructs a :class:`DeviceMesh` with ``device_type`` from an + existing :class:`ProcessGroup` or a list of existing :class:`ProcessGroup`. + + The constructed device mesh has number of dimensions equal to the + number of groups passed. For example, if a single process group is passed in, + the resulted DeviceMesh is a 1D mesh. If a list of 2 process groups is passed in, + the resulted DeviceMesh is a 2D mesh. + + If more than one group is passed, then the ``mesh`` and ``mesh_dim_names`` arguments + are required. The order of the process groups passed in determines the topology of + the mesh. For example, the first process group will be the 0th dimension of the DeviceMesh. + The `mesh` tensor passed in must have the same number of dimensions as the number of process + groups passed in, and the order of the dimensions in the `mesh` tensor must match the order + in the process groups passed in. + + Args: + group (ProcessGroup or list[ProcessGroup]): the existing ProcessGroup + or a list of existing ProcessGroups. + device_type (str): The device type of the mesh. Currently supports: "cpu", + "cuda/cuda-like". Passing in a device type with a GPU index, such as "cuda:0", + is not allowed. + mesh (torch.Tensor or ArrayLike, optional): A multi-dimensional array or an + integer tensor describing the layout of devices, where the IDs are global IDs + of the default process group. Default is None. + mesh_dim_names (tuple[str, ...], optional): A tuple of mesh dimension names to assign + to each dimension of the multi-dimensional array describing the layout of devices. + Its length must match the length of `mesh_shape`. Each string in `mesh_dim_names` + must be unique. Default is None. + + Returns: + DeviceMesh: A :class:`DeviceMesh` object representing the device layout. + """ + + # 1D scenario + if isinstance(group, ProcessGroup): + group_ranks = get_process_group_ranks(group) + if ( + isinstance(mesh, torch.Tensor) and mesh.tolist() != group_ranks + ) or ( + mesh is not None + and not isinstance(mesh, torch.Tensor) + and mesh != group_ranks + ): + raise ValueError( + f"Invalid mesh {str(mesh)} for ProcessGroup with ranks {group_ranks}" + ) + mesh = torch.tensor(group_ranks, device="cpu", dtype=torch.int) + device_mesh = DeviceMesh( + device_type, + mesh, + mesh_dim_names=mesh_dim_names, + _init_backend=False, + ) + device_mesh._dim_group_names = [group.group_name] + device_mesh._pg_registry[group.group_name] = group + return device_mesh + + # nD scenario + groups = list(group) + if len(groups) == 0: + raise ValueError("Expects at least one ProcessGroup to be passed") + if mesh is None: + raise ValueError("Must pass mesh if passing multiple ProcessGroups") + if mesh_dim_names is None: + raise ValueError( + "Must pass mesh_dim_names if passing multiple ProcessGroups" + ) + # When init a DeviceMesh with multiple ProcessGroups directly, we need to make sure + # the mesh tensor is contiguous. Otherwise, the layout we inferred from the mesh tensor + # will have larger span than the actual tensor. This is just internal implementation detail + # and does not affect user facing behavior. + mesh = ( + mesh.detach().to(dtype=torch.int, device="cpu") + if isinstance(mesh, torch.Tensor) + else torch.tensor(mesh, device="cpu", dtype=torch.int) + ) + if mesh.ndim != len(groups): + raise ValueError( + "Expects mesh with ndim equal to number of ProcessGroups but got " + f"mesh {mesh.tolist()} and {len(groups)} ProcessGroups" + ) + device_mesh = DeviceMesh( + device_type, mesh, mesh_dim_names=mesh_dim_names, _init_backend=False + ) + device_mesh._dim_group_names = [group.group_name for group in groups] + for group in groups: + device_mesh._pg_registry[group.group_name] = group + return device_mesh + + def size(self, mesh_dim: int | None = None) -> int: + if mesh_dim is not None: + return self._layout[mesh_dim].numel() + return self._layout.numel() + + @property + def ndim(self) -> int: + return len(self._layout) + + @property + def shape(self) -> tuple[int, ...]: + return self._layout.top_level_sizes + + def get_rank(self) -> int: + """ + Returns the current global rank. + """ + return get_rank() + + def get_local_rank(self, mesh_dim: int | str | None = None) -> int: + """ + Returns the local rank of the given mesh_dim of the DeviceMesh. + + Args: + mesh_dim (str/int, optional): it can be the name of the mesh dimension or the index + of the mesh dimension. Default is None. + + Returns: + An integer denotes the local rank. + + The following program runs on each process/rank in an SPMD manner. In this example, we have 2 + hosts with 4 GPUs each. + Calling mesh_2d.get_local_rank(mesh_dim=0) on rank 0, 1, 2, 3 would return 0. + Calling mesh_2d.get_local_rank(mesh_dim=0) on rank 4, 5, 6, 7 would return 1. + Calling mesh_2d.get_local_rank(mesh_dim=1) on rank 0, 4 would return 0. + Calling mesh_2d.get_local_rank(mesh_dim=1) on rank 1, 5 would return 1. + Calling mesh_2d.get_local_rank(mesh_dim=1) on rank 2, 6 would return 2. + Calling mesh_2d.get_local_rank(mesh_dim=1) on rank 3, 7 would return 3. + + Example:: + + >>> # xdoctest: +SKIP("no rank") + >>> from torch.distributed.device_mesh import DeviceMesh + >>> + >>> # Initialize device mesh as (2, 4) to represent the topology + >>> # of cross-host(dim 0), and within-host (dim 1). + >>> mesh = DeviceMesh(device_type="cuda", mesh=[[0, 1, 2, 3],[4, 5, 6, 7]]) + """ + if self.ndim > 1 and mesh_dim is None: + raise RuntimeError( + f"Found the DeviceMesh have {len(self._layout)} dimensions", + "Optional kwarg `mesh_dim` needs to be specified when device_mesh.ndim > 1.", + ) + elif mesh_dim is None: + mesh_dim = 0 + + mesh_dim_group = not_none(self.get_group(mesh_dim)) + if not isinstance(mesh_dim_group, ProcessGroup): + raise AssertionError( + "We expect ProcessGroup before calling `get_rank`!" + ) + return not_none(get_rank(mesh_dim_group)) + + def _is_current_rank_part_of_mesh(self) -> bool: + """ + Return True if the current rank is part of this mesh. + + When a DeviceMesh is created with a subset of ranks (a sub-mesh), + ranks not included in the mesh are "non-participating ranks". These + ranks: + + - Return None from get_coordinate() + - Hold empty tensors as their local DTensor representation + - Skip computation during DTensor dispatch (returning default values) + - Skip collective operations during redistribute + - Return 0 cost for redistribute cost calculations + + This allows DTensor operations to execute correctly across the entire + process group while only performing actual work on participating ranks. + """ + return self._coordinate_on_dim is not None + + def get_coordinate(self) -> tuple[int, ...] | None: + """ + Return the relative indices of this rank relative to all + dimensions of the mesh. If this rank is not part of the mesh, return None. + """ + return self._coordinate_on_dim + + def _sym_get_coordinate(self, index: int) -> IntLikeType: + import torch.compiler.config as config + from torch._guards import detect_fake_mode + + if ( + not config.compile_on_one_rank + or not (fake_mode := detect_fake_mode()) + or not fake_mode.shape_env + ): + # This is only valid when the current rank is part of the mesh. + if self._coordinate_on_dim is None: + raise AssertionError + return self._coordinate_on_dim[index] + + # This will cause the ops to be registered - so don't let RUFF + # delete this import because it thinks it's unused... + from ._ops import device_mesh # noqa: F401 + + # Temporarily turn off tracing while we lift the constant + # rank_map to a list so it can be a constant in the graph. + with torch._subclasses.fake_tensor.unset_fake_temporarily(): + rank_map_list = self._rank_map.tolist() + rank_map = torch.tensor(rank_map_list, device="cpu", dtype=torch.int) + full_mesh = self._layout.remap_to_tensor(rank_map) + + return torch.ops.device_mesh._runtime_compute_coordinate_on_dim( + full_mesh, index + ) + + def _flatten( + self, + mesh_dim_name: str | None = None, + backend_override: str + | C10dBackend.Options + | tuple[str, C10dBackend.Options] + | None = None, + ) -> "DeviceMesh": + """ + Returns a 1D DeviceMesh by flattening the current DeviceMesh. + + If no mesh_dim_name is provided, the default is a string concatenating the mesh_dim_names of the + given submesh with each mesh_dim_name separated by "_". For example, if we have a 3D mesh + DeviceMesh([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], mesh_dim_names=("dp", "cp", "tp")), calling + mesh_3d["dp", "cp"]._flatten() will create a 1D submesh DeviceMesh([0, 2, 4, 6], mesh_dim_names=("dp_cp",)) + on rank 0, 2, 4, 6 and a 1D submesh DeviceMesh([1, 3, 5, 7], mesh_dim_names=("dp_cp",)) on rank 1, 3, 5, 7. + + After the flattened dimension is created, to access the flattened dimension in mesh_3d, one can use the + existing slicing method to obtain the flattened mesh through calling mesh_3d["dp_cp"]. + """ + if not self._mesh_dim_names: + raise RuntimeError( + "Cannot flatten a DeviceMesh without mesh_dim_names!" + ) + + # Compute mesh_dim_name first if not provided + if not mesh_dim_name: + mesh_dim_name = "_".join(not_none(self._mesh_dim_names)) + + # Flatten a 1D device mesh into its original mesh_dim_name will return itself. + if self.ndim == 1 and mesh_dim_name in not_none(self._mesh_dim_names): + return self + + if backend_override is not None: + (backend_override_tuple,) = _normalize_backend_override( + {0: backend_override}, 1 + ) + else: + backend_override_tuple = (None, None) + + self._create_flatten_mesh(mesh_dim_name, backend_override_tuple) + + return self._get_root_mesh()._flatten_mapping[mesh_dim_name] + + def _create_unflatten_mesh( + self, + dim: int, + mesh_sizes: tuple[int, ...], + mesh_dim_names: tuple[str, ...], + backend_override: tuple[ + tuple[str | None, C10dBackend.Options | None], ... + ] = ((None, None),), + ) -> "DeviceMesh": + inner_layout = _MeshLayout.from_sizes_strides(tuple(mesh_sizes)) + + if inner_layout.numel() != self._layout[dim].numel(): + raise ValueError( + f"The product of {mesh_sizes=} is {inner_layout.numel()}, " + f"but the original dimension at dim={dim} has size {self._layout[dim].numel()}. " + f"These must be equal for unflatten to work correctly." + ) + + partial_layout = self._layout[dim].composition(inner_layout) + unflattened_layout = self._layout.splice(dim, dim + 1, partial_layout) + unflattened_mesh_dim_names = list(not_none(self.mesh_dim_names)) + unflattened_mesh_dim_names[dim : dim + 1] = list(mesh_dim_names) + + root_mesh = self._get_root_mesh() + res_mesh = DeviceMesh( + self.device_type, + _layout=unflattened_layout, + _rank_map=root_mesh._rank_map, + mesh_dim_names=tuple(unflattened_mesh_dim_names), + _root_mesh=root_mesh, + _init_backend=False, + ) + + # If original mesh has initiated its backend, we need to initialize the backend + # of unflatten dims as well. + # TODO: To make backend init more efficient with cute layout representation and support + # per dim backend init. + if hasattr(self, "_dim_group_names"): + dim_group_names = self._dim_group_names.copy() + new_group_names = self._init_process_groups( + partial_layout, + root_mesh._rank_map, + mesh_dim_names, + backend_override, + ) + dim_group_names[dim : dim + 1] = new_group_names + res_mesh._dim_group_names = dim_group_names + # Populate root mesh's pg registry with new groups + for name in new_group_names: + pg = _resolve_process_group(name) + if pg is not None: + root_mesh._pg_registry[name] = pg + + return res_mesh + + def _unflatten( + self, + dim: int | str, + mesh_sizes: tuple[int, ...], + mesh_dim_names: tuple[str, ...], + backend_override: dict[ + str, str | C10dBackend.Options | tuple[str, C10dBackend.Options] + ] + | None = None, + ) -> "DeviceMesh": + """ + Returns a DeviceMesh by unflatten the current DeviceMesh. + + This api can be used to unflatten a N-D DeviceMesh into N-1+len(mesh_sizes)-D meshes or submeshes. + The dim is the dimension to be unflattened which can be either a string or an integer. + + The mesh_sizes is a tuple which specifies the shape of the mesh unflatten into for the given dim. + The mesh_dim_names is a list of strings which specifies the names of the dimensions of the mesh unflatten into. + Its length must match the length of mesh_sizes. + + For example, if we have a 1D mesh DeviceMesh([0, 1, 2, 3, 4, 5, 6, 7], mesh_dim_names=("world")), + calling mesh_1d._unflatten(0, (2, 2, 4), ["dp", "pp", "tp"]) will create a 3D mesh + DeviceMesh([[[0, 1], [2, 3]], [[4, 5], [6, 7]]], mesh_dim_names=("dp", "cp", "tp")). + + Note that after calling the unflatten, there is no access to the unflattened dimension in mesh_1d, one can only + use the newly unflattened mesh to slice out the unflattened mesh dims. + """ + if isinstance(dim, int) and dim >= self.ndim: + raise ValueError( + f"dim {dim} specified in `_unflatten` is out of range {self.ndim}" + ) + elif isinstance(dim, str) and dim not in not_none(self.mesh_dim_names): + raise ValueError( + f"dim {dim} specified in `_unflatten` is not in {self.mesh_dim_names}" + ) + + if len(mesh_sizes) != len(mesh_dim_names): + raise RuntimeError( + "mesh_dim_names must have same length as mesh_sizes in _unflatten!" + ) + + if isinstance(dim, str): + dim = not_none(self.mesh_dim_names).index(dim) + + if backend_override is not None: + backend_override_tuple = tuple( + _normalize_backend_override( + backend_override, # type: ignore[arg-type] + len(mesh_sizes), + mesh_dim_names, + ) + ) + else: + backend_override_tuple = ((None, None),) * len(mesh_dim_names) + + return self._create_unflatten_mesh( + dim, + mesh_sizes, + mesh_dim_names, + backend_override_tuple, + ) + + @staticmethod + def _concatenate(device_mesh_list: list["DeviceMesh"]) -> "DeviceMesh": + concat_dim_names: list[str] = [] + concat_axes: list[_FlatLayout] = [] + concat_dim_group_name: list[GroupName] = [] + flatten_rank_map = device_mesh_list[0]._flatten_rank_map + for dm in device_mesh_list: + concat_axes.extend(dm._layout) + concat_dim_names.extend(not_none(dm.mesh_dim_names)) + concat_dim_group_name.extend(not_none(dm._dim_group_names)) + # Concatenate device mesh having different root mesh tensors are meaningless + # because the concatenated indices should be indexed by the same root mesh tensor. + if dm._flatten_rank_map != flatten_rank_map: + raise RuntimeError( + "Cannot concatenate DeviceMeshes derived from different device meshes" + ) + concat_mesh_layout = _MeshLayout(concat_axes) + if not concat_mesh_layout.collapse().check_orthogonal(): + raise RuntimeError( + f"Cannot concatenate overlapping meshes: {device_mesh_list}" + ) + res_mesh = DeviceMesh( + device_mesh_list[0].device_type, + _layout=concat_mesh_layout, + _rank_map=device_mesh_list[0]._rank_map, + mesh_dim_names=tuple(concat_dim_names), + _root_mesh=device_mesh_list[0]._get_root_mesh(), + _init_backend=False, + ) + res_mesh._dim_group_names = concat_dim_group_name + return res_mesh + + def _normalize_backend_override( + backend_override: dict[ + int | str, + str | C10dBackend.Options | tuple[str, C10dBackend.Options], + ], + ndim: int, + mesh_dim_names: tuple[str, ...] | None = None, + ) -> Iterator[BackendConfig]: + if mesh_dim_names is None: + mesh_dim_names = () + for dim_idx, dim_name in zip_longest(range(ndim), mesh_dim_names): + if dim_name is not None and dim_name in backend_override: + if dim_idx in backend_override: + raise RuntimeError( + f"Found redundant dim index {dim_idx} and " + f"name {dim_name} in backend_override" + ) + val = backend_override.pop(dim_name) + elif dim_idx in backend_override: + val = backend_override.pop(dim_idx) + else: + yield (None, None) + continue + + if isinstance(val, str): + yield (val, None) + elif isinstance(val, C10dBackend.Options): + yield (None, val) + else: + yield val + + if backend_override: + raise RuntimeError( + f"Found invalid keys in backend_override: got {list(backend_override.keys())}, " + f"expected integers in range [0, {ndim}) or one of {mesh_dim_names}" + ) + + def init_device_mesh( + device_type: str, + mesh_shape: tuple[int, ...], + *, + mesh_dim_names: tuple[str, ...] | None = None, + backend_override: dict[ + int | str, str | C10dBackend.Options | tuple[str, C10dBackend.Options] + ] + | None = None, + ) -> DeviceMesh: + """ + Initializes a `DeviceMesh` based on `device_type`, `mesh_shape`, and `mesh_dim_names` parameters. + + This creates a DeviceMesh with an n-dimensional array layout, where `n` is the length of `mesh_shape`. + If `mesh_dim_names` is provided, each dimension is labeled as `mesh_dim_names[i]`. + + .. note:: + `init_device_mesh` follows SPMD programming model, meaning the same PyTorch Python program + runs on all processes/ranks in the cluster. Ensure `mesh_shape` (the dimensions of the nD array + describing device layout) is identical across all ranks. Inconsistent `mesh_shape` may lead to hanging. + + .. note:: + If no process group is found, init_device_mesh will initialize distributed process group/groups + required for distributed communications behind the scene. + + Args: + device_type (str): The device type of the mesh. Currently supports: "cpu", "cuda/cuda-like", "xpu". + Passing in a device type with a GPU index, such as "cuda:0", is not allowed. + mesh_shape (Tuple[int]): A tuple defining the dimensions of the multi-dimensional array + describing the layout of devices. + mesh_dim_names (tuple[str, ...], optional): A tuple of mesh dimension names to assign to each dimension + of the multi-dimensional array describing the layout of devices. Its length must match the length + of `mesh_shape`. Each string in `mesh_dim_names` must be unique. + backend_override (Dict[int | str, tuple[str, Options] | str | Options], optional): Overrides for some or all of + the ProcessGroups that will be created for each mesh dimension. Each key can be either the index of a + dimension or its name (if mesh_dim_names is provided). Each value can be a tuple containing the name + of the backend and its options, or just one of these two components (in which case the other will be + set to its default value). + + Returns: + DeviceMesh: A :class:`DeviceMesh` object representing the device layout. + + Example:: + + >>> # xdoctest: +SKIP("no rank") + >>> from torch.distributed.device_mesh import init_device_mesh + >>> + >>> mesh_1d = init_device_mesh("cuda", mesh_shape=(8,)) + >>> mesh_2d = init_device_mesh("cuda", mesh_shape=(2, 8), mesh_dim_names=("dp", "tp")) + + """ + if mesh_dim_names is not None: + if len(set(mesh_dim_names)) != len(mesh_dim_names): + raise RuntimeError( + "Each mesh_dim_name must be unique. " + f"Found repeated mesh_dim_name in mesh_dim_names {mesh_dim_names}" + ) + + if len(mesh_shape) != len(mesh_dim_names): + raise RuntimeError( + "mesh_shape and mesh_dim_names should have same length! " + f"Found len(mesh_dim_names): {len(mesh_dim_names)} and len(mesh_shape):{len(mesh_shape)}." + ) + + if backend_override is not None: + backend_override_tuple = tuple( + _normalize_backend_override( + backend_override, len(mesh_shape), mesh_dim_names + ) + ) + else: + backend_override_tuple = None + + # assume valid device types are all letters + if device_type and not device_type.isalpha(): + raise RuntimeError( + f"Device type with index is not supported but got {device_type}. ", + "If you maintained a 'torch.device' object, it's recommended to pass in 'device.type'.", + ) + + layout = _MeshLayout.from_sizes_strides(tuple(mesh_shape)) + # Always initialize the (identity) rank map on CPU, regardless of what the + # external device type has been set to be (e.g. meta) + with torch.device("cpu"): + rank_map = torch.arange(layout.numel(), dtype=torch.int) + device_mesh = DeviceMesh( + device_type=device_type, + _layout=layout, + _rank_map=rank_map, + mesh_dim_names=mesh_dim_names, + backend_override=backend_override_tuple, + ) + + return device_mesh + + +_distributed_opaque_types_registered = False + + +def _device_mesh_reconstruct_fn( + mesh: "CustomClassBase", + get_tracked_proxy: Callable[["CustomClassBase"], "torch.fx.Proxy | None"], + tracer: Any, +) -> "torch.fx.Proxy | None": + """Reconstruct a DeviceMesh submesh from a tracked ancestor mesh. + + Called by PythonKeyTracer when make_fx encounters a DeviceMesh that isn't + tracked (e.g. a submesh captured by a backward closure). Looks for any + tracked mesh that shares the same root and contains the target dim names, + then emits a call_function node that derives the submesh via _get_submesh. + """ + if not isinstance(mesh, DeviceMesh): + raise AssertionError("DeviceMesh expected") + + root_mesh = mesh._get_root_mesh() + + # Only submeshes can be reconstructed; root meshes must already be tracked. + if mesh is root_mesh: + return None + + dim_names = mesh._mesh_dim_names + if dim_names is None: + return None + + # Ensure the custom ops are registered + from torch.distributed._ops import device_mesh as _dm_ops # noqa: F401 + + # Try the root mesh first (original path). + ancestor_proxy = get_tracked_proxy(root_mesh) + ancestor_dim_names = root_mesh._mesh_dim_names + + # If root isn't tracked, search for any tracked DeviceMesh that shares + # the same root AND contains all our dim names. This handles the case + # where e.g. a concatenated (fsdp, tp) mesh is a graph input (from + # DTensor.__tensor_flatten__) but neither root nor the individual + # submeshes are tracked directly. + if ancestor_proxy is None: + from torch._library.fake_class_registry import FakeScriptObject + + for tracked_obj, proxy in tracer.opaque_tracker.items(): + real_obj = ( + tracked_obj.real_obj + if isinstance(tracked_obj, FakeScriptObject) + else tracked_obj + ) + if not isinstance(real_obj, DeviceMesh) or real_obj is mesh: + continue + if real_obj._get_root_mesh() is not root_mesh: + continue + tracked_dim_names = real_obj._mesh_dim_names + if tracked_dim_names is None: + continue + if all(n in tracked_dim_names for n in dim_names): + ancestor_proxy = proxy + ancestor_dim_names = tracked_dim_names + break + + if ancestor_proxy is None or ancestor_dim_names is None: + return None + + # Convert our dim names to indices into the ancestor mesh's dim names + mesh_dims = [ancestor_dim_names.index(n) for n in dim_names] + + # Dispatch through the custom op with proxy mode active so that + # meta["val"] is set and the result is tracked in opaque_tracker. + return torch.ops.device_mesh._get_submesh(ancestor_proxy, mesh_dims) + + +def _register_distributed_opaque_types(): + """ + Register DeviceMesh as an opaque type for torch.compile. + This must happen before any custom ops that use DeviceMesh in their schema. + Called lazily to avoid circular import issues. + """ + global _distributed_opaque_types_registered + if _distributed_opaque_types_registered: + return + _distributed_opaque_types_registered = True + + from torch._library.opaque_object import MemberType, register_custom_class + + _register_process_group_opaque_type() + + register_custom_class( + DeviceMesh, + typ="symbolic", + reconstruct_fn=_device_mesh_reconstruct_fn, + guard_fn=lambda obj: [ + obj._flatten_rank_map, + obj._layout, + obj._device_type, + obj._mesh_dim_names, + obj._thread_id, + ], + members={ + # USE_REAL: Evaluate these with the real object at compile time + # and bake the result as a constant + "_flatten_rank_map": MemberType.USE_REAL, + "_layout": MemberType.USE_REAL, + "_device_type": MemberType.USE_REAL, + "_mesh_dim_names": MemberType.USE_REAL, + "_thread_id": MemberType.USE_REAL, + "get_rank": MemberType.USE_REAL, + "size": MemberType.USE_REAL, + "get_coordinate": MemberType.USE_REAL, + "get_local_rank": MemberType.USE_REAL, + "__eq__": MemberType.USE_REAL, + "__ne__": MemberType.USE_REAL, + "ndim": MemberType.USE_REAL, + "shape": MemberType.USE_REAL, + "mesh_dim_names": MemberType.USE_REAL, + "get_group": MemberType.INLINED, + "_pg_registry": MemberType.INLINED, + "_hash": MemberType.USE_REAL, + "_create_flatten_mesh": MemberType.USE_REAL, + "_coordinate_on_dim": MemberType.USE_REAL, + "_dim_group_names": MemberType.USE_REAL, + "_flatten_mapping": MemberType.USE_REAL, + "_rank_map": MemberType.USE_REAL, + "_root_mesh": MemberType.USE_REAL, + "device_type": MemberType.USE_REAL, + "mesh": MemberType.USE_REAL, + "_flatten": MemberType.INLINED, + "_unflatten": MemberType.USE_REAL, + "_is_current_rank_part_of_mesh": MemberType.USE_REAL, + "_sym_get_coordinate": MemberType.USE_REAL, + "_get_mesh_dim_by_name": MemberType.USE_REAL, + "_get_root_mesh": MemberType.INLINED, + "__getitem__": MemberType.INLINED, + "_get_slice_mesh_layout": MemberType.INLINED, + "_create_sub_mesh": MemberType.INLINED, + }, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py b/.venv/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py new file mode 100644 index 0000000000000000000000000000000000000000..0734c8ead6676377c77e90e36a2bcaceceddc799 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py @@ -0,0 +1,8055 @@ +# mypy: allow-untyped-defs +"""Distributed Collective Communication (c10d).""" + +import builtins +import collections.abc +import contextlib +import ctypes +import hashlib +import io +import itertools +import logging +import operator +import os +import pickle +import sys +import time +import traceback +import types +import warnings +from collections.abc import Callable, Sequence +from datetime import timedelta +from typing import ( + cast, + Final, + Literal, + NamedTuple, + NewType, + overload, + Protocol, + TYPE_CHECKING, + TypeAlias, + TypeVar, +) +from typing_extensions import deprecated, NotRequired, TypedDict, TypeIs + +import torch +from torch._C import _DistStoreError as DistStoreError +from torch._C._distributed_c10d import ( + _DistributedBackendOptions, + _register_process_group, + _resolve_process_group, + _unregister_all_process_groups, + _unregister_process_group, + AllgatherOptions, + AllreduceCoalescedOptions, + AllreduceOptions, + AllToAllOptions, + Backend as C10DBackend, + BarrierOptions, + BroadcastOptions, + DebugLevel, + GatherOptions, + get_debug_level, + NanCheckHook, + PrefixStore, + ProcessGroup, + ReconfigureOptions, + ReduceOp, + ReduceOptions, + ReduceScatterOptions, + ScatterOptions, + Store, + Window, + Work, +) +from torch._utils_internal import set_pytorch_distributed_envs_from_justknobs +from torch.monitor import _WaitCounter +from torch.overrides import handle_torch_function, has_torch_function +from torch.utils._typing_utils import not_none + +from . import config as dist_config +from .c10d_logger import _exception_logger, _time_logger +from .constants import default_pg_nccl_timeout, default_pg_timeout +from .rendezvous import register_rendezvous_handler, rendezvous # noqa: F401 + + +def _register_process_group_opaque_type() -> None: + from torch._library.opaque_object import ( + is_custom_class, + MemberType, + register_custom_class, + ) + + if is_custom_class(ProcessGroup): + return + + register_custom_class( + ProcessGroup, + typ="symbolic", + members={ + "size": MemberType.USE_REAL, + "rank": MemberType.USE_REAL, + "_get_backend_name": MemberType.USE_REAL, + "group_name": MemberType.USE_REAL, + "group_desc": MemberType.USE_REAL, + "__eq__": MemberType.USE_REAL, + "__ne__": MemberType.USE_REAL, + }, + ) + + +__all__ = [ + "Backend", + "BackendConfig", + "GroupMember", + "P2POp", + "all_gather", + "all_gather_coalesced", + "all_gather_object", + "all_reduce", + "all_reduce_coalesced", + "all_to_all", + "all_to_all_single", + "barrier", + "batch_isend_irecv", + "broadcast", + "send_object_list", + "recv_object_list", + "broadcast_object_list", + "destroy_process_group", + "gather", + "gather_single", + "gather_into_tensor", + "gather_object", + "get_backend_config", + "get_backend", + "get_backend_impl", + "get_default_backend_for_device", + "get_rank", + "get_world_size", + "get_pg_count", + "group", + "init_process_group", + "irecv", + "is_gloo_available", + "is_initialized", + "is_mpi_available", + "is_backend_available", + "is_nccl_available", + "is_torchelastic_launched", + "is_ucc_available", + "is_xccl_available", + "isend", + "monitored_barrier", + "new_group", + "new_subgroups", + "new_subgroups_by_enumeration", + "recv", + "reduce", + "reduce_scatter", + "scatter", + "scatter_object_list", + "send", + "supports_complex", + "AllreduceCoalescedOptions", + "AllreduceOptions", + "AllToAllOptions", + "BarrierOptions", + "BroadcastOptions", + "GatherOptions", + "GroupName", + "PrefixStore", + "ProcessGroup", + "ReduceOp", + "ReduceOptions", + "ReduceScatterOptions", + "ScatterOptions", + "Store", + "DebugLevel", + "get_debug_level", + "Work", + "default_pg_timeout", + "get_group_rank", + "get_global_rank", + "get_process_group_ranks", + "reduce_op", + "all_gather_single", + "all_gather_into_tensor", + "reduce_scatter_single", + "reduce_scatter_tensor", + "get_node_local_rank", + "set_timeout", + "split_group", + "shrink_group", + "record_comm", +] + +_MPI_AVAILABLE = True +_NCCL_AVAILABLE = True +_GLOO_AVAILABLE = True +_UCC_AVAILABLE = True +_XCCL_AVAILABLE = True + +try: + try: + # pyrefly: ignore [missing-import] + from torchcomms._comms import ( + _BackendWrapper, + _is_backend_registered as _torchcomms_is_backend_registered, + ) + except ImportError: + # pyrefly: ignore [missing-import] + from torchcomms._backend_wrapper import _BackendWrapper + + def _torchcomms_is_backend_registered(backend: str) -> bool: + return False + + # pyrefly: ignore [missing-import] + from torchcomms import is_backend_built as _torchcomms_is_backend_built, new_comm + + # pyrefly: ignore [missing-import] + from torchcomms.hooks import FlightRecorderHook + + _TORCHCOMM_AVAILABLE = True +except ImportError: + _TORCHCOMM_AVAILABLE = False + + def _torchcomms_is_backend_built(backend: str) -> bool: + return False + + def _torchcomms_is_backend_registered(backend: str) -> bool: + return False + + +def _use_torchcomms_enabled() -> bool: + """Check if torchcomms is enabled via config.""" + return _TORCHCOMM_AVAILABLE and dist_config.use_torchcomms + + +def _is_torchcomms_backend(backend: str) -> bool: + return _use_torchcomms_enabled() and ( + _torchcomms_is_backend_registered(backend) + or _torchcomms_is_backend_built(backend) + ) + + +def _torchcomms_handles_backend(backend) -> bool: + """True if TorchComms can create a comm for *backend*. + + Backends TorchComms doesn't own -- custom c10d plugins such as ``mooncake`` + or ``ucc`` -- must fall through to the normal ProcessGroup path even when + TorchComms is enabled, rather than being routed through ``new_comm`` / + ``split_group`` (which cannot construct them). + + ``backend`` may be ``None`` (inherits the parent's TorchComms-owned + backend), a bare name (``"nccl"``), or a device-qualified string + (``"cpu:gloo,cuda:nccl"``). A built-in backend reports via + ``is_backend_built``; a backend dynamically registered through + ``torchcomms.register_backend`` (e.g. a Python adapter) reports via + ``_is_backend_registered`` -- so registering such an adapter is enough to + move that backend onto the native TorchComms path with no change here. + """ + if not _TORCHCOMM_AVAILABLE: + return False + if backend is None: + return True + for part in str(backend).lower().split(","): + part = part.strip() + name = part.split(":", 1)[1] if ":" in part else part + if not name: + continue + if not ( + _torchcomms_is_backend_registered(name) + or _torchcomms_is_backend_built(name) + ): + return False + return True + + +def _pg_options_to_hints(pg_options: object) -> dict[str, str] | None: + """Convert ProcessGroupNCCL.Options to a torchcomms hints dict.""" + if pg_options is None: + return None + + hints: dict[str, str] = {} + + if hasattr(pg_options, "config"): + config = pg_options.config + for attr in dir(config): + if not attr.startswith("_"): + try: + value = getattr(config, attr) + if not callable(value): + hints[attr] = str(value) + except Exception: + pass + + for attr in dir(pg_options): + if not attr.startswith("_") and attr != "config": + try: + value = getattr(pg_options, attr) + if not callable(value): + hints[attr] = str(value) + except Exception: + pass + + return hints if hints else None + + +_pickler = pickle.Pickler +_unpickler = pickle.Unpickler + +# Object collectives can serialize via torch.save/torch.load(weights_only=True) +# by passing weights_only=True. Exceptions and stack summaries are commonly +# transmitted for error propagation (e.g. torch.distributed.checkpoint), so +# allowlist them. +torch.serialization.add_safe_globals( + [ + t + for t in vars(builtins).values() + if isinstance(t, type) and issubclass(t, BaseException) + ] + + [traceback.StackSummary, traceback.FrameSummary] +) + +GroupName = NewType("GroupName", str) + + +# Change __module__ of all imported types from torch._C._distributed_c10d that are public +def _export_c_types() -> None: + _public_types_to_change_module = [ + AllreduceCoalescedOptions, + AllreduceOptions, + AllToAllOptions, + BarrierOptions, + BroadcastOptions, + GatherOptions, + PrefixStore, + ProcessGroup, + ReduceOp, + ReduceOptions, + ReduceScatterOptions, + ScatterOptions, + Store, + DebugLevel, + get_debug_level, + Work, + ] + for type in _public_types_to_change_module: + type.__module__ = "torch.distributed.distributed_c10d" + + +_export_c_types() + +try: + from torch._C._distributed_c10d import ProcessGroupMPI + + ProcessGroupMPI.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupMPI"] +except ImportError: + _MPI_AVAILABLE = False + +try: + from torch._C._distributed_c10d import ProcessGroupNCCL + + ProcessGroupNCCL.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupNCCL"] +except ImportError: + _NCCL_AVAILABLE = False + +try: + # In-tree NCCL backend built on the torchcomms engine (selected via the + # "nccl2" backend / entry point). Available whenever NCCL is built. + from torch._C._distributed_c10d import ProcessGroupNCCL2 + + ProcessGroupNCCL2.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupNCCL2"] +except ImportError: + pass + +try: + # Lazy variant of "nccl2": collectives run on a primary communicator while + # each point-to-point peer pair gets its own lazily-created 2-rank comm + # (selected via the "nccl-lazy" backend / entry point). + from torch._C._distributed_c10d import ProcessGroupNCCLLazy + + ProcessGroupNCCLLazy.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupNCCLLazy"] +except ImportError: + pass + +try: + from torch._C._distributed_c10d import _ProcessGroupWrapper, ProcessGroupGloo + + ProcessGroupGloo.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupGloo"] +except ImportError: + _GLOO_AVAILABLE = False + +try: + from torch._C._distributed_c10d import ProcessGroupUCC + + ProcessGroupUCC.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupUCC"] +except ImportError: + _UCC_AVAILABLE = False + +try: + from torch._C._distributed_c10d import ProcessGroupXCCL + + ProcessGroupXCCL.__module__ = "torch.distributed.distributed_c10d" + __all__ += ["ProcessGroupXCCL"] +except ImportError: + _XCCL_AVAILABLE = False + + +if TYPE_CHECKING: + from torch._C._distributed_c10d import ( # noqa: TC004 + _ProcessGroupWrapper, + ProcessGroupGloo, + ProcessGroupMPI, + ProcessGroupNCCL, + ProcessGroupNCCL2, + ProcessGroupUCC, + ProcessGroupXCCL, + ) + +logger = logging.getLogger(__name__) + +PG_WRAPPER_STORE_PREFIX = "pg_wrapper" + + +# Some reduce ops are not supported by complex numbers and will result in an error. +# We currently provide complex support to the distributed API by viewing +# complex tensors as real (torch.view_as_real), meaning that calling +# these unsupported ops will return garbage values rather than error out. +# (e.g. max(2+3i, 3+2i) = 3+3i) +# We'd like calls to unsupported ops to error out accordingly, +# rather than returning garbage values. +_ReduceOp = ReduceOp | ReduceOp.RedOpType +_NonGroupMember: TypeAlias = Literal[-100] +_T = TypeVar("_T") + + +class _TorchComm(Protocol): + def finalize(self) -> None: ... + + +class _WorldPGConfig(TypedDict): + pg_name: GroupName + pg_desc: str + backend_config: str + ranks: list[int] + group_size: int + group_count: int + + +class _PGConfig(TypedDict): + pg_name: str + pg_desc: str + backend_config: str + pg_size: int + ranks: list[int] + + +def supports_complex(reduceOp: _ReduceOp) -> bool: + """Return true if reduce ops is supported. False otherwise.""" + denyList = [ + ReduceOp.MAX, + ReduceOp.MIN, + ReduceOp.PRODUCT, + ReduceOp.BAND, + ReduceOp.BOR, + ReduceOp.BXOR, + ] + return reduceOp not in denyList + + +# TODO refactor into enum/strenum +class Backend(str): # noqa: SLOT000 + """ + An enum-like class for backends. + + Available backends: GLOO, NCCL, UCC, MPI, XCCL, FAKE, and other registered backends. + + The values of this class are lowercase strings, e.g., ``"gloo"``. They can + be accessed as attributes, e.g., ``Backend.NCCL``. + + This class can be directly called to parse the string, e.g., + ``Backend(backend_str)`` will check if ``backend_str`` is valid, and + return the parsed lowercase string if so. It also accepts uppercase strings, + e.g., ``Backend("GLOO")`` returns ``"gloo"``. + + .. note:: The entry ``Backend.UNDEFINED`` is present but only used as + initial value of some fields. Users should neither use it directly + nor assume its existence. + """ + + UNDEFINED = "undefined" + GLOO = "gloo" + NCCL = "nccl" + UCC = "ucc" + MPI = "mpi" + XCCL = "xccl" + FAKE = "fake" + + class _BackendPlugin(NamedTuple): + creator_fn: Callable[..., C10DBackend | ProcessGroup | None] + extended_api: bool + + _plugins: dict[str, _BackendPlugin] = {} + + backend_list = [UNDEFINED, GLOO, NCCL, XCCL, UCC, MPI, FAKE] + + # 3rd-party devices can register the default backend support here + default_device_backend_map: dict[str, str] = { + "cpu": GLOO, + "cuda": NCCL, + "xpu": XCCL, + "mps": GLOO, + } + + backend_capability: dict[str, list[str]] = { + GLOO: ["cpu", "cuda"], + NCCL: ["cuda"], + XCCL: ["xpu"], + UCC: ["cpu", "cuda"], + MPI: ["cpu", "cuda"], + FAKE: ["cpu", "cuda", "hpu", "xpu"], + } + + backend_type_map: dict[str, ProcessGroup.BackendType] = { + UNDEFINED: ProcessGroup.BackendType.UNDEFINED, + GLOO: ProcessGroup.BackendType.GLOO, + NCCL: ProcessGroup.BackendType.NCCL, + XCCL: ProcessGroup.BackendType.XCCL, + UCC: ProcessGroup.BackendType.UCC, + MPI: ProcessGroup.BackendType.MPI, + FAKE: ProcessGroup.BackendType.CUSTOM, + } + + def __new__(cls, name: str) -> str: + """Create and return a new instance of the class.""" + if not isinstance(name, str): + raise ValueError("Backend constructor parameter must be string-ish") + value = getattr(Backend, name.upper(), Backend.UNDEFINED) + + if value == Backend.UNDEFINED: + value = name.lower() + return value + + @classmethod + def _ensure_backend_registered(cls, name: str) -> None: + normalized_name = name.lower() + upper_name = normalized_name.upper() + if upper_name in cls._plugins: + return + + from importlib.metadata import entry_points + + backend_entrypoint = None + for entrypoint in entry_points(group="torch.distributed.backends"): + if entrypoint.name.lower() == normalized_name: + backend_entrypoint = entrypoint + break + + if backend_entrypoint is None: + return + + try: + registrar = backend_entrypoint.load() + except Exception as err: + raise RuntimeError( + f"Failed to load the c10d backend entry point: {normalized_name}" + ) from err + + if not callable(registrar): + raise TypeError("backend entry point must load a callable registrar") + + registrar() + if upper_name not in cls._plugins: + raise RuntimeError( + f"c10d backend entry point {normalized_name} did not register a backend" + ) + + @classmethod + def register_backend( + cls, + name: str, + func: Callable[..., C10DBackend | ProcessGroup | None], + extended_api: bool = False, + devices: str | list[str] | None = None, + *, + _backend_type: ProcessGroup.BackendType | None = None, + ) -> None: + """ + Register a new backend with the given name and instantiating function. + + This class method is used by 3rd party ``ProcessGroup`` extension to + register new backends. + + Args: + name (str): Backend name of the ``ProcessGroup`` extension. It + should match the one in ``init_process_group()``. + func (function): Function handler that instantiates the backend. + The function should be implemented in the backend + extension and takes four arguments, including + ``store``, ``rank``, ``world_size``, and ``timeout``. + extended_api (bool, optional): Whether the backend supports extended argument structure. + Default: ``False``. If set to ``True``, the backend + will get an instance of ``c10d::DistributedBackendOptions``, and + a process group options object as defined by the backend implementation. + devices (str or list of str, optional): device type this backend + supports, e.g. "cpu", "cuda", etc. If `None`, + assumes CPU and the current accelerator. + + .. note:: This support of 3rd party backend is experimental and subject to change. + + """ + if isinstance(devices, str): + devices = [devices] + + # This takes care of CUSTOM Out-of-tree backend types, update in backend_list indicates availability + normalized_name = name.lower() + if not hasattr(Backend, normalized_name.upper()): + setattr(Backend, normalized_name.upper(), normalized_name) + if normalized_name not in Backend.backend_list: + Backend.backend_list.append(normalized_name) + + backend_type = _backend_type or ProcessGroup.BackendType.CUSTOM + Backend.backend_type_map[normalized_name] = backend_type + + if devices is None: + # This is more of a backward support for groups like `threaded`: + # assume default devices CPU and the current accelerator, but warn + warnings.warn( + f"Device capability of {name} unspecified, assuming `cpu` and " + "the current accelerator. Please specify it via the `devices` argument of " + "`register_backend`.", + stacklevel=3, + ) + device_list = ["cpu"] + if (acc := torch.accelerator.current_accelerator()) is not None: + device_list.append(acc.type) + else: + device_list = devices + + Backend.backend_capability[normalized_name] = device_list + if devices is not None: + for device in device_list: + current = Backend.default_device_backend_map.get(device) + # Allow remapping from fake backend to actual backend (e.g., HPU from fake to HCCL) + # but prevent fake backend from claiming devices + if current is None or (current == "fake" and normalized_name != "fake"): + Backend.default_device_backend_map[device] = normalized_name + + Backend._plugins[normalized_name.upper()] = Backend._BackendPlugin( + func, + extended_api, + ) + + +def _create_mpi_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend | None: + if not is_mpi_available(): + raise RuntimeError( + "Distributed package doesn't have MPI built in. MPI is only included " + "if you build PyTorch from source on a host that has MPI installed." + ) + return ProcessGroupMPI.create(opts.global_ranks_in_group) + + +def _create_gloo_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_gloo_available(): + raise RuntimeError("Distributed package doesn't have Gloo built in") + backend_class = ProcessGroupGloo( + opts.store, + opts.group_rank, + opts.group_size, + # pyrefly: ignore [bad-argument-type] + timeout=opts.timeout, + enable_reconfigure=opts.enable_reconfigure, + ) + backend_class.options.global_ranks_in_group = opts.global_ranks_in_group + backend_class.options.group_name = opts.group_id + return backend_class + + +def _create_nccl_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_nccl_available(): + raise RuntimeError("Distributed package doesn't have NCCL built in") + if backend_options is not None: + if not isinstance(backend_options, ProcessGroupNCCL.Options): + raise AssertionError( + "Expected backend_options argument to be of type ProcessGroupNCCL.Options" + ) + if backend_options._timeout != opts.timeout: + warnings.warn( + "backend_options._timeout was specified, " + "but timeout kwarg has a default value that will always override it. ", + stacklevel=2, + ) + else: + # default backend_options for NCCL + backend_options = ProcessGroupNCCL.Options() + backend_options.is_high_priority_stream = False + # pyrefly: ignore [bad-argument-type] + backend_options._timeout = opts.timeout + + if opts.split_from: + if not isinstance(opts.split_from, ProcessGroupNCCL): + raise AssertionError("Expected split_from to be ProcessGroupNCCL") + backend_options.split_from = opts.split_from + backend_options.split_color = _process_group_color(opts.global_ranks_in_group) + backend_options.global_ranks_in_group = opts.global_ranks_in_group + backend_options.group_name = opts.group_id + if opts.enable_reconfigure: + backend_options.enable_reconfigure = True + backend_class = ProcessGroupNCCL( + opts.store, opts.group_rank, opts.group_size, backend_options + ) + return backend_class + + +def _nccl2_options( + backend_options: object | None, +) -> "ProcessGroupNCCL.Options": + if backend_options is None: + return ProcessGroupNCCL.Options() + if not isinstance(backend_options, ProcessGroupNCCL.Options): + raise AssertionError( + "Expected backend_options argument to be of type ProcessGroupNCCL.Options" + ) + return backend_options + + +def _create_nccl2_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_nccl_available(): + raise RuntimeError("Distributed package doesn't have NCCL built in") + pg_options = _nccl2_options(backend_options) + # pyrefly: ignore [bad-argument-type] + pg_options._timeout = opts.timeout + pg_options.global_ranks_in_group = opts.global_ranks_in_group + pg_options.group_name = opts.group_id + if opts.enable_reconfigure: + pg_options.enable_reconfigure = True + if opts.split_from: + if not isinstance(opts.split_from, ProcessGroupNCCL2): + raise AssertionError("Expected split_from to be ProcessGroupNCCL2") + return opts.split_from.split( # pyrefly: ignore[missing-attribute] + opts.store, opts.global_ranks_in_group, pg_options + ) + return ProcessGroupNCCL2(opts.store, opts.group_rank, opts.group_size, pg_options) + + +def _create_nccl_lazy_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_nccl_available(): + raise RuntimeError("Distributed package doesn't have NCCL built in") + pg_options = _nccl2_options(backend_options) + # pyrefly: ignore [bad-argument-type] + pg_options._timeout = opts.timeout + pg_options.global_ranks_in_group = opts.global_ranks_in_group + pg_options.group_name = opts.group_id + if opts.enable_reconfigure: + pg_options.enable_reconfigure = True + if opts.split_from: + if not isinstance(opts.split_from, ProcessGroupNCCLLazy): + raise AssertionError("Expected split_from to be ProcessGroupNCCLLazy") + return opts.split_from.split( # pyrefly: ignore[missing-attribute] + opts.store, opts.global_ranks_in_group, pg_options + ) + return ProcessGroupNCCLLazy( + opts.store, opts.group_rank, opts.group_size, pg_options + ) + + +def _create_ucc_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_ucc_available(): + raise RuntimeError("Distributed package doesn't have UCC built in") + backend_class = ProcessGroupUCC( + opts.store, + opts.group_rank, + opts.group_size, + # pyrefly: ignore [bad-argument-type] + timeout=opts.timeout, + ) + return backend_class + + +def _create_xccl_process_group( + opts: _DistributedBackendOptions, backend_options: object | None +) -> C10DBackend: + if not is_xccl_available(): + raise RuntimeError("Distributed package doesn't have XCCL built in") + backend_options = ProcessGroupXCCL.Options() + backend_options.global_ranks_in_group = opts.global_ranks_in_group + backend_options.group_name = opts.group_id + # pyrefly: ignore [bad-argument-type] + backend_options._timeout = opts.timeout + if opts.enable_reconfigure: + backend_options.enable_reconfigure = True + backend_class = ProcessGroupXCCL( + opts.store, opts.group_rank, opts.group_size, backend_options + ) + return backend_class + + +def _register_builtin_mpi_backend() -> None: + Backend.register_backend( + Backend.MPI, + _create_mpi_process_group, + extended_api=True, + devices=Backend.backend_capability[Backend.MPI], + _backend_type=ProcessGroup.BackendType.MPI, + ) + + +def _register_builtin_gloo_backend() -> None: + Backend.register_backend( + Backend.GLOO, + _create_gloo_process_group, + extended_api=True, + devices=Backend.backend_capability[Backend.GLOO], + _backend_type=ProcessGroup.BackendType.GLOO, + ) + + +def _register_builtin_nccl_backend() -> None: + creator_fn = ( + _create_nccl2_process_group + if os.environ.get("TORCH_DIST_USE_NCCL2") == "1" + else _create_nccl_process_group + ) + Backend.register_backend( + Backend.NCCL, + creator_fn, + extended_api=True, + devices=Backend.backend_capability[Backend.NCCL], + _backend_type=ProcessGroup.BackendType.NCCL, + ) + + +def _register_builtin_nccl_legacy_backend() -> None: + Backend.register_backend( + "nccl-legacy", + _create_nccl_process_group, + extended_api=True, + devices=["cuda"], + _backend_type=ProcessGroup.BackendType.CUSTOM, + ) + + +def _register_builtin_nccl2_backend() -> None: + Backend.register_backend( + "nccl2", + _create_nccl2_process_group, + extended_api=True, + devices=["cuda"], + _backend_type=ProcessGroup.BackendType.CUSTOM, + ) + + +def _register_builtin_nccl_lazy_backend() -> None: + Backend.register_backend( + "nccl-lazy", + _create_nccl_lazy_process_group, + extended_api=True, + devices=["cuda"], + _backend_type=ProcessGroup.BackendType.CUSTOM, + ) + + +def _register_builtin_ucc_backend() -> None: + Backend.register_backend( + Backend.UCC, + _create_ucc_process_group, + extended_api=True, + devices=Backend.backend_capability[Backend.UCC], + _backend_type=ProcessGroup.BackendType.UCC, + ) + + +def _register_builtin_xccl_backend() -> None: + Backend.register_backend( + Backend.XCCL, + _create_xccl_process_group, + extended_api=True, + devices=Backend.backend_capability[Backend.XCCL], + _backend_type=ProcessGroup.BackendType.XCCL, + ) + + +class BackendConfig: + """Backend configuration class.""" + + def __init__(self, backend: str | Backend) -> None: + """Init.""" + self.device_backend_map: dict[str, str] = {} + backend = str(backend) + normalized_backend = backend.lower() + + if backend == Backend.UNDEFINED: + # Detect the accelerator on the machine. If no accelerator is + # available, it returns CPU. + device_type = ( + acc.type + if (acc := torch.accelerator.current_accelerator(check_available=True)) + else "cpu" + ) + try: + backend_str = Backend.default_device_backend_map[device_type] + self.device_backend_map[device_type] = Backend(backend_str) + except KeyError: + raise ValueError( + f"We detected accelerator {device_type} on your machine. " + f"But we don't know which communication backend to use for this accelerator. " + f"Please specify the `backend` argument in the `init_process_group` call." + ) from None + elif ":" not in normalized_backend: + # "device:backend" forms register each backend in the per-pair loop + # below; registering here would only trigger a useless entry-point + # scan for the whole comma-separated string. + Backend._ensure_backend_registered(backend) + + if backend == Backend.UNDEFINED: + pass + elif normalized_backend in Backend.backend_list: + # Cases for when backend is a single string (without device types) + # e.g. "nccl", "gloo", "ucc", "mpi" + supported_devices = Backend.backend_capability[normalized_backend] + backend_val = Backend(backend) + + self.device_backend_map = dict.fromkeys(supported_devices, backend_val) + elif ":" not in normalized_backend: + # User specified a single backend name whose device capability is + # unknown, assuming it can support the default devices of PyTorch + # (CPU and the current accelerator) + warnings.warn( + f"Device capability of {backend} unknown, assuming `cpu` and " + "the current accelerator. You can specify it in `device:backend` format in " + "`init_process_group` call.", + stacklevel=2, + ) + backend_val = Backend(backend) + device_types = ["cpu"] + if (acc := torch.accelerator.current_accelerator()) is not None: + device_types.append(acc.type) + self.device_backend_map = dict.fromkeys(device_types, backend_val) + logger.info("Using backend config: %s", self.device_backend_map) + return + elif ":" in normalized_backend: + # Backend specified in "device:backend" format + # make sure the backend string is in the correct format + # "{device_type1}:{backend1},{device_type2}:{backend2}" + # e.g. "cpu:gloo,cuda:nccl" + backend_str_error_message = f"""The custom backend string argument is invalid: {backend}. + Custom backend string is an experimental feature where the backend string must be in the format: + ":,:...". e.g. 'cpu:gloo,cuda:nccl,xpu:xccl'""" + + # parse the backend string and populate the device_backend_map + for device_backend_pair_str in normalized_backend.split(","): + device_backend_pair = device_backend_pair_str.split(":") + if len(device_backend_pair) != 2: + raise ValueError( + f"Invalid device:backend pairing: \ + {device_backend_pair_str}. {backend_str_error_message}" + ) + device, backend = device_backend_pair + if device in self.device_backend_map: + raise ValueError( + f"Duplicate device type {device} \ + in backend string: {backend}. {backend_str_error_message}" + ) + Backend._ensure_backend_registered(backend) + self.device_backend_map[device] = Backend(backend) + logger.info("Using backend config: %s", self.device_backend_map) + + def __repr__(self) -> str: + """Return all the device:backend pairs separated by commas.""" + return ",".join( + f"{device}:{backend}" for device, backend in self.device_backend_map.items() + ) + + def get_device_backend_map(self) -> dict[str, str]: + """Return backend map of the device.""" + return self.device_backend_map + + +def _parse_backend_string( + backend: str, + available_devices: set[str], +) -> dict[str, str]: + """Parse a backend string into a ``{device_type: backend_name}`` dict. + + Accepts either the merged form (``"cpu:gloo,cuda:nccl"``) or a simple + backend name (``"nccl"``). For the simple form, the backend is mapped to + every device for which it is registered as the default in + ``Backend.default_device_backend_map``. + + For the simple form, only device types in *available_devices* are + included in the result. This prevents bare backend names like ``"gloo"`` + from expanding to device types not present in the caller's context (e.g. + ``"mps"`` on a CUDA-only system). The merged form is returned exactly as + named so callers can validate explicitly-requested devices against their + own context. + + Unlike :class:`BackendConfig`, this does not synthesize defaults for + unspecified devices — callers get back exactly what was named. + """ + backend = backend.lower() + if ":" in backend: + device_backends: dict[str, str] = {} + for part in backend.split(","): + pieces = part.split(":") + if len(pieces) != 2: + raise ValueError( + f"Invalid device:backend pairing '{part}' in '{backend}'. " + "Expected format: ':,:...'" + ) + device, be = pieces[0].strip(), pieces[1].strip() + if device in device_backends: + raise ValueError( + f"Duplicate device type '{device}' in backend string '{backend}'" + ) + Backend._ensure_backend_registered(be) + device_backends[device] = be + return device_backends + + Backend._ensure_backend_registered(backend) + device_types = [ + device + for device, be in Backend.default_device_backend_map.items() + if be == backend and device in available_devices + ] + if not device_types: + raise ValueError( + f"Unknown backend '{backend}'. Specify a 'device:backend' string " + f"or use one of: {sorted(Backend.default_device_backend_map.values())}" + ) + return dict.fromkeys(device_types, backend) + + +def _get_default_backend_type_for_backend_config( + backend_config: BackendConfig, + device_id: torch.device | None = None, +) -> ProcessGroup.BackendType: + def _resolve_backend_type(backend: str) -> ProcessGroup.BackendType: + # Must mirror _new_process_group_helper's per-device + # `backend_type_map.get(backend_str, CUSTOM)`, or the default type + # names a backend that never gets registered. + known_type = Backend.backend_type_map.get(backend) + if known_type is not None: + return known_type + if _is_torchcomms_backend(backend): + return ProcessGroup.BackendType.CUSTOM + # Unreachable for a group that actually gets built + return ProcessGroup.BackendType.GLOO + + device_backend_map = backend_config.device_backend_map + if not device_backend_map: + return ProcessGroup.BackendType.GLOO + + accelerator = torch.accelerator.current_accelerator() + + def _device_priority(device: str) -> int: + # Prefer this group's accelerator backend over the host one, because + # ``ProcessGroup::barrier()`` derives its tensor device from the default + # backend type -- a host default silently downgrades barrier() on a + # mixed "cpu:,:" group to a CPU barrier. + if device_id is not None and device == device_id.type: + return 0 + if accelerator is not None and device == accelerator.type: + return 1 + if device != "cpu": + return 2 + return 3 + + # `min` keeps the first device of the winning priority, so ties resolve in + # the order the devices were named in the backend string. + device = min(device_backend_map, key=_device_priority) + return _resolve_backend_type(str(device_backend_map[device])) + + +class _reduce_op: + r""" + Deprecated enum-like class. + + For reduction operations: ``SUM``, ``PRODUCT``, ``MIN``, and ``MAX``. + + :class:`~torch.distributed.ReduceOp` is recommended to use instead. + """ + + def __init__(self) -> None: + # __members__ is a dict storing key-value pairs for enum classes + for k, v in ReduceOp.RedOpType.__members__.items(): + setattr(self, k, v) + self.__members__ = ReduceOp.RedOpType.__members__ + + @deprecated( + "`torch.distributed.reduce_op` is deprecated, " + "please use `torch.distributed.ReduceOp` instead", + category=FutureWarning, + ) + def __getattribute__(self, key: str) -> object: + return object.__getattribute__(self, key) + + +reduce_op = _reduce_op() + + +class P2POp: + """ + A class to build point-to-point operations for ``batch_isend_irecv``. + + This class builds the type of P2P operation, communication buffer, peer rank, + Process Group, and tag. Instances of this class will be passed to + ``batch_isend_irecv`` for point-to-point communications. + + Args: + op (Callable): A function to send data to or receive data from a peer process. + The type of ``op`` is either ``torch.distributed.isend`` or + ``torch.distributed.irecv``. + tensor (Tensor): Tensor to send or receive. + peer (int, optional): Destination or source rank. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + tag (int, optional): Tag to match send with recv. + group_peer (int, optional): Destination or source rank. + """ + + def __init__( + self, + op: Callable[..., Work | None], + tensor: torch.Tensor, + peer: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_peer: int | None = None, + ) -> None: + """Init.""" + self.op = op + self.tensor = tensor + self.group = _group_or_default_group(group) + self.peer = _canonicalize_group_rank( + self.group, peer, group_peer, return_global=True + ) + self.tag = tag + self.group_peer = _canonicalize_group_rank(self.group, peer, group_peer) + + def __new__( + cls, + op: Callable[..., Work | None], + tensor: torch.Tensor, + peer: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_peer: int | None = None, + ) -> "P2POp": + """Create and return a new instance of the class.""" + _check_op(op) + _check_single_tensor(tensor, "tensor") + + return object.__new__(cls) + + def __repr__(self) -> str: + my_group_rank = get_rank(self.group) + op_name = self.op.__name__ + group_name = self.group.group_name if self.group else "default_pg" + if "send" in op_name: + s = my_group_rank + d = self.group_peer + elif "recv" in op_name: + s = self.group_peer + d = my_group_rank + else: + return super().__repr__() + + return f"P2POp({op_name} pg={group_name}, group_src={s}, group_dst={d}, {self.tensor.shape}, {self.tensor.dtype})" + + +class _CollOp: + """ + A class to capture collective operations. + + Args: + op (Callable): A collective function, e.g. ``torch.distributed.all_reduce``. + tensor (Tensor): Tensor to operate on. + dst_tensor (Tensor, optional): Provided when source and destination tensors are not the same. + redop (ReduceOp, optional): reduce operation. + root (int, optional): root of broadcast or reduce. + """ + + def __init__( + self, + op: Callable[..., Work | None], + tensor: torch.Tensor, + dst_tensor: torch.Tensor | None = None, + redop: _ReduceOp | None = None, + root: int | None = None, + ) -> None: + self.op = op + self.tensor = tensor + self.dst_tensor = dst_tensor + self.redop = redop + self.root = root + + +# DO NOT USE THESE FIELDS DIRECTLY. +# Use them through the _world object to make sure the _world override mechanism +_pg_map: dict[ProcessGroup, tuple[str, Store]] = {} +_pg_names: dict[ProcessGroup, GroupName] = {} +_pg_group_ranks: dict[ProcessGroup | _NonGroupMember, dict[int, int]] = {} +# For a pg, it is a map from ProcessGroup to BackendConfig +_pg_backend_config: dict[ProcessGroup, str] = {} +_group_count = 0 +_tags_to_pg: dict[str, list[ProcessGroup]] = {} +_pg_to_tag: dict[ProcessGroup, str] = {} +_backend: str | None = None + + +class _World: + """ + Container class for c10d process group state. + + This is used during registration and lookup of PG state. + + .. warning:: This is an experimental API intended to expose the inner workings + of c10d and is subject to change.. + """ + + def __init__(self) -> None: + self._default_pg = None + self._pg_coalesce_state: dict[ProcessGroup, list[_CollOp]] = {} + self._comms: list[_TorchComm] = [] + + @property + def default_pg(self) -> ProcessGroup | None: + """ + Process group that includes all ranks of the cluster. + + This default ProcessGroup is used by c10d APIs when a ProcessGroup is needed + but None is provided. + """ + return self._default_pg + + @default_pg.setter + def default_pg(self, value: ProcessGroup | None) -> None: + self._default_pg = value + + @property + def pg_map(self) -> dict[ProcessGroup, tuple[str, Store]]: + """ + Provide Mapping from ProcessGroup to backend name and store. + + For NCCL and GLOO pg, it is a map from ProcessGroup to (Backend, Store) + For MPI pg, it is a map from ProcessGroup to (Backend, None) + + TODO don't expose the map, expose fine grained ops + """ + global _pg_map + return _pg_map + + @property + def pg_names(self) -> dict[ProcessGroup, GroupName]: + """ + Process group's names, map from ProcessGroup to str. + + TODO don't expose the map, expose fine grained ops + """ + global _pg_names + return _pg_names + + @property + def pg_group_ranks( + self, + ) -> dict[ProcessGroup | _NonGroupMember, dict[int, int]]: + """ + Process group's global rank to local rank mapping. + + TODO don't expose the map, expose fine grained ops + """ + global _pg_group_ranks + return _pg_group_ranks + + @property + def pg_backend_config(self) -> dict[ProcessGroup, str]: + """ + Process group's backend config. + + TODO don't expose the map, expose fine grained ops + """ + global _pg_backend_config + return _pg_backend_config + + @property + def group_count(self) -> int: + """ + Process group count for default naming. + + TODO don't expose group_count, use something else instead + """ + global _group_count + return _group_count + + @group_count.setter + def group_count(self, value: int) -> None: + """Use to compute the name of ProcessGroups when using global synchronization.""" + global _group_count + _group_count = value + + @property + def tags_to_pg(self) -> dict[str, list[ProcessGroup]]: + global _tags_to_pg + return _tags_to_pg + + @property + def pg_to_tag(self) -> dict[ProcessGroup, str]: + global _pg_to_tag + return _pg_to_tag + + @property + def pg_coalesce_state(self) -> dict[ProcessGroup, list[_CollOp]]: + return self._pg_coalesce_state + + @property + def comms(self) -> list[_TorchComm]: + return self._comms + + @property + def pg_config_info(self) -> list[_WorldPGConfig]: + """ + Return a list of dict with process groups and backends. + + Along with their unique IDs and configurations (types and ranks). + """ + config_info: list[_WorldPGConfig] = [] + default_pg_size = _get_group_size(None) + for pg in self.pg_map: + ranks = self.pg_group_ranks[pg] + config_info.append( + { + "pg_name": self.pg_names[pg], + "pg_desc": pg.group_desc, + "backend_config": self.pg_backend_config[pg], + "ranks": ( + list(ranks.keys()) if len(ranks) != default_pg_size else [] + ), # 'ranks' is an empty list when all ranks are involved in a pg + "group_size": len(ranks), + "group_count": self.group_count, + } + ) + return config_info + + +_world = _World() +"""Holds the singleton instance of ``_World`` used by c10. Experimental extension point to override it""" + + +class _WorldMeta(type): + """ + Meta class of ``group`` and ``GroupMember``. + + Allows them to have the class property ``WORLD``. + """ + + # Points to the default PG once initialized. + @property + def WORLD(cls) -> ProcessGroup | None: + return _world.default_pg + + @WORLD.setter + def WORLD(cls, pg: ProcessGroup | None) -> None: + _world.default_pg = pg + + +class group(metaclass=_WorldMeta): + """Group class. Placeholder.""" + + +class GroupMember(metaclass=_WorldMeta): + """Group member class.""" + + NON_GROUP_MEMBER: Final[_NonGroupMember] = -100 + + +def _get_default_timeout(backend: str) -> timedelta: + # see note on nccl vs other backend timeout (constants.py) + if backend == Backend.NCCL: + if not isinstance(default_pg_nccl_timeout, timedelta): + # TODO moco benchmark on CPU initializes pgnccl backend today, triggered this assert in CI before it was + # changed to be a warning. We should fix the moco model. + warnings.warn( + "Attempted to get default timeout for nccl backend, but NCCL support is not compiled", + stacklevel=2, + ) + return default_pg_timeout + return default_pg_nccl_timeout + else: + return default_pg_timeout + + +def _check_valid_timeout(timeout: object) -> None: + if not isinstance(timeout, timedelta): + raise TypeError( + f"Expected timeout argument to be of type datetime.timedelta, got {timeout}" + ) + + +# Default process group state +_default_pg_init_method: str | None = None + +STORE_BASED_BARRIER_PREFIX = "store_based_barrier_key" + + +def _get_object_coll_device(group: ProcessGroup | None = None) -> str: + """ + .. note:: This is an internal helper and does not have backward + compatibility, please use with caution. + + Return the device type to use with ``group`` for object collectives or + barrier. + + There are selection rules: + 1. If user specifies exactly one backend in ``init_process_group`` call: + use that backend + 2. Else if user specifies multiple "device:backend" pairs in init_process_group: + If "cpu" is among those pairs, use "cpu" (because the object is in cpu memory); + Otherwise, use the first backend (sort of a random pick). + + Args: + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Returns: + str: The device type to use for object collective with ``group``. + + """ + group = group or _get_default_group() + + if not isinstance(group, ProcessGroup): + warnings.warn( + f"You are using a Backend {type(group)} as a ProcessGroup. " + "This usage is deprecated since PyTorch 2.0. Please use a public API " + "of PyTorch Distributed instead.", + stacklevel=2, + ) + # Provide backward compatibility to cases where `group` passed in is + # actually a Backend (like `ProcessGroupGloo`) rather than a + # `ProcessGroup` in PT 2.0 sense + if isinstance(group, ProcessGroupGloo): + # RPC uses Gloo for object collectives + return "cpu" + else: + raise ValueError(f"Expecting a ProcessGroup, but got a {type(group)}.") + + """ + ``group._device_types`` is a property pybind that returns the devices + ("cpu", "cuda", etc) supported by ``group``. Can be multiple if the + ``group`` supports multiple devices. + """ + devices = group._device_types + + if len(devices) == 1: + # User fixed exactly one backend in `init_process_group` + return devices[0].type + elif len(devices) == 0: + # No backend has been registered with this PG (maybe because no + # collective has been run?) We pick cpu as the default and hopefully + # this would lazily init Gloo or other available cpu backend. + return "cpu" + elif torch.device("cpu") in devices: + # There are multiple backends in this PG and cpu is among them. + # cpu is preferred as the object is in cpu memory. No need for device + # copy. + return "cpu" + else: + # No cpu in the backend list. Randomly pick the first backend + return devices[0].type + + +def _get_pg_default_device(group: ProcessGroup | None = None) -> torch.device: + """ + .. note:: This method will be deprecated, it only stays for + backward-compatibility reason. Alternatives: + + - If you need to find a device for object collectives, please use + `_get_object_coll_device(group)`. + + - If you need to query the device types supported by group, please use + `_device_capability(group)`. + + Return the device type registered with ``group``. + + For example, if `init_process_group("nccl", ...)` was called, the returned + value would be `torch.device("cuda")`. + + Errors out if no device has been registered. + + Args: + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Returns: + torch.device: The device type registered with ``group``. + """ + + warnings.warn( + "`_get_pg_default_device` will be deprecated, it only stays for " + "backward-compatibility reason. If you need to find a device for object " + "collectives, please use `_get_object_coll_device`. If you need to query " + "the device types supported by group, please use " + "`_device_capability(group)`. ", + stacklevel=2, + ) + group = group or _get_default_group() + + if not isinstance(group, ProcessGroup): + # Provide backward compatibility to cases where `group` passed in is + # actually a Backend (like `ProcessGroupGloo`) rather than a + # `ProcessGroup` in PT 2.0 sense + warnings.warn( + f"You are using a Backend {type(group)} as a ProcessGroup. " + "This usage is deprecated since PyTorch 2.0. Please use a public API " + "of PyTorch Distributed instead.", + FutureWarning, + stacklevel=3, + ) + # Most users create Gloo with private API for object collectives + return torch.device("cpu") + + """ + ``group._device_types`` is a property pybind that returns the devices + ("cpu", "cuda", etc) supported by ``group``. Can be multiple if the + ``group`` supports multiple devices. + """ + devices = group._device_types + + if len(devices) == 1: + # User fixed exactly one backend in `init_process_group` + return devices[0] + elif len(devices) == 0: + raise RuntimeError( + "Default device not found, because no backend has been registered " + "with this ProcessGroup." + ) + else: + # There are multiple backends in this PG. + if torch.device("cpu") in devices: + rv = torch.device("cpu") + else: + rv = devices[0] + warnings.warn( + "Multiple backends are registered with this ProcessGroup. We cannot " + f"determine which one is the default. Returning {rv}. " + "Please consider using other APIs.", + stacklevel=2, + ) + return rv + + +def _device_capability(group: ProcessGroup | None = None) -> list[str]: + """ + Return the device type(s) supported by ``group``. + + Args: + group (ProcessGroup, optional): The process group to query. If None, + the default process group will be used. + + Returns: + List[str]: A list of device types supported by ``group``. + """ + group = group or _get_default_group() + return [device.type for device in group._device_types] + + +@_time_logger +def _store_based_barrier( + rank: int, + store: Store, + group_name: GroupName, + rendezvous_count: int, + timeout: timedelta, + logging_interval: timedelta = timedelta(seconds=10), +) -> None: + """ + Store based barrier for synchronizing processes. + + Barrier based on store which is used for synchronizing processes after + ``init_process_group`` or ``new_group``. Intended to be used only with + those two methods and is not a generic alternative to ``barrier()``. + """ + store_key = f"{STORE_BASED_BARRIER_PREFIX}:{group_name}" + store.add(store_key, 1) + logger.debug("Added key: %s to store for rank: %s", store_key, rank) + + # Now wait for all workers to check in with the store. + world_size = rendezvous_count + worker_count = store.add(store_key, 0) + + last_worker_key = f"{store_key}:last_worker" + if worker_count == world_size: + store.set(last_worker_key, "1") + + # adjust the timeout to be at least 10secs + 1sec per thousand ranks to reduce the odds of timeout + # this value was empirically found while scale testing. + logging_interval = max(logging_interval, timedelta(seconds=10 + world_size / 1000)) + + start = time.time() + while True: + try: + # This will throw an exception after the logging_interval in which we print out + # the status of the group or time out officially, throwing runtime error + store.wait([last_worker_key], logging_interval) + break + except RuntimeError as e: + worker_count = store.add(store_key, 0) + # Print status periodically to keep track. + logger.debug( + "Waiting in store based barrier to initialize process group for %s seconds" + "rank: %s, key: %s (world_size=%s, num_workers_joined=%s, timeout=%s error=%s)", + time.time() - start, + rank, + store_key, + world_size, + worker_count, + timeout, + e, + ) + + if timedelta(seconds=(time.time() - start)) > timeout: + raise DistStoreError( # noqa: B904 + "Timed out initializing process group in store based barrier on " + f"rank {rank}, for key: {store_key} (world_size={world_size}, " + f"num_workers_joined={worker_count}, timeout={timeout} error={e})" + ) + + logger.info( + "Rank %s: Completed store-based barrier for key:%s with %s nodes.", + rank, + store_key, + world_size, + ) + + +def _rank_not_in_group( + group: ProcessGroup | C10DBackend | _NonGroupMember | None, +) -> TypeIs[_NonGroupMember]: + """Check if the current process's rank is not in a given group.""" + if group is None: + return False + return group == GroupMember.NON_GROUP_MEMBER + + +def _warn_not_in_group(op_name: str) -> None: + global_rank = -1 if GroupMember.WORLD is None else GroupMember.WORLD.rank() + warnings.warn( + f"Running {op_name} on global rank {global_rank} which does not " + "belong to the given group.", + stacklevel=2, + ) + + +def get_group_rank(group: ProcessGroup, global_rank: int) -> int: + """ + Translate a global rank into a group rank. + + ``global_rank`` must be part of ``group`` otherwise this raises RuntimeError. + + Args: + group (ProcessGroup): ProcessGroup to find the relative rank. + global_rank (int): Global rank to query. + + Returns: + Group rank of ``global_rank`` relative to ``group`` + + N.B. calling this function on the default process group returns identity + """ + if group is GroupMember.WORLD: + return global_rank + if group not in _world.pg_group_ranks: + raise ValueError( + f"Group {group} is not registered, please create group with torch.distributed.new_group API" + ) + group_ranks = _world.pg_group_ranks[group] + if global_rank not in group_ranks: + raise ValueError(f"Global rank {global_rank} is not part of group {group}") + + return group_ranks[global_rank] + + +def get_global_rank(group: ProcessGroup, group_rank: int) -> int: + """ + Translate a group rank into a global rank. + + ``group_rank`` must be part of `group` otherwise this raises RuntimeError. + + Args: + group (ProcessGroup): ProcessGroup to find the global rank from. + group_rank (int): Group rank to query. + + Returns: + Global rank of ``group_rank`` relative to ``group`` + + N.B. calling this function on the default process group returns identity + """ + if group is GroupMember.WORLD: + return group_rank + if group not in _world.pg_group_ranks: + raise ValueError( + f"Group {group} is not registered, please create group with torch.distributed.new_group API" + ) + for rank, grp_rank in _world.pg_group_ranks[group].items(): + if grp_rank == group_rank: + return rank + raise ValueError(f"Group rank {group_rank} is not part of group {group}") + + +# TODO: remove this once the ecosystem moves away from it. +@deprecated( + "`torch.distributed.distributed_c10d._get_global_rank` is deprecated, " + "please use `torch.distributed.distributed_c10d.get_global_rank` instead", + category=FutureWarning, +) +def _get_global_rank(group: ProcessGroup, rank: int) -> int: + """Use get_global_rank as this method is deprecated.""" + return get_global_rank(group, rank) + + +def get_process_group_ranks(group: ProcessGroup | None) -> list[int]: + """ + Get all ranks associated with ``group``. + + Args: + group (Optional[ProcessGroup]): ProcessGroup to get all ranks from. + If None, the default process group will be used. + + Returns: + List of global ranks ordered by group rank. + """ + return list(_world.pg_group_ranks[group or _get_default_group()].keys()) + + +def _get_group_size(group: ProcessGroup | None) -> int: + """Get a given group's world size.""" + if group is GroupMember.WORLD or group is None: + default_pg = _get_default_group() + return default_pg.size() + return group.size() + + +def _get_group_size_by_name(group_name: GroupName | ProcessGroup) -> int: + if isinstance(group_name, str): + # pyrefly: ignore[bad-argument-type] # pyrefly bug + group_name = _resolve_process_group(group_name) + return group_name.size() + + +def _resolve_group_name_by_ranks_and_tag(ranks: list[int], tag: str) -> GroupName: + # TODO(yifu): remove this function once ranks + tag is not a supported + # identifier for process group for functional collectives. + group = _find_pg_by_ranks_and_tag(tag, ranks) + if group is None: + raise ValueError("") + return group.group_name + + +def _check_single_tensor(param: object, param_name: str) -> None: + """Check that the parameter ``param_name`` is a single tensor.""" + if not isinstance(param, torch.Tensor): + raise TypeError( + f"""Invalid function argument. Expected parameter `{param_name}` of type torch.Tensor + but got {type(param)} instead.""" + ) + + +def _check_tensor_list(param: object, param_name: str) -> None: + """Check that the parameter ``param_name`` is a list of tensors.""" + if not isinstance(param, list): + raise TypeError( + f"""Invalid function argument. Expected parameter `{param_name}` of type List[torch.Tensor] + but got {type(param)} instead.""" + ) + elif not all(isinstance(p, torch.Tensor) for p in param): + raise TypeError( + f"""Invalid function argument. Expected parameter `{param_name}` of type List[torch.Tensor] + but got {type(param)} with elements of type {[type(p) for p in param]}.""" + ) + + +def _group_or_default_group(group: ProcessGroup | None = None) -> ProcessGroup: + if group is None or group is GroupMember.WORLD: + group = _get_default_group() + return group + + +def _canonicalize_group_rank( + group: ProcessGroup, + global_rank: int | None = None, + group_rank: int | None = None, + return_global: bool = False, +) -> int: + """ + Helper method to take _either_ a global rank or a group rank and produce a group rank. + + If 'return_global' is true, produce a global rank instead of a group rank. + """ + + if group_rank is not None: + if global_rank is not None: + raise ValueError("Can't specify both group_rank and global_rank") + if return_global: + return get_global_rank(group, group_rank) + else: + if global_rank is None: + raise ValueError("Must specify global_rank or group_rank") + if return_global: + return global_rank + group_rank = get_group_rank(group, global_rank) + return group_rank + + +def _check_not_self_rank(group: ProcessGroup, rank: int, rank_type: str) -> None: + if group.rank() == rank: + raise ValueError( + f"Invalid {rank_type} rank: {rank_type} rank should not be the same as " + "the rank of the current process." + ) + + +def _as_iterable( + obj: torch.Tensor | list[torch.Tensor], +) -> collections.abc.Iterable[torch.Tensor]: + return obj if isinstance(obj, list) else (obj,) + + +def _ensure_all_tensors_same_dtype( + *tensors: torch.Tensor | list[torch.Tensor], +) -> None: + last_dtype = None + + for tensor in itertools.chain.from_iterable(map(_as_iterable, tensors)): + tensor_dtype = tensor.dtype + # Mixing complex and its element type is allowed + if tensor_dtype.is_complex: + tensor_dtype = ( + torch.float32 if tensor_dtype == torch.complex64 else torch.complex128 + ) + + if last_dtype is None: + last_dtype = tensor_dtype + else: + if last_dtype != tensor_dtype: + raise ValueError( + "Invalid usage of tensors with different dtypes" + f"Found {last_dtype} and {tensor.dtype}" + ) + + +def _check_op(op: Callable[..., Work | None]) -> None: + """Check that the ``op`` is either isend or irecv.""" + if op not in [isend, irecv]: + raise ValueError( + "Invalid ``op``. Expected ``op`` " + "to be of type ``torch.distributed.isend`` or " + "``torch.distributed.irecv``." + ) + + +def _check_p2p_op_list(p2p_op_list: list[P2POp]) -> None: + """ + Check that the ``p2p_op_list`` is a list of P2POp instances. + + Also, check that all ops use the same group. + """ + if not isinstance(p2p_op_list, list) or not all( + isinstance(p2p_op, P2POp) for p2p_op in p2p_op_list + ): + raise ValueError( + "Invalid ``p2p_op_list``. Each op is expected to " + "to be of type ``torch.distributed.P2POp``." + ) + + group = p2p_op_list[0].group + if not all(group == p2p_op.group for p2p_op in p2p_op_list): + raise ValueError("All ops need to use the same group.") + + +def is_mpi_available() -> bool: + """Check if the MPI backend is available.""" + return _MPI_AVAILABLE + + +def is_nccl_available() -> bool: + """Check if the NCCL backend is available.""" + return _NCCL_AVAILABLE + + +def is_gloo_available() -> bool: + """Check if the Gloo backend is available.""" + return _GLOO_AVAILABLE + + +def is_ucc_available() -> bool: + """Check if the UCC backend is available.""" + return _UCC_AVAILABLE + + +def is_xccl_available() -> bool: + """Check if the XCCL backend is available.""" + return _XCCL_AVAILABLE + + +def _check_single_backend_availability(backend_name: str) -> bool: + """ + Helper function to check if a single backend is available. + """ + available_func = getattr( + torch.distributed, f"is_{str(backend_name).lower()}_available", None + ) + if available_func: + return available_func() + Backend._ensure_backend_registered(str(backend_name)) + return str(backend_name).lower() in Backend.backend_list + + +def is_backend_available(backend: str) -> bool: + """ + Check backend availability. + + Checks if the given backend is available and supports the built-in backends or + third-party backends through function ``Backend.register_backend``. + + Args: + backend (str): Backend name. + Returns: + bool: Returns true if the backend is available otherwise false. + """ + # If the backend has an ``is_backend_available`` function, return the result of that function directly + if ":" in backend.lower(): # composite backend like "cpu:gloo" + backend_config = BackendConfig(Backend(backend)) + device_backend_map = backend_config.get_device_backend_map() + return all( + _check_single_backend_availability(str(backend_name)) + for backend_name in device_backend_map.values() + ) + else: + # Handle simple backend strings like "nccl", "gloo" + return _check_single_backend_availability(backend) + + +def is_initialized() -> bool: + """Check if the default process group has been initialized.""" + return GroupMember.WORLD is not None + + +def is_torchelastic_launched() -> bool: + """ + Check whether this process was launched with ``torch.distributed.elastic`` (aka torchelastic). + + The existence of ``TORCHELASTIC_RUN_ID`` environment + variable is used as a proxy to determine whether the current process + was launched with torchelastic. This is a reasonable proxy since + ``TORCHELASTIC_RUN_ID`` maps to the rendezvous id which is always a + non-null value indicating the job id for peer discovery purposes.. + """ + return os.getenv("TORCHELASTIC_RUN_ID") is not None + + +def _is_barrier_after_init() -> int: + # Environment variable to control whether process group should perform a + # barrier after its init. Default value is 0, i.e. no barrier. If you + # experience issue with this setting, you may set + # `TORCH_DIST_INIT_BARRIER=1` to add the barrier. + return int(os.getenv("TORCH_DIST_INIT_BARRIER", "0")) + + +def _get_default_group() -> ProcessGroup: + """Get the default process group created by init_process_group.""" + if not is_initialized(): + raise ValueError( + "Default process group has not been initialized, " + "please make sure to call init_process_group." + ) + if TYPE_CHECKING: + return not_none(GroupMember.WORLD) + else: + return GroupMember.WORLD + + +def _get_default_store() -> Store: + """Get the default store created by init_process_group.""" + if not is_initialized(): + raise ValueError( + "Default process group has not been initialized, " + "please make sure to call init_process_group." + ) + default_pg = _get_default_group() + _, default_store = _world.pg_map[default_pg] + return default_store + + +def _update_default_pg(pg: ProcessGroup | None) -> None: + _world.default_pg = pg + rank = pg.rank() if pg is not None and pg != GroupMember.NON_GROUP_MEMBER else -1 + torch._C._distributed_c10d._set_global_rank(rank) + + +def get_backend_config(group: ProcessGroup | None = None) -> str: + """ + Return the backend configuration of the given process group. + + Args: + group (ProcessGroup, optional): The process group to work on. The + default is the general main process group. If another specific group + is specified, the calling process must be part of :attr:`group`. + + Returns: + The backend configuration of the given process group as a lower case string. + + """ + pg = group or _get_default_group() + if _rank_not_in_group(pg): + raise ValueError("Invalid process group specified") + backend_config = _world.pg_backend_config.get(pg) + return str(not_none(backend_config)) + + +def get_backend(group: ProcessGroup | None = None) -> str: + """ + Return the backend of the given process group. + + Args: + group (ProcessGroup, optional): The process group to work on. The + default is the general main process group. If another specific group + is specified, the calling process must be part of :attr:`group`. + + Returns: + The backend of the given process group as a lower case string. + + """ + pg = group or _get_default_group() + if _rank_not_in_group(pg): + raise ValueError("Invalid process group specified") + + pg_store = _world.pg_map.get(pg, None) + if pg_store is None: + raise ValueError( + f"Process group {pg} is not initialized in the world group map. Please initialize the group first." + ) + + return Backend(not_none(pg_store)[0]) + + +def get_backend_impl( + group: str | ProcessGroup | None = None, + device: torch.device | None = None, +) -> torch._C._distributed_c10d.Backend: + r"""get_backend_impl(group=None, device=None) -> torch._C._distributed_c10d.Backend + + Return the underlying backend implementation of the given process group. + + This allows custom backends to expose backend-specific methods for + experimentation purposes. + + .. warning:: + This API bypasses ``torch.compile`` tracing and other hooks. Backend + methods are experimental and subject to breakage without warning. + + Args: + group (str or ProcessGroup, optional): The process group or process + group name to work on. The default is the general main process + group. If another specific group is specified, the calling process + must be part of :attr:`group`. + device (:class:`torch.device`, optional): The device used to select a + backend implementation. If ``None``, this returns the bound device + backend or the single backend shared by the registered devices. + Default: ``None``. + + Returns: + torch._C._distributed_c10d.Backend: The backend implementation for + the given process group and device. + """ + if isinstance(group, str): + pg = _resolve_process_group(GroupName(group)) + else: + pg = group or _get_default_group() + + if _rank_not_in_group(pg): + raise ValueError("Invalid process group specified") + + if device is not None: + return pg.get_backend(device) + + bound_device_id = pg.bound_device_id + if bound_device_id is not None: + return pg.get_backend(bound_device_id) + + devices = pg._device_types + if not devices: + return pg.get_backend(torch.device("cpu")) + + backend_impl = pg.get_backend(devices[0]) + for backend_device in devices[1:]: + if pg.get_backend(backend_device) is not backend_impl: + raise ValueError( + "Process group has multiple backend implementations; pass " + "the device argument to select one." + ) + return backend_impl + + +def get_default_backend_for_device(device: str | torch.device) -> str: + """ + Return the default backend for the given device. + + Args: + device (Union[str, torch.device]): The device to get the default backend for. + + Returns: + The default backend for the given device as a lower case string. + + """ + if isinstance(device, torch.device): + device_str = device.type + else: + device_str = torch.device(device).type + + backend = Backend.default_device_backend_map.get(device_str) + if backend is None: + raise ValueError(f"Default backend not registered for device : {device}") + + return backend + + +def _get_process_group_uid(pg: ProcessGroup) -> int: + backend = None + try: + backend = pg._get_backend( + torch.accelerator.current_accelerator() or torch.device("cpu") + ) + except RuntimeError: + pass + if is_nccl_available() and isinstance(backend, ProcessGroupNCCL): + return backend.uid + return -1 + + +def _get_pg_config(group: ProcessGroup | None = None) -> _PGConfig: + """ + Return the pg configuration of the given process group. + + """ + pg = group or _get_default_group() + return { + "pg_name": _get_process_group_name(pg), + "pg_desc": pg.group_desc, + "backend_config": get_backend_config(pg), + "pg_size": _get_group_size(pg), + "ranks": get_process_group_ranks(pg), + } + + +def _get_all_pg_configs() -> list[_PGConfig]: + """ + Return the pg configuration of all the process groups. + + """ + config_info: list[_PGConfig] = [_get_pg_config(pg) for pg in _world.pg_map] + return config_info + + +def get_pg_count() -> int: + """ + Return the number of process groups. + + """ + return _world.group_count + + +def get_node_local_rank(fallback_rank: int | None = None) -> int: + """ + Return the local rank of the current process relative to the node. + + Semantically, this is a useful concept for mapping processes to devices. + For example, on a node with 8 accelerator you could use the node local rank to decide + which accelerator device to bind the process to. + + In practice, the actual assignment of node local ranks is handled by the process launcher outside of pytorch, + and communicated via the `LOCAL_RANK` environment variable. + + Torchrun will automatically populate `LOCAL_RANK`, but other launchers may not. If `LOCAL_RANK` is unspecified, + this API will fall back to the provided kwarg 'fallback_rank' if specified, otherwise it will raise an error. The + intent is to allow writing an application that runs either in single or multi device contexts without error. + + """ + if "LOCAL_RANK" in os.environ: + return int(os.environ["LOCAL_RANK"]) + elif fallback_rank is not None: + return int(fallback_rank) + raise RuntimeError( + "LOCAL_RANK is not in the environment. Consider passing fallback_rank to allow `get_node_local_rank` to work, " + "assuming you are not running in a multi-device context and want the code to run locally instead." + ) + + +def _add_ephemeral_timeout_for_all_pgs(timeout: timedelta) -> None: + """ + This API adds an ephemeral timeout extension for all PGs locally + on one rank. The timeout gets reset when the first collective issued + after the API is called completes. Work issued before this call retains its + original timeout. Backends that do not support ephemeral extensions ignore + this call. + NOTE: While this feature + provides flexibility in specific scenarios, it introduces statefulness + to timeout setting. Therefore, it is advisable to use this API sparingly + and consider alternative approaches, such as directly setting the timeout + or utilizing a barrier collective (one can set any timeout to the barrier), + whenever feasible. + + Args: + timeout (timedelta): The delta of timeout to extend. + + Returns: + None. + """ + for pg in _world.pg_map: + pg._add_ephemeral_timeout(timeout) + + +def set_timeout(timeout: timedelta, group: ProcessGroup | None = None) -> None: + """ + Set the default timeout for all future operations on a process group. + + This overrides the timeout configured when the group was created (via + :func:`init_process_group` or :func:`new_group`). The new timeout is + forwarded to every backend registered with :attr:`group` -- for example both + the Gloo and NCCL backends of a group spanning CPU and CUDA devices. Backends + that do not support changing their timeout (such as MPI and UCC) emit a + warning and leave their timeout unchanged. + + Args: + timeout (timedelta): Timeout to set for operations executed against the + process group. The default configured at initialization time is 10 + minutes for NCCL and 30 minutes for other backends. For NCCL this is + the duration after which collectives are aborted asynchronously and + the process crashes; this is necessary because CUDA execution is + async, so it is not safe to keep running user code once an async NCCL + operation has failed (subsequent CUDA operations might run on + corrupted data). When ``TORCH_NCCL_BLOCKING_WAIT`` is set, the process + blocks and waits for this timeout instead. + group (ProcessGroup, optional): The process group to work on. The default + is the general main process group. If another specific group is + specified, the calling process must be part of :attr:`group`. + + Raises: + ValueError: If the calling process is not part of :attr:`group`. + + Returns: + None + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> import torch.distributed as dist + >>> from datetime import timedelta + >>> # Shorten the timeout of the default process group to 30 seconds. + >>> dist.set_timeout(timedelta(seconds=30)) + """ + if group is None: + group = _get_default_group() + if _rank_not_in_group(group): + raise ValueError("Invalid process group specified") + if not isinstance(group, ProcessGroup): + raise AssertionError(f"Expected ProcessGroup, got {type(group)}") + group.set_timeout(timeout) + + +@deprecated( + "`torch.distributed.distributed_c10d._set_pg_timeout` is deprecated, " + "please use `torch.distributed.set_timeout` instead", + category=FutureWarning, +) +def _set_pg_timeout(timeout: timedelta, group: ProcessGroup | None = None) -> None: + """Use set_timeout as this method is deprecated.""" + set_timeout(timeout, group) + + +@_exception_logger +@_time_logger +def init_process_group( + backend: str | None = None, + init_method: str | None = None, + timeout: timedelta | None = None, + world_size: int = -1, + rank: int = -1, + store: Store | None = None, + group_name: str = "", + pg_options: object | None = None, + device_id: torch.device | int | None = None, + _ranks: list[int] | None = None, + enable_reconfigure: bool = False, +) -> None: + """ + Initialize the default distributed process group. + + This will also initialize the distributed package. + + There are 2 main ways to initialize a process group: + 1. Specify ``store``, ``rank``, and ``world_size`` explicitly. + 2. Specify ``init_method`` (a URL string) which indicates where/how + to discover peers. Optionally specify ``rank`` and ``world_size``, + or encode all required parameters in the URL and omit them. + + If neither is specified, ``init_method`` is assumed to be "env://". + + + Args: + backend (str or Backend, optional): The backend to use. Depending on + build-time configurations, valid values include ``mpi``, ``gloo``, + ``nccl``, ``ucc``, ``xccl`` or one that is registered by a third-party + plugin. + Since 2.6, if ``backend`` is not provided, c10d will use a backend + registered for the device type indicated by the `device_id` kwarg + (if provided). The known default registrations today are: ``nccl`` + for ``cuda``, ``gloo`` for ``cpu``, ``xccl`` for ``xpu``. + If neither ``backend`` nor ``device_id`` is provided, c10d will + detect the accelerator on the run-time machine and use a backend + registered for that detected accelerator (or ``cpu``). + This field can be given as a lowercase string (e.g., ``"gloo"``), + which can also be accessed via :class:`Backend` attributes (e.g., + ``Backend.GLOO``). + If using multiple processes per machine with ``nccl`` backend, each + process must have exclusive access to every GPU it uses, as sharing + GPUs between processes can result in deadlock or NCCL invalid usage. + ``ucc`` backend is experimental. + Default backend for the device can be queried with + :func:`get_default_backend_for_device`. + init_method (str, optional): URL specifying how to initialize the + process group. Default is "env://" if no + ``init_method`` or ``store`` is specified. + Mutually exclusive with ``store``. + world_size (int, optional): Number of processes participating in + the job. Required if ``store`` is specified. + rank (int, optional): Rank of the current process (it should be a + number between 0 and ``world_size``-1). + Required if ``store`` is specified. + store(Store, optional): Key/value store accessible to all workers, used + to exchange connection/address information. + Mutually exclusive with ``init_method``. + timeout (timedelta, optional): Timeout for operations executed against + the process group. Default value is 10 minutes for NCCL and 30 minutes for other backends. + This is the duration after which collectives will be aborted asynchronously and the process will crash. + This is done since CUDA execution is async and it is no longer safe to continue executing user code since + failed async NCCL operations might result in subsequent CUDA operations running on corrupted data. + When TORCH_NCCL_BLOCKING_WAIT is set, the process will block and wait for this timeout. + + group_name (str, optional, deprecated): Group name. This argument is ignored + pg_options (ProcessGroupOptions, optional): process group options + specifying what additional options need to be passed in during + the construction of specific process groups. As of now, the only + options we support is ``ProcessGroupNCCL.Options`` for the ``nccl`` + backend, ``is_high_priority_stream`` can be specified so that + the nccl backend can pick up high priority cuda streams when + there're compute kernels waiting. For other available options to config nccl, + See https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/api/types.html#ncclconfig-t + device_id (torch.device | int, optional): a single, specific device + this process will work on, allowing for backend-specific + optimizations. Currently this has two effects, only under + NCCL: the communicator is immediately formed (calling + ``ncclCommInit*`` immediately rather than the normal lazy + call) and sub-groups will use ``ncclCommSplit`` when + possible to avoid unnecessary overhead of group creation. If you + want to know NCCL initialization error early, you can also use this + field. If an `int` is provided, the API assumes that the accelerator + type at compile time will be used. + _ranks: The ranks in the process group. If provided, the process + group name will be the hash of all the ranks in the group. + enable_reconfigure (bool, optional): If ``True``, create the backend in + the reconfigure (fault tolerance) regime. The communicator is not + initialized until :meth:`ProcessGroup.reconfigure` is called. + Backends that do not support reconfigure ignore this flag. Default + is ``False``. + + .. note:: To enable ``backend == Backend.MPI``, PyTorch needs to be built from source + on a system that supports MPI. + + .. note:: Support for multiple backends is experimental. Currently when no backend is + specified, both ``gloo`` and ``nccl`` backends will be created. The ``gloo`` backend + will be used for collectives with CPU tensors and the ``nccl`` backend will be used + for collectives with CUDA tensors. A custom backend can be specified by passing in + a string with format ":,:", e.g. + "cpu:gloo,cuda:custom_backend". + + """ + + global _world + + global _backend + global _default_pg_init_method + + if GroupMember.WORLD is not None: + raise ValueError("trying to initialize the default process group twice!") + + set_pytorch_distributed_envs_from_justknobs() + + # Depending on the import order, some trace_rules functions may be evaluated + # during the import phase. In such a case, these functions may not correctly + # add the distributed related rules due to import circular dependency. + # We need to clear the lru_cache during the runtime to ensure the correctness + # of these trace_rules. + # + # Since this API must be called before all distributed code being compiled, + # clearing the cache here should be safe. + if "torch._dynamo" in sys.modules: + torch._dynamo.trace_rules.clear_lru_cache() + + if not ((store is None) or (init_method is None)): + raise AssertionError("Cannot specify both init_method and store.") + + if store is not None: + if not world_size > 0: + raise AssertionError("world_size must be positive if using store") + if not rank >= 0: + raise AssertionError("rank must be non-negative if using store") + elif init_method is None: + init_method = "env://" + + # Get the compile-time accelerator type. + # None indicates no accelerator support. + acc = torch.accelerator.current_accelerator() + + # Auto complete device id + if isinstance(device_id, int): + if acc is None: + raise ValueError( + "device_id is an int, but no accelerator support is found from the current compilation. " + "Please use a different compiled version that supports your accelerator." + ) + device_id = torch.device(acc.type, device_id) + + # Sanity check device_id + if device_id is not None and device_id.type != "cpu": + # Type + if acc is None or device_id.type != acc.type: + raise ValueError( + f"device_id {device_id} does not match the current compilation's accelerator support: {acc}. " + "Please use a different compiled version that supports your accelerator." + ) + # Index + if device_id.index is None: + raise ValueError("Please use a device_id with index.") + # Range + if device_id.index >= torch.accelerator.device_count(): + raise ValueError( + f"device_id {device_id} is out of range. Please use a device index less than " + f"the number of accelerators available: {torch.accelerator.device_count()}." + ) + + logger.info("Using device: %s", device_id) + + # If user did not provide a backend string but provided a device id, e.g. + # >>> init_process_group(device_id=device) + # we try to figure out the backend name based on the device type. + if backend is None and device_id is not None: + # Note: 3rd-party devices can register default backend through the + # default map below. + backend = Backend.default_device_backend_map.get(device_id.type) + + # If we still cannot figure it out, e.g. + # >>> init_process_group() + # we set it to `undefined` and rely on lazy init. + if backend is None: + backend = "undefined" + + # TorchComms needs device-qualified backend strings so that split_group + # can create subgroups for each device type. Auto-qualify bare backend + # names (e.g. "nccl" → "cpu:gloo,cuda:nccl") so callers don't have to. + if ( + _use_torchcomms_enabled() + and device_id is not None + and ":" not in backend + and backend not in (Backend.UNDEFINED, Backend.MPI, Backend.FAKE) + and _torchcomms_handles_backend(backend) + ): + bare = backend.lower() + qualified: dict[str, str] = {} + for dev, be in Backend.default_device_backend_map.items(): + if be == bare: + qualified[dev] = bare + if not qualified: + qualified[device_id.type] = bare + if "cpu" not in qualified: + cpu_be = Backend.default_device_backend_map.get("cpu") + if cpu_be: + qualified["cpu"] = cpu_be + backend = ",".join(f"{d}:{b}" for d, b in qualified.items()) + + # Normalize the backend name. + backend = Backend(backend) + + if timeout is None: + timeout = _get_default_timeout(backend) + + _check_valid_timeout(timeout) + + """ + Group name is not visible to users unless they access + internals of c10d. This means we can ignore the value + they provide as it not exposed in a public way. + """ + if _ranks is None or len(_ranks) == 0: + group_name = _process_group_name([], use_hashed_name=False) + else: + group_name = _process_group_name(_ranks, use_hashed_name=True) + if backend == Backend.MPI: + if world_size != -1 or rank != -1: + warnings.warn( + f"For MPI backend, world_size ({world_size}) and rank ({rank}) " + "are ignored since they are assigned by the " + "MPI runtime.", + stacklevel=2, + ) + + default_pg, _ = _new_process_group_helper( + -1, + -1, + [], + backend, + Store(), # Placeholder value since store cannot be None + group_name, + timeout=timeout, + group_desc="default_pg", + enable_reconfigure=enable_reconfigure, + ) + else: + # backward compatible API + if store is None: + if backend == Backend.FAKE: + from torch.testing._internal.distributed.fake_pg import FakeStore + + store = FakeStore() + else: + rendezvous_iterator = rendezvous( + not_none(init_method), rank, world_size, timeout=timeout + ) + store, rank, world_size = next(rendezvous_iterator) + store.set_timeout(timeout) + + # Use a PrefixStore to avoid accidental overrides of keys used by + # different systems (e.g. RPC) in case the store is multi-tenant. + store = PrefixStore("default_pg", store) + + default_pg, _ = _new_process_group_helper( + world_size, + rank, + [], + backend, + store, + group_name, + backend_options=pg_options, + timeout=timeout, + device_id=device_id, + group_desc="default_pg", + enable_reconfigure=enable_reconfigure, + ) + + if not isinstance(default_pg, ProcessGroup): + raise AssertionError("Default process group creation returned a nonmember") + _update_default_pg(default_pg) + + _world.pg_group_ranks[default_pg] = {i: i for i in range(default_pg.size())} + _backend = _world.pg_map[not_none(GroupMember.WORLD)][0] + _default_pg_init_method = init_method + + old_hook = sys.excepthook + excepthook_prefix = f"[rank{get_rank()}]" + + def _distributed_excepthook( + exc_type: type[BaseException], + exc_value: BaseException, + exc_traceback: types.TracebackType | None, + ) -> None: + old_stderr = sys.stderr + sys.stderr = buf = io.StringIO() + try: + old_hook(exc_type, exc_value, exc_traceback) + finally: + sys.stderr = old_stderr + msg = buf.getvalue() + msg = "\n".join( + f"{excepthook_prefix}: {s}" if s != "" else "" for s in msg.split("\n") + ) + sys.stderr.write(msg) + sys.stderr.flush() + + sys.excepthook = _distributed_excepthook + + if _is_barrier_after_init() == 1: + # barrier at the end to ensure that once we return from this method, all + # process groups including global variables (if any) are updated + # correctly on all ranks. + # Update 04/2023: for large-scale runs, this barrier (esp. store-based + # barrier) may be costly and/or unscalable. Also, in a lot of cases, + # these barriers may be unnecessary, as proven by a green CI after + # removal. An environment variable `TORCH_DIST_INIT_BARRIER` has been + # added which enables this barrier only when set to 1. + logger.debug( + "Performing barrier after ProcessGroup initialization since " + "TORCH_DIST_INIT_BARRIER = 1" + ) + if backend == Backend.MPI: + # MPI backend doesn't use store. + barrier() + else: + if store is None: + raise AssertionError("Default process group store is not initialized") + # Use store based barrier here since barrier() used a bunch of + # default devices and messes up NCCL internal state. + _store_based_barrier(rank, store, group_name, world_size, timeout) + + +def _get_split_source(pg: ProcessGroup) -> C10DBackend | None: + split_from = None + if pg.bound_device_id: + split_from = pg._get_backend(pg.bound_device_id) + elif pg is _world.default_pg: + try: + split_from = pg._get_backend( + torch.accelerator.current_accelerator() or torch.device("cpu") + ) + except RuntimeError: + # no cuda device associated with this backend + pass + + if not split_from or not split_from.supports_splitting: + return None + + # If necessary, find a backend to split from by peeling process + # group wrappers from our potentially wrapped process group. + while is_gloo_available() and isinstance(split_from, _ProcessGroupWrapper): + split_from = split_from.wrapped_pg + + return split_from + + +def _new_process_group_helper( + group_size: int, + group_rank: int | None, + global_ranks_in_group: list[int], + backend: str, + store: Store, + group_name: GroupName, + backend_options: object | None = None, + timeout: timedelta | None = None, + pg_tag: str | None = None, + device_id: torch.device | None = None, + group_desc: str | None = None, + enable_reconfigure: bool = False, +) -> tuple[ProcessGroup | _NonGroupMember, Store | None]: + """ + Create a new distributed process group. + + This function must be called by ALL processes in the global group, even if + the calling process is not part of the newly created group. In that case, + this function returns GroupMember.NON_GROUP_MEMBER. + + This function is called with ``global_ranks_in_group == []`` for the default group. + """ + global _world + + if group_name in _world.pg_names.values(): + raise ValueError( + "The specified group name has already been " + "created, please use a different group name" + ) + + if device_id is not None and (device_id.index is None or device_id.type == "cpu"): + raise ValueError( + "init_process_group device_id parameter must be an accelerator with an index" + ) + + # Note: _new_process_group_helper is only called from init_process_group, which always provides a timeout value + _check_valid_timeout(timeout) + timeout = not_none(timeout) + + if pg_tag not in [None, ""]: + # creating with the same tag and rank set results in the same underlying PG + existing_group = _find_pg_by_ranks_and_tag(pg_tag, global_ranks_in_group) + if existing_group: + _, prefix_store = _world.pg_map[existing_group] + return existing_group, prefix_store + + group_desc = "undefined" if group_desc is None else group_desc + + # The list of group ranks is empty if we're creating the default group. + is_default_group = len(global_ranks_in_group) == 0 + + # nccl and potentially other backends allow creation of + # communicators based on pre-existing ones, which can save + # initialization time. Due to lazy initialization of + # communicators in some backends, we have to be careful and only + # split when we *know* the default PG has already started communicator initialization. + # We know this if we have bound a device id to the default pg (eager initialized). + if is_initialized() and _get_default_group().bound_device_id: + split_from = _get_split_source(_get_default_group()) + else: + split_from = None + + # If this is a subgroup (which means group_ranks is specified), + # we check if the current process is a member of the new group. + if not is_default_group: + global_rank = _get_default_group().rank() + if global_rank not in global_ranks_in_group: + # If we are using `ncclCommSplit` (or similar split from + # other APIs) to create the communicator, we will need to + # call `ncclCommSplit` on *all* ranks in this new group's + # parent group, even those not in the new group. This is + # a requirement of the NCCL API as otherwise we would get + # out of sync. + # + # Under TorchComms, subgroups are built with `new_comm` (a + # members-only store rendezvous), not by splitting the parent, so + # non-members must NOT issue a no-color split -- doing so would + # desync against members that never split. + if split_from and not _use_torchcomms_enabled(): + split_from.perform_nocolor_split( # pyrefly: ignore[missing-attribute] + _get_default_group().bound_device_id + ) + return GroupMember.NON_GROUP_MEMBER, None + + if group_rank is None: + raise AssertionError("Group rank must be specified for group members") + + prefix_store = PrefixStore(f"{group_name}/", store) + # The backend for PG will be set later based on what's inside BackendConfig + # and timeout are set in each backend's option. + pg: ProcessGroup = ProcessGroup( + prefix_store, + group_rank, + group_size, + ) + backend_config = BackendConfig(backend) + pg_backend_set = False + # Set the default backend when single backend is passed in. + if "," not in str(backend) and ":" not in str(backend): + if backend not in Backend.backend_type_map: + raise AssertionError(f"Unknown backend type {backend}") + if backend != Backend.UNDEFINED: + pg._set_default_backend(Backend.backend_type_map[backend]) + pg_backend_set = True + # In order to correctly call pg._has_hooks(), we should set the default backend + # when multi backend is passed in + if not pg_backend_set: + pg._set_default_backend( + _get_default_backend_type_for_backend_config(backend_config, device_id) + ) + + if device_id: + pg.bound_device_id = device_id + backend_class: C10DBackend | ProcessGroup + for device, backend_str in backend_config.get_device_backend_map().items(): + # Use the group name as prefix in the default store, such that + # a single store can be reused by multiple groups. + backend_prefix_store = PrefixStore(f"{device}/", prefix_store) + + if ( + _use_torchcomms_enabled() + and backend_str not in [Backend.FAKE] + and _torchcomms_handles_backend(backend_str) + ): + torch_device = torch.device(device) + # Pass this rank's actual device WITH its index. A device-type-only + # torch.device(device) makes the TorchComms bootstrap default the + # device to (group-local rank % device_count) -- correct only for the + # world group (group-local == global rank). For a subgroup the + # group-local rank differs from the rank's physical device, so the + # comm (and its lazy P2P pair comms) would be created on the wrong + # device, causing illegal memory access. The default PG's + # bound_device_id is this rank's device for every group it joins. + if ( + device_id is not None + and device_id.index is not None + and device_id.type == torch_device.type + ): + torch_device = device_id + logger.warning( + "Using TorchComms backend (enabled via %s) for device %s with backend %s", + "TORCH_DISTRIBUTED_USE_TORCHCOMMS env var" + if os.environ.get("TORCH_DISTRIBUTED_USE_TORCHCOMMS") + else "dist_config.use_torchcomms", + torch_device, + backend_str, + ) + # `persistent_store=true` tells torchcomms to reuse the c10d-side + # `backend_prefix_store` directly instead of constructing its own + # TCPStore via StoreManager (which would otherwise require an + # explicit MASTER_ADDR/MASTER_PORT and conflict with the c10d + # rendezvous store on rapid re-binds). + hints: dict[str, str] = {"persistent_store": "true"} + if backend_options is not None: + extra = _pg_options_to_hints(backend_options) + if extra: + hints.update(extra) + # new_comm has no rank/size params -- the TorchComms bootstrap reads + # them from TORCHCOMM_RANK/SIZE. Seed from this group's rank/size so + # non-Torchrun launchers (which TorchComms cannot auto-detect, e.g. + # process-spawning inference servers) work without each caller having + # to set these. Save/restore around the call (single-threaded here). + _tc_saved = ( + os.environ.get("TORCHCOMM_RANK"), + os.environ.get("TORCHCOMM_SIZE"), + ) + os.environ["TORCHCOMM_RANK"] = str(group_rank) + os.environ["TORCHCOMM_SIZE"] = str(group_size) + try: + comm = new_comm( + backend_str, + torch_device, + name=group_name, + store=backend_prefix_store, + hints=hints, + ) + finally: + for _k, _v in zip(("TORCHCOMM_RANK", "TORCHCOMM_SIZE"), _tc_saved): + if _v is None: + os.environ.pop(_k, None) + else: + os.environ[_k] = _v + buffer_size = os.environ.get( + "TORCH_FR_BUFFER_SIZE", + os.environ.get("TORCH_NCCL_TRACE_BUFFER_SIZE", "0"), + ) + recorder = FlightRecorderHook(max_entries=int(buffer_size)) + recorder.register_with_comm(comm) + # Keep a reference so the comm outlives this function scope. + _world.comms.append(comm) + group_name = GroupName(group_name) + backend_class = _BackendWrapper(comm) + # Use the underlying backend's BackendType so distinct torchcomms + # backends (e.g. gloo vs nccl in a "cpu:gloo,cuda:nccl" PG) don't + # collide in ProcessGroup::setBackend's backendTypeToBackend_ map + # — that collision silently aliases the second device to the first + # backend and trips the bound_device_id consistency check. + backend_type = Backend.backend_type_map.get( + backend_str, ProcessGroup.BackendType.CUSTOM + ) + else: + Backend._ensure_backend_registered(backend_str) + if backend_str.upper() not in Backend._plugins: + raise AssertionError(f"Unknown c10d backend type {backend_str.upper()}") + + backend_plugin = Backend._plugins[backend_str.upper()] + creator_fn = backend_plugin.creator_fn + backend_type = Backend.backend_type_map.get( + backend_str, ProcessGroup.BackendType.CUSTOM + ) + + if not backend_plugin.extended_api: + backend_result = creator_fn( + backend_prefix_store, group_rank, group_size, timeout + ) + else: + dist_backend_opts = _DistributedBackendOptions() + dist_backend_opts.store = backend_prefix_store + dist_backend_opts.group_rank = group_rank + dist_backend_opts.group_size = group_size + # pyrefly: ignore [bad-argument-type] + dist_backend_opts.timeout = timeout + dist_backend_opts.group_id = group_name + dist_backend_opts.global_ranks_in_group = global_ranks_in_group + dist_backend_opts.process_group = pg + dist_backend_opts.split_from = split_from + dist_backend_opts.enable_reconfigure = enable_reconfigure + backend_result = creator_fn(dist_backend_opts, backend_options) + + if backend_result is None: + return GroupMember.NON_GROUP_MEMBER, None + backend_class = backend_result + + # If the type is a subclass of ProcessGroup then return this process group immediately + # TODO: This defaults to the old behavior for PythonProcessGroups which overwrites the + # ProcessGroup instance + if isinstance(backend_class, ProcessGroup): + pg = backend_class # type: ignore[assignment] + break + + # Process group wrapper initialization for supported PGs when TORCH_DISTRIBUTED_DEBUG is set + if ( + backend_str in [Backend.GLOO, Backend.NCCL, Backend.XCCL, Backend.UCC] + or backend_str.upper() in Backend._plugins + ): + # In debug mode and if GLOO is available, wrap in a wrapper PG that + # enables enhanced collective checking for debuggability. + if get_debug_level() == DebugLevel.DETAIL: + if not is_gloo_available(): + logger.info( + """TORCH_DISTRIBUTED_DEBUG was set to DETAIL, but + GLOO is not available. Build with Gloo to + create a wrapper process group in debug mode + to aid collective desynchronization debugging.""" + ) + else: + backend_class = _create_process_group_wrapper( + wrapped_pg=backend_class, + store_prefix=group_name, + store=backend_prefix_store, + rank=group_rank, + world_size=group_size, + # pyrefly: ignore [bad-argument-type] + timeout=timeout, + ) + elif ( + get_debug_level() == DebugLevel.DETAIL + or os.environ.get("TORCH_DISTRIBUTED_DEBUG", "") == "DETAIL" + ) and _use_torchcomms_enabled(): + logger.warning( + "(Backend %s) Debug Mode not supported for torchcomms.", backend_str + ) + + # register only a single backend when all get_device_backend_map values are the same + if len(set(backend_config.get_device_backend_map().values())) == 1: + for device in backend_config.get_device_backend_map(): + pg._register_backend(torch.device(device), backend_type, backend_class) + + # break out of outer loop to not create any more backends + break + + pg._register_backend(torch.device(device), backend_type, backend_class) + + # set group_name and group_desc to backend + if group_name is None: + raise AssertionError("group_name must not be None") + if group_desc is None: + raise AssertionError("group_desc must not be None") + pg._set_group_name(group_name) + pg._set_group_desc(group_desc) + + # Backend-agnostic NaN checking, for backends without a native checker + # (ProcessGroupNCCL consumes TORCH_NCCL_NAN_CHECK itself). The group owns the + # hook, so there is no handle to keep alive here. + if os.environ.get("TORCH_DIST_NAN_CHECK", "0") == "1": + NanCheckHook.attach(pg) + + if device_id and pg._get_backend(device_id).supports_splitting: + eager_backend = pg._get_backend(device_id) + eager_backend.eager_connect_single_device(device_id) + + # update global state + if pg_tag in [None, ""]: + # Default-tagged PG: also added to the "" default tag. + _register_pg_in_world( + pg, + backend_name=backend, + store=prefix_store, + group_name=group_name, + backend_config=str(backend_config), + add_to_default_tag=True, + ) + else: + # User-tagged PG: only added to the "user:{pg_tag}" tag. + _register_pg_in_world( + pg, + backend_name=backend, + store=prefix_store, + group_name=group_name, + backend_config=str(backend_config), + pg_tag=f"user:{pg_tag}", + ) + return pg, prefix_store + + +def destroy_process_group( + group: ProcessGroup | _NonGroupMember | None = None, +) -> None: + """ + Destroy a given process group, and deinitialize the distributed package. + + Args: + group (ProcessGroup, optional): The process group to be destroyed, if + group.WORLD is given, all process + groups including the default one will + be destroyed. + """ + global _world + + if group == GroupMember.NON_GROUP_MEMBER: + return + + if group is None: + pg = GroupMember.WORLD + else: + pg = group + + if pg is None: + raise AssertionError("Process group cannot be None") + if _world.pg_map.get(pg, None) is None: + raise ValueError("Invalid process group specified") + + # When users register Python onCompletion hooks, those hooks will run on a + # different thread than the main thread. Today, the ProcessGroup dtor does + # wait for that thread. However, the dtor might finish after the Python + # Interpreter exits. After that grabbing the GIL for the Python hook will crash. + # We can either revive the interpreter when running hooks or keep the main one + # alive until all works and hooks are done. The current implementation does the + # latter. Therefore, we explicitly call _wait_for_pending_works() here to wait + # for the pending hooks to finish. + if type(pg) is ProcessGroup and pg._has_hooks(): + pg._wait_for_pending_works() + + if group is None or group == GroupMember.WORLD: + for comm in _world.comms: + comm.finalize() + _world.comms.clear() + + # shutdown all backends in the order of pg names. shutting down in order because + # ncclCommAbort() was a 'collective' call in some versions of NCCL. + for pg_to_shutdown in sorted( + _world.pg_names, key=lambda x: _world.pg_names[x], reverse=True + ): + pg_to_shutdown.shutdown() + + _update_default_pg(None) + _world.pg_map.clear() + _world.pg_names.clear() + _world.pg_group_ranks.clear() + _world.pg_backend_config.clear() + _world.pg_to_tag.clear() + _world.tags_to_pg.clear() + _world.pg_coalesce_state.clear() + _unregister_all_process_groups() + + # when process group doesn't have an explicit name (only WORLD (default) + # process group can have an explicit name), we use global _world.group_count + # to generate the name. We need to reset the counter on destruction to + # allow consistent value to be generated when we re-create process + # groups after some trainers recover from failure + # + # We only reset this when WORLD is being destroyed because if this + # process group is in good state, we aren't dealing with failures. + _world.group_count = 0 + else: + if _TORCHCOMM_AVAILABLE: + # A single comm may be shared across multiple device types (e.g. a + # gloo group reports both 'cuda' and 'cpu' device types backed by the + # same _BackendWrapper). Deduplicate by comm identity so we finalize + # each comm exactly once — finalize() is not idempotent and raises + # "already finalized" on a second call. + finalized_comm_ids: set[int] = set() + for device_type in pg._device_types: + backend = pg._get_backend(device_type) + if isinstance(backend, _BackendWrapper): + comm = backend.get_comm() + if id(comm) not in finalized_comm_ids: + comm.finalize() + finalized_comm_ids.add(id(comm)) + if finalized_comm_ids: + _world.comms[:] = [ + comm for comm in _world.comms if id(comm) not in finalized_comm_ids + ] + pg.shutdown() + del _world.pg_map[pg] + del _world.pg_names[pg] + del _world.pg_group_ranks[pg] + del _world.pg_backend_config[pg] + if pg in _world.pg_coalesce_state: + warnings.warn( + "Some coalesced collectives haven't been launched when " + "ProcessGroup is destroyed. They will be cleaned.", + stacklevel=2, + ) + del _world.pg_coalesce_state[pg] + + tag = _world.pg_to_tag.get(pg) + del _world.pg_to_tag[pg] + if tag is not None: + try: + _world.tags_to_pg[tag].remove(pg) + if tag.startswith("ptd:"): + _world.tags_to_pg[""].remove(pg) + except Exception: + pass + _unregister_process_group(pg.group_name) + + +def _abort_process_group( + group: ProcessGroup | _NonGroupMember | None = None, +) -> None: + """ + Abort a given process group. If group.WORLD (i.e. `None`) is given, all + process groups including the default one will be aborted. + + Args: + group (ProcessGroup, optional): The process group to be aborted. + + .. note:: this API is experimental and currently only works with the NCCL + backend. + + .. note:: this API should be used with `TORCH_NCCL_ASYNC_ERROR_HANDLING` + turned off (i.e. set to 0). Otherwise, ProcessGroupNCCL's watchdog may + automatically handle errors or timeouts for you including aborting the + ProcessGroup. + """ + global _world + + if group == GroupMember.NON_GROUP_MEMBER: + return + + pg = group or GroupMember.WORLD + + if pg is None: + raise AssertionError("Process group cannot be None") + if _world.pg_map.get(pg, None) is None: + raise ValueError("Invalid process group specified or has been destroyed.") + + device = torch.accelerator.current_accelerator() or torch.device("cpu") + try: + backend = pg._get_backend(device) + except RuntimeError: + backend = None + + if group is None or group == GroupMember.WORLD: + # Abort all backends within a coalescing region (ncclGroupStart|End + # semantic for NCCL). This ensures that different communicators' + # abort calls won't deadlock each other. + # For details, please see: https://github.com/pytorch/pytorch/issues/119797 + # getattr instead of a direct attribute access because custom Python + # backends are ProcessGroup subclasses, which don't expose the + # Backend.supports_coalescing property. + coalescing_device = ( + device if getattr(backend, "supports_coalescing", False) else None + ) + # Drop any pending coalesced ops; everything is being aborted anyway + # and _coalescing_manager requires an empty op list on entry. + _world.pg_coalesce_state.pop(pg, None) + with _coalescing_manager(pg, coalescing_device): + for pg_to_abort in sorted( + _world.pg_names, key=lambda x: _world.pg_names[x], reverse=True + ): + pg_to_abort.abort() + + _update_default_pg(None) + _world.pg_map.clear() + _world.pg_names.clear() + _world.pg_group_ranks.clear() + _world.pg_backend_config.clear() + _world.pg_to_tag.clear() + _world.tags_to_pg.clear() + _world.pg_coalesce_state.clear() + _unregister_all_process_groups() + + # when process group doesn't have an explicit name (only WORLD (default) + # process group can have an explicit name), we use global _world.group_count + # to generate the name. We need to reset the counter on destruction to + # allow consistent value to be generated when we re-create process + # groups after some trainers recover from failure + # + # We only reset this when WORLD is being destroyed because if this + # process group is in good state, we aren't dealing with failures. + _world.group_count = 0 + else: + pg.abort() + del _world.pg_map[pg] + del _world.pg_names[pg] + del _world.pg_group_ranks[pg] + del _world.pg_backend_config[pg] + if pg in _world.pg_coalesce_state: + warnings.warn( + "Some coalesced collectives haven't been launched when " + "ProcessGroup is aborted. They will be cleaned.", + stacklevel=2, + ) + del _world.pg_coalesce_state[pg] + + tag = _world.pg_to_tag.get(pg) + del _world.pg_to_tag[pg] + if tag is not None: + try: + _world.tags_to_pg[tag].remove(pg) + if tag.startswith("ptd:"): + _world.tags_to_pg[""].remove(pg) + except Exception: + pass + _unregister_process_group(pg.group_name) + + +def get_rank(group: ProcessGroup | None = None) -> int: + """ + Return the rank of the current process in the provided ``group``, default otherwise. + + Rank is a unique identifier assigned to each process within a distributed + process group. They are always consecutive integers ranging from 0 to + ``world_size``. + + Args: + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Returns: + The rank of the process group + -1, if not part of the group + + """ + if _rank_not_in_group(group): + return -1 + + default_pg = _get_default_group() + if group is None or group is GroupMember.WORLD: + return default_pg.rank() + + return get_group_rank(group, default_pg.rank()) + + +def get_world_size(group: ProcessGroup | None = None) -> int: + """ + Return the number of processes in the current process group. + + Args: + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Returns: + The world size of the process group + -1, if not part of the group + + """ + if _rank_not_in_group(group): + return -1 + + return _get_group_size(group) + + +def isend( + tensor: torch.Tensor, + dst: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_dst: int | None = None, +) -> Work | None: + """ + Send a tensor asynchronously. + + .. warning:: + Modifying ``tensor`` before the request completes causes undefined + behavior. + + .. warning:: + ``tag`` is not supported with the NCCL backend. + + Unlike send, which is blocking, isend allows src == dst rank, i.e. send to self. + + Args: + tensor (Tensor): Tensor to send. + dst (int): Destination rank on global process group (regardless of ``group`` argument) + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + tag (int, optional): Tag to match send with remote recv + group_dst (int, optional): Destination rank on ``group``. Invalid to specify both ``dst`` and ``group_dst`` + + Returns: + A distributed request object. + None, if not part of the group + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + isend, + relevant_args, + tensor, + dst=dst, + group=group, + tag=tag, + group_dst=group_dst, + ) + + group = _group_or_default_group(group) + group_dst = _canonicalize_group_rank(group, dst, group_dst) + _check_single_tensor(tensor, "tensor") + if _rank_not_in_group(group): + _warn_not_in_group("isend") + return None + + if tensor.is_complex(): + tensor = torch.view_as_real(tensor) + + return group.send([tensor], group_dst, tag) + + +def irecv( + tensor: torch.Tensor, + src: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_src: int | None = None, +) -> Work | None: + """ + Receives a tensor asynchronously. + + .. warning:: + ``tag`` is not supported with the NCCL backend. + + Unlike recv, which is blocking, irecv allows src == dst rank, i.e. recv from self. + + Args: + tensor (Tensor): Tensor to fill with received data. + src (int, optional): Source rank on global process group (regardless of ``group`` argument). + Will receive from any process if unspecified. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + tag (int, optional): Tag to match recv with remote send + group_src (int, optional): Destination rank on ``group``. Invalid to specify both ``src`` and ``group_src``. + + Returns: + A distributed request object. + None, if not part of the group + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + irecv, + relevant_args, + tensor, + src=src, + group=group, + tag=tag, + group_src=group_src, + ) + + _check_single_tensor(tensor, "tensor") + if _rank_not_in_group(group): + _warn_not_in_group("irecv") + return None + + if tensor.is_complex(): + tensor = torch.view_as_real(tensor) + + group = _group_or_default_group(group) + if src is None and group_src is None: + return group.recv_anysource([tensor], tag) + else: + group_src = _canonicalize_group_rank(group, src, group_src) + return group.recv([tensor], group_src, tag) + + +@_exception_logger +def send( + tensor: torch.Tensor, + dst: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_dst: int | None = None, +) -> None: + """ + Send a tensor synchronously. + + .. warning:: + ``tag`` is not supported with the NCCL backend. + + Args: + tensor (Tensor): Tensor to send. + dst (int): Destination rank on global process group (regardless of ``group`` argument). + Destination rank should not be the same as the rank of the current process. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + tag (int, optional): Tag to match send with remote recv + group_dst (int, optional): Destination rank on ``group``. Invalid to specify both ``dst`` and ``group_dst``. + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + send, + relevant_args, + tensor, + dst=dst, + group=group, + tag=tag, + group_dst=group_dst, + ) + + group = _group_or_default_group(group) + group_dst = _canonicalize_group_rank(group, dst, group_dst) + _check_not_self_rank(group, group_dst, "destination") + work = isend(tensor, group=group, tag=tag, group_dst=group_dst) + if work is not None: + work.wait() + + +@_exception_logger +def recv( + tensor: torch.Tensor, + src: int | None = None, + group: ProcessGroup | None = None, + tag: int = 0, + group_src: int | None = None, +) -> int: + """ + Receives a tensor synchronously. + + .. warning:: + ``tag`` is not supported with the NCCL backend. + + Args: + tensor (Tensor): Tensor to fill with received data. + src (int, optional): Source rank on global process group (regardless of ``group`` argument). + Will receive from any process if unspecified. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + tag (int, optional): Tag to match recv with remote send + group_src (int, optional): Destination rank on ``group``. Invalid to specify both ``src`` and ``group_src``. + + Returns: + Sender rank + -1, if not part of the group + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + recv, + relevant_args, + tensor, + src=src, + group=group, + tag=tag, + group_src=group_src, + ) + + work = irecv(tensor, src=src, group=group, tag=tag, group_src=group_src) + if work is None: + return -1 + work.wait() + if src is None: + if group_src is None: + group_src = work._source_rank() + group = _group_or_default_group(group) + _check_not_self_rank(group, group_src, "source") + src = get_global_rank(group, group_src) + return src + + +class _IllegalWork(Work): + def __getattribute__(self, name: str) -> object: + if name in [ + "is_success", + "exception", + "wait", + "source_rank", + "_source_rank", + "result", + "synchronize", + ]: + raise ValueError(f"Illegal to call {name} on IllegalWork object") + + +class _CoalescingManager: + def __init__(self) -> None: + self.works: list[Work] = [] + + def append(self, work: Work | None = None) -> None: + if work: + self.works.append(work) + + def wait(self) -> None: + for work in self.works: + work.wait() + + +@contextlib.contextmanager +def _coalescing_manager( + group: ProcessGroup | None = None, + device: torch.device | None = None, + async_ops: bool = False, +) -> collections.abc.Iterator[_CoalescingManager]: + """ + Context manager used to coalesce collectives or P2P operations when possible. + + Args: + group (`ProcessGroup`, optional): The process group to work on. If None, + the default process group will be used. + device (`torch.device`, optional): Default is None, set to a device if + there isn't a `**_coalesced` implementation by the backend. + async_ops (`bool`, optional): whether the coalesced ops are async ops. + + Examples: + >>> # xdoctest: +SKIP("no rank") + >>> # Synchronous ops + >>> with _coalescing_manager(): + >>> for i in range(num_colls): + >>> dist.all_reduce(tensors[i]) + >>> # Asynchronous ops + >>> with _coalescing_manager(async_ops=True) as cm: + >>> for i in range(num_colls): + >>> dist.all_reduce(tensors[i]) + >>> cm.wait() + + .. warning:: + :func:`_coalescing_manager` currently do not support coalescing + all-reduces with different reduce operators, e.g. `ReduceOp.SUM` mixed + with `ReduceOp.PRODUCT`. + """ + group = group or _get_default_group() + op_list = _world.pg_coalesce_state.setdefault(group, []) + if op_list: + raise ValueError( + "ProcessGroup has non-empty op list at the start of coalescing" + ) + if device: + group._start_coalescing(device) + cm = _CoalescingManager() + yield cm + work = None + op_list = _world.pg_coalesce_state.pop(group) + if op_list: + # Collectives supporting "Fast Path" coalescing are captured. + # See implementation in corresponding collective APIs. + # Currently supported: + # - coalesced `all_reduce` + # - coalesced `all_gather_single` + # - coalesced `reduce_scatter_single` + op0 = op_list[0].op + if any(op.op is not op0 for op in op_list): + raise RuntimeError( + "Coalescing manager requires all collectives to be the same type, " + f"but got mixed types: {set(op.op.__name__ for op in op_list)}" # noqa: C401 + ) + + if op0 is all_reduce: + tensors = [op.tensor for op in op_list] + all_reduce_opts = AllreduceCoalescedOptions() + all_reduce_opts.reduceOp = not_none(op_list[0].redop) + all_reduce_opts.asyncOp = async_ops + work = group.allreduce_coalesced(tensors, all_reduce_opts) + elif op0 is all_gather_single: + inputs = [] + outputs = [] + for op in op_list: + inputs.append(op.tensor) + outputs.append(not_none(op.dst_tensor)) + all_gather_opts = AllgatherOptions() + all_gather_opts.asyncOp = async_ops + work = group.all_gather_single_coalesced(outputs, inputs, all_gather_opts) + elif op0 is reduce_scatter_single: + inputs = [] + outputs = [] + for op in op_list: + inputs.append(op.tensor) + outputs.append(not_none(op.dst_tensor)) + reduce_opts = ReduceScatterOptions() + reduce_opts.reduceOp = not_none(op_list[0].redop) + reduce_opts.asyncOp = async_ops + work = group.reduce_scatter_single_coalesced(outputs, inputs, reduce_opts) + else: + raise AssertionError( + f"Coalescing manager does not support fast-path coalescing of {op0}, " + f"yet {op0} is still recorded in op list. This is an internal error of c10d." + ) + + if device: + # Old style of letting each coll inside the context manager to call into C++ counterpart via python binding + work = group._end_coalescing(device) + + if async_ops: + cm.append(work) + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + + +class _TimeEstimator: + def __init__(self) -> None: + self.estimated_time: float | None = None + + +@contextlib.contextmanager +def _time_estimator( + group: ProcessGroup | None = None, + device: torch.device | None = None, +) -> collections.abc.Iterator[_TimeEstimator]: + """ + Context manager used to estimate time of collectives. + Within the context manager, nothing is actually run and the backend just simulates + the collective time only. + + Args: + group (`ProcessGroup`, optional): The process group to work on. If None, + the default process group will be used. + device (`torch.device`, optional): Default is None, set to a device if + there isn't a `**_coalesced` implementation by the backend. + + Examples: + >>> # xdoctest: +SKIP("no rank") + >>> # Synchronous ops + >>> with _time_estimator() as cm: + >>> for i in range(num_colls): + >>> dist.all_reduce(tensors[i]) + >>> # estimate time is stored in cm.estimated_time + + .. warning:: + :func:`_time_estimator` currently only support NCCL backend but it can + easily be extended to other backends. + + Also a NCCL communicator needs to be created because only with a real communicator can we do accurate estimation. + The communicator internally has knowledge about the links it runs on + (e.g. intra-node or inter-node, whether the links are NVLink or PCI-e or IB). + """ + # TODO: We need to also support torch inductor for the time estimator. + group = group or _get_default_group() + device = device or _get_pg_default_device(group) + backend = group._get_backend(device) + if not backend.supports_time_estimate: + raise NotImplementedError( + f"collective time estimator is not supported in the current version of backend {backend}" + ) + backend._start_time_estimate() # type: ignore[attr-defined] + cm = _TimeEstimator() + yield cm + cm.estimated_time = backend._end_time_estimate() # type: ignore[attr-defined] + + +def batch_isend_irecv(p2p_op_list: list[P2POp]) -> list[Work]: + """ + Send or Receive a batch of tensors asynchronously and return a list of requests. + + Process each of the operations in ``p2p_op_list`` and return the corresponding + requests. NCCL, Gloo, and UCC backend are currently supported. + + Args: + p2p_op_list: A list of point-to-point operations(type of each operator is + ``torch.distributed.P2POp``). All operations in the list are treated as + a single batch -- the relative ordering of sends vs. receives in the + list does not matter and will not cause deadlocks. For example, + ``[recv_from_1, send_to_1]`` is equivalent to ``[send_to_1, recv_from_1]``. + + The ordering **does** matter for multiple operations between the same + pair of ranks in the same direction: if rank 0 calls + ``[send_to_1(A), send_to_1(B)]``, rank 1 must call + ``[recv_from_0(A), recv_from_0(B)]`` in the same order to ensure + the correct tensors are matched. + + Returns: + A list of distributed request objects returned by calling the corresponding + op in the op_list. + + Examples: + >>> # xdoctest: +SKIP("no rank") + >>> send_tensor = torch.arange(2, dtype=torch.float32) + 2 * rank + >>> recv_tensor = torch.randn(2, dtype=torch.float32) + >>> send_op = dist.P2POp(dist.isend, send_tensor, (rank + 1) % world_size) + >>> recv_op = dist.P2POp( + ... dist.irecv, recv_tensor, (rank - 1 + world_size) % world_size + ... ) + >>> reqs = batch_isend_irecv([send_op, recv_op]) + >>> for req in reqs: + >>> req.wait() + >>> recv_tensor + tensor([2, 3]) # Rank 0 + tensor([0, 1]) # Rank 1 + + .. note:: Note that when this API is used with the NCCL PG backend, users must set + the current GPU device with `torch.cuda.set_device`, otherwise it will + lead to unexpected hang issues. + + In addition, if this API is the first collective call in the ``group`` + passed to ``dist.P2POp``, all ranks of the ``group`` must participate in + this API call; otherwise, the behavior is undefined. If this API call is + not the first collective call in the ``group``, batched P2P operations + involving only a subset of ranks of the ``group`` are allowed. + """ + _check_p2p_op_list(p2p_op_list) + group = p2p_op_list[0].group + if group is None: + group = _get_default_group() + device = p2p_op_list[0].tensor.device + + def peer_kwarg(op: P2POp) -> dict[str, int]: + key = "group_dst" if op.op is isend else "group_src" + return {key: op.group_peer} + + if isinstance(group, ProcessGroup) and getattr( + group._get_backend(device), "supports_coalescing", False + ): + # NCCL style coalescing + with _coalescing_manager(group, device, async_ops=True) as cm: + for p2p_op in p2p_op_list: + p2p_op.op( + p2p_op.tensor, + group=p2p_op.group, + tag=p2p_op.tag, + **peer_kwarg(p2p_op), + ) + + return cm.works + else: + # backend not support coalescing + reqs = [] + for p2p_op in p2p_op_list: + work = p2p_op.op( + p2p_op.tensor, + group=p2p_op.group, + tag=p2p_op.tag, + **peer_kwarg(p2p_op), + ) + if work: + reqs.append(work) + return reqs + + +def _is_fp8(tensor: torch.Tensor) -> bool: + return tensor.dtype in ( + torch.float8_e4m3fn, + torch.float8_e4m3fnuz, + torch.float8_e5m2, + torch.float8_e5m2fnuz, + ) + + +@_exception_logger +def broadcast( + tensor: torch.Tensor, + src: int | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, + group_src: int | None = None, +) -> Work | None: + """ + Broadcasts the tensor to the whole group. + + ``tensor`` must have the same number of elements in all processes + participating in the collective. + + Args: + tensor (Tensor): Data to be sent if ``src`` is the rank of current + process, and tensor to be used to save received data otherwise. + src (int): Source rank on global process group (regardless of ``group`` argument). + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + group_src (int): Source rank on ``group``. Must specify one of ``group_src`` + and ``src`` but not both. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + broadcast, + relevant_args, + tensor, + src=src, + group=group, + async_op=async_op, + group_src=group_src, + ) + + group = _group_or_default_group(group) + group_src = _canonicalize_group_rank(group, src, group_src, return_global=False) + _check_single_tensor(tensor, "tensor") + if _rank_not_in_group(group): + _warn_not_in_group("broadcast") + return None + + opts = BroadcastOptions() + opts.rootRank = group_src + opts.rootTensor = 0 + opts.asyncOp = async_op + sm90_or_more = not ( + tensor.is_cuda and torch.cuda.get_device_capability(tensor.device)[0] >= 9 + ) + if tensor.is_complex(): + tensor = torch.view_as_real(tensor) + elif _is_fp8(tensor) and not sm90_or_more: + # FP8 is supported by NCCL on sm90+, use workaround for older GPUs + tensor = tensor.view(torch.uint8) + work = group.broadcast([tensor], opts) + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@overload +def all_reduce( + tensor: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + *, + async_op: Literal[True], +) -> Work: ... + + +@overload +def all_reduce( + tensor: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: ... + + +@_exception_logger +def all_reduce( + tensor: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Reduces the tensor data across all machines in a way that all get the final result. + + After the call ``tensor`` is going to be bitwise identical in all processes. + + Complex tensors are supported. + + Args: + tensor (Tensor): Input and output of the collective. The function + operates in-place. + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + Examples: + >>> # xdoctest: +SKIP("no rank") + >>> # All tensors below are of torch.int64 type. + >>> # We have 2 process groups, 2 ranks. + >>> device = torch.device(f"cuda:{rank}") + >>> tensor = torch.arange(2, dtype=torch.int64, device=device) + 1 + 2 * rank + >>> tensor + tensor([1, 2], device='cuda:0') # Rank 0 + tensor([3, 4], device='cuda:1') # Rank 1 + >>> dist.all_reduce(tensor, op=ReduceOp.SUM) + >>> tensor + tensor([4, 6], device='cuda:0') # Rank 0 + tensor([4, 6], device='cuda:1') # Rank 1 + + >>> # All tensors below are of torch.cfloat type. + >>> # We have 2 process groups, 2 ranks. + >>> tensor = torch.tensor( + ... [1 + 1j, 2 + 2j], dtype=torch.cfloat, device=device + ... ) + 2 * rank * (1 + 1j) + >>> tensor + tensor([1.+1.j, 2.+2.j], device='cuda:0') # Rank 0 + tensor([3.+3.j, 4.+4.j], device='cuda:1') # Rank 1 + >>> dist.all_reduce(tensor, op=ReduceOp.SUM) + >>> tensor + tensor([4.+4.j, 6.+6.j], device='cuda:0') # Rank 0 + tensor([4.+4.j, 6.+6.j], device='cuda:1') # Rank 1 + + """ + # Dynamo has built-in logic to map legacy distributed ops to functional collectives. + # Let's redirect to a torch function mode that can mimic this logic outside Dynamo + # (e.g., non-strict export implements such a torch function mode). + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + all_reduce, + relevant_args, + tensor, + op=op, + group=group, + async_op=async_op, # pyrefly: ignore[bad-argument-type] + ) + + _check_single_tensor(tensor, "tensor") + if _rank_not_in_group(group): + _warn_not_in_group("all_reduce") + return None + + if tensor.is_complex(): + if not supports_complex(op): + raise ValueError(f"all_reduce does not support {op} on complex tensors") + tensor = torch.view_as_real(tensor) + + opts = AllreduceOptions() + opts.reduceOp = op + opts.asyncOp = async_op + if group is None: + group = _get_default_group() + + if group in _world.pg_coalesce_state: + # We are in coalescing context, do not issue single operation, just append a collective representation + coll = _CollOp(all_reduce, tensor, None, op, None) + _world.pg_coalesce_state[group].append(coll) + if async_op: + return _IllegalWork() + else: + return None + + work = group.allreduce([tensor], opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +@deprecated( + "`torch.distributed.all_reduce_coalesced` will be deprecated. If you must " + "use it, please revisit our documentation later at " + "https://pytorch.org/docs/main/distributed.html#collective-functions", + category=FutureWarning, +) +def all_reduce_coalesced( + tensors: list[torch.Tensor], + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> torch.Future | None: + """ + WARNING: at this time individual shape checking is not implemented across nodes. + + For example, if the rank 0 node passes [torch.rand(4), torch.rand(2)] and the + rank 1 node passes [torch.rand(2), torch.rand(2), torch.rand(2)], the allreduce + operation will proceed without complaint and return erroneous outputs. This lack + of shape checking results in significant performance improvements but users of this + function should take extra care to ensure that each node passes in tensors whose + shapes match across nodes. + + Reduces each tensor in tensors (residing on the same device) across all machines + in such a way that all get the final result. + + After the call each tensor in tensors is going to bitwise identical + in all processes. + + Complex tensors are supported. + + Args: + tensors (Union[List[Tensor], Tensor]): Input and output of the collective. + The function operates in-place. + op (Optional[ReduceOp]): One of the values from + ``torch.distributed.ReduceOp`` enum. Specifies an operation used for + element-wise reductions. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (Optional[bool]): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + """ + if isinstance(tensors, torch.Tensor): + tensors = [tensors] + relevant_args = tuple(tensors) if isinstance(tensors, (list, tuple)) else (tensors,) + if has_torch_function(relevant_args): + return handle_torch_function( + all_reduce_coalesced, + relevant_args, + tensors, + op=op, + group=group, + async_op=async_op, + ) + + _check_tensor_list(tensors, "tensor") + _ensure_all_tensors_same_dtype(tensors) + if _rank_not_in_group(group): + _warn_not_in_group("all_reduce_coalesced") + return + + if any(t.is_complex() for t in tensors) and not supports_complex(op): + raise ValueError(f"all_reduce does not support {op} on complex tensors") + + tensors = [t if not t.is_complex() else torch.view_as_real(t) for t in tensors] + + opts = AllreduceCoalescedOptions() + opts.reduceOp = op + opts.asyncOp = async_op + group = group or _get_default_group() + work = group.allreduce_coalesced(tensors, opts) + + if async_op: + return work.get_future() + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + + +@_exception_logger +def reduce( + tensor: torch.Tensor, + dst: int | None = None, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, + group_dst: int | None = None, +) -> Work | None: + """ + Reduces the tensor data across all machines. + + Only the process with rank ``dst`` is going to receive the final result. + + Args: + tensor (Tensor): Input and output of the collective. The function + operates in-place. + dst (int): Destination rank on global process group (regardless of ``group`` argument) + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + group_dst (int): Destination rank on ``group``. Must specify one of ``group_dst`` + and ``dst`` but not both. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + reduce, + relevant_args, + tensor, + dst=dst, + op=op, + group=group, + async_op=async_op, + group_dst=group_dst, + ) + + group = _group_or_default_group(group) + group_dst = _canonicalize_group_rank(group, dst, group_dst, return_global=False) + _check_single_tensor(tensor, "tensor") + if _rank_not_in_group(group): + _warn_not_in_group("reduce") + return None + + opts = ReduceOptions() + opts.reduceOp = op + opts.rootRank = group_dst + opts.asyncOp = async_op + work = group.reduce([tensor], opts) + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +def _object_to_tensor( + obj: object, + device: torch.device | str, + group: ProcessGroup | None, + weights_only: bool = False, +) -> tuple[torch.Tensor, torch.Tensor]: + with _WaitCounter("pytorch.wait_counter.c10d._object_to_tensor").guard(): + f = io.BytesIO() + if weights_only: + torch.save(obj, f) + else: + _pickler(f).dump(obj) + byte_storage = torch.ByteStorage._from_buffer(f.getvalue()) # type: ignore[attr-defined] + # Do not replace `torch.ByteTensor` or `torch.LongTensor` with torch.tensor and specifying dtype. + # Otherwise, it will cause 100X slowdown. + # See: https://github.com/pytorch/pytorch/issues/65696 + byte_tensor = torch.ByteTensor(byte_storage).to(device) + if get_debug_level() == DebugLevel.DETAIL and is_nccl_available(): + backend = get_backend(group) + if backend == Backend.NCCL: + hash = torch._C._distributed_c10d._hash_tensors([byte_tensor]) + logger.warning( + "_object_to_tensor size: %s hash value: %s", + byte_tensor.numel(), + hash, + ) + local_size = torch.LongTensor([byte_tensor.numel()]).to(device) + return byte_tensor, local_size + + +def _tensor_to_object( + tensor: torch.Tensor, + tensor_size: torch.Tensor | int, + group: ProcessGroup | None, + weights_only: bool = False, +) -> object: + with _WaitCounter("pytorch.wait_counter.c10d._tensor_to_object").guard(): + if get_debug_level() == DebugLevel.DETAIL and is_nccl_available(): + backend = get_backend(group) + if backend == Backend.NCCL: + hash = torch._C._distributed_c10d._hash_tensors([tensor]) + logger.warning( + "_tensor_to_object size: %s hash value: %s", tensor.numel(), hash + ) + tensor = tensor.cpu() + buf = tensor.numpy().tobytes()[:tensor_size] + if weights_only: + return torch.load(io.BytesIO(buf), weights_only=True) + return _unpickler(io.BytesIO(buf)).load() + + +@_exception_logger +def all_gather_object( + object_list: list[_T], + obj: _T, + group: ProcessGroup | None = None, + weights_only: bool = False, +) -> None: + """ + Gathers picklable objects from the whole group into a list. + + Similar to :func:`all_gather`, but Python objects can be passed in. + Note that the object must be picklable in order to be gathered. + + Args: + object_list (list[Any]): Output list. It should be correctly sized as the + size of the group for this collective and will contain the output. + obj (Any): Pickable Python object to be broadcast from current process. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + weights_only (bool, optional): If ``True``, objects are serialized with + ``torch.save`` and deserialized with ``torch.load(weights_only=True)``, + which restricts deserialization to safe types. If ``False``, ``pickle`` + is used, which is insecure with untrusted data. All ranks must pass + the same value. Default is ``False``. + + Returns: + None. If the calling rank is part of this group, the output of the + collective will be populated into the input ``object_list``. If the + calling rank is not part of the group, the passed in ``object_list`` will + be unmodified. + + .. note:: Note that this API differs slightly from the :func:`all_gather` + collective since it does not provide an ``async_op`` handle and thus + will be a blocking call. + + .. note:: For NCCL-based processed groups, internal tensor representations + of objects must be moved to the GPU device before communication takes + place. In this case, the device used is given by + ``torch.cuda.current_device()`` and it is the user's responsibility to + ensure that this is set so that each rank has an individual GPU, via + ``torch.cuda.set_device()``. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`all_gather_object` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`all_gather_object` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`all_gather` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> # Assumes world_size of 3. + >>> gather_objects = ["foo", 12, {1: 2}] # any picklable object + >>> output = [None for _ in gather_objects] + >>> dist.all_gather_object(output, gather_objects[dist.get_rank()]) + >>> output + ['foo', 12, {1: 2}] + """ + if _rank_not_in_group(group): + _warn_not_in_group("all_gather_object") + return + + current_device = _get_object_coll_device(group) + input_tensor, local_size = _object_to_tensor( + obj, current_device, group, weights_only + ) + + # Gather all local sizes. This is so that we can find the max size, and index + # until the correct size when deserializing the tensors. + group_size = get_world_size(group=group) + object_sizes_tensor = torch.zeros( + group_size, dtype=torch.long, device=current_device + ) + object_size_list = [ + object_sizes_tensor[i].unsqueeze(dim=0) for i in range(group_size) + ] + # Allgather tensor sizes + all_gather(object_size_list, local_size, group=group) + max_object_size = int(max(object_size_list).item()) # type: ignore[type-var] + # Resize tensor to max size across all ranks. + input_tensor.resize_(max_object_size) + coalesced_output_tensor = torch.empty( + max_object_size * group_size, dtype=torch.uint8, device=current_device + ) + # Output tensors are nonoverlapping views of coalesced_output_tensor + output_tensors = [ + coalesced_output_tensor[max_object_size * i : max_object_size * (i + 1)] + for i in range(group_size) + ] + all_gather(output_tensors, input_tensor, group=group) + # Deserialize outputs back to object. + for i, tensor in enumerate(output_tensors): + tensor = tensor.type(torch.uint8) + tensor_size = object_size_list[i] + object_list[i] = cast( + _T, _tensor_to_object(tensor, tensor_size, group, weights_only) + ) + + +@_exception_logger +def gather_object( + obj: _T, + object_gather_list: list[_T] | None = None, + dst: int | None = None, + group: ProcessGroup | None = None, + group_dst: int | None = None, + weights_only: bool = False, +) -> None: + """ + Gathers picklable objects from the whole group in a single process. + + Similar to :func:`gather`, but Python objects can be passed in. Note that the + object must be picklable in order to be gathered. + + Args: + obj (Any): Input object. Must be picklable. + object_gather_list (list[Any]): Output list. On the ``dst`` rank, it + should be correctly sized as the size of the group for this + collective and will contain the output. Must be ``None`` on non-dst + ranks. (default is ``None``) + dst (int, optional): Destination rank on global process group (regardless of ``group`` argument). + (If both ``dst`` and ``group_dst`` are None, default is global rank 0) + group: (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + group_dst (int, optional): Destination rank on ``group``. Invalid to specify both ``dst`` and ``group_dst`` + weights_only (bool, optional): If ``True``, objects are serialized with + ``torch.save`` and deserialized with ``torch.load(weights_only=True)``, + which restricts deserialization to safe types. If ``False``, ``pickle`` + is used, which is insecure with untrusted data. All ranks must pass + the same value. Default is ``False``. + + Returns: + None. On the ``dst`` rank, ``object_gather_list`` will contain the + output of the collective. + + .. note:: Note that this API differs slightly from the gather collective + since it does not provide an async_op handle and thus will be a blocking + call. + + .. note:: For NCCL-based processed groups, internal tensor representations + of objects must be moved to the GPU device before communication takes + place. In this case, the device used is given by + ``torch.cuda.current_device()`` and it is the user's responsibility to + ensure that this is set so that each rank has an individual GPU, via + ``torch.cuda.set_device()``. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`gather_object` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`gather_object` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`gather` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> # Assumes world_size of 3. + >>> gather_objects = ["foo", 12, {1: 2}] # any picklable object + >>> output = [None for _ in gather_objects] + >>> dist.gather_object( + ... gather_objects[dist.get_rank()], + ... output if dist.get_rank() == 0 else None, + ... dst=0 + ... ) + >>> # On rank 0 + >>> output + ['foo', 12, {1: 2}] + """ + group = _group_or_default_group(group) + if dst is None and group_dst is None: + dst = 0 + group_dst = _canonicalize_group_rank(group, dst, group_dst, return_global=False) + if _rank_not_in_group(group): + _warn_not_in_group("gather_object") + return + + # Ensure object_gather_list is specified appropriately. + my_group_rank = group.rank() + _validate_output_list_for_rank(my_group_rank, group_dst, object_gather_list) + current_device = _get_object_coll_device(group) + input_tensor, local_size = _object_to_tensor( + obj, current_device, group, weights_only + ) + + # Gather all local sizes. This is so that we can find the max size, and index + # until the correct size when deserializing the tensors. + group_size = get_world_size(group=group) + object_sizes_tensor = torch.zeros( + group_size, dtype=torch.long, device=current_device + ) + object_size_list = [ + object_sizes_tensor[i].unsqueeze(dim=0) for i in range(group_size) + ] + # Allgather tensor sizes. An all-gather is needed here despite this being a + # gather, since each rank needs to broadcast a tensor of the same (maximal) + # size. + all_gather(object_size_list, local_size, group=group) + max_object_size = int(max(object_size_list).item()) # type: ignore[type-var] + # Resize tensor to max size across all ranks. + input_tensor.resize_(max_object_size) + # Avoid populating output tensors if the result won't be gathered on this rank. + if my_group_rank == group_dst: + coalesced_output_tensor = torch.empty( + max_object_size * group_size, dtype=torch.uint8, device=current_device + ) + # Output tensors are nonoverlapping views of coalesced_output_tensor + output_tensors = [ + coalesced_output_tensor[max_object_size * i : max_object_size * (i + 1)] + for i in range(group_size) + ] + # All ranks call gather with equal-sized tensors. + gather( + input_tensor, + gather_list=output_tensors if my_group_rank == group_dst else None, # type: ignore[possibly-undefined] + group_dst=group_dst, + group=group, + ) + if my_group_rank != group_dst: + return + + if object_gather_list is None: + raise AssertionError("Must provide object_gather_list on dst rank") + # pyrefly: ignore [unbound-name] + for i, tensor in enumerate(output_tensors): + tensor = tensor.type(torch.uint8) + tensor_size = object_size_list[i] + object_gather_list[i] = cast( + _T, _tensor_to_object(tensor, tensor_size, group, weights_only) + ) + + +@_exception_logger +def send_object_list( + object_list: Sequence[object], + dst: int | None = None, + group: ProcessGroup | None = None, + device: torch.device | None = None, + group_dst: int | None = None, + use_batch: bool = False, + weights_only: bool = False, +) -> None: + """ + Sends picklable objects in ``object_list`` synchronously. + + Similar to :func:`send`, but Python objects can be passed in. + Note that all objects in ``object_list`` must be picklable in order to be + sent. + + Args: + object_list (List[Any]): List of input objects to sent. + Each object must be picklable. Receiver must provide lists of equal sizes. + dst (int): Destination rank to send ``object_list`` to. + Destination rank is based on global process group (regardless of ``group`` argument) + group: (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + device (``torch.device``, optional): If not None, the objects are + serialized and converted to tensors which are moved to the + ``device`` before sending. Default is ``None``. + group_dst (int, optional): Destination rank on ``group``. + Must specify one of ``dst`` and ``group_dst`` but not both + use_batch (bool, optional): If True, use batch p2p operations instead of + regular send operations. This avoids initializing 2-rank communicators and + uses existing entire group communicators. See batch_isend_irecv for usage and + assumptions. Default is ``False``. + weights_only (bool, optional): If ``True``, objects are serialized with + ``torch.save``, which the receiver can deserialize with + ``torch.load(weights_only=True)``. If ``False``, ``pickle`` is used, + which is insecure with untrusted data. Must match the value passed + to :func:`recv_object_list` on the receiver. Default is ``False``. + Returns: + ``None``. + + .. note:: For NCCL-based process groups, internal tensor representations + of objects must be moved to the GPU device before communication takes + place. In this case, the device used is given by + ``torch.cuda.current_device()`` and it is the user's responsibility to + ensure that this is set so that each rank has an individual GPU, via + ``torch.cuda.set_device()``. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`send_object_list` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`send_object_list` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`send` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> # Assumes backend is not NCCL + >>> device = torch.device("cpu") + >>> if dist.get_rank() == 0: + >>> # Assumes world_size of 2. + >>> objects = ["foo", 12, {1: 2}] # any picklable object + >>> dist.send_object_list(objects, dst=1, device=device) + >>> else: + >>> objects = [None, None, None] + >>> dist.recv_object_list(objects, src=0, device=device) + >>> objects + ['foo', 12, {1: 2}] + """ + group = _group_or_default_group(group) + group_dst = _canonicalize_group_rank(group, dst, group_dst) + _check_not_self_rank(group, group_dst, "destination") + + if _rank_not_in_group(group): + _warn_not_in_group("send_object_list") + return + + # Current device selection. + # To preserve backwards compatibility, ``device`` is default to ``None`` + # in which case we run current logic of device selection, i.e. + # ``current_device`` is CUDA if backend is NCCL otherwise CPU device. In the + # case it is not ``None`` we move the size and object tensors to be + # sent to this device. + current_device = device or _get_object_coll_device(group) + # Serialize object_list elements to tensors on src rank. + tensor_list, size_list = zip( + *[ + _object_to_tensor(obj, current_device, group, weights_only) + for obj in object_list + ] + ) + object_sizes_tensor = torch.cat(size_list) + + # Send object sizes + if use_batch: + batch_isend_irecv( + [P2POp(isend, object_sizes_tensor, group_peer=group_dst, group=group)] + ).pop().wait() + else: + send(object_sizes_tensor, group_dst=group_dst, group=group) + + # Concatenate and send serialized object tensors + # Note: torch.cat will do an extra memory copy to the current device, if the tensor_list + # has only one element, we can skip the copy. + if len(tensor_list) == 1: # type: ignore[possibly-undefined] + object_tensor = tensor_list[0] + else: + object_tensor = torch.cat(tensor_list) + + if use_batch: + batch_isend_irecv( + [P2POp(isend, object_tensor, group_peer=group_dst, group=group)] + ).pop().wait() + else: + send(object_tensor, group_dst=group_dst, group=group) + + +@_exception_logger +def recv_object_list( + object_list: list[_T], + src: int | None = None, + group: ProcessGroup | None = None, + device: torch.device | None = None, + group_src: int | None = None, + use_batch: bool = False, + weights_only: bool = False, +) -> int: + """ + Receives picklable objects in ``object_list`` synchronously. + + Similar to :func:`recv`, but can receive Python objects. + + Args: + object_list (List[Any]): List of objects to receive into. + Must provide a list of sizes equal to the size of the list being sent. + src (int, optional): Source rank from which to recv ``object_list``. + Source rank is based on global process group (regardless of ``group`` argument) + Will receive from any rank if set to None. Default is ``None``. + group: (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + device (``torch.device``, optional): If not None, receives on this device. + Default is ``None``. + group_src (int, optional): Destination rank on ``group``. Invalid to specify both ``src`` and ``group_src``. + use_batch (bool, optional): If True, use batch p2p operations instead of + regular send operations. This avoids initializing 2-rank communicators and + uses existing entire group communicators. See batch_isend_irecv for usage and + assumptions. Default is ``False``. + weights_only (bool, optional): If ``True``, objects are deserialized with + ``torch.load(weights_only=True)``, which restricts deserialization to + safe types. If ``False``, ``pickle`` is used, which is insecure with + untrusted data. Must match the value passed to + :func:`send_object_list` on the sender. Default is ``False``. + + Returns: + Sender rank. -1 if rank is not part of the group. If rank is part of the group, + ``object_list`` will contain the sent objects from ``src`` rank. + + .. note:: For NCCL-based process groups, internal tensor representations + of objects must be moved to the GPU device before communication takes + place. In this case, the device used is given by + ``torch.cuda.current_device()`` and it is the user's responsibility to + ensure that this is set so that each rank has an individual GPU, via + ``torch.cuda.set_device()``. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`recv_object_list` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`recv_object_list` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`recv` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> # Assumes backend is not NCCL + >>> device = torch.device("cpu") + >>> if dist.get_rank() == 0: + >>> # Assumes world_size of 2. + >>> objects = ["foo", 12, {1: 2}] # any picklable object + >>> dist.send_object_list(objects, dst=1, device=device) + >>> else: + >>> objects = [None, None, None] + >>> dist.recv_object_list(objects, src=0, device=device) + >>> objects + ['foo', 12, {1: 2}] + """ + group = _group_or_default_group(group) + group_src = _canonicalize_group_rank(group, src, group_src) + _check_not_self_rank(group, group_src, "source") + + if _rank_not_in_group(group): + _warn_not_in_group("recv_object_list") + return -1 + + # Current device selection. + # To preserve backwards compatibility, ``device`` is default to ``None`` + # in which case we run current logic of device selection, i.e. + # ``current_device`` is CUDA if backend is NCCL otherwise CPU device. In the + # case it is not ``None`` we move the size and object tensors to be + # received to this device. + current_device = device or _get_object_coll_device(group) + object_sizes_tensor = torch.empty( + len(object_list), dtype=torch.long, device=current_device + ) + + # Receive object sizes + if use_batch: + work = batch_isend_irecv( + [ + P2POp( + irecv, + object_sizes_tensor, + group_peer=group_src, + group=group, + ) + ] + ).pop() + work.wait() + rank_sizes = get_global_rank(group, group_src) + else: + rank_sizes = recv(object_sizes_tensor, group=group, group_src=group_src) + + # Tensor to receive serialized objects into. + object_tensor = torch.empty( # type: ignore[call-overload] + torch.sum(object_sizes_tensor).item(), # type: ignore[arg-type] + dtype=torch.uint8, + device=current_device, + ) + + if use_batch: + work = batch_isend_irecv( + [ + P2POp( + irecv, + object_tensor, + group_peer=group_src, + group=group, + ) + ] + ).pop() + work.wait() + rank_objects = get_global_rank(group, group_src) + else: + rank_objects = recv(object_tensor, group=group, group_src=group_src) + if rank_sizes != rank_objects: + raise AssertionError("Mismatch in return ranks for object sizes and objects.") + # Deserialize objects using their stored sizes. + offset = 0 + for i, obj_size in enumerate(object_sizes_tensor): + obj_view = object_tensor[offset : offset + obj_size] + obj_view = obj_view.type(torch.uint8) + offset += obj_size + object_list[i] = cast( + _T, _tensor_to_object(obj_view, obj_size, group, weights_only) + ) + return rank_objects + + +@_exception_logger +def broadcast_object_list( + object_list: list[_T], + src: int | None = None, + group: ProcessGroup | None = None, + device: torch.device | None = None, + group_src: int | None = None, + weights_only: bool = False, +) -> None: + """ + Broadcasts picklable objects in ``object_list`` to the whole group. + + Similar to :func:`broadcast`, but Python objects can be passed in. + Note that all objects in ``object_list`` must be picklable in order to be + broadcasted. + + Args: + object_list (List[Any]): List of input objects to broadcast. + Each object must be picklable. Only objects on the ``src`` rank will + be broadcast, but each rank must provide lists of equal sizes. + src (int): Source rank from which to broadcast ``object_list``. + Source rank is based on global process group (regardless of ``group`` argument) + group: (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + device (``torch.device``, optional): If not None, the objects are + serialized and converted to tensors which are moved to the + ``device`` before broadcasting. Default is ``None``. + group_src (int): Source rank on ``group``. Must not specify one of ``group_src`` + and ``src`` but not both. + weights_only (bool, optional): If ``True``, objects are serialized with + ``torch.save`` and deserialized with ``torch.load(weights_only=True)``, + which restricts deserialization to safe types. If ``False``, ``pickle`` + is used, which is insecure with untrusted data. All ranks must pass + the same value. Default is ``False``. + + Returns: + ``None``. If rank is part of the group, ``object_list`` will contain the + broadcasted objects from ``src`` rank. + + .. note:: For NCCL-based process groups, internal tensor representations + of objects must be moved to the GPU device before communication takes + place. In this case, the device used is given by + ``torch.cuda.current_device()`` and it is the user's responsibility to + ensure that this is set so that each rank has an individual GPU, via + ``torch.cuda.set_device()``. + + .. note:: Note that this API differs slightly from the :func:`broadcast` + collective since it does not provide an ``async_op`` handle and thus + will be a blocking call. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`broadcast_object_list` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`broadcast_object_list` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`broadcast` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> if dist.get_rank() == 0: + >>> # Assumes world_size of 3. + >>> objects = ["foo", 12, {1: 2}] # any picklable object + >>> else: + >>> objects = [None, None, None] + >>> # Assumes backend is not NCCL + >>> device = torch.device("cpu") + >>> dist.broadcast_object_list(objects, src=0, device=device) + >>> objects + ['foo', 12, {1: 2}] + """ + group = _group_or_default_group(group) + if src is None and group_src is None: + src = 0 + group_src = _canonicalize_group_rank(group, src, group_src, return_global=False) + if _rank_not_in_group(group): + _warn_not_in_group("broadcast_object_list") + return + + # Current device selection. + # To preserve backwards compatibility, ``device`` is default to ``None`` + # in which case we run current logic of device selection, i.e. + # ``current_device`` is CUDA if backend is NCCL otherwise CPU device. In the + # case it is not ``None`` we move the size and object tensors to be + # broadcasted to this device. + current_device = device or _get_object_coll_device(group) + my_group_rank = group.rank() + # Serialize object_list elements to tensors on src rank. + if my_group_rank == group_src: + tensor_list, size_list = zip( + *[ + _object_to_tensor(obj, current_device, group, weights_only) + for obj in object_list + ] + ) + object_sizes_tensor = torch.cat(size_list) + else: + object_sizes_tensor = torch.empty( + len(object_list), dtype=torch.long, device=current_device + ) + + # Broadcast object sizes + broadcast(object_sizes_tensor, group_src=group_src, group=group) + + # Concatenate and broadcast serialized object tensors + # Note: torch.cat will do an extra memory copy to the current device, if the tensor_list + # has only one element, we can skip the copy. + if my_group_rank == group_src: + if len(tensor_list) == 1: # type: ignore[possibly-undefined] + # pyrefly: ignore [unbound-name] + object_tensor = tensor_list[0] + else: + # pyrefly: ignore [unbound-name] + object_tensor = torch.cat(tensor_list) + else: + object_tensor = torch.empty( # type: ignore[call-overload] + torch.sum(object_sizes_tensor).item(), # type: ignore[arg-type] + dtype=torch.uint8, + device=current_device, + ) + + broadcast(object_tensor, group_src=group_src, group=group) + # Deserialize objects using their stored sizes. + offset = 0 + if my_group_rank != group_src: + for i, obj_size in enumerate(object_sizes_tensor): + obj_view = object_tensor[offset : offset + obj_size] + obj_view = obj_view.type(torch.uint8) + offset += obj_size + object_list[i] = cast( + _T, _tensor_to_object(obj_view, obj_size, group, weights_only) + ) + + +@_exception_logger +def scatter_object_list( + scatter_object_output_list: list[_T], + scatter_object_input_list: Sequence[object] | None = None, + src: int | None = None, + group: ProcessGroup | None = None, + group_src: int | None = None, + weights_only: bool = False, +) -> None: + """ + Scatters picklable objects in ``scatter_object_input_list`` to the whole group. + + Similar to :func:`scatter`, but Python objects can be passed in. On + each rank, the scattered object will be stored as the first element of + ``scatter_object_output_list``. Note that all objects in + ``scatter_object_input_list`` must be picklable in order to be scattered. + + Args: + scatter_object_output_list (List[Any]): Non-empty list whose first + element will store the object scattered to this rank. + scatter_object_input_list (List[Any], optional): List of input objects to scatter. + Each object must be picklable. Only objects on the ``src`` rank will + be scattered, and the argument can be ``None`` for non-src ranks. + src (int): Source rank from which to scatter ``scatter_object_input_list``. + Source rank is based on global process group (regardless of ``group`` argument). + (If both ``src`` and ``group_src`` are None, default is global rank 0) + group: (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. Default is ``None``. + group_src (int, optional): Source rank on ``group``. Invalid to specify both ``src`` and ``group_src`` + weights_only (bool, optional): If ``True``, objects are serialized with + ``torch.save`` and deserialized with ``torch.load(weights_only=True)``, + which restricts deserialization to safe types. If ``False``, ``pickle`` + is used, which is insecure with untrusted data. All ranks must pass + the same value. Default is ``False``. + + Returns: + ``None``. If rank is part of the group, ``scatter_object_output_list`` + will have its first element set to the scattered object for this rank. + + .. note:: Note that this API differs slightly from the scatter collective + since it does not provide an ``async_op`` handle and thus will be a + blocking call. + + .. warning:: + Object collectives have a number of serious performance and scalability + limitations. See :ref:`object_collectives` for details. + + .. warning:: + :func:`scatter_object_list` with ``weights_only=False`` uses ``pickle`` + module implicitly, which is known to be insecure. It is possible to + construct malicious pickle data which will execute arbitrary code during + unpickling. Only call this function with data you trust. + + .. warning:: + Calling :func:`scatter_object_list` with GPU tensors is not well supported + and inefficient as it incurs GPU -> CPU transfer since tensors would be + pickled. Please consider using :func:`scatter` instead. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> if dist.get_rank() == 0: + >>> # Assumes world_size of 3. + >>> objects = ["foo", 12, {1: 2}] # any picklable object + >>> else: + >>> # Can be any list on non-src ranks, elements are not used. + >>> objects = [None, None, None] + >>> output_list = [None] + >>> dist.scatter_object_list(output_list, objects, src=0) + >>> # Rank i gets objects[i]. For example, on rank 2: + >>> output_list + [{1: 2}] + """ + group = _group_or_default_group(group) + if src is None and group_src is None: + src = 0 + group_src = _canonicalize_group_rank(group, src, group_src, return_global=False) + if _rank_not_in_group(group): + _warn_not_in_group("scatter_object_list") + return + + if ( + not isinstance(scatter_object_output_list, list) + or len(scatter_object_output_list) < 1 + ): + raise ValueError( + "Expected argument scatter_object_output_list to be a list of size at least 1." + ) + + my_group_rank = group.rank() + pg_device = _get_object_coll_device(group) + if my_group_rank == group_src: + if scatter_object_input_list is None: + raise ValueError( + "source rank must provide non-None scatter_object_input_list" + ) + tensor_list, tensor_sizes = zip( + *[ + _object_to_tensor(obj, pg_device, group, weights_only) + for obj in scatter_object_input_list + ] + ) + tensor_list, tensor_sizes = list(tensor_list), list(tensor_sizes) + + # Src rank broadcasts the maximum tensor size. This is because all ranks are + # expected to call into scatter() with equal-sized tensors. + max_tensor_size = max(tensor_sizes) # type: ignore[possibly-undefined] + for tensor in tensor_list: # type: ignore[possibly-undefined] + tensor.resize_(max_tensor_size) + else: + max_tensor_size = torch.tensor([0], dtype=torch.long, device=pg_device) + broadcast(max_tensor_size, group_src=group_src, group=group) + + # Scatter actual serialized objects + # pyrefly: ignore [no-matching-overload] + output_tensor = torch.empty( + max_tensor_size.item(), dtype=torch.uint8, device=pg_device + ) + scatter( + output_tensor, + scatter_list=None if my_group_rank != group_src else tensor_list, # type: ignore[possibly-undefined] + group_src=group_src, + group=group, + ) + + # Scatter per-object sizes to trim tensors when deserializing back to object + obj_tensor_size = torch.tensor([0], dtype=torch.long, device=pg_device) + scatter( + obj_tensor_size, + scatter_list=None if my_group_rank != group_src else tensor_sizes, # type: ignore[possibly-undefined] + group_src=group_src, + group=group, + ) + + # Deserialize back to object + scatter_object_output_list[0] = cast( + _T, + _tensor_to_object(output_tensor, obj_tensor_size, group, weights_only), + ) + + +@overload +def all_gather( + tensor_list: list[torch.Tensor], + tensor: torch.Tensor, + group: ProcessGroup | C10DBackend | None = None, + *, + async_op: Literal[True], +) -> Work: ... + + +@overload +def all_gather( + tensor_list: list[torch.Tensor], + tensor: torch.Tensor, + group: ProcessGroup | C10DBackend | None = None, + async_op: bool = False, +) -> Work | None: ... + + +@_exception_logger +def all_gather( + tensor_list: list[torch.Tensor], + tensor: torch.Tensor, + group: ProcessGroup | C10DBackend | None = None, + async_op: bool = False, +) -> Work | None: + """ + Gathers tensors from the whole group in a list. + + Complex and uneven sized tensors are supported. + + Args: + tensor_list (list[Tensor]): Output list. It should contain + correctly-sized tensors to be used for output of the collective. + Uneven sized tensors are supported. + tensor (Tensor): Tensor to be broadcast from current process. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + Examples: + >>> # xdoctest: +SKIP("need process group init") + >>> # All tensors below are of torch.int64 dtype. + >>> # We have 2 process groups, 2 ranks. + >>> device = torch.device(f"cuda:{rank}") + >>> tensor_list = [ + ... torch.zeros(2, dtype=torch.int64, device=device) for _ in range(2) + ... ] + >>> tensor_list + [tensor([0, 0], device='cuda:0'), tensor([0, 0], device='cuda:0')] # Rank 0 + [tensor([0, 0], device='cuda:1'), tensor([0, 0], device='cuda:1')] # Rank 1 + >>> tensor = torch.arange(2, dtype=torch.int64, device=device) + 1 + 2 * rank + >>> tensor + tensor([1, 2], device='cuda:0') # Rank 0 + tensor([3, 4], device='cuda:1') # Rank 1 + >>> dist.all_gather(tensor_list, tensor) + >>> tensor_list + [tensor([1, 2], device='cuda:0'), tensor([3, 4], device='cuda:0')] # Rank 0 + [tensor([1, 2], device='cuda:1'), tensor([3, 4], device='cuda:1')] # Rank 1 + + >>> # All tensors below are of torch.cfloat dtype. + >>> # We have 2 process groups, 2 ranks. + >>> tensor_list = [ + ... torch.zeros(2, dtype=torch.cfloat, device=device) for _ in range(2) + ... ] + >>> tensor_list + [tensor([0.+0.j, 0.+0.j], device='cuda:0'), tensor([0.+0.j, 0.+0.j], device='cuda:0')] # Rank 0 + [tensor([0.+0.j, 0.+0.j], device='cuda:1'), tensor([0.+0.j, 0.+0.j], device='cuda:1')] # Rank 1 + >>> tensor = torch.tensor( + ... [1 + 1j, 2 + 2j], dtype=torch.cfloat, device=device + ... ) + 2 * rank * (1 + 1j) + >>> tensor + tensor([1.+1.j, 2.+2.j], device='cuda:0') # Rank 0 + tensor([3.+3.j, 4.+4.j], device='cuda:1') # Rank 1 + >>> dist.all_gather(tensor_list, tensor) + >>> tensor_list + [tensor([1.+1.j, 2.+2.j], device='cuda:0'), tensor([3.+3.j, 4.+4.j], device='cuda:0')] # Rank 0 + [tensor([1.+1.j, 2.+2.j], device='cuda:1'), tensor([3.+3.j, 4.+4.j], device='cuda:1')] # Rank 1 + + """ + # Dynamo has built-in logic to map legacy distributed ops to functional collectives. + # Let's redirect to a torch function mode that can mimic this logic outside Dynamo + # (e.g., non-strict export implements such a torch function mode). + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + all_gather, + relevant_args, + tensor_list, + tensor, + group=group, + async_op=async_op, # pyrefly: ignore[bad-argument-type] + ) + + _check_tensor_list(tensor_list, "tensor_list") + _check_single_tensor(tensor, "tensor") + _ensure_all_tensors_same_dtype(tensor_list, tensor) + if _rank_not_in_group(group): + _warn_not_in_group("all_gather") + return None + + tensor_list = [ + t if not t.is_complex() else torch.view_as_real(t) for t in tensor_list + ] + tensor = tensor if not tensor.is_complex() else torch.view_as_real(tensor) + + group = group or _get_default_group() + opts = AllgatherOptions() + opts.asyncOp = async_op + work = group.allgather( # pyrefly: ignore[missing-attribute] + [tensor_list], [tensor], opts + ) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +def all_gather_single( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: ProcessGroup | C10DBackend | None = None, + async_op: bool = False, +) -> Work | None: + """ + Gather tensors from all ranks and put them in a single output tensor. + + This function requires all tensors to be the same size on each process. + + Args: + output_tensor (Tensor): Output tensor to accommodate tensor elements + from all ranks. It must be correctly sized to have one of the + following forms: + (i) a concatenation of all the input tensors along the primary + dimension; for definition of "concatenation", see ``torch.cat()``; + (ii) a stack of all the input tensors along the primary dimension; + for definition of "stack", see ``torch.stack()``. + Examples below may better explain the supported output forms. + input_tensor (Tensor): Tensor to be gathered from current rank. + Different from the ``all_gather`` API, the input tensors in this + API must have the same size across all ranks. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + Examples: + >>> # xdoctest: +SKIP("need process group init") + >>> # All tensors below are of torch.int64 dtype and on CUDA devices. + >>> # We have two ranks. + >>> device = torch.device(f"cuda:{rank}") + >>> tensor_in = torch.arange(2, dtype=torch.int64, device=device) + 1 + 2 * rank + >>> tensor_in + tensor([1, 2], device='cuda:0') # Rank 0 + tensor([3, 4], device='cuda:1') # Rank 1 + >>> # Output in concatenation form + >>> tensor_out = torch.zeros(world_size * 2, dtype=torch.int64, device=device) + >>> dist.all_gather_single(tensor_out, tensor_in) + >>> tensor_out + tensor([1, 2, 3, 4], device='cuda:0') # Rank 0 + tensor([1, 2, 3, 4], device='cuda:1') # Rank 1 + >>> # Output in stack form + >>> tensor_out2 = torch.zeros(world_size, 2, dtype=torch.int64, device=device) + >>> dist.all_gather_single(tensor_out2, tensor_in) + >>> tensor_out2 + tensor([[1, 2], + [3, 4]], device='cuda:0') # Rank 0 + tensor([[1, 2], + [3, 4]], device='cuda:1') # Rank 1 + """ + # Dynamo has built-in logic to map legacy distributed ops to functional collectives. + # Let's redirect to a torch function mode that can mimic this logic outside Dynamo + # (e.g., non-strict export implements such a torch function mode). + relevant_args = (input_tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + all_gather_single, + relevant_args, + output_tensor, + input_tensor, + group=group, + async_op=async_op, + ) + + _check_single_tensor(input_tensor, "input_tensor") + _check_single_tensor(output_tensor, "output_tensor") + if _rank_not_in_group(group): + _warn_not_in_group("all_gather_single") + return None + + output_tensor = ( + output_tensor + if not output_tensor.is_complex() + else torch.view_as_real(output_tensor) + ) + input_tensor = ( + input_tensor + if not input_tensor.is_complex() + else torch.view_as_real(input_tensor) + ) + + opts = AllgatherOptions() + opts.asyncOp = async_op + + group = group or _get_default_group() + + if group in _world.pg_coalesce_state: + # We are in coalescing context, do not issue single operation, just append a collective representation + coll = _CollOp(all_gather_single, input_tensor, output_tensor) + _world.pg_coalesce_state[group].append(coll) + if async_op: + return _IllegalWork() + else: + return None + + work = group.all_gather_single( # pyrefly: ignore[missing-attribute] + output_tensor, input_tensor, opts + ) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +@deprecated( + "`torch.distributed.all_gather_into_tensor` is deprecated. " + "Please use `torch.distributed.all_gather_single` instead.", + category=FutureWarning, +) +def all_gather_into_tensor( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Gather tensors from all ranks and put them in a single output tensor. + + .. warning:: + `all_gather_into_tensor` is deprecated. Users should use + `all_gather_single` instead. + + """ + return all_gather_single(output_tensor, input_tensor, group, async_op) + + +@_exception_logger +@deprecated( + "`torch.distributed._all_gather_base` is a private function and will be deprecated. " + "Please use `torch.distributed.all_gather_single` instead.", + category=FutureWarning, +) +def _all_gather_base( + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Single tensor all gather. Gathers a single tensor from all ranks, and puts them in a single output tensor. + + Args: + output_tensor (Tensor): Output tensor. It should contain + correctly-sized tensors to be used for output of the collective. + input_tensor (Tensor): Tensor to be broadcast from current process. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + .. warning:: + `_all_gather_base` is a private function. Users should use + `all_gather_single` instead. + + """ + return all_gather_single(output_tensor, input_tensor, group, async_op) + + +@_exception_logger +@deprecated( + "`torch.distributed.all_gather_coalesced` will be deprecated. If you must use it, " + "please revisit our documentation later at " + "https://pytorch.org/docs/main/distributed.html#collective-functions", + category=FutureWarning, +) +def all_gather_coalesced( + output_tensor_lists: list[list[torch.Tensor]], + input_tensor_list: list[torch.Tensor], + group: ProcessGroup | None = None, + async_op: bool = False, +) -> torch.Future | None: + """ + Gathers input tensors from the whole group in a list in a coalesced manner. + + Complex tensors are supported. + + Args: + output_tensor_lists (list[list[Tensor]]): Output list. It should contain + correctly-sized tensors to be used for output of the collective. + input_tensor_list (list[Tensor]): Tensors to be broadcast from + current process. At least one tensor has to be non empty. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + Example:: + + # we have 2 process groups, 2 ranks. + # rank 0 passes: + input_tensor_list = [[[1, 1], [1, 1]], [2], [3, 3]] + output_tensor_lists = [ + [[[-1, -1], [-1, -1]], [-1], [-1, -1]], + [[[-1, -1], [-1, -1]], [-1], [-1, -1]], + ] + # rank 1 passes: + input_tensor_list = [[[3, 3], [3, 3]], [5], [1, 1]] + output_tensor_lists = [ + [[[-1, -1], [-1, -1]], [-1], [-1, -1]], + [[[-1, -1], [-1, -1]], [-1], [-1, -1]], + ] + # both rank 0 and 1 get: + output_tensor_lists = [ + [[[1, 1], [1, 1]], [2], [3, 3]], + [[[3, 3], [3, 3]], [5], [1, 1]], + ] + + WARNING: at this time individual shape checking is not implemented across nodes. + For example, if the rank 0 node passes [torch.rand(4), torch.rand(2)] and the + rank 1 node passes [torch.rand(2), torch.rand(2), torch.rand(2)], the + all_gather_coalesced operation will proceed without complaint and return + erroneous outputs. This lack of shape checking results in significant + performance improvements but users of this function should take extra care + to ensure that each node passes in tensors whose shapes match across nodes. + """ + relevant_args = ( + tuple(input_tensor_list) + if isinstance(input_tensor_list, (list, tuple)) + else (input_tensor_list,) + ) + if has_torch_function(relevant_args): + return handle_torch_function( + all_gather_coalesced, + relevant_args, + output_tensor_lists, + input_tensor_list, + group=group, + async_op=async_op, + ) + + # We only check basic compatibility with C++ params here, C++ code will + # do shape and type checking. + if _rank_not_in_group(group): + _warn_not_in_group("all_gather_coalesced") + return + _check_tensor_list(input_tensor_list, "input_tensor_list") + _ensure_all_tensors_same_dtype(input_tensor_list) + if not isinstance(output_tensor_lists, list): + raise TypeError( + "Invalid function argument: output_tensor_lists should be a list" + ) + for output_tensor_list in output_tensor_lists: + _check_tensor_list(output_tensor_list, "output_tensor_lists") + _ensure_all_tensors_same_dtype(output_tensor_list) + + output_tensor_lists = [ + [t if not t.is_complex() else torch.view_as_real(t) for t in l] + for l in output_tensor_lists + ] + input_tensor_list = [ + t if not t.is_complex() else torch.view_as_real(t) for t in input_tensor_list + ] + + group = group or _get_default_group() + opts = AllgatherOptions() + opts.asyncOp = async_op + work = group.allgather_coalesced(output_tensor_lists, input_tensor_list, opts) + + if async_op: + return work.get_future() + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + + +def _validate_output_list_for_rank( + my_rank: int, dst: int, gather_list: Sequence[object] | None +) -> None: + if dst == my_rank: + if not gather_list: + raise ValueError( + "Argument ``gather_list`` must be specified on destination rank." + ) + elif gather_list: + raise ValueError( + "Argument ``gather_list`` must NOT be specified on non-destination ranks." + ) + + +@_exception_logger +def gather( + tensor: torch.Tensor, + gather_list: list[torch.Tensor] | None = None, + dst: int | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, + group_dst: int | None = None, +) -> Work | None: + """ + Gathers a list of tensors in a single process. + + This function requires all tensors to be the same size on each process. + + Args: + tensor (Tensor): Input tensor. + gather_list (list[Tensor], optional): List of appropriately, + same-sized tensors to use for gathered data + (default is None, must be specified on the destination rank) + dst (int, optional): Destination rank on global process group (regardless of ``group`` argument). + (If both ``dst`` and ``group_dst`` are None, default is global rank 0) + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + group_dst (int, optional): Destination rank on ``group``. Invalid to specify both ``dst`` and ``group_dst`` + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + .. note:: Note that all Tensors in gather_list must have the same size. + + Example:: + >>> # xdoctest: +SKIP("no rank") + >>> # We have 2 process groups, 2 ranks. + >>> tensor_size = 2 + >>> device = torch.device(f'cuda:{rank}') + >>> tensor = torch.ones(tensor_size, device=device) + rank + >>> if dist.get_rank() == 0: + >>> gather_list = [torch.zeros_like(tensor, device=device) for i in range(2)] + >>> else: + >>> gather_list = None + >>> dist.gather(tensor, gather_list, dst=0) + >>> # Rank 0 gets gathered data. + >>> gather_list + [tensor([1., 1.], device='cuda:0'), tensor([2., 2.], device='cuda:0')] # Rank 0 + None # Rank 1 + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + gather, + relevant_args, + tensor, + gather_list=gather_list, + dst=dst, + group=group, + async_op=async_op, + group_dst=group_dst, + ) + + _check_single_tensor(tensor, "tensor") + + # Parameter ``gather_list`` may be left unspecified on non-dst ranks. + if gather_list: + _check_tensor_list(gather_list, "gather_list") + else: + gather_list = [] + _ensure_all_tensors_same_dtype(tensor, gather_list) + group = _group_or_default_group(group) + if _rank_not_in_group(group): + _warn_not_in_group("gather") + return None + if dst is None and group_dst is None: + dst = 0 + group_dst = _canonicalize_group_rank(group, dst, group_dst, return_global=False) + my_group_rank = group.rank() + _validate_output_list_for_rank(my_group_rank, group_dst, gather_list) + if group_dst == my_group_rank: + output_tensors = [gather_list] + elif _use_torchcomms_enabled(): + # TorchComms BackendWrapper::gather requires exactly one entry in + # outputTensors on all ranks (including non-dst), so we pass [[]] instead + # of [] to satisfy the TORCH_CHECK(outputTensors.size() == 1) guard. + output_tensors = [[]] + else: + output_tensors = [] + input_tensors = [tensor] + + opts = GatherOptions() + opts.rootRank = group_dst + opts.asyncOp = async_op + work = group.gather(output_tensors, input_tensors, opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +def gather_single( + tensor: torch.Tensor, + gather_tensor: torch.Tensor | None = None, + dst: int | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, + group_dst: int | None = None, +): + """ + Gather the input tensor from all ranks into a single output tensor on ``dst``. + + This is the single-output-tensor analog of :func:`gather`: instead of + filling a Python list of per-rank tensors on the destination rank, each + rank's contribution is written directly into a single, correctly-sized + ``gather_tensor``. Backends that support a native gather-into-tensor path + (e.g. NCCL >= 2.28.3 via ``ncclGather``) avoid the extra per-rank copy that + :func:`gather` incurs. + + This function requires ``tensor`` to be the same size on each process. + + Args: + tensor (Tensor): Input tensor to be gathered from the current rank. + gather_tensor (Tensor, optional): Output tensor to accommodate the + gathered contributions from all ranks. It must be contiguous and + sized to hold ``world_size`` copies of ``tensor``; only its total + number of elements is validated, so either a flat concatenation + (``world_size * tensor.numel()``) or a stack + (``[world_size, *tensor.shape]``) works, as both share the same + contiguous layout. A non-contiguous output (e.g. a strided stacked + view) raises a contiguity error rather than a shape error. Only + required (and only used) on the destination rank. + dst (int, optional): Destination rank on global process group (regardless + of ``group`` argument). (If both ``dst`` and ``group_dst`` are None, + default is global rank 0) + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + group_dst (int, optional): Destination rank on ``group``. Invalid to + specify both ``dst`` and ``group_dst`` + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + Example:: + >>> # xdoctest: +SKIP("no rank") + >>> # We have 2 process groups, 2 ranks. + >>> device = torch.device(f"cuda:{rank}") + >>> tensor = torch.arange(2, dtype=torch.int64, device=device) + 1 + 2 * rank + >>> if dist.get_rank() == 0: + >>> gather_tensor = torch.zeros(2 * 2, dtype=torch.int64, device=device) + >>> else: + >>> gather_tensor = None + >>> dist.gather_single(tensor, gather_tensor, dst=0) + >>> gather_tensor + tensor([1, 2, 3, 4], device='cuda:0') # Rank 0 + None # Rank 1 + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + gather_single, + relevant_args, + tensor, + gather_tensor=gather_tensor, + dst=dst, + group=group, + async_op=async_op, + group_dst=group_dst, + ) + + _check_single_tensor(tensor, "tensor") + group = _group_or_default_group(group) + if _rank_not_in_group(group): + _warn_not_in_group("gather_single") + return + if dst is None and group_dst is None: + dst = 0 + group_dst = _canonicalize_group_rank(group, dst, group_dst, return_global=False) + my_group_rank = group.rank() + + if group_dst == my_group_rank: + if gather_tensor is None: + raise ValueError("gather_tensor must be specified on the destination rank") + _check_single_tensor(gather_tensor, "gather_tensor") + _ensure_all_tensors_same_dtype(tensor, gather_tensor) + output_tensor = gather_tensor + else: + # gather_tensor is unused on non-destination ranks; pass an empty + # placeholder so the single-tensor op signature is satisfied. + output_tensor = tensor.new_empty(0) + + opts = GatherOptions() + opts.rootRank = group_dst + opts.asyncOp = async_op + work = group.gather_single(output_tensor, tensor, opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + + +@_exception_logger +@deprecated( + "`torch.distributed.gather_into_tensor` is deprecated. " + "Please use `torch.distributed.gather_single` instead.", + category=FutureWarning, +) +def gather_into_tensor( + tensor: torch.Tensor, + gather_tensor: torch.Tensor | None = None, + dst: int | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, + group_dst: int | None = None, +): + """ + Gather the input tensor from all ranks into a single output tensor on ``dst``. + + .. warning:: + `gather_into_tensor` is deprecated. Users should use `gather_single` + instead. + + """ + return gather_single(tensor, gather_tensor, dst, group, async_op, group_dst) + + +@_exception_logger +def scatter( + tensor: torch.Tensor, + scatter_list: list[torch.Tensor] | None = None, + src: int | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, + group_src: int | None = None, +) -> Work | None: + """ + Scatters a list of tensors to all processes in a group. + + Each process will receive exactly one tensor and store its data in the + ``tensor`` argument. + + Complex tensors are supported. + + Args: + tensor (Tensor): Output tensor. + scatter_list (list[Tensor]): List of tensors to scatter (default is + None, must be specified on the source rank) + src (int): Source rank on global process group (regardless of ``group`` argument). + (If both ``src`` and ``group_src`` are None, default is global rank 0) + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + group_src (int, optional): Source rank on ``group``. Invalid to specify both ``src`` and ``group_src`` + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + .. note:: Note that all Tensors in scatter_list must have the same size. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> tensor_size = 2 + >>> device = torch.device(f'cuda:{rank}') + >>> output_tensor = torch.zeros(tensor_size, device=device) + >>> if dist.get_rank() == 0: + >>> # Assumes world_size of 2. + >>> # Only tensors, all of which must be the same size. + >>> t_ones = torch.ones(tensor_size, device=device) + >>> t_fives = torch.ones(tensor_size, device=device) * 5 + >>> scatter_list = [t_ones, t_fives] + >>> else: + >>> scatter_list = None + >>> dist.scatter(output_tensor, scatter_list, src=0) + >>> # Rank i gets scatter_list[i]. + >>> output_tensor + tensor([1., 1.], device='cuda:0') # Rank 0 + tensor([5., 5.], device='cuda:1') # Rank 1 + + """ + relevant_args = (tensor,) + if has_torch_function(relevant_args): + return handle_torch_function( + scatter, + relevant_args, + tensor, + scatter_list=scatter_list, + src=src, + group=group, + async_op=async_op, + group_src=group_src, + ) + + _check_single_tensor(tensor, "tensor") + # Parameter ``scatter_list`` may be left unspecified on non-src ranks. + if scatter_list: + _check_tensor_list(scatter_list, "scatter_list") + else: + scatter_list = [] + _ensure_all_tensors_same_dtype(tensor, scatter_list) + group = _group_or_default_group(group) + if src is None and group_src is None: + src = 0 + group_src = _canonicalize_group_rank(group, src, group_src, return_global=False) + if _rank_not_in_group(group): + _warn_not_in_group("scatter") + return None + scatter_list = [ + t if not t.is_complex() else torch.view_as_real(t) for t in scatter_list + ] + tensor = tensor if not tensor.is_complex() else torch.view_as_real(tensor) + + my_group_rank = group.rank() + if group_src == my_group_rank: + if not scatter_list: + raise ValueError( + "Argument ``scatter_list`` must be specified on source rank." + ) + input_tensors = [scatter_list] + output_tensors = [tensor] + else: + if scatter_list: + raise ValueError( + "Argument ``scatter_list`` must NOT be specified on non-source ranks." + ) + input_tensors = [] + output_tensors = [tensor] + + opts = ScatterOptions() + opts.rootRank = group_src + opts.asyncOp = async_op + work = group.scatter(output_tensors, input_tensors, opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@overload +def reduce_scatter( + output: torch.Tensor, + input_list: list[torch.Tensor], + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + *, + async_op: Literal[True], +) -> Work: ... + + +@overload +def reduce_scatter( + output: torch.Tensor, + input_list: list[torch.Tensor], + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: ... + + +@_exception_logger +def reduce_scatter( + output: torch.Tensor, + input_list: list[torch.Tensor], + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Reduces, then scatters a list of tensors to all processes in a group. + + Args: + output (Tensor): Output tensor. + input_list (list[Tensor]): List of tensors to reduce and scatter. + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + """ + relevant_args = (output,) + if has_torch_function(relevant_args): + return handle_torch_function( + reduce_scatter, + relevant_args, + output, + input_list, + op=op, + group=group, + async_op=async_op, # pyrefly: ignore[bad-argument-type] + ) + + _check_single_tensor(output, "output") + _check_tensor_list(input_list, "input_list") + _ensure_all_tensors_same_dtype(output, input_list) + if _rank_not_in_group(group): + _warn_not_in_group("reduce_scatter") + return None + + opts = ReduceScatterOptions() + opts.reduceOp = op + opts.asyncOp = async_op + + group = group or _get_default_group() + work = group.reduce_scatter([output], [input_list], opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@overload +def reduce_scatter_single( + output: torch.Tensor, + input: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + *, + async_op: Literal[True], +) -> Work: ... + + +@overload +def reduce_scatter_single( + output: torch.Tensor, + input: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: ... + + +@_exception_logger +def reduce_scatter_single( + output: torch.Tensor, + input: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Reduces, then scatters a tensor to all ranks in a group. + + Args: + output (Tensor): Output tensor. It should have the same size across all + ranks. + input (Tensor): Input tensor to be reduced and scattered. Its size + should be output tensor size times the world size. The input tensor + can have one of the following shapes: + (i) a concatenation of the output tensors along the primary + dimension, or + (ii) a stack of the output tensors along the primary dimension. + For definition of "concatenation", see ``torch.cat()``. + For definition of "stack", see ``torch.stack()``. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + Examples: + >>> # xdoctest: +SKIP("need process group init") + >>> # All tensors below are of torch.int64 dtype and on CUDA devices. + >>> # We have two ranks. + >>> device = torch.device(f"cuda:{rank}") + >>> tensor_out = torch.zeros(2, dtype=torch.int64, device=device) + >>> # Input in concatenation form + >>> tensor_in = torch.arange(world_size * 2, dtype=torch.int64, device=device) + >>> tensor_in + tensor([0, 1, 2, 3], device='cuda:0') # Rank 0 + tensor([0, 1, 2, 3], device='cuda:1') # Rank 1 + >>> dist.reduce_scatter_single(tensor_out, tensor_in) + >>> tensor_out + tensor([0, 2], device='cuda:0') # Rank 0 + tensor([4, 6], device='cuda:1') # Rank 1 + >>> # Input in stack form + >>> tensor_in = torch.reshape(tensor_in, (world_size, 2)) + >>> tensor_in + tensor([[0, 1], + [2, 3]], device='cuda:0') # Rank 0 + tensor([[0, 1], + [2, 3]], device='cuda:1') # Rank 1 + >>> dist.reduce_scatter_single(tensor_out, tensor_in) + >>> tensor_out + tensor([0, 2], device='cuda:0') # Rank 0 + tensor([4, 6], device='cuda:1') # Rank 1 + + """ + # Dynamo has built-in logic to map legacy distributed ops to functional collectives. + # Let's redirect to a torch function mode that can mimic this logic outside Dynamo + # (e.g., non-strict export implements such a torch function mode). + relevant_args = (input,) + if has_torch_function(relevant_args): + return handle_torch_function( + reduce_scatter_single, + relevant_args, + output, + input, + op=op, + group=group, + async_op=async_op, # pyrefly: ignore[bad-argument-type] + ) + + _check_single_tensor(output, "output") + _check_single_tensor(input, "input") + + if _rank_not_in_group(group): + _warn_not_in_group("reduce_scatter_single") + return None + + opts = ReduceScatterOptions() + opts.reduceOp = op + opts.asyncOp = async_op + + group = group or _get_default_group() + + # Check if we are in coalescing context + # If we are, do not issue single operation, just append a collective representation + if group in _world.pg_coalesce_state: + coll = _CollOp(reduce_scatter_single, input, output, op, None) + _world.pg_coalesce_state[group].append(coll) + if async_op: + return _IllegalWork() + else: + return None + + work = group.reduce_scatter_single(output, input, opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +@deprecated( + "`torch.distributed.reduce_scatter_tensor` is deprecated. " + "Please use `torch.distributed.reduce_scatter_single` instead.", + category=FutureWarning, +) +def reduce_scatter_tensor( + output: torch.Tensor, + input: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Reduces, then scatters a tensor to all ranks in a group. + + .. warning:: + `reduce_scatter_tensor` is deprecated. Users should use + `reduce_scatter_single` instead. + + """ + return reduce_scatter_single(output, input, op, group, async_op) + + +@deprecated( + "`torch.distributed._reduce_scatter_base` is a private function and will be deprecated. " + "Please use `torch.distributed.reduce_scatter_single` instead.", + category=FutureWarning, +) +def _reduce_scatter_base( + output: torch.Tensor, + input: torch.Tensor, + op: _ReduceOp = ReduceOp.SUM, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Reduces, then scatters a flattened tensor to all processes in a group. + + Args: + output (Tensor): Output tensor. + input (Tensor): Input tensor that is of size output tensor size times world size + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + .. warning:: + `_reduce_scatter_base` is a private function. Users should use + `reduce_scatter_single` instead. + + """ + return reduce_scatter_single(output, input, op, group, async_op) + + +@_exception_logger +def all_to_all_single( + output: torch.Tensor, + input: torch.Tensor, + output_split_sizes: list[int] | None = None, + input_split_sizes: list[int] | None = None, + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Split input tensor and then scatter the split list to all processes in a group. + + Later the received tensors are concatenated from all the processes in the group + and returned as a single output tensor. + + Complex tensors are supported. + + Args: + output (Tensor): Gathered concatenated output tensor. + input (Tensor): Input tensor to scatter. + output_split_sizes: (list[Int], optional): Output split sizes for dim 0 + if specified None or empty, dim 0 of ``output`` tensor must divide + equally by ``world_size``. + input_split_sizes: (list[Int], optional): Input split sizes for dim 0 + if specified None or empty, dim 0 of ``input`` tensor must divide + equally by ``world_size``. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + .. warning:: + `all_to_all_single` is experimental and subject to change. + + Examples: + >>> # xdoctest: +SKIP("Undefined rank") + >>> input = torch.arange(4) + rank * 4 + >>> input + tensor([0, 1, 2, 3]) # Rank 0 + tensor([4, 5, 6, 7]) # Rank 1 + tensor([8, 9, 10, 11]) # Rank 2 + tensor([12, 13, 14, 15]) # Rank 3 + >>> output = torch.empty([4], dtype=torch.int64) + >>> dist.all_to_all_single(output, input) + >>> output + tensor([0, 4, 8, 12]) # Rank 0 + tensor([1, 5, 9, 13]) # Rank 1 + tensor([2, 6, 10, 14]) # Rank 2 + tensor([3, 7, 11, 15]) # Rank 3 + + >>> # Essentially, it is similar to following operation: + >>> scatter_list = list(input.chunk(world_size)) + >>> gather_list = list(output.chunk(world_size)) + >>> for i in range(world_size): + >>> dist.scatter(gather_list[i], scatter_list if i == rank else [], src = i) + + >>> # Another example with uneven split + >>> input + tensor([0, 1, 2, 3, 4, 5]) # Rank 0 + tensor([10, 11, 12, 13, 14, 15, 16, 17, 18]) # Rank 1 + tensor([20, 21, 22, 23, 24]) # Rank 2 + tensor([30, 31, 32, 33, 34, 35, 36]) # Rank 3 + >>> input_splits + [2, 2, 1, 1] # Rank 0 + [3, 2, 2, 2] # Rank 1 + [2, 1, 1, 1] # Rank 2 + [2, 2, 2, 1] # Rank 3 + >>> output_splits + [2, 3, 2, 2] # Rank 0 + [2, 2, 1, 2] # Rank 1 + [1, 2, 1, 2] # Rank 2 + [1, 2, 1, 1] # Rank 3 + >>> output = ... + >>> dist.all_to_all_single(output, input, output_splits, input_splits) + >>> output + tensor([ 0, 1, 10, 11, 12, 20, 21, 30, 31]) # Rank 0 + tensor([ 2, 3, 13, 14, 22, 32, 33]) # Rank 1 + tensor([ 4, 15, 16, 23, 34, 35]) # Rank 2 + tensor([ 5, 17, 18, 24, 36]) # Rank 3 + + + >>> # Another example with tensors of torch.cfloat type. + >>> input = torch.tensor( + ... [1 + 1j, 2 + 2j, 3 + 3j, 4 + 4j], dtype=torch.cfloat + ... ) + 4 * rank * (1 + 1j) + >>> input + tensor([1+1j, 2+2j, 3+3j, 4+4j]) # Rank 0 + tensor([5+5j, 6+6j, 7+7j, 8+8j]) # Rank 1 + tensor([9+9j, 10+10j, 11+11j, 12+12j]) # Rank 2 + tensor([13+13j, 14+14j, 15+15j, 16+16j]) # Rank 3 + >>> output = torch.empty([4], dtype=torch.int64) + >>> dist.all_to_all_single(output, input) + >>> output + tensor([1+1j, 5+5j, 9+9j, 13+13j]) # Rank 0 + tensor([2+2j, 6+6j, 10+10j, 14+14j]) # Rank 1 + tensor([3+3j, 7+7j, 11+11j, 15+15j]) # Rank 2 + tensor([4+4j, 8+8j, 12+12j, 16+16j]) # Rank 3 + """ + # Dynamo has built-in logic to map legacy distributed ops to functional collectives. + # Let's redirect to a torch function mode that can mimic this logic outside Dynamo + # (e.g., non-strict export implements such a torch function mode). + relevant_args = (input,) + if has_torch_function(relevant_args): + return handle_torch_function( + all_to_all_single, + relevant_args, + output, + input, + output_split_sizes=output_split_sizes, + input_split_sizes=input_split_sizes, + group=group, + async_op=async_op, + ) + + if _rank_not_in_group(group): + _warn_not_in_group("all_to_all_single") + return None + + opts = AllToAllOptions() + opts.asyncOp = async_op + _check_single_tensor(output, "output") + _check_single_tensor(input, "input") + _ensure_all_tensors_same_dtype(output, input) + + if input.is_complex(): + input = torch.view_as_real(input) + if output.is_complex(): + output = torch.view_as_real(output) + + output_split_sizes = [] if output_split_sizes is None else output_split_sizes + input_split_sizes = [] if input_split_sizes is None else input_split_sizes + + group = group or _get_default_group() + work = group.all_to_all_single( + output, input, output_split_sizes, input_split_sizes, opts + ) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@_exception_logger +def all_to_all( + output_tensor_list: list[torch.Tensor], + input_tensor_list: list[torch.Tensor], + group: ProcessGroup | None = None, + async_op: bool = False, +) -> Work | None: + """ + Scatters list of input tensors to all processes in a group and return gathered list of tensors in output list. + + Complex tensors are supported. + + Args: + output_tensor_list (list[Tensor]): List of tensors to be gathered one + per rank. + input_tensor_list (list[Tensor]): List of tensors to scatter one per rank. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group. + + .. warning:: + `all_to_all` is experimental and subject to change. + + Examples: + >>> # xdoctest: +SKIP("Undefined rank") + >>> input = torch.arange(4) + rank * 4 + >>> input = list(input.chunk(4)) + >>> input + [tensor([0]), tensor([1]), tensor([2]), tensor([3])] # Rank 0 + [tensor([4]), tensor([5]), tensor([6]), tensor([7])] # Rank 1 + [tensor([8]), tensor([9]), tensor([10]), tensor([11])] # Rank 2 + [tensor([12]), tensor([13]), tensor([14]), tensor([15])] # Rank 3 + >>> output = list(torch.empty([4], dtype=torch.int64).chunk(4)) + >>> dist.all_to_all(output, input) + >>> output + [tensor([0]), tensor([4]), tensor([8]), tensor([12])] # Rank 0 + [tensor([1]), tensor([5]), tensor([9]), tensor([13])] # Rank 1 + [tensor([2]), tensor([6]), tensor([10]), tensor([14])] # Rank 2 + [tensor([3]), tensor([7]), tensor([11]), tensor([15])] # Rank 3 + + >>> # Essentially, it is similar to following operation: + >>> scatter_list = input + >>> gather_list = output + >>> for i in range(world_size): + >>> dist.scatter(gather_list[i], scatter_list if i == rank else [], src=i) + + >>> input + tensor([0, 1, 2, 3, 4, 5]) # Rank 0 + tensor([10, 11, 12, 13, 14, 15, 16, 17, 18]) # Rank 1 + tensor([20, 21, 22, 23, 24]) # Rank 2 + tensor([30, 31, 32, 33, 34, 35, 36]) # Rank 3 + >>> input_splits + [2, 2, 1, 1] # Rank 0 + [3, 2, 2, 2] # Rank 1 + [2, 1, 1, 1] # Rank 2 + [2, 2, 2, 1] # Rank 3 + >>> output_splits + [2, 3, 2, 2] # Rank 0 + [2, 2, 1, 2] # Rank 1 + [1, 2, 1, 2] # Rank 2 + [1, 2, 1, 1] # Rank 3 + >>> input = list(input.split(input_splits)) + >>> input + [tensor([0, 1]), tensor([2, 3]), tensor([4]), tensor([5])] # Rank 0 + [tensor([10, 11, 12]), tensor([13, 14]), tensor([15, 16]), tensor([17, 18])] # Rank 1 + [tensor([20, 21]), tensor([22]), tensor([23]), tensor([24])] # Rank 2 + [tensor([30, 31]), tensor([32, 33]), tensor([34, 35]), tensor([36])] # Rank 3 + >>> output = ... + >>> dist.all_to_all(output, input) + >>> output + [tensor([0, 1]), tensor([10, 11, 12]), tensor([20, 21]), tensor([30, 31])] # Rank 0 + [tensor([2, 3]), tensor([13, 14]), tensor([22]), tensor([32, 33])] # Rank 1 + [tensor([4]), tensor([15, 16]), tensor([23]), tensor([34, 35])] # Rank 2 + [tensor([5]), tensor([17, 18]), tensor([24]), tensor([36])] # Rank 3 + + >>> # Another example with tensors of torch.cfloat type. + >>> input = torch.tensor( + ... [1 + 1j, 2 + 2j, 3 + 3j, 4 + 4j], dtype=torch.cfloat + ... ) + 4 * rank * (1 + 1j) + >>> input = list(input.chunk(4)) + >>> input + [tensor([1+1j]), tensor([2+2j]), tensor([3+3j]), tensor([4+4j])] # Rank 0 + [tensor([5+5j]), tensor([6+6j]), tensor([7+7j]), tensor([8+8j])] # Rank 1 + [tensor([9+9j]), tensor([10+10j]), tensor([11+11j]), tensor([12+12j])] # Rank 2 + [tensor([13+13j]), tensor([14+14j]), tensor([15+15j]), tensor([16+16j])] # Rank 3 + >>> output = list(torch.empty([4], dtype=torch.int64).chunk(4)) + >>> dist.all_to_all(output, input) + >>> output + [tensor([1+1j]), tensor([5+5j]), tensor([9+9j]), tensor([13+13j])] # Rank 0 + [tensor([2+2j]), tensor([6+6j]), tensor([10+10j]), tensor([14+14j])] # Rank 1 + [tensor([3+3j]), tensor([7+7j]), tensor([11+11j]), tensor([15+15j])] # Rank 2 + [tensor([4+4j]), tensor([8+8j]), tensor([12+12j]), tensor([16+16j])] # Rank 3 + + """ + relevant_args = ( + tuple(input_tensor_list) + if isinstance(input_tensor_list, (list, tuple)) + else (input_tensor_list,) + ) + if has_torch_function(relevant_args): + return handle_torch_function( + all_to_all, + relevant_args, + output_tensor_list, + input_tensor_list, + group=group, + async_op=async_op, + ) + + if _rank_not_in_group(group): + _warn_not_in_group("all_to_all") + return None + + opts = AllToAllOptions() + opts.asyncOp = async_op + _check_tensor_list(output_tensor_list, "output_tensor_list") + _check_tensor_list(input_tensor_list, "input_tensor_list") + _ensure_all_tensors_same_dtype(output_tensor_list, input_tensor_list) + + input_tensor_list = [ + t if not t.is_complex() else torch.view_as_real(t) for t in input_tensor_list + ] + output_tensor_list = [ + t if not t.is_complex() else torch.view_as_real(t) for t in output_tensor_list + ] + + group = group or _get_default_group() + work = group.alltoall(output_tensor_list, input_tensor_list, opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +@overload +def barrier( + group: ProcessGroup | None = GroupMember.WORLD, + *, + async_op: Literal[True], + device_ids: list[int] | None = None, + timeout: timedelta | None = None, +) -> Work: ... + + +@overload +def barrier( + group: ProcessGroup | None = GroupMember.WORLD, + async_op: bool = False, + device_ids: list[int] | None = None, + timeout: timedelta | None = None, +) -> Work | None: ... + + +@_exception_logger +def barrier( + group: ProcessGroup | None = GroupMember.WORLD, + async_op: bool = False, + device_ids: list[int] | None = None, + timeout: timedelta | None = None, +) -> Work | None: + """ + Synchronize all processes. + + This collective blocks processes until the whole group enters this function, + if async_op is False, or if async work handle is called on wait(). + + Args: + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + async_op (bool, optional): Whether this op should be an async op + device_ids ([int], optional): List of device/GPU ids. Only one id is expected. + timeout (datetime.timedelta, optional): Timeout for barrier. + If ``None``, the default process group timeout will be used. + + Returns: + Async work handle, if async_op is set to True. + None, if not async_op or if not part of the group + + .. note:: `ProcessGroupNCCL` now blocks the cpu thread till the completion of the barrier collective. + .. note:: `ProcessGroupNCCL` implements barrier as an all_reduce of a 1-element tensor. A device must be chosen + for allocating this tensor. The device choice is made by checking in this order (1) the first device passed to + `device_ids` arg of barrier if not None, (2) the device passed to init_process_group if not None, (3) the device + that was first used with this process group, if another collective with tensor inputs has been performed, (4) + the device index indicated by the global rank mod local device count. + """ + group = group or _get_default_group() + + if _rank_not_in_group(group): + _warn_not_in_group("barrier") + return None + + opts = BarrierOptions() + opts.asyncOp = async_op + if timeout is not None: + opts.timeout = timeout + # Detect the accelerator on the machine. If no accelerator is available, it + # returns CPU. + device = torch._C._get_accelerator() + if isinstance(device_ids, list): + opts.device_ids = device_ids + # use only the first device id + opts.device = torch.device(device.type, device_ids[0]) + elif getattr(group, "bound_device_id", None) is not None: + # Use device id from `init_process_group(device_id=...)` + opts.device = group.bound_device_id # type: ignore[assignment] + elif device.type == "cpu" or _get_object_coll_device(group) == "cpu": + opts.device = torch.device("cpu") + else: + # Use the current device set by the user. If user did not set any, this + # may use default device 0, causing issues like hang or all processes + # creating context on device 0. + opts.device = device + if group.rank() == 0: + warnings.warn( # warn only once + "barrier(): using the device under current context. " + "You can specify `device_id` in `init_process_group` to mute this warning.", + stacklevel=2, + ) + + work = group.barrier(opts=opts) + + if async_op: + return work + elif ( + work is not None + ): # Backward compatible with backends that don't sync at CPP level + work.wait() + # Otherwise, the backend has sync'ed at CPP level + return None + + +def monitored_barrier( + group: ProcessGroup | None = GroupMember.WORLD, + timeout: timedelta | float | None = None, + wait_all_ranks: bool = False, +) -> None: + """ + Synchronize processes similar to ``torch.distributed.barrier``, but consider a configurable timeout. + + It is able to report ranks that did not pass this barrier within the provided timeout. + Specifically, for non-zero ranks, will block until a send/recv is processed from rank 0. + Rank 0 will block until all send /recv from other ranks are processed, and will report + failures for ranks that failed to respond in time. Note that if one rank does not reach the + monitored_barrier (for example due to a hang), all other ranks would fail in monitored_barrier. + + This collective will block all processes/ranks in the group, until the + whole group exits the function successfully, making it useful for debugging + and synchronizing. However, it can have a performance impact and should only + be used for debugging or scenarios that require full synchronization points + on the host-side. For debugging purposes, this barrier can be inserted + before the application's collective calls to check if any ranks are + desynchronized. + + .. note:: Note that this collective is only supported with the GLOO backend. + + Args: + group (ProcessGroup, optional): The process group to work on. If + ``None``, the default process group will be used. + timeout (datetime.timedelta, optional): Timeout for monitored_barrier. + If ``None``, the default process group timeout will be used. + wait_all_ranks (bool, optional): Whether to collect all failed ranks or + not. By default, this is ``False`` and ``monitored_barrier`` on rank 0 + will throw on the first failed rank it encounters in order to fail + fast. By setting ``wait_all_ranks=True`` ``monitored_barrier`` will + collect all failed ranks and throw an error containing information + about all failed ranks. + + Returns: + ``None``. + + Example:: + >>> # xdoctest: +SKIP("need process group init") + >>> # Note: Process group initialization omitted on each rank. + >>> import torch.distributed as dist + >>> if dist.get_rank() != 1: + >>> dist.monitored_barrier() # Raises exception indicating that + >>> # rank 1 did not call into monitored_barrier. + >>> # Example with wait_all_ranks=True + >>> if dist.get_rank() == 0: + >>> dist.monitored_barrier(wait_all_ranks=True) # Raises exception + >>> # indicating that ranks 1, 2, ... world_size - 1 did not call into + >>> # monitored_barrier. + """ + # Need to call rank not in group before using the group, otherwise + # "Invalid process group" error is raised. + if _rank_not_in_group(group): + _warn_not_in_group("monitored_barrier") + return + + # monitored_barrier runs on a CPU backend (GLOO, or its torchcomms + # BackendWrapper equivalent). Require the group to have a CPU backend rather + # than matching the "gloo" name, so a device-qualified group (e.g. + # "cpu:gloo,cuda:nccl") is accepted for its CPU backend. + group_to_use = group or _get_default_group() + if torch.device("cpu") not in group_to_use._device_types: + raise ValueError( + "monitored_barrier requires a group with a CPU-capable backend (e.g. GLOO)." + ) + + if timeout is None: + timeout = _get_default_timeout(get_backend(group)) + elif isinstance(timeout, float): + # TODO(whc) apparently some existing test case for monitored_barrier passes in a timeout in float format? + warnings.warn( + "Please specify timeout arg as a timedelta. " + f"Converting current value of {timeout} assuming it represents seconds", + stacklevel=2, + ) + timeout = timedelta(seconds=timeout) + + _check_valid_timeout(timeout) + + return group_to_use.monitored_barrier( # type:ignore[attr-defined] + timeout, wait_all_ranks=wait_all_ranks + ) + + +def _create_process_group_wrapper( + wrapped_pg: C10DBackend, + store_prefix: str, + store: Store, + rank: int, + world_size: int, + timeout: timedelta = default_pg_timeout, +) -> C10DBackend: + if not is_gloo_available(): + raise AssertionError("ProcessGroupWrapper unsupported without GLOO backend.") + + # (whc) this appears to be just for the gloo backend? if so, `default_pg_timeout` is appropriate... + + # Create a separate prefix store for the helper process group. + prefix = f"{PG_WRAPPER_STORE_PREFIX}:{store_prefix}" + store = PrefixStore(prefix, store) + helper_pg = ProcessGroupGloo(store, rank, world_size, timeout=timeout) + # Wrap the underlying pg with ProcessGroupWrapper. + wrapped_pg = _ProcessGroupWrapper(wrapped_pg, helper_pg) + return wrapped_pg + + +# helper function for hashing a list of ranks to a unique string +def _hash_ranks_to_str(ranks: Sequence[int]) -> str: + rank_join: str = "_".join(map(str, ranks)) + # Disambiguate multiple PGs with the same rank composition. The salt MUST be + # identical across all ranks that create this group, otherwise ranks compute + # different group names for the same group. Backends that use the group name + # as a rendezvous store prefix (e.g. Gloo split's connectFullMesh) then key + # off different prefixes and deadlock. len(_world.pg_names) is NOT safe here: + # it diverges across ranks after an earlier asymmetric-membership new_group() + # (non-member ranks register fewer PGs). _world.group_count is incremented by + # _process_group_name() on every rank that reaches it (before the member + # check), so it stays consistent and monotonic across ranks. + unique_str = "_".join([rank_join, str(_world.group_count)]) + return hashlib.sha1(bytes(unique_str, "utf-8"), usedforsecurity=False).hexdigest() + + +# Takes a list of ranks and computes an integer color +def _process_group_color(ranks: list[int]) -> int: + # Convert list to tuple to make it hashable + # pyrefly: ignore [bad-assignment] + ranks = tuple(ranks) + hash_value = hash(ranks) + # Split color must be: + # - a non-negative integer; + # - a type compatible with C's int because we are pybinding to the latter. + # Thus, we limit the hash value within c_int's max value. + max_c_int = 2 ** (ctypes.sizeof(ctypes.c_int) * 8 - 1) + color = abs(hash_value) % max_c_int + return color + + +def _process_group_name(ranks: Sequence[int], use_hashed_name: bool) -> GroupName: + # Create name for a process group. + global _world + if use_hashed_name: + pg_name = GroupName(_hash_ranks_to_str(ranks)) + else: + pg_name = GroupName(str(_world.group_count)) + # Increment on BOTH paths so group_count advances once per group-creation + # call on every rank that reaches here. This keeps it a collective-consistent, + # monotonic counter usable as the uniqueness salt in _hash_ranks_to_str + # (see the comment there). Names need only be unique, not contiguous, so the + # hashed path consuming counter values is harmless for the non-hashed names. + _world.group_count += 1 + return pg_name + + +def _get_backend_from_str(backend: str | None = None) -> str: + # Default to the same backend as the global process group + # if backend is not specified. + if not backend: + backend = get_backend(_get_default_group()) + return Backend(backend) + + +def _is_safe_to_split() -> bool: + """ + Checks if it is safe to split the any process group in the world. + This is only safe if the default pg has a bound device id, otherwise + users must be aware that a pg is only splittable after the first collective is + issued. + """ + return _get_default_group().bound_device_id is not None + + +@_time_logger +def split_group( + parent_pg: ProcessGroup | None = None, + split_ranks: list[list[int]] | None = None, + timeout: timedelta | None = None, + pg_options: C10DBackend.Options | None = None, + group_desc: str | None = None, + backend: str | Backend | None = None, +) -> ProcessGroup | _NonGroupMember: + """ + Create a new process group split from the given parent process group. + + warning:: This is an experimental API. Only the ``NCCL`` and custom plugin backends + are supported. Other backends will raise an error. + Users of this API must guarantee that all ranks in the parent group enter this API call, + and the split of the sub groups is the same across all ranks in the parent group. + + Args: + parent_pg (ProcessGroup, optional): The parent process group. If None, + the default process group will be used. Users need to guarantee that + the parent group is fully initialized (e.g, communicators are initialized) + split_ranks (list[list[int]]): the split ranks, which is a list of list of ranks. + Users need to make sure the validity of the split ranks such that one + split (represented by one inner list of ints) does not overlap with any other split. + Note that the ranks in each split is the group rank (instead of global rank) + in the parent pg. For example, if the parent group has 4 ranks, and split_ranks can be + [[0, 1], [2, 3]]. Note [[0,1]] is also a valid split, in which case ranks 2, 3 would + return a non-group member. + The order of ranks within each split is preserved: position in the + list determines the group rank in the new group. All ranks must pass + the same ordering. + timeout (timedelta, optional): see `init_process_group` for details and default value. + pg_options (ProcessGroupOptions, optional): Additional options need to be passed in during + the construction of specific process groups. i.e.``is_high_priority_stream`` + can be specified so that process group can pick up high priority cuda streams. + group_desc (str, optional): a string to describe the process group. + backend (str or Backend, optional): selects a subset of the parent process + group's per-device backends to retain in the child group. The string + follows the same format as ``init_process_group`` (e.g. ``"nccl"`` or + ``"cpu:gloo,cuda:nccl"``). Each device type in the requested backend + must be present in the parent group, and the backend name for that + device type must exactly match the parent's. The child's default + backend's device type must be in the requested set. If ``None`` + (default), the child inherits the full backend configuration of the + parent. + + Returns: + ProcessGroup if the current rank is within one split/subgroup given by split_ranks, + or GroupMember.NON_GROUP_MEMBER if the current rank is not part of any + split in ``split_ranks``. + + """ + # check inputs + if split_ranks is None or len(split_ranks) == 0: + raise ValueError("split_ranks cannot be None or empty") + + global _world + default_pg = _get_default_group() + device_id = default_pg.bound_device_id + if not device_id and not _use_torchcomms_enabled(): + raise RuntimeError( + "No device associated with the default pg, not safe to split any process groups" + ) + global_rank = default_pg.rank() + global_world_size = default_pg.size() + + if not parent_pg: + parent_pg = default_pg + if parent_pg not in _world.pg_group_ranks: + raise ValueError(f"Group {parent_pg} is not registered") + + parent_global_to_group_ranks = _world.pg_group_ranks[parent_pg] + parent_group_to_global_ranks = { + group_rank: global_rank + for global_rank, group_rank in parent_global_to_group_ranks.items() + } + + if global_rank not in parent_global_to_group_ranks: + raise ValueError( + f"Global rank {global_rank} is not part of the parent group {parent_pg}" + ) + + parent_group_rank = parent_global_to_group_ranks[global_rank] + + if torch.accelerator.is_available(): + parent_backend = parent_pg._get_backend( + torch.accelerator.current_accelerator() # pyrefly: ignore[bad-argument-type] + ) + elif _use_torchcomms_enabled(): + # torchcomms supports CPU/gloo splitting; no accelerator is required. + parent_backend = parent_pg._get_backend( + torch.device("cpu") # pyrefly: ignore[bad-argument-type] + ) + else: + raise RuntimeError( + "No backend for the parent process group or its backend does not support splitting" + ) + + # if the parent backend does not support splitting, raise error + # currently this API only support NCCL and XCCL backend + if ( + not parent_backend or not parent_backend.supports_splitting + ) and not _use_torchcomms_enabled(): + raise RuntimeError( + "No backend for the parent process group or its backend does not support splitting" + ) + + # set the group_desc before the color or no_color split + if hasattr(parent_backend, "comm_split_count") and group_desc is None: + group_desc = f"{parent_pg.group_desc}:split:{parent_backend.comm_split_count()}" # type: ignore[attr-defined] + + parent_backend_str, _ = _world.pg_map[parent_pg] + # same type of backend as the parent process group + pg_backend = Backend(parent_backend_str) + backend_config = BackendConfig(pg_backend) + + # `backend` (extension API) lets the caller select a subset of the parent + # group's per-device backends to retain in the child group. The C++ split + # loop honors this filter so unwanted backends are never split. + device_types_filter: list[torch.device] | None = None + if backend is not None: + # The parent's per-device backends are what BackendConfig expanded its + # backend string to when the group was built (see + # _new_process_group_helper), restricted to what actually got + # registered. Re-deriving them from Backend.default_device_backend_map + # answers a different question -- "which devices default to this + # backend" -- and gets a bare parent string wrong: a "gloo" parent runs + # gloo on cuda too, and "fake" is no device's default at all. + parent_devices = {d.type for d in parent_pg._device_types} + parent_device_backends = { + device_type: str(be) + for device_type, be in backend_config.get_device_backend_map().items() + if device_type in parent_devices + } + requested_backend_str = str(backend).lower() + if ":" in requested_backend_str: + requested_device_backends = _parse_backend_string( + requested_backend_str, available_devices=parent_devices + ) + else: + # A bare backend name selects every parent device running it. + Backend._ensure_backend_registered(requested_backend_str) + requested_device_backends = { + device_type: be + for device_type, be in parent_device_backends.items() + if be == requested_backend_str + } + if not requested_device_backends: + raise ValueError( + f"Requested backend '{requested_backend_str}' is not present " + f"in the parent process group (parent backends: " + f"{parent_device_backends})" + ) + for device_type, requested_be in requested_device_backends.items(): + if device_type not in parent_device_backends: + raise ValueError( + f"Requested backend for device '{device_type}' is not " + f"present in the parent process group (parent backends: " + f"{parent_device_backends})" + ) + parent_be = parent_device_backends[device_type] + if parent_be != requested_be: + raise ValueError( + f"Backend mismatch for device '{device_type}': parent has " + f"'{parent_be}', requested '{requested_be}'. Backend names " + f"must match exactly when filtering split_group backends." + ) + device_types_filter = [ + torch.device(device_type) for device_type in requested_device_backends + ] + pg_backend = Backend(str(backend)) + backend_config = BackendConfig(pg_backend) + + # pg_options is left as-is when the caller did not pass any: ProcessGroup:: + # splitGroup gives each device's backend a copy of *that backend's* options, + # which is what the child needs. Substituting the default (accelerator) + # backend's options here would hand e.g. ProcessGroupNCCL::Options to the + # gloo leg of a "cpu:gloo,cuda:nccl" group, which rejects them and silently + # falls back to defaults, dropping the caller's timeout and group_name. + + # this timeout defaulting/validation is used for all the new_groups/new_subgroups variants, + # which may just pass their timeout value (or None) + if timeout is None: + timeout = _get_default_timeout(pg_backend) + _check_valid_timeout(timeout) + + # find my group of ranks and my group local rank in split_ranks + # for ranks which are not in any split PGs, we just pass in this the first split group + # and GroupMember.NON_GROUP_MEMBER will be returned. + my_group = split_ranks[0] + + for split_group in split_ranks: + if len(split_group) == 0: + raise ValueError("the split group cannot be empty") + if len(split_group) > global_world_size: + raise ValueError( + "the split group's size should be less or equal to the world_size set by init_process_group" + ) + if len(split_group) != len(set(split_group)): + raise ValueError("the split group cannot have duplicate ranks") + if parent_group_rank in split_group: + my_group = split_group + break + + # use_hashed_name is True to ensure that subgroups have unique names. + # This is needed as some backends (e.g. Gloo) use the group name as a + # PrefixStore prefix for initialization of splits. Thus, names have to be + # unique to avoid key collisions. + group_name = _process_group_name(my_group, use_hashed_name=True) + split_pg = parent_pg.split_group( + my_group, + timeout=timeout, + opts=pg_options, + group_name=group_name, + group_desc=group_desc, + device_types=device_types_filter, + ) + if split_pg is None: + return GroupMember.NON_GROUP_MEMBER + + global_ranks_in_my_group = [parent_group_to_global_ranks[rank] for rank in my_group] + split_pg.bound_device_id = device_id # type: ignore[union-attr] + + # `backend` may have filtered the accelerator's leg out of the child, so + # look the child's backend up on a device the child actually has. + accelerator = torch.accelerator.current_accelerator() + split_devices = {d.type for d in split_pg._device_types} + if accelerator is not None and accelerator.type in split_devices: + split_backend_device = accelerator + elif "cpu" in split_devices: + split_backend_device = torch.device("cpu") + else: + raise RuntimeError( + "No backend for the parent process group or its backend does not support splitting" + ) + split_backend_class = split_pg._get_backend(split_backend_device) + + if split_pg.group_name != group_name: + raise AssertionError( + f"group name should be set to {group_name} but got {split_pg.group_name}" + ) + + # update global state + _register_pg_in_world( + split_pg, + backend_name=pg_backend, + store=split_pg.get_group_store(), + group_name=group_name, + backend_config=str(backend_config), + rank_mapping={ + global_rank: group_rank + for group_rank, global_rank in enumerate(global_ranks_in_my_group) + }, + ) + + if _use_torchcomms_enabled(): + # pyrefly: ignore [missing-attribute] + _world.comms.append(split_backend_class.get_comm()) + return split_pg + + +@_time_logger +def new_group( + ranks: Sequence[int] | None = None, + timeout: timedelta | None = None, + backend: str | Backend | None = None, + pg_options: object | None = None, + use_local_synchronization: bool = False, + group_desc: str | None = None, + device_id: torch.device | None = None, + sort_ranks: bool = True, +) -> ProcessGroup | _NonGroupMember: + """ + Create a new distributed group. + + This function requires that all processes in the main group (i.e. all + processes that are part of the distributed job) enter this function, even + if they are not going to be members of the group. Additionally, groups + should be created in the same order in all processes. + + .. warning:: + Safe concurrent usage: + When using multiple process groups with the ``NCCL`` backend, the user + must ensure a globally consistent execution order of collectives across + ranks. + + If multiple threads within a process issue collectives, explicit + synchronization is necessary to ensure consistent ordering. + + When using async variants of torch.distributed communication APIs, + a work object is returned and the communication kernel is + enqueued on a separate CUDA stream, allowing overlap of communication + and computation. Once one or more async ops have been issued on one process + group, they must be synchronized with other cuda streams by calling `work.wait()` + before using another process group. + + See `Using multiple NCCL communicators concurrently + ` + for more details. + + Args: + ranks (list[int]): List of ranks of group members. If ``None``, will be + set to all ranks. Default is ``None``. + timeout (timedelta, optional): see `init_process_group` for details and default value. + backend (str or Backend, optional): The backend to use. Depending on + build-time configurations, valid values are ``gloo`` and ``nccl``. + By default uses the same backend as the global group. This field + should be given as a lowercase string (e.g., ``"gloo"``), which can + also be accessed via :class:`Backend` attributes (e.g., + ``Backend.GLOO``). If ``None`` is passed in, the backend + corresponding to the default process group will be used. Default is + ``None``. + pg_options (ProcessGroupOptions, optional): process group options + specifying what additional options need to be passed in during + the construction of specific process groups. i.e. for the ``nccl`` + backend, ``is_high_priority_stream`` can be specified so that + process group can pick up high priority cuda streams. For other available options to config nccl, + See https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/api/types.html#ncclconfig-tuse_local_synchronization + (bool, optional): perform a group-local barrier at the end of the process group creation. + This is different in that non-member ranks don't need to call into API and don't + join the barrier. + group_desc (str, optional): a string to describe the process group. + device_id (torch.device, optional): a single, specific device + to "bind" this process to, The `new_group` call will try to initialize + a communication backend immediately for the device if this field is given. + sort_ranks (bool, optional): If ``True`` (the default), sort the + ``ranks`` list before creating the group. If ``False``, preserve + the caller-provided rank order so that position in the ``ranks`` + list determines group rank. All processes must pass the identical + ``ranks`` list. Default is ``True``. + + Returns: + A handle of distributed group that can be given to collective calls or + GroupMember.NON_GROUP_MEMBER if the rank is not part of ``ranks``. + + N.B. use_local_synchronization doesn't work with MPI. + + N.B. While use_local_synchronization=True can be significantly faster with larger + clusters and small process groups, care must be taken since it changes cluster behavior + as non-member ranks don't join the group barrier(). + + N.B. use_local_synchronization=True can lead to deadlocks when each rank creates + multiple overlapping process groups. To avoid that, make sure all ranks follow the + same global creation order. + + N.B. When TorchComms is enabled (``torch.distributed.config.use_torchcomms`` + / ``TORCH_DISTRIBUTED_USE_TORCHCOMMS=1``), subgroups are created directly via + TorchComms' ``new_comm`` (the normal ``_new_group_with_tag`` path), not by + splitting the parent communicator. Pass ``backend="nccl-lazy"`` to build a + per-peer, lazily-initialized group (a dedicated comm + stream per send/recv + peer, like ``ProcessGroupNCCL``) so concurrent P2P to different peers can + overlap; pass the default / ``"nccl"`` backend for an eager group. + """ + return _new_group_with_tag( + ranks, + timeout, + backend, + pg_options, + None, + use_local_synchronization=use_local_synchronization, + group_desc=group_desc, + device_id=device_id, + sort_ranks=sort_ranks, + ) + + +def _new_group_with_tag( + ranks: Sequence[int] | None = None, + timeout: timedelta | None = None, + backend: str | Backend | None = None, + backend_options: object | None = None, + pg_tag: str | None = None, + use_local_synchronization: bool = False, + group_desc: str | None = None, + device_id: torch.device | None = None, + sort_ranks: bool = True, +) -> ProcessGroup | _NonGroupMember: + """ + Variant of ``new_group`` that exposes tag creation. + + :: N.B. The mechanism is experimental and tied to the functional collectives effort, see + ``torch.distributed._functional_collectives`` for reference on how to use it. + """ + global _world + + default_pg = _get_default_group() + if device_id is None: + device_id = default_pg.bound_device_id + elif default_pg.bound_device_id is not None: + if device_id != default_pg.bound_device_id: + raise AssertionError( + "Mismatched bound device between new pg and the default pg." + ) + default_backend, default_store = _world.pg_map[default_pg] + global_rank = default_pg.rank() + global_world_size = default_pg.size() + + # Default to the same backend as the global process group + # if the backend is not specified. + if not backend: + backend = default_backend + backend = Backend(backend) + + # this timeout defaulting/validation is used for all the new_groups/new_subgroups variants, + # which may just pass their timeout value (or None) + if timeout is None: + timeout = _get_default_timeout(backend) + _check_valid_timeout(timeout) + + if ranks is not None: + try: + ranks = [operator.index(rank) for rank in ranks] + except TypeError as error: + raise TypeError("ranks must be a sequence of integers") from error + + if use_local_synchronization: + # MPI backend doesn't have a way for us to perform a partial sync + if backend == Backend.MPI: + raise ValueError( + "MPI backend doesn't support use_local_synchronization=True" + ) + if ranks is not None and get_rank() not in ranks: + return GroupMember.NON_GROUP_MEMBER + + # checks the input ranks + if ranks is not None: + if sort_ranks: + ranks = sorted(ranks) + if len(set(ranks)) != len(ranks): + raise ValueError( + f"ranks list must not contain duplicate entries, got {ranks}" + ) + group_world_size = len(ranks) + if group_world_size > global_world_size: + raise ValueError( + "the new group's world size should be less or " + "equal to the world size set by " + "init_process_group" + ) + # check ranks' sanity + for rank in ranks: + if rank < 0 or rank >= global_world_size: + raise ValueError( + f"Rank {rank} is out of range. Valid ranks are 0 to {global_world_size - 1} " + f"(world_size={global_world_size})" + ) + if global_rank in ranks: + group_rank = ranks.index(global_rank) + else: + group_rank = None + else: + ranks = list(range(global_world_size)) + group_world_size = global_world_size + group_rank = global_rank + + group_name = _process_group_name(ranks, use_hashed_name=use_local_synchronization) + + # If the default PG implements new_group, delegate to it. This allows + # custom Python PG subclasses to handle subgroup creation in a single + # call, avoiding the per-device creator iteration in + # _new_process_group_helper. + if hasattr(default_pg, "new_group") and callable(default_pg.new_group): + pg_or_none = default_pg.new_group( + ranks, + timeout=timeout, + backend=backend, + pg_options=backend_options, + group_name=group_name, + group_desc=group_desc, + ) + if pg_or_none is None: + return GroupMember.NON_GROUP_MEMBER + + pg: ProcessGroup = pg_or_none # pyrefly: ignore[bad-assignment] + _register_pg_in_world( + pg, + backend_name=backend, + store=PrefixStore(f"{group_name}/", default_store), + group_name=group_name, + backend_config=str(BackendConfig(backend)), + rank_mapping={ + global_rank: group_rank for group_rank, global_rank in enumerate(ranks) + }, + ) + return pg + + pg_or_nonmember, pg_store = _new_process_group_helper( + group_world_size, + group_rank, + ranks, + backend, + default_store, + group_name, + backend_options=backend_options, + timeout=timeout, + pg_tag=pg_tag, + device_id=device_id, + group_desc=group_desc, + ) + + # Create the global rank to group rank mapping + _world.pg_group_ranks[pg_or_nonmember] = { + global_rank: group_rank for group_rank, global_rank in enumerate(ranks) + } + + if _is_barrier_after_init() == 1: + # barrier at the end to ensure that once we return from this method, all + # process groups including global variables (if any) are updated + # correctly on all ranks. + # Update 04/2023: for large-scale runs, this barrier (esp. store-based + # barrier) may be costly and/or unscalable. Also, in a lot of cases, + # these barriers may be unnecessary, as proven by a green CI after + # removal. An environment variable `TORCH_DIST_INIT_BARRIER` has been + # added which enables this barrier only when set to 1. + logger.info( + "Performing barrier after ProcessGroup initialization since " + "TORCH_DIST_INIT_BARRIER = 1" + ) + if backend == Backend.MPI: + # MPI doesn't have store. + barrier() + else: + if use_local_synchronization: + if pg_store is None: + raise AssertionError("Local process group store is not initialized") + barrier_store = pg_store + else: + barrier_store = default_store + world_size = len(ranks) if use_local_synchronization else get_world_size() + # Use store based barrier here since barrier() used a bunch of + # default devices and messes up NCCL internal state. + _store_based_barrier( + global_rank, barrier_store, group_name, world_size, timeout + ) + + return pg_or_nonmember + + +def new_subgroups( + group_size: int | None = None, + group: ProcessGroup | None = None, + timeout: timedelta | None = None, + backend: str | Backend | None = None, + pg_options: object | None = None, + group_desc: str | None = None, +) -> tuple[ProcessGroup, list[ProcessGroup | _NonGroupMember]]: + """ + Create subgroups of equal size. + + By default, it creates intra-machine subgroups, + where each of which contains all the ranks of a machine, based on the assumption + that each machine has the same number of devices. + + This is a convenience API that calls ``new_group`` to generate multiple subgroups. + It requires that all processes in the main group (i.e. all + processes that are part of the distributed job) enter this function, even + if they are not going to be members of the group. + + .. warning:: + If ``group_size`` is passed in, the world size must be divisible by ``group_size``. + If no ``group_size`` is passed in, it believe that you are creating a group based + on CUDA and determining the group size by number of CUDA devices, and if not all + the machines have the same number of devices, the subgroup division will be + different across nodes and can cause unexpected behaviors. Therefore, if you are + creating a subgroup that does not depend on CUDA (such as Gloo on CPU), please + pass in ``group_size`` correctly. + + .. warning:: + See warning `Safe concurrent usage` for `new_group` API for important details about + using multiple process groups concurrently in a safe manner. + + Args: + group_size (int, optional): The size of each subgroup. If ``None``, + the default subgroup size is equal to the number of devices on each machine, + based on the assumption that each machine has exactly the same + number of devices. Default is ``None``. + group (ProcessGroup, optional): The process group to work on. If + ``None``, the default process group will be used. Default is ``None``. + timeout (timedelta, optional): see `init_process_group` for details and default value. + backend (str or Backend, optional): The backend to use. Depending on + build-time configurations, valid values are ``gloo`` and ``nccl``. + By default uses the same backend as the global group. This field + should be given as a lowercase string (e.g., ``"gloo"``), which can + also be accessed via :class:`Backend` attributes (e.g., + ``Backend.GLOO``). If ``None`` is passed in, the backend + corresponding to the default process group will be used. Default is + ``None``. + pg_options (ProcessGroupOptions, optional): process group options + specifying what additional options need to be passed in during + the construction of specific process groups. i.e. for the ``nccl`` + backend, ``is_high_priority_stream`` can be specified so that + process group can pick up high priority cuda streams. + group_desc (str, optional): A string describing the group. Each subgroup will + inherit its group_desc + + Returns: + The subgroup containing the current rank, and all the subgroups used for cleanup. + + Examples: + >>> # Create intra-machine subgroups. + >>> # xdoctest: +SKIP("need process group init") + >>> cur_subgroup, subgroups = dist.new_subgroups() + >>> # Allreduce within the machine. + >>> rank = dist.get_rank() + >>> tensor = torch.ones(1, device=rank) * rank + >>> dist.all_reduce(tensor, group=cur_subgroup) + >>> tensor + tensor([28]) # Assume 8 CUDA devices per machine. 28 is sum(range(8)). + >>> # Cleanup. + >>> for subgroup in subgroups: + >>> dist.destroy_process_group(subgroup) + """ + if group_size is None: + if not torch.accelerator.is_available(): + raise ValueError( + "Default group size only takes effect when CUDA/XPU is available." + "If your subgroup using a backend that does not depend on CUDA/XPU," + "please pass in 'group_size' correctly." + ) + group_size = torch.accelerator.device_count() + if group_size <= 0: + raise ValueError(f"The arg 'group_size' ({group_size}) must be positive") + + world_size = get_world_size(group=group) + if world_size < group_size: + raise ValueError( + f"The arg 'group_size' ({group_size}) must not exceed the world size ({world_size})" + ) + if world_size % group_size != 0: + raise ValueError( + f"The world size ({world_size}) must be divisible by '{group_size=}'" + ) + + # TODO: Use itertools.batched(get_process_group_ranks(group=group), group_size) instead when Python 3.12 is supported. + ranks = get_process_group_ranks(group=group) + ranks_per_subgroup_list = [ + ranks[i : i + group_size] for i in range(0, len(ranks), group_size) + ] + subgroup, subgroups = new_subgroups_by_enumeration( + ranks_per_subgroup_list, + timeout=timeout, + backend=backend, + pg_options=pg_options, + group_desc=group_desc, + ) + if not isinstance(subgroup, ProcessGroup): + raise AssertionError("Current rank was not assigned to a subgroup") + return subgroup, subgroups + + +def new_subgroups_by_enumeration( + ranks_per_subgroup_list: list[list[int]], + timeout: timedelta | None = None, + backend: str | Backend | None = None, + pg_options: object | None = None, + group_desc: str | None = None, +) -> tuple[ + ProcessGroup | _NonGroupMember | None, + list[ProcessGroup | _NonGroupMember], +]: + """ + Create subgroups by dividing the global world. + + The division is specified by a nested list of ranks. The subgroups cannot have + overlap, and some ranks may not have to be in any subgroup. + + This is a convenience API that calls ``new_group`` to generate multiple subgroups. + It requires that all processes in the main group (i.e. all + processes that are part of the distributed job) enter this function, even + if they are not going to be members of the group. + + .. warning:: + See warning `Safe concurrent usage` for `new_group` API for important details about + using multiple process groups concurrently in a safe manner. + + Args: + ranks_per_subgroup_list (list[list[int]]): A nested list of ranks of + group members. + timeout (timedelta, optional): see `init_process_group` for details and default value. + backend (str or Backend, optional): The backend to use. Depending on + build-time configurations, valid values are ``gloo`` and ``nccl``. + By default uses the same backend as the global group. This field + should be given as a lowercase string (e.g., ``"gloo"``), which can + also be accessed via :class:`Backend` attributes (e.g., + ``Backend.GLOO``). If ``None`` is passed in, the backend + corresponding to the default process group will be used. Default is + ``None``. + pg_options (ProcessGroupOptions, optional): process group options + specifying what additional options need to be passed in during + the construction of specific process groups. i.e. for the ``nccl`` + backend, ``is_high_priority_stream`` can be specified so that + process group can pick up high priority cuda streams. + group_desc (str, optional): A string describing the group. Each subgroup will + inherit its group_desc. + + Returns: + The subgroup containing the current rank, and all the subgroups used for cleanup. + + Examples: + >>> # Create two subgroups, where each has 2 processes. + >>> # xdoctest: +SKIP("need process group init") + >>> cur_subgroup, subgroups = dist.new_subgroups(ranks=[[0, 2], [1, 3]]) + >>> rank = dist.get_rank() + >>> tensor = torch.ones(1, device=rank) * rank + >>> dist.all_reduce(tensor, group=cur_subgroup) + >>> tensor + tensor([2]) # Subgroup 0: ranks 0 and 2 + tensor([4]) # Subgroup 1: ranks 1 and 3 + """ + if ranks_per_subgroup_list is None or len(ranks_per_subgroup_list) == 0: + raise ValueError("The arg 'ranks_per_subgroup_list' cannot be empty") + + subgroups: list[ProcessGroup | _NonGroupMember] = [] + cur_subgroup: ProcessGroup | _NonGroupMember | None = None + # Create a mapping from rank to subgroup to check if there is any subgroup overlap. + rank_to_ranks_dict: dict[int, list[int]] = {} + for ranks in ranks_per_subgroup_list: + subgroup = new_group( + ranks=ranks, + timeout=timeout, + backend=backend, + pg_options=pg_options, + group_desc=group_desc, + ) + subgroups.append(subgroup) + my_rank = get_rank() + for rank in ranks: + if rank in rank_to_ranks_dict: + raise ValueError( + f"Rank {rank} has appeared in both subgroup {rank_to_ranks_dict[rank]} and {ranks}" + ) + rank_to_ranks_dict[rank] = ranks + if my_rank == rank: + cur_subgroup = subgroup + logger.info("Rank %s is assigned to subgroup %s", rank, ranks) + + return cur_subgroup, subgroups + + +def _find_pg_by_ranks_and_tag(tag: str, ranks: list[int]) -> ProcessGroup | None: + if len(tag) > 0 and not tag.startswith("ptd:") and not tag.startswith("user:"): + tag = f"user:{tag}" + + for group in _world.tags_to_pg.get(tag, []): + if group.size() != len(ranks): + continue + + group_ranks = get_process_group_ranks(group) + good = all(r in group_ranks for r in ranks) + if good: + return group + return None + + +def _find_or_create_pg_by_ranks_and_tag( + tag: str, ranks: list[int], stride: int +) -> ProcessGroup: + if len(ranks) % stride != 0: + raise ValueError( + f"Ranks length ({len(ranks)}) must be divisible by stride ({stride})" + ) + + my_rank = get_rank() + my_ranks = None + + if stride == len(ranks): + my_ranks = ranks.copy() + if my_rank not in my_ranks: + raise AssertionError("rankset doesn't include the current node") + else: + for i in range(0, len(ranks), stride): + rank_set = ranks[i : i + stride] + if my_rank in rank_set: + my_ranks = rank_set + if my_ranks is None: + raise AssertionError("rankset doesn't include the current node") + + my_ranks = sorted(my_ranks) + + pg = _find_pg_by_ranks_and_tag(tag, my_ranks) + if pg is not None: + return pg + if tag == "": + raise ValueError("Cannot automatically create PG with empty tag") + # TODO copy settings and timeout from default PG + new_group = _new_group_with_tag(my_ranks, pg_tag=tag) + if new_group == GroupMember.NON_GROUP_MEMBER: + raise AssertionError( + f"Rank {my_rank} was not included in process group {my_ranks}" + ) + return new_group + + +def _get_group_tag(pg: ProcessGroup) -> str: + """Return the tag associated with ``pg``.""" + tag = _world.pg_to_tag[pg] + tag = tag.removeprefix("user:") + return tag + + +def _get_process_group_name(pg: ProcessGroup) -> str: + return _world.pg_names.get(pg, "None") + + +def _get_process_group_store(pg: ProcessGroup) -> Store: + return _world.pg_map[pg][1] + + +# Shrink flags for process group backends +SHRINK_DEFAULT = 0x00 +SHRINK_ABORT = 0x01 + + +class _ShrinkGroupInfo(TypedDict): + process_group: ProcessGroup + is_default_group: bool + group_size: int + current_rank: int + group_name: str + pg_options_override: NotRequired[C10DBackend.Options | None] + + +class _GroupMetadata(TypedDict): + backend_name: str + store: Store + backend_config: str + original_group_name: str + new_group_name: GroupName + bound_device_id: torch.device | None + + +@_time_logger +def shrink_group( + ranks_to_exclude: list[int], + group: ProcessGroup | None = None, + shrink_flags: int = SHRINK_DEFAULT, + pg_options: C10DBackend.Options | None = None, +) -> ProcessGroup: + """ + Shrinks a process group by excluding specified ranks. + + Creates and returns a new, smaller process group comprising only the ranks + from the original group that were not in the ``ranks_to_exclude`` list. + + Args: + ranks_to_exclude (List[int]): A list of ranks from the original + ``group`` to exclude from the new group. + group (ProcessGroup, optional): The process group to shrink. If ``None``, + the default process group is used. Defaults to ``None``. + shrink_flags (int, optional): Flags to control the shrinking behavior. + Can be ``SHRINK_DEFAULT`` (default) or ``SHRINK_ABORT``. + ``SHRINK_ABORT`` will attempt to terminate ongoing operations + in the parent communicator before shrinking. + Defaults to ``SHRINK_DEFAULT``. + pg_options (ProcessGroupOptions, optional): Backend-specific options to apply + to the shrunken process group. If provided, the backend will use + these options when creating the new group. If omitted, the new group + inherits defaults from the parent. + + Returns: + ProcessGroup: a new group comprised of the remaining ranks. If the + default group was shrunk, the returned group becomes the new default group. + + Raises: + TypeError: if the group’s backend does not support shrinking. + ValueError: if ``ranks_to_exclude`` is invalid (empty, out of bounds, + duplicates, or excludes all ranks). + RuntimeError: if an excluded rank calls this function or the backend + fails the operation. + + Notes: + - Only non-excluded ranks should call this function; excluded ranks + must not participate in the shrink operation. + - Shrinking the default group destroys all other process groups since + rank reassignment makes them inconsistent. + """ + # Step 1: Validate input parameters with comprehensive error checking + _validate_shrink_inputs(ranks_to_exclude, shrink_flags) + + # Step 2: Get target group and essential properties + target_group_info = _prepare_shrink_target_group(group) + + # Step 3: Validate backend requirements and availability + backend_impl = _validate_shrink_backend_requirements(target_group_info) + + # Step 4: Validate ranks against group and check for duplicates + excluded_ranks_set = _validate_and_process_excluded_ranks( + ranks_to_exclude, target_group_info + ) + + # Step 5: Execute the actual shrink operation (backend-specific) + new_backend = backend_impl.shrink( + sorted(excluded_ranks_set), + shrink_flags, + pg_options, + ) + + # Step 6: Handle cleanup and creation of new process group + target_group_info["pg_options_override"] = pg_options + return _finalize_shrunk_group(target_group_info, excluded_ranks_set, new_backend) + + +def _validate_shrink_inputs(ranks_to_exclude: list[int], shrink_flags: int) -> None: + """Validate input parameters for shrink_group.""" + if not isinstance(ranks_to_exclude, list): + raise TypeError( + f"ranks_to_exclude must be a list, but got {type(ranks_to_exclude).__name__}. " + f"Example: [1, 3, 5] to exclude ranks 1, 3, and 5." + ) + + if not ranks_to_exclude: + raise ValueError( + "ranks_to_exclude cannot be empty. To shrink a group, you must specify at least " + "one rank to exclude. Example: [failed_rank_id]" + ) + + # Validate shrink_flags with clear explanation of valid values + valid_flags = [SHRINK_DEFAULT, SHRINK_ABORT] + if not isinstance(shrink_flags, int) or shrink_flags not in valid_flags: + raise ValueError( + f"Invalid shrink_flags value: {shrink_flags}. Must be one of: " + f"SHRINK_DEFAULT ({SHRINK_DEFAULT}) or SHRINK_ABORT ({SHRINK_ABORT}). " + f"Use SHRINK_ABORT to abort ongoing operations before shrinking." + ) + + +def _prepare_shrink_target_group( + group: ProcessGroup | None, +) -> _ShrinkGroupInfo: + """Prepare and validate the target group for shrinking.""" + target_pg = group if group is not None else _get_default_group() + + # Cache frequently accessed properties to avoid repeated calls + group_size = int(target_pg.size()) + group_info: _ShrinkGroupInfo = { + "process_group": target_pg, + "is_default_group": (target_pg == _get_default_group()), + "group_size": group_size, + "current_rank": target_pg.rank(), + "group_name": _get_process_group_name(target_pg), + } + + # Validate that we have a valid process group + if group_size <= 1: + raise ValueError( + f"Cannot shrink a process group with size {group_size}. " + f"Group must have at least 2 ranks to support shrinking." + ) + + return group_info + + +def _validate_shrink_backend_requirements( + group_info: _ShrinkGroupInfo, +) -> C10DBackend: + """Return the backend implementation for the target group or raise if unsupported.""" + target_pg = group_info["process_group"] + group_name = group_info["group_name"] + + # Get the group's backend directly via ProcessGroup API. Prefer a bound + # device if present, otherwise try the current accelerator then fall back to CPU. + try: + preferred_device = getattr(target_pg, "bound_device_id", None) + if preferred_device is not None: + backend_impl = target_pg._get_backend(preferred_device) + else: + # Try CUDA first if available, else CPU + try: + backend_impl = target_pg._get_backend( + torch.accelerator.current_accelerator() or torch.device("cpu") + ) + except Exception: + backend_impl = target_pg._get_backend(torch.device("cpu")) + except RuntimeError as e: + raise RuntimeError( + f"Cannot access device backend for process group '{group_name}'. " + f"Ensure the process group was initialized with a compatible device backend and devices are available." + ) from e + + try: + supports = bool(backend_impl.supports_shrinking) + except Exception: + supports = False + if not supports: + raise TypeError( + f"Process group backend for '{group_name}' does not support shrinking operations." + ) + + return backend_impl + + +def _validate_and_process_excluded_ranks( + ranks_to_exclude: list[int], group_info: _ShrinkGroupInfo +) -> set[int]: + """Validate excluded ranks and convert to set for efficient operations.""" + group_size = group_info["group_size"] + current_rank = group_info["current_rank"] + + # Use set for O(1) duplicate detection and membership testing + excluded_ranks_set: set[int] = set() + + # Validate each rank with detailed error messages + for i, rank in enumerate(ranks_to_exclude): + if not isinstance(rank, int): + raise TypeError( + f"All elements in ranks_to_exclude must be integers. " + f"Element at index {i} is {type(rank).__name__}: {rank}" + ) + + if not (0 <= rank < group_size): + raise ValueError( + f"Rank {rank} at index {i} is out of bounds for group size {group_size}. " + f"Valid ranks are in range [0, {group_size - 1}]." + ) + + if rank in excluded_ranks_set: + raise ValueError( + f"Duplicate rank {rank} found in ranks_to_exclude at index {i}. " + f"Each rank can only be excluded once." + ) + + excluded_ranks_set.add(rank) + + # Ensure we don't exclude all ranks + if len(excluded_ranks_set) >= group_size: + raise ValueError( + f"Cannot exclude all {group_size} ranks from process group. " + f"At least one rank must remain. Excluding {len(excluded_ranks_set)} ranks." + ) + + # Critical check: current rank should not be in excluded list + if current_rank in excluded_ranks_set: + raise RuntimeError( + f"Current rank {current_rank} is in the exclusion list and should not call shrink_group(). " + f"Only non-excluded ranks should participate in the shrinking operation. " + f"Excluded ranks should terminate their processes instead." + ) + + return excluded_ranks_set + + +def _finalize_shrunk_group( + group_info: _ShrinkGroupInfo, + excluded_ranks_set: set[int], + new_backend: C10DBackend, +) -> ProcessGroup: + """Clean up old group and create new shrunk process group.""" + target_pg = group_info["process_group"] + is_default_group = group_info["is_default_group"] + + # Handle default group dependencies - destroy other groups first + if is_default_group: + _destroy_all_other_groups(exclude_group=target_pg) + + # Gather original group metadata before cleanup + original_group_metadata = _extract_group_metadata(target_pg) + + # Calculate remaining ranks efficiently + original_ranks = get_process_group_ranks(target_pg) + remaining_ranks = [ + rank for rank in original_ranks if rank not in excluded_ranks_set + ] + + # Clean up the original group + _cleanup_original_group(target_pg, is_default_group) + + # Create and configure the new process group + new_pg = _create_shrunk_process_group( + new_backend, remaining_ranks, original_group_metadata, is_default_group + ) + + # Register the new group in global state + if is_default_group: + _update_default_pg(new_pg) + + # Update global state with new group information + rank_mapping = { + global_rank: group_rank + for group_rank, global_rank in enumerate(remaining_ranks) + } + _register_pg_in_world( + pg=new_pg, + backend_name=original_group_metadata["backend_name"], + store=original_group_metadata["store"], + group_name=original_group_metadata["new_group_name"], + backend_config=original_group_metadata["backend_config"], + rank_mapping=rank_mapping, + ) + + return new_pg + + +def _extract_group_metadata(target_pg: ProcessGroup) -> _GroupMetadata: + """Extract metadata from the original group before cleanup.""" + original_backend_name, original_store = _world.pg_map[target_pg] + original_backend_config = _world.pg_backend_config.get(target_pg, "") + original_group_name = _get_process_group_name(target_pg) + + # Extract device binding information before cleanup to avoid accessing destroyed group + bound_device_id = None + if hasattr(target_pg, "bound_device_id"): + bound_device_id = target_pg.bound_device_id + + # Generate new group name for the shrunk group; hash for uniqueness across backends + remaining_ranks = list(get_process_group_ranks(target_pg)) + new_group_name = _process_group_name(remaining_ranks, use_hashed_name=True) + + return { + "backend_name": original_backend_name, + "store": original_store, + "backend_config": original_backend_config, + "original_group_name": original_group_name, + "new_group_name": new_group_name, + "bound_device_id": bound_device_id, # Safe to access after cleanup + } + + +def _cleanup_original_group(target_pg: ProcessGroup, is_default_group: bool) -> None: + """Clean up the original process group safely.""" + try: + destroy_process_group(target_pg) + except Exception: + group_type = "default" if is_default_group else "non-default" + logger.warning( + "Failed to destroy %s group during shrinking", group_type, exc_info=True + ) + + # Ensure global state cleanup even if destroy_process_group fails + _cleanup_process_group_global_state(target_pg) + + +def _create_shrunk_process_group( + new_backend: C10DBackend, + remaining_ranks: list[int], + metadata: _GroupMetadata, + is_default_group: bool, +) -> ProcessGroup: + """Create and configure the new shrunk process group.""" + # Create new group properties + new_group_rank = new_backend.rank() + new_group_size = new_backend.size() + group_name = metadata["new_group_name"] + + # Generate descriptive group description + if is_default_group: + group_desc = "default:shrunken" + else: + group_desc = f"{metadata['original_group_name']}:shrunk" + + # Create process group with new communicator (clone the parent store like split does) + prefix_store = PrefixStore( + f"{group_name}/", + metadata["store"].clone(), + ) + new_pg = ProcessGroup(prefix_store, new_group_rank, new_group_size) + + # Configure backend using the device type of the new backend's bound device if available, + # otherwise derive from the original group's bound device or fall back to CPU. + backend_device = metadata.get("bound_device_id") + if backend_device is None: + # Default to CPU if no bound device is present + backend_device = torch.device("cpu") + + # Choose backend enum based on device type + if backend_device.type == "cuda": + backend_type = ProcessGroup.BackendType.NCCL + elif backend_device.type == "xpu": + backend_type = ProcessGroup.BackendType.XCCL + else: + backend_type = ProcessGroup.BackendType.GLOO + + new_pg._register_backend(backend_device, backend_type, new_backend) + new_pg._set_default_backend(backend_type) + + # Inherit device binding from original group if it was bound + bound_device_id = metadata.get("bound_device_id") + if bound_device_id is not None: + new_pg.bound_device_id = bound_device_id + + # Set group metadata + new_pg._set_group_name(group_name) + new_pg._set_group_desc(group_desc) + + # Persist backend configuration overrides (if provided via shrink_group) + backend_config_override = metadata.get("backend_config") + if backend_config_override is not None: + # Store for introspection/debugging and potential backend hooks + _world.pg_backend_config[new_pg] = backend_config_override + + return new_pg + + +def _destroy_all_other_groups(exclude_group: ProcessGroup | None = None) -> None: + """ + Destroy all process groups except the excluded group and clean up all global state. + + This is necessary when shrinking the default group because global ranks + are reassigned by NCCL, making all existing process groups inconsistent. + + Note: Uses abort for non-collective cleanup since excluded ranks may not + participate in collective operations. Backend cleanup is handled independently per group. + + Args: + exclude_group (ProcessGroup, optional): Process group to exclude from destruction. + If None, destroys all process groups. + """ + # Get list of groups to destroy (avoid modifying dict while iterating) + groups_to_destroy: list[ProcessGroup] = [] + for pg in list(_world.pg_group_ranks.keys()): + if not isinstance(pg, ProcessGroup): + continue + if exclude_group is not None and pg == exclude_group: + continue + groups_to_destroy.append(pg) + + # Warn user about automatic destruction + if groups_to_destroy: + group_names = [_get_process_group_name(pg) for pg in groups_to_destroy] + logger.warning( + "Shrinking default group will destroy %d other process groups: %s. " + "This is necessary because shrinking the default group reassigns global ranks, " + "making existing groups inconsistent.", + len(groups_to_destroy), + ", ".join(group_names), + ) + + # Destroy each group and clean up global state + for pg in groups_to_destroy: + try: + # First call abort_process_group which handles the C++ cleanup non-collectively + _abort_process_group(pg) + except Exception: + # Log but don't fail - some groups might already be destroyed + logger.warning( + "Failed to abort process group %s", + _get_process_group_name(pg), + exc_info=True, + ) + + # Ensure all global state is cleaned up even if _abort_process_group fails + # or doesn't clean up everything + _cleanup_process_group_global_state(pg) + + +def _cleanup_process_group_global_state(pg: ProcessGroup) -> None: + """ + Clean up all global state associated with a process group. + + This function ensures complete cleanup of process group state from all + global dictionaries and registries, even if destroy_process_group fails + or doesn't clean up everything. This is critical when destroying multiple + groups to prevent inconsistent state. + + The cleanup removes the process group from: + - _world.pg_map (backend and store mapping) + - _world.pg_names (group name mapping) + - _world.pg_group_ranks (rank mappings) + - _world.pg_backend_config (backend configuration) + - _world.tags_to_pg and _world.pg_to_tag (tag mappings) + - _world.pg_coalesce_state (coalescing state) + - C++ internal registries via _unregister_process_group + + Args: + pg (ProcessGroup): The process group to clean up. + """ + try: + # Clean up main process group mappings + _world.pg_map.pop(pg, None) + _world.pg_group_ranks.pop(pg, None) + _world.pg_backend_config.pop(pg, None) + + # Clean up process group name mapping + group_name = _world.pg_names.pop(pg, None) + + # Clean up tag mappings + pg_tag = _world.pg_to_tag.pop(pg, None) + if pg_tag is not None and pg_tag in _world.tags_to_pg: + try: + _world.tags_to_pg[pg_tag].remove(pg) + # Remove the tag entry if list is empty + if not _world.tags_to_pg[pg_tag]: + _world.tags_to_pg.pop(pg_tag, None) + except (ValueError, KeyError): + # Process group was already removed from the list + pass + + # Clean up any registered process group names using C++ unregister function + if group_name is not None: + try: + _unregister_process_group(group_name) + except Exception: + # Process group name might not be registered or already unregistered + pass + + # Clean up coalesce state if present + _world.pg_coalesce_state.pop(pg, None) + + except Exception: + # Log cleanup failures but don't propagate - we want to continue with other cleanups + logger.warning( + "Failed to fully clean up global state for process group", exc_info=True + ) + + +def _register_pg_in_world( + pg: ProcessGroup, + backend_name: str, + store: Store, + group_name: GroupName, + backend_config: str, + rank_mapping: dict[int, int] | None = None, + pg_tag: str | None = None, + user_tag: str | None = None, + add_to_default_tag: bool = False, +) -> None: + """ + Update all global state dictionaries for a process group. + + This helper function consolidates the common pattern of updating multiple + global state dictionaries when creating or modifying process groups. + + Args: + pg (ProcessGroup): The process group to update state for. + backend_name (str): Backend name for pg_map. + store (Store): Store instance for pg_map. + group_name (str): Group name for pg_names and registration. + backend_config (str): Backend configuration string. + rank_mapping (Dict[int, int], optional): Global rank to group rank mapping. + If None, skips updating pg_group_ranks. + pg_tag (str, optional): Process group tag. If None, defaults to f"ptd:{group_name}". + user_tag (str, optional): User-provided tag for special tag handling. + If provided, creates "user:{user_tag}" tag and also adds to default "". + add_to_default_tag (bool): When True and user_tag is None, also appends + pg to the default "" tag in addition to pg_tag. Ignored when + user_tag is set (the "" tag is always appended in that case). + """ + # Update main process group mappings + _world.pg_map[pg] = (backend_name, store) + _world.pg_names[pg] = group_name + _world.pg_backend_config[pg] = backend_config + + # Register the process group name + _register_process_group(group_name, pg) + + # Update rank mapping if provided + if rank_mapping is not None: + _world.pg_group_ranks[pg] = rank_mapping + + # Handle tag management + if pg_tag is None: + pg_tag = f"ptd:{group_name}" + + if user_tag is not None: + # Special handling for user-provided tags + # Add to default "" tag first + _world.tags_to_pg.setdefault("", []).append(pg) + # Then create user-specific tag + user_pg_tag = f"user:{user_tag}" + _world.tags_to_pg.setdefault(user_pg_tag, []).append(pg) + _world.pg_to_tag[pg] = user_pg_tag + else: + # Standard process group tag + if add_to_default_tag: + _world.tags_to_pg.setdefault("", []).append(pg) + _world.tags_to_pg.setdefault(pg_tag, []).append(pg) + _world.pg_to_tag[pg] = pg_tag + + +@contextlib.contextmanager +def record_comm(name: str) -> collections.abc.Iterator[None]: + """Context manager to set a custom profiling name for communication collectives. + + When active, all c10d collectives issued within this context will use ``name`` + as their profiling title in the Work base class, overriding the default + backend-specific name (e.g. ``nccl:all_reduce``). This works across all + backends without per-backend or per-collective changes. + + Args: + name (str): The profiling name to associate with collectives. + + Example:: + >>> # xdoctest: +SKIP("undefined vars") + >>> with dist.record_comm("FSDP::all_gather (layer1)"): + ... dist.all_gather_into_tensor(output, input, group=pg) + """ + prev = torch._C._distributed_c10d._get_comm_profiling_name() + torch._C._distributed_c10d._set_comm_profiling_name(name) + try: + yield + finally: + torch._C._distributed_c10d._set_comm_profiling_name(prev) + + +def _supports_reconfigure(group: ProcessGroup | None = None) -> bool: + """ + Return whether ``group`` supports the reconfigure-based fault tolerance API. + + .. warning:: + This API is experimental and subject to change or removal. + + Args: + group (ProcessGroup, optional): The process group to query. If ``None``, + the default process group is used. + """ + pg = group or _get_default_group() + return pg.supports_reconfigure + + +def _get_reconfigure_handle(group: ProcessGroup | None = None) -> str: + """ + Return an opaque reconfigure handle for ``group``. + + The handle encodes the information peers exchange out-of-band to + (re)initialize the communicator via :func:`_reconfigure`. + + .. warning:: + This API is experimental and subject to change or removal. + + Args: + group (ProcessGroup, optional): The process group to query. If ``None``, + the default process group is used. + """ + pg = group or _get_default_group() + return pg.get_reconfigure_handle() + + +def _reconfigure( + uuid: int, + handles: set[str] | list[str], + group: ProcessGroup | None = None, + timeout: timedelta | None = None, + hints: dict[str, str] | None = None, +) -> Work: + """ + Reconfigure ``group`` with a new set of peers for fault tolerance. + + .. warning:: + This API is experimental and subject to change or removal. + + Args: + uuid (int): Uniquely identifies this instance of the communicator. Pass + a fresh value on every (re)initialization. + handles (set[str] or list[str]): One reconfigure handle per peer, as + returned by :func:`_get_reconfigure_handle`. A list assigns ranks by + position; a set lets the backend choose the rank assignment. + group (ProcessGroup, optional): The process group to reconfigure. If + ``None``, the default process group is used. + timeout (timedelta, optional): How long to allow reconfiguration before + failing. ``None`` uses the backend's default timeout. + hints (dict[str, str], optional): Backend-specific configuration. + + Returns: + Work: An async work handle for the reconfigure operation. + + Example:: + >>> # xdoctest: +SKIP("requires out-of-band rendezvous") + >>> dist.init_process_group("gloo", enable_reconfigure=True) + >>> # Every peer receives the same fresh UUID and rank-ordered handles. + >>> uuid, handles = rendezvous_reconfigure(dist._get_reconfigure_handle()) + >>> dist._reconfigure(uuid=uuid, handles=handles).wait() + """ + pg = group or _get_default_group() + opts = ReconfigureOptions() + opts.uuid = uuid + opts.handles = handles + if timeout is not None: + opts.timeout = timeout + if hints is not None: + opts.hints = hints + return pg.reconfigure(opts) + + +def _supports_window(group: ProcessGroup | None = None) -> bool: + """ + Return whether ``group`` supports one-sided (RMA) window operations. + + Args: + group (ProcessGroup, optional): The process group to query. If ``None``, + the default process group is used. + """ + pg = group or _get_default_group() + return pg.supports_window + + +def _new_window( + tensor: torch.Tensor | None = None, + group: ProcessGroup | None = None, +) -> Window: + """ + Create a new one-sided (RMA) communication window on ``group``. + + This is a collective operation. Every rank in ``group`` must call it in + the same order. If ``tensor`` is omitted, the subsequent + ``Window.tensor_register`` call is also collective and must be called by + every rank in the same order. + + Args: + tensor (torch.Tensor, optional): If provided, the tensor is registered + with the new window. + group (ProcessGroup, optional): The process group to create the window + on. If ``None``, the default process group is used. + + Returns: + Window: The new one-sided communication window. + """ + pg = group or _get_default_group() + return pg.new_window(tensor) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..54a481a90a660dfe482fbcb1f0f7d80374e743ef --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__init__.py @@ -0,0 +1,77 @@ +#!/usr/bin/env/python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" + +Torchelastic agent and user worker failover contract: + +**TL;DR;**: + +* TE(torchelastic) expects user workers to finish with the 5 minutes drift +* It is better to design DDP app to fail for all workers, rather than a single one. +* TE does not synchronize number of restarts between agents +* TE re-rendezvous does not trigger restart decrease +* When a single agent finishes its job(successfully or not), it will close rendezvous. + If other agents still have workers in progress, they will be terminated. +* Based on above, scale down does not work if at least single agent finishes the job. +* When Scale up is detected by agents, it will not decrease ``max_restarts`` + + +In general TE(torchelastic) can launch arbitrary user code, but there is some +clarifications need to be done around what failover mechanism torchelastic +provides and what failover mechanism it expects from user workers. + +Torchelastic currently supports DDP style applications. That means that +TE expects *ALL* workers finish approximately at the same time. In practice, +it is nearly impossible to guarantee that all workers in arbitrary +DDP application finish at the time, so TE provides a finalization barrier +that waits for TIMEOUT(5 minutes) for worker finalization. + +**Worker Failure** + +When worker fails, TE will check the number of restarts +available, if there is more than 0 restarts, TE will start a new rendezvous +round and restart the worker process. New rendezvous round will other +TE agents to terminate their workers. + +.. note:: The TE agent does not synchronize restarts between themselves. + When a single agent performs restart, it will trigger a local ``max_restarts`` + decrease, other agent will not decrease their ``max_restarts``. + the user to run the distributed application locally on a dev host. + +A single worker failure can cause the whole cluster to fail: +If a single worker is constantly failing, it will cause the TE agent +``max_restarts`` to go to zero. This will cause an agent to finish its +work and close rendezvous. If there are any other workers on different +agents, they will be terminated. + + +**Re-Rendezvous** + +Re-rendezvous occurs when TE agents detect a new node +trying to joint a cluster. TE will not decrease ``max_restarts``. TE agents +will terminate its workers and start a new rendezvous round. + +Note about DynamicRendezvous(etcd-v2, c10d-experimental): If the rendezvous +has already max_nodes, the new node won't be added to the wait list right +away since there is no need to tear down a rendezvous that is already fully +utilized. The new node will wait until its timeout (600 secs by default) +and periodically check the number of participants. If the number becomes +less than max_nodes, it will be added to the wait list; otherwise, it will time out after 600 secs. + +*Scale up event*. When scale up event happens, torchelastic rendezvous +will detect that there are new nodes trying to join. Torchelastic agent +will stop all workers and perform re-rendezvous. Note: when scale up event +happens, *``max_restarts``* will *not* decrease. + +*Scale down event*. When scale down event happens, rendezvous will not +notify the torchelastic agent about it. If TE agent launched with ``max_restarts=0`` , +it relies on the underlying scheduler to handle job restart. If the ``max_restarts>0`` , +TE agent will terminate workers and start a new rdzv round, which is a *Scale up event*. + +""" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c3284b4d84a3e0dafea41eef66cb6159d5cd752e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/control_plane.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/control_plane.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..22fae85c2239aa4f37a2ef5d447dfa9dce18f40a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/__pycache__/control_plane.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc7812b476132b3fba3fae8555250e6f2d27c369 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..c16fc7f1e629975e6f85003eec994bf6c342c8f8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__init__.py @@ -0,0 +1,41 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +The elastic agent is the control plane of torchelastic. + +It is a process that launches and manages underlying worker processes. +The agent is responsible for: + +1. Working with distributed torch: the workers are started with all the + necessary information to successfully and trivially call + ``torch.distributed.init_process_group()``. + +2. Fault tolerance: monitors workers and upon detecting worker failures + or unhealthiness, tears down all workers and restarts everyone. + +3. Elasticity: Reacts to membership changes and restarts workers with the new + members. + +The simplest agents are deployed per node and work with local processes. +A more advanced agent can launch and manage workers remotely. Agents can +be completely decentralized, making decisions based on the workers it manages. +Or can be coordinated, communicating to other agents (that manage workers +in the same job) to make a collective decision. +""" + +from .api import ( + ElasticAgent, + RunResult, + SimpleElasticAgent, + Worker, + WorkerGroup, + WorkerSpec, + WorkerState, +) +from .local_elastic_agent import TORCHELASTIC_ENABLE_FILE_TIMER, TORCHELASTIC_TIMER_FILE diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d3a1cb2a44c66c32397bb6dfd6e1792b763530a4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7888ae281c40a760ad724cb322b949fb7001b66c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/health_check_server.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/health_check_server.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b59e299b841eec14c9e67180b2c8482f9841c561 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/health_check_server.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/local_elastic_agent.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/local_elastic_agent.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..548ee9e77d396c2ebdfcb0e5110319233dc5cb55 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/__pycache__/local_elastic_agent.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/api.py new file mode 100644 index 0000000000000000000000000000000000000000..c1b39fc5949d2875f9f600ccbcea2197172ff21c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/api.py @@ -0,0 +1,1013 @@ +# mypy: ignore-errors + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import abc +import json +import os +import signal +import socket +import time +import traceback +import warnings +from collections import defaultdict +from collections.abc import Callable +from contextlib import contextmanager +from dataclasses import dataclass, field +from enum import Enum +from typing import Any + +import torch.distributed.elastic.rendezvous as rdzv +import torch.distributed.elastic.utils.store as store_util +from torch.distributed.elastic.events import Event, EventSource, record +from torch.distributed.elastic.metrics import prof, put_metric +from torch.distributed.elastic.multiprocessing import ProcessFailure, SignalException +from torch.distributed.elastic.rendezvous import RendezvousGracefulExitError +from torch.distributed.elastic.utils.logging import get_logger +from torch.numa.binding import NumaOptions + + +__all__ = [ + "WorkerSpec", + "Worker", + "WorkerState", + "WorkerGroup", + "RunResult", + "ElasticAgent", + "SimpleElasticAgent", +] +_TERMINAL_STATE_SYNC_ID = "torchelastic/agent/terminal_state" + +DEFAULT_ROLE = "default" +logger = get_logger(__name__) + + +@dataclass +class WorkerSpec: + """ + Blueprint information about a particular type of worker. + + For a given role, there must only exist a single worker spec. + Worker spec is expected to be homogeneous across all nodes (machine), + that is each node runs the same number of workers for a particular spec. + + Args: + role: user-defined role for the workers with this spec + local_world_size: number local workers to run + fn: (deprecated use entrypoint instead) + entrypoint: worker function or command + args: arguments to pass to ``entrypoint`` + rdzv_handler: handles rdzv for this set of workers + max_restarts: number of max retries for the workers + monitor_interval: monitor status of workers every ``n`` seconds + master_port: fixed port to run the c10d store on rank 0 + if not specified then will choose a random free port + master_addr: fixed master_addr to run the c10d store on rank 0 + if not specified then will choose hostname on agent rank 0 + redirects: redirect std streams to a file, + selectively redirect for a particular + local rank by passing a map + tee: tees the specified std stream(s) to console + file, + selectively tee for a particular local rank by passing a map, + takes precedence over ``redirects`` settings. + event_log_handler: name of the event logging handler as registered in + `elastic/events/handlers.py `_. + duplicate_stdout_filters: If non-empty, duplicates stdout to a file containing only lines + that match _any_ of the filter strings. + duplicate_stderr_filters: If non-empty, duplicates stderr to a file containing only lines + that match _any_ of the filter strings. + virtual_local_rank: Enable virtual local rank mode for workers (defaults to False). + When enabled, LOCAL_RANK is set to 0 for all workers and + CUDA_VISIBLE_DEVICES is adjusted so each worker accesses its + assigned GPU at device index 0. + """ + + role: str + local_world_size: int + rdzv_handler: rdzv.RendezvousHandler + fn: Callable | None = None + # TODO @kiuk - make entrypoint a required field + entrypoint: Callable | str | None = None + args: tuple = () + max_restarts: int = 3 + monitor_interval: float = 0.1 + master_port: int | None = None + master_addr: str | None = None + local_addr: str | None = None + event_log_handler: str = "null" + numa_options: NumaOptions | None = None + duplicate_stdout_filters: list[str] | None = None + duplicate_stderr_filters: list[str] | None = None + virtual_local_rank: bool = False + + def __post_init__(self): + if self.local_world_size <= 0: + raise AssertionError + if self.monitor_interval <= 0: + raise AssertionError + + if self.fn: + warnings.warn( + "WorkerSpec.fn will be deprecated," + " please use WorkerSpec.entrypoint instead", + stacklevel=2, + category=DeprecationWarning, + ) + self.entrypoint = self.fn + if not self.entrypoint: + raise AssertionError + + def get_entrypoint_name(self): + """Get the entry point name. + + If the entrypoint is a function (e.g. ``Callable``) returns its ``__qualname__`` + else if the entrypoint is a binary (e.g. ``str``), returns the binary name. + """ + if isinstance(self.entrypoint, str): + return os.path.basename(self.entrypoint) + else: + if self.entrypoint is None: + raise AssertionError + return self.entrypoint.__qualname__ + + +class Worker: + """A worker instance. + + Contrast this with ``WorkerSpec`` that represents the specifications of a + worker. A ``Worker`` is created from a ``WorkerSpec``. A ``Worker`` is to + a ``WorkerSpec`` as an object is to a class. + + The ``id`` of the worker is interpreted + by the specific implementation of ``ElasticAgent``. For a local + agent, it could be the ``pid (int)`` of the worker, for a remote + agent it could be encoded as ``host:port (string)``. + + Args: + id (Any): uniquely identifies a worker (interpreted by the agent) + local_rank (int): local rank of the worker + global_rank (int): global rank of the worker + role_rank (int): rank of the worker across all workers that have the same role + world_size (int): number of workers (globally) + role_world_size (int): number of workers that have the same role + """ + + __slots__ = [ + "id", + "local_rank", + "global_rank", + "role_rank", + "world_size", + "role_world_size", + ] + + def __init__( + self, + local_rank: int, + global_rank: int = -1, + role_rank: int = -1, + world_size: int = -1, + role_world_size: int = -1, + ): + # unique identifier for this worker + self.id: Any = None + + # rank of the worker among workers with the same role being monitored + # by the same ``agent`` instance. + self.local_rank: int = local_rank + + # rank of the worker among all the workers across all roles + # across all ``agent`` instances. + # Global rank is not stable between re-rendezvous. + self.global_rank: int = global_rank + + # rank of the worker among all the workers with the same role + # across all ``agent`` instances. + # Role rank is not stable between re-rendezvous. + self.role_rank: int = role_rank + + # total number of workers (globally). Due to elasticity + # the world size may change between re-rendezvous. + self.world_size: int = world_size + + # total number of workers that share the same role. Due to elasticity + # the role world size may change between re-rendezvous. + self.role_world_size: int = role_world_size + + def __str__(self): + return ( + f"local_rank={self.local_rank},global_rank={self.global_rank}" + f",role_rank={self.role_rank},world_size={self.world_size}" + f",role_world_size={self.role_world_size}" + ) + + def __repr__(self): + return str(self) + + +class WorkerState(str, Enum): + """A state of the ``WorkerGroup``. + + Workers in a worker group change state as a unit. If a single worker + in a worker group fails the entire set is considered failed:: + + UNKNOWN - agent lost track of worker group state, unrecoverable + INIT - worker group object created not yet started + HEALTHY - workers running and healthy + UNHEALTHY - workers running and unhealthy + STOPPED - workers stopped (interrupted) by the agent + SUCCEEDED - workers finished running (exit 0) + FAILED - workers failed to successfully finish (exit !0) + + + A worker group starts from an initial ``INIT`` state, + then progresses to ``HEALTHY`` or ``UNHEALTHY`` states, + and finally reaches a terminal ``SUCCEEDED`` or ``FAILED`` state. + + Worker groups can be interrupted and temporarily put into ``STOPPED`` state + by the agent. Workers in ``STOPPED`` state are scheduled to be restarted + in the near future by the agent. Some examples of workers being put into + ``STOPPED`` state are: + + 1. Worker group failure|unhealthy observed + 2. Membership change detected + + When actions (start, stop, rdzv, retry, etc) on worker group fail + and results in the action being partially applied to the worker group + the state will be ``UNKNOWN``. Typically this happens on uncaught/unhandled + exceptions during state change events on the agent. The agent is not + expected to recover worker groups in ``UNKNOWN`` state and is better off + self terminating and allowing the job manager to retry the node. + """ + + UNKNOWN = "UNKNOWN" + INIT = "INIT" + HEALTHY = "HEALTHY" + UNHEALTHY = "UNHEALTHY" + STOPPED = "STOPPED" + SUCCEEDED = "SUCCEEDED" + FAILED = "FAILED" + + @staticmethod + def is_running(state: "WorkerState") -> bool: + """Return the state of the Worker. + + Returns: + True if the worker state represents workers still running + (e.g. that the process exists but not necessarily healthy). + """ + return state in {WorkerState.HEALTHY, WorkerState.UNHEALTHY} + + +class WorkerGroup: + """A set of ``Worker`` instances. + + The class defines a set of ``Worker`` instances for the given ``WorkerSpec`` managed by ``ElasticAgent``. Whether the worker + group contains cross instance workers or not depends on the implementation of the agent. + """ + + __slots__ = [ + "spec", + "workers", + "store", + "group_rank", + "group_world_size", + "state", + "master_addr", + "master_port", + ] + + def __init__(self, spec: WorkerSpec): + self.spec = spec + self.workers = [Worker(local_rank=i) for i in range(self.spec.local_world_size)] + + # assigned after rdzv + self.store = None + self.group_rank = None + self.group_world_size = None + self.master_addr = None + self.master_port = None + + self.state = WorkerState.INIT + + +class _RoleInstanceInfo: + """The class is used by the agent to exchange the information with other agents. + + The information is used to determine the rank of the workers that agent + manages in heterogeneous environments, where different agents can have + different number of workers. + """ + + __slots__ = ["role", "rank", "local_world_size"] + + def __init__(self, role: str, rank: int, local_world_size: int): + r"""Initialize the agent class instance. + + Args: + role (str): user-defined role for the workers with this spec + rank (int): the rank of the agent + local_world_size (int): number of local workers to run + """ + self.role = role + self.rank = rank + self.local_world_size = local_world_size + + def serialize(self) -> bytes: + dict_data = { + "role": self.role, + "rank": self.rank, + "local_world_size": self.local_world_size, + } + return json.dumps(dict_data).encode(encoding="UTF-8") + + @staticmethod + def deserialize(data: bytes): + dict_data = json.loads(data.decode(encoding="UTF-8")) + return _RoleInstanceInfo( + dict_data["role"], dict_data["rank"], dict_data["local_world_size"] + ) + + @staticmethod + def compare(obj1, obj2) -> int: + if obj1.role == obj2.role: + return obj1.rank - obj2.rank + elif obj1.role > obj2.role: + return 1 + else: + return -1 + + @staticmethod + def find_role_boundaries(roles_infos: list, role: str) -> tuple[int, int]: + start_idx, end_idx = -1, -1 + for idx, role_info in enumerate(roles_infos): + if role_info.role == role: + if start_idx == -1: + start_idx = idx + end_idx = idx + return (start_idx, end_idx) + + +@dataclass +class RunResult: + """Return results of the worker executions. + + Run results follow an "all-or-nothing" policy where the run is successful if and + only if ALL local workers managed by this agent complete successfully. + + If the result is successful (e.g. ``is_failed() = False``) then the ``return_values`` + field contains the outputs (return values) of the workers managed by THIS agent mapped + by their GLOBAL ranks. That is ``result.return_values[0]`` is the return value of + global rank 0. + + .. note:: ``return_values`` are only meaningful for when the worker entrypoint + is a function. Workers specified as a binary entrypoint do not canonically + have a return value and the ``return_values`` field is meaningless and + may be empty. + + If ``is_failed()`` returns ``True`` then the ``failures`` field contains the + failure information, again, mapped by the GLOBAL rank of the worker that failed. + + The keys in ``return_values`` and ``failures`` are mutually exclusive, that is, + a worker's final state can only be one of: succeeded, failed. Workers intentionally + terminated by the agent according to the agent's restart policy, are not represented + in either ``return_values`` nor ``failures``. + """ + + state: WorkerState + return_values: dict[int, Any] = field(default_factory=dict) + failures: dict[int, ProcessFailure] = field(default_factory=dict) + + def is_failed(self) -> bool: + return self.state == WorkerState.FAILED + + +def _get_fq_hostname() -> str: + return socket.getfqdn(socket.gethostname()) + + +class ElasticAgent(abc.ABC): + """An agent process responsible for managing one or more worker processes. + + The worker processes are assumed to be regular distributed PyTorch scripts. + When the worker process is created by the agent, the agent provides the + necessary information for the worker processes to properly initialize + a torch process group. + + The exact deployment topology and ratio of agent-to-worker is dependent + on the specific implementation of the agent and the user's job placement + preferences. For instance, to run a distributed training job on GPU with + 8 trainers (one per GPU) one can: + + 1. Use 8 x single GPU instances, place an agent per instance, managing + 1 worker per agent. + 2. Use 4 x double GPU instances, place an agent per instance, managing + 2 workers per agent. + 3. Use 2 x quad GPU instances, place an agent per instance, managing + 4 workers per agent. + 4. Use 1 x 8 GPU instance, place an agent per instance, managing + 8 workers per agent. + + Usage + :: + + group_result = agent.run() + if group_result.is_failed(): + # workers failed + failure = group_result.failures[0] + logger.exception("worker 0 failed with exit code : %s", failure.exit_code) + else: + return group_result.return_values[0] # return rank 0's results + + """ + + @abc.abstractmethod + def run(self, role: str = DEFAULT_ROLE) -> RunResult: + """Run the agent. + + Supports retrying the worker group on failures up to ``max_restarts``. + + Returns: + The result of the execution, containing the return values or + failure details for each worker mapped by the worker's global rank. + + Raises: + Exception - any other failures NOT related to worker process + """ + raise NotImplementedError + + @abc.abstractmethod + def get_worker_group(self, role: str = DEFAULT_ROLE) -> WorkerGroup: + """Return the ``WorkerGroup`` for the given ``role``. + + Note that the worker group is a mutable object and hence in a + multi-threaded/process environment it may change state. + Implementers are encouraged (but not required) to return + a defensive read-only copy. + """ + raise NotImplementedError + + +class SimpleElasticAgent(ElasticAgent): + """An ``ElasticAgent`` that manages one particular type of worker role. + + An ``ElasticAgent`` that manages workers (``WorkerGroup``) for a single ``WorkerSpec`` + such as one particular type of worker role. + """ + + def __init__( + self, + spec: WorkerSpec, + exit_barrier_timeout: float = 300, + shutdown_timeout: int = 30, + ): + self._worker_group = WorkerGroup(spec) + self._remaining_restarts = self._worker_group.spec.max_restarts + self._store = None + self._exit_barrier_timeout = exit_barrier_timeout + self._shutdown_timeout = shutdown_timeout + self._total_execution_time = 0 + self._in_exit_barrier: bool = False + + def get_worker_group(self, role: str = DEFAULT_ROLE) -> WorkerGroup: + return self._worker_group + + @abc.abstractmethod + def _start_workers(self, worker_group: WorkerGroup) -> dict[int, Any]: + r"""Start ``worker_group.spec.local_world_size`` number of workers. + + This is according to worker spec for the worker group . + Returns a map of ``local_rank`` to worker ``id``. + """ + raise NotImplementedError + + @abc.abstractmethod + def _stop_workers(self, worker_group: WorkerGroup) -> None: + r"""Stop all workers in the given worker group. + + Implementers must deal with workers in all states defined by + ``WorkerState``. That is, it must gracefully handle stopping + non-existent workers, unhealthy (stuck) workers, etc. + """ + raise NotImplementedError + + @abc.abstractmethod + def _monitor_workers(self, worker_group: WorkerGroup) -> RunResult: + r"""Check on the workers for the ``worker_group``. + + This function also returns the new state of the worker group. + """ + raise NotImplementedError + + @abc.abstractmethod + def _shutdown( + self, death_sig: signal.Signals = signal.SIGTERM, timeout: int = 30 + ) -> None: + """Clean up any resources that were allocated during the agent's work. + + Args: + death_sig: Signal to send to the child process, SIGTERM is default + timeout: Time to wait for graceful shutdown before sending SIGKILL + """ + raise NotImplementedError + + @prof + def _rendezvous(self, worker_group: WorkerGroup) -> None: + r"""Run rendezvous for the workers specified by the worker spec. + + Assigns workers a new global rank and world size. + Updates the rendezvous store for the worker group. + """ + spec = worker_group.spec + + with self.record_duration("RENDEZVOUS"): + rdzv_info = spec.rdzv_handler.next_rendezvous() + store = rdzv_info.store + group_rank = rdzv_info.rank + group_world_size = rdzv_info.world_size + + # master_addr/master_port could be explicitly overridden + # TODO: BC - specific to static rdzv and can be simplified further + master_addr = spec.master_addr or rdzv_info.bootstrap_store_info.master_addr + master_port = spec.master_port or rdzv_info.bootstrap_store_info.master_port + + self._store = store + + with self.record_duration("ASSIGN_WORKER_RANKS"): + workers = self._assign_worker_ranks( + store, group_rank, group_world_size, spec + ) + worker_group.workers = workers + worker_group.store = store + worker_group.group_rank = group_rank + worker_group.group_world_size = group_world_size + worker_group.master_addr = master_addr + worker_group.master_port = master_port + + restart_count = spec.max_restarts - self._remaining_restarts + + logger.info( + "[%(role)s] Rendezvous complete for workers. Result:\n" + " restart_count=%(restart_count)s\n" + " master_addr=%(master_addr)s\n" + " master_port=%(master_port)s\n" + " group_rank=%(group_rank)s\n" + " group_world_size=%(group_world_size)s\n" + " local_ranks=%(local_ranks)s\n" + " role_ranks=%(role_ranks)s\n" + " global_ranks=%(global_ranks)s\n" + " role_world_sizes=%(role_world_sizes)s\n" + " global_world_sizes=%(global_world_sizes)s\n" + " event_log_handler=%(event_log_handler)s\n", + { + "role": spec.role, + "restart_count": restart_count, + "master_addr": master_addr, + "master_port": master_port, + "group_rank": group_rank, + "group_world_size": group_world_size, + "local_ranks": [worker.local_rank for worker in workers], + "role_ranks": [worker.role_rank for worker in workers], + "global_ranks": [worker.global_rank for worker in workers], + "role_world_sizes": [worker.role_world_size for worker in workers], + "global_world_sizes": [worker.world_size for worker in workers], + "event_log_handler": spec.event_log_handler, + }, + ) + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _assign_worker_ranks( + self, store, group_rank: int, group_world_size: int, spec: WorkerSpec + ) -> list[Worker]: + """Determine proper ranks for worker processes. + + Fast Path: when all workers have the same role and world size. We calculate + the global rank to be group_rank * group_world_size + local_rank. And the + `role_world_size` is the same as `global_world_size`. No TCP store is used in + this case. This is only enabled when users set the environment variable + `TORCH_ELASTIC_WORKER_IDENTICAL` to 1. + + Time complexity: each worker O(1), overall O(1) + + Slow Path: when workers have different roles and world sizes. We use the + following algorithm: + + 1. Each agent writes its configuration(group_rank, group_world_size + , num_workers) to the common store. + 2. The rank 0 agent reads all the role_info from the store and + determines each agents worker ranks. + 3. Determine the global rank: the global rank of the workers is computed + by cumulative sum of the local_world_size for all workers in front of it. + For efficiency reasons each worker is assigned a base global rank + such that its workers are in the range [base_global_rank, + base_global_rank + local_world_size). + 4. Determine the role rank: The role rank is determined using the algorithms + in the point 3 with the exception that the ranks are calculated with + respect to the role name. + 5. The rank 0 agent writes the assigned ranks to the store. + 6. Each agent reads the assigned ranks from the store. + + Time complexity: each worker O(1), rank0 O(n), overall O(n) + """ + + if os.environ.get("TORCH_ELASTIC_WORKER_IDENTICAL", "0") == "1": + global_world_size = group_world_size * spec.local_world_size + base_global_rank = group_rank * spec.local_world_size + base_role_rank = base_global_rank + role_world_size = global_world_size + else: + ROLE_INFO_PREFIX = "torchelastic/role_info/" + ASSIGNED_RANKS_PREFIX = "torchelastic/assigned_ranks/" + + agent_role_info = _RoleInstanceInfo( + spec.role, group_rank, spec.local_world_size + ) + store.set(f"{ROLE_INFO_PREFIX}{group_rank}", agent_role_info.serialize()) + + # tcp store is collocated with rank 0 so we can use it to do extra compute to reduce overall # of operations. + if group_rank == 0: + role_infos_bytes = store.multi_get( + [f"torchelastic/role_info/{i}" for i in range(group_world_size)] + ) + role_infos = [ + _RoleInstanceInfo.deserialize(info_bytes) + for info_bytes in role_infos_bytes + ] + + role_sizes = defaultdict(lambda: 0) + global_size = 0 + for role_info in role_infos: + role_sizes[role_info.role] += role_info.local_world_size + global_size += role_info.local_world_size + + base_global_rank = 0 + role_ranks = defaultdict(lambda: 0) + + keys = [] + values = [] + for i, role_info in enumerate(role_infos): + keys.append(f"{ASSIGNED_RANKS_PREFIX}{i}") + values.append( + json.dumps( + [ + base_global_rank, + global_size, + role_ranks[role_info.role], + role_sizes[role_info.role], + ] + ) + ) + + base_global_rank += role_info.local_world_size + role_ranks[role_info.role] += role_info.local_world_size + + store.multi_set(keys, values) + + # get will block until the data is available in the store. + ( + base_global_rank, + global_world_size, + base_role_rank, + role_world_size, + ) = json.loads(store.get(f"{ASSIGNED_RANKS_PREFIX}{group_rank}")) + + workers = [] + for local_rank in range(spec.local_world_size): + worker = Worker( + local_rank=local_rank, + global_rank=base_global_rank + local_rank, + role_rank=base_role_rank + local_rank, + world_size=global_world_size, + role_world_size=role_world_size, + ) + workers.append(worker) + return workers + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _initialize_workers(self, worker_group: WorkerGroup) -> None: + r"""Start a fresh set of workers for the worker_group. + + Essentially, a rendezvous followed by a ``start_workers``. + The caller should first call ``_stop_workers()`` to stop running workers + prior to calling this method. + + Optimistically sets the state of the worker group that + just started as ``HEALTHY`` and delegates the actual monitoring + of state to ``_monitor_workers()`` method + """ + role = worker_group.spec.role + logger.info("[%s] Rendezvous'ing worker group", role) + + # TODO after stopping workers, wait at least monitor_interval*2 for + # workers on different nodes to fail on a collective op before waiting + # on the rdzv barrier, this way we ensure that nodes enter rdzv + # at around the same time and reduce false positive rdzv timeout errors + self._rendezvous(worker_group) + + logger.info("[%s] Starting worker group", role) + worker_ids = self._start_workers(worker_group) + for local_rank, w_id in worker_ids.items(): + worker = worker_group.workers[local_rank] + worker.id = w_id + record( + self._construct_event("START", EventSource.WORKER, worker), + worker_group.spec.event_log_handler, + ) + + worker_group.state = WorkerState.HEALTHY + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _restart_workers(self, worker_group: WorkerGroup) -> None: + """Restart (stops, rendezvous, starts) all local workers in the group.""" + role = worker_group.spec.role + logger.info("[%s] Stopping worker group", role) + self._stop_workers(worker_group) + worker_group.state = WorkerState.STOPPED + self._initialize_workers(worker_group) + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def run(self, role: str = DEFAULT_ROLE) -> RunResult: + start_time = time.monotonic() + shutdown_called: bool = False + try: + result = self._invoke_run(role) + self._total_execution_time = int(time.monotonic() - start_time) + self._record_metrics(result) + self._record_worker_events(result) + return result + except RendezvousGracefulExitError as e: + logger.info("Rendezvous gracefully exited: %s", e) + except SignalException as e: + logger.warning("Received %s death signal, shutting down workers", e.sigval) + self._shutdown(e.sigval, timeout=self._shutdown_timeout) + shutdown_called = True + raise + finally: + if not shutdown_called: + self._shutdown(timeout=self._shutdown_timeout) + # record the execution time in case there were any exceptions during run. + self._total_execution_time = int(time.monotonic() - start_time) + + def get_event_failed(self) -> Event: + return self._construct_event( + state="FAILED", + source=EventSource.AGENT, + raw_error=traceback.format_exc(), + ) + + def get_event_succeeded(self) -> Event: + return self._construct_event( + state="SUCCEEDED", + source=EventSource.AGENT, + ) + + def _record_worker_events(self, result: RunResult) -> None: + for worker in self._worker_group.workers: + failure = result.failures.get(worker.global_rank) + state: str = self._get_worker_state(worker, result) + raw_error = json.dumps(failure.error_file_data) if failure else None + exit_code = failure.exitcode if failure else None + worker_pid = failure.pid if failure else None + record( + self._construct_event( + state=state, + source=EventSource.WORKER, + worker=worker, + raw_error=raw_error, + exit_code=exit_code, + worker_pid=worker_pid, + ), + self._worker_group.spec.event_log_handler, + ) + + def _get_worker_state(self, worker: Worker, result: RunResult) -> str: + failure = result.failures.get(worker.global_rank) + if result.state in {WorkerState.UNHEALTHY, WorkerState.FAILED} and not failure: + # The worker got terminated by the torchelastic agent via SIGTERM signal + return "TERMINATED" + elif failure or worker.global_rank in result.return_values: + return result.state.value + else: + raise ValueError(f"Unknown worker: {worker.global_rank}") + + @contextmanager + def record_duration(self, state: str): + start_time = time.perf_counter() + try: + yield + finally: + end_time = time.perf_counter() + duration_ms = (end_time - start_time) * 1000 + record( + self._construct_event( + state=state, source=EventSource.AGENT, duration_ms=duration_ms + ), + self._worker_group.spec.event_log_handler, + ) + + def _construct_event( + self, + state: str, + source: EventSource, + worker: Worker | None = None, + raw_error: str | None = None, + duration_ms: float | None = None, + exit_code: int | None = None, + worker_pid: int | None = None, + ) -> Event: + wg = self._worker_group + spec = wg.spec + md = { + "group_world_size": wg.group_world_size, + "entry_point": spec.get_entrypoint_name(), + } + if worker: + md["local_rank"] = (worker.local_rank,) + md["role_rank"] = (worker.role_rank,) + md["role_world_size"] = (worker.role_world_size,) + md["exit_code"] = (exit_code,) + md["worker_pid"] = (worker_pid,) + global_rank = worker.global_rank + worker_id = str(worker.id) + else: + global_rank = None + worker_id = None + md_str = json.dumps(md) + metadata = { + "run_id": spec.rdzv_handler.get_run_id(), + "global_rank": global_rank, + "group_rank": wg.group_rank, + "worker_id": worker_id, + "role": spec.role, + "hostname": _get_fq_hostname(), + "state": state, + "total_run_time": self._total_execution_time, + "rdzv_backend": spec.rdzv_handler.get_backend(), + "raw_error": raw_error, + "metadata": md_str, + "agent_restarts": spec.max_restarts - self._remaining_restarts, + "duration_ms": duration_ms, + } + + return Event( + f"torchelastic.worker.status.{state}", source=source, metadata=metadata + ) + + def _record_metrics(self, group_results: RunResult): + is_failed = group_results.is_failed() + self._record_flakiness_metric(is_failed) + spec = self._worker_group.spec + restarts_happened = self._remaining_restarts != spec.max_restarts + put_metric(f"workers.{spec.role}.run_total", 1) + self._record_metric_with_condition( + "run_success_with_retries", not is_failed and restarts_happened + ) + self._record_metric_with_condition( + "run_success_no_retries", not is_failed and not restarts_happened + ) + self._record_metric_with_condition( + "run_failed_with_retries", is_failed and restarts_happened + ) + self._record_metric_with_condition( + "run_failed_no_retries", is_failed and not restarts_happened + ) + + def _record_metric_with_condition(self, metric_name, condition): + spec = self._worker_group.spec + if condition: + put_metric(f"workers.{spec.role}.{metric_name}", 1) + else: + put_metric(f"workers.{spec.role}.{metric_name}", 0) + + def _record_flakiness_metric(self, is_failed: bool = False): + if is_failed: + flakiness = 100.0 + else: + spec = self._worker_group.spec + flakiness = 100.0 - 100.0 * (self._remaining_restarts + 1) / ( + spec.max_restarts + 1 + ) + spec = self._worker_group.spec + + put_metric(f"workers.{spec.role}.flakiness", int(flakiness)) + + def _invoke_run(self, role: str = DEFAULT_ROLE) -> RunResult: + # NOTE: currently only works for a single role + + spec = self._worker_group.spec + role = spec.role + + logger.info( + "[%s] starting workers for entrypoint: %s", role, spec.get_entrypoint_name() + ) + + self._initialize_workers(self._worker_group) + monitor_interval = spec.monitor_interval + rdzv_handler = spec.rdzv_handler + + while True: + if self._worker_group.state == WorkerState.INIT: + raise AssertionError + time.sleep(monitor_interval) + run_result = self._monitor_workers(self._worker_group) + state = run_result.state + self._worker_group.state = state + + put_metric(f"workers.{role}.remaining_restarts", self._remaining_restarts) + put_metric(f"workers.{role}.{state.name.lower()}", 1) + + if state == WorkerState.SUCCEEDED: + logger.info( + "[%s] worker group successfully finished." + " Waiting %s seconds for other agents to finish.", + role, + self._exit_barrier_timeout, + ) + self._exit_barrier() + return run_result + elif state in {WorkerState.UNHEALTHY, WorkerState.FAILED}: + if self._remaining_restarts > 0: + logger.info( + "[%s] Worker group %s. " + "%s/%s attempts left;" + " will restart worker group", + role, + state.name, + self._remaining_restarts, + spec.max_restarts, + ) + self._remaining_restarts -= 1 + self._restart_workers(self._worker_group) + else: + self._stop_workers(self._worker_group) + self._worker_group.state = WorkerState.FAILED + return run_result + elif state == WorkerState.HEALTHY: + # membership changes do not count as retries + num_nodes_waiting = rdzv_handler.num_nodes_waiting() + group_rank = self._worker_group.group_rank + if num_nodes_waiting > 0: + logger.info( + "[%s] Detected %s " + "new nodes from group_rank=%s; " + "will restart worker group", + role, + num_nodes_waiting, + group_rank, + ) + self._restart_workers(self._worker_group) + else: + raise Exception( # noqa: TRY002 + f"[{role}] Worker group in {state.name} state" + ) + + def _exit_barrier(self): + """ + Define a barrier that keeps the agent process alive until all workers finish. + + Wait for ``exit_barrier_timeout`` seconds for all agents to finish + executing their local workers (either successfully or not). This + acts as a safety guard against user scripts that terminate at different + times. + """ + logger.info( + "Local worker group finished (%s). " + "Waiting %s seconds for other agents to finish", + self._worker_group.state, + self._exit_barrier_timeout, + ) + start = time.time() + self._in_exit_barrier = True + try: + store_util.barrier( + store=self._store, + world_size=self._worker_group.group_world_size, + key_prefix=_TERMINAL_STATE_SYNC_ID, + barrier_timeout=self._exit_barrier_timeout, + ) + logger.info( + "Done waiting for other agents. Elapsed: %s seconds", + time.time() - start, + ) + except SignalException as e: + logger.warning("Got termination signal: %s", e.sigval) + raise + except Exception: + logger.exception( + "Error waiting on exit barrier. Elapsed: %s seconds", + time.time() - start, + ) + finally: + self._in_exit_barrier = False diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/health_check_server.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/health_check_server.py new file mode 100644 index 0000000000000000000000000000000000000000..a371be053c36a62d0e23d82d5550c962531129eb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/health_check_server.py @@ -0,0 +1,69 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from collections.abc import Callable + +from torch.distributed.elastic.utils.logging import get_logger + + +log = get_logger(__name__) + +__all__ = ["HealthCheckServer", "create_healthcheck_server"] + + +class HealthCheckServer: + """ + Interface for health check monitoring server, which can be extended + by starting tcp/http server on the specified port. + + Args: + + alive_callback: Callable[[], int], callback to last progress time of agent + + port: int, port number to start tcp/http server + + timeout: int, timeout seconds to decide agent is alive/dead + """ + + _alive_callback: Callable[[], int] + _port: int + _timeout: int + + def __init__( + self, alive_callback: Callable[[], int], port: int, timeout: int + ) -> None: + self._alive_callback = alive_callback + self._port = port + self._timeout = timeout + + def start(self) -> None: + """ + Unsupported functionality for PyTorch, doesn't start any health check server + """ + log.warning("No health check server started") + + def stop(self) -> None: + """ + Function to stop health check server + """ + log.info("Stopping noop health check server.") + + @property + def alive_callback(self) -> Callable[[], int]: + return self._alive_callback + + +def create_healthcheck_server( + alive_callback: Callable[[], int], + port: int, + timeout: int, +) -> HealthCheckServer: + """ + creates health check server object + """ + return HealthCheckServer(alive_callback, port, timeout) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py new file mode 100644 index 0000000000000000000000000000000000000000..a7554de0b92177f3491f7e8f3e4e07fdd9a94ab5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/agent/server/local_elastic_agent.py @@ -0,0 +1,655 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from __future__ import annotations + +import json +import os +import signal +import socket +import tempfile +import time +import uuid +from string import Template +from typing import Any, TYPE_CHECKING + +import torch.distributed.elastic.timer as timer +from torch._utils_internal import justknobs_check +from torch.distributed.elastic import events +from torch.distributed.elastic.agent.server.api import ( + RunResult, + SimpleElasticAgent, + WorkerGroup, + WorkerSpec, + WorkerState, +) +from torch.distributed.elastic.agent.server.health_check_server import ( + create_healthcheck_server, + HealthCheckServer, +) +from torch.distributed.elastic.metrics.api import prof +from torch.distributed.elastic.multiprocessing import ( + LogsSpecs, + PContext, + start_processes, +) +from torch.distributed.elastic.utils import macros +from torch.distributed.elastic.utils.logging import get_logger +from torch.distributed.elastic.utils.process_state import ( + is_uninterruptible_state, + read_proc_state, +) + + +if TYPE_CHECKING: + from collections.abc import Callable + + from torch.distributed.elastic.events.api import EventMetadataValue + +logger = get_logger(__name__) + +__all__ = [ + "LocalElasticAgent", + "TORCHELASTIC_ENABLE_FILE_TIMER", + "TORCHELASTIC_TIMER_FILE", + "TORCHELASTIC_HEALTH_CHECK_PORT", + "TORCHELASTIC_UNINTERRUPTIBLE_STATE_TIMEOUT", +] + +TORCHELASTIC_ENABLE_FILE_TIMER = "TORCHELASTIC_ENABLE_FILE_TIMER" +TORCHELASTIC_HEALTH_CHECK_PORT = "TORCHELASTIC_HEALTH_CHECK_PORT" +TORCHELASTIC_TIMER_FILE = "TORCHELASTIC_TIMER_FILE" +# Seconds a worker may spend in Linux uninterruptible kernel sleep (D-state) +# before the agent marks the worker group UNHEALTHY. Used as the default when +# ``uninterruptible_state_timeout`` is not passed to ``LocalElasticAgent``. +# Unset / non-positive disables the check. Linux-only. +TORCHELASTIC_UNINTERRUPTIBLE_STATE_TIMEOUT = ( + "TORCHELASTIC_UNINTERRUPTIBLE_STATE_TIMEOUT" +) + + +def _resolve_uninterruptible_state_timeout(explicit: float | None) -> float: + """Resolve the uninterruptible-state timeout once. + + Precedence: explicit constructor arg > ``TORCHELASTIC_UNINTERRUPTIBLE_STATE_TIMEOUT`` + env var > 0 (disabled). Non-positive / unparsable values disable the check. + """ + if explicit is not None: + return max(0.0, float(explicit)) + raw = os.environ.get(TORCHELASTIC_UNINTERRUPTIBLE_STATE_TIMEOUT, "") + if not raw: + return 0.0 + try: + value = float(raw) + except ValueError: + return 0.0 + return max(0.0, value) + + +class _AliveCallbackProxy: + """Mutable callback wrapper for the health check server. + + The C++ pybind ``HealthCheckThriftServer`` binds its ``alive_callback`` + at construction time and cannot update it afterward. This proxy is + created *before* the health check server so it can be passed as the + callback. Initially it returns ``time.time()`` (signalling "alive"). + After the agent is constructed, :meth:`set_delegate` wires it to + ``agent._get_alive_time`` for real liveness tracking. + """ + + def __init__(self) -> None: + self._delegate: Callable[[], int] | None = None + + def __call__(self) -> int: + if self._delegate is not None: + return self._delegate() + return int(time.time()) + + def set_delegate(self, delegate: Callable[[], int]) -> None: + self._delegate = delegate + + +class LocalElasticAgent(SimpleElasticAgent): + """An implementation of :py:class:`torchelastic.agent.server.ElasticAgent` that handles host-local workers. + + This agent is deployed per host and is configured to spawn ``n`` workers. + When using GPUs, ``n`` maps to the number of GPUs available on the host. + + The local agent does not communicate to other local agents deployed on + other hosts, even if the workers may communicate inter-host. The worker id + is interpreted to be a local process. The agent starts and stops all worker + processes as a single unit. + + + The worker function and argument passed to the worker function must be + python multiprocessing compatible. To pass multiprocessing data structures + to the workers you may create the data structure in the same multiprocessing + context as the specified ``start_method`` and pass it as a function argument. + + The ``exit_barrier_timeout`` specifies the amount of time (in seconds) to wait + for other agents to finish. This acts as a safety net to handle cases where + workers finish at different times, to prevent agents from viewing workers + that finished early as a scale-down event. It is strongly advised that the + user code deal with ensuring that workers are terminated in a synchronous + manner rather than relying on the exit_barrier_timeout. + + A named pipe based watchdog can be enabled in ```LocalElasticAgent``` if an + environment variable ``TORCHELASTIC_ENABLE_FILE_TIMER`` with value 1 has + been defined in the ```LocalElasticAgent``` process. + Optionally, another environment variable ```TORCHELASTIC_TIMER_FILE``` + can be set with a unique file name for the named pipe. If the environment + variable ```TORCHELASTIC_TIMER_FILE``` is not set, ```LocalElasticAgent``` + will internally create a unique file name and set it to the environment + variable ```TORCHELASTIC_TIMER_FILE```, and this environment variable will + be propagated to the worker processes to allow them to connect to the same + named pipe that ```LocalElasticAgent``` uses. + + Logs are written to the specified log directory. Each log line will be by default + prefixed by ``[${role_name}${local_rank}]:`` (e.g. ``[trainer0]: foobar``). + Log prefixes can be customized by passing a `template string + `_ as the + ``log_line_prefix_template`` argument. + The following macros (identifiers) are substituted at runtime: + ``${role_name}, ${local_rank}, ${rank}, ${hostname}``. + For example, to prefix each log line with + global rank instead of the local rank, set ``log_line_prefix_template = "[${rank}]:``. + ``${hostname}`` expands to the name of the node the agent runs on, which + identifies the offending node in a multi-node job; for example + ``log_line_prefix_template = "${hostname}:${rank}: "`` renders as + ``r12i0n8:3: foobar``. + + + Example launching function + + :: + + def trainer(args) -> str: + return "do train" + + def main(): + start_method="spawn" + shared_queue= multiprocessing.get_context(start_method).Queue() + spec = WorkerSpec( + role="trainer", + local_world_size=nproc_per_process, + entrypoint=trainer, + args=("foobar",), + ...) + agent = LocalElasticAgent(spec, start_method) + results = agent.run() + + if results.is_failed(): + print("trainer failed") + else: + print(f"rank 0 return value: {results.return_values[0]}") + # prints -> rank 0 return value: do train + + Example launching binary + + :: + + def main(): + spec = WorkerSpec( + role="trainer", + local_world_size=nproc_per_process, + entrypoint="/usr/local/bin/trainer", + args=("--trainer-args", "foobar"), + ...) + agent = LocalElasticAgent(spec) + results = agent.run() + + if not results.is_failed(): + print("binary launches do not have return values") + + """ + + def __init__( + self, + spec: WorkerSpec, + logs_specs: LogsSpecs, + start_method="spawn", + exit_barrier_timeout: float = 300, + log_line_prefix_template: str | None = None, + shutdown_timeout: int = 30, + health_check_server: HealthCheckServer | None = None, + uninterruptible_state_timeout: float | None = None, + ): + super().__init__(spec, exit_barrier_timeout, shutdown_timeout) + self._start_method = start_method + self._pcontext: PContext | None = None + self._rdzv_handler = spec.rdzv_handler + self._log_line_prefix_template = log_line_prefix_template + self._worker_watchdog: timer.FileTimerServer | None = None + self._logs_specs = logs_specs + self._health_check_server = health_check_server + # Resolve the uninterruptible-state timeout once at construction time. + # Explicit constructor arg wins; otherwise fall back to the env var. + # Non-positive / unparsable values disable the check. + self._uninterruptible_state_timeout: float = ( + _resolve_uninterruptible_state_timeout(uninterruptible_state_timeout) + ) + # Maps PID -> monotonic timestamp when the PID was first observed in + # an uninterruptible (D-state) sleep. Cleared per-PID when the worker + # leaves that state or exits. + self._uninterruptible_state_first_seen: dict[int, float] = {} + + def _setup_local_watchdog(self, envs: dict[int, dict[str, str]]) -> None: + enable_watchdog_env_name = TORCHELASTIC_ENABLE_FILE_TIMER + watchdog_enabled = os.getenv(enable_watchdog_env_name) + watchdog_file_env_name = TORCHELASTIC_TIMER_FILE + watchdog_file_path = os.getenv(watchdog_file_env_name) + if watchdog_enabled is not None and str(watchdog_enabled) == "1": + if watchdog_file_path is None: + watchdog_file_path = os.path.join( + tempfile.gettempdir(), "watchdog_timer_" + str(uuid.uuid4()) + ) + logger.info("Starting a FileTimerServer with %s ...", watchdog_file_path) + if not envs: + logger.warning( + "Empty envs variables, using empty run_id for FileTimerServer" + ) + run_id = "" + else: + run_id = envs[0]["TORCHELASTIC_RUN_ID"] + self._worker_watchdog = timer.FileTimerServer( + file_path=watchdog_file_path, + run_id=run_id, + max_interval=0.1, + daemon=True, + log_event=self._log_watchdog_event, + ) + self._worker_watchdog.start() + logger.info("FileTimerServer started") + else: + logger.info( + "Environment variable '%s' not found. Do not start FileTimerServer.", + enable_watchdog_env_name, + ) + # Propagate the watchdog file env to worker processes + if watchdog_file_path is not None: + for worker_env in envs.values(): + worker_env[watchdog_file_env_name] = watchdog_file_path + + @staticmethod + def _get_current_time_secs() -> int: + return int(time.time()) + + def _get_alive_time(self) -> int: + """Return the last progress time from the watchdog, or the current time. + + This callback is passed to the health check server at startup and + is called on every TW health check poll. During initialization + (before rendezvous and worker launch), the watchdog does not exist + yet, so we return the current time to signal the agent is alive. + Once workers are running and the watchdog is active, we delegate + to the watchdog's ``get_last_progress_time`` for real liveness + tracking. + + During the exit barrier wait, workers have finished and the watchdog + progress time is stale. We return the current time to prevent TW + from killing the task while agents coordinate shutdown. + """ + if self._in_exit_barrier: + return int(time.time()) + if self._worker_watchdog is not None: + return self._worker_watchdog.get_last_progress_time() + return int(time.time()) + + def _setup_healthcheck(self) -> None: + healthcheck_port_env_name = TORCHELASTIC_HEALTH_CHECK_PORT + healthcheck_port = os.getenv(healthcheck_port_env_name) + if healthcheck_port is not None: + logger.info( + "Found healthcheck port %s: %s", + healthcheck_port_env_name, + healthcheck_port, + ) + + if justknobs_check( + "ai_infra/pytorch_distributed:torchelastic_enable_healthcheck_before_rendezvous", + default=False, + ): + # New behavior: idempotent guard + dynamic callback that + # returns current time before watchdog exists and delegates + # to watchdog once workers are running. + if self._health_check_server is not None: + return + alive_callback = self._get_alive_time + else: + # Original behavior: pick callback based on watchdog state + # at call time (only called from _start_workers where + # watchdog is already set up). + if self._worker_watchdog is None: + logger.info( + "FileTimerServer doesn't exist, using current time as dummy callback" + ) + alive_callback = LocalElasticAgent._get_current_time_secs + else: + alive_callback = self._worker_watchdog.get_last_progress_time + + try: + healthcheck_port_as_int = int(healthcheck_port) + self._health_check_server = create_healthcheck_server( + alive_callback=alive_callback, + port=healthcheck_port_as_int, + timeout=60, + ) + self._health_check_server.start() + except ValueError: + logger.info( + "Invalid healthcheck port value: '%s', expecting integer. Not starting healthcheck server.", + healthcheck_port, + ) + else: + logger.info( + "Environment variable '%s' not found. Do not start health check.", + healthcheck_port_env_name, + ) + + def _get_fq_hostname(self) -> str: + return socket.getfqdn(socket.gethostname()) + + def _log_watchdog_event( + self, + name: str, + request: timer.FileTimerRequest | None, + ) -> None: + wg = self._worker_group + spec = wg.spec + md = {"watchdog_event": name} + if request is not None: + md["worker_pid"] = str(request.worker_pid) + md["scope_id"] = request.scope_id + md["expiration_time"] = str(request.expiration_time) + md["signal"] = str(request.signal) + md_str = json.dumps(md) + state = "RUNNING" + metadata: dict[str, EventMetadataValue] = { + "run_id": spec.rdzv_handler.get_run_id(), + "global_rank": None, + "group_rank": wg.group_rank, + "worker_id": None, + "role": spec.role, + "hostname": self._get_fq_hostname(), + "state": state, + "total_run_time": self._total_execution_time, + "rdzv_backend": spec.rdzv_handler.get_backend(), + "raw_error": None, + "metadata": md_str, + "agent_restarts": spec.max_restarts - self._remaining_restarts, + } + # Note: The 'metadata' field of the Event is converted to a TorchelasticStatusLogEntry later. + # The 'name' field of the Event is NOT used in the TorchelasticStatusLogEntry. + event = events.Event( + name=name, source=events.EventSource.AGENT, metadata=metadata + ) + events.record(event, self._worker_group.spec.event_log_handler) + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _stop_workers(self, worker_group: WorkerGroup) -> None: + self._shutdown() + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _start_workers(self, worker_group: WorkerGroup) -> dict[int, Any]: + spec = worker_group.spec + store = worker_group.store + if store is None: + raise AssertionError + restart_count = spec.max_restarts - self._remaining_restarts + + use_agent_store: bool = spec.rdzv_handler.use_agent_store + logger.info("use_agent_store: %s", use_agent_store) + + args: dict[int, tuple] = {} + envs: dict[int, dict[str, str]] = {} + log_line_prefixes: dict[int, str] | None = ( + {} if self._log_line_prefix_template else None + ) + # Short name (e.g. "r12i0n8") rather than _get_fq_hostname(): the fq name + # eats horizontal space in every log line, and degrades to an unhelpful + # reverse-DNS record (e.g. "...ip6.arpa") when the node has no PTR entry. + hostname = socket.gethostname() if self._log_line_prefix_template else "" + for worker in worker_group.workers: + local_rank = worker.local_rank + worker_env = { + "RANK": str(worker.global_rank), + "GROUP_RANK": str(worker_group.group_rank), + "ROLE_RANK": str(worker.role_rank), + "ROLE_NAME": spec.role, + "LOCAL_WORLD_SIZE": str(spec.local_world_size), + "WORLD_SIZE": str(worker.world_size), + "GROUP_WORLD_SIZE": str(worker_group.group_world_size), + "ROLE_WORLD_SIZE": str(worker.role_world_size), + "MASTER_ADDR": worker_group.master_addr, + "MASTER_PORT": str(worker_group.master_port), + "TORCHELASTIC_RESTART_COUNT": str(restart_count), + "TORCHELASTIC_MAX_RESTARTS": str(spec.max_restarts), + "TORCHELASTIC_RUN_ID": spec.rdzv_handler.get_run_id(), + "TORCHELASTIC_USE_AGENT_STORE": str(use_agent_store), + "TORCH_NCCL_ASYNC_ERROR_HANDLING": os.getenv( + "TORCH_NCCL_ASYNC_ERROR_HANDLING", str(1) + ), + } + self._set_local_rank_env(worker_env, local_rank, spec) + if "OMP_NUM_THREADS" in os.environ: + worker_env["OMP_NUM_THREADS"] = os.environ["OMP_NUM_THREADS"] + + if self._log_line_prefix_template: + log_line_prefix = Template( + self._log_line_prefix_template + ).safe_substitute( + role_name=spec.role, + rank=worker.global_rank, + local_rank=local_rank, + hostname=hostname, + ) + # pyrefly: ignore [unsupported-operation] + log_line_prefixes[local_rank] = log_line_prefix + + # pyrefly: ignore [unsupported-operation] + envs[local_rank] = worker_env + worker_args = list(spec.args) + worker_args = macros.substitute(worker_args, str(local_rank)) + args[local_rank] = tuple(worker_args) + + self._setup_local_watchdog(envs=envs) + self._setup_healthcheck() + + if spec.entrypoint is None: + raise AssertionError + if self._logs_specs is None: + raise AssertionError + self._pcontext = start_processes( + name=spec.role, + entrypoint=spec.entrypoint, + args=args, + envs=envs, + logs_specs=self._logs_specs, + log_line_prefixes=log_line_prefixes, + start_method=self._start_method, + numa_options=spec.numa_options, + duplicate_stdout_filters=spec.duplicate_stdout_filters, + duplicate_stderr_filters=spec.duplicate_stderr_filters, + ) + + return self._pcontext.pids() + + def _set_local_rank_env( + self, worker_env: dict[str, str | None], local_rank: int, spec: WorkerSpec + ) -> None: + # Set CUDA_VISIBLE_DEVICES and LOCAL_RANK based on virtual_local_rank mode. + # Virtual mode: Each worker sees only its assigned GPU as device 0, LOCAL_RANK=0 + # Traditional mode: Workers see all GPUs, LOCAL_RANK matches actual local rank + + if spec.virtual_local_rank: + # Set LOCAL_RANK=0 and use CUDA_VISIBLE_DEVICES to control the actual GPU access. + + worker_env["LOCAL_RANK"] = "0" + + # Map local_rank through existing CUDA_VISIBLE_DEVICES + # HIP uses CUDA_VISIBLE_DEVICES as a compatibility hack: + # https://rocm.docs.amd.com/en/latest/conceptual/gpu-isolation.html#cuda-visible-devices + parent_visible_devices = os.getenv("CUDA_VISIBLE_DEVICES") + if parent_visible_devices is not None: + # Parse comma-separated list of GPU IDs + available_gpus = parent_visible_devices.split(",") + if local_rank >= len(available_gpus): + raise ValueError( + f"local_rank {local_rank} exceeds available GPUs in " + f"CUDA_VISIBLE_DEVICES={parent_visible_devices}" + ) + + visible_gpu = available_gpus[local_rank].strip() + else: + # No restriction, use local_rank directly + visible_gpu = str(local_rank) + + worker_env["CUDA_VISIBLE_DEVICES"] = visible_gpu + return + + # In traditional mode, don't override CUDA_VISIBLE_DEVICES + # (inherit from parent environment) + worker_env["LOCAL_RANK"] = str(local_rank) + + if "CUDA_VISIBLE_DEVICES" in os.environ: + worker_env["CUDA_VISIBLE_DEVICES"] = os.environ["CUDA_VISIBLE_DEVICES"] + + def _shutdown( + self, death_sig: signal.Signals = signal.SIGTERM, timeout: int = 30 + ) -> None: + if self._worker_watchdog is not None: + self._worker_watchdog.stop() + self._worker_watchdog = None + if self._health_check_server is not None: + self._health_check_server.stop() + self._health_check_server = None + if self._pcontext: + self._pcontext.close(death_sig, timeout) + + def _check_uninterruptible_state_timeout( + self, worker_group: WorkerGroup, timeout: float + ) -> RunResult | None: + """Return UNHEALTHY when any worker has been in Linux uninterruptible + sleep (D-state) for at least ``timeout`` seconds; otherwise update + bookkeeping and return None. + """ + if self._pcontext is None: + return None + role = worker_group.spec.role + live_pids = set(self._pcontext.pids().values()) + # Drop bookkeeping for pids that have exited since the last check. + for pid in list(self._uninterruptible_state_first_seen): + if pid not in live_pids: + self._uninterruptible_state_first_seen.pop(pid, None) + + timed_out: list[tuple[int, float]] = [] + for pid in live_pids: + elapsed = self._update_uninterruptible_dwell(pid, role, timeout) + if elapsed is not None and elapsed >= timeout: + timed_out.append((pid, elapsed)) + + if not timed_out: + return None + for pid, elapsed in timed_out: + logger.error( + "[%s] Worker pid=%s stuck in uninterruptible sleep (D-state)" + " for %.1fs (>= %.1fs); marking worker group UNHEALTHY and" + " disabling restarts.", + role, + pid, + elapsed, + timeout, + ) + # Disable restarts: a wedged worker is holding a kernel resource + # (GPU, NIC) that a fresh worker on the same host would conflict + # with. Force the supervising loop into the _stop_workers branch + # so the agent can exit promptly. + self._remaining_restarts = 0 + return RunResult(state=WorkerState.UNHEALTHY) + + def _update_uninterruptible_dwell( + self, pid: int, role: str, timeout: float + ) -> float | None: + """Update bookkeeping for ``pid`` and return how long (seconds) it has + been continuously in uninterruptible sleep, or ``None`` if it isn't + (or its state can't be read). + """ + state = read_proc_state(pid) + if state is None: + return None + if not is_uninterruptible_state(state): + self._uninterruptible_state_first_seen.pop(pid, None) + return None + first = self._uninterruptible_state_first_seen.get(pid) + if first is None: + self._uninterruptible_state_first_seen[pid] = time.monotonic() + logger.warning( + "[%s] Worker pid=%s entered uninterruptible sleep (D-state);" + " will mark UNHEALTHY if it remains for %.1fs.", + role, + pid, + timeout, + ) + return 0.0 + return time.monotonic() - first + + # pyre-fixme[56]: Pyre was not able to infer the type of the decorator + # `torch.distributed.elastic.metrics.prof`. + @prof + def _monitor_workers(self, worker_group: WorkerGroup) -> RunResult: + role = worker_group.spec.role + worker_pids = {w.id for w in worker_group.workers} + if self._pcontext is None: + raise AssertionError + pc_pids = set(self._pcontext.pids().values()) + if worker_pids != pc_pids: + logger.error( + "[%s] worker pids do not match process_context pids." + " Expected: %s, actual: %s", + role, + worker_pids, + pc_pids, + ) + return RunResult(state=WorkerState.UNKNOWN) + + result = self._pcontext.wait(0) + if result: + if result.is_failed(): + # map local rank failure to global rank + worker_failures = {} + for local_rank, failure in result.failures.items(): + worker = worker_group.workers[local_rank] + worker_failures[worker.global_rank] = failure + return RunResult( + state=WorkerState.FAILED, + failures=worker_failures, + ) + else: + # copy ret_val_queue into a map with global ranks + workers_ret_vals = {} + for local_rank, ret_val in result.return_values.items(): + worker = worker_group.workers[local_rank] + workers_ret_vals[worker.global_rank] = ret_val + return RunResult( + state=WorkerState.SUCCEEDED, + return_values=workers_ret_vals, + ) + else: + timeout = self._uninterruptible_state_timeout + if timeout > 0: + ustate_result = self._check_uninterruptible_state_timeout( + worker_group, timeout + ) + if ustate_result is not None: + return ustate_result + return RunResult(state=WorkerState.HEALTHY) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/control_plane.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/control_plane.py new file mode 100644 index 0000000000000000000000000000000000000000..18cc7fea2cd89a9cc720f2bbaa15631be7273f28 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/control_plane.py @@ -0,0 +1,53 @@ +import os +from collections.abc import Generator +from contextlib import contextmanager, ExitStack + +from torch.distributed.elastic.multiprocessing.errors import record + + +__all__ = [ + "worker_main", +] + +TORCH_WORKER_SERVER_SOCKET = "TORCH_WORKER_SERVER_SOCKET" + + +@contextmanager +def _worker_server(socket_path: str) -> Generator[None, None, None]: + from torch._C._distributed_c10d import _WorkerServer + + server = _WorkerServer(socket_path) + try: + yield + finally: + server.shutdown() + + +@record +@contextmanager +def worker_main() -> Generator[None, None, None]: + """ + This is a context manager that wraps your main entry function. This combines + the existing ``errors.record`` logic as well as a new ``_WorkerServer`` that + exposes handlers via a unix socket specified by + ``TORCH_WORKER_SERVER_SOCKET``. + + Example + + :: + + @worker_main() + def main(): + pass + + + if __name__ == "__main__": + main() + + """ + with ExitStack() as stack: + socket_path = os.environ.get(TORCH_WORKER_SERVER_SOCKET) + if socket_path is not None: + stack.enter_context(_worker_server(socket_path)) + + yield diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..0b8437b2efec50c68e68a705a43b4d65a7c8e65a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__init__.py @@ -0,0 +1,167 @@ +#!/usr/bin/env/python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Module contains events processing mechanisms that are integrated with the standard python logging. + +Example of usage: + +:: + + from torch.distributed.elastic import events + + event = events.Event( + name="test_event", source=events.EventSource.WORKER, metadata={...} + ) + events.get_logging_handler(destination="console").info(event) + +""" + +import inspect +import logging +import os +import socket +import traceback +from typing import Optional + +from torch.distributed.elastic.events.handlers import get_logging_handler + +from .api import Event, EventMetadataValue, EventSource, NodeState, RdzvEvent + + +_events_loggers: dict[str, logging.Logger] = {} + + +def _get_or_create_logger(destination: str = "null") -> logging.Logger: + """ + Construct python logger based on the destination type or extends if provided. + + Available destination could be found in ``handlers.py`` file. + The constructed logger does not propagate messages to the upper level loggers, + e.g. root logger. This makes sure that a single event can be processed once. + + Args: + destination: The string representation of the event handler. + Available handlers found in ``handlers`` module + """ + global _events_loggers + + if destination not in _events_loggers: + _events_logger = logging.getLogger(f"torchelastic-events-{destination}") + _events_logger.setLevel(os.environ.get("LOGLEVEL", "INFO")) + # Do not propagate message to the root logger + _events_logger.propagate = False + + logging_handler = get_logging_handler(destination) + _events_logger.addHandler(logging_handler) + + # Add the logger to the global dictionary + _events_loggers[destination] = _events_logger + + return _events_loggers[destination] + + +def record(event: Event, destination: str = "null") -> None: + _get_or_create_logger(destination).info(event.serialize()) + + +def record_rdzv_event(event: RdzvEvent) -> None: + _get_or_create_logger("dynamic_rendezvous").info(event.serialize()) + + +def construct_and_record_rdzv_event( + run_id: str, + message: str, + node_state: NodeState, + name: str = "", + hostname: str = "", + pid: int | None = None, + master_endpoint: str = "", + local_id: int | None = None, + rank: int | None = None, +) -> None: + """ + Initialize rendezvous event object and record its operations. + + Args: + run_id (str): The run id of the rendezvous. + message (str): The message describing the event. + node_state (NodeState): The state of the node (INIT, RUNNING, SUCCEEDED, FAILED). + name (str): Event name. (E.g. Current action being performed). + hostname (str): Hostname of the node. + pid (Optional[int]): The process id of the node. + master_endpoint (str): The master endpoint for the rendezvous store, if known. + local_id (Optional[int]): The local_id of the node, if defined in dynamic_rendezvous.py + rank (Optional[int]): The rank of the node, if known. + Returns: + None + Example: + >>> # See DynamicRendezvousHandler class + >>> def _record( + ... self, + ... message: str, + ... node_state: NodeState = NodeState.RUNNING, + ... rank: Optional[int] = None, + ... ) -> None: + ... construct_and_record_rdzv_event( + ... name=f"{self.__class__.__name__}.{get_method_name()}", + ... run_id=self._settings.run_id, + ... message=message, + ... node_state=node_state, + ... hostname=self._this_node.addr, + ... pid=self._this_node.pid, + ... local_id=self._this_node.local_id, + ... rank=rank, + ... ) + """ + # We don't want to perform an extra computation if not needed. + if isinstance(get_logging_handler("dynamic_rendezvous"), logging.NullHandler): + return + + # Set up parameters. + if not hostname: + hostname = socket.getfqdn() + if not pid: + pid = os.getpid() + + # Determines which file called this function. + callstack = inspect.stack() + filename = "no_file" + if len(callstack) > 1: + stack_depth_1 = callstack[1] + filename = os.path.basename(stack_depth_1.filename) + if not name: + name = stack_depth_1.function + + # Delete the callstack variable. If kept, this can mess with python's + # garbage collector as we are holding on to stack frame information in + # the inspect module. + del callstack + + # Set up error trace if this is an exception + if node_state == NodeState.FAILED: + error_trace = traceback.format_exc() + else: + error_trace = "" + + # Initialize event object + event = RdzvEvent( + name=f"{filename}:{name}", + run_id=run_id, + message=message, + hostname=hostname, + pid=pid, + node_state=node_state, + master_endpoint=master_endpoint, + rank=rank, + local_id=local_id, + error_trace=error_trace, + ) + + # Finally, record the event. + record_rdzv_event(event) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5e52e698e1e869dad0334576b710104c1a005266 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..39b7ffb6c324523316d4739c66ec6b197b0042fe Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..02c0ff9c56fc59fc43b68a38c32fd609c6f38098 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/__pycache__/handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/api.py new file mode 100644 index 0000000000000000000000000000000000000000..9638f98b402c41d391ac0190325d9bcfb695d0c1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/api.py @@ -0,0 +1,116 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import json +from dataclasses import asdict, dataclass, field +from enum import Enum +from typing import Union + + +__all__ = ["EventSource", "Event", "NodeState", "RdzvEvent"] + +EventMetadataValue = str | int | float | bool | None + + +class EventSource(str, Enum): + """Known identifiers of the event producers.""" + + AGENT = "AGENT" + WORKER = "WORKER" + + +@dataclass +class Event: + """ + The class represents the generic event that occurs during the torchelastic job execution. + + The event can be any kind of meaningful action. + + Args: + name: event name. + source: the event producer, e.g. agent or worker + timestamp: timestamp in milliseconds when event occurred. + metadata: additional data that is associated with the event. + """ + + name: str + source: EventSource + timestamp: int = 0 + metadata: dict[str, EventMetadataValue] = field(default_factory=dict) + + def __str__(self): + return self.serialize() + + @staticmethod + def deserialize(data: Union[str, "Event"]) -> "Event": + if isinstance(data, Event): + return data + if isinstance(data, str): + data_dict = json.loads(data) + data_dict["source"] = EventSource[data_dict["source"]] # type: ignore[possibly-undefined] + # pyrefly: ignore [unbound-name] + return Event(**data_dict) + + def serialize(self) -> str: + return json.dumps(asdict(self)) + + +class NodeState(str, Enum): + """The states that a node can be in rendezvous.""" + + INIT = "INIT" + RUNNING = "RUNNING" + SUCCEEDED = "SUCCEEDED" + FAILED = "FAILED" + + +@dataclass +class RdzvEvent: + """ + Dataclass to represent any rendezvous event. + + Args: + name: Event name. (E.g. Current action being performed) + run_id: The run id of the rendezvous + message: The message describing the event + hostname: Hostname of the node + pid: The process id of the node + node_state: The state of the node (INIT, RUNNING, SUCCEEDED, FAILED) + master_endpoint: The master endpoint for the rendezvous store, if known + rank: The rank of the node, if known + local_id: The local_id of the node, if defined in dynamic_rendezvous.py + error_trace: Error stack trace, if this is an error event. + """ + + name: str + run_id: str + message: str + hostname: str + pid: int + node_state: NodeState + master_endpoint: str = "" + rank: int | None = None + local_id: int | None = None + error_trace: str = "" + + def __str__(self): + return self.serialize() + + @staticmethod + def deserialize(data: Union[str, "RdzvEvent"]) -> "RdzvEvent": + if isinstance(data, RdzvEvent): + return data + if isinstance(data, str): + data_dict = json.loads(data) + data_dict["node_state"] = NodeState[data_dict["node_state"]] # type: ignore[possibly-undefined] + # pyrefly: ignore [unbound-name] + return RdzvEvent(**data_dict) + + def serialize(self) -> str: + return json.dumps(asdict(self)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..30d925353253d5bab4c4780f298e7fa68a4409e5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/events/handlers.py @@ -0,0 +1,21 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging + + +_log_handlers: dict[str, logging.Handler] = { + "console": logging.StreamHandler(), + "dynamic_rendezvous": logging.NullHandler(), + "null": logging.NullHandler(), +} + + +def get_logging_handler(destination: str = "null") -> logging.Handler: + global _log_handlers + return _log_handlers[destination] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..75edb816e2a1feb6494f2da98e852c326d121e31 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__init__.py @@ -0,0 +1,168 @@ +#!/usr/bin/env/python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Metrics API. + +**Overview**: + +The metrics API in torchelastic is used to publish telemetry metrics. +It is designed to be used by torchelastic's internal modules to +publish metrics for the end user with the goal of increasing visibility +and helping with debugging. However you may use the same API in your +jobs to publish metrics to the same metrics ``sink``. + +A ``metric`` can be thought of as timeseries data +and is uniquely identified by the string-valued tuple +``(metric_group, metric_name)``. + +torchelastic makes no assumptions about what a ``metric_group`` is +and what relationship it has with ``metric_name``. It is totally up +to the user to use these two fields to uniquely identify a metric. + +.. note:: The metric group ``torchelastic`` is reserved by torchelastic for + platform level metrics that it produces. + For instance torchelastic may output the latency (in milliseconds) + of a re-rendezvous operation from the agent as + ``(torchelastic, agent.rendezvous.duration.ms)`` + +A sensible way to use metric groups is to map them to a stage or module +in your job. You may also encode certain high level properties +the job such as the region or stage (dev vs prod). + +**Publish Metrics**: + +Using torchelastic's metrics API is similar to using python's logging +framework. You first have to configure a metrics handler before +trying to add metric data. + +The example below measures the latency for the ``calculate()`` function. + +:: + + import time + import torch.distributed.elastic.metrics as metrics + + # makes all metrics other than the one from "my_module" to go /dev/null + metrics.configure(metrics.NullMetricsHandler()) + metrics.configure(metrics.ConsoleMetricsHandler(), "my_module") + + + def my_method(): + start = time.time() + calculate() + end = time.time() + metrics.put_metric("calculate_latency", int(end - start), "my_module") + +You may also use the torch.distributed.elastic.metrics.prof` decorator +to conveniently and succinctly profile functions + +:: + + # -- in module examples.foobar -- + + import torch.distributed.elastic.metrics as metrics + + metrics.configure(metrics.ConsoleMetricsHandler(), "foobar") + metrics.configure(metrics.ConsoleMetricsHandler(), "Bar") + + + @metrics.prof + def foo(): + pass + + + class Bar: + @metrics.prof + def baz(): + pass + +``@metrics.prof`` will publish the following metrics +:: + + .success - 1 if the function finished successfully + .failure - 1 if the function threw an exception + .duration.ms - function duration in milliseconds + +**Configuring Metrics Handler**: + +`torch.distributed.elastic.metrics.MetricHandler` is responsible for emitting +the added metric values to a particular destination. Metric groups can be +configured with different metric handlers. + +By default torchelastic emits all metrics to ``/dev/null``. +By adding the following configuration metrics, +``torchelastic`` and ``my_app`` metric groups will be printed out to +console. + +:: + + import torch.distributed.elastic.metrics as metrics + + metrics.configure(metrics.ConsoleMetricHandler(), group="torchelastic") + metrics.configure(metrics.ConsoleMetricHandler(), group="my_app") + +**Writing a Custom Metric Handler**: + +If you want your metrics to be emitted to a custom location, implement +the `torch.distributed.elastic.metrics.MetricHandler` interface +and configure your job to use your custom metric handler. + +Below is a toy example that prints the metrics to ``stdout`` + +:: + + import torch.distributed.elastic.metrics as metrics + + + class StdoutMetricHandler(metrics.MetricHandler): + def emit(self, metric_data): + ts = metric_data.timestamp + group = metric_data.group_name + name = metric_data.name + value = metric_data.value + print(f"[{ts}][{group}]: {name}={value}") + + + metrics.configure(StdoutMetricHandler(), group="my_app") + +Now all metrics in the group ``my_app`` will be printed to stdout as: + +:: + + [1574213883.4182858][my_app]: my_metric= + [1574213940.5237644][my_app]: my_metric= + +""" + +from typing import Optional + +from .api import ( + configure, + ConsoleMetricHandler, + get_elapsed_time_ms, + getStream, + MetricData, + MetricHandler, + MetricsConfig, + NullMetricHandler, + prof, + profile, + publish_metric, + put_metric, +) + + +def initialize_metrics(cfg: MetricsConfig | None = None): + pass + + +try: + from torch.distributed.elastic.metrics.static_init import * # type: ignore[import] # noqa: F403 +except ModuleNotFoundError: + pass diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..63d3d98b0dc42be8942b5d5067a2fce5dabc02e2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c201010bcbd3fcccda3044e1c9883d991ce6c344 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/api.py new file mode 100644 index 0000000000000000000000000000000000000000..102049481538d15a7fe995a8602ba45d6842303e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/metrics/api.py @@ -0,0 +1,216 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import abc +import time +from collections import namedtuple +from functools import wraps +from typing_extensions import deprecated + + +__all__ = [ + "MetricsConfig", + "MetricHandler", + "ConsoleMetricHandler", + "NullMetricHandler", + "MetricStream", + "configure", + "getStream", + "prof", + "profile", + "put_metric", + "publish_metric", + "get_elapsed_time_ms", + "MetricData", +] + +MetricData = namedtuple("MetricData", ["timestamp", "group_name", "name", "value"]) + + +class MetricsConfig: + __slots__ = ["params"] + + def __init__(self, params: dict[str, str] | None = None): + self.params = params + if self.params is None: + self.params = {} + + +class MetricHandler(abc.ABC): + @abc.abstractmethod + def emit(self, metric_data: MetricData): + pass + + +class ConsoleMetricHandler(MetricHandler): + def emit(self, metric_data: MetricData): + print( + f"[{metric_data.timestamp}][{metric_data.group_name}]: {metric_data.name}={metric_data.value}" + ) + + +class NullMetricHandler(MetricHandler): + def emit(self, metric_data: MetricData): + pass + + +class MetricStream: + def __init__(self, group_name: str, handler: MetricHandler): + self.group_name = group_name + self.handler = handler + + def add_value(self, metric_name: str, metric_value: int): + self.handler.emit( + MetricData(time.time(), self.group_name, metric_name, metric_value) + ) + + +_metrics_map: dict[str, MetricHandler] = {} +_default_metrics_handler: MetricHandler = NullMetricHandler() + + +# pyre-fixme[9]: group has type `str`; used as `None`. +def configure(handler: MetricHandler, group: str | None = None): + if group is None: + global _default_metrics_handler + # pyre-fixme[9]: _default_metrics_handler has type `NullMetricHandler`; used + # as `MetricHandler`. + _default_metrics_handler = handler + else: + _metrics_map[group] = handler + + +def getStream(group: str): + handler = _metrics_map.get(group, _default_metrics_handler) + return MetricStream(group, handler) + + +def _get_metric_name(fn): + qualname = fn.__qualname__ + split = qualname.split(".") + if len(split) == 1: + module = fn.__module__ + if module: + return module.split(".")[-1] + "." + split[0] + else: + return split[0] + else: + return qualname + + +def prof(fn=None, group: str = "torchelastic"): + r""" + @profile decorator publishes duration.ms, count, success, failure metrics for the function that it decorates. + + The metric name defaults to the qualified name (``class_name.def_name``) of the function. + If the function does not belong to a class, it uses the leaf module name instead. + + Usage + + :: + + @metrics.prof + def x(): + pass + + + @metrics.prof(group="agent") + def y(): + pass + """ + + def wrap(f): + @wraps(f) + def wrapper(*args, **kwargs): + key = _get_metric_name(f) + try: + start = time.time() + result = f(*args, **kwargs) + put_metric(f"{key}.success", 1, group) + except Exception: + put_metric(f"{key}.failure", 1, group) + raise + finally: + put_metric(f"{key}.duration.ms", get_elapsed_time_ms(start), group) # type: ignore[possibly-undefined] + return result + + return wrapper + + if fn: + return wrap(fn) + else: + return wrap + + +@deprecated("Deprecated, use `@prof` instead", category=FutureWarning) +def profile(group=None): + """ + @profile decorator adds latency and success/failure metrics to any given function. + + Usage + + :: + + @metrics.profile("my_metric_group") + def some_function(): + """ + + def wrap(func): + @wraps(func) + def wrapper(*args, **kwargs): + try: + start_time = time.time() + result = func(*args, **kwargs) + # pyrefly: ignore [bad-argument-type] + publish_metric(group, f"{func.__name__}.success", 1) + except Exception: + # pyrefly: ignore [bad-argument-type] + publish_metric(group, f"{func.__name__}.failure", 1) + raise + finally: + publish_metric( + # pyrefly: ignore [bad-argument-type] + group, + f"{func.__name__}.duration.ms", + get_elapsed_time_ms(start_time), # type: ignore[possibly-undefined] + ) + return result + + return wrapper + + return wrap + + +def put_metric(metric_name: str, metric_value: int, metric_group: str = "torchelastic"): + """ + Publish a metric data point. + + Usage + + :: + + put_metric("metric_name", 1) + put_metric("metric_name", 1, "metric_group_name") + """ + getStream(metric_group).add_value(metric_name, metric_value) + + +@deprecated( + "Deprecated, use `put_metric(metric_group)(metric_name, metric_value)` instead", + category=FutureWarning, +) +def publish_metric(metric_group: str, metric_name: str, metric_value: int): + metric_stream = getStream(metric_group) + metric_stream.add_value(metric_name, metric_value) + + +def get_elapsed_time_ms(start_time_in_seconds: float): + """Return the elapsed time in millis from the given start time.""" + end_time = time.time() + return int((end_time - start_time_in_seconds) * 1000) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a387ba40a49554495d777db0d60caad6aa1f18b3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__init__.py @@ -0,0 +1,256 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Library that launches and manages ``n`` copies of worker subprocesses either specified by a function or a binary. + +For functions, it uses ``torch.multiprocessing`` (and therefore python +``multiprocessing``) to spawn/fork worker processes. For binaries it uses python +``subprocessing.Popen`` to create worker processes. + + +Usage 1: Launching two trainers as a function + +:: + + from torch.distributed.elastic.multiprocessing import Std, start_processes + + + def trainer(a, b, c): + pass # train + + + # runs two trainers + # LOCAL_RANK=0 trainer(1,2,3) + # LOCAL_RANK=1 trainer(4,5,6) + ctx = start_processes( + name="trainer", + entrypoint=trainer, + args={0: (1, 2, 3), 1: (4, 5, 6)}, + envs={0: {"LOCAL_RANK": 0}, 1: {"LOCAL_RANK": 1}}, + log_dir="/tmp/foobar", + redirects=Std.ALL, # write all worker stdout/stderr to a log file + tee={0: Std.ERR}, # tee only local rank 0's stderr to console + ) + + # waits for all copies of trainer to finish + ctx.wait() + +Usage 2: Launching 2 echo workers as a binary + +:: + + # same as invoking + # echo hello + # echo world > stdout.log + ctx = start_processes( + name="echo" + entrypoint="echo", + log_dir="/tmp/foobar", + args={0: "hello", 1: "world"}, + redirects={1: Std.OUT}, + ) + +Just like ``torch.multiprocessing``, the return value of the function +:func:`start_processes` is a process context (:class:`api.PContext`). If a function +was launched, a :class:`api.MultiprocessContext` is returned and if a binary +was launched a :class:`api.SubprocessContext` is returned. Both are specific +implementations of the parent :class:`api.PContext` class. +""" + +from collections.abc import Callable +from typing import Optional, Union + +from torch.distributed.elastic.multiprocessing.api import ( + _validate_full_rank, + DefaultLogsSpecs, + LogsDest, + LogsSpecs, + MultiprocessContext, + PContext, + ProcessFailure, + RunProcsResult, + SignalException, + Std, + SubprocessContext, + to_map, +) +from torch.distributed.elastic.utils.logging import get_logger +from torch.numa.binding import NumaOptions + + +__all__ = [ + "start_processes", + "MultiprocessContext", + "PContext", + "ProcessFailure", + "RunProcsResult", + "SignalException", + "Std", + "LogsDest", + "LogsSpecs", + "DefaultLogsSpecs", + "SubprocessContext", + "to_map", +] + + +def start_processes( + name: str, + entrypoint: Callable | str, + args: dict[int, tuple], + envs: dict[int, dict[str, str]], + logs_specs: LogsSpecs, + log_line_prefixes: dict[int, str] | None = None, + start_method: str = "spawn", + numa_options: NumaOptions | None = None, + duplicate_stdout_filters: list[str] | None = None, + duplicate_stderr_filters: list[str] | None = None, +) -> PContext: + """ + Start ``n`` copies of ``entrypoint`` processes with the provided options. + + ``entrypoint`` is either a ``Callable`` (function) or a ``str`` (binary). + The number of copies is determined by the number of entries for ``args`` and + ``envs`` arguments, which need to have the same key set. + + ``args`` and ``env`` parameters are the arguments and environment variables + to pass down to the entrypoint mapped by the replica index (local rank). + All local ranks must be accounted for. + That is, the keyset should be ``{0,1,...,(nprocs-1)}``. + + .. note:: When the ``entrypoint`` is a binary (``str``), ``args`` can only be strings. + If any other type is given, then it is casted to a string representation + (e.g. ``str(arg1)``). Furthermore, a binary failure will only write + an ``error.json`` error file if the main function is annotated with + ``torch.distributed.elastic.multiprocessing.errors.record``. For function launches, + this is done by default and there is no need to manually annotate + with the ``@record`` annotation. + + Inside ``logs_specs``, ``redirects`` and ``tee`` are bitmasks specifying which std + stream(s) to redirect to a log file in the ``log_dir``. Valid mask values are defined + in ``Std``. To redirect/tee only certain local ranks, pass ``redirects`` as a map + with the key as the local rank to specify the redirect behavior for. + Any missing local ranks will default to ``Std.NONE``. + + ``duplicate_stdout_filters`` and ``duplicate_stderr_filters``, if non-empty, + duplicate stdouts and stderrs respectively specified in ``logs_specs``'s ``tee`` + to a file containing only lines that match _any_ of the filter strings. The log + file is aggregated across all ranks selected by ``tee``. + + ``tee`` acts like the unix "tee" command in that it redirects + prints to console. + To avoid worker stdout/stderr from printing to console, use the ``redirects`` parameter. + + For each process, the ``log_dir`` will contain: + + #. ``{local_rank}/error.json``: if the process failed, a file with the error info + #. ``{local_rank}/stdout.log``: if ``redirect & STDOUT == STDOUT`` + #. ``{local_rank}/stderr.log``: if ``redirect & STDERR == STDERR`` + #. ``filtered_stdout.log``: if ``duplicate_stdout_filters`` is non-empty + #. ``filtered_stderr.log``: if ``duplicate_stderr_filters`` is non-empty + + .. note:: It is expected that the ``log_dir`` exists, is empty, and is a directory. + + Example: + :: + + log_dir = "/tmp/test" + + # ok; two copies of foo: foo("bar0"), foo("bar1") + start_processes( + name="trainer", + entrypoint=foo, + args:{0:("bar0",), 1:("bar1",), + envs:{0:{}, 1:{}}, + log_dir=log_dir + ) + + # invalid; envs missing for local rank 1 + start_processes( + name="trainer", + entrypoint=foo, + args:{0:("bar0",), 1:("bar1",), + envs:{0:{}}, + log_dir=log_dir + ) + + # ok; two copies of /usr/bin/touch: touch file1, touch file2 + start_processes( + name="trainer", + entrypoint="/usr/bin/touch", + args:{0:("file1",), 1:("file2",), + envs:{0:{}, 1:{}}, + log_dir=log_dir + ) + + # caution; arguments casted to string, runs: + # echo "1" "2" "3" and echo "[1, 2, 3]" + start_processes( + name="trainer", + entrypoint="/usr/bin/echo", + args:{0:(1,2,3), 1:([1,2,3],), + envs:{0:{}, 1:{}}, + log_dir=log_dir + ) + + Args: + name: a human readable short name that describes what the processes are + (used as header when tee'ing stdout/stderr outputs) + entrypoint: either a ``Callable`` (function) or ``cmd`` (binary) + args: arguments to each replica + envs: env vars to each replica + log_dir: directory used to write log files + start_method: multiprocessing start method (spawn, fork, forkserver) + ignored for binaries + logs_specs: defines ``log_dir``, ``redirects``, and ``tee``. + inside ``logs_specs``: + - redirects: which std streams to redirect to a log file + - tee: which std streams to redirect + print to console + local_ranks_filter: which ranks' logs to print to console + duplicate_stdout_filters: filters for the duplicated stdout logs + duplicate_stderr_filters: filters for the duplicated stderr logs + + """ + + nprocs = len(args) + _validate_full_rank(args, nprocs, "args") + _validate_full_rank(envs, nprocs, "envs") + + context: PContext + if isinstance(entrypoint, str): + context = SubprocessContext( + name=name, + entrypoint=entrypoint, + args=args, + envs=envs, + duplicate_stdout_filters=duplicate_stdout_filters, + duplicate_stderr_filters=duplicate_stderr_filters, + logs_specs=logs_specs, + log_line_prefixes=log_line_prefixes, + numa_options=numa_options, + ) + else: + context = MultiprocessContext( + name=name, + entrypoint=entrypoint, + args=args, + envs=envs, + duplicate_stdout_filters=duplicate_stdout_filters, + duplicate_stderr_filters=duplicate_stderr_filters, + log_line_prefixes=log_line_prefixes, + start_method=start_method, + logs_specs=logs_specs, + numa_options=numa_options, + ) + + try: + context.start() + return context + except Exception: + context.close() + raise diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..560115a46bf159df5317ff815085c6fe16cdfdc6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..41ff11448b29d5db59e6dc1a299e0b805b069328 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/redirects.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/redirects.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fa018133e64d50fff73fd9d3c37beb64e57fac2a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/redirects.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/tail_log.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/tail_log.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3a9bf040a0f7c30192246c1e0847f0ace63c8e72 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/__pycache__/tail_log.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/api.py new file mode 100644 index 0000000000000000000000000000000000000000..76c9917d04c4b04234ed53dcac2c679322442d4e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/api.py @@ -0,0 +1,1070 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import abc +import logging +import os +import re +import shutil +import signal +import subprocess +import sys +import tempfile +import threading +import time +from abc import ABC, abstractmethod +from collections.abc import Callable +from contextlib import nullcontext +from dataclasses import dataclass, field +from enum import IntFlag +from multiprocessing import synchronize +from types import FrameType +from typing import Any, TextIO, Union + +import torch.multiprocessing as mp +from torch.distributed.elastic.multiprocessing.errors import ProcessFailure, record +from torch.distributed.elastic.multiprocessing.redirects import ( + redirect_stderr, + redirect_stdout, +) +from torch.distributed.elastic.multiprocessing.subprocess_handler import ( + get_subprocess_handler, + SubprocessHandler, +) +from torch.distributed.elastic.multiprocessing.tail_log import TailLog +from torch.numa.binding import _maybe_wrap_with_numa_binding, NumaOptions + + +IS_WINDOWS = sys.platform == "win32" +IS_MACOS = sys.platform == "darwin" + + +logger = logging.getLogger(__name__) + +__all__ = [ + "DefaultLogsSpecs", + "SignalException", + "Std", + "to_map", + "RunProcsResult", + "PContext", + "get_std_cm", + "MultiprocessContext", + "SubprocessContext", + "LogsDest", + "LogsSpecs", +] + + +class SignalException(Exception): + """ + Exception is raised inside the torchelastic agent process by the termination handler + if the death signal got received by the process. + """ + + def __init__(self, msg: str, sigval: signal.Signals) -> None: + super().__init__(msg) + self.sigval = sigval + + +def _terminate_process_handler(signum: int, frame: FrameType | None) -> None: + """Termination handler that raises exceptions on the main process. + + When the process receives death signal(SIGTERM, SIGINT), this termination handler will + be invoked. It raises the ``SignalException`` exception that should be processed by the + user code. Python does not terminate process after the termination handler is finished, + so the exception should not be silently ignored, otherwise the process will never + be terminated. + """ + sigval = signal.Signals(signum) + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) + + +def _get_kill_signal() -> signal.Signals: + """Get the kill signal. SIGKILL for unix, CTRL_C_EVENT for windows.""" + if IS_WINDOWS: + return signal.CTRL_C_EVENT # type: ignore[attr-defined] + else: + return signal.SIGKILL + + +def _get_default_signal() -> signal.Signals: + """Get the default termination signal. SIGTERM for unix, CTRL_C_EVENT for windows.""" + if IS_WINDOWS: + return signal.CTRL_C_EVENT # type: ignore[attr-defined] + else: + return signal.SIGTERM + + +def _validate_full_rank(d: dict[int, Any], nprocs: int, what: str): + actual_keys = set(d.keys()) + expected_keys = set(range(nprocs)) + + if actual_keys != expected_keys: + raise RuntimeError( + f"{what}, local rank mapping mismatch," + f" expected: {expected_keys}, actual: {actual_keys}" + ) + + +_MAPPING_REGEX = r"^(\d:[0123],)*(\d:[0123])$" +_VALUE_REGEX = r"^[0123]$" + + +class Std(IntFlag): + NONE = 0 + OUT = 1 + ERR = 2 + ALL = OUT | ERR + + @classmethod + def from_str(cls, vm: str) -> Union["Std", dict[int, "Std"]]: + """ + Example: + :: + + from_str("0") -> Std.NONE + from_str("1") -> Std.OUT + from_str("0:3,1:0,2:1,3:2") -> {0: Std.ALL, 1: Std.NONE, 2: Std.OUT, 3: Std.ERR} + + Any other input raises an exception + """ + + def to_std(v: str) -> Std: # type: ignore[return] + s = Std(int(v)) + if s in Std: + return s + # return None -> should NEVER reach here since we regex check input + + if re.match(_VALUE_REGEX, vm): # vm is a number (e.g. 0) + return to_std(vm) + elif re.match(_MAPPING_REGEX, vm): # vm is a mapping (e.g. 0:1,1:2) + d: dict[int, Std] = {} + for m in vm.split(","): + i, v = m.split(":") + d[int(i)] = to_std(v) + return d + else: + raise ValueError( + f"{vm} does not match: <{_VALUE_REGEX}> or <{_MAPPING_REGEX}>" + ) + + +def to_map(val_or_map: Std | dict[int, Std], local_world_size: int) -> dict[int, Std]: + """ + Certain APIs take redirect settings either as a single value (e.g. apply to all + local ranks) or as an explicit user-provided mapping. This method is a convenience + method that converts a value or mapping into a mapping. + + Example: + :: + + to_map(Std.OUT, local_world_size=2) # returns: {0: Std.OUT, 1: Std.OUT} + to_map({1: Std.OUT}, local_world_size=2) # returns: {0: Std.NONE, 1: Std.OUT} + to_map( + {0: Std.OUT, 1: Std.OUT}, local_world_size=2 + ) # returns: {0: Std.OUT, 1: Std.OUT} + """ + if isinstance(val_or_map, Std): + return dict.fromkeys(range(local_world_size), val_or_map) + else: + map = {} + for i in range(local_world_size): + map[i] = val_or_map.get(i, Std.NONE) + return map + + +@dataclass +class LogsDest: + """ + For each log type, holds mapping of local rank ids to file paths. + """ + + stdouts: dict[int, str] = field(default_factory=dict) + stderrs: dict[int, str] = field(default_factory=dict) + tee_stdouts: dict[int, str] = field(default_factory=dict) + tee_stderrs: dict[int, str] = field(default_factory=dict) + error_files: dict[int, str] = field(default_factory=dict) + filtered_stdout: str = field(default_factory=str) + filtered_stderr: str = field(default_factory=str) + + +class LogsSpecs(ABC): + """ + Defines logs processing and redirection for each worker process. + + Args: + log_dir: + Base directory where logs will be written. + redirects: + Streams to redirect to files. Pass a single ``Std`` + enum to redirect for all workers, or a mapping keyed + by local_rank to selectively redirect. + tee: + Streams to duplicate to stdout/stderr. + Pass a single ``Std`` enum to duplicate streams for all workers, + or a mapping keyed by local_rank to selectively duplicate. + """ + + def __init__( + self, + log_dir: str | None = None, + redirects: Std | dict[int, Std] = Std.NONE, + tee: Std | dict[int, Std] = Std.NONE, + local_ranks_filter: set[int] | None = None, + ) -> None: + self._root_log_dir = log_dir + self._redirects = redirects + self._tee = tee + self._local_ranks_filter = local_ranks_filter + + @abstractmethod + def reify( + self, + envs: dict[int, dict[str, str]], + ) -> LogsDest: + """ + Given the environment variables, builds destination of log files for each of the local ranks. + + Envs parameter contains env variables dict for each of the local ranks, where entries are defined in: + :func:`~torchelastic.distributed.elastic.agent.server.local_elastic_agent.LocalElasticAgent._start_workers`. + """ + + @property + @abstractmethod + def root_log_dir(self) -> str: + pass + + +class DefaultLogsSpecs(LogsSpecs): + """ + Default LogsSpecs implementation: + + - `log_dir` will be created if it doesn't exist + - Generates nested folders for each attempt and rank. + """ + + def __init__( + self, + log_dir: str | None = None, + redirects: Std | dict[int, Std] = Std.NONE, + tee: Std | dict[int, Std] = Std.NONE, + local_ranks_filter: set[int] | None = None, + ) -> None: + if log_dir != os.devnull: + if not log_dir: + log_dir = tempfile.mkdtemp(prefix="torchelastic_") + elif not os.path.exists(log_dir): + os.makedirs(log_dir, exist_ok=True) + else: + if os.path.isfile(log_dir): + raise NotADirectoryError(f"log_dir: {log_dir} is a file") + super().__init__(log_dir, redirects, tee, local_ranks_filter) + # initialized only once + self._run_log_dir = None + + @property + def root_log_dir(self) -> str: + return str(self._root_log_dir) + + def _make_log_dir(self, log_dir: str | None, rdzv_run_id: str): + base_log_dir = log_dir or tempfile.mkdtemp(prefix="torchelastic_") + os.makedirs(base_log_dir, exist_ok=True) + dir = tempfile.mkdtemp(prefix=f"{rdzv_run_id}_", dir=base_log_dir) + logger.info("log directory set to: %s", dir) + return dir + + def reify( + self, + envs: dict[int, dict[str, str]], + ) -> LogsDest: + """ + Uses following scheme to build log destination paths: + + - `//attempt_//stdout.log` + - `//attempt_//stderr.log` + - `//attempt_//error.json` + - `//attempt_/filtered_stdout.log` + - `//attempt_/filtered_stderr.log` + """ + nprocs = len(envs) + global_env = {} # use only to query properties that are not dependent on a rank + if nprocs > 0: + global_env = envs[0] + else: + logger.warning( + "Empty envs map provided when defining logging destinations." + ) + # Keys are always defined, but values can be missing in unit tests + run_id = global_env.get("TORCHELASTIC_RUN_ID", "test_run_id") + restart_count = global_env.get("TORCHELASTIC_RESTART_COUNT", "0") + + attempt_log_dir: str = "" + if self._root_log_dir != os.devnull: + if not self._run_log_dir: + self._run_log_dir = self._make_log_dir(self._root_log_dir, run_id) + + attempt_log_dir = os.path.join( + self._run_log_dir, f"attempt_{restart_count}" + ) # type: ignore[call-overload] + shutil.rmtree(attempt_log_dir, ignore_errors=True) + os.makedirs(attempt_log_dir) + + if self._root_log_dir == os.devnull: + attempt_log_dir = os.devnull + + # create subdirs for each local rank in the logs_dir + # logs_dir + # |- 0 + # |- error.json + # |- stdout.log + # |- stderr.log + # |- ... + # |- (nprocs-1) + redirs = to_map(self._redirects, nprocs) + ts = to_map(self._tee, nprocs) + + # to tee stdout/stderr we first redirect into a file + # then tail -f stdout.log/stderr.log so add tee settings to redirects + for local_rank, tee_std in ts.items(): + redirect_std = redirs[local_rank] + redirs[local_rank] = redirect_std | tee_std + + SYS_STREAM = "" # special case to indicate to output to console + stdouts = dict.fromkeys(range(nprocs), SYS_STREAM) + stderrs = dict.fromkeys(range(nprocs), SYS_STREAM) + tee_stdouts: dict[int, str] = {} + tee_stderrs: dict[int, str] = {} + error_files = {} + + for local_rank in range(nprocs): + if attempt_log_dir == os.devnull: + tee_stdouts[local_rank] = os.devnull + tee_stderrs[local_rank] = os.devnull + error_files[local_rank] = os.devnull + envs[local_rank]["TORCHELASTIC_ERROR_FILE"] = "" + else: + clogdir = os.path.join(attempt_log_dir, str(local_rank)) + os.mkdir(clogdir) + + rd = redirs[local_rank] + if (rd & Std.OUT) == Std.OUT: + stdouts[local_rank] = os.path.join(clogdir, "stdout.log") + if (rd & Std.ERR) == Std.ERR: + stderrs[local_rank] = os.path.join(clogdir, "stderr.log") + + t = ts[local_rank] + if t & Std.OUT == Std.OUT: + tee_stdouts[local_rank] = stdouts[local_rank] + if t & Std.ERR == Std.ERR: + tee_stderrs[local_rank] = stderrs[local_rank] + + if ( + self._local_ranks_filter + and local_rank not in self._local_ranks_filter + ): + # If stream is tee'd, only write to file, but don't tail + if local_rank in tee_stdouts: + tee_stdouts.pop(local_rank, None) + if local_rank in tee_stderrs: + tee_stderrs.pop(local_rank, None) + + # If stream is not redirected, don't print + if stdouts[local_rank] == SYS_STREAM: + stdouts[local_rank] = os.devnull + if stderrs[local_rank] == SYS_STREAM: + stderrs[local_rank] = os.devnull + + error_file = os.path.join(clogdir, "error.json") + # pyrefly: ignore [unsupported-operation] + error_files[local_rank] = error_file + logger.info( + "Setting worker%s reply file to: %s", local_rank, error_file + ) + envs[local_rank]["TORCHELASTIC_ERROR_FILE"] = error_file + + return LogsDest( + stdouts, + stderrs, + tee_stdouts, + tee_stderrs, + # pyrefly: ignore [bad-argument-type] + error_files, + os.path.join(attempt_log_dir, "filtered_stdout.log"), + os.path.join(attempt_log_dir, "filtered_stderr.log"), + ) + + def __repr__(self) -> str: + return ( + f"DefaultLogsSpecs(root_log_dir={self._root_log_dir}, redirects={self._redirects}, " + f"tee={self._tee}, local_ranks_filter={self._local_ranks_filter})" + ) + + def __eq__(self, other: object) -> bool: + if not isinstance(other, DefaultLogsSpecs): + return False + + return ( + self._root_log_dir == other._root_log_dir + and self._redirects == other._redirects + and self._tee == other._tee + and self._local_ranks_filter == other._local_ranks_filter + ) + + +@dataclass +class RunProcsResult: + """ + Results of a completed run of processes started with ``start_processes()``. Returned by ``PContext``. + + Note the following: + + 1. All fields are mapped by local rank + 2. ``return_values`` - only populated for functions (not the binaries). + 3. ``stdouts`` - path to stdout.log (empty string if no redirect) + 4. ``stderrs`` - path to stderr.log (empty string if no redirect) + + """ + + return_values: dict[int, Any] = field(default_factory=dict) + failures: dict[int, ProcessFailure] = field(default_factory=dict) + stdouts: dict[int, str] = field(default_factory=dict) + stderrs: dict[int, str] = field(default_factory=dict) + + def is_failed(self) -> bool: + return len(self.failures) > 0 + + +class PContext(abc.ABC): + """ + The base class that standardizes operations over a set of processes that are launched via different mechanisms. + + The name ``PContext`` is intentional to disambiguate with ``torch.multiprocessing.ProcessContext``. + + .. warning:: stdouts and stderrs should ALWAYS be a superset of + tee_stdouts and tee_stderrs (respectively) this is b/c + tee is implemented as a redirect + tail -f + + Args: + duplicate_stdout_filters: + If non-empty, duplicates stdouts specified in ``logs_specs``'s ``tee`` + to a file containing only lines that match _any_ of the filter strings. + The log file is aggregated across all ranks selected by ``tee``. + duplicate_stderr_filters: + If non-empty, duplicates stderrs specified in ``logs_specs``'s ``tee`` + to a file containing only lines that match _any_ of the filter strings. + The log file is aggregated across all ranks selected by ``tee``. + """ + + def __init__( + self, + name: str, + entrypoint: Callable | str, + args: dict[int, tuple], + envs: dict[int, dict[str, str]], + logs_specs: LogsSpecs, + log_line_prefixes: dict[int, str] | None = None, + duplicate_stdout_filters: list[str] | None = None, + duplicate_stderr_filters: list[str] | None = None, + ): + self.name = name + # validate that all mappings have the same number of keys and + # all local ranks are accounted for + nprocs = len(args) + + # TODO log_line_prefixes can be expanded too + logs_dest = logs_specs.reify(envs) + + _validate_full_rank(logs_dest.stdouts, nprocs, "stdouts") + _validate_full_rank(logs_dest.stderrs, nprocs, "stderrs") + + self.entrypoint = entrypoint + self.args = args + self.envs = envs + self.stdouts = logs_dest.stdouts + self.stderrs = logs_dest.stderrs + self.error_files = logs_dest.error_files + self.nprocs = nprocs + self.filtered_stdout: TextIO | None = None + self.filtered_stderr: TextIO | None = None + + self._tail_logs = [ + TailLog(name, logs_dest.tee_stdouts, sys.stdout, log_line_prefixes), + TailLog(name, logs_dest.tee_stderrs, sys.stderr, log_line_prefixes), + ] + + if duplicate_stdout_filters: + self.filtered_stdout = open( # noqa: SIM115 + logs_dest.filtered_stdout, mode="w", errors="replace", buffering=1 + ) + self._tail_logs.append( + TailLog( + name, + logs_dest.tee_stdouts, + self.filtered_stdout, + log_line_prefixes, + log_line_filter=lambda line: any( + needle in line for needle in duplicate_stdout_filters + ), + ) + ) + + if duplicate_stderr_filters: + self.filtered_stderr = open( # noqa: SIM115 + logs_dest.filtered_stderr, mode="w", errors="replace", buffering=1 + ) + self._tail_logs.append( + TailLog( + name, + logs_dest.tee_stderrs, + self.filtered_stderr, + log_line_prefixes, + log_line_filter=lambda line: any( + needle in line for needle in duplicate_stderr_filters + ), + ) + ) + + def start(self) -> None: + """Start processes using parameters defined in the constructor.""" + if threading.current_thread() is threading.main_thread(): + # Register signal handlers for the signals specified in the environment variable + signals_to_handle = os.environ.get( + "TORCHELASTIC_SIGNALS_TO_HANDLE", "SIGTERM,SIGINT,SIGHUP,SIGQUIT" + ) + signal_list = signals_to_handle.split(",") + + for sig_name in signal_list: + try: + sig = getattr(signal, sig_name.strip()) + signal.signal(sig, _terminate_process_handler) + logger.info("Registered signal handler for %s", sig_name) + except (AttributeError, ValueError): + logger.warning( + "Failed to register signal handler for %s", + sig_name, + exc_info=True, + ) + except RuntimeError: + if IS_WINDOWS and sig_name.strip() in [ + "SIGHUP", + "SIGQUIT", + "SIGUSR1", + "SIGUSR2", + ]: + logger.info( + "Signal %s is not supported on Windows, skipping", sig_name + ) + else: + logger.warning( + "Failed to register signal handler for %s", + sig_name, + exc_info=True, + ) + else: + logger.warning( + "Failed to register signal handlers since torchelastic is running on a child thread. " + "This could lead to orphaned worker processes if the torchrun is terminated." + ) + self._start() + for tail_log in self._tail_logs: + tail_log.start() + + @abc.abstractmethod + def _start(self) -> None: + """Start processes using strategy defined in a particular context.""" + raise NotImplementedError + + @abc.abstractmethod + def _poll(self) -> RunProcsResult | None: + """ + Poll the run status of the processes running under this context. + This method follows an "all-or-nothing" policy and returns + a ``RunProcessResults`` object if either all processes complete + successfully or any process fails. Returns ``None`` if + all processes are still running. + """ + raise NotImplementedError + + def wait(self, timeout: float = -1, period: float = 1) -> RunProcsResult | None: + """ + Wait for the specified ``timeout`` seconds, polling every ``period`` seconds + for the processes to be done. Returns ``None`` if the processes are still running + on timeout expiry. Negative timeout values are interpreted as "wait-forever". + A timeout value of zero simply queries the status of the processes (e.g. equivalent + to a poll). + + .. note:: + Multiprocessing library registers SIGTERM and SIGINT signal handlers that raise + ``SignalException`` when the signals received. It is up to the consumer of the code + to properly handle the exception. It is important not to swallow the exception otherwise + the process would not terminate. Example of the typical workflow can be: + + .. code-block:: python + pc = start_processes(...) + try: + pc.wait(1) + .. do some other work + except SignalException as e: + pc.shutdown(e.sigval, timeout=30) + + If SIGTERM or SIGINT occurs, the code above will try to shutdown child processes by propagating + received signal. If child processes will not terminate in the timeout time, the process will send + the SIGKILL. + """ + if timeout == 0: + return self._poll() + + if timeout < 0: + timeout = sys.maxsize + + expiry = time.time() + timeout + while time.time() < expiry: + pr = self._poll() + if pr: + return pr + time.sleep(period) + + return None + + @abc.abstractmethod + def pids(self) -> dict[int, int]: + """Return pids of processes mapped by their respective local_ranks.""" + raise NotImplementedError + + @abc.abstractmethod + def _close(self, death_sig: signal.Signals, timeout: int = 30) -> None: + r""" + Terminates all processes managed by this context and cleans up any + meta resources (e.g. redirect, error_file files). + """ + raise NotImplementedError + + def close(self, death_sig: signal.Signals | None = None, timeout: int = 30) -> None: + r""" + Terminates all processes managed by this context and cleans up any + meta resources (e.g. redirect, error_file files). + + Args: + death_sig: Death signal to terminate processes. + timeout: Time to wait for processes to finish, if process is + still alive after this time, it will be terminated via SIGKILL. + """ + if not death_sig: + death_sig = _get_default_signal() + self._close(death_sig=death_sig, timeout=timeout) + for tail_log in self._tail_logs: + tail_log.stop() + if self.filtered_stdout: + self.filtered_stdout.close() + if self.filtered_stderr: + self.filtered_stderr.close() + + +def get_std_cm(std_rd: str, redirect_fn): + if IS_WINDOWS or IS_MACOS or not std_rd: + return nullcontext() + else: + return redirect_fn(std_rd) + + +def _wrap( + local_rank: int, + fn: Callable, + args: dict[int, tuple], + envs: dict[int, dict[str, str]], + stdout_redirects: dict[int, str], # redirect file for stdout (to console if None) + stderr_redirects: dict[int, str], # redirect file for stderr (to console if None) + ret_vals: dict[int, mp.SimpleQueue], + queue_finished_reading_event: synchronize.Event, + numa_options: NumaOptions | None, +) -> None: + # get the per-rank params up front so we fail fast if no mapping is found + args_ = args[local_rank] + env_ = envs[local_rank] + ret_val_ = ret_vals[local_rank] + + stdout_rd = stdout_redirects[local_rank] + stderr_rd = stderr_redirects[local_rank] + + stdout_cm = get_std_cm(stdout_rd, redirect_stdout) + stderr_cm = get_std_cm(stderr_rd, redirect_stderr) + + for k, v in env_.items(): + os.environ[k] = v + + with stdout_cm, stderr_cm: + fn = _maybe_wrap_with_numa_binding( + fn, device_index=local_rank, numa_options=numa_options + ) + ret = record(fn)(*args_) + ret_val_.put(ret) + queue_finished_reading_event.wait() + + +class MultiprocessContext(PContext): + """``PContext`` holding worker processes invoked as a function.""" + + def __init__( + self, + name: str, + entrypoint: Callable, + args: dict[int, tuple], + envs: dict[int, dict[str, str]], + start_method: str, + logs_specs: LogsSpecs, + log_line_prefixes: dict[int, str] | None = None, + numa_options: NumaOptions | None = None, + duplicate_stdout_filters: list[str] | None = None, + duplicate_stderr_filters: list[str] | None = None, + ): + super().__init__( + name, + entrypoint, + args, + envs, + logs_specs, + log_line_prefixes, + duplicate_stdout_filters, + duplicate_stderr_filters, + ) + + self.start_method = start_method + # each ret_val queue will always contain a single element. + self._ret_vals = { + local_rank: mp.get_context(self.start_method).SimpleQueue() + for local_rank in range(self.nprocs) + } + + # see comments in ``join()`` for what this is + self._return_values: dict[int, Any] = {} + self._pc: mp.ProcessContext | None = None + # Note: set method should ONLY be invoked for the use case when all processes finished + # successfully. If any process died on event.wait() calling set() method will deadlock. + self._worker_finished_event = mp.get_context(self.start_method).Event() + + self._numa_options: NumaOptions | None = numa_options + + def _start(self): + if self._pc: + raise ValueError( + "The process context already initialized." + " Most likely the start method got called twice." + ) + self._pc = mp.start_processes( + fn=_wrap, + args=( + self.entrypoint, + self.args, + self.envs, + self.stdouts, + self.stderrs, + self._ret_vals, + self._worker_finished_event, + self._numa_options, + ), + nprocs=self.nprocs, + join=False, + daemon=False, + start_method=self.start_method, + ) + + def _is_done(self) -> bool: + return len(self._return_values) == self.nprocs + + def _poll(self) -> RunProcsResult | None: + if self._pc is None: + raise AssertionError # assertion for mypy type checker + + try: + # torch.mp.ProcessContext Throws an Exception if some/all of + # worker processes failed + # timeout < 0 checks worker status and return immediately + # Join will never return success since we use synchronize.Event to wait + # for all processes to finish. + self._pc.join(-1) + + # IMPORTANT: we use multiprocessing.Queue to carry worker return values + # back to the parent, the worker process will wait before terminating + # until all the buffered items are fed by the feeder thread to the underlying + # pipe. Hence to prevent deadlocks on large return values, + # we opportunistically try queue.get on each join call + # See: https://docs.python.org/2/library/multiprocessing.html#all-platforms + for local_rank in range(self.nprocs): + return_queue = self._ret_vals[local_rank] + if not return_queue.empty(): + # save the return values temporarily into a member var + self._return_values[local_rank] = return_queue.get() + + if self._is_done(): + # we should ALWAYS have ALL the return values when all the processes are done + self._worker_finished_event.set() + + # At this point workers finished running the user function + # But the child process might still have not exited. Wait for them. + # pc.join() blocks [forever] until "a" proc exits. Loop until all of them exits. + while not self._pc.join(): + logger.debug( + "entrypoint fn finished, waiting for all child procs to exit..." + ) + + _validate_full_rank( + self._return_values, self.nprocs, "return_value queue" + ) + self.close() + return RunProcsResult( + return_values=self._return_values, + stdouts=self.stdouts, + stderrs=self.stderrs, + ) + else: + return None + except (mp.ProcessRaisedException, mp.ProcessExitedException) as e: + failed_local_rank = e.error_index + + # entrypoint for MultiprocessContext will always be a Callable + fn_name = self.entrypoint.__qualname__ # type: ignore[union-attr] + failed_proc = self._pc.processes[failed_local_rank] + error_filepath = self.error_files[failed_local_rank] + + logger.exception( + "failed (exitcode: %s)" + " local_rank: %s (pid: %s)" + " of fn: %s (start_method: %s)", + failed_proc.exitcode, + failed_local_rank, + e.error_pid, + fn_name, + self.start_method, + ) + + self.close() + return RunProcsResult( + failures={ + failed_local_rank: ProcessFailure( + local_rank=failed_local_rank, + pid=e.error_pid, + exitcode=failed_proc.exitcode, + error_file=error_filepath, + ) + }, + stdouts=self.stdouts, + stderrs=self.stderrs, + ) + + def pids(self) -> dict[int, int]: + if self._pc is None: + raise AssertionError # assertion for mypy type checking + return dict(enumerate(self._pc.pids())) + + def _close(self, death_sig: signal.Signals, timeout: int = 30) -> None: + if not self._pc: + return + for proc in self._pc.processes: + if proc.is_alive(): + logger.warning( + "Closing process %s via signal %s", proc.pid, death_sig.name + ) + try: + os.kill(proc.pid, death_sig) + except ProcessLookupError: + # If the process exited because of some reason, + # `ProcessLookupError` will be raised, it is safe to ignore it. + pass + end = time.monotonic() + timeout + for proc in self._pc.processes: + time_to_wait = end - time.monotonic() + if time_to_wait <= 0: + break + proc.join(time_to_wait) + for proc in self._pc.processes: + if proc.is_alive(): + logger.warning( + "Unable to shutdown process %s via %s, forcefully exiting via %s", + proc.pid, + death_sig, + _get_kill_signal(), + ) + try: + os.kill(proc.pid, _get_kill_signal()) + except ProcessLookupError: + # If the process exited because of some reason, + # `ProcessLookupError` will be raised, it is safe to ignore it. + pass + # Bounded join: a process stuck in uninterruptible kernel sleep + # (Linux D-state, common with hung NCCL/GPU collectives) cannot + # be reaped by SIGKILL. Without a timeout the agent itself would + # wedge here and the supervising launcher could never exit. + proc.join(timeout) + if proc.is_alive(): + logger.error( + "Process %s did not exit after %s within %ss;" + " abandoning and continuing teardown." + " Worker may be in uninterruptible kernel sleep" + " (e.g. wedged on GPU/NIC); host may need to be recycled.", + proc.pid, + _get_kill_signal().name, + timeout, + ) + + +class SubprocessContext(PContext): + """``PContext`` holding worker processes invoked as a binary.""" + + def __init__( + self, + name: str, + entrypoint: str, + args: dict[int, tuple], + envs: dict[int, dict[str, str]], + logs_specs: LogsSpecs, + log_line_prefixes: dict[int, str] | None = None, + numa_options: NumaOptions | None = None, + duplicate_stdout_filters: list[str] | None = None, + duplicate_stderr_filters: list[str] | None = None, + ): + super().__init__( + name, + entrypoint, + args, + envs, + logs_specs, + log_line_prefixes, + duplicate_stdout_filters, + duplicate_stderr_filters, + ) + + # state vector; _vdone[local_rank] -> is local_rank finished or not + self._running_local_ranks: set[int] = set(range(self.nprocs)) + self._failures: dict[int, ProcessFailure] = {} + self.subprocess_handlers: dict[int, SubprocessHandler] = {} + self._numa_options: NumaOptions | None = numa_options + + def _start(self): + if self.subprocess_handlers: + raise ValueError( + "The subprocess handlers already initialized. Most likely the start method got called twice." + ) + self.subprocess_handlers = { + local_rank: get_subprocess_handler( + entrypoint=self.entrypoint, # type: ignore[arg-type] # entrypoint is always a str + args=self.args[local_rank], + env=self.envs[local_rank], + stdout=self.stdouts[local_rank], + stderr=self.stderrs[local_rank], + local_rank_id=local_rank, + numa_options=self._numa_options, + ) + for local_rank in range(self.nprocs) + } + + def _capture_process_failures(self, done_local_ranks: set[int]): + for local_rank in self._running_local_ranks: + handler = self.subprocess_handlers[local_rank] + exitcode = handler.proc.poll() + if exitcode is not None: + done_local_ranks.add(local_rank) + if exitcode != 0: # failed or signaled + self._failures[local_rank] = ProcessFailure( + local_rank=local_rank, + pid=handler.proc.pid, + exitcode=exitcode, + error_file=self.error_files[local_rank], + ) + # else: --> succeeded; nothing to do + + def _poll(self) -> RunProcsResult | None: + done_local_ranks: set[int] = set() + self._capture_process_failures(done_local_ranks) + + self._running_local_ranks.difference_update(done_local_ranks) + + # if ALL procs are finished or ANY have failed + if not self._running_local_ranks or self._failures: + self.close() # terminate all running procs + self._capture_process_failures( + done_local_ranks + ) # log sigterms and sigkill exit codes in the self._failures for bookkeeping purposes + + result = RunProcsResult( + failures=self._failures, + stdouts=self.stdouts, + stderrs=self.stderrs, + ) + if result.is_failed(): + first_failure = min(result.failures.values(), key=lambda f: f.timestamp) + logger.error( + "failed (exitcode: %s) local_rank: %s (pid: %s) of binary: %s", + first_failure.exitcode, + first_failure.local_rank, + first_failure.pid, + self.entrypoint, + ) + else: + # Populate return with dummy values. This provides consistency with MultiprocessingHandler + result.return_values = dict.fromkeys(range(self.nprocs)) + + return result + else: # there are no failures and procs still running + return None + + def pids(self) -> dict[int, int]: + return { + local_rank: sh.proc.pid + for local_rank, sh in self.subprocess_handlers.items() + } + + def _close(self, death_sig: signal.Signals, timeout: int = 30) -> None: + if not self.subprocess_handlers: + return + for handler in self.subprocess_handlers.values(): + if handler.proc.poll() is None: + logger.warning( + "Sending process %s closing signal %s", + handler.proc.pid, + death_sig.name, + ) + handler.close(death_sig=death_sig) + end = time.monotonic() + timeout + for handler in self.subprocess_handlers.values(): + time_to_wait = end - time.monotonic() + if time_to_wait <= 0: + break + try: + handler.proc.wait(time_to_wait) + except subprocess.TimeoutExpired: + # Ignore the timeout expired exception, since + # the child process will be forcefully terminated via SIGKILL + pass + for handler in self.subprocess_handlers.values(): + if handler.proc.poll() is None: + logger.warning( + "Unable to shutdown process %s via %s, forcefully exiting via %s", + handler.proc.pid, + death_sig, + _get_kill_signal(), + ) + handler.close(death_sig=_get_kill_signal()) + # Bounded wait: a process stuck in uninterruptible kernel sleep + # (Linux D-state, common with hung NCCL/GPU collectives) cannot + # be reaped by SIGKILL. Without a timeout the agent itself would + # wedge here and the supervising launcher could never exit. + try: + handler.proc.wait(timeout) + except subprocess.TimeoutExpired: + logger.error( + "Process %s did not exit after %s within %ss;" + " abandoning and continuing teardown." + " Worker may be in uninterruptible kernel sleep" + " (e.g. wedged on GPU/NIC);" + " host may need to be recycled.", + handler.proc.pid, + _get_kill_signal().name, + timeout, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a1d3dee0b7ae186fa47a1e6b80ffeea156e37221 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py @@ -0,0 +1,408 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Each host in a distributed PyTorch job runs with a single TorchElastic agent, +and multiple workers (as children processes of the TorchElastic agent). +Since the workers are user-provided (your PyTorch script/job), TorchElastic +has a way to propagate errors on the trainers through the agent and up to the +scheduler, which ultimately informs the end-user about the state of the job +and applies any retry policies. + +TorchElastic categorizes errors into 3 categories: + ++----------------+----------------+--------------------------------------------------------------+ +| Category | Sub-Category | Description | ++================+================+==============================================================+ +| User Error | Input Error | invalid inputs to TorchElastic APIs (e.g. min > max nodes) | +| +----------------+--------------------------------------------------------------+ +| | Worker Failure | any failures on the worker child process | ++----------------+----------------+--------------------------------------------------------------+ +| Platform Error | n/a | failures caused by the agent | ++----------------+----------------+--------------------------------------------------------------+ +| Infra Error | n/a | failures outside the domain of the agent and workers | +| | | (e.g. host failures) | ++----------------+----------------+--------------------------------------------------------------+ + +All errors other than "Worker Failure" are either raised canonically from the +agent process or implicitly or explicitly crash the agent process. So the +standard language (python) provided exception handling strategies apply. + +Worker Failures are special because the exception/failure originates on a different +process from the agent so the error needs to be propagated inter-process +(e.g. the agent cannot simply ``try-catch`` an exception raised on the worker process). + +TorchElastic agents use :func:`torch.distributed.elastic.multiprocessing.start_processes` +to launch the workers which has a simple file based inter-process error propagation +built-in. + +Any function or binary entrypoint decorated with :func:`record` +will write uncaught exceptions (with the trace information) to a file specified by the +environment variable ``TORCHELASTIC_ERROR_FILE``. The parent process (e.g. agent) +sets this env var on each child it launches, then aggregates the error files for all +children, and propagates the one with the **smallest** timestamp (e.g. the **first** error). +""" + +import json +import os +import signal +import socket +import time +from collections.abc import Callable +from dataclasses import dataclass, field +from datetime import datetime +from functools import wraps +from string import Template +from typing import Any, Optional, TypeVar, Union +from typing_extensions import ParamSpec + +from torch.distributed.elastic.utils.logging import get_logger + +from .error_handler import ErrorHandler +from .handlers import get_error_handler + + +__all__ = [ + "ProcessFailure", + "ChildFailedError", + "record", + "ErrorHandler", + "get_error_handler", +] + +logger = get_logger(__name__) + + +JSON = dict[str, Any] + +_EMPTY_ERROR_DATA: dict[str, Any] = {"message": ""} +_NOT_AVAILABLE = "" + +_R = TypeVar("_R") +_P = ParamSpec("_P") + + +@dataclass +class ProcessFailure: + """ + Represent the failed process result. When the worker process fails, it may record failure root cause into the file. + + Tries to read the failure timestamp from the provided ``error_file``, + if the ``error_file`` does not exist, the timestamp is the current + timestamp (seconds since epoch). + + The ``message`` field is a concise explanation of the failure. If + the error file exists then the message is obtained from the error file. + Otherwise one is generated based on the failure signature. + + .. note:: It is assumed that the ``error_file`` is written by + ``torch.distributed.elastic.multiprocessing.errors.error_handler.ErrorHandler``. + Otherwise the behavior is undefined. + + """ + + local_rank: int + pid: int + exitcode: int + error_file: str + error_file_data: JSON = field(init=False) + message: str = field(init=False) + timestamp: int = field(init=False) + + def __post_init__(self): + self.error_file_data = _EMPTY_ERROR_DATA + if os.path.isfile(self.error_file): + try: + with open(self.error_file) as fp: + self.error_file_data = json.load(fp) + logger.debug( + "User process failed with error data: %s", + json.dumps(self.error_file_data, indent=2), + ) + self.message, self.timestamp = self._get_error_data( + self.error_file_data + ) + except Exception: + logger.exception("Failed to parse reply file: %s", self.error_file) + raise + else: + self._set_no_reply_file() + + # make up an informative message if not already present + if not self.message: + # signals typically do not generate an error file message + if self.exitcode < 0: + self.message = ( + f"Signal {-self.exitcode} ({self.signal_name()})" + f" received by PID {self.pid}" + ) + else: + self.error_file_data["errorTraits"] = { + "category": "system_terminated_error", + "retryability": "False", + } + self.message = "To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html" + + def _get_error_data(self, error_file_data: dict[str, Any]) -> tuple[str, int]: + message = error_file_data["message"] + if isinstance(message, str): + timestamp = int(error_file_data.get("timestamp", 0)) + else: + timestamp = int(message["extraInfo"]["timestamp"]) + return (message, timestamp) + + def _set_no_reply_file(self): + self.error_file = _NOT_AVAILABLE + self.error_file_data = _EMPTY_ERROR_DATA + self.message = "" + self.timestamp = int(time.time()) + + def signal_name(self) -> str: + if self.exitcode < 0: + # We don't want to kill the parent process trying to find the signal name. + # if the signal doesn't map to a known name, use not available. + try: + return signal.Signals(-self.exitcode).name + except Exception: + return _NOT_AVAILABLE + else: + return _NOT_AVAILABLE + + def timestamp_isoformat(self): + """Return timestamp in ISO format (YYYY-MM-DD_HH:MM:SS).""" + return datetime.fromtimestamp(self.timestamp).isoformat(sep="_") + + +GlobalRank = int + +_FAILURE_FORMAT_TEMPLATE = """[${idx}]: + time : ${time} + host : ${hostname} + rank : ${rank} (local_rank: ${local_rank}) + exitcode : ${exitcode} (pid: ${pid}) ${signal_name} + error_file: ${error_file} + traceback : ${message}""" + +# extra new lines before and after are intentional +_MSG_FORMAT_TEMPLATE = """ +${boarder} +${title} +${section} +Failures: +${other_failures} +${section} +Root Cause (first observed failure): +${root_failure} +${boarder}""" + + +class ChildFailedError(Exception): + """ + Special exception type that can be raised from a function annotated with the + ``@record`` decorator to have the child process' (root exception) propagate + up the stack as-is (e.g. without being wrapped in the parent's traceback). + + Useful in cases where the parent is a simple nanny process + and the child (worker) processes are actually doing meaningful compute. + In this case, errors typically occur on the child process as the parent + is not doing anything non-trivial, and child errors should be propagated + to the scheduler for accurate root cause diagnostics. + + .. note:: The propagation relies on error files rather than exception handling to + support both function and binary launches. + + Example: + :: + + # process tree on a host (container) + 0: scheduler-init-process: + |- 1: torchelastic_agent: + |- 2: trainer_0 (ok) + |- 3: trainer_1 (fail) -> error.json + |- ... + |- n+2: trainer_n (ok) + |- n+3: other processes + |- ... + + In the example above, trainer 1's failure (written into error.json) is + the root cause and should be reported to the scheduler's init process. + The torchelastic agent raises a ``ChildFailedError("trainer", {1: "trainer_1/error.json"})`` + upon detecting trainer 1's failure which would propagate the contents + of trainer 1's error file to the scheduler's init process. + """ + + def __init__(self, name: str, failures: dict[GlobalRank, ProcessFailure]): + self.name = name + self.failures = failures + # does not make sense to create a ChildFailedError with no failures + if not self.failures: + raise AssertionError + super().__init__(self.format_msg()) + + def get_first_failure(self) -> tuple[GlobalRank, ProcessFailure]: + rank = min(self.failures.keys(), key=lambda r: self.failures[r].timestamp) + return rank, self.failures[rank] + + def format_msg(self, boarder_delim="=", section_delim="-"): + title = f"{self.name} FAILED" + root_rank, _root_failure = self.get_first_failure() + + root_failure_fmt: str = "" + other_failures_fmt: list[str] = [] + width = len(title) + for idx, (rank, failure) in enumerate(self.failures.items()): + fmt, w = self._format_failure(idx, rank, failure) + width = max(width, w) + if rank == root_rank: + root_failure_fmt = fmt + else: + other_failures_fmt.append(fmt) + + # upper boundary on width + width = min(width, 60) + + return Template(_MSG_FORMAT_TEMPLATE).substitute( + boarder=boarder_delim * width, + title=title, + section=section_delim * width, + root_failure=root_failure_fmt, + other_failures="\n".join(other_failures_fmt or [" "]), + ) + + def _format_failure( + self, idx: int, rank: int, failure: ProcessFailure + ) -> tuple[str, int]: + # failure.message is either a str (when the failure does not generate a traceback - e.g. signals) + # or a dict (json) of the form + # {"message": $ERROR_MSG, "extraInfo": {"py_callstack": $TRACEBACK, timestamp: $TS}} + # so the display logic is: + # 1. if failure.message is not a dict (it is a str) just show it as is + # 2. else try to get the traceback (py_callstack) + # 3. if the traceback is not there, use the message + # 4. if the message is not there show + msg = failure.message + if isinstance(failure.message, dict): + msg = ( + failure.message.get("extraInfo", {}) + .get("py_callstack", failure.message.get("message", "")) + .replace("\n", "\n ") # to properly indent the traceback + ) + + # For signal failures, let the (build-swapped) handler append + # device-fault context (e.g. ROCm GPU faults) to the rendered message. + # Works on the local string, so error_file_data is never mutated, and it + # is a no-op in OSS (the base handler returns the message unchanged). + # Best-effort: never let a handler override break failure rendering. + if failure.exitcode < 0: + try: + msg = get_error_handler().maybe_enrich_signal_failure_message( + msg, failure.error_file + ) + except Exception: + logger.warning("Failed to enrich signal failure message", exc_info=True) + + signal_name = failure.signal_name() + signal_name_str = f" ({signal_name})" if signal_name != _NOT_AVAILABLE else "" + + fmt = Template(_FAILURE_FORMAT_TEMPLATE).substitute( + idx=idx, + time=failure.timestamp_isoformat(), + hostname=socket.getfqdn(), + rank=rank, + local_rank=failure.local_rank, + exitcode=failure.exitcode, + pid=failure.pid, + signal_name=signal_name_str, + error_file=failure.error_file, + message=msg, + ) + width = 0 + for line in fmt.split("\n"): + width = max(width, len(line)) + return fmt, width + + +def record( + fn: Callable[_P, _R], error_handler: ErrorHandler | None = None +) -> Callable[_P, _R | None]: + """ + Syntactic sugar to record errors/exceptions that happened in the decorated + function using the provided ``error_handler``. + + Using this decorator is equivalent to: + + :: + + error_handler = get_error_handler() + error_handler.initialize() + try: + foobar() + except ChildFailedError as e: + _, failure = e.get_first_failure() + error_handler.dump_error_file(failure.error_file, failure.exitcode) + raise + except Exception as e: + error_handler.record_exception(e) + raise + + .. important:: use this decorator once per process at the top level method, + typically this is the main method. + + Example + + :: + + @record + def main(): + pass + + + if __name__ == "__main__": + main() + + """ + if not error_handler: + error_handler = get_error_handler() + + def wrap(f: Callable[_P, _R]) -> Callable[_P, _R | None]: + @wraps(f) + def wrapper(*args: _P.args, **kwargs: _P.kwargs): + if error_handler is None: + raise AssertionError # assertion for mypy type checker + error_handler.initialize() + try: + return f(*args, **kwargs) + except SystemExit as se: + # For run_path based entrypoints, SystemExit with code = 0 will never exit. + # Handling it here by returning a value: + if se.code == 0: + return None + else: + raise + except ChildFailedError as e: + rank, failure = e.get_first_failure() + if failure.error_file != _NOT_AVAILABLE: + error_handler.dump_error_file(failure.error_file, failure.exitcode) + else: + logger.info( + ( + "local_rank %s FAILED with no error file." + " Decorate your entrypoint fn with @record for traceback info." + " See: https://pytorch.org/docs/stable/elastic/errors.html", + rank, + ) + ) + raise + except Exception as e: + error_handler.record_exception(e) + raise + + return wrapper + + return wrap(fn) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d0ce40f35981fa4d65bb08b693c6cdb424aa9609 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/error_handler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/error_handler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8adf2799f9e407e29ca7d2f35f6411d1ece1ee02 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/error_handler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..710067d7d8d1acd382fdff8ef2de890a34691725 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__pycache__/handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/error_handler.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/error_handler.py new file mode 100644 index 0000000000000000000000000000000000000000..21354ba6fd82377ef81947fd73cccdb9e80c2487 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/error_handler.py @@ -0,0 +1,182 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import faulthandler +import json +import logging +import os +import time +import traceback +import warnings +from typing import Any + + +__all__ = ["ErrorHandler"] + +logger = logging.getLogger(__name__) + + +class ErrorHandler: + """ + Write the provided exception object along with some other metadata about + the error in a structured way in JSON format to an error file specified by the + environment variable: ``TORCHELASTIC_ERROR_FILE``. If this environment + variable is not set, then simply logs the contents of what would have been + written to the error file. + + This handler may be subclassed to customize the handling of the error. + Subclasses should override ``initialize()`` and ``record_exception()``. + """ + + def _get_error_file_path(self) -> str | None: + """ + Return the error file path. + + May return ``None`` to have the structured error be logged only. + """ + return os.environ.get("TORCHELASTIC_ERROR_FILE", None) + + def initialize(self) -> None: + """ + Call prior to running code that we wish to capture errors/exceptions. + + Typically registers signal/fault handlers. Users can override this + function to add custom initialization/registrations that aid in + propagation/information of errors/signals/exceptions/faults. + """ + try: + faulthandler.enable(all_threads=True) + except Exception as e: + warnings.warn( + f"Unable to enable fault handler. {type(e).__name__}: {e}", stacklevel=2 + ) + + def _write_error_file(self, file_path: str, error_msg: str) -> None: + """Write error message to the file.""" + try: + with open(file_path, "w") as fp: + fp.write(error_msg) + except Exception as e: + warnings.warn( + f"Unable to write error to file. {type(e).__name__}: {e}", stacklevel=2 + ) + + def record_exception(self, e: BaseException) -> None: + """ + Write structured information about the exception into an error file in JSON format. + + If the error file cannot be determined, then logs the content + that would have been written to the error file. + """ + file = self._get_error_file_path() + if file: + data = { + "message": { + "message": f"{type(e).__name__}: {e}", + "extraInfo": { + "py_callstack": traceback.format_exc(), + "timestamp": str(int(time.time())), + }, + } + } + with open(file, "w") as fp: + json.dump(data, fp) + + def maybe_enrich_signal_failure_message(self, message: str, error_file: str) -> str: + """Hook to enrich a signal (no-traceback) failure message. + + Called from ``ProcessFailure`` when a worker fails by signal (negative + exitcode). Subclasses may override this to append device-specific fault + context (e.g. GPU memory faults) scanned from the worker logs near + ``error_file``, so it surfaces in the propagated failure regardless of + how the failure is later handled. The base implementation is a no-op + that returns ``message`` unchanged. + """ + return message + + def override_error_code_in_rootcause_data( + self, + rootcause_error_file: str, + rootcause_error: dict[str, Any], + error_code: int = 0, + ): + """Modify the rootcause_error read from the file, to correctly set the exit code.""" + if "message" not in rootcause_error: + logger.warning( + "child error file (%s) does not have field `message`. \n" + "cannot override error code: %s", + rootcause_error_file, + error_code, + ) + elif isinstance(rootcause_error["message"], str): + logger.warning( + "child error file (%s) has a new message format. \n" + "skipping error code override", + rootcause_error_file, + ) + else: + rootcause_error["message"]["errorCode"] = error_code + + def dump_error_file(self, rootcause_error_file: str, error_code: int = 0): + """Dump parent error file from child process's root cause error and error code.""" + with open(rootcause_error_file) as fp: + rootcause_error = json.load(fp) + # Override error code since the child process cannot capture the error code if it + # is terminated by signals like SIGSEGV. + if error_code: + self.override_error_code_in_rootcause_data( + rootcause_error_file, rootcause_error, error_code + ) + logger.debug( + "child error file (%s) contents:\n%s", + rootcause_error_file, + json.dumps(rootcause_error, indent=2), + ) + + my_error_file = self._get_error_file_path() + if my_error_file: + # Guard against existing error files + # This can happen when the child is created using multiprocessing + # and the same env var (TORCHELASTIC_ERROR_FILE) is used on the + # parent and child to specify the error files (respectively) + # because the env vars on the child is set in the wrapper function + # and by default the child inherits the parent's env vars, if the child + # process receives a signal before the wrapper function kicks in + # and the signal handler writes to the error file, then the child + # will write to the parent's error file. In this case just log the + # original error file contents and overwrite the error file. + self._rm(my_error_file) + self._write_error_file(my_error_file, json.dumps(rootcause_error)) + logger.info("dumped error file to parent's %s", my_error_file) + else: + logger.error( + "no error file defined for parent, to copy child error file (%s)", + rootcause_error_file, + ) + + def _rm(self, my_error_file): + if os.path.isfile(my_error_file): + # Log the contents of the original file. + with open(my_error_file) as fp: + try: + original = json.dumps(json.load(fp), indent=2) + logger.warning( + "%s already exists" + " and will be overwritten." + " Original contents:\n%s", + my_error_file, + original, + ) + except json.decoder.JSONDecodeError: + logger.warning( + "%s already exists" + " and will be overwritten." + " Unable to load original contents:\n", + my_error_file, + ) + os.remove(my_error_file) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..6721217a41190c2bdd6bf2293540a33c893c145d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/handlers.py @@ -0,0 +1,18 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +# Multiprocessing error-reporting module + + +from torch.distributed.elastic.multiprocessing.errors.error_handler import ErrorHandler + + +__all__ = ["get_error_handler"] + + +def get_error_handler() -> ErrorHandler: + return ErrorHandler() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/redirects.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/redirects.py new file mode 100644 index 0000000000000000000000000000000000000000..12d3ef947e57929dd1ef064900f8768872f485f1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/redirects.py @@ -0,0 +1,229 @@ +# mypy: allow-untyped-defs +# !/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +# Taken and modified from original source: +# https://eli.thegreenplace.net/2015/redirecting-all-kinds-of-stdout-in-python/ +import ctypes +import logging +import os +import sys +from contextlib import contextmanager +from functools import partial + + +IS_WINDOWS = sys.platform == "win32" +IS_MACOS = sys.platform == "darwin" + + +logger = logging.getLogger(__name__) + + +_WIN32_STD_HANDLE = { + "stdout": -11, # STD_OUTPUT_HANDLE + "stderr": -12, # STD_ERROR_HANDLE +} + + +def get_libc(): + if IS_MACOS: + logger.warning("NOTE: Redirects are currently not supported in MacOs.") + return None + elif IS_WINDOWS: + for lib_name in ("ucrtbase", "msvcrt", "msvcr110", "msvcr100"): + try: + lib = ctypes.CDLL(lib_name) + logger.debug("Loaded Windows C runtime: %s", lib_name) + return lib + except OSError: + continue + raise RuntimeError( + "Could not load a C runtime DLL on Windows (tried: ucrtbase, msvcrt, " + "msvcr110, msvcr100). Redirects cannot function without a CRT." + ) + else: + return ctypes.CDLL("libc.so.6") + + +libc = get_libc() + + +def _c_std(stream: str): + if IS_WINDOWS: + stream_index = 2 if stream == "stderr" else 1 + try: + iob_func = libc.__acrt_iob_func + iob_func.restype = ctypes.POINTER(ctypes.c_void_p) + iob_func.argtypes = [ctypes.c_uint] + return iob_func(stream_index) + except AttributeError: + pass + try: + legacy_index = 2 if stream == "stderr" else 1 + iob = (ctypes.POINTER(ctypes.c_void_p) * 3).in_dll(libc, "_iob") + return iob[legacy_index] + except (AttributeError, OSError) as err: + raise RuntimeError( + f"Could not resolve C-runtime FILE* for '{stream}'. " + "Neither __acrt_iob_func nor _iob are available in the loaded CRT." + ) from err + return ctypes.c_void_p.in_dll(libc, stream) + + +def _python_std(stream: str): + return {"stdout": sys.stdout, "stderr": sys.stderr}[stream] + + +_VALID_STD = {"stdout", "stderr"} + + +if IS_WINDOWS: # libc is None on macOS; all of the below is Windows-only + import io as _io + import msvcrt as _msvcrt + + _kernel32 = ctypes.WinDLL("kernel32", use_last_error=True) # type: ignore[attr-defined] + + _crt_dup = libc._dup + _crt_dup2 = libc._dup2 + _crt_dup.restype = ctypes.c_int + _crt_dup.argtypes = [ctypes.c_int] + _crt_dup2.restype = ctypes.c_int + _crt_dup2.argtypes = [ctypes.c_int, ctypes.c_int] + + @contextmanager + def redirect(std: str, to_file: str): + """ + Redirect ``std`` (one of ``"stdout"`` or ``"stderr"``) to a file at ``to_file``. + + On Windows this performs a four-layer redirect: + + 1. ``sys.stdout``/``sys.stderr`` -- rewired to a new TextIOWrapper so + Python's ``print()`` writes to the destination file. + 2. CRT fd (``_dup2``) -- captures C ``printf`` and UCRT ``FILE*`` writers. + 3. Win32 ``SetStdHandle`` -- captures native code using ``WriteFile``/ + ``WriteConsole`` directly, including HIP/ROCm. + 4. ``fflush`` before each switch -- prevents lost output from CRT buffering. + + .. note:: If ROCm/HIP caches the Win32 HANDLE before this redirect runs + (e.g. at ``import torch`` time), set up the redirect *before* + importing torch/ROCm to capture all output. + + Directory of ``to_file`` is assumed to exist. The destination file is + overwritten if it already exists. + """ + if std not in _VALID_STD: + raise ValueError( + f"unknown standard stream <{std}>, must be one of {_VALID_STD}" + ) + + std_fd = 1 if std == "stdout" else 2 + win32_handle_id = _WIN32_STD_HANDLE[std] + orig_sys_std = getattr(sys, std) + orig_fd_dup = _crt_dup(std_fd) + if orig_fd_dup == -1: + raise OSError(f"CRT _dup failed for {std} (fd={std_fd})") + orig_win32_handle = _kernel32.GetStdHandle(win32_handle_id) + + with open(to_file, mode="w+b") as dst: + dst_fd = dst.fileno() + + try: + libc.fflush(_c_std(std)) + except Exception: + pass + try: + orig_sys_std.flush() + except Exception: + pass + + _kernel32.SetStdHandle( + win32_handle_id, + _msvcrt.get_osfhandle(dst_fd), # pyrefly: ignore [missing-attribute] + ) + + if _crt_dup2(dst_fd, std_fd) == -1: + raise OSError(f"CRT _dup2 failed redirecting {std}") + + new_sys_std = _io.TextIOWrapper( + open(dst_fd, mode="wb", closefd=False), # noqa: SIM115 + encoding=orig_sys_std.encoding or "utf-8", + errors="replace", + line_buffering=True, + ) + setattr(sys, std, new_sys_std) + + try: + yield + finally: + try: + new_sys_std.flush() + except Exception: + pass + try: + libc.fflush(_c_std(std)) + except Exception: + pass + + setattr(sys, std, orig_sys_std) + _crt_dup2(orig_fd_dup, std_fd) + os.close(orig_fd_dup) + _kernel32.SetStdHandle(win32_handle_id, orig_win32_handle) + +else: + + @contextmanager + def redirect(std: str, to_file: str): + """ + Redirect ``std`` (one of ``"stdout"`` or ``"stderr"``) to a file in the path specified by ``to_file``. + + This method redirects the underlying std file descriptor (not just python's ``sys.stdout|stderr``). + See usage for details. + + Directory of ``to_file`` is assumed to exist and the destination file + is overwritten if it already exists. + + .. note:: Due to buffering cross source writes are not guaranteed to + appear in wall-clock order. For instance in the example below + it is possible for the C-outputs to appear before the python + outputs in the log file. + + Usage:: + + # syntactic-sugar for redirect("stdout", "tmp/stdout.log") + with redirect_stdout("/tmp/stdout.log"): + print("python stdouts are redirected") + libc = ctypes.CDLL("libc.so.6") + libc.printf(b"c stdouts are also redirected") + os.system("echo system stdouts are also redirected") + + print("stdout restored") + """ + if std not in _VALID_STD: + raise ValueError( + f"unknown standard stream <{std}>, must be one of {_VALID_STD}" + ) + + c_std = _c_std(std) + python_std = _python_std(std) + std_fd = python_std.fileno() + + def _redirect(dst): + libc.fflush(c_std) + python_std.flush() + os.dup2(dst.fileno(), std_fd) + + with os.fdopen(os.dup(std_fd)) as orig_std, open(to_file, mode="w+b") as dst: + _redirect(dst) + try: + yield + finally: + _redirect(orig_std) + + +redirect_stdout = partial(redirect, "stdout") +redirect_stderr = partial(redirect, "stderr") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f56d423ce080fd7c331dc9b43eda58e5370678fc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__init__.py @@ -0,0 +1,16 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +from torch.distributed.elastic.multiprocessing.subprocess_handler.handlers import ( + get_subprocess_handler, +) +from torch.distributed.elastic.multiprocessing.subprocess_handler.subprocess_handler import ( + SubprocessHandler, +) + + +__all__ = ["SubprocessHandler", "get_subprocess_handler"] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..46e9c1ba32dbee809f5f7803b435f2cdc2f2757e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/handlers.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/handlers.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b40fda882ce3047000b15cdd9bd17a8f6172ffa9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/handlers.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/subprocess_handler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/subprocess_handler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..66362e99c4d68f0710c49dcd44befebc4897edf7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/__pycache__/subprocess_handler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..ea1742626e285838485c19911704792510d13fb4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/handlers.py @@ -0,0 +1,33 @@ +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from torch.distributed.elastic.multiprocessing.subprocess_handler.subprocess_handler import ( + SubprocessHandler, +) +from torch.numa.binding import NumaOptions + + +__all__ = ["get_subprocess_handler"] + + +def get_subprocess_handler( + entrypoint: str, + args: tuple, + env: dict[str, str], + stdout: str, + stderr: str, + local_rank_id: int, + numa_options: NumaOptions | None = None, +) -> SubprocessHandler: + return SubprocessHandler( + entrypoint=entrypoint, + args=args, + env=env, + stdout=stdout, + stderr=stderr, + local_rank_id=local_rank_id, + numa_options=numa_options, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/subprocess_handler.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/subprocess_handler.py new file mode 100644 index 0000000000000000000000000000000000000000..28b18a962868e97a6aeaeb8eb41c4294ff128847 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/subprocess_handler/subprocess_handler.py @@ -0,0 +1,89 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import os +import signal +import sys +from subprocess import Popen +from typing import Any + +from torch.numa.binding import _maybe_wrap_command_args_with_numa_binding, NumaOptions + + +__all__ = ["SubprocessHandler"] + +IS_WINDOWS = sys.platform == "win32" + + +def _get_default_signal() -> signal.Signals: + """Get the default termination signal. SIGTERM for unix, CTRL_C_EVENT for windows.""" + if IS_WINDOWS: + return signal.CTRL_C_EVENT # type: ignore[attr-defined] + else: + return signal.SIGTERM + + +class SubprocessHandler: + """ + Convenience wrapper around python's ``subprocess.Popen``. Keeps track of + meta-objects associated to the process (e.g. stdout and stderr redirect fds). + """ + + def __init__( + self, + entrypoint: str, + args: tuple, + env: dict[str, str], + stdout: str | None, + stderr: str | None, + local_rank_id: int, + numa_options: NumaOptions | None, + ): + self._stdout = open(stdout, "w") if stdout else None # noqa: SIM115 + self._stderr = open(stderr, "w") if stderr else None # noqa: SIM115 + # inherit parent environment vars + env_vars = os.environ.copy() + env_vars.update(env) + + args_str = (entrypoint, *[str(e) for e in args]) + args_str = _maybe_wrap_command_args_with_numa_binding( + args_str, + device_index=local_rank_id, + numa_options=numa_options, + ) + + self.local_rank_id = local_rank_id + + self.proc: Popen = self._popen(args_str, env_vars) + + def _popen(self, args: tuple, env: dict[str, str]) -> Popen: + kwargs: dict[str, Any] = {} + if not IS_WINDOWS: + kwargs["start_new_session"] = True + + return Popen( + # pyre-fixme[6]: Expected `Union[typing.Sequence[Union[_PathLike[bytes], + # _PathLike[str], bytes, str]], bytes, str]` for 1st param but got + # `Tuple[str, *Tuple[Any, ...]]`. + args=args, + env=env, + stdout=self._stdout, + stderr=self._stderr, + **kwargs, + ) + + def close(self, death_sig: signal.Signals | None = None) -> None: + if not death_sig: + death_sig = _get_default_signal() + if IS_WINDOWS: + self.proc.send_signal(death_sig) + else: + os.killpg(self.proc.pid, death_sig) + if self._stdout: + self._stdout.close() + if self._stderr: + self._stderr.close() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/tail_log.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/tail_log.py new file mode 100644 index 0000000000000000000000000000000000000000..72ed4da6f17ecc17435af2dbef1753b3b3fafcf5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/tail_log.py @@ -0,0 +1,167 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging +import os +import time +from collections.abc import Callable +from concurrent.futures.thread import ThreadPoolExecutor +from threading import Event +from typing import TextIO, TYPE_CHECKING + + +if TYPE_CHECKING: + from concurrent.futures._base import Future + +__all__ = ["tail_logfile", "TailLog"] + +logger = logging.getLogger(__name__) + + +def tail_logfile( + header: str, + file: str, + dst: TextIO, + finished: Event, + interval_sec: float, + log_line_filter: Callable[[str], bool] | None = None, +): + while not os.path.exists(file): + if finished.is_set(): + return + time.sleep(interval_sec) + + with open(file, errors="replace") as fp: + while True: + line = fp.readline() + + if line: + if log_line_filter and log_line_filter(line): + dst.write(f"{header}{line}") + else: # reached EOF + if finished.is_set(): + # log line producer is finished + break + else: + # log line producer is still going + # wait for a bit before looping again + time.sleep(interval_sec) + + +class TailLog: + """ + Tail the given log files. + + The log files do not have to exist when the ``start()`` method is called. The tail-er will gracefully wait until + the log files are created by the producer and will tail the contents of the + log files until the ``stop()`` method is called. + + .. warning:: ``TailLog`` will wait indefinitely for the log file to be created! + + Each log file's line will be suffixed with a header of the form: ``[{name}{idx}]:``, + where the ``name`` is user-provided and ``idx`` is the index of the log file + in the ``log_files`` mapping. ``log_line_prefixes`` can be used to override the + header for each log file. + + Usage: + + :: + + log_files = {0: "/tmp/0_stdout.log", 1: "/tmp/1_stdout.log"} + tailer = TailLog("trainer", log_files, sys.stdout).start() + # actually run the trainers to produce 0_stdout.log and 1_stdout.log + run_trainers() + tailer.stop() + + # once run_trainers() start writing the ##_stdout.log files + # the tailer will print to sys.stdout: + # >>> [trainer0]:log_line1 + # >>> [trainer1]:log_line1 + # >>> [trainer0]:log_line2 + # >>> [trainer0]:log_line3 + # >>> [trainer1]:log_line2 + + .. note:: Due to buffering log lines between files may not necessarily + be printed out in order. You should configure your application's + logger to suffix each log line with a proper timestamp. + + """ + + def __init__( + self, + name: str, + log_files: dict[int, str], + dst: TextIO, + log_line_prefixes: dict[int, str] | None = None, + interval_sec: float = 0.1, + log_line_filter: Callable[[str], bool] = (lambda _: True), + ): + n = len(log_files) + self._threadpool = None + if n > 0: + self._threadpool = ThreadPoolExecutor( + max_workers=n, + thread_name_prefix=f"{self.__class__.__qualname__}_{name}", + ) + + self._name = name + self._dst = dst + self._log_files = log_files + self._log_line_prefixes = log_line_prefixes + self._log_line_filter = log_line_filter + self._finished_events: dict[int, Event] = { + local_rank: Event() for local_rank in log_files + } + self._futs: list[Future] = [] + self._interval_sec = interval_sec + self._stopped = False + + def start(self) -> "TailLog": + if not self._threadpool or not self._dst: + return self + + for local_rank, file in self._log_files.items(): + header = f"[{self._name}{local_rank}]:" + if self._log_line_prefixes and local_rank in self._log_line_prefixes: + header = self._log_line_prefixes[local_rank] + self._futs.append( + self._threadpool.submit( + tail_logfile, + header=header, + file=file, + dst=self._dst, + finished=self._finished_events[local_rank], + interval_sec=self._interval_sec, + log_line_filter=self._log_line_filter, + ) + ) + return self + + def stop(self) -> None: + for finished in self._finished_events.values(): + finished.set() + + for local_rank, f in enumerate(self._futs): + try: + f.result() + except Exception as e: + logger.exception( + "error in log tailer for %s%s. %s", + self._name, + local_rank, + e.__class__.__qualname__, + ) + + if self._threadpool: + self._threadpool.shutdown(wait=True) + + self._stopped = True + + def stopped(self) -> bool: + return self._stopped diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f1a07d51e8f0bf1b40c998113b1e19ea227b3f1f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__init__.py @@ -0,0 +1,163 @@ +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +In the context of Torch Distributed Elastic we use the term *rendezvous* to +refer to a particular functionality that combines a **distributed +synchronization** primitive with **peer discovery**. + +It is used by Torch Distributed Elastic to gather participants of a training +job (i.e. nodes) such that they all agree on the same list of participants and +everyone's roles, as well as make a consistent collective decision on when +training can begin/resume. + +Torch Distributed Elastic rendezvous provides the following critical +functionalities: + +**Barrier**: + +Nodes performing rendezvous will all block until the rendezvous is considered +complete - this happens when at least ``min`` total number of nodes have joined +the rendezvous barrier (for the same job). This also implies the barrier is not +necessarily of fixed size. + +There's an additional small waiting time after reaching ``min`` number of +nodes - this is used to ensure the rendezvous is not completed "too quickly" +(which could potentially exclude additional nodes attempting to join at +approximately the same time). + +If ``max`` number of nodes is gathered at the barrier, the rendezvous is +completed immediately. + +There's also an overall timeout which causes the rendezvous to fail if ``min`` +number of nodes is never reached - this is meant to be a simple fail-safe to +help release partially allocated job resources, in case there's a problem with +the resource manager, and is meant to be interpreted as non-retryable. + +**Exclusivity**: + +A simple distributed barrier would not be sufficient, as we also need to ensure +that only one group of nodes exists at any given time (for a given job). In +other words, new nodes (i.e. joining late) should not be able to form a parallel +independent group of workers for the same job. + +Torch Distributed Elastic rendezvous ensures that if a group of nodes has +already completed a rendezvous (and hence might already be training), then +additional "late" nodes attempting to rendezvous will only announce themselves +as waiting, and will have to wait until the (previously completed) existing +rendezvous is destroyed first. + +**Consistency**: + +When a rendezvous is completed, all its members will agree on the job membership +and everyone's role in it. This role is represented using an integer, called +rank, that is between 0 and world size. + +Note that ranks are *not stable*, in the sense that the same node can be +assigned a different rank in the next (re-)rendezvous. + +**Fault-tolerance**: + +Torch Distributed Elastic rendezvous is designed to tolerate node failures +during the rendezvous process. Should a process crash (or lose network +connectivity, etc), between joining the rendezvous and it being completed, then +a re-rendezvous with remaining healthy nodes will happen automatically. + +A node can also fail *after* it has completed (or *has been observed* by other +nodes to have completed) the rendezvous - this scenario will be handled by the +Torch Distributed Elastic ``train_loop`` instead (where it will also trigger a +re-rendezvous). + +**Shared key-value store**: + +When the rendezvous is completed, a shared key-value store is created and +returned. This store implements a ``torch.distributed.Store`` API (see +`distributed communication docs +`__). + +This store is only shared by the members of the completed rendezvous. It +is intended to be used by Torch Distributed Elastic to exchange information +necessary to initialize job control and data-planes. + +**Waiting workers and rendezvous closing**: + +Torch Distributed Elastic rendezvous handler object provides additional +functionalities, which are technically not part of the rendezvous process: + +1. Querying how many workers arrived late at the barrier, who can participate in + *next* rendezvous. + +2. Setting the rendezvous *closed* to signal all nodes not to participate in + next rendezvous. + +**DynamicRendezvousHandler**: + +Torch Distributed Elastic comes with the :py:class:`.DynamicRendezvousHandler` +class that implements the rendezvous mechanism described above. It is a backend- +agnostic type that expects a particular :py:class:`.RendezvousBackend` instance +to be specified during construction. + +Torch distributed users can either implement their own backend type or use one +of the following implementations that come with PyTorch: + +- :py:class:`.C10dRendezvousBackend`: Uses a C10d store (by default + ``TCPStore``) as the rendezvous backend. The main advantage of using a C10d + store is that it requires no 3rd-party dependency (such as etcd) to establish + a rendezvous. +- :py:class:`.EtcdRendezvousBackend`: Supersedes the legacy + :py:class:`.EtcdRendezvousHandler` class. Passing an + :py:class:`.EtcdRendezvousBackend` instance to + :py:class:`.DynamicRendezvousHandler` is functionally equivalent to + instantiating an :py:class:`.EtcdRendezvousHandler`. + + :: + + store = TCPStore("localhost") + + backend = C10dRendezvousBackend(store, "my_run_id") + + rdzv_handler = DynamicRendezvousHandler.from_backend( + run_id="my_run_id", store=store, backend=backend, min_nodes=2, max_nodes=4 + ) +""" + +from .api import ( + rendezvous_handler_registry, + RendezvousClosedError, + RendezvousConnectionError, + RendezvousError, + RendezvousGracefulExitError, + RendezvousHandler, + RendezvousHandlerCreator, + RendezvousHandlerRegistry, + RendezvousInfo, + RendezvousParameters, + RendezvousStateError, + RendezvousStoreInfo, + RendezvousTimeoutError, +) +from .registry import _register_default_handlers, _register_out_of_tree_handlers + + +_register_default_handlers() +_register_out_of_tree_handlers() + + +__all__ = [ + "RendezvousClosedError", + "RendezvousConnectionError", + "RendezvousError", + "RendezvousGracefulExitError", + "RendezvousHandler", + "RendezvousHandlerCreator", + "RendezvousHandlerRegistry", + "RendezvousInfo", + "RendezvousParameters", + "RendezvousStateError", + "RendezvousStoreInfo", + "RendezvousTimeoutError", + "rendezvous_handler_registry", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..76745d9471dc778b0dce4f8a2936f06b53d1bb44 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/_etcd_stub.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/_etcd_stub.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..afa527890c246f7d346569cfceecaa82c8092124 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/_etcd_stub.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5b70d6efb01fa558ea278e5678c188909e9e662b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/c10d_rendezvous_backend.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/c10d_rendezvous_backend.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a6368c0f2c2409d41a31b223b88d069adac2d889 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/c10d_rendezvous_backend.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/dynamic_rendezvous.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/dynamic_rendezvous.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a588376d7c221eb2d19303e84d46e045bc9525eb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/dynamic_rendezvous.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..83f84436a91bab18a5fa18eedf6a2089b8cc99af Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous_backend.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous_backend.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9622e26ca6a311f73717757717c4fea3c1b247ec Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_rendezvous_backend.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_server.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_server.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5e91edb07d2ce359aaab3fbb196b07760b44fa49 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_server.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_store.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_store.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..86039041f83c63569f063126f22153b84907af48 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/etcd_store.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/registry.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/registry.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..624af263bb1bc6ba15e2f8efdd0f5fc2a9e35df1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/registry.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/static_tcp_rendezvous.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/static_tcp_rendezvous.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b9a242b68227791f1ab1ddb19ca496ad637205d7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/static_tcp_rendezvous.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c9257d7f7e255429e73c9f8433f5c89e8cdfc7e0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/_etcd_stub.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/_etcd_stub.py new file mode 100644 index 0000000000000000000000000000000000000000..5890a97c672a61b5678e66b006ba173fe7668286 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/_etcd_stub.py @@ -0,0 +1,75 @@ +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from typing import Any + + +""" +This file is not meant to be used directly. It serves as a stub to allow +other files to be safely imported without requiring the installation of +the 'etcd' library. The classes and methods here raise exceptions to +indicate that the real 'etcd' module is needed. +""" + + +class EtcdStubError(ImportError): + """Custom exception to indicate that the real etcd module is required.""" + + def __init__(self) -> None: + super().__init__("The 'etcd' module is required but not installed.") + + +class EtcdAlreadyExist(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdCompareFailed(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdKeyNotFound(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdWatchTimedOut(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdEventIndexCleared(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdException(Exception): + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + +class EtcdResult: + def __init__(self) -> None: + raise EtcdStubError + + +class Client: + def __init__(self, *args: Any, **kwargs: Any) -> None: + raise EtcdStubError + + def read(self, key: str) -> None: + raise EtcdStubError + + def write( + self, key: str, value: Any, ttl: int | None = None, **kwargs: Any + ) -> None: + raise EtcdStubError + + def test_and_set( + self, key: str, value: Any, prev_value: Any, ttl: int | None = None + ) -> None: + raise EtcdStubError diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/api.py new file mode 100644 index 0000000000000000000000000000000000000000..166a0fbb517a9da4a61a2038a9ace6c24b696c0c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/api.py @@ -0,0 +1,391 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import socket +from abc import ABC, abstractmethod +from collections.abc import Callable +from dataclasses import dataclass +from typing import Any, ClassVar + +from torch.distributed import Store +from torch.distributed.elastic.utils.distributed import get_free_port + + +__all__ = [ + "RendezvousClosedError", + "RendezvousConnectionError", + "RendezvousError", + "RendezvousGracefulExitError", + "RendezvousHandler", + "RendezvousHandlerCreator", + "RendezvousHandlerRegistry", + "RendezvousInfo", + "RendezvousParameters", + "RendezvousStateError", + "RendezvousStoreInfo", + "RendezvousTimeoutError", + "rendezvous_handler_registry", +] + + +class RendezvousError(Exception): + """Represents the base type for rendezvous errors.""" + + +class RendezvousClosedError(RendezvousError): + """Raised when a rendezvous is closed.""" + + +class RendezvousTimeoutError(RendezvousError): + """Raised when a rendezvous did not complete on time.""" + + +class RendezvousConnectionError(RendezvousError): + """Raised when the connection to a rendezvous backend has failed.""" + + +class RendezvousStateError(RendezvousError): + """Raised when the state of a rendezvous is corrupt.""" + + +class RendezvousGracefulExitError(RendezvousError): + """Raised when node wasn't included in rendezvous and gracefully exits. + + Exception is a mechanism to exit the stack, however does not mean a failure. + """ + + +@dataclass +class RendezvousStoreInfo: + """Store address and port that can be used to bootstrap trainer distributed comms""" + + MASTER_ADDR_KEY: ClassVar[str] = "MASTER_ADDR" + MASTER_PORT_KEY: ClassVar[str] = "MASTER_PORT" + master_addr: str + master_port: int + + @staticmethod + def build( + rank: int, + store: Store, + local_addr: str | None, + server_port: int | None = None, + ) -> "RendezvousStoreInfo": + """Factory method, finds unused new port on rank0 host and addr/port info with all ranks. + + If master_addr/master_port is known (useful when sharing existing tcp store server) use the constructor. + + Args: + rank: rank of the current node + store: store to use for rendezvous + local_addr: address of the current node, if not provided will be resolved from hostname + server_port: port of the TCPStore server, when the TCPStore is shared. + """ + # TODO swap to collectives comms API + if rank == 0: + addr = local_addr or socket.getfqdn() + # When TCPStore is not shared, we fallback to get_free_port. + port = server_port or get_free_port() + store.set( + RendezvousStoreInfo.MASTER_ADDR_KEY, + addr.encode(encoding="UTF-8"), # type: ignore[arg-type] + ) + store.set( + RendezvousStoreInfo.MASTER_PORT_KEY, + str(port).encode(encoding="UTF-8"), # type: ignore[arg-type] + ) + + addr = store.get(RendezvousStoreInfo.MASTER_ADDR_KEY).decode(encoding="UTF-8") + port = int( + store.get(RendezvousStoreInfo.MASTER_PORT_KEY).decode(encoding="UTF-8") + ) + return RendezvousStoreInfo(master_addr=addr, master_port=port) + + +class RendezvousInfo: + """Holds the information about the rendezvous.""" + + def __init__( + self, + store: Store, + rank: int, + world_size: int, + bootstrap_store_info: RendezvousStoreInfo, + ): + self._store = store + self._rank = rank + self._world_size = world_size + self._bootstrap_store_info = bootstrap_store_info + + @property + def store(self) -> Store: + """Store used by torchelastic control plane""" + return self._store + + @property + def rank(self) -> int: + """Rank within a group""" + return self._rank + + @property + def world_size(self) -> int: + """Global group size""" + return self._world_size + + @property + def bootstrap_store_info(self) -> RendezvousStoreInfo | None: + """Store information that can used by trainer code to bootstrap distributed comms.""" + return self._bootstrap_store_info + + +class RendezvousHandler(ABC): + """Main rendezvous interface. + + Note: + Distributed Torch users normally **do not** need to implement their own + ``RendezvousHandler``. An implementation based on C10d Store is already + provided, and is recommended for most users. + """ + + @abstractmethod + def get_backend(self) -> str: + """Return the name of the rendezvous backend.""" + + @property + def use_agent_store(self) -> bool: + """Indicates that store reference returned by :py:meth:`next_rendezvous` can be shared with user + applications and will be available during application lifecycle. + + Rendezvous handler impl will share store details as instance of :py:class:`RendezvousStoreInfo`. + Applications as a convention use `MASTER_ADDR`/`MASTER_PORT` env variables to lookup the store. + """ + return False + + @abstractmethod + def next_rendezvous(self) -> RendezvousInfo: + """Main entry-point into the rendezvous barrier. + + Blocks until the rendezvous is complete and the current process is + included in the formed worker group, or a timeout occurs, or the + rendezvous was marked closed. + + Returns: + Instance of :py:class:`RendezvousInfo`. + + Raises: + RendezvousClosedError: + The rendezvous is closed. + RendezvousConnectionError: + The connection to the rendezvous backend has failed. + RendezvousStateError: + The rendezvous state is corrupt. + RendezvousTimeoutError: + The rendezvous did not complete on time. + """ + + @abstractmethod + def is_closed(self) -> bool: + """Check whether the rendezvous has been closed. + + A closed rendezvous means all future attempts to re-rendezvous within + same job will fail. + + ``is_closed()`` and :py:meth:`set_closed` have semantics of eventual + propagation and should not be used for synchronization. The intention is + that if at least one node decides the job is finished, it will close the + rendezvous, and other nodes will soon observe this and stop running as + well. + """ + + @abstractmethod + def set_closed(self): + """Mark the rendezvous as closed.""" + + @abstractmethod + def num_nodes_waiting(self) -> int: + """Return the number of nodes who arrived late at the rendezvous + barrier, hence were not included in the current worker group. + + Callers should periodically call this method to check whether new + nodes are waiting to join the job and if so admit them by calling + :py:meth:`next_rendezvous()` (re-rendezvous). + """ + + @abstractmethod + def get_run_id(self) -> str: + """Return the run id of the rendezvous. + + The run id is a user-defined id that uniquely identifies an instance of + a distributed application. It typically maps to a job id and is used to + allow nodes to join the correct distributed application. + """ + + @abstractmethod + def shutdown(self) -> bool: + """Close all resources that were open for the rendezvous. + + Example:: + + rdzv_handler = ... + try: + store, rank, world_size = rdzv_handler.next_rendezvous() + finally: + rdzv_handler.shutdown() + """ + + +class RendezvousParameters: + """Hold the parameters to construct a :py:class:`RendezvousHandler`. + + Args: + backend: + The name of the backend to use to handle the rendezvous. + endpoint: + The endpoint of the rendezvous, usually in form [:]. + run_id: + The id of the rendezvous. + min_nodes: + The minimum number of nodes to admit to the rendezvous. + max_nodes: + The maximum number of nodes to admit to the rendezvous. + local_addr: + The address of the local node. + **kwargs: + Additional parameters for the specified backend. + """ + + def __init__( + self, + backend: str, + endpoint: str, + run_id: str, + min_nodes: int, + max_nodes: int, + local_addr: str | None = None, + **kwargs, + ): + if not backend: + raise ValueError("The rendezvous backend name must be a non-empty string.") + + if min_nodes < 1: + raise ValueError( + f"The minimum number of rendezvous nodes ({min_nodes}) must be greater than zero." + ) + if max_nodes < min_nodes: + raise ValueError( + f"The maximum number of rendezvous nodes ({max_nodes}) must be greater than or " + f"equal to the minimum number of rendezvous nodes ({min_nodes})." + ) + + self.backend = backend + self.endpoint = endpoint + self.run_id = run_id + self.min_nodes = min_nodes + self.max_nodes = max_nodes + self.config = kwargs + self.local_addr = local_addr + + def get(self, key: str, default: Any = None) -> Any: + """Return the value for ``key`` if ``key`` exists, else ``default``.""" + return self.config.get(key, default) + + def get_as_bool(self, key: str, default: bool | None = None) -> bool | None: + """Return the value for ``key`` as a ``bool``.""" + value = self.get(key, default) + if value is None or isinstance(value, bool): + return value + if isinstance(value, int): + if value == 1: + return True + if value == 0: + return False + elif isinstance(value, str): + if value.lower() in ["1", "true", "t", "yes", "y"]: + return True + if value.lower() in ["0", "false", "f", "no", "n"]: + return False + raise ValueError( + f"The rendezvous configuration option '{key}' does not represent a valid boolean value." + ) + + def get_as_int(self, key: str, default: int | None = None) -> int | None: + """Return the value for ``key`` as an ``int``.""" + value = self.get(key, default) + if value is None: + return value + try: + return int(value) + except ValueError as e: + raise ValueError( + f"The rendezvous configuration option '{key}' does not represent a valid integer " + "value." + ) from e + + +RendezvousHandlerCreator = Callable[[RendezvousParameters], RendezvousHandler] + + +class RendezvousHandlerRegistry: + """Represent a registry of :py:class:`RendezvousHandler` backends.""" + + _registry: dict[str, RendezvousHandlerCreator] + + def __init__(self) -> None: + self._registry = {} + + def register(self, backend: str, creator: RendezvousHandlerCreator) -> None: + """Register a new rendezvous backend. + + Args: + backend: + The name of the backend. + creator: + The callback to invoke to construct the + :py:class:`RendezvousHandler`. + """ + if not backend: + raise ValueError("The rendezvous backend name must be a non-empty string.") + + current_creator: RendezvousHandlerCreator | None + try: + current_creator = self._registry[backend] + except KeyError: + current_creator = None + + if current_creator is not None and current_creator != creator: + raise ValueError( + f"The rendezvous backend '{backend}' cannot be registered with '{creator}' as it " + f"is already registered with '{current_creator}'." + ) + + self._registry[backend] = creator + + def create_handler(self, params: RendezvousParameters) -> RendezvousHandler: + """Create a new :py:class:`RendezvousHandler`.""" + try: + creator = self._registry[params.backend] + except KeyError as e: + raise ValueError( + f"The rendezvous backend '{params.backend}' is not registered. Did you forget " + f"to call `{self.register.__name__}`?" + ) from e + + handler = creator(params) + + # Do some sanity check. + if handler.get_backend() != params.backend: + raise RuntimeError( + f"The rendezvous backend '{handler.get_backend()}' does not match the requested " + f"backend '{params.backend}'." + ) + + return handler + + +# The default global registry instance used by launcher scripts to instantiate +# rendezvous handlers. +rendezvous_handler_registry = RendezvousHandlerRegistry() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py new file mode 100644 index 0000000000000000000000000000000000000000..9100fef6fb763078333ce76def4f5b39d93d5eab --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py @@ -0,0 +1,270 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import binascii +import logging +import os +import tempfile +from base64 import b64decode, b64encode +from datetime import timedelta +from typing import Any, cast + +from torch.distributed import FileStore, Store, TCPStore +from torch.distributed.elastic.events import construct_and_record_rdzv_event, NodeState + +from .api import ( + RendezvousConnectionError, + RendezvousError, + RendezvousParameters, + RendezvousStateError, +) +from .dynamic_rendezvous import RendezvousBackend, Token +from .utils import _matches_machine_hostname, parse_rendezvous_endpoint + + +logger = logging.getLogger(__name__) + +# default port for the TCP store +DEFAULT_PORT = 29400 + + +class C10dRendezvousBackend(RendezvousBackend): + """Represents a C10d-backed rendezvous backend. + + Args: + store: + The :py:class:`torch.distributed.Store` instance to use to + communicate with the C10d store. + run_id: + The run id of the rendezvous. + """ + + # See the explanation in the __init__ method. + _NULL_SENTINEL = "Y2FuaW1hZGFt" + + _store: Store + _key: str + + def __init__(self, store: Store, run_id: str) -> None: + if not run_id: + raise ValueError("The run id must be a non-empty string.") + + self._store = store + + self._key = "torch.rendezvous." + run_id + + # The read operation of a store blocks the caller until the specified + # key becomes available. This behavior makes it tricky to use a store + # as a regular key-value dictionary. + # + # As a workaround we initially set a sentinel value as the rendezvous + # state. Whenever this value gets returned we treat it as a None. + self._call_store("compare_set", self._key, "", self._NULL_SENTINEL) + + @property + def name(self) -> str: + """See base class.""" + return "c10d" + + def get_state(self) -> tuple[bytes, Token] | None: + """See base class.""" + base64_state: bytes = self._call_store("get", self._key) + + return self._decode_state(base64_state) + + def set_state( + self, state: bytes, token: Token | None = None + ) -> tuple[bytes, Token, bool] | None: + """See base class.""" + base64_state_str: str = b64encode(state).decode() + + if token: + # Shortcut if we know for sure that the token is not valid. + if not isinstance(token, bytes): + result = self.get_state() + if result is not None: + return *result, False + return None + + token = token.decode() + else: + token = self._NULL_SENTINEL + + base64_state: bytes = self._call_store( + "compare_set", self._key, token, base64_state_str + ) + + state_token_pair = self._decode_state(base64_state) + if state_token_pair is None: + return None + + new_state, new_token = state_token_pair + + # C10d Store's compare_set method does not offer an easy way to find out + # whether our write attempt was successful. As a brute-force solution we + # perform a bitwise comparison of our local state and the remote state. + return new_state, new_token, new_state == state + + def _call_store(self, store_op: str, *args, **kwargs) -> Any: + try: + return getattr(self._store, store_op)(*args, **kwargs) + except (ValueError, RuntimeError, TimeoutError) as exc: + raise RendezvousConnectionError( + "The connection to the C10d store has failed. See inner exception for details." + ) from exc + + def _decode_state(self, base64_state: bytes) -> tuple[bytes, Token] | None: + if base64_state == self._NULL_SENTINEL.encode(): + return None + + try: + state = b64decode(base64_state) + except binascii.Error as exc: + raise RendezvousStateError( + "The state object is corrupt. See inner exception for details." + ) from exc + + return state, base64_state + + +def _create_tcp_store(params: RendezvousParameters) -> TCPStore: + host, port = parse_rendezvous_endpoint(params.endpoint, default_port=DEFAULT_PORT) + + cfg_is_host = params.get_as_bool("is_host") + # If the user has explicitly specified whether our process should host + # the store, respect it. + if cfg_is_host is not None: + is_host = cfg_is_host + # Otherwise try to determine whether we are the host based on our hostname + # and IP address. + else: + is_host = _matches_machine_hostname(host) + + # The timeout + read_timeout = cast(int, params.get_as_int("read_timeout", 60)) + if read_timeout <= 0: + raise ValueError("The read timeout must be a positive integer.") + + # In specific cases we attempt to instantiate the store twice. For details + # see the explanation in the except clause below. + for is_server in [is_host, False]: + try: + store = TCPStore( + host, + port, + is_master=is_server, + multi_tenant=True, + timeout=timedelta(seconds=read_timeout), + ) + + if is_server: + msg = f"Process {os.getpid()} hosts the TCP store for the C10d rendezvous backend." + construct_and_record_rdzv_event( + run_id=params.run_id, message=msg, node_state=NodeState.INIT + ) + logger.info(msg) + + break + except (ValueError, RuntimeError, TimeoutError) as exc: + # If we heuristically inferred the value of is_host as True and our + # first attempt to instantiate the TCP store has failed, try it one + # more time with is_host set to False. As an edge case there can be + # more than one process that is part of the same rendezvous on this + # machine and only one of them will eventually host the store. + + if not is_server or cfg_is_host is not None: + raise RendezvousConnectionError( + "The connection to the C10d store has failed. See inner exception for details." + ) from exc + + return store # type: ignore[possibly-undefined] + + +def _create_file_store(params: RendezvousParameters) -> FileStore: + # If a user specifies an endpoint, we treat it as a path to a file. + if params.endpoint: + path = params.endpoint + else: + try: + # The temporary file is readable and writable only by the user of + # this process. + _, path = tempfile.mkstemp() + except OSError as exc: + raise RendezvousError( + "The file creation for C10d store has failed. See inner exception for details." + ) from exc + + try: + store = FileStore(path) + except (ValueError, RuntimeError) as exc: + raise RendezvousConnectionError( + "The connection to the C10d store has failed. See inner exception for details." + ) from exc + + return store + + +def create_backend(params: RendezvousParameters) -> tuple[C10dRendezvousBackend, Store]: + """Create a new :py:class:`C10dRendezvousBackend` from the specified parameters. + + +--------------+-----------------------------------------------------------+ + | Parameter | Description | + +==============+===========================================================+ + | store_type | The type of the C10d store. The currently supported types | + | | are "tcp" and "file" which correspond to | + | | :py:class:`torch.distributed.TCPStore` and | + | | :py:class:`torch.distributed.FileStore`, respectively. | + | | Defaults to "tcp". | + +--------------+-----------------------------------------------------------+ + | read_timeout | The read timeout, in seconds, for store operations. | + | | Defaults to 60 seconds. | + | | | + | | Note this only applies to | + | | :py:class:`torch.distributed.TCPStore`. It is not relevant| + | | to :py:class:`torch.distributed.FileStore` which does not | + | | take in timeout as a parameter. | + +--------------+-----------------------------------------------------------+ + | is_host | A boolean value indicating whether this backend instance | + | | will host the C10d store. If not specified it will be | + | | inferred heuristically by matching the hostname or the IP | + | | address of this machine against the specified rendezvous | + | | endpoint. Defaults to ``None``. | + | | | + | | Note that this configuration option only applies to | + | | :py:class:`torch.distributed.TCPStore`. In normal | + | | circumstances you can safely skip it; the only time when | + | | it is needed is if its value cannot be correctly | + | | determined (e.g. the rendezvous endpoint has a CNAME as | + | | the hostname or does not match the FQDN of the machine). | + +--------------+-----------------------------------------------------------+ + """ + # As of today we only support TCPStore and FileStore. Other store types do + # not have the required functionality (e.g. compare_set) yet. + store_type = params.get("store_type", "tcp").strip().lower() + store: Store + + try: + if store_type == "file": + store = _create_file_store(params) + elif store_type == "tcp": + store = _create_tcp_store(params) + else: + raise ValueError( + "Invalid store type given. Currently only supports file and tcp." + ) + + backend = C10dRendezvousBackend(store, params.run_id) + + except Exception as e: + construct_and_record_rdzv_event( + message=f"{type(e).__name__}: {str(e)}", + run_id=params.run_id, + node_state=NodeState.FAILED, + ) + raise + + return backend, store diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py new file mode 100644 index 0000000000000000000000000000000000000000..0e31e0946e3f501b83f117587137babcde11ead7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py @@ -0,0 +1,1458 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import inspect +import logging +import os +import pickle +import socket +import threading +import time +import weakref +from abc import ABC, abstractmethod +from collections.abc import Callable +from dataclasses import dataclass +from datetime import datetime, timedelta, timezone +from enum import Enum +from typing import Any + +import torch.distributed as dist +from torch.distributed import Store +from torch.distributed.elastic.events import construct_and_record_rdzv_event, NodeState + +from .api import ( + RendezvousClosedError, + RendezvousError, + RendezvousGracefulExitError, + RendezvousHandler, + RendezvousInfo, + RendezvousParameters, + RendezvousStateError, + RendezvousStoreInfo, + RendezvousTimeoutError, +) +from .utils import _delay, _PeriodicTimer + + +__all__ = [ + "RendezvousBackend", + "RendezvousTimeout", + "RendezvousSettings", + "DynamicRendezvousHandler", + "create_handler", +] + +logger = logging.getLogger(__name__) + + +def get_method_name(depth=2): + if len(inspect.stack()) > depth: + return inspect.stack()[depth].function + return "no_method_name" + + +Token = Any +"""Represent an opaque fencing token used by the rendezvous backend.""" + + +class RendezvousBackend(ABC): + """Represent a backend that holds the rendezvous state.""" + + @property + @abstractmethod + def name(self) -> str: + """Get the name of the backend.""" + + @abstractmethod + def get_state(self) -> tuple[bytes, Token] | None: + """Get the rendezvous state. + + Returns: + A tuple of the encoded rendezvous state and its fencing token or + ``None`` if no state is found in the backend. + + Raises: + RendezvousConnectionError: + The connection to the backend has failed. + RendezvousStateError: + The rendezvous state is corrupt. + """ + + @abstractmethod + def set_state( + self, state: bytes, token: Token | None = None + ) -> tuple[bytes, Token, bool] | None: + """Set the rendezvous state. + + The new rendezvous state is set conditionally: + + - If the specified ``token`` matches the fencing token stored in the + backend, the state will be updated. The new state will be returned + to the caller along with its fencing token. + - If the specified ``token`` does not match the fencing token stored + in the backend, the state won't be updated; instead the existing + state along with its fencing token will be returned to the caller. + - If the specified ``token`` is ``None``, the new state will be set + only if there is no existing state in the backend. Either the new + state or the existing state along with its fencing token will be + returned to the caller. + + Args: + state: + The encoded rendezvous state. + token: + An optional fencing token that was retrieved by a previous call + to :py:meth:`get_state` or ``set_state()``. + + Returns: + A tuple of the serialized rendezvous state, its fencing token, and + a boolean value indicating whether our set attempt succeeded. + + Raises: + RendezvousConnectionError: + The connection to the backend has failed. + RendezvousStateError: + The rendezvous state is corrupt. + """ + + +class RendezvousTimeout: + """Hold the timeout configuration of a rendezvous. + + Args: + join: + The time within which the rendezvous is expected to complete. + last_call: + An additional wait amount before completing the rendezvous once the + rendezvous has the minimum number of required participants. + close: + The time within which the rendezvous is expected to close after a + call to :py:meth:`RendezvousHandler.set_closed` or + :py:meth:`RendezvousHandler.shutdown`. + heartbeat: + The time within which a keep-alive heartbeat is expected to + complete. + """ + + _ZERO = timedelta(0) + + _DEFAULT_TIMEOUTS = { + "join": timedelta(seconds=600), + "last_call": timedelta(seconds=30), + "close": timedelta(seconds=30), + "heartbeat": timedelta(seconds=5), + } + + _join: timedelta + _last_call: timedelta + _close: timedelta + _heartbeat: timedelta + + def __init__( + self, + join: timedelta | None = None, + last_call: timedelta | None = None, + close: timedelta | None = None, + heartbeat: timedelta | None = None, + ) -> None: + self._set_timeouts( + join=join, last_call=last_call, close=close, heartbeat=heartbeat + ) + + @property + def join(self) -> timedelta: + """Get the join timeout.""" + return self._join + + @property + def last_call(self) -> timedelta: + """Get the last call timeout.""" + return self._last_call + + @property + def close(self) -> timedelta: + """Get the close timeout.""" + return self._close + + @property + def heartbeat(self) -> timedelta: + """Get the keep-alive heartbeat timeout.""" + return self._heartbeat + + def _set_timeouts(self, **timeouts: timedelta | None): + for name, timeout in timeouts.items(): + if timeout is None: + timeout = self._DEFAULT_TIMEOUTS[name] + if timeout <= self._ZERO: + raise ValueError(f"The {name} timeout ({timeout}) must be positive.") + setattr(self, "_" + name, timeout) + + +@dataclass(repr=False, eq=False, frozen=True) +class RendezvousSettings: + """Hold the settings of the rendezvous. + + Attributes: + run_id: + The run id of the rendezvous. + min_nodes: + The minimum number of nodes to admit to the rendezvous. + max_nodes: + The maximum number of nodes to admit to the rendezvous. + timeout: + The timeout configuration of the rendezvous. + keep_alive_interval: + The amount of time a node waits before sending a heartbeat to keep + it alive in the rendezvous. + keep_alive_max_attempt: + The maximum number of failed heartbeat attempts after which a node + is considered dead. + """ + + run_id: str + min_nodes: int + max_nodes: int + timeout: RendezvousTimeout + keep_alive_interval: timedelta + keep_alive_max_attempt: int + + +@dataclass(eq=True, order=True, frozen=True) +class _NodeDesc: + """Describe a node in the rendezvous. + + Attributes: + addr: + The FQDN of the node or user specified local node address. + pid: + The id of the process in which the rendezvous handler runs. + local_id: + A process-wide unique id. + """ + + addr: str + pid: int + local_id: int + + def __repr__(self) -> str: + return f"{self.addr}_{self.pid}_{self.local_id}" + + +class _NodeDescGenerator: + """Generate node descriptors. + + A node descriptor is a combination of an FQDN, a process id, and an auto- + incremented integer that uniquely identifies a node in the rendezvous. + """ + + _lock: threading.Lock + _local_id: int + + def __init__(self) -> None: + self._lock = threading.Lock() + + # An integer that is incremented with each call to generate(). + self._local_id = 0 + + def generate(self, local_addr: str | None = None) -> _NodeDesc: + # This method can be called by multiple threads concurrently; therefore, + # we must increment the integer atomically. + with self._lock: + local_id = self._local_id + + self._local_id += 1 + + return _NodeDesc(local_addr or socket.getfqdn(), os.getpid(), local_id) + + +class _RendezvousState: + """Hold the state of a rendezvous. + + Attributes: + round: + The current round of the rendezvous. + complete: + A boolean value indicating whether the current round of the + rendezvous is complete. + deadline: + The time at which the current round of the rendezvous will be + considered complete if it is still waiting for nodes to join. + closed: + A boolean value indicating whether the rendezvous is closed. + participants: + A dictionary of the participants and their corresponding ranks. + wait_list: + A set of nodes that are waiting to participate in the next round of + the rendezvous. + redundancy_list: + A set of nodes that are redundant in the current round and can join + the next rendezvous without triggering re-rendezvous. + last_heartbeats: + A dictionary containing each node's last heartbeat time. + """ + + round: int + complete: bool + deadline: datetime | None + closed: bool + participants: dict[_NodeDesc, int] + wait_list: set[_NodeDesc] + redundancy_list: set[_NodeDesc] + last_heartbeats: dict[_NodeDesc, datetime] + + def __init__(self) -> None: + self.round = 0 + self.complete = False + self.deadline = None + self.closed = False + self.participants = {} + self.wait_list = set() + self.redundancy_list = set() + self.last_heartbeats = {} + + +def _remove_participant_epilogue( + state: _RendezvousState, settings: RendezvousSettings +) -> None: + if state.complete: + # If we do not have any participants left, move to the next round. + if not state.participants: + msg = "No participants left in the rendezvous, marking rendezvous as incomplete" + logger.debug(msg) + state.complete = False + + state.round += 1 + else: + if len(state.participants) < settings.min_nodes: + msg = ( + f"Number of participants {len(state.participants)} less than " + f"min_nodes {settings.min_nodes}, clearing deadline in state" + ) + logger.debug(msg) + state.deadline = None + + +class _RendezvousStateHolder(ABC): + """Hold the shared rendezvous state synced with other nodes.""" + + @property + @abstractmethod + def state(self) -> _RendezvousState: + """Get the local state.""" + + @abstractmethod + def sync(self) -> bool | None: + """Read or writes the latest state. + + Returns: + A boolean value indicating whether the local state, in case marked + as dirty, was successfully synced with other nodes. + """ + + @abstractmethod + def mark_dirty(self) -> None: + """Mark the local state as dirty.""" + + +class _BackendRendezvousStateHolder(_RendezvousStateHolder): + """Hold the rendezvous state synced with other nodes via a backend. + + Args: + backend: + The rendezvous backend to use. + settings: + The rendezvous settings. + cache_duration: + The amount of time, in seconds, to cache the last rendezvous state + before requesting it from the backend again. + """ + + _backend: RendezvousBackend + _state: _RendezvousState + _settings: RendezvousSettings + _cache_duration: int + _token: Token + _dirty: bool + _last_sync_time: float + _dead_nodes: list[_NodeDesc] + + def __init__( + self, + backend: RendezvousBackend, + settings: RendezvousSettings, + cache_duration: int = 1, + ) -> None: + self._backend = backend + self._state = _RendezvousState() + self._settings = settings + self._cache_duration = cache_duration + self._token = None + self._dirty = False + self._last_sync_time = -1 + self._dead_nodes = [] + + def _record(self, message: str, node_state: NodeState = NodeState.RUNNING): + construct_and_record_rdzv_event( + name=f"{self.__class__.__name__}.{get_method_name()}", + run_id=self._settings.run_id, + message=message, + node_state=node_state, + ) + + @property + def state(self) -> _RendezvousState: + """See base class.""" + return self._state + + def sync(self) -> bool | None: + """See base class.""" + state_bits: bytes | None = None + + token = None + + has_set: bool | None + + if self._dirty: + has_set = False + + state_bits = pickle.dumps(self._state) + + set_response = self._backend.set_state(state_bits, self._token) + if set_response is not None: + state_bits, token, has_set = set_response + else: + has_set = None + + if self._cache_duration > 0: + # Avoid overloading the backend if we are asked to retrieve the + # state repeatedly. Try to serve the cached state. + if self._last_sync_time >= max( + time.monotonic() - self._cache_duration, 0 + ): + return None + + get_response = self._backend.get_state() + if get_response is not None: + state_bits, token = get_response + + if state_bits is not None: + try: + self._state = pickle.loads(state_bits) + except pickle.PickleError as exc: + raise RendezvousStateError( + "The rendezvous state is corrupt. See inner exception for details." + ) from exc + else: + self._state = _RendezvousState() + + if has_set and self._dead_nodes and logger.isEnabledFor(logging.DEBUG): + node_list = ", ".join(f"'{dead_node}'" for dead_node in self._dead_nodes) + + msg = ( + f"As part of the sync operation the node(s) {node_list} have been removed from the " + f"rendezvous '{self._settings.run_id}' since they had no heartbeat." + ) + self._record(message=msg) + logger.debug(msg) + + self._token = token + + self._dirty = False + + self._last_sync_time = time.monotonic() + + self._sanitize() + + return has_set + + def _sanitize(self) -> None: + state = self._state + + expire_time = datetime.now(timezone.utc) - ( + self._settings.keep_alive_interval * self._settings.keep_alive_max_attempt + ) + + # Filter out the dead nodes. + self._dead_nodes = [ + node + for node, last_heartbeat in state.last_heartbeats.items() + if last_heartbeat < expire_time + ] + + participant_removed = False + + for dead_node in self._dead_nodes: + msg = f"Detected dead node '{dead_node}', removing it from the rendezvous" + logger.debug(msg) + del state.last_heartbeats[dead_node] + + try: + del state.participants[dead_node] + + participant_removed = True + except KeyError: + pass + + try: + state.wait_list.remove(dead_node) + except KeyError: + pass + + try: + state.redundancy_list.remove(dead_node) + except KeyError: + pass + + if participant_removed: + # Common epilogue shared with the _remove_from_participants() + # function of _DistributedRendezvousOpExecutor. + _remove_participant_epilogue(state, self._settings) + + def mark_dirty(self) -> None: + """See base class. + + If the local rendezvous state is dirty, the next sync call will try to + write the changes back to the backend. However this attempt might fail + if another node, which had the same state, also made changes and wrote + them before us. + """ + self._dirty = True + + +class _Action(Enum): + """Specifies the possible actions based on the state of the rendezvous.""" + + KEEP_ALIVE = 1 + ADD_TO_PARTICIPANTS = 2 + ADD_TO_WAIT_LIST = 3 + ADD_TO_REDUNDANCY_LIST = 4 + REMOVE_FROM_PARTICIPANTS = 5 + REMOVE_FROM_WAIT_LIST = 6 + REMOVE_FROM_REDUNDANCY_LIST = 7 + MARK_RENDEZVOUS_COMPLETE = 8 + MARK_RENDEZVOUS_CLOSED = 9 + SYNC = 10 + ERROR_CLOSED = 11 + ERROR_TIMEOUT = 12 + FINISH = 13 + + +class _RendezvousContext: + """Holds the context of the rendezvous. + + Attributes: + node: + The node descriptor associated with the current rendezvous handler + instance. + state: + The current state of the rendezvous. + settings: + The rendezvous settings. + """ + + node: _NodeDesc + state: _RendezvousState + settings: RendezvousSettings + + def __init__( + self, node: _NodeDesc, state: _RendezvousState, settings: RendezvousSettings + ) -> None: + self.node = node + self.state = state + self.settings = settings + + +class _RendezvousOpExecutor(ABC): + """Execute rendezvous operations.""" + + @abstractmethod + def run( + self, + state_handler: Callable[[_RendezvousContext, float], _Action], + deadline: float, + update_deadline: Callable[[timedelta], float] | None = None, + ) -> None: + """Execute a rendezvous operation. + + An operation is run inside a state machine and is expected to transition + the rendezvous from one state to another. + + Args: + state_handler: + A callable that is expected to return the next state transition + action based on the current state of the rendezvous. + deadline: + The time, in seconds, at which the operation will be considered + timed-out. + update_deadline: + Function to generate a new operation deadline if the current + node may participate in the next rendezvous. + """ + + +class _DistributedRendezvousOpExecutor(_RendezvousOpExecutor): + """Execute rendezvous operations using a shared state. + + Args: + node: + The node descriptor associated with the current rendezvous handler + instance. + state_holder: + The ``RendezvousStateHolder`` to use to sync the rendezvous state + with other nodes. + settings: + The rendezvous settings. + """ + + _node: _NodeDesc + _state: _RendezvousState + _state_holder: _RendezvousStateHolder + _settings: RendezvousSettings + + def __init__( + self, + node: _NodeDesc, + state_holder: _RendezvousStateHolder, + settings: RendezvousSettings, + ) -> None: + self._node = node + self._state_holder = state_holder + self._settings = settings + + def _record(self, message: str, node_state: NodeState = NodeState.RUNNING) -> None: + construct_and_record_rdzv_event( + name=f"{self.__class__.__name__}.{get_method_name()}", + run_id=self._settings.run_id, + message=message, + node_state=node_state, + hostname=self._node.addr, + pid=self._node.pid, + local_id=self._node.local_id, + ) + + def run( + self, + state_handler: Callable[[_RendezvousContext, float], _Action], + deadline: float, + update_deadline: Callable[[timedelta], float] | None = None, + ) -> None: + """See base class.""" + action = None + while action != _Action.FINISH: + # Reads or writes the latest rendezvous state shared by all nodes in + # the rendezvous. Note that our local changes might get overridden + # by another node if that node synced its changes before us. + has_set = self._state_holder.sync() + if has_set is not None: + if has_set: + msg = ( + f"The node '{self._node}' has successfully synced its local changes with " + f"other nodes in the rendezvous '{self._settings.run_id}'." + ) + else: + msg = ( + f"The node '{self._node}' has a stale state and failed to sync its local " + f"changes with other nodes in the rendezvous '{self._settings.run_id}'." + ) + + self._record(message=msg) + logger.debug(msg) + + self._state = self._state_holder.state + + ctx = _RendezvousContext(self._node, self._state, self._settings) + + # Determine the next action to take based on the current state of + # the rendezvous. + action = state_handler(ctx, deadline) + + if action == _Action.FINISH: + continue + + if action == _Action.ERROR_CLOSED: + raise RendezvousClosedError + + if action == _Action.ERROR_TIMEOUT: + raise RendezvousTimeoutError + + if action == _Action.SYNC: + # Delay the execution by one second to avoid overloading the + # backend if we are asked to poll for state changes. + _delay(seconds=1) + else: + if action == _Action.KEEP_ALIVE: + self._keep_alive() + elif action == _Action.ADD_TO_PARTICIPANTS: + self._add_to_participants() + elif action == _Action.ADD_TO_WAIT_LIST: + self._add_to_wait_list() + elif action == _Action.ADD_TO_REDUNDANCY_LIST: + self._add_to_redundancy_list() + elif action == _Action.REMOVE_FROM_PARTICIPANTS: + self._remove_from_participants() + elif action == _Action.REMOVE_FROM_WAIT_LIST: + self._remove_from_wait_list() + elif action == _Action.REMOVE_FROM_REDUNDANCY_LIST: + self._remove_from_redundancy_list() + # update deadline since the node may participate in rendezvous process + if update_deadline: + deadline = update_deadline(self._settings.timeout.join) + elif action == _Action.MARK_RENDEZVOUS_COMPLETE: + self._mark_rendezvous_complete() + elif action == _Action.MARK_RENDEZVOUS_CLOSED: + self._mark_rendezvous_closed() + + # Attempt to sync our changes back to other nodes. + self._state_holder.mark_dirty() + + def _keep_alive(self) -> None: + msg = ( + f"The node '{self._node}' updated its keep-alive heartbeat time for the rendezvous " + f"'{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + self._state.last_heartbeats[self._node] = datetime.now(timezone.utc) + + def _add_to_participants(self) -> None: + msg = ( + f"The node '{self._node}' added itself to the participants of round " + f"{self._state.round} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + state = self._state + + try: + state.wait_list.remove(self._node) + except KeyError: + pass + + # The ranks of the participants will be set once the rendezvous is + # complete. + state.participants[self._node] = 0 + + self._keep_alive() + + if len(state.participants) == self._settings.min_nodes: + state.deadline = ( + datetime.now(timezone.utc) + self._settings.timeout.last_call + ) + + if len(state.participants) == self._settings.max_nodes: + self._mark_rendezvous_complete() + + def _add_to_wait_list(self) -> None: + msg = ( + f"The node '{self._node}' added itself to the wait list of round " + f"{self._state.round + 1} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + if self._node in self._state.redundancy_list: + self._state.redundancy_list.remove(self._node) + self._state.wait_list.add(self._node) + + self._keep_alive() + + def _add_to_redundancy_list(self) -> None: + msg = ( + f"The node '{self._node}' added itself to the redundancy list of round " + f"{self._state.round + 1} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + self._state.redundancy_list.add(self._node) + + self._keep_alive() + + def _remove_from_participants(self) -> None: + msg = ( + f"The node '{self._node}' removed itself from the participants of round " + f"{self._state.round} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + state = self._state + + del state.participants[self._node] + + del state.last_heartbeats[self._node] + + # Common epilogue shared with the sanitizer() function of + # _BackendRendezvousStateHolder. + _remove_participant_epilogue(state, self._settings) + + def _remove_from_wait_list(self) -> None: + msg = ( + f"The node '{self._node}' removed itself from the wait list of round " + f"{self._state.round + 1} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + self._state.wait_list.remove(self._node) + + del self._state.last_heartbeats[self._node] + + def _remove_from_redundancy_list(self) -> None: + msg = ( + f"The node '{self._node}' removed itself from the redundant list of round " + f"{self._state.round + 1} of the rendezvous '{self._settings.run_id}'. Pending sync." + ) + self._record(message=msg) + logger.debug(msg) + + self._state.redundancy_list.remove(self._node) + + del self._state.last_heartbeats[self._node] + + def _mark_rendezvous_complete(self) -> None: + msg = ( + f"The node '{self._node}' marked round {self._state.round} of the rendezvous " + f"'{self._settings.run_id}' as complete. Pending sync." + ) + self._record(message=msg, node_state=NodeState.SUCCEEDED) + logger.debug(msg) + + state = self._state + + state.complete = True + state.deadline = None + + # Assign the ranks. + for rank, node in enumerate(sorted(state.participants)): + state.participants[node] = rank + + def _mark_rendezvous_closed(self) -> None: + msg = ( + f"The node '{self._node}' marked the rendezvous '{self._settings.run_id}' as closed. " + "Pending sync." + ) + self._record(message=msg, node_state=NodeState.SUCCEEDED) + logger.debug(msg) + + self._state.closed = True + + +def _should_keep_alive(ctx: _RendezvousContext) -> bool: + """Determine whether a keep-alive heartbeat should be sent.""" + try: + last_heartbeat = ctx.state.last_heartbeats[ctx.node] + except KeyError: + return False + + return ( + last_heartbeat <= datetime.now(timezone.utc) - ctx.settings.keep_alive_interval + ) + + +class _RendezvousExitOp: + """Represent a rendezvous exit operation.""" + + def __call__(self, ctx: _RendezvousContext, deadline: float) -> _Action: + if ctx.node in ctx.state.participants: + if time.monotonic() > deadline: + return _Action.ERROR_TIMEOUT + return _Action.REMOVE_FROM_PARTICIPANTS + return _Action.FINISH + + +class _RendezvousJoinOp: + """Represent a rendezvous join operation.""" + + def __call__(self, ctx: _RendezvousContext, deadline: float) -> _Action: + state = ctx.state + + # A closed rendezvous means that it no longer accepts new nodes. + if state.closed: + if ctx.node in state.redundancy_list: + msg = f"The rendezvous '{ctx.settings.run_id}' is closed, terminating pending rendezvous." + raise RendezvousGracefulExitError(msg) + return _Action.ERROR_CLOSED + + if ctx.node in state.redundancy_list: + msg = f"The node {ctx.node} is in redundancy list" + logger.debug(msg) + # don't apply the timeout logic here, since we want to allow the node to rejoin + if len(state.participants) == ctx.settings.max_nodes: + if _should_keep_alive(ctx): + return _Action.KEEP_ALIVE + else: + return _Action.SYNC + else: + # transition to waiting state that will respect timeouts. + msg = f"The node {ctx.node} is removed from redundancy list" + logger.debug(msg) + return _Action.REMOVE_FROM_REDUNDANCY_LIST + + is_participant = ctx.node in state.participants + + # If we are part of the rendezvous and it is already complete there is + # no further action to take. + if state.complete and is_participant: + return _Action.FINISH + + now = time.monotonic() + if now > deadline: + rollback_period = 5 # 5 seconds + + # If we still have time to rollback (a short period on top of the + # operation deadline), try to remove ourself from the rendezvous. + # It is okay if we can't though as our keep-alive will eventually + # expire. + if now <= deadline + rollback_period: + # If we are part of the rendezvous, it means we couldn't find + # enough participants to complete it on time. + if is_participant: + return _Action.REMOVE_FROM_PARTICIPANTS + # If we are in the wait list, it means we couldn't wait till the + # next round of the rendezvous. + if ctx.node in state.wait_list: + return _Action.REMOVE_FROM_WAIT_LIST + return _Action.ERROR_TIMEOUT + + if state.complete: + # If we are here, it means we are not part of the rendezvous. In + # case the rendezvous has capacity for additional participants add + # ourself to the wait list for the next round. + if len(state.participants) < ctx.settings.max_nodes: + if ctx.node not in state.wait_list: + return _Action.ADD_TO_WAIT_LIST + elif len(state.participants) >= ctx.settings.max_nodes: + if ( + ctx.node not in state.redundancy_list + and ctx.node not in state.wait_list + ): + return _Action.ADD_TO_REDUNDANCY_LIST + elif is_participant: + # If the rendezvous has enough number of participants including us, + # check whether we have passed the rendezvous deadline. If yes, + # complete it. + if ( + len(state.participants) >= ctx.settings.min_nodes + and len(state.participants) <= ctx.settings.max_nodes + and state.deadline is not None + ): + if state.deadline < datetime.now(timezone.utc): + msg = ( + f"The node '{ctx.node}' marking the rendezvous complete, " + f"quorum established within deadline" + ) + logger.debug(msg) + return _Action.MARK_RENDEZVOUS_COMPLETE + else: + msg = f"The node '{ctx.node}' can't complete rendezvous: deadline reached" + logger.debug(msg) + else: + msg = f"The node '{ctx.node}' can't complete rendezvous: not enough participants" + logger.debug(msg) + else: + # The rendezvous is not complete yet and we are not part of it. Try + # to join. + return _Action.ADD_TO_PARTICIPANTS + + if _should_keep_alive(ctx): + return _Action.KEEP_ALIVE + + # At this point either the rendezvous is not complete, but we are part + # of it, which means we have to wait for other participants to join; or + # the rendezvous is complete, but we are not part of it, which means we + # have to wait for the next round. + return _Action.SYNC + + +class _RendezvousCloseOp: + """Represent a rendezvous close operation.""" + + def __call__(self, ctx: _RendezvousContext, deadline: float) -> _Action: + if ctx.state.closed: + return _Action.FINISH + if time.monotonic() > deadline: + return _Action.ERROR_TIMEOUT + return _Action.MARK_RENDEZVOUS_CLOSED + + +class _RendezvousKeepAliveOp: + """Represent a rendezvous keep-alive update operation.""" + + def __call__(self, ctx: _RendezvousContext, deadline: float) -> _Action: + if _should_keep_alive(ctx): + if time.monotonic() > deadline: + return _Action.ERROR_TIMEOUT + return _Action.KEEP_ALIVE + return _Action.FINISH + + +class DynamicRendezvousHandler(RendezvousHandler): + """Represent a handler that sets up a rendezvous among a set of nodes.""" + + # Static + _node_desc_generator = _NodeDescGenerator() + + _this_node: _NodeDesc + _settings: RendezvousSettings + _backend_name: str + _store: Store + _state_holder: _RendezvousStateHolder + _op_executor: _RendezvousOpExecutor + _heartbeat_lock: threading.Lock + _keep_alive_timer: _PeriodicTimer | None + + @classmethod + def from_backend( + cls, + run_id: str, + store: Store, + backend: RendezvousBackend, + min_nodes: int, + max_nodes: int, + local_addr: str | None = None, + timeout: RendezvousTimeout | None = None, + keep_alive_interval: int = 5, + keep_alive_max_attempt: int = 3, + ): + """Create a new :py:class:`DynamicRendezvousHandler`. + + Args: + run_id: + The run id of the rendezvous. + store: + The C10d store to return as part of the rendezvous. + backend: + The backend to use to hold the rendezvous state. + min_nodes: + The minimum number of nodes to admit to the rendezvous. + max_nodes: + The maximum number of nodes to admit to the rendezvous. + local_addr: + The local node address. + timeout: + The timeout configuration of the rendezvous. + keep_alive_interval: + The amount of time a node waits before sending a heartbeat to keep + it alive in the rendezvous. + keep_alive_max_attempt: + The maximum number of failed heartbeat attempts after which a node + is considered dead. + """ + # We associate each handler instance with a unique node descriptor. + node = cls._node_desc_generator.generate(local_addr) + + settings = RendezvousSettings( + run_id, + min_nodes, + max_nodes, + timeout or RendezvousTimeout(), + keep_alive_interval=timedelta(seconds=keep_alive_interval), + keep_alive_max_attempt=keep_alive_max_attempt, + ) + + state_holder = _BackendRendezvousStateHolder(backend, settings) + + return cls(node, settings, backend.name, store, state_holder) + + def __init__( + self, + node: _NodeDesc, + settings: RendezvousSettings, + backend_name: str, + store: Store, + state_holder: _RendezvousStateHolder, + ) -> None: + if not settings.run_id: + raise ValueError("The run id must be a non-empty string.") + + if settings.min_nodes < 1: + raise ValueError( + f"The minimum number of nodes ({settings.min_nodes}) must be greater than zero." + ) + + if settings.max_nodes < settings.min_nodes: + raise ValueError( + f"The maximum number of nodes ({settings.max_nodes}) must be greater than or equal " + f"to the minimum number of nodes ({settings.min_nodes})." + ) + + self._this_node = node + + self._settings = settings + + self._backend_name = backend_name + + self._store = store + + self._state_holder = state_holder + + self._op_executor = _DistributedRendezvousOpExecutor( + self._this_node, self._state_holder, self._settings + ) + + self._heartbeat_lock = threading.Lock() + + self._keep_alive_timer = None + + # Cached shared store server reference + self._shared_tcp_store_server: dist.Store | None = None + + self._bootstrap_store_info: RendezvousStoreInfo | None = None + + def _record( + self, + message: str, + node_state: NodeState = NodeState.RUNNING, + rank: int | None = None, + ) -> None: + construct_and_record_rdzv_event( + name=f"{self.__class__.__name__}.{get_method_name()}", + run_id=self._settings.run_id, + message=message, + node_state=node_state, + hostname=self._this_node.addr, + pid=self._this_node.pid, + local_id=self._this_node.local_id, + rank=rank, + ) + + def _create_tcp_store_server(self, master_addr, master_port) -> dist.TCPStore: + return dist.TCPStore( + host_name=master_addr, + port=master_port, + is_master=True, + multi_tenant=True, + ) + + @property + def settings(self) -> RendezvousSettings: + """Get the settings of the rendezvous.""" + return self._settings + + def get_backend(self) -> str: + """See base class.""" + return self._backend_name + + @property + def use_agent_store(self) -> bool: + """See base class.""" + return os.getenv("TORCH_DISABLE_SHARE_RDZV_TCP_STORE", "0") != "1" + + def next_rendezvous(self) -> RendezvousInfo: + """See base class.""" + msg = ( + f"The node '{self._this_node}' attempts to join the next round of the rendezvous " + f"'{self._settings.run_id}'." + ) + self._record(message=msg) + logger.info(msg) + + try: + self._stop_heartbeats() + + # Delay the execution for a small random amount of time if this is our + # first run. This will slightly skew the rendezvous attempts across the + # nodes and reduce the load on the backend. + if self._state_holder.state.round == 0: + _delay(seconds=(0, 0.3)) + + exit_op = _RendezvousExitOp() + join_op = _RendezvousJoinOp() + + deadline = self._get_deadline(self._settings.timeout.join) + self._op_executor.run(exit_op, deadline) + self._op_executor.run(join_op, deadline, self._get_deadline) + + self._start_heartbeats() + + rank, world_size = self._get_world() + store = self._get_store() + + except Exception as e: + self._record( + message=f"{type(e).__name__}: {str(e)}", + node_state=NodeState.FAILED, + ) + raise + + msg = ( + f"The node '{self._this_node}' has joined round {self._state_holder.state.round} of " + f"the rendezvous '{self._settings.run_id}' as rank {rank} in a world of size " + f"{world_size}." + ) + self._record(message=msg, rank=rank) + logger.info(msg) + + # opt-out option of TCPStore sharing + if os.getenv("TORCH_DISABLE_SHARE_RDZV_TCP_STORE", "0") == "1": + bootstrap_store_info = RendezvousStoreInfo.build( + rank, store, local_addr=self._this_node.addr + ) + return RendezvousInfo( + store, + rank, + world_size, + bootstrap_store_info, + ) + + # This will only be hit when TCPStore sharing is enabled. + # Rebuild bootstrap store info when this is the first rendezvous or + # when the node became rank 0 but hasn't created a TCP store server + # yet (rank changed between rounds due to elastic membership changes). + if self._bootstrap_store_info is None or ( + rank == 0 and self._shared_tcp_store_server is None + ): + server_port = 0 + if rank == 0: + self._shared_tcp_store_server = self._create_tcp_store_server( + self._this_node.addr, server_port + ) + server_port = self._shared_tcp_store_server.port + self._bootstrap_store_info = RendezvousStoreInfo.build( + rank, + store, + local_addr=self._this_node.addr, + server_port=server_port, # For non-0 rank, this is a no-op + ) + + if self._bootstrap_store_info is None: + raise AssertionError + if rank == 0: + if self._shared_tcp_store_server is None: + raise AssertionError + + return RendezvousInfo( + store, + rank, + world_size, + self._bootstrap_store_info, # type: ignore[assignment] + ) + + def is_closed(self) -> bool: + """See base class.""" + try: + with self._heartbeat_lock: + self._state_holder.sync() + + return self._state_holder.state.closed + + except Exception as e: + self._record( + message=f"{type(e).__name__}: {str(e)}", + node_state=NodeState.FAILED, + ) + raise + + def set_closed(self) -> None: + """See base class.""" + try: + with self._heartbeat_lock: + self._close() + except Exception as e: + self._record( + message=f"{type(e).__name__}: {str(e)}", + node_state=NodeState.FAILED, + ) + raise + + def num_nodes_waiting(self) -> int: + """See base class.""" + try: + with self._heartbeat_lock: + self._state_holder.sync() + + return len(self._state_holder.state.wait_list) + + except Exception as e: + self._record( + message=f"{type(e).__name__}: {str(e)}", + node_state=NodeState.FAILED, + ) + raise + + def get_run_id(self) -> str: + """See base class.""" + return self._settings.run_id + + def shutdown(self) -> bool: + """See base class.""" + self._stop_heartbeats() + + try: + self._close() + + return True + except RendezvousError as ex: + msg = ( + f"The node '{self._this_node}' has failed to shutdown the rendezvous " + f"'{self._settings.run_id}' due to an error of type {type(ex).__name__}." + ) + self._record(message=msg, node_state=NodeState.FAILED) + logger.warning(msg) + + return False + except Exception as e: + self._record( + message=f"{type(e).__name__}: {str(e)}", + node_state=NodeState.FAILED, + ) + raise + + def _close(self) -> None: + op = _RendezvousCloseOp() + + deadline = self._get_deadline(self._settings.timeout.close) + + self._op_executor.run(op, deadline) + + msg = f"The node '{self._this_node}' has closed the rendezvous '{self._settings.run_id}'." + self._record(message=msg, node_state=NodeState.SUCCEEDED) + logger.info(msg) + + @staticmethod + def _keep_alive_weak(weak_self) -> None: + self = weak_self() + if self is not None: + self._keep_alive() + + def _keep_alive(self) -> None: + with self._heartbeat_lock: + op = _RendezvousKeepAliveOp() + + deadline = self._get_deadline(self._settings.timeout.heartbeat) + + try: + self._op_executor.run(op, deadline) + + msg = ( + f"The node '{self._this_node}' has sent a keep-alive heartbeat to the rendezvous " + f"'{self._settings.run_id}'." + ) + self._record(message=msg) + logger.debug(msg) + except RendezvousError as ex: + msg = ( + f"The node '{self._this_node}' has failed to send a keep-alive heartbeat to the " + f"rendezvous '{self._settings.run_id}' due to an error of type {type(ex).__name__}." + ) + self._record(message=msg, node_state=NodeState.FAILED) + logger.warning(msg) + + def _start_heartbeats(self) -> None: + self._keep_alive_timer = _PeriodicTimer( + self._settings.keep_alive_interval, self._keep_alive_weak, weakref.ref(self) + ) + + self._keep_alive_timer.set_name( + f"RendezvousKeepAliveTimer_{self._this_node.local_id}" + ) + + self._keep_alive_timer.start() + + def _stop_heartbeats(self) -> None: + if self._keep_alive_timer is None: + return + + self._keep_alive_timer.cancel() + + def _get_world(self) -> tuple[int, int]: + state = self._state_holder.state + + return state.participants[self._this_node], len(state.participants) + + def _wrap_store(self, store: Store) -> Store: + key_prefix = ( + f"torch.rendezvous.{self._settings.run_id}.{self._state_holder.state.round}" + ) + + return dist.PrefixStore(key_prefix, store) + + def _get_store(self) -> Store: + return self._wrap_store(self._store) + + def _get_deadline(self, timeout: timedelta) -> float: + return time.monotonic() + timeout.total_seconds() + + +def _get_timeout(params: RendezvousParameters, key: str) -> timedelta | None: + timeout = params.get_as_int(key + "_timeout") + if timeout is None: + return None + return timedelta(seconds=timeout) + + +def create_handler( + store: Store, backend: RendezvousBackend, params: RendezvousParameters +) -> DynamicRendezvousHandler: + """Create a new :py:class:`DynamicRendezvousHandler` from the specified parameters. + + Args: + store: + The C10d store to return as part of the rendezvous. + backend: + The backend to use to hold the rendezvous state. + + +-------------------+------------------------------------------------------+ + | Parameter | Description | + +===================+======================================================+ + | join_timeout | The total time, in seconds, within which the | + | | rendezvous is expected to complete. Defaults to 600 | + | | seconds. | + +-------------------+------------------------------------------------------+ + | last_call_timeout | An additional wait amount, in seconds, before | + | | completing the rendezvous once the minimum number of | + | | nodes has been reached. Defaults to 30 seconds. | + +-------------------+------------------------------------------------------+ + | close_timeout | The time, in seconds, within which the rendezvous is | + | | expected to close after a call to | + | | :py:meth:`RendezvousHandler.set_closed` or | + | | :py:meth:`RendezvousHandler.shutdown`. Defaults to | + | | 30 seconds. | + +-------------------+------------------------------------------------------+ + | heartbeat | The time, in seconds, within which a keep-alive | + | | heartbeat is expected to complete | + +-------------------+------------------------------------------------------+ + """ + try: + timeout = RendezvousTimeout( + _get_timeout(params, "join"), + _get_timeout(params, "last_call"), + _get_timeout(params, "close"), + _get_timeout(params, "heartbeat"), + ) + keep_alive_interval = params.get_as_int("keep_alive_interval", 5) + if keep_alive_interval is None: + raise TypeError( + "You passed 'keep_alive_interval=None' as a rendezvous configuration option" + ) + keep_alive_max_attempt = params.get_as_int("keep_alive_max_attempt", 3) + if keep_alive_max_attempt is None: + raise TypeError( + "You passed 'keep_alive_max_attempt=None' as a rendezvous configuration option" + ) + + return DynamicRendezvousHandler.from_backend( + params.run_id, + store, + backend, + params.min_nodes, + params.max_nodes, + params.local_addr, + timeout, + keep_alive_interval=keep_alive_interval, + keep_alive_max_attempt=keep_alive_max_attempt, + ) + except Exception as e: + construct_and_record_rdzv_event( + message=f"{type(e).__name__}: {str(e)}", + run_id=params.run_id, + node_state=NodeState.FAILED, + ) + raise diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous.py new file mode 100644 index 0000000000000000000000000000000000000000..5398bdf19447f145a2e2ed7d4ba57c064c7e3328 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous.py @@ -0,0 +1,1081 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import json +import logging +import sys +import threading +import time + + +try: + import etcd # type: ignore[import] +except ModuleNotFoundError: + from . import _etcd_stub as etcd + +from torch.distributed.elastic.rendezvous import ( + RendezvousClosedError, + RendezvousError, + RendezvousHandler, + RendezvousInfo, + RendezvousParameters, + RendezvousStoreInfo, + RendezvousTimeoutError, +) + +from .etcd_store import cas_delay, EtcdStore +from .utils import parse_rendezvous_endpoint + + +__all__ = [ + "EtcdRendezvousRetryableFailure", + "EtcdRendezvousRetryImmediately", + "EtcdRendezvousHandler", + "EtcdRendezvous", + "create_rdzv_handler", +] + +_log_fmt = logging.Formatter("%(levelname)s %(asctime)s %(message)s") +_log_handler = logging.StreamHandler(sys.stderr) +_log_handler.setFormatter(_log_fmt) + +logger = logging.getLogger(__name__) +logger.propagate = False +logger.setLevel(logging.INFO) +logger.addHandler(_log_handler) + + +# Retryable failure exception means we were too late to make +# a desired state transition (e.g. because of a race condition), +# and should now restart from the beginning. +# A small delay is recommended to avoid spamming Etcd. +class EtcdRendezvousRetryableFailure(Exception): + pass + + +# Similar to retryable failure, but the new state we observed suggests we +# can re-try immediately, i.e. without a need for "safety delay". +class EtcdRendezvousRetryImmediately(Exception): + pass + + +# Default timeout for the rendezvous. +_DEFAULT_TIMEOUT: int = 600 # 10 minutes + +# Additional waiting time after reaching the minimum number of nodes +# in case the rendezvous is elastic (min != max). +_DEFAULT_LAST_CALL_TIMEOUT: int = 30 # 30 seconds + +# Various constants used internally in EtcdRendezvous +CONST_ETCD_SETUP_TTL = 5 +CONST_ETCD_FROZEN_TTL = 10 +CONST_ETCD_JOINABLE_EPHEMERAL_TTL = 10 + +# Ephemeral node TTL for worker's keep-alive key: +CONST_WORKER_KEEPALIVE_TTL = 10 + +# TTL for the ephemeral run_id-specific directory. All rendezvous state data +# for a specific run_id (job instance) is contained within directory. +# Its only role is to clean-up rendezvous data from old runs (for the case when +# etcd server is persistent), and has no affect on correctness, but should be +# larger than any timeouts that a worker process is expected to survive: +CONST_RUNID_SUBROOT_TTL = 7200 # 2 hours + + +class EtcdRendezvousHandler(RendezvousHandler): + """ + Implements a + :py:class:`torch.distributed.elastic.rendezvous.RendezvousHandler` interface + backed by + :py:class:`torch.distributed.elastic.rendezvous.etcd_rendezvous.EtcdRendezvous`. + ``EtcdRendezvousHandler`` uses a URL to configure the type of rendezvous to + use and to pass implementation specific configurations to the rendezvous + module. The basic etcd rendezvous configuration URL looks like the following + :: + + etcd://:/?min_workers=&max_workers= # noqa: W605 + + -- example -- + + etcd://localhost:2379/1234?min_workers=1&max_workers=3 + + The URL above is interpreted as follows: + + 1. Use the rendezvous handler that is registered with the ``etcd`` + scheme + 2. The ``etcd`` endpoint to use is ``localhost:2379`` + 3. ``job_id == 1234`` is used as the prefix in etcd (this allows one to + share a common etcd server for multiple jobs so long as the + ``job_ids`` are guaranteed to be unique). Note that the job id can be + any string (e.g. does not need to be a number) as long as it is + unique. + 4. ``min_workers=1`` and ``max_workers=3`` specifies a range for + membership size - Torch Distributed Elastic starts running the job as + long as the cluster size is greater than or equal to ``min_workers`` + and admits up to ``max_workers`` into the cluster. + + Below are a full list of the parameters that can be passed to etcd + rendezvous: + + +--------------------------------------------+--------------------------+ + | Parameter | Description | + +============================================+==========================+ + | min_workers | minimum number of | + | | workers for the | + | | rendezvous to be valid | + +--------------------------------------------+--------------------------+ + | max_workers | maximum number of | + | | workers to admit | + +--------------------------------------------+--------------------------+ + | timeout | total timeout within | + | | which next_rendezvous is | + | | expected to succeed | + | | (default 600s) | + +--------------------------------------------+--------------------------+ + | last_call_timeout | additional wait amount | + | | ("last call") after min | + | | number of workers has | + | | been reached (defaults | + | | to 30s) | + +--------------------------------------------+--------------------------+ + | etcd_prefix | path prefix (from etcd | + | | root), inside which all | + | | etcd nodes will be | + | | created (defaults to | + | | ``/torchelastic/p2p``) | + +--------------------------------------------+--------------------------+ + """ + + def __init__(self, rdzv_impl: "EtcdRendezvous", local_addr: str | None): + """ + Args: + rdzv_impl: the implementation of the rendezvous + local_addr: the local address of the current node + """ + + self._rdzv_impl = rdzv_impl + self._local_addr = local_addr + + def __del__(self): + # TODO: look into using weakref here instead. + del self._rdzv_impl + + def get_backend(self) -> str: + return "etcd" + + def next_rendezvous(self): + rdzv_version, rank, world_size = self._rdzv_impl.rendezvous_barrier() + + logger.info("Creating EtcdStore as the c10d::Store implementation") + store = self._rdzv_impl.setup_kv_store(rdzv_version) + + bootstrap_store_info = RendezvousStoreInfo.build( + rank, store, local_addr=self._local_addr + ) + return RendezvousInfo(store, rank, world_size, bootstrap_store_info) + + def is_closed(self): + try: + _, state = self._rdzv_impl.get_rdzv_state() + return state["status"] == "closed" + except etcd.EtcdKeyNotFound: + # No rendezvous state, so it cannot be closed. + return False + + def set_closed(self): + self._rdzv_impl.set_closed() + + def num_nodes_waiting(self): + try: + _, state = self._rdzv_impl.get_rdzv_state() + if state["status"] == "final": + return state["num_workers_waiting"] + except etcd.EtcdKeyNotFound: + pass + return 0 + + def get_run_id(self) -> str: + return self._rdzv_impl._run_id + + def shutdown(self) -> bool: + try: + self.set_closed() + return True + except BaseException: + logger.warning("Shutdown failed", exc_info=True) + return False + + +# TODO: we should probably handle a few additional errors, +# like EtcdLeaderElectionInProgress and EtcdWatcherCleared. These are +# only relevant for multi-node Etcd ensemble. A simple retry would work, +# but is verbose to add everywhere. Consider wrapping the client calls +# into auto-retry for these errors? +# +class EtcdRendezvous: + """A rendezvous implementation that uses `etcd `__ as the backend store.""" + + def __init__( + self, + client, + prefix, + run_id, + num_min_workers, + num_max_workers, + timeout, + last_call_timeout, + ): + self.client = client + logger.info("Etcd machines: %s", self.client.machines) + + self._prefix = prefix + self._run_id = run_id + self._num_min_workers = num_min_workers + self._num_max_workers = num_max_workers + self._timeout = timeout + self._last_call_timeout = last_call_timeout + + # For cleaning up TTL refresher threads (for ephemeral keys) + self._lease_run_id_stop = None + self._lease_this_rank_stop = None + + if not self._prefix.endswith("/"): + self._prefix += "/" + + # Setup a permanent prefix dir, if didn't exist + if self._prefix != "/": + self.create_path_if_not_exists(self._prefix) + + # Lease a "sub-root" node specific to this job instance (run_id) + self.create_path_if_not_exists(self.get_path(""), ttl=CONST_RUNID_SUBROOT_TTL) + self._lease_run_id_stop = self.setup_lease_renewal( + self.get_path(""), ttl=CONST_RUNID_SUBROOT_TTL + ) + + # Subdir for all rendezvous work + self.create_path_if_not_exists(self.get_path("/rdzv")) + + # Create a rendezvous version counter, if doesn't exist + try: + self.client.write( + key=self.get_path("/rdzv/version_counter"), value="0", prevExist=False + ) + except etcd.EtcdAlreadyExist: + pass + + def __del__(self): + # TODO: look into using weakref here instead. + if self._lease_run_id_stop is not None: + self._lease_run_id_stop.set() + + if self._lease_this_rank_stop is not None: + self._lease_this_rank_stop.set() + + def rendezvous_barrier(self): + """ + Main entry point for next rendezvous. + + This method is blocking until rendezvous succeeds or a timeout occurs. + + Returns: + ``(rdzv_version, rank, world_size)`` + + Raises: + RendezvousTimeoutError - timeout waiting for rendezvous + RendezvousClosedError - rendezvous is or was closed while waiting + RendezvousError - other persistent errors that + render the rendezvous non-retryable + """ + self._rendezvous_deadline = time.time() + self._timeout + while True: + if time.time() > self._rendezvous_deadline: + raise RendezvousTimeoutError + + logger.info("Attempting to join next rendezvous") + try: + # Dis-own our lease in the previous rendezvous, if exists + if self._lease_this_rank_stop is not None: + self._lease_this_rank_stop.set() + + return self.init_phase() + + except EtcdRendezvousRetryImmediately: + # The type of failure suggests we can retry without delay + pass + + except EtcdRendezvousRetryableFailure: + # In case of retryable failure, wait a small delay + # to avoid spamming etcd + time.sleep(1) + + except RendezvousTimeoutError: + logger.info("Rendezvous timeout occurred in EtcdRendezvousHandler") + raise + + except RendezvousClosedError: + logger.info( + "Rendezvous for run_id=%s was observed to be closed", self._run_id + ) + raise + + except RendezvousError: + raise + + except Exception as e: + # In case of a general exception, wait a small delay + # to avoid spamming etcd + # FIXME: there are a few things that fall under this like + # etcd.EtcdKeyNotFound, etc, which could be handled more explicitly. + logger.info("Rendezvous attempt failed, will retry. Reason: %s", e) + time.sleep(1) + + def init_phase(self): + """ + Initially, the rendezvous state is expected to be one of: + + 1. empty (non-existent) - in this case we try to create a new one. + 2. joinable - we try to join it. + 3. final - we announce ourselves as waiting, and go into monitoring mode + + Any other state is considered transitional, and will be retried after + a short delay. + + Returns: + ``(rdzv_version, rank, world_size)`` + + Raises: + RendezvousClosedError - current rendezvous was/is closed + EtcdRendezvousRetryableFailure - observed some intermediate + state, which is best handled by retrying later + """ + try: + active_version = self.try_create_rendezvous() + state = json.loads(active_version.value) + logger.info("New rendezvous state created: %s", state) + except etcd.EtcdAlreadyExist: + active_version, state = self.get_rdzv_state() + # Note: it is possible for above query to fail (etcd.EtcdKeyNotFound), + # but this is ok for us - just means we'll restart from beginning. + logger.info("Observed existing rendezvous state: %s", state) + + if state["status"] == "closed": + raise RendezvousClosedError + + if state["status"] == "joinable": + return self.join_phase(state["version"]) + + if state["status"] == "final": + self.handle_existing_rendezvous(state["version"]) + raise EtcdRendezvousRetryImmediately + + self.try_wait_for_state_change(etcd_index=active_version.etcd_index + 1) + raise EtcdRendezvousRetryableFailure + + def join_phase(self, expected_version): + """ + We observed a rendezvous state in 'joinable' state, and attempt to join this + particular version, and then wait for all other peers to join. + """ + # Failure to join will propagate an exception, causing a re-entry. + active_version, this_rank = self.join_rendezvous(expected_version) + state = json.loads(active_version.value) + logger.info( + "Joined rendezvous version %s as rank %s. Full state: %s", + state["version"], + this_rank, + state, + ) + + # If this worker was first to reach num_min_workers requirement, + # and rendezvous is still joinable (therefore it is elastic), + # then this worker will be responsible for waiting out the "last call" + # timeout and closing (i.e. transitioning to 'frozen') the rendezvous + # afterwards. + # As a safety against a potential failure of this worker (during the + # last call timeout), the rendezvous state is made ephemeral + # when min_num_workers is reached. + + if this_rank == self._num_min_workers - 1 and state["status"] == "joinable": + logger.info("Rank %s is responsible for join last call.", this_rank) + last_call_deadline = time.time() + self._last_call_timeout + self.handle_join_last_call(expected_version, last_call_deadline) + logger.info("Rank %s finished join last call.", this_rank) + + # Wait for rendezvous state to be frozen, which means a fixed set of peers + logger.info("Waiting for remaining peers.") + active_version = self.wait_for_peers(expected_version) + state = json.loads(active_version.value) + + if state["version"] != expected_version: + raise AssertionError("Logic error: failed to observe version mismatch") + + return self.confirm_phase(expected_version, this_rank) + + def confirm_phase(self, expected_version, this_rank): + """ + Once the rendezvous state transitions from 'joinable' to 'frozen', + we have every participant confirm their membership and setup per-member + keep-alive TTL keys, and then wait for all other participants to confirm, + which would then successfully conclude this rendezvous. + """ + logger.info("All peers arrived. Confirming membership.") + self.confirm_membership(expected_version, this_rank) + + logger.info("Waiting for confirmations from all peers.") + active_version = self.wait_for_final(expected_version) + state = json.loads(active_version.value) + + logger.info( + "Rendezvous version %s is complete. Final state: %s", + state["version"], + state, + ) + + # Rendezvous version number; our rank in it; world size + return state["version"], this_rank, len(state["participants"]) + + def handle_existing_rendezvous(self, expected_version): + """ + Handle the case when there's an existing (state 'final) rendezvous already + in place, and we have to announce ourselves waiting, and wait until + the next rendezvous opportunity. + """ + # If state is 'final' -> increment num_workers_waiting + # Then, observe state changes: + # 1. if it's no longer final -> bail out and re-try + # 2. if keep alives are missing, destroy it and bail out. + active_state = self.announce_self_waiting(expected_version) + logger.info( + "Added self to waiting list. Rendezvous full state: %s", active_state.value + ) + + self.wait_for_rendezvous_to_free(expected_version) + logger.info( + "Previously existing rendezvous state changed. Will re-try joining." + ) + + def try_create_rendezvous(self): + """ + Create new rendezvous state or raise an exception that indicates an unexpected state (e.g. already exists). + + Raises: + RendezvousError - on unexpected state + """ + # Initially active_version is ephemeral - this is to handle the + # possibility that might fail to complete the setup transaction, + # i.e. the transition "setup" -> "joinable". + active_version = self.client.write( + key=self.get_path("/rdzv/active_version"), + value=json.dumps({"status": "setup"}), + prevExist=False, + ttl=CONST_ETCD_SETUP_TTL, + ) + + try: + version_counter = self.client.get(self.get_path("/rdzv/version_counter")) + version_counter.value = str(int(version_counter.value) + 1) + self.client.update(version_counter) + except (etcd.EtcdKeyNotFound, etcd.EtcdCompareFailed) as e: + raise RendezvousError( + "Unexpected state of EtcdRendezvousHandler, worker needs to die." + ) from e + + # Any failure below results in declaring a retryable rendezvous failure. + # The ephemeral /rdzv/active_version will expire and someone can then + # re-try the setup process. + + # Create directory node for participant data + self.client.write( + key=self.get_path(f"/rdzv/v_{version_counter.value}"), + value=None, + dir=True, + prevExist=False, + ) + + # Publish rendezvous version and signal it is ready-to-be-joined. + # If rendezvous was set closed just before this, a retry will happen, + # where the closed condition will be handled. + return self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps( + { + "status": "joinable", + "version": version_counter.value, + "participants": [], + } + ), + prev_value=active_version.value, + ) + + def join_rendezvous(self, expected_version): + """Helper method for the join phase.""" + # Use compare-and-swap to add self to rendezvous state: + while True: + cas_delay() + active_version, state = self.get_rdzv_state() + + if state["status"] != "joinable": + raise EtcdRendezvousRetryableFailure( + "Rendezvous state became non-joinable before we could join. " + "Must join next one." + ) + + if state["version"] != expected_version: + raise EtcdRendezvousRetryImmediately( + "Rendezvous version changed. Must try join the new one." + ) + + if len(state["participants"]) >= self._num_max_workers: + raise AssertionError( + "Logic error: joinable rendezvous should always have space left" + ) + + this_rank = len(state["participants"]) + state["participants"].append(this_rank) + + # When reaching min workers, or changing state to frozen, we'll set + # the active_version node to be ephemeral. + set_ttl: int | None = None + if len(state["participants"]) == self._num_max_workers: + state["status"] = "frozen" + state["keep_alives"] = [] + set_ttl = CONST_ETCD_FROZEN_TTL + elif len(state["participants"]) >= self._num_min_workers: + set_ttl = CONST_ETCD_JOINABLE_EPHEMERAL_TTL + + try: + # Compare-and-swap. + active_version = self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps(state), + prev_value=active_version.value, + ttl=set_ttl, + ) + # We succeeded joining. + return active_version, this_rank + + except etcd.EtcdCompareFailed: + logger.info("Join rendezvous CAS unsuccessful, retrying") + + def wait_for_peers(self, expected_version): + """Helper method for the join phase.""" + active_version, state = self.get_rdzv_state() + while True: + if state["status"] == "frozen" and state["version"] == expected_version: + # Success, all peers arrived. + return active_version + + elif state["status"] == "joinable" and state["version"] == expected_version: + # Continue waiting for any interesting events. + active_version, state = self.try_wait_for_state_change( + etcd_index=active_version.etcd_index + 1 + ) + + else: + # No valid transition possible at this point + raise EtcdRendezvousRetryableFailure( + "Rendezvous state transition no longer possible. Must re-enter." + ) + + def confirm_membership(self, expected_version, this_rank): + """Helper method for the confirm phase.""" + # Compare-and-swap loop + while True: + cas_delay() + active_version, state = self.get_rdzv_state() + + if state["status"] != "frozen": + raise EtcdRendezvousRetryImmediately( + "Rendezvous no longer frozen, before we confirmed. " + "Must join next one" + ) + if state["version"] != expected_version: + raise EtcdRendezvousRetryImmediately( + "Rendezvous version changed. Must try join the new one." + ) + + this_lease_key = self.get_path( + f"/rdzv/v_{expected_version}/rank_{this_rank}" + ) + self.client.set(this_lease_key, value=None, ttl=CONST_WORKER_KEEPALIVE_TTL) + + state["keep_alives"].append(this_lease_key) + if len(state["keep_alives"]) == len(state["participants"]): + # Everyone confirmed (this rank is last to do so) + state["status"] = "final" + state["num_workers_waiting"] = 0 + finalize = True + else: + finalize = False + + try: + # Compare-and-swap. If new state is still frozen, keep it ephemeral. + active_version = self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps(state), + prev_value=active_version.value, + ttl=None if finalize else CONST_ETCD_FROZEN_TTL, + ) + + self._lease_this_rank_stop = self.setup_lease_renewal( + this_lease_key, ttl=CONST_WORKER_KEEPALIVE_TTL + ) + return active_version + + except etcd.EtcdCompareFailed: + logger.info("Confirm membership CAS unsuccessful, retrying") + + def wait_for_final(self, expected_version): + """Helper method for the confirm phase.""" + active_version, state = self.get_rdzv_state() + while True: + if state["status"] == "final" and state["version"] == expected_version: + # Success. This rendezvous is final, and we accept it. + return active_version + + elif state["status"] == "frozen" and state["version"] == expected_version: + # Continue waiting for any interesting events. + active_version, state = self.try_wait_for_state_change( + etcd_index=active_version.etcd_index + 1 + ) + + else: + # No valid transition possible at this point + raise EtcdRendezvousRetryableFailure( + "Rendezvous state transition no longer possible. Must re-enter." + ) + + def announce_self_waiting(self, expected_version): + """ + Announce this worker is waiting (via num_workers_waiting counter) to join next + rendezvous, but only if state and version match. + """ + while True: + cas_delay() + active_version, state = self.get_rdzv_state() + + if state["status"] != "final" or state["version"] != expected_version: + raise EtcdRendezvousRetryImmediately + + # Increment counter to signal an additional waiting worker. + state["num_workers_waiting"] += 1 + + try: + active_version = self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps(state), + prev_value=active_version.value, + ) + return active_version + + except etcd.EtcdCompareFailed: + logger.info("Announce self as waiting CAS unsuccessful, retrying") + + def wait_for_rendezvous_to_free(self, expected_version): + """ + When there's an existing valid rendezvous in state 'final', we have to wait until the next opportunity to join. + + Such opportunity may come from: + + 1. rendezvous state changed by someone else, in which case we unblock and retry. + 2. rendezvous becomes invalid because at least one member failed to renew their + leased keep_alive node. We detect this, and destroy the rendezvous. + """ + active_version, state = self.get_rdzv_state() + while True: + if state["status"] != "final" or state["version"] != expected_version: + return + + # Check if current rendezvous state is valid, in the sense that all + # its members are alive (renewing their lease). + # If not, try destroy this rendezvous, so a new one can be created. + alive_members = self.client.get( + self.get_path(f"/rdzv/v_{expected_version}") + ) + keep_alive_keys = [ch.key for ch in alive_members.children] + + for key in state["keep_alives"]: + if key not in keep_alive_keys: + # This participant didn't renew their lease. We'll declare this + # rendezvous version as dead (but only if it hadn't changed) + logger.info("Keep-alive key %s is not renewed.", key) + logger.info( + "Rendezvous version %s is incomplete. ", expected_version + ) + logger.info("Attempting to destroy it.") + + # Compare-and-delete operation. Throws if compare failed, + # which means rendezvous was already destroyed/re-created/closed, + # and we can try to re-enter the barrier. + self.client.delete( + key=self.get_path("/rdzv/active_version"), + prevValue=active_version.value, + ) + + logger.info( + "Destroyed rendezvous version %s successfully.", + expected_version, + ) + + # We can return (and retry) immediately + return + + # Existing rendezvous seems valid, no reason to destroy it. + # We just have to wait until something changes and re-check. + try: + overall_timeout = ( + max(self._rendezvous_deadline - time.time(), 0.0) + 1.0 + ) + self.client.watch( + key=self.get_path("/rdzv"), + index=active_version.etcd_index + 1, + recursive=True, + timeout=overall_timeout, + ) + except (etcd.EtcdEventIndexCleared, etcd.EtcdWatchTimedOut): + pass + + if time.time() > self._rendezvous_deadline: + raise RendezvousTimeoutError + active_version, state = self.get_rdzv_state() + + def handle_join_last_call(self, expected_version, deadline): + """ + After we reach min number of workers, one particular worker takes on the + responsibility of waiting an additional timeout before closing the join window. + If the worker responsible for this fails, the rendezvous will be destroyed due + to expiring TTL, and the other participants will re-rendezvous. + + Here we expect to see state + Exit gracefully if either: + + 1. state becomes + 2. timeout happens (reaching deadline), in which case + we try the transition to + + Exit with exception otherwise. + """ + active_version, state = self.get_rdzv_state() + while True: + if state["status"] == "frozen" and state["version"] == expected_version: + # Worker set became frozen before last-call timeout. This is possible + # when num_max_workers is reached before the timeout. + return + + if state["status"] != "joinable" or state["version"] != expected_version: + raise EtcdRendezvousRetryableFailure( + "Rendezvous state transition no longer possible. Must re-enter." + ) + + # If timeout occurred, attempt a state transition (joinable -> frozen) + if time.time() >= deadline: + state["status"] = "frozen" + state["keep_alives"] = [] + try: + active_version = self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps(state), + prev_value=active_version.value, + ttl=CONST_ETCD_FROZEN_TTL, + ) + # We successfully made this rendezvous frozen. + return + except etcd.EtcdCompareFailed: + logger.info( + "Join last-call transition CAS unsuccessful. Will retry" + ) + cas_delay() + active_version, state = self.get_rdzv_state() + continue + + # Timeout did not occur, so we must refresh TTL, and wait for + # further changes. Note: we only want TTL to be refreshed if + # state is still joinable, hence we use CAS for that here, + # even though we don't change any of the data. + try: + active_version = self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=active_version.value, + prev_value=active_version.value, + ttl=CONST_ETCD_JOINABLE_EPHEMERAL_TTL, + ) + + # Minimize "oversleeping": + timeout = min( + CONST_ETCD_JOINABLE_EPHEMERAL_TTL / 2, + deadline - time.time() + 1.0, # Oversleeping by 1s is ok. + ) + active_version, state = self.try_wait_for_state_change( + etcd_index=active_version.etcd_index + 1, timeout=timeout + ) + except etcd.EtcdCompareFailed: + logger.info("Join last-call TTL refresh CAS unsuccessful, will retry") + cas_delay() + active_version, state = self.get_rdzv_state() + + def set_closed(self): + """ + Mark rendezvous 'closed' for current run_id, which is used to signal other + participants to not attempt to perform (re-)rendezvous. This is useful + when one of the workers decides the job is complete. + """ + while True: + active_version, state = self.get_rdzv_state() + + if state["status"] == "closed": + # Already closed by someone else. + return + + state["status"] = "closed" + try: + self.client.test_and_set( + key=self.get_path("/rdzv/active_version"), + value=json.dumps(state), + prev_value=active_version.value, + ) + return + + except etcd.EtcdCompareFailed: + logger.info("Set closed CAS unsuccessful, retrying") + cas_delay() + + def get_rdzv_state(self): + active_version = self.client.get(key=self.get_path("/rdzv/active_version")) + return active_version, json.loads(active_version.value) + + def try_wait_for_state_change(self, etcd_index, timeout=None): + # Don't sleep past the overall deadline (at least more than by 1s) + overall_timeout = max(self._rendezvous_deadline - time.time(), 0.0) + 1.0 + timeout = overall_timeout if timeout is None else min(timeout, overall_timeout) + + try: + self.client.watch( + self.get_path("/rdzv/active_version"), index=etcd_index, timeout=timeout + ) + except (etcd.EtcdEventIndexCleared, etcd.EtcdWatchTimedOut): + pass + + if time.time() > self._rendezvous_deadline: + raise RendezvousTimeoutError + + # Unfortunately, we have to do another fetch in order to get last etcd_index. + return self.get_rdzv_state() + + def get_path(self, path): + if not path.startswith("/"): + path = "/" + path + + return f"{self._prefix}run_{self._run_id}{path}" + + def create_path_if_not_exists(self, full_path, ttl=None): + try: + self.client.write( + key=full_path, value=None, dir=True, prevExist=False, ttl=ttl + ) + except etcd.EtcdAlreadyExist: + pass + + def setup_lease_renewal(self, full_path, ttl): + # NOTE: For ephemeral key TTL renewal (~lease) to work correctly, + # make sure you don't call any long-blocking methods that do not + # release the Python's GIL! An example of this is calling a pybind11 + # extension function that is blocking / long-running, but is not + # doing a scoped release of the GIL. + def lease_worker(client, path, ttl, stop_event): + while True: + try: + client.refresh(path, ttl=ttl) + except etcd.EtcdKeyNotFound: + break + except ConnectionRefusedError: + # This error usually occurs during test when the server already got terminated but the + # python garbage collector have not yet invoked the __del__ method. + break + + if stop_event.wait(timeout=ttl / 2): + break + + lease_stop_event = threading.Event() + lease_thread = threading.Thread( + target=lease_worker, args=(self.client, full_path, ttl, lease_stop_event) + ) + + lease_thread.daemon = True + lease_thread.start() + + return lease_stop_event + + def store_extra_data(self, rdzv_version, key, value): + node = self.get_path(f"/rdzv/v_{rdzv_version}/extra_data") + try: + # If first time we are storing anything: + extra_data = self.client.write( + key=node, value=json.dumps({key: value}), prevExist=False + ) + return + except etcd.EtcdAlreadyExist: + pass + + # CAS loop, to make sure we don't lose concurrent stores. + while True: + # We never delete extra_data. Failure here should be fatal, no special handling. + extra_data = self.client.get(node) + + new_extra_data_value = json.loads(extra_data.value) + new_extra_data_value[key] = value + + try: + extra_data = self.client.test_and_set( + key=node, + value=json.dumps(new_extra_data_value), + prev_value=extra_data.value, + ) + return + except etcd.EtcdCompareFailed: + logger.info("Store extra_data CAS unsuccessful, retrying") + time.sleep(0.1) + + def load_extra_data(self, rdzv_version, key, timeout=None): + # 'extra_data' node itself, and the directory it is located in: + node = self.get_path(f"/rdzv/v_{rdzv_version}/extra_data") + node_dir = self.get_path(f"/rdzv/v_{rdzv_version}") + + # TODO: implement timeout + # https://github.com/pytorch/elastic/issues/12 + while True: + # Combined wait for the node itself, and the key inside it. + root = self.client.get(node_dir) + + # Find the extra_data node, if it exists + extra_data = [n for n in root.children if n.key == node] + if len(extra_data) > 1: + raise AssertionError + + # Node for extra_data exists, check the desired key inside it. + if len(extra_data) == 1: + extra_data_dict = json.loads(extra_data[0].value) + if key in extra_data_dict: + return extra_data_dict[key] + + # The 'extra_data' node doesn't exist, or they key isn't published yet. + # Wait for interesting events on the extra_data node and retry. + try: + self.client.watch(node, index=root.etcd_index + 1) + except (etcd.EtcdEventIndexCleared, etcd.EtcdWatchTimedOut): + pass + + def setup_kv_store(self, rdzv_version): + store_path = self.get_path(f"/rdzv/v_{rdzv_version}/kv") + self.create_path_if_not_exists(store_path) + return EtcdStore(etcd_client=self.client, etcd_store_prefix=store_path) + + +def _create_etcd_client(params: RendezvousParameters) -> etcd.Client: + """Create a new ``etcd.Client`` from the specified ``RendezvousParameters``.""" + hostname, port = parse_rendezvous_endpoint(params.endpoint, 2379) + + # The communication protocol + protocol = params.config.get("protocol") + if protocol is None: + protocol = "http" + else: + if protocol != "http" and protocol != "https": + raise ValueError("The etcd protocol must be HTTP or HTTPS.") + + # The SSL client certificate + ssl_cert = params.config.get("cert") + if ssl_cert is not None: + cert_key = params.config.get("key") + if cert_key is not None: + # The etcd client expects the certificate key as the second element + # of the `cert` tuple. + ssl_cert = (ssl_cert, cert_key) + + # The root certificate + ca_cert = params.config.get("cacert") + + return etcd.Client( + hostname, + port, + protocol=protocol, + cert=ssl_cert, + ca_cert=ca_cert, + allow_reconnect=True, + ) + + +# Handler for torch.distributed "static" registration +def create_rdzv_handler(params: RendezvousParameters) -> RendezvousHandler: + """ + Usage: + + :: + + rdzv_params = RendezvousParameters( + backend="etcd", + endpoint="192.168.0.42:2379", + run_id="123", + min_nodes=4, + max_nodes=8, + timeout=300, + last_call_timeout=30, + etcd_prefix="custom_prefix", + protocol="https", + cacert="/etc/kubernetes/certs/ca.crt", + cert="/etc/kubernetes/certs/client.crt", + key="/etc/kubernetes/certs/client.key") + # -- or -- + rdzv_params = RendezvousParameters( + backend="etcd", + endpoint="192.168.0.42:2379", + run_id="123", + min_nodes=4, + max_nodes=8) + + etcd_rdzv_handler = create_etcd_rendezvous_handler(rdzv_params) + + + Where: + run_id - unique id for this training job instance, + min_nodes - min number of workers expected to join the rendezvous, + max_nodes - max number of workers allowed to join the rendezvous, + defaults to min_workers is not specified. + timeout - total timeout within which next_rendezvous is expected to + succeed; a RendezvousTimeoutError is raised otherwise; + Defaults is 600 (10 minutes). + last_call_timeout - additional wait amount ("last call") after + min number of workers has been reached. + Defaults to 30 seconds. + etcd_prefix - path prefix (from etcd root), inside which all + etcd nodes will be created. + Default is "/torchelastic/p2p". + protocol - http (default) or https to access etcd. + cacert - CA cert to access etcd, only makes sense with https. + cert - client cert to access etcd, only makes sense with https. + key - client key to access etcd, only makes sense with https. + """ + client = _create_etcd_client(params) + + etcd_prefix = params.get("etcd_prefix", "/torchelastic/p2p") + + rdzv = EtcdRendezvous( + client=client, + prefix=etcd_prefix, + run_id=params.run_id, + num_min_workers=params.min_nodes, + num_max_workers=params.max_nodes, + timeout=params.get_as_int("timeout", _DEFAULT_TIMEOUT), + last_call_timeout=params.get_as_int( + "last_call_timeout", _DEFAULT_LAST_CALL_TIMEOUT + ), + ) + return EtcdRendezvousHandler( + rdzv_impl=rdzv, + local_addr=params.local_addr, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous_backend.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous_backend.py new file mode 100644 index 0000000000000000000000000000000000000000..e41fc006f64c910c8410405f1f59694d24abd629 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_rendezvous_backend.py @@ -0,0 +1,214 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import binascii +from base64 import b64decode, b64encode +from typing import cast + +import urllib3.exceptions # type: ignore[import] + + +try: + import etcd # type: ignore[import] +except ModuleNotFoundError: + from . import _etcd_stub as etcd + +from torch.distributed import Store + +from .api import RendezvousConnectionError, RendezvousParameters, RendezvousStateError +from .dynamic_rendezvous import RendezvousBackend, Token +from .etcd_store import EtcdStore +from .utils import parse_rendezvous_endpoint + + +class EtcdRendezvousBackend(RendezvousBackend): + """Represents an etcd-based rendezvous backend. + + Args: + client: + The ``etcd.Client`` instance to use to communicate with etcd. + run_id: + The run id of the rendezvous. + key_prefix: + The path under which to store the rendezvous state in etcd. + ttl: + The TTL of the rendezvous state. If not specified, defaults to two hours. + """ + + _DEFAULT_TTL = 7200 # 2 hours + + _client: etcd.Client + _key: str + _ttl: int + + def __init__( + self, + client: etcd.Client, + run_id: str, + key_prefix: str | None = None, + ttl: int | None = None, + ) -> None: + if not run_id: + raise ValueError("The run id must be a non-empty string.") + + self._client = client + + if key_prefix: + self._key = key_prefix + "/" + run_id + else: + self._key = run_id + + if ttl and ttl > 0: + self._ttl = ttl + else: + self._ttl = self._DEFAULT_TTL + + @property + def name(self) -> str: + """See base class.""" + return "etcd-v2" + + def get_state(self) -> tuple[bytes, Token] | None: + """See base class.""" + try: + result = self._client.read(self._key) + except etcd.EtcdKeyNotFound: + return None + except (etcd.EtcdException, urllib3.exceptions.TimeoutError) as exc: + raise RendezvousConnectionError( + "The connection to etcd has failed. See inner exception for details." + ) from exc + + return self._decode_state(result) + + def set_state( + self, state: bytes, token: Token | None = None + ) -> tuple[bytes, Token, bool] | None: + """See base class.""" + base64_state = b64encode(state).decode() + + kwargs = {} + + def get_state(): + result = self.get_state() + if result is not None: + return *result, False + return None + + if token: + try: + token = int(token) + except ValueError: + return get_state() + + if token: + kwargs["prevIndex"] = token + else: + kwargs["prevExist"] = False + + try: + result = self._client.write(self._key, base64_state, self._ttl, **kwargs) + except (etcd.EtcdAlreadyExist, etcd.EtcdCompareFailed): + result = None + except (etcd.EtcdException, urllib3.exceptions.TimeoutError) as exc: + raise RendezvousConnectionError( + "The connection to etcd has failed. See inner exception for details." + ) from exc + + if result is None: + return get_state() + + tmp = *self._decode_state(result), True + return tmp + + def _decode_state(self, result: etcd.EtcdResult) -> tuple[bytes, Token]: + # pyrefly: ignore [missing-attribute] + base64_state = result.value.encode() + + try: + state = b64decode(base64_state) + except binascii.Error as exc: + raise RendezvousStateError( + "The state object is corrupt. See inner exception for details." + ) from exc + + # pyrefly: ignore [missing-attribute] + return state, result.modifiedIndex + + +def _create_etcd_client(params: RendezvousParameters) -> etcd.Client: + host, port = parse_rendezvous_endpoint(params.endpoint, default_port=2379) + + # The timeout + read_timeout = cast(int, params.get_as_int("read_timeout", 60)) + if read_timeout <= 0: + raise ValueError("The read timeout must be a positive integer.") + + # The communication protocol + protocol = params.get("protocol", "http").strip().lower() + if protocol != "http" and protocol != "https": + raise ValueError("The protocol must be HTTP or HTTPS.") + + # The SSL client certificate + ssl_cert = params.get("ssl_cert") + if ssl_cert: + ssl_cert_key = params.get("ssl_cert_key") + if ssl_cert_key: + # The etcd client expects the certificate key as the second element + # of the `cert` tuple. + ssl_cert = (ssl_cert, ssl_cert_key) + + # The root certificate + ca_cert = params.get("ca_cert") + + try: + return etcd.Client( + host, + port, + read_timeout=read_timeout, + protocol=protocol, + cert=ssl_cert, + ca_cert=ca_cert, + allow_reconnect=True, + ) + except (etcd.EtcdException, urllib3.exceptions.TimeoutError) as exc: + raise RendezvousConnectionError( + "The connection to etcd has failed. See inner exception for details." + ) from exc + + +def create_backend(params: RendezvousParameters) -> tuple[EtcdRendezvousBackend, Store]: + """Create a new :py:class:`EtcdRendezvousBackend` from the specified parameters. + + +--------------+-----------------------------------------------------------+ + | Parameter | Description | + +==============+===========================================================+ + | read_timeout | The read timeout, in seconds, for etcd operations. | + | | Defaults to 60 seconds. | + +--------------+-----------------------------------------------------------+ + | protocol | The protocol to use to communicate with etcd. Valid | + | | values are "http" and "https". Defaults to "http". | + +--------------+-----------------------------------------------------------+ + | ssl_cert | The path to the SSL client certificate to use along with | + | | HTTPS. Defaults to ``None``. | + +--------------+-----------------------------------------------------------+ + | ssl_cert_key | The path to the private key of the SSL client certificate | + | | to use along with HTTPS. Defaults to ``None``. | + +--------------+-----------------------------------------------------------+ + | ca_cert | The path to the root SSL authority certificate. Defaults | + | | to ``None``. | + +--------------+-----------------------------------------------------------+ + """ + client = _create_etcd_client(params) + + backend = EtcdRendezvousBackend( + client, params.run_id, key_prefix="/torch/elastic/rendezvous" + ) + + store = EtcdStore(client, "/torch/elastic/store") + + return backend, store diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_server.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_server.py new file mode 100644 index 0000000000000000000000000000000000000000..c4f30608a71dc0110f518d7bef7c9ddbf5eb71de --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_server.py @@ -0,0 +1,248 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import atexit +import logging +import os +import shlex +import shutil +import socket +import subprocess +import tempfile +import time +from typing import TextIO + + +try: + import etcd # type: ignore[import] +except ModuleNotFoundError: + pass + + +logger = logging.getLogger(__name__) + + +def find_free_port(): + """ + Find a free port and binds a temporary socket to it so that the port can be "reserved" until used. + + .. note:: the returned socket must be closed before using the port, + otherwise an ``address already in use`` error will happen. + The socket should be held and closed as close to the + consumer of the port as possible since otherwise, there + is a greater chance of race-condition where a different + process may see the port as being free and take it. + + Returns: a socket bound to the reserved free port + + Usage:: + + sock = find_free_port() + port = sock.getsockname()[1] + sock.close() + use_port(port) + """ + addrs = socket.getaddrinfo( + host="localhost", port=None, family=socket.AF_UNSPEC, type=socket.SOCK_STREAM + ) + + for addr in addrs: + family, type, proto, _, _ = addr + try: + s = socket.socket(family, type, proto) + s.bind(("localhost", 0)) + s.listen(0) + return s + except OSError as e: + s.close() # type: ignore[possibly-undefined] + print(f"Socket creation attempt failed: {e}") + raise RuntimeError("Failed to create a socket") + + +def stop_etcd(subprocess, data_dir: str | None = None): + if subprocess and subprocess.poll() is None: + logger.info("stopping etcd server") + subprocess.terminate() + subprocess.wait() + + if data_dir: + logger.info("deleting etcd data dir: %s", data_dir) + shutil.rmtree(data_dir, ignore_errors=True) + + +class EtcdServer: + """ + .. note:: tested on etcd server v3.4.3. + + Starts and stops a local standalone etcd server on a random free + port. Useful for single node, multi-worker launches or testing, + where a sidecar etcd server is more convenient than having to + separately setup an etcd server. + + This class registers a termination handler to shutdown the etcd + subprocess on exit. This termination handler is NOT a substitute for + calling the ``stop()`` method. + + The following fallback mechanism is used to find the etcd binary: + + 1. Uses env var TORCHELASTIC_ETCD_BINARY_PATH + 2. Uses ``/bin/etcd`` if one exists + 3. Uses ``etcd`` from ``PATH`` + + Usage + :: + + server = EtcdServer("/usr/bin/etcd", 2379, "/tmp/default.etcd") + server.start() + client = server.get_client() + # use client + server.stop() + + Args: + etcd_binary_path: path of etcd server binary (see above for fallback path) + """ + + def __init__(self, data_dir: str | None = None): + self._port = -1 + self._host = "localhost" + + root = os.path.dirname(__file__) + default_etcd_bin = os.path.join(root, "bin/etcd") + self._etcd_binary_path = os.environ.get( + "TORCHELASTIC_ETCD_BINARY_PATH", default_etcd_bin + ) + if not os.path.isfile(self._etcd_binary_path): + self._etcd_binary_path = "etcd" + + self._base_data_dir = ( + data_dir if data_dir else tempfile.mkdtemp(prefix="torchelastic_etcd_data") + ) + self._etcd_cmd = None + self._etcd_proc: subprocess.Popen | None = None + + def _get_etcd_server_process(self) -> subprocess.Popen: + if not self._etcd_proc: + raise RuntimeError( + "No etcd server process started. Call etcd_server.start() first" + ) + else: + return self._etcd_proc + + def get_port(self) -> int: + """Return the port the server is running on.""" + return self._port + + def get_host(self) -> str: + """Return the host the server is running on.""" + return self._host + + def get_endpoint(self) -> str: + """Return the etcd server endpoint (host:port).""" + return f"{self._host}:{self._port}" + + def start( + self, + timeout: int = 60, + num_retries: int = 3, + stderr: int | TextIO | None = None, + ) -> None: + """ + Start the server, and waits for it to be ready. When this function returns the server is ready to take requests. + + Args: + timeout: time (in seconds) to wait for the server to be ready + before giving up. + num_retries: number of retries to start the server. Each retry + will wait for max ``timeout`` before considering it as failed. + stderr: the standard error file handle. Valid values are + `subprocess.PIPE`, `subprocess.DEVNULL`, an existing file + descriptor (a positive integer), an existing file object, and + `None`. + + Raises: + TimeoutError: if the server is not ready within the specified timeout + """ + curr_retries = 0 + while True: + try: + data_dir = os.path.join(self._base_data_dir, str(curr_retries)) + os.makedirs(data_dir, exist_ok=True) + return self._start(data_dir, timeout, stderr) + except Exception as e: + curr_retries += 1 + stop_etcd(self._etcd_proc) + logger.warning( + "Failed to start etcd server, got error: %s, retrying", e + ) + if curr_retries >= num_retries: + shutil.rmtree(self._base_data_dir, ignore_errors=True) + raise + atexit.register(stop_etcd, self._etcd_proc, self._base_data_dir) + + def _start( + self, data_dir: str, timeout: int = 60, stderr: int | TextIO | None = None + ) -> None: + sock = find_free_port() + sock_peer = find_free_port() + self._port = sock.getsockname()[1] + peer_port = sock_peer.getsockname()[1] + + etcd_cmd = shlex.split( + " ".join( + [ + self._etcd_binary_path, + "--enable-v2", + "--data-dir", + data_dir, + "--listen-client-urls", + f"http://{self._host}:{self._port}", + "--advertise-client-urls", + f"http://{self._host}:{self._port}", + "--listen-peer-urls", + f"http://{self._host}:{peer_port}", + ] + ) + ) + + logger.info("Starting etcd server: [%s]", etcd_cmd) + + sock.close() + sock_peer.close() + self._etcd_proc = subprocess.Popen(etcd_cmd, close_fds=True, stderr=stderr) + self._wait_for_ready(timeout) + + def get_client(self): + """Return an etcd client object that can be used to make requests to this server.""" + return etcd.Client( + host=self._host, port=self._port, version_prefix="/v2", read_timeout=10 + ) + + def _wait_for_ready(self, timeout: int = 60) -> None: + client = etcd.Client( + host=f"{self._host}", port=self._port, version_prefix="/v2", read_timeout=5 + ) + max_time = time.time() + timeout + + while time.time() < max_time: + if self._get_etcd_server_process().poll() is not None: + # etcd server process finished + exitcode = self._get_etcd_server_process().returncode + raise RuntimeError( + f"Etcd server process exited with the code: {exitcode}" + ) + try: + logger.info("etcd server ready. version: %s", client.version) + return + except Exception: + time.sleep(1) + raise TimeoutError("Timed out waiting for etcd server to be ready!") + + def stop(self) -> None: + """Stop the server and cleans up auto generated resources (e.g. data dir).""" + logger.info("EtcdServer stop method called") + stop_etcd(self._etcd_proc, self._base_data_dir) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_store.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_store.py new file mode 100644 index 0000000000000000000000000000000000000000..09e588ff5b709b5ad9648c989046fcaafc383d27 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/etcd_store.py @@ -0,0 +1,216 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import datetime +import random +import time +from base64 import b64decode, b64encode + +# pyre-ignore[21]: Could not find name `Store` in `torch.distributed`. +from torch.distributed import Store + + +try: + import etcd # type: ignore[import] +except ModuleNotFoundError: + from . import _etcd_stub as etcd + + +# Delay (sleep) for a small random amount to reduce CAS failures. +# This does not affect correctness, but will reduce requests to etcd server. +def cas_delay(): + time.sleep(random.uniform(0, 0.1)) + + +# pyre-fixme[11]: Annotation `Store` is not defined as a type. +class EtcdStore(Store): + """ + Implement a c10 Store interface by piggybacking on the rendezvous etcd instance. + + This is the store object returned by ``EtcdRendezvous``. + """ + + def __init__( + self, + etcd_client, + etcd_store_prefix, + # Default timeout same as in c10d/Store.hpp + timeout: datetime.timedelta | None = None, + ): + super().__init__() # required for pybind trampoline. + + self.client = etcd_client + self.prefix = etcd_store_prefix + + if timeout is not None: + self.set_timeout(timeout) + + if not self.prefix.endswith("/"): + self.prefix += "/" + + def set(self, key, value): + """ + Write a key/value pair into ``EtcdStore``. + + Both key and value may be either Python ``str`` or ``bytes``. + """ + self.client.set(key=self.prefix + self._encode(key), value=self._encode(value)) + + def get(self, key) -> bytes: + """ + Get a value by key, possibly doing a blocking wait. + + If key is not immediately present, will do a blocking wait + for at most ``timeout`` duration or until the key is published. + + + Returns: + value ``(bytes)`` + + Raises: + LookupError - If key still not published after timeout + """ + b64_key = self.prefix + self._encode(key) + kvs = self._try_wait_get([b64_key]) + + if kvs is None: + raise LookupError(f"Key {key} not found in EtcdStore") + + return self._decode(kvs[b64_key]) + + def add(self, key, num: int) -> int: + """ + Atomically increment a value by an integer amount. + + The integer is represented as a string using base 10. If key is not present, + a default value of ``0`` will be assumed. + + Returns: + the new (incremented) value + + + """ + b64_key = self._encode(key) + # c10d Store assumes value is an integer represented as a decimal string + try: + # Assume default value "0", if this key didn't yet: + node = self.client.write( + key=self.prefix + b64_key, + value=self._encode(str(num)), # i.e. 0 + num + prevExist=False, + ) + return int(self._decode(node.value)) + except etcd.EtcdAlreadyExist: + pass + + while True: + # Note: c10d Store does not have a method to delete keys, so we + # can be sure it's still there. + node = self.client.get(key=self.prefix + b64_key) + new_value = self._encode(str(int(self._decode(node.value)) + num)) + try: + node = self.client.test_and_set( + key=node.key, value=new_value, prev_value=node.value + ) + return int(self._decode(node.value)) + except etcd.EtcdCompareFailed: + cas_delay() + + # pyrefly: ignore [bad-override] + def wait(self, keys, override_timeout: datetime.timedelta | None = None): + """ + Wait until all of the keys are published, or until timeout. + + Raises: + LookupError - if timeout occurs + """ + b64_keys = [self.prefix + self._encode(key) for key in keys] + kvs = self._try_wait_get(b64_keys, override_timeout) + if kvs is None: + raise LookupError("Timeout while waiting for keys in EtcdStore") + # No return value on success + + def check(self, keys) -> bool: + """Check if all of the keys are immediately present (without waiting).""" + b64_keys = [self.prefix + self._encode(key) for key in keys] + kvs = self._try_wait_get( + b64_keys, + override_timeout=datetime.timedelta(microseconds=1), # as if no wait + ) + return kvs is not None + + # + # Encode key/value data in base64, so we can store arbitrary binary data + # in EtcdStore. Input can be `str` or `bytes`. + # In case of `str`, utf-8 encoding is assumed. + # + def _encode(self, value) -> str: + if type(value) is bytes: + return b64encode(value).decode() + elif type(value) is str: + return b64encode(value.encode()).decode() + raise ValueError("Value must be of type str or bytes") + + # + # Decode a base64 string (of type `str` or `bytes`). + # Return type is `bytes`, which is more convenient with the Store interface. + # + def _decode(self, value) -> bytes: + if type(value) is bytes: + return b64decode(value) + elif type(value) is str: + return b64decode(value.encode()) + raise ValueError("Value must be of type str or bytes") + + # + # Get all of the (base64-encoded) etcd keys at once, or wait until all the keys + # are published or timeout occurs. + # This is a helper method for the public interface methods. + # + # On success, a dictionary of {etcd key -> etcd value} is returned. + # On timeout, None is returned. + # + def _try_wait_get(self, b64_keys, override_timeout=None): + timeout = self.timeout if override_timeout is None else override_timeout # type: ignore[attr-defined] + deadline = time.time() + timeout.total_seconds() + + while True: + # Read whole directory (of keys), filter only the ones waited for + all_nodes = None + try: + all_nodes = self.client.get(key=self.prefix) + req_nodes = { + node.key: node.value + for node in all_nodes.children + if node.key in b64_keys + } + + if len(req_nodes) == len(b64_keys): + # All keys are available + return req_nodes + except etcd.EtcdKeyNotFound: + pass + + watch_timeout = deadline - time.time() + if watch_timeout <= 0: + return None + + try: + index = all_nodes.etcd_index + 1 if all_nodes else 0 + self.client.watch( + key=self.prefix, + recursive=True, + timeout=watch_timeout, + index=index, + ) + except etcd.EtcdWatchTimedOut: + if time.time() >= deadline: + return None + else: + continue + except etcd.EtcdEventIndexCleared: + continue diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/registry.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/registry.py new file mode 100644 index 0000000000000000000000000000000000000000..71b62c9316df17d2391b9fe80fd5082bac290ee2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/registry.py @@ -0,0 +1,96 @@ +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging +from importlib.metadata import entry_points + +from .api import ( + rendezvous_handler_registry as handler_registry, + RendezvousHandler, + RendezvousParameters, +) +from .dynamic_rendezvous import create_handler + + +log = logging.getLogger(__name__) + +__all__ = ["get_rendezvous_handler"] + + +def _create_static_handler(params: RendezvousParameters) -> RendezvousHandler: + from . import static_tcp_rendezvous + + return static_tcp_rendezvous.create_rdzv_handler(params) + + +def _create_etcd_handler(params: RendezvousParameters) -> RendezvousHandler: + from . import etcd_rendezvous + + return etcd_rendezvous.create_rdzv_handler(params) + + +def _create_etcd_v2_handler(params: RendezvousParameters) -> RendezvousHandler: + from .etcd_rendezvous_backend import create_backend + + backend, store = create_backend(params) + + return create_handler(store, backend, params) + + +def _create_c10d_handler(params: RendezvousParameters) -> RendezvousHandler: + from .c10d_rendezvous_backend import create_backend + + backend, store = create_backend(params) + + return create_handler(store, backend, params) + + +def _register_default_handlers() -> None: + handler_registry.register("etcd", _create_etcd_handler) + handler_registry.register("etcd-v2", _create_etcd_v2_handler) + handler_registry.register("c10d", _create_c10d_handler) + handler_registry.register("static", _create_static_handler) + + +def _register_out_of_tree_handlers() -> None: + discovered_handler_generators = entry_points(group="torchrun.handlers") + + for handler_generator in discovered_handler_generators: + try: + # pyrefly: ignore [bad-index] + get_handler = discovered_handler_generators[handler_generator.name].load() + handler_registry.register(handler_generator.name, get_handler()) + except Exception: + log.warning( + "Exception while registering out of tree plugin %s: ", + handler_generator.name, + exc_info=True, + ) + + +def get_rendezvous_handler(params: RendezvousParameters) -> RendezvousHandler: + """ + Obtain a reference to a :py:class`RendezvousHandler`. + + Custom rendezvous handlers can be registered by + + :: + + from torch.distributed.elastic.rendezvous import rendezvous_handler_registry + from torch.distributed.elastic.rendezvous.registry import get_rendezvous_handler + + + def create_my_rdzv(params: RendezvousParameters): + return MyCustomRdzv(params) + + + rendezvous_handler_registry.register("my_rdzv_backend_name", create_my_rdzv) + + my_rdzv_handler = get_rendezvous_handler( + "my_rdzv_backend_name", RendezvousParameters + ) + """ + return handler_registry.create_handler(params) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/static_tcp_rendezvous.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/static_tcp_rendezvous.py new file mode 100644 index 0000000000000000000000000000000000000000..52b68000530889b6be1a8ec78ea762f6e5817975 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/static_tcp_rendezvous.py @@ -0,0 +1,128 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import datetime +import logging +from typing import cast + +from torch.distributed import PrefixStore, Store, TCPStore +from torch.distributed.elastic.rendezvous import ( + RendezvousHandler, + RendezvousInfo, + RendezvousParameters, + RendezvousStoreInfo, +) +from torch.distributed.elastic.rendezvous.utils import parse_rendezvous_endpoint + + +__all__ = ["StaticTCPRendezvous", "create_rdzv_handler"] + +logger = logging.getLogger(__name__) + +_default_timeout_seconds = 600 + + +class StaticTCPRendezvous(RendezvousHandler): + """ + Static rendezvous that is a wrapper around the TCPStore. + + Creates TCPStore based on the input parameters with the + listener on the agent with group_rank=0 + """ + + def __init__( + self, + master_addr: str, + master_port: int, + rank: int, + world_size: int, + run_id: str, + timeout: int, + ): + self.master_addr = master_addr + self.master_port = master_port + self.rank = rank + self.world_size = world_size + self.run_id = run_id + self.timeout = datetime.timedelta(seconds=timeout) + self._store: Store | None = None + + def get_backend(self) -> str: + return "static" + + @property + def use_agent_store(self) -> bool: + return True + + def next_rendezvous(self) -> RendezvousInfo: + logger.info("Creating TCPStore as the c10d::Store implementation") + is_master = self.rank == 0 + if not self._store: + self._store = TCPStore( # type: ignore[call-arg] + self.master_addr, + self.master_port, + self.world_size, + is_master, + self.timeout, + multi_tenant=True, + ) + store = PrefixStore(self.run_id, self._store) + # TCPStore server instance is used by trainer code + bootstrap_store_info = RendezvousStoreInfo(self.master_addr, self.master_port) + return RendezvousInfo( + store, + self.rank, + self.world_size, + bootstrap_store_info, + ) + + def is_closed(self): + return False + + def set_closed(self): + pass + + def num_nodes_waiting(self): + return 0 + + def get_run_id(self) -> str: + return self.run_id + + def shutdown(self) -> bool: + return True + + +def create_rdzv_handler(params: RendezvousParameters) -> RendezvousHandler: + if "rank" not in params.config: + raise ValueError( + "rank is absent in RendezvousParameters." + "Try add --node-rank to the cmd request" + ) + endpoint = params.endpoint.strip() + if not endpoint: + raise ValueError( + "endpoint is absent in RendezvousParameters" + "Try add --master-port and --master-addr to the cmd request" + ) + master_addr, master_port = parse_rendezvous_endpoint(endpoint, -1) + if master_port == -1: + raise ValueError( + f"Port is absent in endpoint: {endpoint}. Try launching with --master-port" + ) + world_size = params.max_nodes + rank = cast(int, params.config.get("rank")) + run_id = params.run_id + if "timeout" in params.config: + timeout = int(params.config["timeout"]) + else: + timeout = _default_timeout_seconds + + return StaticTCPRendezvous( + master_addr, master_port, rank, world_size, run_id, timeout + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..2164cab7b1f74e6b65f0d24c00b17a9af47f4810 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/rendezvous/utils.py @@ -0,0 +1,285 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import ipaddress +import random +import re +import socket +import time +import weakref +from collections.abc import Callable +from datetime import timedelta +from threading import Event, Thread +from typing import Any + + +__all__ = ["parse_rendezvous_endpoint"] + + +def _parse_rendezvous_config(config_str: str) -> dict[str, str]: + """Extract key-value pairs from a rendezvous configuration string. + + Args: + config_str: + A string in format =,...,=. + """ + config: dict[str, str] = {} + + config_str = config_str.strip() + if not config_str: + return config + + key_values = config_str.split(",") + for kv in key_values: + key, *values = kv.split("=", 1) + + key = key.strip() + if not key: + raise ValueError( + "The rendezvous configuration string must be in format " + "=,...,=." + ) + + value: str | None + if values: + value = values[0].strip() + else: + value = None + if not value: + raise ValueError( + f"The rendezvous configuration option '{key}' must have a value specified." + ) + + config[key] = value + return config + + +def _try_parse_port(port_str: str) -> int | None: + """Try to extract the port number from ``port_str``.""" + if port_str and re.match(r"^[0-9]{1,5}$", port_str): + return int(port_str) + return None + + +def parse_rendezvous_endpoint( + endpoint: str | None, default_port: int +) -> tuple[str, int]: + """Extract the hostname and the port number from a rendezvous endpoint. + + Args: + endpoint: + A string in format [:]. + default_port: + The port number to use if the endpoint does not include one. + + Returns: + A tuple of hostname and port number. + """ + if endpoint is not None: + endpoint = endpoint.strip() + + if not endpoint: + return ("localhost", default_port) + + # An endpoint that starts and ends with brackets represents an IPv6 address. + if endpoint[0] == "[" and endpoint[-1] == "]": + host, *rest = endpoint, *[] + else: + host, *rest = endpoint.rsplit(":", 1) + + # Sanitize the IPv6 address. + if len(host) > 1 and host[0] == "[" and host[-1] == "]": + host = host[1:-1] + + if len(rest) == 1: + port = _try_parse_port(rest[0]) + if port is None or port >= 2**16: + raise ValueError( + f"The port number of the rendezvous endpoint '{endpoint}' must be an integer " + "between 0 and 65536." + ) + else: + port = default_port + + if not re.match(r"^[\w\.:-]+$", host): + raise ValueError( + f"The hostname of the rendezvous endpoint '{endpoint}' must be a dot-separated list of " + "labels, an IPv4 address, or an IPv6 address." + ) + + return host, port + + +def _matches_machine_hostname(host: str) -> bool: + """Indicate whether ``host`` matches the hostname of this machine. + + This function compares ``host`` to the hostname as well as to the IP + addresses of this machine. Note that it may return a false negative if this + machine has CNAME records beyond its FQDN or IP addresses assigned to + secondary NICs. + """ + if host == "localhost": + return True + + try: + addr = ipaddress.ip_address(host) + except ValueError: + addr = None + + if addr and addr.is_loopback: + return True + + try: + host_addr_list = socket.getaddrinfo( + host, None, proto=socket.IPPROTO_TCP, flags=socket.AI_CANONNAME + ) + except (ValueError, socket.gaierror) as _: + host_addr_list = [] + + host_ip_list = [host_addr_info[4][0] for host_addr_info in host_addr_list] + + this_host = socket.gethostname() + if host == this_host: + return True + + addr_list = socket.getaddrinfo( + this_host, None, proto=socket.IPPROTO_TCP, flags=socket.AI_CANONNAME + ) + for addr_info in addr_list: + # If we have an FQDN in the addr_info, compare it to `host`. + if addr_info[3] and addr_info[3] == host: + return True + + # Otherwise if `host` represents an IP address, compare it to our IP + # address. + if addr and addr_info[4][0] == str(addr): + return True + + # If the IP address matches one of the provided host's IP addresses + if addr_info[4][0] in host_ip_list: + return True + + return False + + +def _delay(seconds: float | tuple[float, float]) -> None: + """Suspend the current thread for ``seconds``. + + Args: + seconds: + Either the delay, in seconds, or a tuple of a lower and an upper + bound within which a random delay will be picked. + """ + if isinstance(seconds, tuple): + seconds = random.uniform(*seconds) + # Ignore delay requests that are less than 10 milliseconds. + if seconds >= 0.01: + time.sleep(seconds) + + +class _PeriodicTimer: + """Represent a timer that periodically runs a specified function. + + Args: + interval: + The interval, in seconds, between each run. + function: + The function to run. + """ + + # The state of the timer is held in a separate context object to avoid a + # reference cycle between the timer and the background thread. + class _Context: + interval: float + function: Callable[..., None] + args: tuple[Any, ...] + kwargs: dict[str, Any] + stop_event: Event + + _name: str | None + _thread: Thread | None + _finalizer: weakref.finalize | None + + # The context that is shared between the timer and the background thread. + _ctx: _Context + + def __init__( + self, + interval: timedelta, + function: Callable[..., None], + *args: Any, + **kwargs: Any, + ) -> None: + self._name = None + + self._ctx = self._Context() + self._ctx.interval = interval.total_seconds() + self._ctx.function = function # type: ignore[assignment] + self._ctx.args = args or () + self._ctx.kwargs = kwargs or {} + self._ctx.stop_event = Event() + + self._thread = None + self._finalizer = None + + @property + def name(self) -> str | None: + """Get the name of the timer.""" + return self._name + + def set_name(self, name: str) -> None: + """Set the name of the timer. + + The specified name will be assigned to the background thread and serves + for debugging and troubleshooting purposes. + """ + if self._thread: + raise RuntimeError("The timer has already started.") + + self._name = name + + def start(self) -> None: + """Start the timer.""" + if self._thread: + raise RuntimeError("The timer has already started.") + + self._thread = Thread( + target=self._run, + name=self._name or "PeriodicTimer", + args=(self._ctx,), + daemon=True, + ) + + # We avoid using a regular finalizer (a.k.a. __del__) for stopping the + # timer as joining a daemon thread during the interpreter shutdown can + # cause deadlocks. The weakref.finalize is a superior alternative that + # provides a consistent behavior regardless of the GC implementation. + self._finalizer = weakref.finalize( + self, self._stop_thread, self._thread, self._ctx.stop_event + ) + + # We do not attempt to stop our background thread during the interpreter + # shutdown. At that point we do not even know whether it still exists. + self._finalizer.atexit = False + + self._thread.start() + + def cancel(self) -> None: + """Stop the timer at the next opportunity.""" + if self._finalizer: + self._finalizer() + + @staticmethod + def _run(ctx) -> None: + while not ctx.stop_event.wait(ctx.interval): + ctx.function(*ctx.args, **ctx.kwargs) + + @staticmethod + def _stop_thread(thread, stop_event): + stop_event.set() + + thread.join() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..1b8eb72b520bbd92aa9be936070393e405cb4ed3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__init__.py @@ -0,0 +1,44 @@ +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Expiration timers are set up on the same process as the agent and +used from your script to deal with stuck workers. When you go into +a code-block that has the potential to get stuck you can acquire +an expiration timer, which instructs the timer server to kill the +process if it does not release the timer by the self-imposed expiration +deadline. + +Usage:: + + import torchelastic.timer as timer + import torchelastic.agent.server as agent + + def main(): + start_method = "spawn" + message_queue = mp.get_context(start_method).Queue() + server = timer.LocalTimerServer(message, max_interval=0.01) + server.start() # non-blocking + + spec = WorkerSpec( + fn=trainer_func, + args=(message_queue,), + ...) + agent = agent.LocalElasticAgent(spec, start_method) + agent.run() + + def trainer_func(message_queue): + timer.configure(timer.LocalTimerClient(message_queue)) + with timer.expires(after=60): # 60 second expiry + # do some work + +In the example above if ``trainer_func`` takes more than 60 seconds to +complete, then the worker process is killed and the agent retries the worker group. +""" + +from .api import configure, expires, TimerClient, TimerRequest, TimerServer +from .file_based_local_timer import FileTimerClient, FileTimerRequest, FileTimerServer +from .local_timer import LocalTimerClient, LocalTimerServer diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6cf0e4dc2fe756838234ab8be0a1603bb9ad51de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9b0450c88d9a649408930b741f2194f0c18c21dd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/debug_info_logging.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/debug_info_logging.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..160f257705d7a5cd4f2daddea4fbbd9fad36f828 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/debug_info_logging.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/file_based_local_timer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/file_based_local_timer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b8ec32703e5100d7923e2522bb7ed69fb5e1dd28 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/file_based_local_timer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/local_timer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/local_timer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d903c2082a485188597ac67545932334be8f181a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/__pycache__/local_timer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/api.py new file mode 100644 index 0000000000000000000000000000000000000000..c3cec4d01b16d9305dcd5d0371e540b85122a4fd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/api.py @@ -0,0 +1,281 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import abc +import logging +import threading +import time +from contextlib import contextmanager +from inspect import getframeinfo, stack +from typing import Any + + +__all__ = [ + "TimerRequest", + "TimerClient", + "RequestQueue", + "TimerServer", + "configure", + "expires", +] + +logger = logging.getLogger(__name__) + + +class TimerRequest: + """ + Data object representing a countdown timer acquisition and release + that is used between the ``TimerClient`` and ``TimerServer``. + A negative ``expiration_time`` should be interpreted as a "release" + request. + + .. note:: the type of ``worker_id`` is implementation specific. + It is whatever the TimerServer and TimerClient implementations + have on to uniquely identify a worker. + """ + + __slots__ = ["worker_id", "scope_id", "expiration_time"] + + def __init__(self, worker_id: Any, scope_id: str, expiration_time: float): + self.worker_id = worker_id + self.scope_id = scope_id + self.expiration_time = expiration_time + + def __eq__(self, other): + if isinstance(other, TimerRequest): + return ( + self.worker_id == other.worker_id + and self.scope_id == other.scope_id + and self.expiration_time == other.expiration_time + ) + return False + + +class TimerClient(abc.ABC): + """ + Client library to acquire and release countdown timers by communicating + with the TimerServer. + """ + + @abc.abstractmethod + def acquire(self, scope_id: str, expiration_time: float) -> None: + """ + Acquires a timer for the worker that holds this client object + given the scope_id and expiration_time. Typically registers + the timer with the TimerServer. + """ + + @abc.abstractmethod + def release(self, scope_id: str): + """ + Releases the timer for the ``scope_id`` on the worker this + client represents. After this method is + called, the countdown timer on the scope is no longer in effect. + """ + + +class RequestQueue(abc.ABC): + """ + Consumer queue holding timer acquisition/release requests + """ + + @abc.abstractmethod + def size(self) -> int: + """ + Returns the size of the queue at the time this method is called. + Note that by the time ``get`` is called the size of the queue + may have increased. The size of the queue should not decrease + until the ``get`` method is called. That is, the following assertion + should hold: + + size = q.size() + res = q.get(size, timeout=0) + assert size == len(res) + + -- or -- + + size = q.size() + res = q.get(size * 2, timeout=1) + assert size <= len(res) <= size * 2 + """ + + @abc.abstractmethod + def get(self, size: int, timeout: float) -> list[TimerRequest]: + """ + Gets up to ``size`` number of timer requests in a blocking fashion + (no more than ``timeout`` seconds). + """ + + +class TimerServer(abc.ABC): + """ + Entity that monitors active timers and expires them + in a timely fashion. This server is responsible for + reaping workers that have expired timers. + """ + + def __init__( + self, request_queue: RequestQueue, max_interval: float, daemon: bool = True + ): + """ + :param request_queue: Consumer ``RequestQueue`` + :param max_interval: max time (in seconds) to wait + for an item in the request_queue + :param daemon: whether to run the watchdog thread as a daemon + """ + super().__init__() + self._request_queue = request_queue + self._max_interval = max_interval + self._daemon = daemon + self._watchdog_thread: threading.Thread | None = None + self._stop_signaled = False + + @abc.abstractmethod + def register_timers(self, timer_requests: list[TimerRequest]) -> None: + """ + Processes the incoming timer requests and registers them with the server. + The timer request can either be an acquire-timer or release-timer request. + Timer requests with a negative expiration_time should be interpreted + as a release-timer request. + """ + + @abc.abstractmethod + def clear_timers(self, worker_ids: set[Any]) -> None: + """ + Clears all timers for the given ``worker_ids``. + """ + + @abc.abstractmethod + def get_expired_timers(self, deadline: float) -> dict[str, list[TimerRequest]]: + """ + Returns all expired timers for each worker_id. An expired timer + is a timer for which the expiration_time is less than or equal to + the provided deadline. + """ + + @abc.abstractmethod + def _reap_worker(self, worker_id: Any) -> bool: + """ + Reaps the given worker. Returns True if the worker has been + successfully reaped, False otherwise. If any uncaught exception + is thrown from this method, the worker is considered reaped + and all associated timers will be removed. + """ + + def _reap_worker_no_throw(self, worker_id: Any) -> bool: + """ + Wraps ``_reap_worker(worker_id)``, if an uncaught exception is + thrown, then it considers the worker as reaped. + """ + try: + return self._reap_worker(worker_id) + except Exception: + logger.exception( + "Uncaught exception thrown from _reap_worker(), " + "check that the implementation correctly catches exceptions", + ) + return True + + def _watchdog_loop(self): + while not self._stop_signaled: + try: + self._run_watchdog() + except Exception: + logger.exception("Error running watchdog") + + def _run_watchdog(self): + batch_size = max(1, self._request_queue.size()) + timer_requests = self._request_queue.get(batch_size, self._max_interval) + self.register_timers(timer_requests) + now = time.time() + reaped_worker_ids = set() + for worker_id, expired_timers in self.get_expired_timers(now).items(): + logger.info( + "Reaping worker_id=[%s]. Expired timers: %s", + worker_id, + self._get_scopes(expired_timers), + ) + if self._reap_worker_no_throw(worker_id): + logger.info("Successfully reaped worker=[%s]", worker_id) + reaped_worker_ids.add(worker_id) + else: + logger.error( + "Error reaping worker=[%s]. Will retry on next watchdog.", worker_id + ) + self.clear_timers(reaped_worker_ids) + + def _get_scopes(self, timer_requests): + return [r.scope_id for r in timer_requests] + + def start(self) -> None: + logger.info( + "Starting %s... max_interval=%s, daemon=%s", + type(self).__name__, + self._max_interval, + self._daemon, + ) + self._watchdog_thread = threading.Thread( + target=self._watchdog_loop, daemon=self._daemon + ) + logger.info("Starting watchdog thread...") + self._watchdog_thread.start() + + def stop(self) -> None: + logger.info("Stopping %s", type(self).__name__) + self._stop_signaled = True + if self._watchdog_thread: + logger.info("Stopping watchdog thread...") + self._watchdog_thread.join(self._max_interval) + self._watchdog_thread = None + else: + logger.info("No watchdog thread running, doing nothing") + + +_timer_client: TimerClient | None = None + + +def configure(timer_client: TimerClient): + """ + Configures a timer client. Must be called before using ``expires``. + """ + global _timer_client + _timer_client = timer_client + logger.info("Timer client configured to: %s", type(_timer_client).__name__) + + +@contextmanager +def expires(after: float, scope: str | None = None, client: TimerClient | None = None): + """ + Acquires a countdown timer that expires in ``after`` seconds from now, + unless the code-block that it wraps is finished within the timeframe. + When the timer expires, this worker is eligible to be reaped. The + exact meaning of "reaped" depends on the client implementation. In + most cases, reaping means to terminate the worker process. + Note that the worker is NOT guaranteed to be reaped at exactly + ``time.now() + after``, but rather the worker is "eligible" for being + reaped and the ``TimerServer`` that the client talks to will ultimately + make the decision when and how to reap the workers with expired timers. + + Usage:: + + torch.distributed.elastic.timer.configure(LocalTimerClient()) + with expires(after=10): + torch.distributed.all_reduce(...) + """ + if client is None: + if _timer_client is None: + raise RuntimeError("Configure timer client before using countdown timers.") + client = _timer_client + if scope is None: + # grab the caller file + lineno + caller = getframeinfo(stack()[1][0]) + scope = f"{caller.filename}#{caller.lineno}" + expiration = time.time() + after + client.acquire(scope, expiration) + try: + yield + finally: + client.release(scope) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/debug_info_logging.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/debug_info_logging.py new file mode 100644 index 0000000000000000000000000000000000000000..e385d91283a7b610f00397bfa4bc4800a89761ca --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/debug_info_logging.py @@ -0,0 +1,24 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + + +from torch.distributed.elastic.utils.logging import get_logger + + +logger = get_logger(__name__) + +__all__ = ["log_debug_info_for_expired_timers"] + + +def log_debug_info_for_expired_timers( + run_id: str, + expired_timers: dict[int, list[str]], +): + if expired_timers: + logger.info("Timers expired for run:[%s] [%s].", run_id, expired_timers) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/file_based_local_timer.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/file_based_local_timer.py new file mode 100644 index 0000000000000000000000000000000000000000..5855efefcc85342378c273657fed27b37160a6ba --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/file_based_local_timer.py @@ -0,0 +1,444 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import io +import json +import os +import select +import signal +import sys +import threading +import time +from collections.abc import Callable +from typing import TypeVar +from typing_extensions import ParamSpec + +from torch.distributed.elastic.timer.api import TimerClient, TimerRequest +from torch.distributed.elastic.timer.debug_info_logging import ( + log_debug_info_for_expired_timers, +) +from torch.distributed.elastic.utils.logging import get_logger + + +_P = ParamSpec("_P") +_R = TypeVar("_R") + +__all__ = ["FileTimerClient", "FileTimerRequest", "FileTimerServer"] + +logger = get_logger(__name__) + + +def _retry(max_retries: int, sleep_time: float) -> Callable: + """ + A simple retry wrapper. + + Args: + max_retries: int, the maximum number of retries. + sleep_time: float, the time to sleep between retries. + """ + + def wrapper(func: Callable[_P, _R]) -> Callable[_P, _R]: + def wrapper(*args: _P.args, **kwargs: _P.kwargs): + for i in range(max_retries): + try: + return func(*args, **kwargs) + except Exception: + logger.exception("Error running %s. Retrying...", func.__name__) + if i < max_retries - 1: + time.sleep(sleep_time) + else: + raise + + return wrapper + + return wrapper + + +class FileTimerRequest(TimerRequest): + """ + Data object representing a countdown timer acquisition and release + that is used between the ``FileTimerClient`` and ``FileTimerServer``. + A negative ``expiration_time`` should be interpreted as a "release" + request. + ``signal`` is the signal to reap the worker process from the server + process. + """ + + __slots__ = ["version", "signal"] + + def __init__( + self, worker_pid: int, scope_id: str, expiration_time: float, signal: int = 0 + ) -> None: + super().__init__( + worker_id=worker_pid, scope_id=scope_id, expiration_time=expiration_time + ) + self.version = 1 + self.signal = signal + + @property + def worker_pid(self) -> int: + return self.worker_id + + def __eq__(self, other) -> bool: + if isinstance(other, FileTimerRequest): + return ( + super().__eq__(other) + and self.version == other.version + and self.signal == other.signal + ) + return False + + def to_json(self) -> str: + return json.dumps( + { + "version": self.version, + "pid": self.worker_pid, + "scope_id": self.scope_id, + "expiration_time": self.expiration_time, + "signal": self.signal, + }, + ) + + +class FileTimerClient(TimerClient): + """ + Client side of ``FileTimerServer``. This client is meant to be used + on the same host that the ``FileTimerServer`` is running on and uses + pid to uniquely identify a worker. + This client uses a named_pipe to send timer requests to the + ``FileTimerServer``. This client is a producer while the + ``FileTimerServer`` is a consumer. Multiple clients can work with + the same ``FileTimerServer``. + + Args: + + file_path: str, the path of a FIFO special file. ``FileTimerServer`` + must have created it by calling os.mkfifo(). + + signal: signal, the signal to use to kill the process. Using a + negative or zero signal will not kill the process. + """ + + def __init__( + self, + file_path: str, + signal=(signal.SIGKILL if sys.platform != "win32" else signal.CTRL_C_EVENT), # type: ignore[attr-defined] + ) -> None: + super().__init__() + self._file_path = file_path + self.signal = signal + + @_retry(max_retries=10, sleep_time=0.1) + def _open_non_blocking(self) -> io.TextIOWrapper | None: + # The server may have crashed or may haven't started yet. + # In such case, calling open() in blocking model blocks the client. + # To avoid such issue, open it in non-blocking mode, and an OSError will + # be raised if the server is not there. + fd = os.open(self._file_path, os.O_WRONLY | os.O_NONBLOCK) + return os.fdopen(fd, "wt") + + def _send_request(self, request: FileTimerRequest) -> None: + try: + file = self._open_non_blocking() + except Exception as e: + raise BrokenPipeError( + "Could not send the FileTimerRequest because FileTimerServer is not available." + ) from e + with file: + json_request = request.to_json() + # Write request with no greater than select.PIPE_BUF is guarantee to be atomic. + if len(json_request) > select.PIPE_BUF: + raise RuntimeError( + f"FileTimerRequest larger than {select.PIPE_BUF} bytes " + f"is not supported: {json_request}" + ) + file.write(json_request + "\n") + + def acquire(self, scope_id: str, expiration_time: float) -> None: + self._send_request( + request=FileTimerRequest( + worker_pid=os.getpid(), + scope_id=scope_id, + expiration_time=expiration_time, + signal=self.signal, + ), + ) + + def release(self, scope_id: str) -> None: + self._send_request( + request=FileTimerRequest( + worker_pid=os.getpid(), scope_id=scope_id, expiration_time=-1, signal=0 + ), + ) + + +class FileTimerServer: + """ + Server that works with ``FileTimerClient``. Clients are expected to be + running on the same host as the process that is running this server. + Each host in the job is expected to start its own timer server locally + and each server instance manages timers for local workers (running on + processes on the same host). + + Args: + + file_path: str, the path of a FIFO special file to be created. + + max_interval: float, max interval in seconds for each watchdog loop. + + daemon: bool, running the watchdog thread in daemon mode or not. + A daemon thread will not block a process to stop. + log_event: Callable[[Dict[str, str]], None], an optional callback for + logging the events in JSON format. + """ + + def __init__( + self, + file_path: str, + run_id: str, + max_interval: float = 10, + daemon: bool = True, + log_event: Callable[[str, FileTimerRequest | None], None] | None = None, + ) -> None: + self._file_path = file_path + self._run_id = run_id + self._max_interval = max_interval + self._daemon = daemon + self._timers: dict[tuple[int, str], FileTimerRequest] = {} + self._stop_signaled = False + self._watchdog_thread: threading.Thread | None = None + + self._is_client_started = False + if os.path.exists(self._file_path): + os.remove(self._file_path) + os.mkfifo(self._file_path) + # For test only. Count the number of requests received. + self._request_count = 0 + # For test only. Process all requests and stop the server. + self._run_once = False + self._log_event = ( + log_event if log_event is not None else lambda name, request: None + ) + self._last_progress_time = int(time.time()) + + def start(self) -> None: + logger.info( + "Starting %s... max_interval=%s, daemon=%s, file_path=%s", + type(self).__name__, + self._max_interval, + self._daemon, + self._file_path, + ) + self._watchdog_thread = threading.Thread( + target=self._watchdog_loop, daemon=self._daemon + ) + logger.info("Starting watchdog thread...") + self._watchdog_thread.start() + self._log_event("watchdog started", None) + + def stop(self) -> None: + logger.info("Stopping %s", type(self).__name__) + self._stop_signaled = True + if self._watchdog_thread: + logger.info("Stopping watchdog thread...") + self._watchdog_thread.join(self._max_interval) + self._watchdog_thread = None + else: + logger.info("No watchdog thread running, doing nothing") + if os.path.exists(self._file_path): + os.remove(self._file_path) + self._log_event("watchdog stopped", None) + + def run_once(self) -> None: + self._run_once = True + if self._watchdog_thread: + logger.info("Stopping watchdog thread...") + self._watchdog_thread.join() + self._watchdog_thread = None + else: + logger.info("No watchdog thread running, doing nothing") + if os.path.exists(self._file_path): + os.remove(self._file_path) + + @staticmethod + def is_process_running(pid: int): + """ + function to check process is running or not + """ + try: + # Check if the process exists and we can send signals to it + os.kill(pid, 0) + return True + except OSError: + return False + + def _watchdog_loop(self) -> None: + # Open the pipe in blocking mode blocks the server thread. + # This is fine for the following reasons: + # 1. No client case usually does not happen. + # 2. We are running the watchdog loop in a separate daemon + # thread, which will not block the process to stop. + try: + with open(self._file_path) as fd: + self._is_client_started = True + while not self._stop_signaled: + try: + run_once = self._run_once + self._run_watchdog(fd) + if run_once: + break + self._last_progress_time = int(time.time()) + except Exception: + logger.exception("Error running watchdog") + + except Exception: + logger.exception("Could not open the FileTimerServer pipe") + raise + + def _run_watchdog(self, fd: io.TextIOWrapper) -> None: + timer_requests = self._get_requests(fd, self._max_interval) + self.register_timers(timer_requests) + now = time.time() + reaped_worker_pids = set() + kill_process = False + reap_signal = 0 + + all_expired_timers = self.get_expired_timers(now) + log_debug_info_for_expired_timers( + self._run_id, + { + pid: [expired_timer.to_json() for expired_timer in expired_timers] + for pid, expired_timers in all_expired_timers.items() + }, + ) + + for worker_pid, expired_timers in all_expired_timers.items(): + logger.info( + "Reaping worker_pid=[%s]. Expired timers: %s", + worker_pid, + self._get_scopes(expired_timers), + ) + reaped_worker_pids.add(worker_pid) + # In case we have multiple expired timers, we find the first timer + # with a valid signal (>0) in the expiration time order. + expired_timers.sort(key=lambda timer: timer.expiration_time) + signal = 0 + expired_timer = None + for timer in expired_timers: + self._log_event("timer expired", timer) + if timer.signal > 0: + signal = timer.signal + expired_timer = timer + break + if signal <= 0: + logger.info( + "No signal specified with worker=[%s]. Do not reap it.", worker_pid + ) + continue + if self._reap_worker(worker_pid, signal): + logger.info( + "Successfully reaped worker=[%s] with signal=%s", worker_pid, signal + ) + self._log_event("kill worker process", expired_timer) + kill_process = True + reap_signal = signal + else: + logger.error( + "Error reaping worker=[%s]. Will retry on next watchdog.", + worker_pid, + ) + if kill_process and reap_signal > 0: + logger.info( + "Terminating the server process=[%s] because of expired timers", + os.getpid(), + ) + self._reap_worker(os.getpid(), reap_signal) + + self.clear_timers(reaped_worker_pids) + + def _get_scopes(self, timer_requests: list[FileTimerRequest]) -> list[str]: + return [r.scope_id for r in timer_requests] + + def _get_requests( + self, fd: io.TextIOWrapper, max_interval: float + ) -> list[FileTimerRequest]: + start = time.time() + requests = [] + while not self._stop_signaled or self._run_once: + # For named pipe, readline() is blocking when at least one writer opens. + # It returns only when flush() is called at the writer side. + # Note that flush() is automatically called inside close(). + # After the last writer closes, readline() is not blocking. + # It will return an empty string when it's at end-of-file. + # Since the client side always opens the pipe, writes a message and closes + # the pipe immediately, the readline() call below is not blocking for long. + json_request = fd.readline() + if len(json_request) == 0: + if self._run_once: + break + time.sleep(min(max_interval, 1)) + else: + request = json.loads(json_request) + pid = request["pid"] + scope_id = request["scope_id"] + expiration_time = request["expiration_time"] + signal = request["signal"] + requests.append( + FileTimerRequest( + worker_pid=pid, + scope_id=scope_id, + expiration_time=expiration_time, + signal=signal, + ) + ) + now = time.time() + if now - start > max_interval: + break + return requests + + def register_timers(self, timer_requests: list[FileTimerRequest]) -> None: + for request in timer_requests: + pid = request.worker_pid + scope_id = request.scope_id + expiration_time = request.expiration_time + self._request_count += 1 + + key = (pid, scope_id) + # negative expiration is a proxy for a release call + if expiration_time < 0: + if key in self._timers: + del self._timers[key] + else: + self._timers[key] = request + + def clear_timers(self, worker_pids: set[int]) -> None: + for pid, scope_id in list(self._timers.keys()): + if pid in worker_pids or not FileTimerServer.is_process_running(pid): + del self._timers[(pid, scope_id)] + + def get_expired_timers(self, deadline: float) -> dict[int, list[FileTimerRequest]]: + # pid -> [timer_requests...] + expired_timers: dict[int, list[FileTimerRequest]] = {} + for request in self._timers.values(): + if request.expiration_time <= deadline: + expired_scopes = expired_timers.setdefault(request.worker_pid, []) + expired_scopes.append(request) + return expired_timers + + def _reap_worker(self, worker_pid: int, signal: int) -> bool: + try: + os.kill(worker_pid, signal) + return True + except ProcessLookupError: + logger.info("Process with pid=%s does not exist. Skipping", worker_pid) + return True + except Exception: + logger.exception("Error terminating pid=%s", worker_pid) + return False + + def get_last_progress_time(self) -> int: + return self._last_progress_time if self._is_client_started else int(time.time()) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/local_timer.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/local_timer.py new file mode 100644 index 0000000000000000000000000000000000000000..5e66ef3fae34958422c1160bfdc1994b13bf1553 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/timer/local_timer.py @@ -0,0 +1,128 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import logging +import multiprocessing as mp +import os +import signal +import time +from queue import Empty +from typing import Any + +from .api import RequestQueue, TimerClient, TimerRequest, TimerServer + + +__all__ = ["LocalTimerClient", "MultiprocessingRequestQueue", "LocalTimerServer"] + +logger = logging.getLogger(__name__) + + +class LocalTimerClient(TimerClient): + """ + Client side of ``LocalTimerServer``. This client is meant to be used + on the same host that the ``LocalTimerServer`` is running on and uses + pid to uniquely identify a worker. This is particularly useful in situations + where one spawns a subprocess (trainer) per GPU on a host with multiple + GPU devices. + """ + + def __init__(self, mp_queue): + super().__init__() + self._mp_queue = mp_queue + + def acquire(self, scope_id, expiration_time): + pid = os.getpid() + acquire_request = TimerRequest(pid, scope_id, expiration_time) + self._mp_queue.put(acquire_request) + + def release(self, scope_id): + pid = os.getpid() + release_request = TimerRequest(pid, scope_id, -1) + self._mp_queue.put(release_request) + + +class MultiprocessingRequestQueue(RequestQueue): + """ + A ``RequestQueue`` backed by python ``multiprocessing.Queue`` + """ + + def __init__(self, mp_queue: mp.Queue): + super().__init__() + self._mp_queue = mp_queue + + def size(self) -> int: + return self._mp_queue.qsize() + + def get(self, size, timeout: float) -> list[TimerRequest]: + requests = [] + wait = timeout + for _ in range(size): + start = time.time() + + try: + r = self._mp_queue.get(block=True, timeout=wait) + except Empty: + break + + requests.append(r) + wait = wait - (time.time() - start) + if wait <= 0: + break + + return requests + + +class LocalTimerServer(TimerServer): + """ + Server that works with ``LocalTimerClient``. Clients are expected to be + subprocesses to the parent process that is running this server. Each host + in the job is expected to start its own timer server locally and each + server instance manages timers for local workers (running on processes + on the same host). + """ + + def __init__( + self, mp_queue: mp.Queue, max_interval: float = 60, daemon: bool = True + ): + super().__init__(MultiprocessingRequestQueue(mp_queue), max_interval, daemon) + self._timers: dict[tuple[Any, str], TimerRequest] = {} + + def register_timers(self, timer_requests: list[TimerRequest]) -> None: + for request in timer_requests: + pid = request.worker_id + scope_id = request.scope_id + expiration_time = request.expiration_time + + # negative expiration is a proxy for a release call + if expiration_time < 0: + self._timers.pop((pid, scope_id), None) + else: + self._timers[(pid, scope_id)] = request + + def clear_timers(self, worker_ids: set[int]) -> None: + for pid, scope_id in list(self._timers.keys()): + if pid in worker_ids: + self._timers.pop((pid, scope_id)) + + def get_expired_timers(self, deadline: float) -> dict[Any, list[TimerRequest]]: + # pid -> [timer_requests...] + expired_timers: dict[Any, list[TimerRequest]] = {} + for request in self._timers.values(): + if request.expiration_time <= deadline: + expired_scopes = expired_timers.setdefault(request.worker_id, []) + expired_scopes.append(request) + return expired_timers + + def _reap_worker(self, worker_id: int) -> bool: + try: + os.kill(worker_id, signal.SIGKILL) + return True + except ProcessLookupError: + logger.info("Process with pid=%s does not exist. Skipping", worker_id) + return True + except Exception: + logger.exception("Error terminating pid=%s", worker_id) + return False diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..8fe3ff97ad4ff663fc9f18147ccb07a7e33dcbda --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__init__.py @@ -0,0 +1,9 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from .api import get_env_variable_or_raise, get_socket_with_port, macros diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d0bad742e9123ba5c838e0bf60aeb2e76124b060 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..41c1599b7fd7b1055dee68320cdd69a441239ae4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/distributed.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/distributed.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bcd2f065aca1420a2c7ce7f6cc36888fccbd8572 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/distributed.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/log_level.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/log_level.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ba2f929db42c1ff84abb6b8c41f8c99bfe294027 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/log_level.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/logging.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/logging.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c1483131044a67570a83b9282b10d59b9cb56630 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/logging.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/process_state.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/process_state.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d06801741774eba00ae41cb92d33b45c0114bcea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/process_state.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/store.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/store.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c188ff6966bd65fb1410fb0f36ca8f64d846b541 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/__pycache__/store.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/api.py new file mode 100644 index 0000000000000000000000000000000000000000..2b881137047c23789a061a719437a43b1743959f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/api.py @@ -0,0 +1,62 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import os +import socket +from string import Template +from typing import Any + + +def get_env_variable_or_raise(env_name: str) -> str: + r""" + Tries to retrieve environment variable. Raises ``ValueError`` + if no environment variable found. + + Args: + env_name (str): Name of the env variable + """ + value = os.environ.get(env_name, None) + if value is None: + msg = f"Environment variable {env_name} expected, but not set" + raise ValueError(msg) + return value + + +def get_socket_with_port() -> socket.socket: + addrs = socket.getaddrinfo( + host="localhost", port=None, family=socket.AF_UNSPEC, type=socket.SOCK_STREAM + ) + for addr in addrs: + family, type, proto, _, _ = addr + s = socket.socket(family, type, proto) + try: + s.bind(("localhost", 0)) + s.listen(0) + return s + except OSError: + s.close() + raise RuntimeError("Failed to create a socket") + + +class macros: + """ + Defines simple macros for caffe2.distributed.launch cmd args substitution + """ + + local_rank = "${local_rank}" + + @staticmethod + def substitute(args: list[Any], local_rank: str) -> list[str]: + args_sub = [] + for arg in args: + if isinstance(arg, str): + sub = Template(arg).safe_substitute(local_rank=local_rank) + args_sub.append(sub) + else: + args_sub.append(arg) + return args_sub diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..cc2a7a0978d9aa85330f1b4bc0173159dadf10be --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__init__.py @@ -0,0 +1,10 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from .cycling_iterator import CyclingIterator +from .elastic_distributed_sampler import ElasticDistributedSampler diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c12d17c12262884efc4d54ea7dfb1e47e27ea1c5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/cycling_iterator.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/cycling_iterator.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..677bc435058051c9973597c6da0af2fa58713f1f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/cycling_iterator.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/elastic_distributed_sampler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/elastic_distributed_sampler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..aabd19e2855edb570ba57ae049a67f61b11fbe61 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/__pycache__/elastic_distributed_sampler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/cycling_iterator.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/cycling_iterator.py new file mode 100644 index 0000000000000000000000000000000000000000..291a04226db79c77b3bde4cec239e45b31be81b5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/cycling_iterator.py @@ -0,0 +1,57 @@ +#!/usr/bin/env python3 + +from collections.abc import Callable, Iterator +from typing import TypeVar +from typing_extensions import Self + + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +_T = TypeVar("_T") + +__all__ = ["CyclingIterator"] + + +class CyclingIterator(Iterator[_T]): + """ + An iterator decorator that cycles through the + underlying iterator "n" times. Useful to "unroll" + the dataset across multiple training epochs. + + The generator function is called as ``generator_fn(epoch)`` + to obtain the underlying iterator, where ``epoch`` is a + number less than or equal to ``n`` representing the ``k``th cycle + + For example if ``generator_fn`` always returns ``[1,2,3]`` + then ``CyclingIterator(n=2, generator_fn)`` will iterate through + ``[1,2,3,1,2,3]`` + """ + + def __init__( + self, + n: int, + generator_fn: Callable[[int], Iterator[_T]], + start_epoch: int = 0, + ): + self._n = n + self._epoch = start_epoch + self._generator_fn = generator_fn + self._iter = generator_fn(self._epoch) + + def __iter__(self) -> Self: + return self + + def __next__(self) -> _T: + try: + return next(self._iter) + except StopIteration as eod: # eod == end of data + if self._epoch < self._n - 1: + self._epoch += 1 + self._iter = self._generator_fn(self._epoch) + return self.__next__() + else: + raise eod diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/elastic_distributed_sampler.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/elastic_distributed_sampler.py new file mode 100644 index 0000000000000000000000000000000000000000..4889ce0b07bf3fc2a0cf317cd179d7c3179fbe94 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/data/elastic_distributed_sampler.py @@ -0,0 +1,95 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import math +from collections.abc import Iterator, Sized +from typing import cast, TypeVar + +import torch +from torch.utils.data import Dataset +from torch.utils.data.distributed import DistributedSampler + + +T = TypeVar("T") + +__all__ = ["ElasticDistributedSampler"] + + +class ElasticDistributedSampler(DistributedSampler[T]): + """ + Sampler that restricts data loading to a subset of + the dataset for elastic training. + + It is especially useful in conjunction with + :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each + process can pass a DistributedSampler instance as a DataLoader sampler, + and load a subset of the original dataset that is exclusive to it. + + .. note:: + Dataset is assumed to be of constant size. + + Args: + dataset: Dataset used for sampling. + num_replicas (optional): Number of processes participating in + distributed training. + rank (optional): Rank of the current process within num_replicas. + start_index (optional): Which index of the dataset to start sampling from + """ + + def __init__( + self, + dataset: Dataset[T], + num_replicas: int | None = None, + rank: int | None = None, + start_index: int = 0, + ): + super().__init__(dataset=dataset, num_replicas=num_replicas, rank=rank) + if not isinstance(dataset, Sized): + raise TypeError("Dataset must be an instance of collections.abc.Sized") + + # Cast to Sized for mypy + + sized_dataset = cast(Sized, dataset) + + if start_index >= len(sized_dataset): + raise ValueError( + f"Start index {start_index} should be less than dataset size {len(sized_dataset)}" + ) + + self.start_index = start_index + sized_dataset = cast(Sized, self.dataset) + self.num_samples = math.ceil( + float(len(sized_dataset) - self.start_index) / self.num_replicas + ) + self.total_size = self.num_samples * self.num_replicas + + def __iter__(self) -> Iterator[T]: + # deterministically shuffle based on epoch + g = torch.Generator() + g.manual_seed(self.epoch) + sized_dataset = cast(Sized, self.dataset) + indices = ( + torch.randperm(len(sized_dataset) - self.start_index, generator=g) + .add(self.start_index) + .tolist() + ) + + # add extra samples to make it evenly divisible + indices += indices[: (self.total_size - len(indices))] + if len(indices) != self.total_size: + raise AssertionError + + # subsample + indices = indices[self.rank : self.total_size : self.num_replicas] + if len(indices) != self.num_samples: + raise AssertionError + + return iter(indices) + + def __len__(self) -> int: + return self.num_samples diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/distributed.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/distributed.py new file mode 100644 index 0000000000000000000000000000000000000000..1ceafc09aa15eec94fa9ad2c9b560418900caeaf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/distributed.py @@ -0,0 +1,178 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import datetime +import os +import socket +from contextlib import closing + +import torch.distributed as dist +from torch.distributed.elastic.utils.logging import get_logger +from torch.distributed.elastic.utils.store import barrier + + +__all__ = ["create_c10d_store", "get_free_port", "get_socket_with_port"] + +logger = get_logger(__name__) + +_ADDRESS_IN_USE = "address already in use" +_SOCKET_TIMEOUT = "Socket Timeout" + +_TCP_STORE_INIT = "_tcp_store/num_members" + + +def create_c10d_store( + is_server: bool, + server_addr: str, + server_port: int = -1, + world_size: int = 1, + timeout: float = (60 * 10), # 10 min + wait_for_workers: bool = True, + retries=3, + use_libuv: bool | None = None, +): + if use_libuv is not None: + logger.warning( + "argument use_libuv is deprecated and ignored. Set USE_LIBUV environment " + 'variable to "0" to disable libuv, or "1" to enable it. If the env var ' + "is not set, libuv will be used by default." + ) + + # check os.environ for use_libuv + use_libuv = os.environ.get("USE_LIBUV", "1") == "1" # libuv is the default option + + if server_port == -1 and world_size > 1: + raise ValueError( + f"server_port must be specified when world_size > 1, got server_port={server_port}, world_size={world_size}" + ) + + if server_port != -1: + logger.info("sever_port: %s, specified, ignoring retries", server_port) + + # only retry when server_port is NOT static + attempt = retries if server_port == -1 else 1 + while True: + if server_port != -1: + port = server_port + else: + port = get_free_port() + + logger.info( + "Creating c10d store on %s:%s\n" + " world_size : %s\n" + " is_server : %s\n" + " timeout(sec): %s\n" + " use_libuv : %s\n", + server_addr, + port, + world_size, + is_server, + timeout, + use_libuv, + ) + + try: + store = dist.TCPStore( + host_name=server_addr, + port=port, + world_size=world_size, + is_master=is_server, + timeout=datetime.timedelta(seconds=timeout), + wait_for_workers=wait_for_workers, + use_libuv=use_libuv, + ) + # skips full rank check when we don't have to wait for all workers + if wait_for_workers: + _check_full_rank(store, world_size, timeout=timeout) + logger.info("Successfully created c10d store") + return store + except dist.DistNetworkError as e: + if _ADDRESS_IN_USE in str(e).lower(): # this will only happen on the server + if attempt < retries: + logger.warning( + "port: %s already in use, attempt: [%s/%s]", + port, + attempt, + retries, + ) + attempt += 1 + else: + raise dist.DistNetworkError( + f"on {server_addr}, port: {port} already in use" + ) from e + else: + raise + + +def _check_full_rank(store, world_size, timeout): + try: + barrier(store, world_size, key_prefix=_TCP_STORE_INIT, barrier_timeout=timeout) + except RuntimeError as e: + if str(e) == _SOCKET_TIMEOUT: + raise TimeoutError( + f"timed out waiting for all {world_size} members to join" + ) from e + else: + raise + + +def get_free_port(): + """ + Returns an unused port on localhost. + + This function finds an unused port on localhost by opening to socket to bind + to a port and then closing it. + + Returns: + int: an unused port on localhost + + Example: + >>> # xdoctest: +SKIP("Nondeterministic") + >>> get_free_port() + 63976 + + .. note:: + The port returned by :func:`get_free_port` is not reserved and may be + taken by another process after this function returns. + """ + sock = get_socket_with_port() + with closing(sock): + return sock.getsockname()[1] + + +def get_socket_with_port() -> socket.socket: + """ + Returns a free port on localhost that is "reserved" by binding a temporary + socket on it. Close the socket before passing the port to the entity + that requires it. Usage example + + :: + + sock = _get_socket_with_port() + with closing(sock): + port = sock.getsockname()[1] + sock.close() + # there is still a race-condition that some other process + # may grab this port before func() runs + func(port) + """ + + addrs = socket.getaddrinfo( + host="localhost", port=None, family=socket.AF_UNSPEC, type=socket.SOCK_STREAM + ) + for addr in addrs: + family, type, proto, _, _ = addr + s = socket.socket(family, type, proto) + try: + s.bind(("localhost", 0)) + s.listen(0) + return s + except OSError as e: + s.close() + logger.warning("Socket creation attempt failed.", exc_info=e) + raise RuntimeError("Failed to create a socket") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/log_level.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/log_level.py new file mode 100644 index 0000000000000000000000000000000000000000..87ea0f7d64182488b40fd7fed6965ce57ec475a0 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/log_level.py @@ -0,0 +1,14 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + + +def get_log_level() -> str: + """ + Return default log level for pytorch. + """ + return "WARNING" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/logging.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/logging.py new file mode 100644 index 0000000000000000000000000000000000000000..dbc0979d4f547c6deae213177a6eff4206936eed --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/logging.py @@ -0,0 +1,70 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import inspect +import logging +import os +import warnings + +from torch.distributed.elastic.utils.log_level import get_log_level + + +def get_logger(name: str | None = None) -> logging.Logger: + """ + Util function to set up a simple logger that writes + into stderr. The loglevel is fetched from the LOGLEVEL + env. variable or WARNING as default. The function will use the + module name of the caller if no name is provided. + + Args: + name: Name of the logger. If no name provided, the name will + be derived from the call stack. + """ + + # Derive the name of the caller, if none provided + # Use depth=2 since this function takes up one level in the call stack + return _setup_logger(name or _derive_module_name(depth=2)) + + +def _setup_logger(name: str | None = None) -> logging.Logger: + logger = logging.getLogger(name) + logger.setLevel(os.environ.get("LOGLEVEL", get_log_level())) + return logger + + +def _derive_module_name(depth: int = 1) -> str | None: + """ + Derives the name of the caller module from the stack frames. + + Args: + depth: The position of the frame in the stack. + """ + try: + stack = inspect.stack() + if depth >= len(stack): + raise AssertionError + # FrameInfo is just a named tuple: (frame, filename, lineno, function, code_context, index) + frame_info = stack[depth] + + module = inspect.getmodule(frame_info[0]) + if module: + module_name = module.__name__ + else: + # inspect.getmodule(frame_info[0]) does NOT work (returns None) in + # binaries built with @mode/opt + # return the filename (minus the .py extension) as modulename + filename = frame_info[1] + module_name = os.path.splitext(os.path.basename(filename))[0] + return module_name + except Exception as e: + warnings.warn( + f"Error deriving logger module name, using . Exception: {e}", + RuntimeWarning, + stacklevel=2, + ) + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/process_state.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/process_state.py new file mode 100644 index 0000000000000000000000000000000000000000..d2bbca39e5bbd181cb455f771abbf64bddb29bf1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/process_state.py @@ -0,0 +1,54 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Helpers for inspecting Linux process state from ``/proc``. + +These are intentionally agent-agnostic so they can be reused outside of +``torch.distributed.elastic.agent``. +""" + +from __future__ import annotations + + +__all__ = ["read_proc_state", "is_uninterruptible_state"] + + +def read_proc_state(pid: int) -> str | None: + """Read the process state char from ``/proc//stat``. Linux-only. + + Returns the single-letter state (e.g. ``'R'``, ``'S'``, ``'D'``, + ``'Z'``) or ``None`` if the file is unreadable (non-Linux, process + gone, permission denied, or parse error). The ``comm`` field can + contain spaces and parentheses, so the parser keys off the last + ``)`` rather than splitting on whitespace. + """ + try: + with open(f"/proc/{pid}/stat") as f: + data = f.read() + rparen = data.rfind(")") + if rparen == -1: + return None + # Fields after comm are space-separated; state is field 3 overall + # i.e. the first token after the closing ')'. + rest = data[rparen + 1 :].split() + if not rest: + return None + return rest[0] + except OSError: + return None + + +def is_uninterruptible_state(state: str | None) -> bool: + """Return True if ``state`` is Linux uninterruptible sleep ('D'). + + Includes the ``'D+'`` (foreground) variant; the state field is + a single char in the kernel format we read, but we accept any + string that starts with ``'D'`` defensively. + """ + return state is not None and state.startswith("D") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/store.py b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/store.py new file mode 100644 index 0000000000000000000000000000000000000000..00f37f0b12c2be77001d8a709504087fde58ddc1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/elastic/utils/store.py @@ -0,0 +1,226 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from collections.abc import Callable, Iterable +from contextlib import contextmanager +from datetime import timedelta + +import torch + + +DistStoreError = torch._C._DistStoreError + +_NUM_MEMBERS = "/num_members" +_LAST_MEMBER_CHECKIN = "/last_member" +_TRACE = "/TRACE" +_TRACING_GATE = "/TRACING_GATE" +_MAX_TRACE_MISSING_RANKS = 16 + + +__all__ = ["store_timeout", "get_all", "synchronize", "barrier"] + + +@contextmanager +def store_timeout(store, timeout: float): + """ + This sets the timeout and then restores the old timeout when the context + manager exits. + + Args: + store: the store to set the timeout on + timeout: the timeout to set + """ + + old_timeout = store.timeout + store.set_timeout(timedelta(seconds=timeout)) + yield + store.set_timeout(old_timeout) + + +def get_all(store, rank: int, prefix: str, world_size: int): + r""" + Given a store and a prefix, the method goes through the array of keys + of the following format: ``{prefix}{idx}``, where idx is in a range + from 0 to size, and tries to retrieve the data. + + The Rank0 process waits at the end to make sure all other processes + finished the procedure before exiting. + + Usage + + :: + + values = get_all(store, "torchelastic/data", 3) + value1 = values[0] # retrieves the data for key torchelastic/data0 + value2 = values[1] # retrieves the data for key torchelastic/data1 + value3 = values[2] # retrieves the data for key torchelastic/data2 + + """ + data_arr = store.multi_get([f"{prefix}{idx}" for idx in range(world_size)]) + + barrier_key = _barrier_nonblocking( + store=store, + world_size=world_size, + key_prefix=f"{prefix}/finished", + ) + if rank == 0: + # Rank0 runs the TCPStore daemon, as a result it needs to exit last. + # Otherwise, the barrier may timeout if rank0 process finished the work + # before other processes finished `get_all` method + store.wait([barrier_key]) + + return data_arr + + +def synchronize( + store, + data: bytes, + rank: int, + world_size: int, + key_prefix: str, + timeout: float = 300, +) -> list[bytes]: + """ + Synchronizes ``world_size`` agents between each other using the underlying c10d store. + The ``data`` will be available on each of the agents. + + Note: The data on the path is not deleted, as a result there can be stale data if + you use the same key_prefix twice. + + Time complexity: O(N) per worker, O(N^2) globally. + """ + with store_timeout(store, timeout): + store.set(f"{key_prefix}{rank}", data) + agent_data = get_all(store, rank, key_prefix, world_size) + return agent_data + + +def _try_detecting_missing_ranks( + store, + world_size: int, + key_prefix: str, + rank: int, + rank_decoder: Callable[[int], str], + trace_timeout: float, +) -> Iterable[str] | None: + store.set(f"{key_prefix}{rank}{_TRACE}", "") + + def _find_missing_ranks(): + missing_rank_info = set() + ranks_missing = 0 + for i in range(1, world_size): + # reduce noise, assuming in general 8 ranks per node + # It is valuable to know that 1 or >1 nodes have timed-out. + if ranks_missing >= _MAX_TRACE_MISSING_RANKS: + break + try: + if ranks_missing == 0: + store.wait( + [f"{key_prefix}{i}{_TRACE}"], timedelta(seconds=trace_timeout) + ) + else: + # use a shortest timeout, some ranks have failed to check-in + store.wait([f"{key_prefix}{i}{_TRACE}"], timedelta(milliseconds=1)) + except DistStoreError: + ranks_missing += 1 + missing_rank_info.add(rank_decoder(i)) + return missing_rank_info + + def _checkin(): + try: + store.wait([f"{key_prefix}{_TRACING_GATE}"]) + return [f"[]"] + except DistStoreError: + # in case rank0 is the source of the timeout, original exception will be raised + return None + + if rank == 0: + missing_rank_info = _find_missing_ranks() + store.set(f"{key_prefix}{_TRACING_GATE}", "") + return missing_rank_info + else: + return _checkin() + + +def _barrier_nonblocking(store, world_size: int, key_prefix: str) -> str: + """ + Does all the non-blocking operations for a barrier and returns the final key + that can be waited on. + """ + num_members_key = key_prefix + _NUM_MEMBERS + last_member_key = key_prefix + _LAST_MEMBER_CHECKIN + + idx = store.add(num_members_key, 1) + if idx == world_size: + store.set(last_member_key, "") + + return last_member_key + + +def barrier( + store, + world_size: int, + key_prefix: str, + barrier_timeout: float = 300, + rank: int | None = None, + rank_tracing_decoder: Callable[[int], str] | None = None, + trace_timeout: float = 10, +) -> None: + """ + A global lock between agents. This will pause all workers until at least + ``world_size`` workers respond. + + This uses a fast incrementing index to assign waiting ranks and a success + flag set by the last worker. + + Time complexity: O(1) per worker, O(N) globally. + + Optionally, passing rank will enable tracing of missing ranks on timeouts. + `rank_tracing_decoder` lambda arg can be used to convert rank data + into a more meaningful information at an app level (e.g. hostname). + + Note: Since the data is not removed from the store, the barrier can be used + once per unique ``key_prefix``. + """ + + if rank is None: + if rank_tracing_decoder is not None: + raise AssertionError("Tracing requires rank information") + + with store_timeout(store, barrier_timeout): + last_member_key = _barrier_nonblocking( + store=store, world_size=world_size, key_prefix=key_prefix + ) + try: + store.wait([last_member_key]) + except DistStoreError as e: + if rank is None: + raise e + else: + missing_ranks = _try_detecting_missing_ranks( + store, + world_size, + key_prefix, + rank, + rank_tracing_decoder or (lambda x: str(x)), + trace_timeout, + ) + if missing_ranks is not None: + raise DistStoreError( + "Timed out waiting on barrier on " + "rank {}, for key prefix: {} (world_size={}, missing_ranks={}, timeout={})".format( + rank, + key_prefix, + world_size, + f"[{', '.join(missing_ranks)}]", + barrier_timeout, + ) + ) from None + else: + raise e diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..afbdcadf0f9d04215b55493aa634237c212df191 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/fr_trace.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/fr_trace.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5bf8c481c9bbc2c2f4e1f8ddbf96d787f56e8cb7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/__pycache__/fr_trace.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..93a7989ab3d5878573038aab53f6a826059c07ef Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/builder.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/builder.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..59cd322ab7b12d132e2d99336e774f2f4e660e44 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/builder.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/config_manager.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/config_manager.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4af4e763edb8261f481cd36c56e1777ca753679f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/config_manager.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/fr_logger.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/fr_logger.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d7f4c12b6f307aa45831fa77f2554ae377afc3df Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/fr_logger.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/loader.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/loader.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e1edbb7bf55d26a82ee3dc11436635494ef472e8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/loader.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/types.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/types.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..11602ae59e935dc82f1116bda03f8f46ec99faed Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/types.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eb19c61fde344999ac19127f24f899f6fbb70739 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/builder.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/builder.py new file mode 100644 index 0000000000000000000000000000000000000000..d7c63e2abd30d9e74f23304572e5423eb3673adb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/builder.py @@ -0,0 +1,467 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import argparse +import ast +import copy +import os +import sys +from typing import Any # type: ignore[attr-defined] + +from torch.distributed.flight_recorder.components.fr_logger import FlightRecorderLogger +from torch.distributed.flight_recorder.components.types import ( + Collective, + Database, + EntryState, + Group, + MatchStateRecord, + Membership, + NCCLCall, + Op, + Traceback, +) +from torch.distributed.flight_recorder.components.utils import ( + add_stack_id_in_entries, + align_trace_from_beginning, + check_current_entry_match, + check_no_missing_dump_files, + check_version, + error_analysis, + find_coalesced_group as find_coalesced_group_p2p_only, + find_coalesced_group_with_non_p2p, + get_version_detail, + just_print_entries, + match_coalesced_groups as match_coalesced_groups_p2p_only, + match_coalesced_groups_with_non_p2p, +) + + +__all__ = [ + "build_groups_memberships", + "build_collectives", + "transform_ft", + "build_db", +] + +# Set up logging +logger: FlightRecorderLogger = FlightRecorderLogger() + + +try: + from tabulate import tabulate as _tabulate +except ModuleNotFoundError: + logger.warning("tabulate is not installed. Proceeding without it.") + + # Define a no-op tabulate function + def _tabulate(data: Any, headers: Any = None) -> Any: # type: ignore[misc] + return data + + +""" +Flat DB builder +""" + + +def build_groups_memberships( + pg_config: Any, +) -> tuple[ + list[Group], + dict[Any, Group], + list[Membership], + dict[str, set[Any]], + dict[tuple[str, int], str], +]: + """ + pg_config: { + global_rank: { + (pg_guid, desc, ranks) + } + } + + `pg_guid` is a system generated id, but depending on the mode of PG creation it could be a globally incrementing int + or a hash of the ranks. See `_process_group_name` in distributed_c10d.py. + `desc` is provided by the user (optionally) and should be 'meaningful' (e.g. TP/PP/DP group) + `ranks` is a list of the 'global ranks' that are members of the PG. + + (pg_guid, desc, ranks) tuples are appended lazily to the flight buffer when `getNCCLComm` is called on a PG and + the `enabled_` flag is true for that PG. + - the order of calling (init_process_group, new_group, etc) does not affect the order of the tuples in the list + + Returns: + `groups`: a groups table where each row is a Group namedtuple. + `_groups`: a dict that is indexed by pg_guid with Group namedtuple as value. + `memberships`: a membership table where each row is a Membership namedtuple. + `_memberships`: a dict that is indexed by pg_guid with set of ranks (int) as value. + `_pg_guids`: a dict that is indexed by (pg_uid, global_rank) with pg_guid as value. + """ + # flat lists for return + groups = [] + memberships = [] + + # dicts for faster cross-rank validation + _groups = {} + _memberships = {} + _pg_guids = {} + for global_rank in pg_config: + for pg_uid in pg_config[global_rank]: + desc = pg_config[global_rank][pg_uid]["desc"] + ranks = ast.literal_eval(pg_config[global_rank][pg_uid]["ranks"]) + # With the adoption of the split_group API, we can have multiple PGs with the same pg_guid (PG Name) + # So we need to add the hash of all its ranks within the PG as well. + # Also guid must be a string because `_process_group_name` returns a string. + pg_guid = pg_uid + str(hash(frozenset(ranks))) + _pg_guids[(pg_uid, global_rank)] = pg_guid + if isinstance(ranks, str): + # TODO Bug in FR data format? ranks is '[0, 1,...]' + ranks = ast.literal_eval(ranks) + + if pg_guid not in _groups: + groups.append(Group(id=pg_guid, desc=desc, size=len(ranks))) + for rank in ranks: + memberships.append(Membership(group_id=pg_guid, global_rank=rank)) + _groups[pg_guid] = groups[-1] + _memberships[pg_guid] = set(ranks) + else: + # validation across ranks + if _groups[pg_guid].desc != desc: + raise AssertionError( + f"mismatch in desc {_groups[pg_guid].desc} vs {desc} for group {pg_guid}" + ) + if _memberships[pg_guid] != set(ranks): + raise AssertionError( + f"mismatch in membership for group {pg_guid}" + f" {_memberships[pg_guid]} vs {set(ranks)}" + ) + return groups, _groups, memberships, _memberships, _pg_guids + + +def build_collectives( + all_entries: dict[int, list[dict[str, Any]]], + _groups: dict[str, Group], + _memberships: dict[str, set[Any]], + _pg_guids: dict[tuple[str, int], str], + version: str, + mismatch_cap: int = 10, +) -> tuple[list[Traceback], list[Collective], list[NCCLCall]]: + """ + groups, memberships are the non-flat dicts that are indexable + all_entries is a raw dict from the original dumps: + + all_entries: { + global_rank: [ + { + record_id: ordered id of the event in the trace buffer + pg_id: ProcessGroupNCCL::uid_ + *note: `pg_id` corresponds to nothing in groups table + process_group: (pg_name, desc) + *note: `pg_name`, `desc` corresponds to `pg_id`, `desc` in groups table + collective_seq_id: ordered id for collective operations and coalesced group operations + p2p_seq_id: ordered id for point-to-point operations + op_id: ordered id including individual ops inside coalescing group + profiling_name: descriptive name of the operation + 'time_created_ns', + 'input_sizes', + 'output_sizes', + 'state', + 'time_discovered_started_ns', + 'time_discovered_completed_ns', + 'retired', + 'frames', + } + ] + } + """ + tracebacks: list[Traceback] = [] + + collectives: list[Collective] = [] + nccl_calls: list[NCCLCall] = [] + + # once we find one mismatch, we stop pairing up collectives since the pairing is possibly incorrect + # instead, just record the remaining ops as NCCLCalls + mismatch = {_groups[g].id: 0 for g in _groups} + + # For best effort partial analysis. + dumps_ranks = {int(key) for key in all_entries} + """ + - it doesn't matter what order I put collectives/ncclops into their table. we can later on re-sort it by start time + - there could be multiple options for the "first" collective to pair up (rank 0,1 might do a bcast while rank 2,3 do a bcast) + - within a group, the first collective must be the same on all ranks in the group, then it can be marked as a + collective and removed + """ + while all_entries: + # we greedily match collectives, starting arbitrarily with the trace from the first rank + # later, if we exhaust the first rank, we continue with the next 'first rank' + rank_iter = iter(all_entries) + first_rank = next(rank_iter) + other_ranks = list(rank_iter) + + if len(all_entries[first_rank]) == 0: + all_entries.pop(first_rank) + continue + + # lets match the first collective! we need to know which ranks are involved, and ensure that this same + # collective is also the first one on those ranks within that group + entries = all_entries[first_rank] + current_entry = entries[0] + desc = current_entry["process_group"][1] + # For db build and logs printing, we want to use the original pg_name, not the hash one. + original_pg_name = current_entry["process_group"][0] + pg_name = _pg_guids[(original_pg_name, first_rank)] + expected_ranks = set(_memberships[pg_name]) + entry_state = EntryState(current_entry, expected_ranks) + match_record = MatchStateRecord( + expected_ranks=expected_ranks, + other_ranks=other_ranks, + entry_state=entry_state, + candidate_ranks={first_rank}, + candidate_idx={}, + found_ranks=set(), + found_idx={}, + errors=set(), + ) + + major_v, minor_v = get_version_detail(version) + find_coalesced_group = ( + find_coalesced_group_p2p_only + if major_v <= 2 and minor_v < 7 + else find_coalesced_group_with_non_p2p + ) + maybe_coalesced_group = find_coalesced_group( + pg_name, entries, _pg_guids, first_rank + ) + if len(maybe_coalesced_group) > 1: + num_coalesced_entries = len(maybe_coalesced_group) + # We need a copy of the original expected ranks to avoid modifying it. + candidate_ranks = copy.deepcopy(expected_ranks) + done_ranks = set() + all_coalesced_entries = {} + while candidate_ranks: + curr = candidate_ranks.pop() + done_ranks.add(curr) + grp = ( + find_coalesced_group(pg_name, all_entries[curr], _pg_guids, curr) # type: ignore[index] + if curr in all_entries # type: ignore[comparison-overlap] + else [] + ) + all_coalesced_entries[curr] = grp + for _, entry in grp: + op = Op(entry, _memberships, pg_name) + peer = None + if op.type == "send": + if op._src_g != curr: + raise AssertionError( + f"Send src error: {curr} expected but {op._src_g} is set" + ) + peer = op._dst_g + elif op.type == "recv": + if op._dst_g != curr: + raise AssertionError( + f"Recv dst error: {curr} expected but {op._dst_g} is set" + ) + peer = op._src_g + if peer and peer not in done_ranks: + candidate_ranks.add(peer) + + if major_v <= 2 and minor_v < 7: + match = match_coalesced_groups_p2p_only( + all_coalesced_entries, + group_size=_groups[pg_name].size, + groups=_groups, + memberships=_memberships, + _pg_guids=_pg_guids, + ) + else: + match = match_coalesced_groups_with_non_p2p( + copy.deepcopy( + all_coalesced_entries + ), # We want to keep a copy for cleanup. + pg_info=(pg_name, desc), + memberships=_memberships, + _pg_guids=_pg_guids, + mismatch=mismatch, + dumps_ranks=dumps_ranks, + version=version, + collectives=collectives, + match_record=match_record, + ) + + if match and mismatch[pg_name] == 0: + # We treat coalesced collectives as a single collective. + # TODO: we need to surface a merged collective info like input/output sizes to users. + collectives.append( + match_record.entry_state.to_collective(len(collectives)) + ) + else: + mismatch[pg_name] += 1 + for r in all_coalesced_entries: + idx_map = {r: i for i, _ in reversed(all_coalesced_entries[r])} # noqa: B035 + nccl_calls.extend( + reversed( + match_record.entry_state.to_nccl_call( + all_entries, + idx_map, + len(nccl_calls), + collectives[-1].id if match else None, + ) + ) + ) + # This extra cleanup is needed because we need to pop all collectives within a coalesced collective. + for i, k in idx_map.items(): + for _ in range(1, num_coalesced_entries): + try: + all_entries[i].pop(k) + except IndexError: + # In the case of a missing rank symptom that a rank didn't schedule the coalesced collective, + # we should not fail the analysis script here. + pass + else: + # Iterate through all the ranks and check if there is a mismatch for the current entry. + check_current_entry_match( + all_entries, + _pg_guids, + (pg_name, desc), + current_entry, + _memberships, + mismatch, + match_record, + ) + + # Use heuristics to decide what type of errors and error messages we should print. + error_analysis( + all_entries, + match_record, + dumps_ranks, + first_rank, + current_entry, + mismatch, + get_version_detail(version), + pg_name, + ) + + # at this point there are 3 possibilities + # 1. we found a match on all the ranks that are members of the group + # -> we create a Collective and remove the individual entries from their original lists + if match_record.found_ranks == expected_ranks and mismatch[pg_name] == 0: + collectives.append( + match_record.entry_state.to_collective(len(collectives)) + ) + idx_map = { + r: match_record.found_idx[r] if r != first_rank else 0 + for r in match_record.found_ranks + } + nccl_calls.extend( + match_record.entry_state.to_nccl_call( + all_entries, idx_map, len(nccl_calls), collectives[-1].id + ) + ) + + # 2. we found a partial match but some ranks are missing + # 3. we found no match + # -> since it's not a complete collective, no entry goes into collectives but we still record a nccl call + # TODO should there be a way to mark 'mismatches'? + else: + logger.debug("appending a non-matching collective") + idx_map = { + r: match_record.candidate_idx[r] if r != first_rank else 0 + for r in match_record.candidate_ranks + } + collectives.append( + match_record.entry_state.to_collective( + len(collectives), + errors=match_record.errors, + idx_map=idx_map, + all_entries=all_entries, + ) + ) + nccl_calls.extend( + match_record.entry_state.to_nccl_call( + all_entries, idx_map, len(nccl_calls), None + ) + ) + + if mismatch[pg_name] > mismatch_cap: + logger.error( + "Too many mismatches for process_group %s: %s aborting", pg_name, desc + ) + break + + return tracebacks, collectives, nccl_calls + + +def transform_ft( + details: dict[str, dict[str, Any]], group_world_size: int +) -> dict[str, dict[str, Any]]: + for dump_key, dump in details.items(): + rank = dump["rank"] + for key, pg_config in dump["pg_config"].items(): + if pg_config["desc"] == "default_pg": + ranks = ast.literal_eval(pg_config["ranks"]) + replica_id = rank // group_world_size + first_rank = replica_id * group_world_size + new_ranks = [r + first_rank for r in ranks] + details[dump_key]["pg_config"][key]["ranks"] = f"{new_ranks}" + + return details + + +def build_db( + details: dict[str, dict[str, Any]], args: argparse.Namespace, version: str +) -> Database: + if args.verbose: + os.environ["FR_TRACE_VERBOSE_OUTPUT"] = "1" + # temporary state used for building database + entries = {} + pg_config = {} + version_by_ranks = {} + for dump in details.values(): + rank = dump["rank"] + entries[rank] = dump["entries"] + version_by_ranks[rank] = dump["version"] + pg_config[rank] = dump["pg_config"] + + # Ensure version is consistent across all ranks. + check_version(version_by_ranks, version) + entries = align_trace_from_beginning(entries) + stack_id_trace_map: dict[str, int] = {} + if args.just_print_entries: + entries, stack_id_trace_map = add_stack_id_in_entries(entries) + + # flattened database + groups, _groups, memberships, _memberships, _pg_guids = build_groups_memberships( + pg_config + ) + logger.debug("built groups, memberships") + + if args.just_print_entries: + just_print_entries( + entries, _groups, _memberships, _pg_guids, args, stack_id_trace_map + ) + sys.exit(0) + + if not args.allow_incomplete_ranks: + check_no_missing_dump_files(entries, memberships) + + tracebacks, collectives, nccl_calls = build_collectives( + entries, _groups, _memberships, _pg_guids, version, args.mismatch_cap + ) + logger.debug("built collectives, nccl_calls") + if args.verbose: + logger.debug("Groups") + logger.debug(_tabulate(groups, headers=Group._fields)) + logger.debug("Memberships") + logger.debug(_tabulate(memberships, headers=Membership._fields)) + logger.debug("Collectives") + logger.debug(_tabulate(collectives, headers=Collective._fields)) + logger.debug("NCCLCalls") + logger.debug(_tabulate(nccl_calls, headers=NCCLCall._fields)) + db = Database( + tracebacks=tracebacks, + collectives=collectives, + ncclcalls=nccl_calls, + groups=groups, + memberships=memberships, + ) + return db diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/config_manager.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/config_manager.py new file mode 100644 index 0000000000000000000000000000000000000000..60de805f20808ea565d3b8bbf1fd60351d07e708 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/config_manager.py @@ -0,0 +1,112 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import argparse +import logging +from collections.abc import Sequence + +from torch.distributed.flight_recorder.components.fr_logger import FlightRecorderLogger + + +__all__ = ["JobConfig"] + + +logger: FlightRecorderLogger = FlightRecorderLogger() + + +class JobConfig: + """ + A helper class to manage the script configuration. + """ + + def __init__(self: "JobConfig"): + self.parser = argparse.ArgumentParser( + description="PyTorch Flight recorder analyzing script." + ) + self.parser.add_argument( + "trace_dir", + nargs="?", + help="Directory containing one trace file per rank, named with _.", + ) + self.parser.add_argument( + "--selected-ranks", + default=None, + nargs="+", + type=int, + help="List of ranks we want to show traces for.", + ) + self.parser.add_argument( + "--allow-incomplete-ranks", + action="store_true", + help=( + "FR trace require all ranks to have dumps for analysis. " + "This flag allows best-effort partial analysis of results " + "and printing of collected data." + ), + ) + self.parser.add_argument( + "--pg-filters", + default=None, + nargs="+", + type=str, + help=( + "List of filter strings, it could be pg name or pg desc. " + "If specified, only show traces for the given pg." + ), + ) + self.parser.add_argument("-o", "--output", default=None) + self.parser.add_argument( + "-p", + "--prefix", + help=( + "Common filename prefix to strip such that rank can be extracted. " + "If not specified, will attempt to infer a common prefix." + ), + default=None, + ) + self.parser.add_argument("-j", "--just_print_entries", action="store_true") + self.parser.add_argument("-v", "--verbose", action="store_true") + self.parser.add_argument("--print_stack_trace", action="store_true") + self.parser.add_argument( + "--mismatch_cap", + type=int, + default=10, + help="Maximum number of mismatches we print (from earliest).", + ) + self.parser.add_argument( + "--transform-ft", + action="store_true", + help="Transform PG config to use global ranks to analyze traces produced by torchft", + ) + self.parser.add_argument( + "--group-world-size", + type=int, + default=None, + help="The number of ranks in 1 torchft replica group. Must be specified if --transform-ft is True", + ) + + def parse_args(self: "JobConfig", args: Sequence[str] | None) -> argparse.Namespace: + # pyrefly: ignore [bad-assignment] + args = self.parser.parse_args(args) + # pyrefly: ignore [missing-attribute] + if args.selected_ranks is not None: + # pyrefly: ignore [missing-attribute] + if not args.just_print_entries: + raise AssertionError( + "Not support selecting ranks without printing entries" + ) + # pyrefly: ignore [missing-attribute] + if args.pg_filters is not None: + # pyrefly: ignore [missing-attribute] + if not args.just_print_entries: + raise AssertionError( + "Not support selecting pg filters without printing entries" + ) + # pyrefly: ignore [missing-attribute] + if args.verbose: + logger.set_log_level(logging.DEBUG) + # pyrefly: ignore [bad-return] + return args diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/fr_logger.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/fr_logger.py new file mode 100644 index 0000000000000000000000000000000000000000..e56634397bff9d6d1ec38eab43f1856f52e02829 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/fr_logger.py @@ -0,0 +1,54 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging +from collections.abc import Callable +from typing import Any + + +__all__ = ["FlightRecorderLogger"] + + +class FlightRecorderLogger: + _instance: Any | None = None + logger: logging.Logger + + def __init__(self) -> None: + self.logger: logging.Logger = logging.getLogger("Flight Recorder") + + def __new__(cls) -> Any: + if cls._instance is None: + cls._instance = super().__new__(cls) + cls._instance.logger = logging.getLogger("Flight Recorder") + cls._instance.logger.setLevel(logging.INFO) + formatter = logging.Formatter("%(message)s") + ch = logging.StreamHandler() + ch.setFormatter(formatter) + cls._instance.logger.addHandler(ch) + return cls._instance + + def set_log_level(self, level: int) -> None: + self.logger.setLevel(level) + + @property + def debug(self) -> Callable[..., None]: + return self.logger.debug + + @property + def info(self) -> Callable[..., None]: + return self.logger.info + + @property + def warning(self) -> Callable[..., None]: + return self.logger.warning + + @property + def error(self) -> Callable[..., None]: + return self.logger.error + + @property + def critical(self) -> Callable[..., None]: + return self.logger.critical diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/loader.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/loader.py new file mode 100644 index 0000000000000000000000000000000000000000..9abd6b4fae9701d4363c84f070a23f1f3f96b7b1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/loader.py @@ -0,0 +1,113 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import argparse +import gc +import os +import pickle +import re +import time +from collections import defaultdict +from collections.abc import Iterator +from contextlib import contextmanager +from typing import Any + +from torch.distributed.flight_recorder.components.fr_logger import FlightRecorderLogger + + +__all__ = [ + "read_dump", + "read_dir", +] + + +logger: FlightRecorderLogger = FlightRecorderLogger() + + +def read_dump(prefix: str, filename: str) -> dict[str, str | int | list[Any]]: + basename = os.path.basename(filename) + + rank = int(basename[len(prefix) :]) + host_name = f"host_rank{rank}" + + with open(filename, "rb") as infile: + dump = pickle.load(infile) + + entries = dump["entries"] + version = dump["version"] + pg_config = dump["pg_config"] + + return { + "host_name": host_name, + "rank": rank, + "entries": entries, + "version": version, + "pg_config": pg_config, + } + + +exp = re.compile(r"([\w\-\_]*?)(\d+)$") + + +def _determine_prefix(files: list[str]) -> str: + """If the user doesn't specify a prefix, but does pass a dir full of similarly-prefixed files, we should be able to + infer the common prefix most of the time. But if we can't confidently infer, just fall back to requiring the user + to specify it + """ + possible_prefixes: defaultdict[str, set[int]] = defaultdict(set) + for f in files: + m = exp.search(f) + if m: + p, r = m.groups() + possible_prefixes[p].add(int(r)) + if len(possible_prefixes) == 1: + prefix = next(iter(possible_prefixes)) + logger.debug("Inferred common prefix %s", prefix) + return prefix + else: + raise ValueError( + "Unable to automatically determine the common prefix for the trace file names. " + "Please specify --prefix argument manually" + ) + + +@contextmanager +def _disable_gc() -> Iterator[None]: + was_enabled = gc.isenabled() + gc.disable() + try: + yield + finally: + if was_enabled: + gc.enable() + + +def read_dir(args: argparse.Namespace) -> tuple[dict[str, dict[str, Any]], str]: + with _disable_gc(): + prefix = args.prefix + details = {} + t0 = time.time() + version = "" + filecount = 0 + if not os.path.isdir(args.trace_dir): + raise AssertionError(f"folder {args.trace_dir} does not exist") + for root, _, files in os.walk(args.trace_dir): + if prefix is None: + prefix = _determine_prefix(files) + for f in files: + if (offset := f.find(prefix)) == -1: + continue + details[f] = read_dump(f[:offset] + prefix, os.path.join(root, f)) + filecount += 1 + if not version: + version = str(details[f]["version"]) + tb = time.time() + if len(details) <= 0: + raise AssertionError( + f"no files loaded from {args.trace_dir} with prefix {prefix}" + ) + logger.debug("loaded %s files in %ss", filecount, tb - t0) + return details, version diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/types.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/types.py new file mode 100644 index 0000000000000000000000000000000000000000..8fc9dcfc4ee8965ac32cefabc837017e8f0f7ea9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/types.py @@ -0,0 +1,688 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import math +import os +from enum import auto, Enum +from typing import ( + _eval_type, # pyrefly: ignore [missing-module-attribute] + Any, + Generic, + NamedTuple, + TypeVar, +) + +from torch.distributed.flight_recorder.components.fr_logger import FlightRecorderLogger + + +__all__ = [ + "Ref", + "TypeInfo", + "MatchState", + "MatchInfo", + "Group", + "Membership", + "Traceback", + "Collective", + "NCCLCall", + "Database", + "EntryState", + "Op", + "MatchStateRecord", +] + + +T = TypeVar("T", bound=NamedTuple) + + +class Ref(Generic[T]): + pass + + +class TypeInfo(NamedTuple): + name: str + fields: list[tuple[str, type]] # type: ignore[type-arg] + + @classmethod + def from_type(cls, c: T) -> "TypeInfo": + if hasattr(c, "__name__"): + name = c.__name__ + else: + name = str(c) + return cls( + name, + [(f, _eval_type(c.__annotations__[f], globals(), {})) for f in c._fields], + ) + + +class MatchState(Enum): + """ + Enum representing the possible states of matching for collective operations. + + - FULLY_MATCHED: Indicates that all aspects of the collective operations match. + - COLLECTIVE_TYPE_MISMATCH: The types of the collective operations differ. + - SIZE_OR_SYNTAX_MISMATCH: There is a mismatch in input/output sizes or violation of collective syntax. + - COLLECTIVE_STATE_MISMATCH: + The states of the collective not same, such as one finished while another just started or scheduled. + - COLLECTIVE_DTYPE_MISMATCH: The data types of the collective input/output differ. + - UNDECIDED: + The match status is ambiguous or cannot be determined, e.g., we might need to check all ranks for alltoall_base. + """ + + FULLY_MATCHED = auto() + COLLECTIVE_TYPE_MISMATCH = auto() + SIZE_OR_SYNTAX_MISMATCH = auto() + COLLECTIVE_STATE_MISMATCH = auto() + COLLECTIVE_DTYPE_MISMATCH = auto() + UNDECIDED = auto() + + +class MatchInfo: + """ + Aside from the match state, we also store some dynamic info for the match such as the culprit rank + or collective state that caused the mismatch. + """ + + def __init__(self, state: MatchState, culprit: str | None = None) -> None: + self._state = state + self.culprit = culprit + + def __str__(self) -> str: + details = f", {self.culprit}" if getattr(self, "culprit", None) else "" + return f"Error type: {self._state.name}{details}" + + @property + def state(self) -> MatchState: + return self._state + + +""" +Schema for flat DB + +TODO schemas not yet implemented +# threads as recorded at termination of process +Threads + id: int + traceback_id: int + process_id: int + +Process: + id: int # Same as world groups RANK + pid: int + hostname: str + +NCCLOp: + # nccl op implementation details (sends/recv) + id: int + nccl_call_id: int + +""" + + +class Group(NamedTuple): + id: str + desc: str + size: int + + +class Membership(NamedTuple): + group_id: str + global_rank: int + + +class Traceback(NamedTuple): + id: int + frames: str + + +class Collective(NamedTuple): + id: int + group_id: str + pass_check: bool + collective_seq_id: int + p2p_seq_id: int + record_id: int + pg_desc: str + collective_name: str + input_sizes: list[list[int]] + output_sizes: list[list[int]] + expected_ranks: set[int] + collective_state: str + collective_frames: list[dict[str, str]] + input_numel: int | None = None + output_numel: int | None = None + missing_ranks: set[int] | None = None + mismatch_collectives: dict[int, "Collective"] | None = None + type_of_mismatch: MatchInfo | None = None + + +class NCCLCall(NamedTuple): + id: int + # pyrefly: ignore [bad-specialization] + collective_id: Ref[Collective] + group_id: str + global_rank: int # technically Ref[Process] once we have it + # pyrefly: ignore [bad-specialization] + traceback_id: Ref[Traceback] + collective_type: str + sizes: list[list[int]] + + +class Database(NamedTuple): + groups: list[Group] + memberships: list[Membership] + tracebacks: list[Traceback] + collectives: list[Collective] + ncclcalls: list[NCCLCall] + + +# TODO: We need to add a schema for the following +types = [ + TypeInfo.from_type(t) # type: ignore[type-var] + for t in [Database, NCCLCall, Collective, Traceback, Membership, Group] + if ( + isinstance(t, type) + and issubclass(t, tuple) + and hasattr(t, "_fields") + and t is not TypeInfo + ) +] + +""" +Stacktrace cache +TODO +""" + + +""" +Collective Matching logic + +NOTE: For now, these collectives need to be supported by NCCL, +https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/overview.html. +""" +COLLECTIVES = { + "broadcast", + "_broadcast_oop", + "reduce", + "_reduce_oop", + "all_gather", + "all_gather_single", + "all_gather_v", + "all_reduce", + "_all_gather_base", + "all_gather_into_tensor_coalesced", + "reduce_scatter", + "reduce_scatter_single", + "reduce_scatter_v", + "reduce_scatter_tensor_coalesced", + "_reduce_scatter_base", + "gather", + "scatter", + "all_to_all", + "all_to_all_single", + "all_to_all_v_single", + "all_reduce_barrier", + "barrier", + "split", + "new_window", + "allreduce_coalesced", + "ALLGATHER_coalesced", + "REDUCE_SCATTER_coalesced", +} + +P2P = { + "send", + "recv", +} + + +class EntryState: + """ + Util class to keep track of the state of an entry and standardize the way we + log the error info during analysis. + """ + + def __init__(self, entry: dict[str, Any], expected_ranks: set[int]) -> None: + self.pg_name = entry["process_group"][0] + self.desc = entry["process_group"][1] + self.pg_desc = ( + f"{self.pg_name}:{self.desc}" if self.desc != "undefined" else self.pg_name + ) + self.profiling_name = entry["profiling_name"] + self.collective_seq_id = entry["collective_seq_id"] + self.p2p_seq_id = entry["p2p_seq_id"] + self.record_id = entry["record_id"] + self.input_sizes = entry["input_sizes"] + self.output_sizes = entry["output_sizes"] + self.collective_state = entry["state"] + self.collective_frames = entry.get("frames", []) + self.expected_ranks = expected_ranks + self.missing_ranks: set[int] + self.input_numel: int + self.output_numel: int + self.errors: set[tuple[int, MatchInfo]] + + def log( + self, + logger: FlightRecorderLogger, + logger_msg: str, + frame_formatter: Any, + total_numel: tuple[int, int] | None = None, + errors: set[tuple[int, MatchInfo]] | None = None, + missing_ranks: set[int] | None = None, + ) -> None: + logger.info( + logger_msg, + self.collective_seq_id, + ) + logger.info("internal record id: %s", self.record_id) + logger.info("group info: %s", self.pg_desc) + logger.info("collective: %s", self.profiling_name) + if missing_ranks: + self.missing_ranks = missing_ranks + logger.info("missing ranks: %s", missing_ranks) + if total_numel: + self.input_numel = total_numel[0] + self.output_numel = total_numel[1] + logger.info("total input numel: %d", total_numel[0]) + logger.info("total output numel: %d", total_numel[1]) + logger.info("input sizes: %s", self.input_sizes) + logger.info("output sizes: %s", self.output_sizes) + logger.info("world size: %d", len(self.expected_ranks)) + logger.info("expected ranks: %s", self.expected_ranks) + logger.info("collective state: %s", self.collective_state) + if errors: + self.errors = errors + error_msg = ", ".join( + f"Culprit rank {error[0]}; {str(error[1])}" for error in errors + ) + logger.info("error msg: %s", error_msg) + logger.info( + "collective stack trace: \n %s", frame_formatter(self.collective_frames) + ) + + def to_collective( + self, + id: int, + errors: set[tuple[int, MatchInfo]] | None = None, + idx_map: dict[int, int] | None = None, + all_entries: dict[int, list[dict[str, Any]]] | None = None, + ) -> Collective: + if not errors: + return Collective( + id=id, + group_id=self.pg_name, + record_id=self.record_id, + pg_desc=self.pg_desc, + pass_check=True, + collective_seq_id=self.collective_seq_id, + p2p_seq_id=self.p2p_seq_id, + collective_name=self.profiling_name, + input_sizes=self.input_sizes, + output_sizes=self.output_sizes, + expected_ranks=self.expected_ranks, + collective_state=self.collective_state, + collective_frames=self.collective_frames, + missing_ranks=getattr(self, "missing_ranks", None), + ) + else: + if idx_map is None: + raise AssertionError("idx_map is None") + if all_entries is None: + raise AssertionError("all_entries is None") + mismatch_collectives = {} + for rank, error in errors: + idx = idx_map[rank] + entry = all_entries[rank][idx] + desc = entry["process_group"][1] + pg_name = entry["process_group"][0] + mismatch_collectives[rank] = Collective( + id=id, + group_id=entry["process_group"][0], + record_id=entry["record_id"], + pg_desc=f"{pg_name}:{desc}" if desc != "undefined" else pg_name, + pass_check=False, + collective_seq_id=entry["collective_seq_id"], + p2p_seq_id=entry["p2p_seq_id"], + collective_name=entry["profiling_name"], + input_sizes=entry["input_sizes"], + output_sizes=entry["output_sizes"], + expected_ranks=self.expected_ranks, + collective_state=entry["state"], + collective_frames=entry.get("frames", []), + type_of_mismatch=error, + ) + return Collective( + id=id, + group_id=self.pg_name, + record_id=self.record_id, + pg_desc=self.pg_desc, + pass_check=False, + collective_seq_id=self.collective_seq_id, + p2p_seq_id=self.p2p_seq_id, + collective_name=self.profiling_name, + input_sizes=self.input_sizes, + output_sizes=self.output_sizes, + expected_ranks=self.expected_ranks, + collective_state=self.collective_state, + collective_frames=self.collective_frames, + input_numel=self.input_numel if hasattr(self, "input_numel") else None, + output_numel=self.output_numel + if hasattr(self, "output_numel") + else None, + missing_ranks=self.missing_ranks + if hasattr(self, "missing_ranks") + else None, + mismatch_collectives=mismatch_collectives, + ) + + def to_nccl_call( + self, + all_entries: dict[int, list[dict[str, Any]]], + idx_map: dict[int, int], + nccl_call_id: int, + collective_id: Any, + ) -> list[NCCLCall]: + result = [] + for i, k in idx_map.items(): + all_entries[i].pop(k) + result.append( + NCCLCall( + id=nccl_call_id, + collective_id=collective_id, + group_id=self.pg_name, # type: ignore[arg-type] + global_rank=i, + traceback_id=0, # type: ignore[arg-type] + collective_type=self.profiling_name, + sizes=self.input_sizes, + ) + ) + nccl_call_id += 1 + return result + + +class Op: + """Parses relevant info about operation out of 'event' dict + + examples of supported `profiling_name`s: + nccl:broadcast + nccl:send 1->2 + nccl:recv 3<-0 + """ + + def __init__( + self, event: dict[Any, Any], memberships: dict[str, set[Any]], pg_name: str + ): + self.profiling_name = event["profiling_name"] + comm_lib_backend, name = self.profiling_name.split(":") + _SUPPORTED_BACKENDS = { + "nccl", + "ncclx", + "xccl", + "gloo", + "rccl", + "rcclx", + "mccl", + "hccl", + } + if comm_lib_backend not in _SUPPORTED_BACKENDS: + raise AssertionError( + f"name formatting error? {comm_lib_backend} not in {_SUPPORTED_BACKENDS}" + ) + parts = name.split(" ") + type = parts[0] + meta = parts[1] if len(parts) == 2 else None + self.state = event["state"] + # Store the hashed pg_name for accessing memberships, and original pg info for display + self.pg_name = pg_name # This is the hashed version used for memberships lookup + self.original_pg_name, self.pg_desc = event["process_group"] + if type not in COLLECTIVES | P2P | {"coalesced"}: + raise AssertionError(f"{type} is not a supported operation") + self.type = type + if type == "send": + if not isinstance(meta, str): + raise AssertionError + s, d = meta.split("->") + self._src, self._dst = int(s), int(d) + elif type == "recv": + if not isinstance(meta, str): + raise AssertionError + d, s = meta.split("<-") + self._dst, self._src = int(d), int(s) + else: + self._src, self._dst = -1, -1 + self._init_global_src_dst(memberships[pg_name]) + self.pg_size = len(memberships[pg_name]) + if type in P2P | COLLECTIVES: + self.input_sizes = event["input_sizes"] + self.output_sizes = event["output_sizes"] + else: + self.input_sizes, self.output_sizes = None, None + self.collective_seq_id = event["collective_seq_id"] + self.stack_id = event.get("stack_id", -1) + self.p2p_seq_id = event["p2p_seq_id"] + self.input_dtypes = event["input_dtypes"] + self.output_dtypes = event["output_dtypes"] + self.time_created_ns = event["time_created_ns"] + self.collective_frames = event.get("frames", []) + self.is_verbose = os.getenv("FR_TRACE_VERBOSE_OUTPUT", "0") == "1" + + def _init_global_src_dst(self, pg_ranks: set[Any]) -> None: + pg_ranks_sorted = sorted(pg_ranks) + self._src_g = pg_ranks_sorted[self._src] if self._src is not None else None + self._dst_g = pg_ranks_sorted[self._dst] if self._dst is not None else None + + @property + def src(self) -> int: + if self.type not in P2P: + raise AssertionError("can't get src of non-p2p op") + return self._src + + @property + def dst(self) -> int: + if self.type not in P2P: + raise AssertionError("can't get dst of non-p2p op") + return self._dst + + def __repr__(self) -> str: + p2p_info = "" + if self.type in P2P: + p2p_info = f"s={self._src_g} d={self._dst_g}" + if self.is_verbose: + verbose_info = ( + f"timestamp_created={self.time_created_ns}", + p2p_info, + f"input_sizes={self.input_sizes}", + f"output_sizes={self.output_sizes}", + f"input_dtypes={self.input_dtypes}", + f"output_dtypes={self.output_dtypes}", + "collective_seq_id | p2p_seq_id=" + f"{self.p2p_seq_id if self.type in P2P else self.collective_seq_id}", + f"pg_name={self.pg_name}", + f"pg_description={self.pg_desc}", + f"pg_size={self.pg_size}", + f"stack_id={self.stack_id}", + f"state={self.state}", + ) + return f"{self.type}(%s)" % ", ".join(s for s in verbose_info if s) + return f"{self.type}(%sinput_sizes={self.input_sizes}, state={self.state})" % ( + f"{p2p_info}, " if p2p_info else "" + ) + + def dtype_mismatch(self, other: "Op") -> bool: + if ( + ( + self.type not in ["scatter", "gather", "broadcast"] + and set(self.input_dtypes) != set(self.output_dtypes) + and self.input_sizes[0] + and self.output_sizes[0] + ) + or ( + self.type not in ["scatter", "broadcast"] + and set(self.input_dtypes) != set(other.input_dtypes) + and self.input_sizes[0] + and other.input_sizes[0] + ) + or ( + self.type != "gather" + and set(self.output_dtypes) != set(other.output_dtypes) + and self.output_sizes[0] + and other.output_sizes[0] + ) + ): + return True + return False + + def match(self, other: "Op") -> MatchInfo: + # TODO: I think this can validly not match, + # e.g. if one PG was used for p2p ops between only some of the peers? + # if self.seq_id != other.seq_id: + # return False + + if self.type == "send": + # TODO: We need more states for p2p ops. + return ( + MatchInfo(MatchState.FULLY_MATCHED) + if ( + other.type == "recv" + and self.src == other.src + and self.dst == other.dst + and self.input_sizes == other.output_sizes + ) + else MatchInfo(MatchState.SIZE_OR_SYNTAX_MISMATCH) + ) + elif self.type == "recv": + return ( + MatchInfo(MatchState.FULLY_MATCHED) + if ( + other.type == "send" + and self.src == other.src + and self.dst == other.dst + and self.output_sizes == other.input_sizes + ) + else MatchInfo(MatchState.SIZE_OR_SYNTAX_MISMATCH) + ) + elif self.type in COLLECTIVES: + if self.type != other.type: + return MatchInfo( + MatchState.COLLECTIVE_TYPE_MISMATCH, + f"Expected collective type: '{self.type}' does not match found collective type: '{other.type}'", + ) + if ( + self.type not in ["all_to_all", "scatter"] + and self.input_sizes != other.input_sizes + ): + return MatchInfo( + MatchState.SIZE_OR_SYNTAX_MISMATCH, + f"Expected input sizes: '{self.input_sizes}' does not match found input sizes: " + f"'{other.input_sizes}'", + ) + if ( + self.type not in ["all_to_all", "gather"] + and self.output_sizes != other.output_sizes + ): + return MatchInfo( + MatchState.SIZE_OR_SYNTAX_MISMATCH, + f"Expected output sizes: '{self.output_sizes}' does not match found output sizes: " + f"'{other.output_sizes}'", + ) + if ( + self.type in ["all_reduce", "allreduce_coalesced"] + and self.input_sizes != other.output_sizes + ): + return MatchInfo( + MatchState.SIZE_OR_SYNTAX_MISMATCH, + f"Expected input sizes: '{self.input_sizes}' does not match found output sizes: '{other.output_sizes}'", + ) + if ( + self.type + in [ + "all_gather", + "all_gather_base", + "all_gather_into_tensor_coalesced", + ] + and math.prod(other.output_sizes[0]) + != math.prod(self.input_sizes[0]) * self.pg_size + ): + return MatchInfo( + MatchState.SIZE_OR_SYNTAX_MISMATCH, + f"Found input numel '{math.prod(other.input_sizes[0])} * pg size {self.pg_size}' " + f"does not match output numel '{math.prod(other.output_sizes[0])}'", + ) + if ( + self.type + in [ + "reduce_scatter", + "_reduce_scatter_base", + "reduce_scatter_tensor_coalesced", + ] + and math.prod(other.input_sizes[0]) + != math.prod(self.output_sizes[0]) * self.pg_size + ): + return MatchInfo( + MatchState.SIZE_OR_SYNTAX_MISMATCH, + f"Found input numel '{math.prod(other.input_sizes[0])}' does not match output numel " + f"'{math.prod(other.output_sizes[0])} * pg size {self.pg_size}'", + ) + if self.dtype_mismatch(other): + return MatchInfo( + MatchState.COLLECTIVE_DTYPE_MISMATCH, + f"Expected dtypes: '{set(self.input_dtypes)}' does not " + f"match found dtype: '{set(self.output_dtypes)}/" + f"{set(other.input_dtypes)}/{set(other.output_dtypes)}'", + ) + if self.state != other.state: + # MatchState() + return MatchInfo( + MatchState.COLLECTIVE_STATE_MISMATCH, + f"Expected state: '{self.state}' does not match found state: '{other.state}'", + ) + if self.type == "all_to_all": + return MatchInfo(MatchState.UNDECIDED) + elif self.type in [ + "coalesced", + "ALLGATHER_coalesced", + "REDUCE_SCATTER_coalesced", + ]: + return ( + MatchInfo(MatchState.FULLY_MATCHED) + if (other.type == self.type) + else MatchInfo(MatchState.SIZE_OR_SYNTAX_MISMATCH) + ) + return MatchInfo(MatchState.FULLY_MATCHED) + + +class MatchStateRecord: + def __init__( + self, + expected_ranks: set[int], + other_ranks: list[int], + entry_state: EntryState, + candidate_ranks: set[int], + candidate_idx: dict[int, int], + found_ranks: set[int], + found_idx: dict[int, int], + errors: set[tuple[int, MatchInfo]], + ) -> None: + self.expected_ranks = expected_ranks + self.other_ranks = other_ranks + self.entry_state = entry_state + self.candidate_ranks = candidate_ranks + self.candidate_idx = candidate_idx + self.found_ranks = found_ranks + self.found_idx = found_idx + self.errors = errors + self.has_undecided_case = False + + def reset_for_coalesced( + self, entry_state: EntryState, candidate_ranks: set[int] + ) -> None: + self.entry_state = entry_state + self.candidate_ranks = candidate_ranks + self.candidate_idx = {} + self.found_ranks = set() + self.found_idx = {} + self.errors = set() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..19d7e658c30f3874c8bc7f8a8607df2f99c78cbd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/components/utils.py @@ -0,0 +1,791 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import argparse +import math +from typing import Any + +from torch.distributed.flight_recorder.components.fr_logger import FlightRecorderLogger +from torch.distributed.flight_recorder.components.types import ( + Collective, + EntryState, + Group, + MatchInfo, + MatchState, + MatchStateRecord, + Membership, + Op, + P2P, +) + + +__all__ = [ + "add_stack_id_in_entries", + "align_trace_from_beginning", + "check_current_entry_match", + "check_no_missing_dump_files", + "check_version", + "error_analysis", + "find_coalesced_group", + "find_coalesced_group_with_non_p2p", + "get_version_detail", + "just_print_entries", + "match_coalesced_groups_with_non_p2p", + "match_coalesced_groups", + "format_frame", + "format_frames", + "match_one_event", + "check_size_alltoall", +] + +logger: FlightRecorderLogger = FlightRecorderLogger() + + +try: + from tabulate import tabulate +except ModuleNotFoundError: + logger.debug("tabulate is not installed. Proceeding without it.") + + +def format_frame(frame: dict[str, str]) -> str: + name = frame["name"] + filename = frame["filename"] + line = frame["line"] + return f"{name} at {filename}:{line}" + + +def format_frames(frames: list[dict[str, str]]) -> str: + formatted_frames = [] + for frame in frames: + formatted_frames.append(format_frame(frame)) + return "\n".join(formatted_frames) + + +def match_one_event( + event_a: dict[Any, Any], + event_b: dict[Any, Any], + memberships: dict[str, set[Any]], + pg_name: str, +) -> MatchInfo: + op_a = Op(event_a, memberships, pg_name) + op_b = Op(event_b, memberships, pg_name) + return op_a.match(op_b) + + +def match_coalesced_groups( + all_rank_events: dict[Any, Any], + group_size: int, + groups: dict[str, Group], + memberships: dict[str, set[Any]], + _pg_guids: dict[tuple[str, int], str], +) -> bool: + """ + all_rank_events: { + rank: [ + (idx, event_dict) + ] + } + + Note: it is possible for event dicts in a coalesced group to be asymmetric. + e.g. the following events lists form a valid coalescing group + events0 [send:1] + events1 [recv:0, send:2] + events2 [recv:1] + + Rule 1: all ops should find a match + Rule 2: relative ordering of sends and recvs in one event list can be arbitrary + e.g. + events1 [recv:0, send:2] —> okay + events1 [send:2, recv:0] —> also okay + Rule 3: sends to the same dest or recvs from the src should be in a consistent order + e.g. + rank0 [send:1 (100B), send:1 (1000B)] + rank1 [recv:0 (1000B), recv:0 (100B)] —> not okay + """ + all_ops = { + rank: [ + Op(e, memberships, _pg_guids[(e["process_group"][0], rank)]) + for i, e in all_rank_events[rank] + ] + for rank in all_rank_events + } + + def visualize_ops( + match: bool, + _pg_guids: dict[tuple[str, int], str], + ) -> None: + all_ops = { + rank: [ + Op(e, memberships, _pg_guids[(e["process_group"][0], rank)]) + for i, e in all_rank_events[rank] + ] + for rank in all_rank_events + } + + i = 0 + row = [] + progress = True + table = [] + while progress: + progress = False + for r in all_ops: + if len(all_ops[r]) > i: + rank, event = all_rank_events[r][i] + # Check if the pg_guid exists for this rank and process group + pg_key = (event["process_group"][0], rank) + if pg_key in _pg_guids: + row.append( + Op( + event, + memberships, + _pg_guids[pg_key], + ) + ) + else: + # Skip this entry if pg_guid mapping doesn't exist + row.append(None) # type: ignore[arg-type] + progress = True + else: + row.append(None) # type: ignore[arg-type] + table.append(row) + row = [] + i += 1 + title = "Match" if match else "MISMATCH" + logger.info("%s \n", title) + logger.info("%s", tabulate(table)) # type: ignore[operator] + + # TODO can't verify seq_id bc there might have been valid seq deltas between ranks even within a pg. + for op_list in all_ops.values(): + if not op_list: + # print("TODO- not sure if its valid for only some ranks in a PG to participate in a coalesced op?") + return False + if op_list[-1].type != "coalesced": + raise AssertionError + op_list.pop(-1) + + while all_ops: + first_rank = next(iter(all_ops)) + my_ops = all_ops[first_rank] + + if len(all_ops[first_rank]) == 0: + all_ops.pop(first_rank) + continue + + # lets match the first collective! we need to know which ranks are involved, and ensure that this same + # collective is also the first one on those ranks within that group + op = my_ops[0] + match_idx = -1 + if op.type in P2P: + dst_global_rank = sorted(memberships[op.pg_name])[op.dst] + peer_ops = all_ops[dst_global_rank] + for i, other in enumerate(peer_ops): + if op.match(other).state == MatchState.FULLY_MATCHED: + match_idx = i + break + elif op.dst == other.src: + # Rule 3 + break + else: + # Rule 1 + continue + else: + raise NotImplementedError("coalesced collective ops") + if match_idx >= 0: + my_ops.pop(0) + peer_ops.pop(match_idx) + else: + visualize_ops(False, _pg_guids) + return False + + visualize_ops(True, _pg_guids) + return True + + +# We enabled the creating FR entry for non-P2P slow path collective ops in v2.7. +def match_coalesced_groups_with_non_p2p( + all_rank_events: dict[Any, Any], + pg_info: tuple[str, str], + memberships: dict[str, set[Any]], + _pg_guids: dict[tuple[str, int], str], + mismatch: dict[str, int], + dumps_ranks: set[int], + version: str, + collectives: list[Collective], + match_record: MatchStateRecord, +) -> bool: + """ + all_rank_events: { + rank: [ + (idx, event_dict) + ] + } + + Note: it is possible for event dicts in a coalesced group to be asymmetric. + e.g. the following events lists form a valid coalescing group + events0 [send:1] + events1 [recv:0, send:2] + events2 [recv:1] + + Rule 1: all ops should find a match + Rule 2: relative ordering of sends and recvs in one event list can be arbitrary + e.g. + events1 [recv:0, send:2] —> okay + events1 [send:2, recv:0] —> also okay + Rule 3: sends to the same dest or recvs from the src should be in a consistent order + e.g. + rank0 [send:1 (100B), send:1 (1000B)] + rank1 [recv:0 (1000B), recv:0 (100B)] —> not okay + """ + all_ops = { + rank: [ + Op(e, memberships, _pg_guids[(e["process_group"][0], rank)]) + for _, e in all_rank_events[rank] + ] + for rank in all_rank_events + } + is_p2p = any(op.type in P2P for ops in all_ops.values() for op in ops) + pg_name = pg_info[0] + + def visualize_ops( + match: bool, + _pg_guids: dict[tuple[str, int], str], + ) -> None: + all_ops = { + rank: [ + Op(e, memberships, _pg_guids[(e["process_group"][0], rank)]) + for _, e in all_rank_events[rank] + ] + for rank in all_rank_events + } + + i = 0 + row = [] + progress = True + table = [] + while progress: + progress = False + for r in all_ops: + if len(all_ops[r]) > i: + rank, event = all_rank_events[r][i] + # Check if the pg_guid exists for this rank and process group + pg_key = (event["process_group"][0], rank) + if pg_key in _pg_guids: + row.append( + Op( + event, + memberships, + _pg_guids[pg_key], + ) + ) + else: + # Skip this entry if pg_guid mapping doesn't exist + row.append(None) # type: ignore[arg-type] + progress = True + else: + row.append(None) # type: ignore[arg-type] + table.append(row) + row = [] + i += 1 + title = "Match" if match else "MISMATCH" + logger.info("%s \n", title) + logger.info("%s", tabulate(table)) # type: ignore[operator] + + # TODO Need to verify no seq_id deltas for P2P ops. + for rank, op_list in all_ops.items(): + if not op_list: + logger.error("Rank %s has an empty op list.", rank) + continue + if op_list[-1].type == "coalesced" and is_p2p: + op_list.pop(-1) + + while all_ops: + first_rank = next(iter(all_ops)) + my_ops = all_ops[first_rank] + + if len(all_ops[first_rank]) == 0: + all_ops.pop(first_rank) + continue + + # lets match the first collective! we need to know which ranks are involved, and ensure that this same + # collective is also the first one on those ranks within that group + op = my_ops[0] + match_idx = -1 + if is_p2p: + dst_global_rank = sorted(memberships[op.pg_name])[op.dst] + peer_ops = all_ops[dst_global_rank] + for i, other in enumerate(peer_ops): + if op.match(other).state == MatchState.FULLY_MATCHED: + match_idx = i + break + elif op.dst == other.src: + # Rule 3 + break + else: + # Rule 1 + continue + if match_idx >= 0: + my_ops.pop(0) + peer_ops.pop(match_idx) + else: + visualize_ops(False, _pg_guids) + return False + else: + all_coalesced_entries = { + rank: [e for _, e in all_rank_events[rank]] for rank in all_rank_events + } + current_entry = all_coalesced_entries[first_rank][0] + my_ops.pop(0) + + match_record.reset_for_coalesced( + EntryState(current_entry, match_record.expected_ranks), + {first_rank}, + ) + + # Iterate through all the ranks and check if there is a mismatch for the current entry. + check_current_entry_match( + all_coalesced_entries, + _pg_guids, + pg_info, + current_entry, + memberships, + mismatch, + match_record, + ) + + # Use heuristics to decide what type of errors and error messages we should print. + error_analysis( + all_coalesced_entries, + match_record, + dumps_ranks, + first_rank, + current_entry, + mismatch, + get_version_detail(version), + pg_info[0], + ) + + # TODO: For now, we only check the correctness of individual collective within a coalesced one in + # this script. We need to merge (e.g, input/output sizes) together + # for downstream consumer. + + # at this point there are 3 possibilities + # 1. we found a match on all the ranks that are members of the group + # -> we create a Collective and remove the individual entries from their original lists + if ( + match_record.found_ranks == match_record.expected_ranks + and mismatch[pg_name] == 0 + ): + # Just pop out this collective. + idx_map = { + r: match_record.found_idx[r] if r != first_rank else 0 + for r in match_record.found_ranks + } + for i, k in idx_map.items(): + all_rank_events[i].pop(k) + for r in match_record.found_ranks: + if r != first_rank: + all_ops[r].pop(0) + + # 2. we found a partial match but some ranks are missing + # 3. we found no match + # -> since its not a complete collective, no entry goes into collectives but we still record a nccl call + else: + logger.debug("Non-matching collective inside coalesced group") + idx_map = { + r: match_record.candidate_idx[r] if r != first_rank else 0 + for r in match_record.candidate_ranks + } + collectives.append( + match_record.entry_state.to_collective( + len(collectives), + errors=match_record.errors, + idx_map=idx_map, + all_entries=all_coalesced_entries, + ) + ) + return False + + if is_p2p: + visualize_ops(True, _pg_guids) + return True + + +def check_size_alltoall(alltoall_cases: list[dict[str, Any]]) -> tuple[bool, int, int]: + input_numel = 0 + output_numel = 0 + for e in alltoall_cases: + input_numel += math.prod(e["input_sizes"][0]) + output_numel += math.prod(e["output_sizes"][0]) + return input_numel != output_numel, input_numel, output_numel + + +def check_current_entry_match( + all_entries: dict[int, list[dict[str, Any]]], + _pg_guids: dict[tuple[str, int], str], + pg_info: tuple[str, str], + current_entry: dict[str, Any], + _memberships: dict[str, set[Any]], + mismatch: dict[str, int], + match_record: MatchStateRecord, +) -> None: + pg_name, desc = pg_info[0], pg_info[1] + for o in match_record.expected_ranks.intersection(set(match_record.other_ranks)): + for i, e in enumerate(all_entries[o]): # type: ignore[index] + # step over ops from other PGs + # only check match state when seq_id matches + if ( + _pg_guids[(e["process_group"][0], o)] == pg_name + and e["process_group"][1] == desc + and e["collective_seq_id"] == match_record.entry_state.collective_seq_id + ): + match_info = match_one_event(current_entry, e, _memberships, pg_name) + if ( + match_info.state in [MatchState.FULLY_MATCHED, MatchState.UNDECIDED] + and mismatch[pg_name] == 0 + ): + match_record.found_ranks.add(o) + match_record.found_idx[o] = i + match_record.has_undecided_case = ( + match_info.state == MatchState.UNDECIDED + ) + else: + match_record.candidate_ranks.add(o) + match_record.candidate_idx[o] = i + if match_info.state not in [ + MatchState.FULLY_MATCHED, + MatchState.UNDECIDED, + ]: + # Here we assume the current rank is not the source of the error. + # But it's possible that the current rank is the culprit, then users will + # see lots of normal ranks reported as culprit. + # TODO: we need to figure out a better way to handle the case mentioned above. + match_record.errors.add((o, match_info)) + break + + +def error_analysis( + all_entries: dict[int, list[dict[str, Any]]], + match_record: MatchStateRecord, + dumps_ranks: set[int], + first_rank: int, + current_entry: dict[str, Any], + mismatch: dict[str, int], + version: tuple[int, int], + pg_name: str, +) -> None: + major_v, minor_v = version[0], version[1] + # case one: not every rank join the collective or in the flight recorder. + if ( + match_record.candidate_ranks | match_record.found_ranks + ) != match_record.expected_ranks and match_record.expected_ranks - ( + match_record.candidate_ranks | match_record.found_ranks + ) <= dumps_ranks: + mismatch[pg_name] += 1 + logger_msg = "Not all ranks joining collective, sequence number: %s" + missing_ranks = match_record.expected_ranks - ( + match_record.candidate_ranks | match_record.found_ranks + ) + match_record.entry_state.log( + logger, logger_msg, format_frames, missing_ranks=missing_ranks + ) + match_record.candidate_ranks.update(match_record.found_ranks) + match_record.candidate_idx.update(match_record.found_idx) + match_record.found_idx.clear() + match_record.found_ranks.clear() + # We didn't see any mismatch and all expected ranks are in the dump. + elif len( + match_record.candidate_ranks + ) == 1 and match_record.expected_ranks.issubset(dumps_ranks): + # case two: alltoall or alltoall_base case. + if match_record.has_undecided_case: + alltoall_cases = [current_entry] + [ + all_entries[o][match_record.found_idx[o]] + for o in match_record.found_ranks + ] + fail_check, total_input_numel, total_output_numel = check_size_alltoall( + alltoall_cases + ) + if major_v <= 2 and minor_v <= 3: + # We don't log the input/output sizes for alltoall before v2.4, + # so we don't consider the size mismatch as an error for now. + fail_check = False + if fail_check: + # When we see errors in all_to_all, it's hard to tell which rank is the source of the error. + mismatch[pg_name] += 1 + logger_msg = ( + "Input/output mismatch in the collective sequence number: %s" + ) + match_record.entry_state.log( + logger, + logger_msg, + format_frames, + total_numel=(total_input_numel, total_output_numel), + ) + match_record.candidate_ranks.update(match_record.found_ranks) + match_record.candidate_idx.update(match_record.found_idx) + match_record.found_idx.clear() + match_record.found_ranks.clear() + match_record.errors.add( + (first_rank, MatchInfo(MatchState.SIZE_OR_SYNTAX_MISMATCH)) + ) + else: + match_record.found_ranks.update(match_record.candidate_ranks) + match_record.found_idx.update(match_record.candidate_idx) + match_record.candidate_idx.clear() + match_record.candidate_ranks.clear() + # case three: all joined and everything matches on all ranks. + else: + match_record.found_ranks.update(match_record.candidate_ranks) + match_record.found_idx.update(match_record.candidate_idx) + match_record.candidate_idx.clear() + match_record.candidate_ranks.clear() + # case four: mismatch cases due to not same type, size mismatch or state mismatch. + elif len(match_record.errors) > 0: + mismatch[pg_name] += 1 + logger_msg = "Collective sequence number: %s has errors" + match_record.entry_state.log( + logger, logger_msg, format_frames, errors=match_record.errors + ) + match_record.candidate_ranks.update(match_record.found_ranks) + match_record.candidate_idx.update(match_record.found_idx) + match_record.found_idx.clear() + match_record.found_ranks.clear() + # partial analysis case when we cannot decide what's wrong with this collective entry. + else: + match_record.candidate_ranks.update(match_record.found_ranks) + match_record.candidate_idx.update(match_record.found_idx) + match_record.found_idx.clear() + match_record.found_ranks.clear() + # if any element in expected_ranks not in dumps_ranks. + if match_record.expected_ranks - dumps_ranks: + mismatch[pg_name] += 1 + logger.info( + "We cannot decide what's wrong with this collective entry " + "because we missed FR dumps from ranks (%s) so we don't have enough " + "information. If you want to debug further use -j to dump all raw trace", + str(match_record.expected_ranks - dumps_ranks), + ) + else: + logger.info( + "No errors found for this collective entry, There could be some " + "other reasons why we see collective timeout." + ) + + +def find_coalesced_group( + pg_name: str, + entries: list[dict[str, Any]], + _pg_guids: dict[tuple[str, int], str], + rank: int, +) -> list[tuple[int, dict[str, Any]]]: + """Given a list of entries, if the collective_seq_id of the first entry matches that of subsequent ones, + build and return a list of entries terminating in a 'coalesced' op entry all sharing a collective_seq_id + """ + found = [] + collective_seq_id = None + for i, e in enumerate(entries): + if _pg_guids[(e["process_group"][0], rank)] != pg_name: + continue + elif collective_seq_id is None: + collective_seq_id = ( + e["p2p_seq_id"] if e["is_p2p"] else e["collective_seq_id"] + ) + found.append((i, e)) + elif not e["is_p2p"] and e["collective_seq_id"] == collective_seq_id: + found.append((i, e)) + elif e["is_p2p"] and e["p2p_seq_id"] == collective_seq_id: + found.append((i, e)) + else: + break + + if len(found) > 1: + if found[-1][1]["profiling_name"] != "nccl:coalesced": + raise AssertionError + return found + return [] + + +# We enabled the creating FR entry for non-P2P slow path collective ops in v2.7. +def find_coalesced_group_with_non_p2p( + pg_name: str, + entries: list[dict[str, Any]], + _pg_guids: dict[tuple[str, int], str], + rank: int, +) -> list[tuple[int, dict[str, Any]]]: + """Given a list of entries, if the collective_seq_id of the first entry matches that of subsequent ones, + build and return a list of entries terminating in a 'coalesced' op entry all sharing a collective_seq_id + """ + found = [] + collective_seq_id = None + for i, e in enumerate(entries): + if _pg_guids[(e["process_group"][0], rank)] != pg_name: + continue + elif collective_seq_id is None: + collective_seq_id = ( + e["p2p_seq_id"] if e["is_p2p"] else e["collective_seq_id"] + ) + found.append((i, e)) + elif not e["is_p2p"] and e["collective_seq_id"] == collective_seq_id: + found.append((i, e)) + elif e["is_p2p"] and e["p2p_seq_id"] == collective_seq_id: + found.append((i, e)) + else: + break + + if len(found) > 1: + name = found[-1][1]["profiling_name"] + if name.startswith("nccl:") and not name.endswith("_coalesced"): + logger.error("Rank %s does not have a coalesced end.", rank) + return found + return [] + + +def just_print_entries( + all_entries: dict[int, list[dict[str, Any]]], + _groups: dict[str, Group], + _memberships: dict[str, set[Any]], + _pg_guids: dict[tuple[str, int], str], + args: argparse.Namespace, + stack_id_trace_map: dict[str, int], +) -> None: + rows = [] + ranks = sorted(all_entries.keys()) + headers = [ + f"Rank {rank}" + for rank in ranks + if args.selected_ranks is None or rank in args.selected_ranks + ] + progress = True + while progress: + progress = False + row = [] + for rank in ranks: + if args.selected_ranks is not None and rank not in args.selected_ranks: + continue + if len(all_entries[rank]) == 0: + row.append("") + else: + entry = all_entries[rank].pop(0) + pg_name = _pg_guids[(entry["process_group"][0], rank)] + if ( + args.pg_filters is None + or entry["process_group"][1] in args.pg_filters + or entry["process_group"][0] in args.pg_filters + ): + row.append(str(Op(entry, _memberships, pg_name))) + else: + row.append("") + progress = True + if progress: + rows.append(row) + + logger.info(tabulate(rows, headers=headers)) + + if stack_id_trace_map and args.print_stack_trace: + headers = ["stack_id", "frame_stack"] + rows = [] + + for frame, stack_id in sorted( + stack_id_trace_map.items(), key=lambda item: item[1] + ): + rows.append([str(stack_id), frame]) + + logger.info(tabulate(rows, headers=headers)) + + +def check_no_missing_dump_files( + entries: dict[int, Any], memberships: list[Membership] +) -> None: + all_ranks = {int(membership.global_rank) for membership in memberships} + dumps_ranks = {int(key) for key in entries} + missing = all_ranks - dumps_ranks + if len(missing) != 0: + raise AssertionError(f"Missing dump files from ranks {missing}") + + +def check_version(version_by_ranks: dict[str, str], version: str) -> None: + for rank, v in version_by_ranks.items(): + if v != version: + raise AssertionError( + f"Rank {rank} has different version {v} from the given version {version}" + ) + + +def get_version_detail(version: str) -> tuple[int, int]: + # pyrefly: ignore [bad-assignment] + version = version.split(".") + if len(version) != 2: + raise AssertionError(f"Invalid version {version}") + major, minor = map(int, version) + return major, minor + + +def add_stack_id_in_entries( + entries: dict[int, list[dict[str, Any]]], +) -> tuple[dict[int, list[dict[str, Any]]], dict[str, int]]: + stack_id = 0 + stack_id_trace_map = {} + for rank in entries: + for dump in entries[rank]: + if dump.get("frames", []): + frames = str(dump["frames"]) + if frames not in stack_id_trace_map: + stack_id_trace_map[frames] = stack_id + dump["stack_id"] = stack_id + stack_id += 1 + else: + dump["stack_id"] = stack_id_trace_map[frames] + else: + dump["stack_id"] = -1 + + return entries, stack_id_trace_map + + +def align_trace_from_beginning( + entries: dict[int, list[dict[str, Any]]], +) -> dict[int, list[dict[str, Any]]]: + """ + Align the trace entries by record ID for entries. + This function takes a dictionary of rank names to lists of trace entries as input. + Each trace entry is a dictionary containing information about a collective operation, + including its unique identifier (`record_id` is monotonically increasing as we write into the ring buffer). + The function finds the largest starting point across all ranks by taking the maximum + `record_id` value of the first entry in each rank. Finally, it filters out any + entries with `record_id` values less than the maximum starting point. + The function returns the updated dictionary of sorted and filtered trace entries. + + Args: + entries (Dict[str, List[Dict[str, Any]]]): A dictionary of rank names to lists of trace entries. + + Returns: + entries (Dict[str, List[Dict[str, Any]]]): Entries sorted by record ID and filtered by the maximum starting point. + """ + + maximum_starting_record_id = 0 + for rank in entries: + # Although this is a ring buffer, we already sort the entries by `record_id` when dumping, we just + # need to find the largest starting point. For example, if the buffer has the following entries: + # Rank 0: [0, 1, 2, 3, 4, 5, 6] + # Rank 1: [1, 2, 3, 4, 5, 6, 7] + # Rank 2: [2, 3, 4, 5, 6, 7, 8] + # Rank 3: [0, 1, 2, 3, 4, 5, None] + # Then we should start from collective 2 not 0 because any collective before, + # we don't have complete records from all ranks so we need to ignore them. + # If we don't have any trace from some ranks, ignore them + # as well. + if len(entries[rank]) == 0: + continue + first_record_id = entries[rank][0]["record_id"] + maximum_starting_record_id = max(maximum_starting_record_id, first_record_id) + + for rank in entries: + entries[rank] = [ + entry + for entry in entries[rank] + if entry["record_id"] >= maximum_starting_record_id + ] + + return entries diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/fr_trace.py b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/fr_trace.py new file mode 100644 index 0000000000000000000000000000000000000000..29bc47cf13b4d6d646f833262e17a2ecf9d36156 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/flight_recorder/fr_trace.py @@ -0,0 +1,69 @@ +#!/usr/bin/env python3 +"""Flight Recorder Trace Analyzer + +This script primarily merges data from individual flight recorder buffers from individual ranks in a +PyTorch Distributed program into a flattened database format that can be used for further analysis. + +However as part of the merging process, it is necessary to perform some analysis in order to match operators +on one rank with corresponding operators on other ranks and register them as one 'collective' entry. During this +process, a significant amount of useful information can already be extracted such as where the first mismatch occurs +in cases of desync (when not all ranks issue a compatible collective in a particular process group). + + +Not Yet Implemented +- TODO- tracebacks aren't implemented + +Known Issues +- Flight Recorder buffer sequence_id information is not sufficient to match collectives and coalesced collectives + unless we have the trace data from the beginning of the program. To enable confident analysis of trace buffers that + do not start from zero (and to simplify the script's matching logic) we need to add more information to the recorder. +- Currently, the script omits checking the 'status' of collectives. We can look for the first 'non completed' + collective easily enough and report that. + +Usage +python fr_trace.py [-o ] + +- Omitting the optional output file will still yield analysis information to stdout +- The output file is a pickle of the flat DB, which may change in format in the future. +- This script is versioned so that we can ensure our future changes to flight recorder are backwards compatible. +""" + +import pickle +from collections.abc import Sequence + +from torch.distributed.flight_recorder.components.builder import build_db, transform_ft +from torch.distributed.flight_recorder.components.config_manager import JobConfig +from torch.distributed.flight_recorder.components.loader import read_dir +from torch.distributed.flight_recorder.components.types import types + + +__all__ = ["main"] + + +def main(args: Sequence[str] | None = None) -> None: + config = JobConfig() + # pyrefly: ignore [bad-assignment] + args = config.parse_args(args) + # pyrefly: ignore [missing-attribute] + if not args.trace_dir: + raise AssertionError("Trace directory trace_dir is required") + # pyrefly: ignore [bad-argument-type] + details, version = read_dir(args) + # pyrefly: ignore [missing-attribute] + if args.transform_ft: + # pyrefly: ignore [missing-attribute] + if not args.group_world_size: + raise AssertionError("World size is required for transform_ft") + # pyrefly: ignore [bad-argument-type] + details = transform_ft(details, args.group_world_size) + # pyrefly: ignore [bad-argument-type] + db = build_db(details, args, version) + # pyrefly: ignore [missing-attribute] + if args.output: + # pyrefly: ignore [no-matching-overload] + with open(args.output, "wb") as f: + pickle.dump((types, db), f) + + +if __name__ == "__main__": + main() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..7c0d55e37d688324b8548f77bdf065fa52a75b83 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__init__.py @@ -0,0 +1,72 @@ +from ._flat_param import FlatParameter as FlatParameter +from ._fully_shard import ( + CPUOffloadPolicy, + DataParallelMeshDims, + FSDPModule, + fully_shard, + MixedPrecisionPolicy, + OffloadPolicy, + register_fsdp_forward_method, + share_comm_ctx, + UnshardHandle, +) +from .fully_sharded_data_parallel import ( + BackwardPrefetch, + CPUOffload, + FullOptimStateDictConfig, + FullStateDictConfig, + FullyShardedDataParallel, + LocalOptimStateDictConfig, + LocalStateDictConfig, + MixedPrecision, + OptimStateDictConfig, + OptimStateKeyType, + ShardedOptimStateDictConfig, + ShardedStateDictConfig, + ShardingStrategy, + StateDictConfig, + StateDictSettings, + StateDictType, +) + + +__all__ = [ + # FSDP1 + "BackwardPrefetch", + "CPUOffload", + "FullOptimStateDictConfig", + "FullStateDictConfig", + "FullyShardedDataParallel", + "LocalOptimStateDictConfig", + "LocalStateDictConfig", + "MixedPrecision", + "OptimStateDictConfig", + "OptimStateKeyType", + "ShardedOptimStateDictConfig", + "ShardedStateDictConfig", + "ShardingStrategy", + "StateDictConfig", + "StateDictSettings", + "StateDictType", + # FSDP2 + "CPUOffloadPolicy", + "DataParallelMeshDims", + "FSDPModule", + "fully_shard", + "MixedPrecisionPolicy", + "OffloadPolicy", + "register_fsdp_forward_method", + "UnshardHandle", + "share_comm_ctx", +] + +# Set namespace for exposed private names +CPUOffloadPolicy.__module__ = "torch.distributed.fsdp" +DataParallelMeshDims.__module__ = "torch.distributed.fsdp" +FSDPModule.__module__ = "torch.distributed.fsdp" +fully_shard.__module__ = "torch.distributed.fsdp" +MixedPrecisionPolicy.__module__ = "torch.distributed.fsdp" +OffloadPolicy.__module__ = "torch.distributed.fsdp" +register_fsdp_forward_method.__module__ = "torch.distributed.fsdp" +UnshardHandle.__module__ = "torch.distributed.fsdp" +share_comm_ctx.__module__ = "torch.distributed.fsdp" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..39634d36ab6b6c6733d77d9e4ca891e64a7c9ce2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_common_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_common_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bdeeaf0df2a7a46be5dea58a489d5150a683b4a8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_common_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_debug_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_debug_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8f380971fbe4e911636e42ef8ce8cd5e2fd06ff0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_debug_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_dynamo_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_dynamo_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1f1b646a6b8b7047d16f50dd06ca127547050623 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_dynamo_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_exec_order_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_exec_order_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2805cad9477172f4aebcdee529fb7f0729776677 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_exec_order_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_flat_param.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_flat_param.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3e33578621bc27138bcab429a4320aa96bbc597a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_flat_param.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ba67e7fbd58a6ac194a1cc8a0dfec8a5450389e1c6c6b819976f607b150539a +size 116811 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_fsdp_extensions.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_fsdp_extensions.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b497997db46d88c3645da1c0ab119a3a30e05858 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_fsdp_extensions.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_init_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_init_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9e8ef88332d5510e7060eb007e0c2777de6a73f0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_init_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_limiter_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_limiter_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6d776973609b373c3ea89e2faa13d792feda7666 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_limiter_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_optim_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_optim_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0eb761e122b7e71e18892663812d8a70401b871d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_optim_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_runtime_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_runtime_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc36752e0e7f7de53500c4e9d1937a0eb738701f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_runtime_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_shard_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_shard_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ed501f1feedd0dffc5e5eb53f5b9c3604a546067 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_shard_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_state_dict_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_state_dict_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8d51bca1f7be5affebd9d68260b38e947575b67c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_state_dict_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_trace_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_trace_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6bff0d7cd173b7c4445ca25e290365cb1199a6ea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_trace_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_traversal_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_traversal_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..db04ce0a72d19d6352fe5e8177a8f31790111f5f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_traversal_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_unshard_param_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_unshard_param_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3fb95572bbb59971be4e0ec133a4ae77f7abf14c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_unshard_param_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_wrap_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_wrap_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bb0ae5fd654a34d24c0a0afc8edd8fb795fae359 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/_wrap_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fe4c80aef5b8ded0404556d878f3ed1e7040c80b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/fully_sharded_data_parallel.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/fully_sharded_data_parallel.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c5fc18d7c5c0057975b4a0c1b7a04bad1566ac7f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/fully_sharded_data_parallel.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45f5b7a3a1d7f7109740c4d8287606e7527cc3ea87d321ae0bba72420b9ccad0 +size 101163 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/sharded_grad_scaler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/sharded_grad_scaler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d3b50b41ba7b60991e3899d051c641c0e1c058cd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/sharded_grad_scaler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/wrap.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/wrap.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7b7a2fd32d99ba32c8cc1f9357a001509102dfc1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/__pycache__/wrap.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_common_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_common_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..ad6d543fc97f5c4377e44a3f3afcd70c4a7886ec --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_common_utils.py @@ -0,0 +1,717 @@ +# mypy: allow-untyped-defs +""" +This file includes private common utilities for FSDP. +""" + +import dataclasses +import logging +import traceback +import warnings +import weakref +from collections.abc import Callable, Generator, Iterable, Iterator +from enum import auto, Enum +from functools import partial +from itertools import chain +from typing import Any, cast, no_type_check, Optional, TYPE_CHECKING + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._flat_param as flat_param_file +import torch.nn as nn +from torch.distributed._composable_state import _get_module_state, _State +from torch.distributed.algorithms._checkpoint.checkpoint_wrapper import ( + _CHECKPOINT_PREFIX, +) +from torch.distributed.utils import _apply_to_tensors +from torch.utils._mode_utils import no_dispatch + +from .api import ( + FullOptimStateDictConfig, + FullStateDictConfig, + OptimStateDictConfig, + ShardingStrategy, + StateDictConfig, + StateDictType, +) + + +if TYPE_CHECKING: + from torch.distributed.device_mesh import DeviceMesh + from torch.distributed.fsdp._fsdp_extensions import FSDPExtensions + + from ._flat_param import FlatParamHandle + + +_MAX_TRAVERSE_DEPTH = 128 + + +def _is_namedtuple(obj: Any) -> bool: + # Mirrors torch.nn.parallel.scatter_gather._is_namedtuple + fields = getattr(type(obj), "_fields", None) + return ( + isinstance(obj, tuple) + and hasattr(obj, "_asdict") + and isinstance(fields, tuple) + and all(isinstance(f, str) for f in fields) + ) + + +def collect_grad_tensors(output: Any) -> tuple[torch.Tensor, ...]: + """ + Recursively collect tensors that require gradients from a nested structure. + + Traverses dict, list, tuple, NamedTuple, and dataclass containers. + Sets and other iterables are *not* traversed (consistent with + ``tree_flatten``). Uses the same traversal order as + :func:`replace_grad_tensors`. + """ + tensors_list: list[torch.Tensor] = [] + _collect_grad_tensors(output, tensors_list) + return tuple(tensors_list) + + +def _collect_grad_tensors( + output: Any, out: list[torch.Tensor], _depth: int = 0 +) -> None: + """Collect grad-requiring tensors in the same order as _replace_grad_tensors.""" + if _depth >= _MAX_TRAVERSE_DEPTH: + raise RuntimeError( + f"collect_grad_tensors exceeded max depth ({_MAX_TRAVERSE_DEPTH}), " + "likely due to a circular reference in the output structure" + ) + # Branch order must mirror _replace_grad_tensors exactly. + # Only dict, list, tuple, NamedTuple, and dataclass are traversed; + # set and other iterables are intentionally skipped (matching tree_flatten). + if torch.is_tensor(output) and output.requires_grad: + out.append(output) + elif _is_namedtuple(output): + # NamedTuple before dataclass to match _replace_grad_tensors ordering. + for item in output: + _collect_grad_tensors(item, out, _depth + 1) + elif dataclasses.is_dataclass(output) and not isinstance(output, type): + for field in dataclasses.fields(output): + _collect_grad_tensors(getattr(output, field.name), out, _depth + 1) + elif isinstance(output, dict): + for v in output.values(): + _collect_grad_tensors(v, out, _depth + 1) + elif isinstance(output, (list, tuple)): + for item in output: + _collect_grad_tensors(item, out, _depth + 1) + + +def replace_grad_tensors(output: Any, tensor_iter: Iterator[torch.Tensor]) -> Any: + """ + Replace grad-requiring tensors in a nested structure using replacements + from tensor_iter. + + Tensors are consumed from tensor_iter in the same traversal order as + :func:`collect_grad_tensors`. Traverses dict, list, tuple, NamedTuple, + and dataclass containers; sets and other iterables are *not* traversed + (consistent with ``tree_flatten``). + + Note: dataclass reconstruction uses ``dataclasses.replace()``, which calls + ``__init__``. Dataclasses with custom ``__init__`` validation, + ``__post_init__`` side effects, or non-standard dict subclass constructors + may not be compatible. In practice, FSDP module outputs are expected to be + shallowly nested, so recursion depth is not a concern. + """ + result = _replace_grad_tensors(output, tensor_iter) + sentinel = object() + leftover = next(tensor_iter, sentinel) + if leftover is not sentinel: + # Count remaining without holding references to all of them + n = 1 + sum(1 for _ in tensor_iter) + raise RuntimeError( + f"{n} replacement tensors were not consumed while processing " + f"{type(output).__qualname__}" + ) + return result + + +def _replace_grad_tensors( + output: Any, tensor_iter: Iterator[torch.Tensor], _depth: int = 0 +) -> Any: + # Branch order must mirror _collect_grad_tensors exactly. + if _depth >= _MAX_TRAVERSE_DEPTH: + raise RuntimeError( + f"replace_grad_tensors exceeded max depth ({_MAX_TRAVERSE_DEPTH}), " + "likely due to a circular reference in the output structure" + ) + if torch.is_tensor(output) and output.requires_grad: + return next(tensor_iter) + elif _is_namedtuple(output): + # NamedTuple before dataclass: a NamedTuple that is also a dataclass + # should be reconstructed via positional args, not dataclasses.replace. + new_items = [] + any_changed = False + for item in output: + new_item = _replace_grad_tensors(item, tensor_iter, _depth + 1) + new_items.append(new_item) + if new_item is not item: + any_changed = True + if any_changed: + return type(output)(*new_items) + return output + elif dataclasses.is_dataclass(output) and not isinstance(output, type): + changes = {} + for field in dataclasses.fields(output): + old_val = getattr(output, field.name) + new_val = _replace_grad_tensors(old_val, tensor_iter, _depth + 1) + if new_val is not old_val: + changes[field.name] = new_val + if changes: + try: + return dataclasses.replace(output, **changes) + except TypeError as e: + raise TypeError( + f"Failed to reconstruct dataclass {type(output).__qualname__} " + f"via dataclasses.replace(). Dataclasses used as FSDP module " + f"inputs/outputs must support dataclasses.replace(): {e}" + ) from None + return output + elif isinstance(output, dict): + new_dict = {} + any_changed = False + for k, v in output.items(): + new_v = _replace_grad_tensors(v, tensor_iter, _depth + 1) + new_dict[k] = new_v + if new_v is not v: + any_changed = True + if any_changed: + return new_dict if type(output) is dict else type(output)(new_dict) + return output + elif isinstance(output, (list, tuple)): + new_items = [] + any_changed = False + for item in output: + new_item = _replace_grad_tensors(item, tensor_iter, _depth + 1) + new_items.append(new_item) + if new_item is not item: + any_changed = True + if any_changed: + typ = type(output) + try: + return typ(new_items) + except TypeError: + # Fall back to base type for subclasses with custom __init__ + return list(new_items) if isinstance(output, list) else tuple(new_items) + return output + else: + return output + + +FSDP_WRAPPED_MODULE = "_fsdp_wrapped_module" +FSDP_PREFIX = FSDP_WRAPPED_MODULE + "." +FSDP_FLATTENED = "_fsdp_flattened" + +# Save a global mapping from module to its input tensor dtype to be populated +# during the forward pre-hook and consumed in the forward post-hook when +# overriding a module's mixed precision +# NOTE: We currently take the last input tensor's dtype in the case of multiple +# floating-point input tensors, which may be incorrect. However, since there is +# not a 1:1 correspondence between input and output tensors, we must use *some* +# heuristic like this to predict the desired output dtype. +_MODULE_TO_INP_DTYPE: weakref.WeakKeyDictionary = weakref.WeakKeyDictionary() + + +class _FSDPDeviceHandle: + """ + This is a simple abstraction for FSDP computing devices, + which enables custom backends that implement CUDA-like + semantics to be integrated with FSDP. + """ + + def __init__(self, device: torch.device, backend: Any = None): + if backend is None: + try: + self.__backend = getattr(torch, device.type) + self.__device = device + except AttributeError as exc: + raise AttributeError( + f"Device '{device}' does not have a corresponding backend registered as 'torch.{device.type}'." + ) from exc + else: + self.__backend = backend + + @classmethod + def from_device(cls, device: torch.device) -> "_FSDPDeviceHandle": + """ + Return a device handle corresponding to the device, and through this handle, + operations with the same semantics as CUDA can be performed on the device. + Just return torch.cuda if the device is cuda to make attribute-access faster. + Custom backend must first register a module with the same name with {device.type} on torch. + """ + if device.type == "cuda": + return cast(_FSDPDeviceHandle, torch.cuda) + elif device.type == "mtia": + return cast(_FSDPDeviceHandle, torch.mtia) + return cls(device) + + def __getattr__(self, name: str, /) -> Any: + try: + return getattr(self.__backend, name) + except AttributeError as exc: + raise AttributeError( + f"Custom backend '{self.__device.type}' not implement 'torch.{self.__device.type}.{name}'" + ) from exc + + +class _UninitializedDeviceHandle(_FSDPDeviceHandle): + def __init__(self) -> None: + pass + + def __getattribute__(self, name: str, /) -> Any: + raise RuntimeError("Trying to use an uninitialized device handle.") + + +class _FSDPState(_State): + def __init__(self) -> None: + # TODO: Move all the attributes to this class to enable typing for + # FSDP/fully_shard. + self._ignored_modules: set[nn.Module] = set() + self._ignored_params: set[nn.Parameter] = set() + # Buffer names are cleaned (without wrapper prefixes) + self._ignored_buffer_names: set[str] = set() + self.process_group: dist.ProcessGroup | None = None + self.rank: int = -1 + self.world_size: int = -1 + self._device_mesh: DeviceMesh | None = None + self.sharding_strategy = ShardingStrategy.FULL_SHARD + self._use_orig_params: bool = False + self.training_state = TrainingState.IDLE + self._unshard_params_ctx: dict[nn.Module, Generator] = {} + self._state_dict_type: StateDictType = StateDictType.FULL_STATE_DICT + self._state_dict_config: StateDictConfig = FullStateDictConfig() + self._optim_state_dict_config: OptimStateDictConfig = FullOptimStateDictConfig() + self._is_root: bool | None = None + self._handle: flat_param_file.FlatParamHandle | None = None + self._fully_sharded_module_to_handle: dict[ + nn.Module, flat_param_file.FlatParamHandle | None + ] = {} + self.compute_device: torch.device | None = None + self._gradient_predivide_factor: int = 0 + self._gradient_postdivide_factor: int = 0 + self._comm_hook: Callable | None = None + self._comm_hook_state: Any | None = None + self._unshard_event: torch.Event | None = None + # Abstract device handle for fsdp compute device. For now, + # the compute device must implement cuda semantics used by fsdp + self._device_handle: _FSDPDeviceHandle = _UninitializedDeviceHandle() + # All following attributes should only be used for root states: + # Save these static lists to avoid the repeated tree traversals + self._all_fsdp_states: list[_FSDPState] = [] + self._all_handles: list[flat_param_file.FlatParamHandle] = [] + self._fsdp_extension: FSDPExtensions | None = None + + +def _get_module_fsdp_state(module: nn.Module) -> _FSDPState | None: + state = _get_module_state(module) + if state is None or not isinstance(state, _FSDPState): + return None + return state + + +def _get_module_fsdp_state_if_fully_sharded_module( + module: nn.Module, +) -> _FSDPState | None: + state = _get_module_fsdp_state(module) + if state is None: + return None + if state == module: # FullyShardedDataParallel module case. + return state + if module in state._fully_sharded_module_to_handle: # fully_shard case. + return state + return None + + +class TrainingState(Enum): + """ + An enum that indicates the state of a ``FullyShardedDataParallel` instance. + """ + + IDLE = auto() + FORWARD_BACKWARD = auto() + SUMMON_FULL_PARAMS = auto() + + +class HandleTrainingState(Enum): + """ + An enum that indicates the state of a ``FlatParamHandle`. + """ + + IDLE = auto() + FORWARD = auto() + BACKWARD_PRE = auto() + BACKWARD_POST = auto() + SUMMON_FULL_PARAMS = auto() + + +def _is_composable(state: _FSDPState): + # TODO: This is a temporary hack for differentiate between code paths. + return not isinstance(state, nn.Module) + + +@no_type_check +def _module_handle(state: _FSDPState, module: nn.Module) -> Optional["FlatParamHandle"]: + """ + Returns the ``FlatParamHandle`` s corresponding to ``module``. This is + the handle that contains some parameter in ``module``. + """ + if _is_composable(state): + # A valid FSDP state may have no managed parameters and hence no + # handles, meaning no entry in `_fully_sharded_module_to_handles` + if state._handle is None: + return None + if module not in state._fully_sharded_module_to_handle: + raise AssertionError( + f"Expects a fully sharded module but got {module} on rank {state.rank}" + ) + return state._fully_sharded_module_to_handle[module] + else: + # NOTE: This assumes `module` is a `FullyShardedDataParallel` instance. + return module._handle + + +@no_type_check +def _has_fsdp_params(state: _FSDPState, module: nn.Module) -> bool: + """Returns if ``module`` has parameters managed by FSDP.""" + return _module_handle(state, module) is not None + + +def _get_sharding_strategy(handle): + """ + Returns the sharding strategy of the handle. + """ + return handle._sharding_strategy if handle else None + + +def clean_tensor_name(tensor_name: str) -> str: + """ + Cleans the parameter or buffer name by removing any module wrapper + prefixes. + """ + tensor_name = tensor_name.replace(FSDP_PREFIX, "") + # TODO: Explicitly replacing the checkpoint wrapper prefix is not ideal as + # it couples `CheckpointWrapper` and FSDP and also does not scale for more + # module wrappers. + tensor_name = tensor_name.replace(_CHECKPOINT_PREFIX, "") + return tensor_name + + +def _set_fsdp_flattened(tensor: torch.Tensor) -> None: + """ + Sets an attribute on ``tensor`` to mark it as flattened by FSDP. This is to + avoid re-flattening it during nested construction. + """ + setattr(tensor, FSDP_FLATTENED, True) + + +def _is_fsdp_flattened(tensor: torch.Tensor) -> bool: + """Returns if ``tensor`` has been marked as flattened by FSDP.""" + return getattr(tensor, FSDP_FLATTENED, False) + + +def _named_parameters_with_duplicates( + module: nn.Module, **kwargs: Any +) -> list[tuple[str, nn.Parameter]]: + """ + This API is required as some modules overwrite `named_parameters()` but do not support + `remove_duplicate`. + """ + if "remove_duplicate" in kwargs: + raise AssertionError( + "_named_parameters_with_duplicates cannot be used with `remove_duplicate` argument." + ) + kwargs["remove_duplicate"] = False + try: + ret = list(module.named_parameters(**kwargs)) + except AssertionError: + kwargs.pop("remove_duplicate") + ret = list(module.named_parameters(**kwargs)) + return ret + + +def _get_param_to_fqns( + model: torch.nn.Module, + dedup_shared_params: bool = True, +) -> dict[nn.Parameter, list[str]]: + """ + Constructs a mapping from parameter to a list of its \"canonical\" FQNs. Here, + we use canonical to mean the fully-qualified name assigned to the parameter + based on its position in the original nn.Module hierarchy before any wrapper + or parallelism has been applied to it. This is in contrast to FQNs that may be + generated after parallelisms or wrappers have been applied to the model. + + Each normal parameter maps to a singleton list containing its FQN, while each + ``FlatParameter`` maps to a list of its original parameter FQNs, which may + have length greater than one. All FQNs are prefixed starting from ``model``. + + In the case where FSDP was applied with ``use_orig_params=True``, there should be no + ``FlatParameter`` s registered to the model's modules and this mapping will only + contain mappings from ``nn.Parameter`` s to singleton FQN lists. + + It is only in the case where FSDP was applied with ``use_orig_params=False`` where + a ``FlatParameter`` will be registered in place of the original parameters and there + will be mappings from each ``FlatParameter`` to lists of FQNs corresponding to the + original parameters. + + Args: + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance). + dedup_shared_params (bool): For shared parameters, if ``True``, only + includes the FQNs corresponding to the first encounter of the + shared parameter in the module traversal; if ``False``, then + includes the FQNs across all encounters. (Default: ``True``) + """ + + def module_fn(module, prefix, tree_level, param_to_fqns): + for param_name, param in _named_parameters_with_duplicates( + module, recurse=False + ): + local_fqns = ( + param._fqns + if isinstance(param, flat_param_file.FlatParameter) + else [param_name] + ) # prefixed from `module` + global_fqns = [ + clean_tensor_name(prefix + name) for name in local_fqns + ] # prefixed from the top level `model` (i.e. including `prefix`) + is_shared_param = param in param_to_fqns + if not is_shared_param: + param_to_fqns[param] = global_fqns + else: + if isinstance(param, flat_param_file.FlatParameter): + # DMP overwrites `named_parameters` and skip (advance to + # the next child module) the wrapped_module (e.g., + # _dmp_wrapped_module and _fsdp_wrapped_module). When a user + # calls `named_child` to traverse the module recursively and + # calls `named_parameters` with `recurse=False`, parameters + # will be traversed more than once. + # This hack is specified designed for DMP + FSDP. We + # overwrite the flat_parameters traversal result to only obtain + # the last one, which happens to be the correct one. + # + # TODO: Remove this hack once DMP + FSDP is not supported. + warnings.warn( + "FlatParameter is being traversed more than once. " + "This case should only happen when using " + "DistributedModelParallel with FullyShardedDataParallel.", + stacklevel=2, + ) + param_to_fqns[param] = global_fqns + elif not dedup_shared_params: + param_to_fqns[param].extend(global_fqns) + + def return_fn(param_to_fqns): + return param_to_fqns + + param_to_unflat_param_names: dict[torch.nn.Parameter, list[str]] = {} + return _apply_to_modules( + model, + module_fn, + return_fn, + [key for key, _ in _named_parameters_with_duplicates(model)], + param_to_unflat_param_names, + ) + + +@no_type_check +def _log_post_backward_hook( + state: _FSDPState, handle: "FlatParamHandle", logger: logging.Logger +) -> None: + # Under TORCH_DISTRIBUTED_DEBUG=INFO, log the module names this hook fires for. + # Below logging of module names this post-bwd hook fires for can help debug certain + # cases where hooks don't fire, such as under certain activation checkpoint configs. + if state._use_orig_params and handle._debug_level == dist.DebugLevel.INFO: + param_fqns = _get_handle_fqns_from_root(state, handle) + logger.warning("FSDP firing post-backward hooks for parameters %s", param_fqns) + + +@no_type_check +def _get_handle_fqns_from_root( + state: _FSDPState, handle: "FlatParamHandle" +) -> list[str] | None: + if handle is None: + return None + param_to_fqn = state._exec_order_data.param_to_fqn + handle_params = handle.flat_param._params # only populated for use_orig_params + param_fqns = [*chain.from_iterable(param_to_fqn[p] for p in handle_params)] + return param_fqns + + +def _apply_to_modules( + root_module: torch.nn.Module, + module_fn: Callable, + return_fn: Callable, + filter_fqns: list[str] | None = None, + *args, + **kwargs, +): + """ + Performs a pre-order traversal of the modules in the hierarchy rooted at + ``root_module``, applying ``module_fn`` at each module and finally + returning a value using ``return_fn``. The traversal constructs the full + module prefix name (e.g. "module.submodule." just like in model state dict) + and makes that available to ``module_fn``. + + ``filter_fqns`` is used because some module may have its own prefix similar + to ``FullyShardedDataParallel`` and the ``named_parameters()`` is overwritten + to remove the prefix. + """ + + # Precompute the set of all prefixes from filter_fqns so that the + # "any FQN starts with new_prefix?" check is O(1) instead of O(N). + filter_prefixes: set[str] | None = None + if filter_fqns is not None: + filter_prefixes = set() + for fqn in filter_fqns: + i = fqn.find(".") + while i != -1: + filter_prefixes.add(fqn[: i + 1]) + i = fqn.find(".", i + 1) + + def f(module: torch.nn.Module, prefix: str, tree_level: int, *args, **kwargs): + # Call the module function before recursing over children (pre-order) + module_fn(module, prefix, tree_level, *args, **kwargs) + for submodule_name, submodule in module.named_children(): + if submodule is None: + continue + new_prefix = prefix + submodule_name + "." + new_tree_level = tree_level + 1 + if filter_prefixes is not None: + if new_prefix not in filter_prefixes: + # DMP's named_parameter() will mess up the traversal with + # ``named_children`` + `named_parameter(recurse=False)``. + # This hack is a must to make the traversal work. + # TODO: Remove this hack once DMP + FSDP is not supported. + # It turns out that recursive wrapping may trigger this as + # well. + if ( + submodule_name == "_fsdp_wrapped_module" + or submodule_name == "_dmp_wrapped_module" + ): + new_prefix = prefix + elif submodule_name == "module": + new_prefix = prefix + f(submodule, new_prefix, new_tree_level, *args, **kwargs) + + f(root_module, "", 0, *args, **kwargs) + return return_fn(*args, **kwargs) + + +@no_type_check +def _assert_in_training_states( + state: _FSDPState, + training_states: list[TrainingState], +) -> None: + """Asserts that FSDP is in the states ``_training_states``.""" + # Raise a `ValueError` instead of using `assert` to ensure that these + # logical assertions run even if `assert`s are disabled + if state.training_state not in training_states: + msg = ( + f"expected to be in states {training_states} but current state is " + f"{state.training_state}" + ) + # Print the error on rank 0 in case this is called in the backward pass + if state.rank == 0: + if isinstance(state, nn.Module): + print(f"Asserting FSDP instance is: {state}") + print(f"ERROR: {msg}") + traceback.print_stack() + raise ValueError(msg) + + +def _get_root_modules(modules: set[nn.Module]) -> set[nn.Module]: + """ + Returns: + Set[nn.Module]: The subset of ``modules`` that are root modules (i.e. + parent-less) with respect to the modules in the set itself. In other + words, these are the modules in ``modules`` that are not the child of + any other module in ``modules``. + """ + root_modules: set[nn.Module] = set() + module_to_submodules = {module: set(module.modules()) for module in modules} + for candidate_module in modules: + is_root_module = True + for module, submodules in module_to_submodules.items(): + is_child_module = ( + candidate_module is not module and candidate_module in submodules + ) + if is_child_module: + is_root_module = False + break + if is_root_module: + root_modules.add(candidate_module) + return root_modules + + +def _override_module_mixed_precision( + root: torch.nn.Module, + module_classes_to_override: Iterable[type[nn.Module]], + wrap_override_dict: dict[str, Any] = {"mixed_precision": None}, # noqa: B006 +) -> set[type[nn.Module]]: + module_classes_to_override = tuple(set(module_classes_to_override)) + # Return a set of the actually overridden module classes + overridden_module_classes: set[type[nn.Module]] = set() + for mod in root.modules(): + if isinstance(mod, module_classes_to_override): + overridden_module_classes.add(type(mod)) + mod._wrap_overrides = wrap_override_dict # type: ignore[assignment] + # TODO: We need to run this mixed precision ignored module in fp32, + # but ensure subsequent modules, that may possibly be running with + # mixed precision, still receive the appropriate precision inputs + # without user having to adjust mixed precision config too much. + # As a result, we attach pre and post forward hooks to up / down + # cast. We should revisit this design. + + def cast_fn( + dtype: torch.dtype, module: nn.Module, x: torch.Tensor + ) -> torch.Tensor: + if not torch.is_floating_point(x) or x.dtype == dtype: + return x + _MODULE_TO_INP_DTYPE[module] = x.dtype + return x.to(dtype) + + def forward_pre_hook(module, args): + return _apply_to_tensors(partial(cast_fn, torch.float32, module), args) + + def forward_post_hook(module, args, output): + # NOTE: If the forward did not have any floating-point tensors, + # then the dtype will not be set for this module, and we do not + # upcast the dtype. + if module in _MODULE_TO_INP_DTYPE: + old_dtype = _MODULE_TO_INP_DTYPE[module] + return _apply_to_tensors( + partial(cast_fn, old_dtype, module), output + ) + + # We intentionally append both of these hooks so that they run after + # all other hooks. + mod.register_forward_pre_hook(forward_pre_hook, prepend=False) + mod.register_forward_hook(forward_post_hook, prepend=False) + return overridden_module_classes + + +def _no_dispatch_record_stream(tensor: torch.Tensor, stream: torch.Stream) -> None: + # FIXME record_stream doesn't work with non-cuda/mtia/xpu tensors + if tensor.device.type not in [ + "cuda", + "mtia", + "xpu", + torch._C._get_privateuse1_backend_name(), + ]: + return + + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + return + # from @ezyang: + # The no_dispatch was added in https://github.com/pytorch/pytorch/pull/88014 cc @fegin + # Looking over the PR, it looks like this is because we don't actually support Stream arguments + # in torch dispatch, so it just chokes. + # If Dynamo is able to answer "are there any torch dispatch modes" active (it should answer False), + # a better version of this would just be to check if there are any modes before disabling dispatch. + # TODO(voz): Extend a dynamo util to answer the above, unify the codepaths here. + tensor.record_stream(stream) + else: + with no_dispatch(): + tensor.record_stream(stream) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_debug_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_debug_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..cf5a411f8c556ff1922775514cb2361a87bb492d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_debug_utils.py @@ -0,0 +1,159 @@ +# mypy: allow-untyped-defs +import logging +import time +from collections import defaultdict +from collections.abc import Iterator +from contextlib import contextmanager +from enum import Enum + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._flat_param as flat_param_file +from torch.distributed.fsdp._common_utils import ( + _apply_to_modules, + _get_module_fsdp_state, + clean_tensor_name, +) + + +logger = logging.getLogger(__name__) + + +class SimpleProfiler: + class Type(str, Enum): + ALL = "all" + ALLGATHER = "all_gather" + ALLGATHER_OBJ = "all_gather_object" + RESHARDING = "resharding" + H2D = "H2D" + D2H = "D2H" + + results: dict[str, float] = defaultdict(float) + profiling: set[str] = set() + + @classmethod + def reset(cls) -> None: + cls.results.clear() + cls.profiling.clear() + + @classmethod + @contextmanager + def profile(cls, profile_type: str) -> Iterator[None]: + if profile_type in cls.profiling: + raise AssertionError( + f"{profile_type} is already being profiled. " + "SimpleProfiler does not support profiling multiple instances at " + "the same time. " + ) + + cls.profiling.add(profile_type) + begin = time.monotonic() + try: + yield + finally: + end = time.monotonic() + cls.results[profile_type] += end - begin + cls.profiling.remove(profile_type) + + @classmethod + def dump_and_reset(cls, msg: str) -> None: + # This cannot be combined with DETAIL distributed log + # as the profiling will be very incorrect. + if dist.get_rank() == 0 and dist.get_debug_level() == dist.DebugLevel.INFO: + logger.info("%s %s", msg, cls.results) + cls.reset() + + +def _get_sharded_module_tree_with_module_name_to_fqns( + model: torch.nn.Module, +) -> tuple[str, dict[str, list[str]]]: + """ + It is used for composable fully_shard() code path, it returns + 1. sharded module tree info: each line represents a submodule name that contains the + submodule's FQN and its submodule class name, if the submodule is sharded by `fully_shard`, + the submodule name will add a postfix with ' FULLY SHARDED'. Each increased tree + level adds 4 spaces before the printed name. A printed sharded module tree info for a toy model + is like this: + [CompositeModel] FULLY SHARDED + l1[Linear] + u1[UnitModule] FULLY SHARDED + u1.l1[Linear] + u1.seq[Sequential] + u1.seq.0[ReLU] + u1.seq.1[Linear] + u1.seq.2[ReLU] + u1.l2[Linear] + u2[UnitModule] FULLY SHARDED + u2.l1[Linear] + u2.seq[Sequential] + u2.seq.0[ReLU] + u2.seq.1[Linear] + u2.seq.2[ReLU] + u2.l2[Linear] + l2[Linear] + 2. a dict mapping from the concated module FQN and class name to a list of its managed + original parameters' FQNs. An example of the dict for the above toy sharded model is like this: + {'[CompositeModel]': ['l1.weight', 'l1.bias', 'l2.weight', 'l2.bias'], + 'u1[UnitModule]': ['u1.l1.weight', 'u1.l1.bias', 'u1.seq.1.weight', 'u1.seq.1.bias', 'u1.l2.weight', 'u1.l2.bias'], + 'u2[UnitModule]': ['u2.l1.weight', 'u2.l1.bias', 'u2.seq.1.weight', 'u2.seq.1.bias', 'u2.l2.weight', 'u2.l2.bias'] + } + All FQNs are prefixed starting from ``model``. + + Args: + model (torch.nn.Module): Root module (which may or may not be passed to + composable `fully_shard()`). + """ + + def module_fn( + module, prefix, tree_level, sharded_tree_info, sharded_module_name_to_fqns + ): + num_spaces = tree_level * 4 + trimed_prefix = ( + prefix[:-1] if (len(prefix) > 0 and prefix[-1] == ".") else prefix + ) + prefixed_module_name = trimed_prefix + "[" + module.__class__.__name__ + "]" + printed_prefixed_module_name = " " * num_spaces + prefixed_module_name + + state = _get_module_fsdp_state(module) + if state is None: + sharded_tree_info[0] += printed_prefixed_module_name + "\n" + return + + handle = state._fully_sharded_module_to_handle.get(module, None) + + if handle: + sharded_tree_info[0] += ( + printed_prefixed_module_name + " FULLY SHARDED" + "\n" + ) + else: + sharded_tree_info[0] += printed_prefixed_module_name + "\n" + + if handle: + param = handle.flat_param + if not isinstance(param, flat_param_file.FlatParameter): + raise AssertionError(f"Expected FlatParameter, got {type(param)}") + global_fqns = [ + clean_tensor_name(prefix + name) for name in param._fqns + ] # prefixed from the top level `model` (i.e. including `prefix`) + + if prefixed_module_name in sharded_module_name_to_fqns: + sharded_module_name_to_fqns[prefixed_module_name].extend(global_fqns) + else: + sharded_module_name_to_fqns[prefixed_module_name] = global_fqns + + def return_fn(sharded_tree_info, sharded_module_name_to_fqns): + return sharded_tree_info[0], sharded_module_name_to_fqns + + # Use List to mutate its value in place while running the recursive functions + sharded_tree_info: list[str] = [ + "", + ] + sharded_module_name_to_fqns: dict[str, list[str]] = {} + return _apply_to_modules( + model, + module_fn, + return_fn, + [key for key, _ in model.named_parameters()], + sharded_tree_info, + sharded_module_name_to_fqns, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_dynamo_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_dynamo_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..77bcd43b63be27da8e8b79f877ce7cb9d67c74b8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_dynamo_utils.py @@ -0,0 +1,43 @@ +import torch.nn as nn + + +def _annotate_modules_for_dynamo( + module: nn.Module, + ignored_modules: set[nn.Module], + use_orig_params: bool, +) -> None: + """ + Annotates the submodules in ``module`` 's tree, except those in + ``ignored_modules``, indicating that the submodules are FSDP-managed and + saving the ``use_orig_params`` setting passed to the FSDP constructor. + """ + for submodule in module.modules(): + if submodule not in ignored_modules: + """[note: Dynamo treats FSDP wrapped modules as UnspecializedNNModule] + + Dynamo doesn't get to see this instance (FullyShardedDataParallel) during tracing, since + it skips tracing all the torch.distributed.fsdp code. + - Why? Running the FSDP code eagerly avoids lots of issues trying to trace complex hooks, and also + gets us graph-breaks on FSDP module boundaries which we want anyway for comm ops. + - However, we _also_ want dynamo to treat the wrapped module inside FSDP 'unspecially' (*), + and we need a way to indicate to dynamo which modules are wrapped by FSDP. + + (*) UnspecializedNNModules in dynamo are traced-through without any assumptions, and with thorough + guards. NNModules otherwise are 'specialized', meaning there is less overhead due to assuming + their code is well-behaved. + + One particular issue with specialized NNModules for FSDP is that the + views created for orig_params are captured into the compiled graph on the first iteration, and while + they are always going to point to the correct flatparameter and give correct results, their order + of creation influences the order of backward execution, preventing overlap of comm and computation + during backward. We need to _use_ the new parameter views created on each forward iteration, in + order for backward to interleave hooks with compute per layer. UnspecializedNNModule lets us achieve + this by capturing the module code more 'functionally' and passing parameters in as inputs each time. + """ + submodule._is_fsdp_managed_module = True # type: ignore[assignment] + + # Dynamo only supports FSDP with use_orig_params=True. + # This is hacky, but I could not think of another way to add an assertion to dynamo + # for this, since Dynamo skips all the FSDP code frames and thus can't inspect the + # FSDP module directly + submodule._fsdp_use_orig_params = use_orig_params # type: ignore[assignment] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_exec_order_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_exec_order_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..d26f54f2f405042d825a6ccefb41b8000056ae52 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_exec_order_utils.py @@ -0,0 +1,365 @@ +# mypy: allow-untyped-defs +import itertools +import warnings +from enum import auto, Enum + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.nn as nn +from torch.distributed.fsdp._common_utils import _FSDPState, _get_param_to_fqns +from torch.distributed.fsdp._flat_param import FlatParamHandle + + +class _ExecOrderWarnStatus(Enum): + """Used internally for execution order validation.""" + + NONE = auto() # no deviation yet + WARNING = auto() # deviated this iteration; currently issuing warnings + WARNED = auto() # deviated in a previous iteration + + +class _ExecOrderData: + """ + This contains the data structures to track the execution order. We track + the pre-forward order on the *first* iteration for forward prefetching + (which thus assumes static graph) and the post-forward order on *every* + iteration for backward prefetching (which thus does not assume static + graph but may be provide an incorrect order). + """ + + def __init__( + self, + debug_level: dist.DebugLevel, + backward_prefetch_limit: int, + forward_prefetch_limit: int, + ) -> None: + # Tracks the (static) pre-forward order for execution order validation + # and forward prefetching + self.handles_pre_forward_order: list[FlatParamHandle] = [] + # Tracks the post-forward order for pre-backward prefetching + self.handles_post_forward_order: list[FlatParamHandle | None] = [] + self._iter = 0 + + # Gives the max number of backward/forward prefetched all-gathers by a + # single module + self._backward_prefetch_limit = backward_prefetch_limit + self._forward_prefetch_limit = forward_prefetch_limit + + # Data structures for execution order validation + self._checking_order: bool = debug_level == dist.DebugLevel.DETAIL + self.process_group: dist.ProcessGroup | None = None + self.world_size: int | None = None + self.all_handles: list[FlatParamHandle] = [] + # Names are prefixed from the root module + self.param_to_fqn: dict[nn.Parameter, list[str]] = {} + # Current index in the pre-forward execution order + self.current_order_index = 0 + self.warn_status = _ExecOrderWarnStatus.NONE + + def init( + self, + state: _FSDPState, + root_module: nn.Module, + process_group: dist.ProcessGroup, + ) -> None: + """ + Initializes the data structures needed for checking the forward order. + This should be called after a root FSDP instance has been set during + lazy initialization. + """ + self.process_group = process_group + self.rank = process_group.rank() + self.world_size = process_group.size() + # Fix an order over the handles, which should be the same across ranks + for handle in traversal_utils._get_fsdp_handles(root_module): + index = len(self.all_handles) + self.all_handles.append(handle) + handle._handle_index = index + self.param_to_fqn = _get_param_to_fqns(root_module) + # TODO (awgu): We can broadcast the metadata of rank 0's `all_handles` + # to check that all ranks have the same handles in the same order. + # https://github.com/pytorch/pytorch/issues/79620 + + @property + def is_first_iter(self) -> bool: + return self._iter == 0 + + def get_handle_to_backward_prefetch( + self, + current_handle: FlatParamHandle, + ) -> FlatParamHandle | None: + """ + Returns a :class:`list` of the handles keys of the handles to backward + prefetch given the current handles key. If there are no valid handles + keys to prefetch, then this returns an empty :class:`list`. + """ + current_index = current_handle._post_forward_index + if current_index is None: + return None + target_index = current_index - 1 + target_handle: FlatParamHandle | None = None + for _ in range(self._backward_prefetch_limit): + if target_index < 0: + break + target_handle = self.handles_post_forward_order[target_index] + target_index -= 1 + return target_handle + + def get_handle_to_forward_prefetch( + self, + current_handle: FlatParamHandle, + ) -> FlatParamHandle | None: + """ + Returns a :class:`list` of the handles keys of the handles to forward + prefetch given the current handles key. If there are no valid handles + keys to prefetch, then this returns an empty :class:`list`. + """ + current_index = current_handle._pre_forward_order_index + if current_index is None: + return None + target_index = current_index + 1 + target_handle: FlatParamHandle | None = None + for _ in range(self._forward_prefetch_limit): + if target_index >= len(self.handles_pre_forward_order): + break + target_handle = self.handles_pre_forward_order[target_index] + target_index += 1 + return target_handle + + def record_post_forward(self, handle: FlatParamHandle | None) -> None: + """ + Records ``handles`` in the post-forward order, where ``handles`` should + be a group of handles used in the same module's forward. If ``handles`` + is empty, then it is omitted. + + Unlike :meth:`record_pre_forward`, this records the order *every* + iteration with the expectation that the recorded order is reset in + :meth:`next_iter`. + """ + if not handle: + return + # Only record the first usage of a handles key + if handle._post_forward_index: + self.handles_post_forward_order.append(handle) + return + index = len(self.handles_post_forward_order) + handle._post_forward_index = index + self.handles_post_forward_order.append(handle) + + def record_pre_forward( + self, handle: FlatParamHandle | None, is_training: bool + ) -> None: + """ + Records ``handles`` in the pre-forward order, where ``handles`` should + be a group of handles used in the same module's forward. If ``handles`` + is empty, then it is omitted. + + On the first iteration, this checks the execution order across ranks. + See :meth:`_check_order` for details. + """ + if not handle: + return + self._check_order(handle, is_training) + # Fix the order after the first iteration and only record the first + # usage of a handles key + if not self.is_first_iter or handle._pre_forward_order_index is not None: + return + index = len(self.handles_pre_forward_order) + handle._pre_forward_order_index = index + self.handles_pre_forward_order.append(handle) + + def _check_order(self, handle: FlatParamHandle, is_training: bool) -> None: + """ + Checks the forward execution order as long as ``is_training`` is + ``True`` since checking in eval mode is not supported. This only checks + if the distributed debug level is DETAIL. + + - On the first iteration, this uses all-gathers to check that all ranks + are all-gathering the same handles and hence ``FlatParameter`` s, + raising an error if not. + - On subsequent iterations, this checks that each rank is locally + consistent with its own forward order from the first iteration, issuing + a warning if not. This issues a warning on the first deviating + iteration and stops warning thereafter. + """ + # Do not check order in eval mode since the post-backward callback does + # not run so it cannot be used to mark the end of an iteration + if not is_training or not self._checking_order: + return + if self.is_first_iter: + msg_prefix = "Forward order differs across ranks:" + optional_local_indices: tuple[int | None, ...] = self._get_handle_indices( + handle + ) + device = handle.device # guaranteed to be non-CPU + num_valid_indices = sum( + (index is not None) for index in optional_local_indices + ) + tensor_kwargs: dict[str, torch.dtype | torch.device] = { + "dtype": torch.int32, + "device": device, + } + world_num_valid_indices = torch.zeros(self.world_size, **tensor_kwargs) # type: ignore[arg-type, call-overload] + local_num_valid_indices = torch.tensor([num_valid_indices], **tensor_kwargs) # type: ignore[arg-type, call-overload] + dist.all_gather_single( + world_num_valid_indices, + local_num_valid_indices, + group=self.process_group, + ) + # Copy entire tensor from D2H once to avoid per element D2H copies + world_num_valid_indices = world_num_valid_indices.cpu() + # Check that all ranks plan to all-gather the same number of + # parameters + # TODO (awgu): Since every module has at most one handle in the + # current implementation, this should never raise the error. + if self.world_size is None: + raise AssertionError("Expected world_size to not be None") + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + # TODO(voz): Don't graph break on this - dynamo hates the n1 != n2 + # tensor comparison control flow. + # https://github.com/pytorch/pytorch/issues/107055 + for (r1, n1), (r2, n2) in itertools.combinations( + ( + (rank, world_num_valid_indices[rank]) + for rank in range(self.world_size) + ), + 2, + ): + if n1 != n2: + raise RuntimeError( + f"{msg_prefix} rank {r1} is all-gathering {n1} parameters " + f"while rank {r2} is all-gathering {n2} parameters" + ) + world_indices = torch.zeros( # type: ignore[call-overload] + self.world_size * num_valid_indices, **tensor_kwargs + ) + local_indices = torch.tensor(optional_local_indices, **tensor_kwargs) # type: ignore[arg-type] + dist.all_gather_single( + world_indices, local_indices, group=self.process_group + ) + # Copy entire tensor from D2H once to avoid per element D2H copies + world_indices = world_indices.cpu() + # Check that all ranks plan to all-gather the same index parameters + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + # TODO(voz): Don't graph break on this - dynamo hates the i1 != i2 + # tensor comparison control flow. + # https://github.com/pytorch/pytorch/issues/107055 + for (r1, i1), (r2, i2) in itertools.combinations( + ( + ( + rank, + world_indices[ + rank * num_valid_indices : (rank + 1) + * num_valid_indices + ], + ) + for rank in range(self.world_size) + ), + 2, + ): + if i1 != i2: + r1_param_names = self._get_names_from_handle_indices(i1) + r2_param_names = self._get_names_from_handle_indices(i2) + raise RuntimeError( + f"{msg_prefix} rank {r1} is all-gathering parameters " + f"for {r1_param_names} while rank {r2} is all-gathering " + f"parameters for {r2_param_names}" + ) + else: + # Only issue warnings on the first deviating iteration and stop + # checking thereafter to avoid flooding the console + if self.warn_status == _ExecOrderWarnStatus.WARNED: + return + msg_prefix = None # non-`None` means we should warn + if self.current_order_index >= len(self.handles_pre_forward_order): + # This iteration sees extra all-gather(s) compared to the first + msg_prefix = ( + "Expected to not all-gather any more parameters in the " + "forward but trying to all-gather parameters for " + ) + else: + expected_handle = self.handles_pre_forward_order[ + self.current_order_index + ] + if expected_handle != handle: + expected_param_names = self._get_names_from_handles(expected_handle) + msg_prefix = ( + f"Expected to all-gather for {expected_param_names} " + "but trying to all-gather parameters for " + ) + if msg_prefix is not None: + param_names = self._get_names_from_handles(handle) + msg_suffix = ( + f"{param_names}" + if param_names + else "a newly-added parameter since construction time" + ) + warnings.warn( + "Forward order differs from that of the first iteration " + f"on rank {self.rank}. Collectives are unchecked and may " + f"give incorrect results or hang.\n{msg_prefix}{msg_suffix}", + stacklevel=2, + ) + self.warn_status = _ExecOrderWarnStatus.WARNING + self.current_order_index += 1 + + def _get_handle_indices( + self, + handle: FlatParamHandle, + ) -> tuple[int | None, ...]: + """ + Returns the handle indices (i.e. indices into ``self.all_handles``) + corresponding to the handles in ``handle``. An entry in the + returned tuple is ``None`` if the handle is invalid. + """ + indices: list[int | None] = [] + if handle: + indices.append(handle._handle_index) + return tuple(indices) + + def _get_names_from_handle_indices( + self, + handle_indices: tuple[int, ...], + ) -> list[list[str]]: + """ + Returns a list of FQNs for each handle in ``handle_indices``. If a + handle index is invalid, then its FQNs are omitted from the returned + list. + """ + fqns: list[list[str]] = [] + for index in handle_indices: + if index is None or index < 0 or index >= len(self.all_handles): + continue + handle = self.all_handles[index] + flat_param = handle.flat_param + fqns.append(self.param_to_fqn[flat_param]) + return fqns + + def _get_names_from_handles( + self, + handle: FlatParamHandle, + ) -> list[list[str]]: + """ + Returns a list of FQNs for each handle in ``handles_key``. If a handle + is invalid, then its FQNs are omitted from the returned list. + """ + fqns: list[list[str]] = [] + if handle: + flat_param = handle.flat_param + if flat_param in self.param_to_fqn: + fqns.append(self.param_to_fqn[flat_param]) + return fqns + + def next_iter(self): + """ + Advances the internal data structures per iteration. This should be + called in the post-backward callback since that marks the true end of + an iteration. + """ + self._iter += 1 + self.handles_post_forward_order.clear() + if self._checking_order: + self.current_order_index = 0 + if self.warn_status == _ExecOrderWarnStatus.WARNING: + self.warn_status = _ExecOrderWarnStatus.WARNED diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_flat_param.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_flat_param.py new file mode 100644 index 0000000000000000000000000000000000000000..a3ce0c298e86699938d6a02c07c409414c75e40e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_flat_param.py @@ -0,0 +1,2845 @@ +# mypy: allow-untyped-defs +import contextlib +import functools +import logging +import os +import warnings +from collections.abc import Callable, Generator, Iterator, Sequence +from enum import auto, Enum +from itertools import accumulate, chain +from typing import Any, cast, NamedTuple, no_type_check + +import torch +import torch.distributed as dist +import torch.nn as nn +import torch.nn.functional as F +from torch import Tensor +from torch.distributed.fsdp._common_utils import ( + _FSDPDeviceHandle, + _named_parameters_with_duplicates, + _no_dispatch_record_stream, + _set_fsdp_flattened, + HandleTrainingState, +) +from torch.distributed.utils import ( + _alloc_storage, + _data_ptr_allocated, + _free_storage, + _p_assert, +) +from torch.nn.parameter import _ParameterMeta # type: ignore[attr-defined] +from torch.testing._internal.distributed.fake_pg import FakeProcessGroup + +from ._fsdp_extensions import ( + _ext_post_unflatten_transform, + _ext_pre_flatten_transform, + FSDPExtensions, +) + + +__all__ = [ + "FlatParameter", + "FlatParamHandle", + "FlatParamShardMetadata", + "ParamInfo", + "SharedParamInfo", + "HandleShardingStrategy", +] + +logger = logging.getLogger(__name__) + + +""" +[Note: Fully Sharded Module] +We define the "fully sharded module" to be the original ``nn.Module`` that owns +a ``FlatParamHandle``. It is the *single* module logically responsible for the +*single* unshard/reshard pair for the handle's ``FlatParameter`` for a given +forward or backward pass. The fully sharded module should be passed to the +``FlatParamHandle`` constructor. + +For the wrapper code path: +- The ``FullyShardedDataParallel`` module wrapping the fully sharded module +runs the unshard/reshard on behalf of the fully sharded module by overriding +``nn.Module.forward``. +- The fully sharded module is exactly the module passed to the +``FullyShardedDataParallel`` constructor's ``module`` argument. + +For the non-wrapper code path: +- Hooks registered on the fully sharded module run the unshard/reshard. +- The fully sharded module may either be the direct argument to ``fully_shard`` +or a submodule chosen by the provided wrapping policy. +""" + +# Environment variable toggling whether to use unsafe `setattr()` for view +# setting in `_use_sharded_views()` and `_use_unsharded_views()` +# We should use 'safe' by default since it respects method overrides, but for +# special cases such as for high CPU overhead or for intentionally bypassing +# checks in the overrides, we may use 'unsafe'. +_FSDP_USE_UNSAFE_SETATTR = "FSDP_USE_UNSAFE_SETATTR" + +# Environment variable toggling whether to check for parameter/gradient +# writeback in case their storages change after FSDP initialization +# We should check by default since it prevents silent correctness errors, but +# since such changes are atypical, we may want to skip the check to save CPU +# overhead, especially since the check happens in the pre-forward and +# pre-backward each iteration. +_FSDP_SKIP_WRITEBACK_CHECK = "FSDP_SKIP_WRITEBACK_CHECK" + +# Env var toggling whether when model is in .eval() mode, should we run in fp32 +# or the reduced precision. +_FSDP_USE_FULL_PREC_IN_EVAL = "FSDP_USE_FULL_PREC_IN_EVAL" + +# Some value to set padding in tensors to for debuggability +_FLAT_PARAM_PADDING_VALUE = 42 + +# Environment variables for disabling the all-gather and reduce-scatter +# communication ops for ablation studies. Note that without these communication +# ops the training won't converge, and you probably need to disable correctness +# checks in your model. +_FSDP_USE_FAKE_ALL_GATHER = "FSDP_USE_FAKE_ALL_GATHER" +_FSDP_USE_FAKE_REDUCE = "FSDP_USE_FAKE_REDUCE" + + +# TODO: Define this for now to avoid circular imports. See if we can remove. +class HandleShardingStrategy(Enum): + FULL_SHARD = auto() + SHARD_GRAD_OP = auto() + NO_SHARD = auto() + HYBRID_SHARD = auto() + _HYBRID_SHARD_ZERO2 = auto() + + +RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES = ( + HandleShardingStrategy.FULL_SHARD, + HandleShardingStrategy.HYBRID_SHARD, +) +NO_RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES = ( + HandleShardingStrategy.SHARD_GRAD_OP, + HandleShardingStrategy._HYBRID_SHARD_ZERO2, +) + + +class ParamInfo(NamedTuple): + """Information for an original parameter.""" + + param_name: str # unprefixed + module: nn.Module + module_name: str + + +class SharedParamInfo(NamedTuple): + """ + Additional information for a shared parameter. + + For each shared parameter, we designate one module and its parameter + variable to be the primary owner, determined as the first one encountered + in the parameter walk. These are prefixed with "prim". The primary module + and parameter do not have their own :class:`SharedParamInfo` instance. + """ + + param_name: str # unprefixed + module: nn.Module + module_name: str + prim_param_name: str # unprefixed + prim_module: nn.Module + prim_module_name: str + + +class _ShardParamInfo(NamedTuple): + """Shard-related information for an original parameter.""" + + in_shard: bool + # Use to index into the sharded flat parameter, e.g. + # `flat_param[offset_in_shard : offset_in_shard + numel_in_shard]` + offset_in_shard: int | None + numel_in_shard: int | None + # Use to get part of the parameter in the local shard from a flattened + # version of the unsharded parameter, e.g. either + # `param.flatten()[intra_param_start_idx : intra_param_end_idx + 1]` or + # `param.as_strided((param.numel(),), (1,))[intra_param_start_idx : intra_param_end_idx + 1]` + intra_param_start_idx: int | None + intra_param_end_idx: int | None # inclusive + + +class FlatParamShardMetadata(NamedTuple): + """ + This holds metadata specific to this rank's shard of the flat parameter. + + Attributes: + param_names (Tuple[str, ...]): Prefixed parameter names of this rank's + shard of the parameters; see :class:`FlatParameter`. + param_shapes (Tuple[torch.Size, ...]): Parameter shapes of this rank's + shard of the parameters; see :class:`FlatParameter`. + param_strides (Tuple[torch.Size, ...]): Parameter strides of this rank's + shard of the parameters; see :class:`FlatParameter`. + param_contiguities (Tuple[bool, ...]): Parameter `.contiguous` call results + of this rank's shard of the parameters; see :class:`FlatParameter`. + param_numels (Tuple[int, ...]): Parameter numels of this rank's shard + of the parameters; see :class:`FlatParameter`. + param_offsets (Tuple[Tuple[int, int], ...]): [start, end] offsets (in + units of numels) giving this rank's part of each flattened + original parameter. + """ + + param_names: tuple[str, ...] + param_shapes: tuple[torch.Size, ...] + param_strides: tuple[tuple[int, ...], ...] + param_contiguities: tuple[bool, ...] + param_numels: tuple[int, ...] + param_offsets: tuple[tuple[int, int], ...] + + +class _FlatParameterMeta(_ParameterMeta): + # Make `isinstance(t, FlatParameter)` return True for custom tensor + # instances that have the _is_flat_param flag for BC + def __instancecheck__(self, instance): + # NB: do NOT test the super implementation + return isinstance(instance, torch.Tensor) and getattr( + instance, "_is_flat_param", False + ) + + +class FlatParameter(nn.Parameter, metaclass=_FlatParameterMeta): + """ + This is the flat parameter used by :class:`FullyShardedDataParallel`. + + It is comprised of one or more original parameters, which are flattened and + concatenated to construct the flat parameter. + + Under the current design, this parameter logically represents both the + unsharded and sharded flat parameter, and its data changes storages + dynamically. + - In the :class:`FullyShardedDataParallel` constructor, the parameter + is initialized as unsharded and then sharded in-place. + - At runtime, the parameter is lazily (re)-initialized. The sharded + parameter data is saved in ``self._local_shard``, and a new ``Tensor`` + ``self._full_param_padded`` is created, which is the all-gather + destination and owns the unsharded parameter storage thereafter. (See + :meth:`FlatParamHandle.init_flat_param_attributes`.) + - Throughout runtime, the parameter data changes storages as needed, + e.g. to the sharded flat parameter, low precision sharded flat + parameter, or the unsharded flat parameter. + + NOTE: Since ``use_orig_params=True`` supports intra-``FlatParameter`` + padding, we have two versions of the per-parameter numels, one that + includes the padding (``_numels_with_padding``) and one that does not + (``_numels``). The former may have length longer than the other data + structures, while the latter has the same length as the number of actual + original parameters like the other per-parameter data structures. + + NOTE: This is not a real class; instead, you will always get a Parameter + back out if you try to create one of these. This is similar to the trick + we implemented for Parameter to get it to work with subclasses; this + is primarily so that FlatParameter supports combination with FakeTensor. + + Attributes: + _unpadded_unsharded_size (torch.Size): Unsharded flat parameter's size + without right-hand-side padding for divisibility by the world size. + For ``use_orig_params=True``, this includes alignment padding. + _padded_unsharded_size (torch.Size): Unsharded flat parameter's size + with right-hand-side padding for divisibility by the world size. + For ``use_orig_params=True``, this includes alignment padding. This + is only set for sharded strategies since they require padding for + the all-gather. + _sharded_size (torch.Size): Sharded flat parameter's size with padding. + This is also set for ``NO_SHARD``, in which case it is the same as + the unsharded sizes. (We omit "padded" because there is no + analogous unpadded one.) + + _num_params (int): Number of original parameters flattened into this + flat parameter. This is the length of the per-parameter data + structures. + _param_infos (Tuple[ParamInfo, ...]): Each parameter's parameter info + entry; see :class:`ParamInfo` for details. + _shapes (Tuple[torch.Size, ...]): Each parameter's original shape. + _strides (Tuple[torch.Size, ...]): Each parameter's original stride. + _contiguities (Tuple[bool, ...]): Each parameter's ``contiguous()`` + call result. + _fqns (Tuple[str, ...]): Each parameter's fully-qualified name (FQN) + prefixed from the ``_fully_sharded_module``. The names are + guaranteed to be unique in the subtree rooted at that module. + _param_extensions (Tuple[Optional[Any], ...]): Each parameter's + extension (i.e. some per-parameter state) used to customize + pre-flatten and post-unflatten behavior or ``None``. This is + experimental, and users should not depend on its existence in the + future. + _numels_with_padding (Tuple[int, ...]): Each parameter's numel + including entries for the padding. This is used to construct views + into the flat parameter via ``torch.split()``. This may have length + longer than ``_num_params``. + _numels (Tuple[int, ...]): Each parameter's numel excluding entries for + padding. This has length equal to ``_num_params``. + _shard_param_infos (Tuple[_ShardParamInfo, ...]): Each parameter's + shard parameter info; see :class:`_ShardParamInfo` for details. + _shared_param_infos (Tuple[SharedParamInfo, ...]): Shared parameter + info entries; see :class:`SharedParamInfo` for details. + _modules (set[nn.Module]): Modules that contain some original parameter + that is flattened into the flat parameter. + + _shard_numel_padded (int): Numel padded for this rank's sharded flat + parameter. + _local_shard (Tensor): Sharded flat parameter with padding if using a + sharded strategy. If using ``NO_SHARD``, then this is the unpadded + unsharded flat parameter, and there is no notion of a sharded flat + parameter or padded unsharded flat parameter. + _full_param_padded (Tensor): Unsharded flat parameter with padding. + This is not defined for ``NO_SHARD``. When using mixed precision + for parameters, this has the low precision. + _full_prec_full_param_padded (Tensor): Full precision unsharded flat + parameter with padding. This is used for unsharding outside of + computation when using mixed precision for parameters. This is + never defined for ``NO_SHARD``. + _post_backward_hook_handle (RemovableHandle): + Flat parameter's post-backward hook handle. (Compile only) + _post_backward_hook_state (Tuple[AccumulateGrad, RemovableHandle]): + Flat parameter's :class:`AccumulateGrad` object and post-backward + hook handle. (Eager only) + _mp_shard (Tensor): Low precision sharded flat parameter with padding. + This is only defined when parameter mixed precision is enabled. For + ``NO_SHARD``, this is used for computation. + _cpu_grad (Tensor): Sharded gradient with padding stored on CPU. + This is only defined when offloading parameters is enabled. + _saved_grad_shard (Tensor): Sharded gradient with padding from previous + iterations for gradient accumulation without :meth:`no_sync`. + + _params (Optional[List[nn.Parameter]]): If ``use_orig_params=True``, + then each original parameter variable; otherwise, ``None``. This + does not include any padding tensors. + _shared_params (Optional[List[nn.Parameter]]): The original shared + parameter variables if ``use_orig_params=True`` and ``None`` + otherwise. + _tensors (Optional[List[Optional[Tensor]]]): This saves the ``Tensor`` + views created in the forward and tracked by autograd when + ``use_orig_params=True`` and is ``None`` otherwise. This is to + preserve those ``Tensor`` variables for the backward to ensure that + the ``FlatParameter`` 's ``AccumulateGrad`` object does not change + in which case the post-backward hook does not run. This is relevant + for cases like reentrant activation checkpointing. + _is_grad_none_mask (Optional[List[bool]]): If ``use_orig_params=True``, + a mask over the original parameters' gradients indicating if it is + logically ``None`` or not; otherwise, ``None``. This does not + include entries for padding. This mask is needed because only some + of the parameters may have ``None`` gradient, in which case the + flat gradient must be non-``None`` and must use zeros to + approximate those original ``None`` gradients. This mask informs + FSDP to set the original parameter gradients to ``None`` (instead + of zeros) as needed. + """ + + _unpadded_unsharded_size: torch.Size + _padded_unsharded_size: torch.Size + _sharded_size: torch.Size + _num_params: int + _param_infos: tuple[ParamInfo, ...] + _shapes: tuple[torch.Size, ...] + _strides: tuple[tuple[int, ...], ...] + _contiguities: tuple[bool, ...] + _fqns: tuple[str, ...] + _param_extensions: tuple[Any | None, ...] + _numels_with_padding: tuple[int, ...] + _numels: tuple[int, ...] + _shard_param_infos: tuple[_ShardParamInfo, ...] + _shared_param_infos: tuple[SharedParamInfo, ...] + _modules: set[nn.Module] + _shard_numel_padded: int + _local_shard: Tensor + _full_param_padded: Tensor + _full_prec_full_param_padded: Tensor + # Eager only + _post_backward_hook_state: tuple[Any, Any] + # Compile only + _post_backward_hook_handle: Any + _mp_shard: Tensor + _cpu_grad: Tensor + _saved_grad_shard: Tensor + _params: list[nn.Parameter] | None + _shared_params: list[nn.Parameter] | None + _tensors: list[Tensor | None] | None + _is_grad_none_mask: list[bool] | None + + _is_padding_mask: list[bool] + + def __new__(cls, data=None, requires_grad=True): + if cls is not FlatParameter: + raise AssertionError("subclasses FlatParameter not supported") + r = nn.Parameter.__new__(nn.Parameter, data, requires_grad) # type: ignore[call-arg] + r._is_flat_param = True # type: ignore[attr-defined] + return r + + # NB: This is not a regular method, because FlatParameters are not actually + # instances of this class (see __new__ above). So you must indirectly + # call this directly through the classmethod. + @classmethod + def _init_metadata( + cls, + self, + param_infos: list[ParamInfo], + numels: list[int], + shapes: list[torch.Size], + strides: list[tuple[int, ...]], + contiguities: list[bool], + fqns: list[str], + shared_param_infos: list[SharedParamInfo], + param_extensions: list[Any | None], + params: list[nn.Parameter] | None, + shared_params: list[nn.Parameter] | None, + is_padding_mask: list[bool], + ) -> None: + """ + Initialize attributes holding metadata about the original parameters comprising the flat parameter. + + We expose this method separate from the constructor to keep the + constructor only responsible for the flat parameter's tensor data. This + method should only be called once per model, while the constructor may + be called multiple times, e.g. when reloading from a checkpoint, in + which case only the tensor data needs to be passed to the constructor. + Since :meth:`load_state_dict` is implemented via :meth:`copy_`, the + metadata is correctly assumed to be unchanged. + + Args: + See the Attributes in the class docstring. + """ + if len(param_infos) != len(shapes): + raise AssertionError( + f"Expected param_infos length {len(param_infos)} to match shapes length {len(shapes)}" + ) + if len(param_infos) != len(strides): + raise AssertionError( + f"Expected param_infos length {len(param_infos)} to match strides length {len(strides)}" + ) + if len(param_infos) != len(contiguities): + raise AssertionError( + f"Expected param_infos length {len(param_infos)} to match contiguities length {len(contiguities)}" + ) + if len(param_infos) != len(fqns): + raise AssertionError( + f"Expected param_infos length {len(param_infos)} to match fqns length {len(fqns)}" + ) + if len(param_infos) != len(param_extensions): + raise AssertionError( + f"Expected param_infos length {len(param_infos)} to match param_extensions length {len(param_extensions)}" + ) + self._num_params = len(param_infos) + self._param_infos = param_infos + self._shapes = shapes + self._strides = strides + self._contiguities = contiguities + self._fqns = fqns + self._param_extensions = param_extensions + self._is_padding_mask = is_padding_mask + + numels_without_padding: list[int] = [] + for numel, is_padding in zip(numels, is_padding_mask): + if not is_padding: + numels_without_padding.append(numel) + self._numels = tuple(numels_without_padding) + self._numels_with_padding = tuple(numels) + if len(self._numels) != self._num_params: + raise AssertionError( + f"Expected _numels length {len(self._numels)} to equal _num_params {self._num_params}" + ) + + self._shared_param_infos = tuple(shared_param_infos) + self._modules = {pi.module for pi in self._param_infos}.union( + {spi.module for spi in self._shared_param_infos} + ) + if (params is None) != (shared_params is None): + raise AssertionError( + "Expected params and shared_params to both be None or both be not None" + ) + if params is not None: + if shared_params is None or len(shared_params) != len(shared_param_infos): + raise AssertionError( + f"Expected shared_params to be not None and have length {len(shared_param_infos)}, got {shared_params}" + ) + self._params = [] + for param, is_padding in zip(params, is_padding_mask): + if not is_padding: + self._params.append(param) + if shared_params is not None: + self._shared_params = shared_params + else: + self._shared_params = [] + # Mark the original parameters to avoid flattening them into + # another `FlatParameter` during recursive construction + for param in chain(self._params, self._shared_params): + _set_fsdp_flattened(param) + self._is_grad_none_mask = [False for _ in range(self._num_params)] + self._tensors = [None for _ in range(self._num_params)] + else: + self._params = None + self._shared_params = None + self._is_grad_none_mask = None + self._tensors = None + self._unpadded_unsharded_size = self.size() + _set_fsdp_flattened(self) + # Tracks whether the `FlatParameter`'s post-backward hook has been + # called to modify the behavior of the post-backward callback + self._post_backward_called = False + + +class FlatParamHandle: + """ + A handle that manages a flat parameter (:class:`FlatParameter`). + + This includes sharding and view management. + + Args: + params (Sequence[nn.Parameter]): The parameters to flatten into the + flat parameter. + fully_sharded_module (nn.Module): See [Note: Fully Sharded Module]. + device (torch.device): The compute and communication device, which + should be a non-CPU device. We refer to it as the compute device. + sharding_strategy (ShardingStrategy): Sharding strategy to apply to + this handle's ``FlatParameter``. + offload_params (bool): Whether to offload the handle's + ``FlatParameter`` to CPU. + mp_param_dtype (Optional[torch.dtype]): Parameter mixed precision + setting passed to the FSDP constructor. + mp_reduce_dtype (Optional[torch.dtype]): Gradient reduction mixed + precision setting passed to the FSDP constructor. + keep_low_precision_grads (bool): Whether to keep gradients in low + precision. + use_orig_params (bool): If ``True``, then FSDP preserves the original + parameter variables and returns them from ``named_parameters()`` + (e.g. to support different optimizer hyperparameters within one + :class:`FlatParameter`). If ``False``, then FSDP reconstructs the + parameters every iteration and returns the :class:`FlatParameter` s + from ``named_parameters()``. + """ + + ################## + # INITIALIZATION # + ################## + def __init__( + self, + params: Sequence[nn.Parameter | Tensor], + fully_sharded_module: nn.Module, + device: torch.device, + sharding_strategy: HandleShardingStrategy, + offload_params: bool, + mp_param_dtype: torch.dtype | None, + mp_reduce_dtype: torch.dtype | None, + keep_low_precision_grads: bool, + process_group: dist.ProcessGroup, + use_orig_params: bool, + *, + fsdp_extension: FSDPExtensions | None = None, + ): + super().__init__() + params = list(params) + if len(params) == 0: + raise ValueError( + f"Cannot construct a {self.__class__.__name__} with an empty parameter list" + ) + self._init_setattr_fns() + self._skip_writeback_check = ( + os.environ.get(_FSDP_SKIP_WRITEBACK_CHECK, "") == "1" + ) + self._use_full_prec_in_eval = ( + os.environ.get(_FSDP_USE_FULL_PREC_IN_EVAL, "") == "1" + ) + self._use_fake_all_gather = os.environ.get(_FSDP_USE_FAKE_ALL_GATHER, "") == "1" + self._use_fake_reduce = os.environ.get(_FSDP_USE_FAKE_REDUCE, "") == "1" + if self._skip_writeback_check: + _warn_skip_writeback_check( + logger, + f"Since {_FSDP_SKIP_WRITEBACK_CHECK}=1, FSDP will not check " + "for parameter or gradient writeback. Changing parameter or " + "gradient storages may lead to silent correctness errors.", + ) + if self._use_fake_all_gather: + _warn_use_fake_all_gather( + logger, + f"Since {_FSDP_USE_FAKE_ALL_GATHER}=1, FSDP will not execute " + "all-gather ops. Your training will be incorrect, but " + "can reveal how much time spent on all-gather ops.", + ) + if self._use_fake_reduce: + _warn_use_fake_reduce( + logger, + f"Since {_FSDP_USE_FAKE_REDUCE}=1, FSDP will not execute " + "reduce-scatter ops. Your training will be incorrect, but " + "can reveal how much time spent on reduce-scatter ops.", + ) + # Only align addresses for `use_orig_params=True` (for now) + align_addresses = use_orig_params + self._init_get_unflat_views_fn(align_addresses) + self.device = device + self._device_handle = _FSDPDeviceHandle.from_device(self.device) + self.process_group = process_group + if self._use_fake_all_gather or self._use_fake_reduce: + self._fake_process_group = FakeProcessGroup._create_internal( + rank=process_group.rank(), world_size=process_group.size() + ) + self.rank = process_group.rank() + self.world_size = process_group.size() + self._sharding_strategy = sharding_strategy + self._offload_params = offload_params + self._use_orig_params = use_orig_params + self._keep_low_precision_grads = keep_low_precision_grads + self._training_state = HandleTrainingState.IDLE + self._debug_level = dist.get_debug_level() + self._fully_sharded_module = fully_sharded_module + # For strategies that do not free after forward, we skip using sharded + # views after forward since the unsharded data exists. We still switch + # `self.flat_param` to point to the sharded flat parameter since what + # it points to parameterizes behavior. We use the following attribute + # to track which tensor data the parameters are unsharded views into. + self._unsharded_flat_param_for_skipped_views: Tensor | None = None + # The index in the state's `all_handles`, which must be the + # same across ranks for the execution order validation to work + self._handle_index: int | None = None + # Index in handles_to_pre_forward_order + self._pre_forward_order_index: int | None = None + # Index in `handles_post_forward_order` + self._post_forward_index: int | None = None + # Used for guarding against mistargeted forward prefetches + self._needs_pre_forward_unshard = False + # Used for guarding against mistargeted backward prefetches + self._needs_pre_backward_unshard = False + # Was the handle prefetched? Set on successful _prefetch_handle and unshard + self._prefetched = False + # Optimistically assume a valid input `params` and set dtype attributes + # before `_init_flat_param()`, which performs the actual validation + self._orig_param_dtype = params[0].dtype + self._init_param_reduce_dtypes(mp_param_dtype, mp_reduce_dtype) + if self._fwd_bwd_param_dtype is None: + raise AssertionError("Expected _fwd_bwd_param_dtype to be not None") # mypy + self._aligned_numel = ( + _get_aligned_numel(unsharded_dtype=self._fwd_bwd_param_dtype) + if align_addresses + else 0 + ) + self._fsdp_extension = fsdp_extension + self._init_flat_param_and_metadata( + params, + fully_sharded_module, + self._aligned_numel, + use_orig_params, # type: ignore[arg-type] + ) + self._use_unsharded_views(as_params=False) + + def __repr__(self): + return f"FlatParamHandle(flat_param.fqns={self.flat_param._fqns})" + + def _init_setattr_fns(self): + use_unsafe_setattr = os.environ.get(_FSDP_USE_UNSAFE_SETATTR, "") == "1" + self._setattr_tensor: Callable[[nn.Module, str, Tensor], None] + self._setattr_param: Callable[[nn.Module, str, nn.Parameter], None] + if use_unsafe_setattr: + self._setattr_tensor = _unsafe_setattr_tensor + self._setattr_param = _unsafe_setattr_param + else: + self._setattr_tensor = _safe_setattr_tensor_or_param + self._setattr_param = _safe_setattr_tensor_or_param + + def _init_get_unflat_views_fn(self, align_addresses: bool): + self._get_unflat_views = ( + self._get_unflat_views_aligned + if align_addresses + else self._get_unflat_views_unaligned + ) + + def _init_flat_param_and_metadata( + self, + params: list[Tensor | nn.Parameter], + module: nn.Module, + aligned_numel: int, + use_orig_params: bool, + ) -> None: + """ + Initialize the ``FlatParameter`` and its metadata. + + NOTE: This should only be called once at construction time, after which + the ``FlatParameter`` metadata is assumed to be static. + + NOTE: The elements of ``params`` should only be ``Tensor`` s when + composing with ``DTensor`` -based tensor parallelism, in which case the + elements may be ``DTensor`` local shards. + """ + if len(params) == 0: + raise ValueError("Expects non-empty `params`") + if aligned_numel < 0: + raise ValueError( + f"Expects non-negative `aligned_numel` but got {aligned_numel}" + ) + ( + dtype, + flat_param_requires_grad, + device, + ) = self._validate_tensors_to_flatten(params) + params_set = set(params) + # For alignment padding, only `numels` gets strictly non-`None` + # elements, and all other lists get `None` elements for padding. + param_infos: list[ParamInfo] = [] + numels: list[int] = [] + shapes: list[torch.Size] = [] + strides: list[tuple[int, ...]] = [] + contiguities: list[bool] = [] + fqns: list[str] = [] + shared_param_infos: list[SharedParamInfo] = [] + shared_param_memo: dict[Tensor | nn.Parameter, tuple[nn.Module, str, str]] = {} + params_to_flatten: list[Tensor | nn.Parameter] = [] + shared_params: list[Tensor | nn.Parameter] = [] + param_extensions: list[Any] = [] + is_padding_mask: list[bool] = [] + total_numel = total_numel_without_padding = 0 + for submodule_name, submodule in module.named_modules(remove_duplicate=False): + for param_name, param in _named_parameters_with_duplicates( + submodule, recurse=False + ): + if param not in params_set: + continue + if param in shared_param_memo: # shared reference + prim_module, prim_module_name, prim_param_name = shared_param_memo[ + param + ] + shared_params.append(param) + shared_param_infos.append( + SharedParamInfo( + param_name, + submodule, + submodule_name, + prim_param_name, + prim_module, + prim_module_name, + ) + ) + else: + if aligned_numel > 0: + numel_to_pad = aligned_numel - (total_numel % aligned_numel) + if numel_to_pad > 0 and numel_to_pad < aligned_numel: + padding_tensor = _construct_padding_tensor( + numel_to_pad, dtype, False, device + ) + params_to_flatten.append(padding_tensor) + is_padding_mask.append(True) + numels.append(numel_to_pad) + total_numel += numel_to_pad + transform_t, extension = _ext_pre_flatten_transform( + param, + self._fsdp_extension, + ) + param = cast(nn.Parameter, transform_t) + param_extensions.append(extension) + shared_param_memo[param] = (submodule, submodule_name, param_name) + params_to_flatten.append(param) + is_padding_mask.append(False) + param_infos.append(ParamInfo(param_name, submodule, submodule_name)) + numels.append(param.numel()) + shapes.append(param.shape) + strides.append(param.stride()) + contiguities.append(_is_truly_contiguous(param)) + fqn = ( + submodule_name + "." + param_name + if submodule_name + else param_name + ) + fqns.append(fqn) + total_numel += param.numel() + total_numel_without_padding += param.numel() + if len(params_to_flatten) == 0: + raise ValueError( + f"`params` were not found in `module`'s tree" + f"params: {params}\nmodule: {module}" + ) + if ( + self.rank == 0 + and aligned_numel > 0 + and total_numel != total_numel_without_padding + ): + logger.debug( + "FSDP FlatParameter address alignment created " + "%s numel of padding (%s vs. %s)", + total_numel - total_numel_without_padding, + total_numel, + total_numel_without_padding, + ) + if aligned_numel > 0: + # Pad to be divisible by world size to avoid a copy for the + # post-backward reduce-scatter + numel_to_pad = self.world_size - (total_numel % self.world_size) + if numel_to_pad > 0 and numel_to_pad < self.world_size: + if self.rank == 0: + logger.info( + "FSDP FlatParameter world size divisibility created " + "%s numel of padding", + numel_to_pad, + ) + padding_tensor = _construct_padding_tensor( + numel_to_pad, dtype, False, device + ) + params_to_flatten.append(padding_tensor) + is_padding_mask.append(True) + numels.append(numel_to_pad) + total_numel += numel_to_pad + # Pass `aligned_numel=0` since we already included padding tensors + self.flat_param: FlatParameter = self.flatten_tensors_into_flat_param( + params_to_flatten, + aligned_numel=0, + requires_grad=flat_param_requires_grad, + ) + FlatParameter._init_metadata( + self.flat_param, + param_infos, + numels, + shapes, + strides, + contiguities, + fqns, + shared_param_infos, + param_extensions, + _convert_to_params(params_to_flatten) if use_orig_params else None, + _convert_to_params(shared_params) if use_orig_params else None, + is_padding_mask, + ) + + def _validate_tensors_to_flatten( + self, tensors: list[Tensor | nn.Parameter] + ) -> tuple: + """Validate the tensors to flatten and returns any necessary metadata.""" + dtype: torch.dtype | None = None + # Return as the logical OR over each tensor's value + flat_param_requires_grad: bool | None = None + device: torch.device | None = None + # For `use_orig_params=True`, permit non-uniform `requires_grad` + for tensor in tensors: + if isinstance(tensor, FlatParameter): + raise ValueError("Cannot flatten a `FlatParameter`") + if dtype is None and not tensor.is_floating_point(): + raise ValueError("Cannot flatten integer dtype tensors") + if dtype is not None and tensor.dtype != dtype: + raise ValueError( + f"Must flatten tensors with uniform dtype but got {dtype} " + f"and {tensor.dtype}" + ) + if ( + not self._use_orig_params + and flat_param_requires_grad is not None + and tensor.requires_grad != flat_param_requires_grad + ): + raise ValueError( + "Must flatten tensors with uniform `requires_grad` when " + "`use_orig_params=False`" + ) + if device is not None and tensor.device != device: + raise ValueError( + "Must flatten tensors on the same device but got both " + f"{device} and {tensor.device}" + ) + dtype = tensor.dtype + flat_param_requires_grad = flat_param_requires_grad or tensor.requires_grad + device = tensor.device + if flat_param_requires_grad is None: + raise AssertionError("Requires non-empty `tensors` list") + return dtype, flat_param_requires_grad, device + + def flatten_tensors( + self, + tensors: list[Tensor], + aligned_numel: int, + ) -> Tensor: + """ + Flatten ``tensors`` into a single flat tensor. + + The flattening optionally includes + padding if ``aligned_numel`` is greater than 0, where ``aligned_numel`` + gives the numel required to have address alignment. + + NOTE: The padding alignment algorithm must be kept in sync with + :meth:`_init_flat_param_metadata`. We separate the two methods because + the initialization happens once, whereas this method may be called + multiple times throughout training (e.g. for checkpointing). + """ + if len(tensors) == 0: + raise ValueError("Expects non-empty `tensors`") + if aligned_numel < 0: + raise ValueError( + f"Expects non-negative `aligned_numel` but got {aligned_numel}" + ) + dtype, _, device = self._validate_tensors_to_flatten(tensors) + flat_tensors: list[Tensor] = [] + if aligned_numel > 0: + total_numel = 0 + for tensor in tensors: + numel_to_pad = aligned_numel - (total_numel % aligned_numel) + if numel_to_pad > 0 and numel_to_pad < aligned_numel: + padding_tensor = _construct_padding_tensor( + numel_to_pad, dtype, False, device + ) + flat_tensors.append(padding_tensor) + total_numel += numel_to_pad + flat_tensors.append( + torch.flatten(_detach_if_needed(tensor)) + if _is_truly_contiguous(tensor) + else _detach_if_needed(tensor).as_strided((tensor.numel(),), (1,)) + ) + total_numel += tensor.numel() + numel_to_pad = self.world_size - (total_numel % self.world_size) + if numel_to_pad > 0 and numel_to_pad < self.world_size: + padding_tensor = _construct_padding_tensor( + numel_to_pad, dtype, False, device + ) + flat_tensors.append(padding_tensor) + total_numel += numel_to_pad + else: + flat_tensors = [ + torch.flatten(_detach_if_needed(tensor)) + if _is_truly_contiguous(tensor) + else _detach_if_needed(tensor).as_strided((tensor.numel(),), (1,)) + for tensor in tensors + ] + return torch.cat(flat_tensors, dim=0) + + def flatten_tensors_into_flat_param( + self, + tensors: list[Tensor], + aligned_numel: int, + requires_grad: bool, + ) -> FlatParameter: + flat_param_data = self.flatten_tensors(tensors, aligned_numel) + return FlatParameter(flat_param_data, requires_grad=requires_grad) + + def _init_param_reduce_dtypes( + self, + mp_param_dtype: torch.dtype | None, + mp_reduce_dtype: torch.dtype | None, + ) -> None: + """ + Initialize param and reduce dtypes. + + Precondition: ``self.flat_param`` is set. This ensures that this + handle's parameters have a single dtype. + + Postcondition: This sets ``self._fwd_bwd_param_dtype`` and + ``self._reduce_dtype``. If ``mp_param_dtype`` or ``mp_reduce_dtype`` + is ``None``, then we assume the original parameter dtype. One special + case is if ``mp_param_dtype`` is not ``None`` and ``mp_reduce_dtype`` + is ``None``, in which case we assume the gradient reduction dtype + matches the forward/backward parameter dtype. + """ + # Save whether these dtypes were specified so that we permit the + # parameter dtype to change up until the lazy initialization + self._low_prec_param_dtype_specified = mp_param_dtype is not None + self._low_prec_reduce_dtype_specified = mp_reduce_dtype is not None + if ( + self._low_prec_param_dtype_specified + and not self._low_prec_reduce_dtype_specified + ): + # Special case: infer gradient reduction mixed precision + self._fwd_bwd_param_dtype = mp_param_dtype + self._reduce_dtype = self._fwd_bwd_param_dtype + else: + self._fwd_bwd_param_dtype = mp_param_dtype or self._orig_param_dtype + self._reduce_dtype = mp_reduce_dtype or self._orig_param_dtype + if self._fwd_bwd_param_dtype is None: + raise AssertionError("Expected _fwd_bwd_param_dtype to be not None") + if self._reduce_dtype is None: + raise AssertionError("Expected _reduce_dtype to be not None") + + ################################### + # SHARD INITIALIZATION & METADATA # + ################################### + @torch.no_grad() + def shard(self): + """ + Shard the handle's ``FlatParameter``. + + This allocates new memory for + the sharded flat parameter and frees the unsharded flat parameter's + storage. + + Postcondition: ``self.flat_param`` is the sharded flat parameter. Shard + metadata attributes are set for all sharding strategies. + """ + flat_param = self.flat_param + if not self.uses_sharded_strategy: + self._init_shard_metadata(0, 0, flat_param.numel() - 1) + else: + _p_assert( + flat_param.storage_offset() == 0, + "The `FlatParameter` is not the sole occupant of its storage", + ) + sharded_flat_param, numel_padded = FlatParamHandle._get_shard( + flat_param, self.rank, self.world_size + ) + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + allocated = flat_param._typed_storage()._size() > 0 + if allocated: + flat_param._typed_storage()._resize_(0) + flat_param.set_(sharded_flat_param) # type: ignore[call-overload] + start_idx = sharded_flat_param.numel() * self.rank + end_idx = sharded_flat_param.numel() * (self.rank + 1) - 1 # inclusive + self._init_shard_metadata(numel_padded, start_idx, end_idx) + if self._use_orig_params: + self._use_sharded_views() + + def _init_shard_metadata( + self, + numel_padded: int, + unsharded_start_idx: int, + unsharded_end_idx: int, + ) -> None: + """ + Initialize shard-related metadata for this rank's shard of the flat parameter. + + This includes ``_sharded_size``, ``_shard_param_infos``, and ``_shard_numel_padded``. + + Args: + numel_padded (int): Numel padded for this rank's sharded flat + parameter. + unsharded_start_idx (int): Start index in the unsharded flat + parameter assigned to this rank. + unsharded_end_idx (int): End index (inclusive) in the unsharded + flat parameter assigned to this rank. + + Precondition: ``self.flat_param`` 's data is the sharded flat + parameter. + """ + flat_param = self.flat_param + flat_param._sharded_size = flat_param.size() # type: ignore[attr-defined] + sharded_flat_param_numel = flat_param.numel() # includes `numel_padded` + _p_assert( + unsharded_start_idx >= 0 and unsharded_start_idx <= unsharded_end_idx, + f"unsharded_start_idx: {unsharded_start_idx} unsharded_end_idx: {unsharded_end_idx}", + ) + _p_assert( + numel_padded <= sharded_flat_param_numel, + f"numel_padded: {numel_padded} " + f"sharded_flat_param_numel: {sharded_flat_param_numel}", + ) + shard_param_infos = self._get_shard_metadata( + unsharded_start_idx, unsharded_end_idx + ) + if len(shard_param_infos) != flat_param._num_params: + raise AssertionError( + f"Expects length {flat_param._num_params} but got {len(shard_param_infos)}" + ) + flat_param._shard_param_infos = shard_param_infos # type: ignore[attr-defined] + flat_param._shard_numel_padded = numel_padded # type: ignore[attr-defined] + + def _get_shard_metadata( + self, + unsharded_start_idx: int, + unsharded_end_idx: int, + ) -> tuple[_ShardParamInfo, ...]: + """ + Compute the shard metadata based on ``unsharded_start_idx`` and ``unsharded_end_idx`` (inclusive). + + ``unsharded_start_idx`` and ``unsharded_end_idx`` give the interval of the + unsharded flat parameter specifying the shard. + """ + flat_param_offsets = self._get_flat_param_offsets() + if len(flat_param_offsets) != len(self.flat_param._numels_with_padding): + raise AssertionError( + f"Expected {len(self.flat_param._numels_with_padding)} but got {len(flat_param_offsets)}" + ) + shard_param_infos: list[_ShardParamInfo] = [] + sharded_flat_param_numel = unsharded_end_idx - unsharded_start_idx + 1 + # `unsharded_param_start_idx` and `unsharded_param_end_idx` are indices + # into the unsharded flat parameter (inclusive) of the given parameter + for ( + (unsharded_param_start_idx, unsharded_param_end_idx), + is_padding, + ) in zip(flat_param_offsets, self.flat_param._is_padding_mask): + if is_padding: + continue + in_sharded_flat_param = ( + unsharded_start_idx <= unsharded_param_end_idx + and unsharded_end_idx >= unsharded_param_start_idx + ) + if not in_sharded_flat_param: + shard_param_info = _ShardParamInfo(False, None, None, None, None) + else: + if unsharded_start_idx <= unsharded_param_start_idx: + # This branch can only happen once since the rank's + # unsharded start index can only intersect one parameter + intra_param_start_idx = 0 + offset_in_shard = unsharded_param_start_idx - unsharded_start_idx + else: + intra_param_start_idx = ( + unsharded_start_idx - unsharded_param_start_idx + ) + offset_in_shard = 0 + if not ( + offset_in_shard >= 0 and offset_in_shard < sharded_flat_param_numel + ): + raise AssertionError( + f"Invalid `offset_in_shard` of {offset_in_shard} for " + f"sharded flat parameter with {sharded_flat_param_numel} numel" + ) + intra_param_end_idx = ( + min(unsharded_param_end_idx, unsharded_end_idx) + - unsharded_param_start_idx + ) + numel_in_shard = intra_param_end_idx - intra_param_start_idx + 1 + shard_param_info = _ShardParamInfo( + True, + offset_in_shard, + numel_in_shard, + intra_param_start_idx, + intra_param_end_idx, + ) + shard_param_infos.append(shard_param_info) + return tuple(shard_param_infos) + + @staticmethod + def _get_unpadded_shard( + tensor: Tensor, + rank: int, + world_size: int, + ) -> tuple[Tensor, int]: + """ + Return the unpadded shard of ``tensor`` for the given ``rank`` and ``world_size``. + + The returned value is a tuple of the shard of ``tensor`` without any + padding and the numel to pad for that shard. + + If ``tensor`` is already flattened or may be viewed in the flattened + shape (which is true in the expected usage), then this method does not + allocate any new tensor memory. + """ + chunks = ( + torch.flatten(tensor).chunk(world_size) + if _is_truly_contiguous(tensor) + else tensor.as_strided((tensor.numel(),), (1,)).chunk(world_size) + ) + if len(chunks) < (rank + 1): + # This rank gets an empty chunk fully padded with zeros since there + # are not enough chunks across ranks + chunk = chunks[0].new_empty(0) + else: + chunk = chunks[rank] + numel_to_pad = chunks[0].numel() - chunk.numel() + if numel_to_pad < 0: + raise AssertionError( + "Chunk's size should be at most the first chunk's size" + ) + return chunk, numel_to_pad + + @staticmethod + def _get_shard( + tensor: Tensor, + rank: int, + world_size: int, + ) -> tuple[Tensor, int]: + """ + Return the shard of ``tensor`` with padding for the given ``rank`` and ``world_size`` and the numel padded for that shard. + + This method allocates new memory (via :meth:`clone`) since the + unsharded ``tensor`` may be deallocated after this method returns. + """ + chunk, numel_to_pad = FlatParamHandle._get_unpadded_shard( + tensor, rank, world_size + ) + shard = chunk.clone() + if numel_to_pad > 0: + shard = F.pad(shard, [0, numel_to_pad]) + return shard, numel_to_pad + + @staticmethod + def _get_sharded_size(tensor: Tensor, rank: int, world_size: int) -> torch.Size: + """ + Return the shape of ``tensor`` after sharding including padding. + + This requires ``tensor`` to have 1D shape and ensures that the returned + shape is 1D. + """ + if len(tensor.shape) != 1: + raise AssertionError(f"Expected 1D tensor shape, got {tensor.shape}") + unpadded_sharded_tensor, numel_to_pad = FlatParamHandle._get_unpadded_shard( + tensor, rank, world_size + ) + unpadded_sharded_size = unpadded_sharded_tensor.size() + if len(unpadded_sharded_size) != 1: + raise AssertionError( + f"Expected 1D unpadded_sharded_size, got {unpadded_sharded_size}" + ) + return torch.Size([unpadded_sharded_size[0] + numel_to_pad]) + + def _get_flat_param_offsets(self) -> list[tuple[int, int]]: + """ + Return [start, end] offsets of each original parameter's flattened data in the unsharded flat parameter (without padding). + + NOTE: The returned list includes elements for alignment padding. + """ + cumulative_sum = list(accumulate(self.flat_param._numels_with_padding)) + starts = [0] + cumulative_sum[:-1] + ends = [end - 1 for end in cumulative_sum] # inclusive + param_offsets = list(zip(starts, ends)) + return param_offsets + + @no_type_check + def shard_metadata( + self, + ) -> FlatParamShardMetadata: + """ + Return the shard-related metadata specific to this rank's shard of the flat parameter. + + NOTE: The returned tuple does not include elements for alignment + padding but does account for the padding. + """ + fqns_list = [] + shapes_list = [] + strides_list = [] + contiguities_list = [] + numels_list = [] + shard_param_offsets = [] + for fqn, shape, stride, contiguous, numel, shard_param_info in zip( + self.flat_param._fqns, + self.flat_param._shapes, + self.flat_param._strides, + self.flat_param._contiguities, + self.flat_param._numels, + self.flat_param._shard_param_infos, + ): + if not shard_param_info.in_shard: + continue + fqns_list.append(fqn) + shapes_list.append(shape) + strides_list.append(stride) + contiguities_list.append(contiguous) + numels_list.append(numel) + shard_param_offsets.append( + ( + shard_param_info.intra_param_start_idx, + shard_param_info.intra_param_end_idx, + ) + ) + return FlatParamShardMetadata( + tuple(fqns_list), + tuple(shapes_list), + tuple(strides_list), + tuple(contiguities_list), + tuple(numels_list), + tuple(shard_param_offsets), + ) + + @no_type_check + @torch.no_grad() + def init_flat_param_attributes(self) -> None: + """ + This initializes some attributes on the handle's ``FlatParameter``. + This should be called during lazy initialization since it requires the + parameter to be on the compute device if not offloading to CPU and we + want to give users the chance to move the parameter appropriately after + the FSDP constructor. + + For each tensor attribute on the ``FlatParameter``, see the unshard and + reshard methods in this class for the allocation and free pattern. + """ + flat_param = self.flat_param + if flat_param.dtype != self._orig_param_dtype: + # Entering this branch means that the user changed the parameter + # dtype after FSDP initialization, in which case we may need to + # refresh some saved dtype attributes (dtypes specified as a part + # of mixed precision take precedence). + if not self._low_prec_param_dtype_specified: + self._fwd_bwd_param_dtype = flat_param.dtype + # For `reduce_dtype`, require `param_dtype` was not specified since + # then we infer the `reduce_dtype` from the specified `param_dtype` + if ( + not self._low_prec_reduce_dtype_specified + and not self._low_prec_param_dtype_specified + ): + self._reduce_dtype = flat_param.dtype + self._orig_param_dtype = flat_param.dtype + cpu_device = torch.device("cpu") + if self._offload_params: + _p_assert( + flat_param.device == cpu_device, + f"Expects the `FlatParameter` to be on CPU when parameter CPU " + f"offloading is enabled, not {flat_param.device}", + ) + else: + self._check_on_compute_device(self.flat_param) + flat_param._local_shard = flat_param.data + if self._offload_params: + # Pin the memory for faster H2D transfer + flat_param._local_shard = flat_param._local_shard.pin_memory() + # Pre-allocate the sharded gradient on CPU to enable non-blocking + # D2H transfer during the backward pass + flat_param._cpu_grad = torch.zeros_like( + flat_param._local_shard, device=cpu_device + ).pin_memory() + if self._uses_param_mixed_precision: + # For parameter mixed precision, we maintain a low precision + # sharded tensor on the compute device to be all-gathered (for + # sharded strategies) or directly used (for `NO_SHARD`) for + # computation. + flat_param._mp_shard = torch.empty_like( + flat_param._local_shard, + device=self.device, + dtype=self._fwd_bwd_param_dtype, + ) + _free_storage(flat_param._mp_shard) + if self.uses_sharded_strategy: + # We maintain a padded unsharded tensor that serves as the + # all-gather destination and owns the original parameter storages. + unsharded_param_dtype = ( + self._fwd_bwd_param_dtype + if self._uses_param_mixed_precision + else flat_param.dtype + ) # use low precision if parameter mixed precision is enabled + padded_unsharded_numel = flat_param.numel() * self.world_size + flat_param._full_param_padded = torch.empty( + padded_unsharded_numel, + device=self.device, + dtype=unsharded_param_dtype, + ) + flat_param._padded_unsharded_size = flat_param._full_param_padded.size() + _free_storage(flat_param._full_param_padded) + + if self._uses_param_mixed_precision: + # For parameter mixed precision, we maintain a full precision + # padded unsharded tensor for when we force full precision. + flat_param._full_prec_full_param_padded = torch.empty( + padded_unsharded_numel, + device=self.device, + dtype=flat_param.dtype, # full precision + ) + _free_storage(flat_param._full_prec_full_param_padded) + + ################### + # UNSHARD/RESHARD # + ################### + def pre_unshard(self) -> bool: + """ + Return ``False`` if this is a no-op and ``True`` otherwise. + + Postcondition: ``self.flat_param`` 's data is on the device for + communication and is what should be all-gathered. This means that it + matches the dtype of the expected unsharded parameter. + """ + if ( + self._training_state == HandleTrainingState.SUMMON_FULL_PARAMS + and self._skipped_use_sharded_views + ): + # Since this path imposes special semantics for the unsharded flat + # parameter (e.g. forcing full precision), use sharded views to + # reuse the existing logic for that special handling + self._use_sharded_views() + ret = False + if self._use_orig_params and not self._skip_writeback_check: + ret = self._writeback_orig_params() + if ( + self.uses_sharded_strategy + and not self._offload_params + and not self.needs_unshard() + ): + pass # no-op + elif self._uses_param_mixed_precision and not self._force_full_precision: + self._use_low_precision_shard() + ret = True + elif self._offload_params and self.flat_param.device != self.device: + # NOTE: This creates a new tensor distinct from any attributes. + self.flat_param_to(self.device, non_blocking=True) + ret = True + self._check_on_compute_device(self.flat_param) + return ret + + def _use_low_precision_shard(self): + """Allocate on the compute device and switch to using the low precision sharded flat parameter.""" + self._check_low_precision_shard() + flat_param = self.flat_param + _alloc_storage( + flat_param._mp_shard, + flat_param._local_shard.size(), # type: ignore[attr-defined] + ) + # `copy_()` implicitly casts to the low precision + flat_param._mp_shard.copy_( # type: ignore[attr-defined] + flat_param._local_shard.to( # type: ignore[attr-defined] + self.device, non_blocking=True + ) + ) + # Invariant: `_mp_shard` is always on the compute device. + flat_param.data = flat_param._mp_shard # type: ignore[attr-defined] + + def unshard(self): + """ + Run the unshard logic. + + This includes all-gathering the flat parameter + and switching to using the unsharded flat parameter. If the handle does + not need unsharding, then this only switches to using the unsharded + flat parameter. For ``NO_SHARD``, this is a no-op. + + If FSDP is in :meth:`summon_full_params` and the handle uses parameter + mixed precision, then the parameter is forced to full precision. + """ + if not self.needs_unshard(): + # Even when not needing an unshard, we should switch to using + # the unsharded flat parameter + unsharded_flat_param = ( + self._get_padded_unsharded_flat_param() + if self.uses_sharded_strategy + else self.flat_param + ) + self._use_unsharded_flat_param(unsharded_flat_param) + return + unsharded_flat_param = self._alloc_padded_unsharded_flat_param() + padded_unsharded_flat_param = self._all_gather_flat_param(unsharded_flat_param) + self._use_unsharded_flat_param(padded_unsharded_flat_param) + + def needs_unshard(self) -> bool: + """Return if the handle's flat parameter needs to be unsharded.""" + if not self.uses_sharded_strategy: + return False + unsharded_flat_param = self._get_padded_unsharded_flat_param() + already_unsharded = _same_storage_size( + unsharded_flat_param, unsharded_flat_param.numel() + ) + return not already_unsharded + + def _alloc_padded_unsharded_flat_param(self): + """ + Allocate the *padded* unsharded flat parameter. + + The unpadded unsharded + flat parameter is always a view into the padded one. This padded + parameter is saved to a different attribute on the ``FlatParameter`` + depending on if we force full precision. + """ + self._check_sharded_strategy() + flat_param = self.flat_param + unsharded_flat_param = self._get_padded_unsharded_flat_param() + self._check_storage_freed(unsharded_flat_param) + _alloc_storage(unsharded_flat_param, flat_param._padded_unsharded_size) # type: ignore[attr-defined] + return unsharded_flat_param + + def _get_padded_unsharded_flat_param(self) -> torch.Tensor: + """ + Return a reference to the padded unsharded flat parameter depending on the calling context. + + This should only be called if using a sharded strategy. + """ + self._check_sharded_strategy() + flat_param = self.flat_param + if self._force_full_precision and self._uses_param_mixed_precision: + # When parameter mixed precision is enabled, we use a different + # tensor as the all-gather destination to preserve the invariant + # that `_full_param_padded` is in the low precision + unsharded_flat_param = flat_param._full_prec_full_param_padded # type: ignore[attr-defined] + _p_assert( + unsharded_flat_param.dtype != self._fwd_bwd_param_dtype, + f"Expects full precision but got {self._fwd_bwd_param_dtype}", + ) + # For no-reshard-after-forward strategies, `_full_param_padded` may + # still be allocated from a previous forward. As we are forcing + # full precision here, the full-precision unsharded copy may be + # modified, invalidating the existing low-precision unsharded copy, + # so we should free it here to ensure a new all-gather for the next + # forward/backward computation to persist the modifications. + if flat_param._full_param_padded.untyped_storage().size() > 0: + _free_storage(flat_param._full_param_padded) + else: + unsharded_flat_param = flat_param._full_param_padded # type: ignore[attr-defined] + return unsharded_flat_param + + def _all_gather_flat_param( + self, + padded_unsharded_flat_param: Tensor, + ) -> Tensor: + """ + All-gather the handle's flat parameter to the destination ``padded_unsharded_flat_param``. + + Then switch to use the all-gathered tensor. + """ + _p_assert( + hasattr(self, "process_group") and hasattr(self, "world_size"), + "Expects a process group and world size to have been set via `shard()`", + ) + sharded_flat_param = self.flat_param.data + expected_numel = sharded_flat_param.numel() * self.world_size + _p_assert( + padded_unsharded_flat_param.numel() == expected_numel, + f"Expects {expected_numel} numel but got {padded_unsharded_flat_param.numel()}", + ) + + pg = ( + self._fake_process_group + if self._use_fake_all_gather + else self.process_group + ) + + # HACK this should be handled by C10D + if sharded_flat_param.is_cpu: # type: ignore[attr-defined] + tensor_list = list( + torch.chunk( + padded_unsharded_flat_param, + dist.get_world_size(pg), # type: ignore[arg-type] + ) + ) + dist.all_gather(tensor_list, sharded_flat_param, group=pg) + else: + dist.all_gather_single( + padded_unsharded_flat_param, + sharded_flat_param, + pg, + ) + + if self._offload_params: + # In case of offloading, `flat_param.data` (i.e. sharded param) is + # created on the pre-unshard stream. We need to hand it over to the + # unshard stream for all-gather + _no_dispatch_record_stream( + sharded_flat_param, + self._device_handle.current_stream(), # unshard_stream + ) + return padded_unsharded_flat_param + + def _use_unsharded_flat_param( + self, + padded_unsharded_flat_param: torch.Tensor, + ) -> None: + """ + Switch to use the *unpadded* unsharded flat parameter. + + This is a view into the *padded* unsharded flat parameter. + """ + unsharded_size = self.flat_param._unpadded_unsharded_size + flat_param_part = padded_unsharded_flat_param[: unsharded_size.numel()] + # slicing [:] is not visible to autograd because of .data + self.flat_param.data = flat_param_part + in_forward = self._training_state == HandleTrainingState.FORWARD + in_pre_backward = self._training_state == HandleTrainingState.BACKWARD_PRE + if self._use_orig_params: + if self._skipped_use_sharded_views and in_pre_backward: + # This call corresponds to the complementary pre-backward + # `_use_unsharded_views()` to the skipped pre-forward + # `_use_sharded_views()`, so we should skip this one too. + return + # We use `Tensor` views in the forward so that they are tracked by + # autograd. We use them in the pre-backward as well to support + # reentrant activation checkpointing, which needs the views to be + # tracked by autograd in the backward pass's recomputed forward. + self._use_unsharded_views( + as_params=(not in_forward and not in_pre_backward) + ) + elif in_forward: + self._use_unsharded_views(as_params=False) + + def post_unshard(self): + """ + Run the post-unshard logic. + + This includes freeing the low precision shard if needed. + """ + if self._uses_param_mixed_precision and self.uses_sharded_strategy: + self._free_low_precision_sharded_param() + self._check_on_compute_device(self.flat_param) + + def _free_low_precision_sharded_param(self): + """Frees the low precision sharded flat parameter.""" + self._check_low_precision_shard() + # `_mp_shard` is allocated in the pre-unshard stream, consumed in the + # unshard stream for sharded strategies, and consumed in both the + # unshard and default streams for `NO_SHARD`. For sharded strategies, + # the current stream here is the unshard stream, and for `NO_SHARD`, + # it is the default stream. For `NO_SHARD`, only recording for the + # default stream suffices since the default stream waits for the + # unshard stream. + _no_dispatch_record_stream( + self.flat_param._mp_shard, + self._device_handle.current_stream(), # type: ignore[attr-defined] + ) + _free_storage(self.flat_param._mp_shard) # type: ignore[attr-defined] + + @torch.no_grad() + def unshard_grad(self): + """ + Unshard the handle's ``FlatParameter``'s gradient. + + If all ranks have + ``None`` gradient, then all original parameters will as well. This + method performs an all-reduce and an all-gather. The additional + all-reduce is tolerable since this method is not meant to be used on + the computation critical path. + + Postcondition: ``_saved_grad_shard`` is defined and contains the value + to set ``flat_param.grad`` after gradients are resharded. + """ + if not self.uses_sharded_strategy: + self._use_unsharded_grad_views() + return + flat_param = self.flat_param + self._check_unsharded(flat_param) + + # Check if all ranks have a `None` gradient + num_grad_none = torch.zeros(1, dtype=torch.int32, device=self.device) + num_grad_none[0] = flat_param.grad is None + dist.all_reduce(num_grad_none, group=self.process_group) + if num_grad_none[0] == self.world_size: + flat_param._saved_grad_shard = None # type: ignore[assignment] + self._use_unsharded_grad_views() + return + + if flat_param.grad is None: + # In the case that only some ranks have `None` gradient, we use + # zeros to approximate as a best effort attempt + if self._debug_level == dist.DebugLevel.INFO: + warnings.warn( + f"[Rank {self.rank}] Only some but not all ranks have a " + "`None` `FlatParameter` gradient, so FSDP is using zeros to " + "approximate those ranks' sharded gradients being `None`", + stacklevel=2, + ) + flat_param._saved_grad_shard = None # type: ignore[assignment] + sharded_grad = torch.zeros(flat_param._sharded_size, device=self.device) # type: ignore[attr-defined] + else: + self._check_sharded(flat_param.grad) + flat_param._saved_grad_shard = flat_param.grad # type: ignore[attr-defined] + sharded_grad = flat_param._saved_grad_shard # type: ignore[attr-defined] + padded_unsharded_grad = torch.empty( + flat_param._padded_unsharded_size, # type: ignore[attr-defined] + device=self.device, + dtype=sharded_grad.dtype, + ) + dist.all_gather_single(padded_unsharded_grad, sharded_grad, self.process_group) + unsharded_size = self.flat_param._unpadded_unsharded_size + flat_param.grad = padded_unsharded_grad[: unsharded_size.numel()].view( + unsharded_size + ) + self._use_unsharded_grad_views() + + def reshard_grad(self): + if self._use_orig_params: + self._use_sharded_grad_views() + if not self.uses_sharded_strategy: + return + self.flat_param.grad = self.flat_param._saved_grad_shard # type: ignore[attr-defined] + delattr(self.flat_param, "_saved_grad_shard") + + def prepare_gradient_for_backward(self): + """ + Prepare the gradient for the backward computation. + + This is done by saving and clearing any existing sharded gradient + in ``.grad`` to enable computing a new unsharded gradient. + """ + _p_assert( + self._training_state + in (HandleTrainingState.BACKWARD_PRE, HandleTrainingState.IDLE), + "Expects to be in `BACKWARD_PRE` or `IDLE` (if prefetching)", + ) + flat_param = self.flat_param + if flat_param.grad is not None and ( + flat_param.grad.size() != flat_param._unpadded_unsharded_size + or flat_param.grad.device != flat_param.device # grad on CPU + ): + self._check_on_compute_device(self.flat_param) + grad_offloaded = flat_param.grad.device != self.device + _p_assert( + not grad_offloaded or self._offload_params, + f"Expects the sharded gradient to be on {self.device} " + f"but got {flat_param.grad.device}", + ) + prev_iter_synced_gradients = ( + flat_param.grad.size() == flat_param._local_shard.size() # type: ignore[attr-defined] + ) + if prev_iter_synced_gradients: + # TODO (awgu): Gradient accumulation outside `no_sync()` + # does not work with CPU offloading. The issue should be + # that, in the post-backward hook, we cannot do an addition + # between a CPU tensor (the existing sharded gradient) and + # a GPU tensor (the new sharded gradient). + if not grad_offloaded: + flat_param._saved_grad_shard = flat_param.grad.data # type: ignore[attr-defined] + sharded_grad = flat_param._saved_grad_shard # type: ignore[attr-defined] + else: + _p_assert( + hasattr(flat_param, "_cpu_grad"), + "`_cpu_grad` should be defined if the gradient is on CPU", + ) + sharded_grad = flat_param._cpu_grad # type: ignore[attr-defined] + # If user specified to keep the gradient in low precision, then + # the gradient may still be of the low precision dtype if the + # user did not set the gradient to `None` after the previous + # backward, in which case FSDP should cast back to the full + # precision dtype so that FSDP can accumulate in that dtype in + # the post-backward hook and assign to `.grad` in that dtype in + # the post-backward callback. + local_shard_dtype = flat_param._local_shard.dtype # type: ignore[attr-defined] + if ( + self._keep_low_precision_grads + and sharded_grad.dtype != local_shard_dtype + ): + sharded_grad.data = sharded_grad.to(local_shard_dtype) + else: + padded_unsharded_size = flat_param._padded_unsharded_size # type: ignore[attr-defined] + _p_assert( + flat_param.grad.size() == padded_unsharded_size, + "Expects `.grad` to be the unsharded gradient in " + f"`no_sync()` with size {padded_unsharded_size} " + f"but got size {flat_param.grad.size()}", + ) + flat_param.grad = None + + def prepare_gradient_for_optim(self): + """Prepare the gradient for optimizer computation by moving the sharded gradient to the ``.grad`` attribute.""" + + def cast_grad_to_param_dtype_if_needed(flat_param): + # TODO (rohan-varma): test for full precision with keep_low_precision_grads + if not self._force_full_precision and self._keep_low_precision_grads: + _p_assert(flat_param.grad is not None, "Unexpected None grad!") + if flat_param.grad.dtype != self._fwd_bwd_param_dtype: + flat_param.grad.data = flat_param.grad.to(self._fwd_bwd_param_dtype) + if self._use_orig_params: + self._use_sharded_grad_views() + + flat_param = self.flat_param + # TODO (awgu): We should replace these conditional checks to encode + # the logical intention more directly. + if hasattr(flat_param, "_cpu_grad"): + # NOTE: This branch includes `NO_SHARD`. + self._check_sharded(flat_param) + self._check_on_cpu(flat_param) + flat_param.grad = flat_param._cpu_grad # type: ignore[attr-defined] + cast_grad_to_param_dtype_if_needed(flat_param) + elif hasattr(flat_param, "_saved_grad_shard"): + self._check_sharded(flat_param) + self._check_on_compute_device(flat_param) + if flat_param._saved_grad_shard is not None: + self._check_on_compute_device(flat_param._saved_grad_shard) # type: ignore[attr-defined] + # If no sharded gradient was computed this iteration, then there is + # no need to forward `_saved_grad_shard` to `grad` + if flat_param._post_backward_called: # type: ignore[attr-defined] + flat_param.grad = flat_param._saved_grad_shard # type: ignore[attr-defined] + if flat_param.grad is not None: + cast_grad_to_param_dtype_if_needed(flat_param) + else: + _p_assert( + not self.uses_sharded_strategy or not flat_param._post_backward_called, # type: ignore[attr-defined] + "All sharded parameters that received a gradient in the " + "post-backward should use `_saved_grad_shard`", + ) + # Delete `_saved_grad_shard` since its existence indicates a previous + # gradient to accumulate with in the post-backward hook + if hasattr(flat_param, "_saved_grad_shard"): + delattr(flat_param, "_saved_grad_shard") + + @contextlib.contextmanager + def to_cpu(self): + """ + Move the unpadded unsharded flat parameter to CPU while in the context and moves it back to the previous device upon exit. + + For now, this assumes the ``FlatParameter`` is the unpadded unsharded flat parameter + since (1) there is no reason to include the padding in the copy and (2) + there is no use case for the sharded flat parameter. + + Precondition: ``self.flat_param`` 's data is the unpadded unsharded + flat parameter on the compute device, and the handle uses a sharded + strategy. + Postcondition: Same as the precondition. + """ + self._check_sharded_strategy() + _p_assert( + self.flat_param.size() == self.flat_param._unpadded_unsharded_size, + f"Expects size {self.flat_param._unpadded_unsharded_size} but got {self.flat_param.size()}", + ) + self._check_on_compute_device(self.flat_param) + # Check that the unpadded unsharded flat parameter is a view into the + # padded unsharded flat parameter as expected + # NOTE: This check is not strictly needed for correctness but is a + # useful sanity check since the tensor should only be used internally. + _p_assert( + _same_storage(self.flat_param, self._get_padded_unsharded_flat_param()), + "Expects the unpadded parameter to be a view into the padded parameter", + ) + cpu_offload_ctx = ( + contextlib.nullcontext() + if self.flat_param.device.type == "cpu" + else self._offload_to_cpu() + ) + with cpu_offload_ctx: + yield + + @contextlib.contextmanager + def _offload_to_cpu(self): + self.flat_param_to(torch.device("cpu")) + self._free_unsharded_flat_param() + try: + yield + finally: + self._restore_unsharded_flat_param_from_cpu() + + def _restore_unsharded_flat_param_from_cpu(self): + _p_assert( + self.flat_param.size() == self.flat_param._unpadded_unsharded_size, + f"Expects size {self.flat_param._unpadded_unsharded_size} but got {self.flat_param.size()}", + ) + padded_unsharded_flat_param = self._alloc_padded_unsharded_flat_param() + # Copy from CPU to the compute device + padded_unsharded_flat_param[: self.flat_param.numel()].copy_(self.flat_param) + self._use_unsharded_flat_param(padded_unsharded_flat_param) + + def reshard(self, free_unsharded_flat_param: bool): + """ + Run the reshard logic. + + This includes freeing the unsharded flat + parameter if ``free_unsharded_flat_param`` and switching to using the + sharded flat parameter. Note that this also implicitly offloads + the sharded flat parameter (if CPU offload is enabled) by pointing + it to the ``_local_shard`` attribute which resides on CPU. + """ + # Switch to the sharded `FlatParameter` before freeing to prevent + # "use-after-free"-type bugs with external profiling tools, where for + # `use_orig_params=True`, the `param` does not point to valid memory + # when setting `param.data = ...` in `_use_sharded_views()`. + self._use_sharded_flat_param() + if free_unsharded_flat_param: + self._free_unsharded_flat_param() + + def post_reshard(self): + """ + Run the post-reshard logic. + + This includes freeing any memory that + can now be freed given that the ``FlatParameter`` points to the full + precision sharded flat parameter. + + Precondition: ``self.flat_param`` 's data points to the full precision + sharded flat parameter. + """ + # For `NO_SHARD`, `_mp_shard` is not freed in the post-unshard since it + # is also the low precision *unsharded* flat parameter. Hence, we delay + # the free until the reshard. + if ( + self._uses_param_mixed_precision + and not self.uses_sharded_strategy + and not self._force_full_precision # did not use the low precision shard + ): + self._free_low_precision_sharded_param() + + def _free_unsharded_flat_param(self): + """ + Free the padded unsharded flat parameter. We allow this + function to be called even when storage is not allocated + + The tensor to free depends + on the calling context since the unshard may have forced full + precision, in which case a different tensor is used. + """ + self._check_sharded_strategy() + unsharded_flat_param = self._get_padded_unsharded_flat_param() + self._check_on_compute_device(unsharded_flat_param) + # Do not free the memory until all ops in the current stream finish + _no_dispatch_record_stream( + unsharded_flat_param, self._device_handle.current_stream() + ) + _free_storage(unsharded_flat_param) + + def _use_sharded_flat_param(self) -> None: + """Switches to using the sharded flat parameter.""" + flat_param = self.flat_param + if self._use_orig_params: + in_forward = self._training_state == HandleTrainingState.FORWARD + skip_use_sharded_views = ( + torch.is_grad_enabled() + and in_forward + and self._sharding_strategy + in NO_RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES + ) + # Only incur the extra `.data` call if needed + if skip_use_sharded_views: + unsharded_flat_param = flat_param.data + if self._offload_params: + device = flat_param._local_shard.device # type: ignore[attr-defined] + _p_assert( + device == torch.device("cpu"), + f"Expects the local shard to be on CPU but got {device}", + ) + flat_param.data = flat_param._local_shard # type: ignore[attr-defined] + if self._use_orig_params: + if skip_use_sharded_views: # type: ignore[possibly-undefined] + self._unsharded_flat_param_for_skipped_views = unsharded_flat_param # type: ignore[possibly-undefined] + else: + self._use_sharded_views() + # For the post-forward reshard, we may try to use sharded gradient + # views (or unsharded gradient views if a gradient was accumulated + # in `no_sync()`), but for the post-backward reshard, we delay the + # call to after the reduce-scatter. + if ( + in_forward # type: ignore[possibly-undefined] + # Skip using gradient views if skipped using sharded views + # since exposing unsharded parameters with sharded gradients + # may be confusing to the user + and not self._skipped_use_sharded_views + ): + # TODO: Change `_unpadded_unsharded_size` if we change the + # gradient to be computed directly with padding. + accumulated_grad_in_no_sync = ( + flat_param.grad is not None + and self.uses_sharded_strategy + and flat_param.grad.shape == flat_param._unpadded_unsharded_size + ) + if accumulated_grad_in_no_sync: + self._use_unsharded_grad_views() + else: + self._use_sharded_grad_views() + + ######### + # VIEWS # + ######### + @no_type_check + def _get_unflat_views_unaligned( + self, + tensor: torch.Tensor | None = None, + ) -> Iterator[Tensor]: + """ + Return unflattened ``Tensor`` views into ``tensor``. + + If `tensor`` is ``None``, ``flat_param`` is used. The unflattening is based + on ``flat_param`` 's metadata. + + Examples for ``tensor`` include ``flat_param.grad`` or unsharded + tensor optimizer state. + """ + flat_param = self.flat_param + if tensor is None: + tensor = flat_param + views = ( + _ext_post_unflatten_transform( + subtensor.view(shape) + if contiguous + else subtensor.as_strided(shape, stride), + param_extension, + self._fsdp_extension, + ) + for (subtensor, shape, stride, contiguous, param_extension) in zip( + torch.split(tensor, flat_param._numels, dim=0), + flat_param._shapes, + flat_param._strides, + flat_param._contiguities, + flat_param._param_extensions, + ) + ) + return views + + @no_type_check + def _get_unflat_views_aligned( + self, + tensor: Tensor | None = None, + ) -> list[Tensor]: + """ + Return unflattened ``Tensor`` views into ``tensor`` with handling for padding. + + This method has the same contract as :meth:`_get_unflat_views_unaligned` + except it checks for ``None`` placeholders representing padding for + alignment, which may incur slightly more CPU overhead. + """ + flat_param = self.flat_param + if tensor is None: + tensor = flat_param + splits: list[Tensor] = torch.split( + tensor, flat_param._numels_with_padding, dim=0 + ) + idx = 0 + views: list[Tensor] = [] + for split, is_padding in zip(splits, flat_param._is_padding_mask): + if is_padding: + continue + views.append( + _ext_post_unflatten_transform( + split.view(flat_param._shapes[idx]) + if flat_param._contiguities[idx] + else split.as_strided( + flat_param._shapes[idx], flat_param._strides[idx] + ), + flat_param._param_extensions[idx], + self._fsdp_extension, + ) + ) + idx += 1 + return views + + @no_type_check + @torch.enable_grad() + def _use_unsharded_views(self, as_params: bool) -> None: + """ + Unflatten the unsharded flat parameter by setting the original parameter variables to be views into it. + + Args: + as_params (bool): If ``True``, then registers the original + parameters as ``nn.Parameter`` s; if ``False``, then registers + the original parameters only as ``Tensor`` s. ``False`` should + be used during forward/backward computation and when hiding the + original parameters from :meth:`nn.Module.named_parameters`. + + Note: + when prefetching for next forward, current forward may be + annotated with `@torch.no_grad()` + `@torch.enable_grad()` ensures non-empty `view.grad_fn` + otherwise `_post_backward_hook` will not get called + """ + flat_param = self.flat_param + self._check_unsharded(flat_param) + views = self._get_unflat_views() + from torch.distributed.tensor import DTensor + + for i, (view, (param_name, module, _)) in enumerate( + zip(views, flat_param._param_infos) + ): + if self._use_orig_params and as_params: + if type(view) is DTensor: + # A `DTensor` `view` is not compatible with assigning + # `param.data = view`, so we cannot preserve the parameter + # variable. + self._setattr_param( + module, + param_name, + nn.Parameter(view, requires_grad=flat_param.requires_grad), + ) + continue + param = self.flat_param._params[i] + self._setattr_param(module, param_name, param) + param.data = view + elif as_params: + self._setattr_param( + module, + param_name, + nn.Parameter(view, requires_grad=flat_param.requires_grad), + ) + else: # `as_params=False` + param_var: Tensor = view + if self._use_orig_params: + if self._training_state == HandleTrainingState.FORWARD: + # Save the `Tensor` for the pre-backward + self.flat_param._tensors[i] = view # save for pre-backward + elif self._training_state == HandleTrainingState.BACKWARD_PRE: + # Use the saved `Tensor` variable from the forward to + # preserve the autograd graph so that the post-backward + # hook fires (e.g. for reentrant AC) + tensor = self.flat_param._tensors[i] + tensor.data = view + param_var = tensor + self._setattr_tensor(module, param_name, param_var) + if ( + self._use_orig_params + and self._training_state == HandleTrainingState.FORWARD + ): + module._parameters[param_name] = param_var + for i, ( + param_name, + module, + _, + prim_param_name, + prim_module, + _, + ) in enumerate(self.flat_param._shared_param_infos): + prim_param: Tensor | nn.Parameter = getattr(prim_module, prim_param_name) + _p_assert( + not as_params or isinstance(prim_param, nn.Parameter), + f"as_params={as_params} type(prim_param)={type(prim_param)}", + ) + if self._use_orig_params and as_params: + shared_param = self.flat_param._shared_params[i] + self._setattr_param(module, param_name, shared_param) + shared_param.data = prim_param + elif as_params: + self._setattr_param(module, param_name, prim_param) + else: + self._setattr_tensor(module, param_name, prim_param) + if ( + self._use_orig_params + and self._training_state == HandleTrainingState.FORWARD + ): + module._parameters[param_name] = prim_param + + @no_type_check + def _use_unsharded_grad_views(self) -> None: + """ + Unflatten the unsharded flat parameter's gradient. + + The original parameter variables' gradients are set to be views into + the unsharded flat parameter's gradient. + """ + # Expects the gradient to be in `flat_param.grad` + if self.flat_param.grad is None: + for param in chain(self.flat_param._params, self.flat_param._shared_params): + param.grad = None + return + self._check_unsharded(self.flat_param.grad) + views = self._get_unflat_views(self.flat_param.grad) + for i, (view, (param_name, module, _)) in enumerate( + zip(views, self.flat_param._param_infos) + ): + _p_assert( + hasattr(module, param_name), + f"{self.flat_param._fqns[i]} is missing", + ) + param = getattr(module, param_name) + if ( + param.shape != view.shape + or param.dtype != view.dtype + or param.device != view.device + ): + # NOTE: This is a hack using `.data` to side step the check + # that parameter/gradient sizes/dtypes/devices match. From + # calling `reshard()`, `param` has the sharded size, has the + # full precision dtype, and if CPU offloading is enabled, is on + # CPU. Thus, one or more of the following cases can hold when + # in `no_sync()`, where `view` is the original parameter's + # gradient: + # 1. `view` can have the unsharded size. + # 2. `view` can have the parameter low precision dtype. + # 3. `view` can be on GPU. + if param.grad is None: + param.grad = torch.empty_like(param) + param.grad.data = view + else: + param.grad = view + for ( + param_name, + module, + module_name, + prim_param_name, + prim_module, + _, + ) in self.flat_param._shared_param_infos: + _p_assert( + hasattr(module, param_name), + f"{module_name + '.' + param_name if module_name else param_name} is missing", + ) + param = getattr(module, param_name) + prim_param = getattr(prim_module, prim_param_name) + if ( + param.shape != prim_param.grad.shape + or param.dtype != prim_param.grad.dtype + or param.device != prim_param.grad.device + ): + # NOTE: This is the same hack to use `.data` to side step the + # size check. + if param.grad is None: + param.grad = torch.empty_like(param) + param.grad.data = prim_param.grad + else: + param.grad = prim_param.grad + + @contextlib.contextmanager + def unflatten_as_params(self) -> Generator: + """ + Unflatten the original parameters. + + The function assumes that the flat parameter is unsharded. When in the context, + unflattens the original parameters as ``nn.Parameter`` views into the + flat parameter, and after the context, restores the original parameters + as ``Tensor`` views into the flat parameter. + """ + self._use_unsharded_views(as_params=True) + try: + yield + finally: + self._use_unsharded_views(as_params=False) + + @no_type_check + @torch.no_grad() + def _use_sharded_views(self) -> None: + """ + Set the original parameter variables' data to be flattened views into the sharded flat parameter. + + The views are kept as flattened to simplify the case where a parameter + is sharded across ranks. Parameters whose data is not present in the + sharded flat parameter have their data set to a size-0 empty tensor. We + do not delete them to ensure to preserve expected behaviors like model + printability. Parameters whose data is present must preserve their + variables to be passable to an optimizer. + """ + self._unsharded_flat_param_for_skipped_views = None + if not self.uses_sharded_strategy: + # For `NO_SHARD`, use the *unflattened* unsharded views since we + # have the unsharded parameter + self._use_unsharded_views(as_params=True) + return + flat_param = self.flat_param + self._check_sharded(flat_param) + # Construct once and reuse for all parameters not in the local shard + size_0_empty_tensor = torch.empty( + 0, + dtype=self.flat_param.dtype, # in case `flat_param` changed dtype + device=self.flat_param.device, + requires_grad=False, + ) + for param, shard_param_info, (param_name, module, _) in zip( + flat_param._params, flat_param._shard_param_infos, flat_param._param_infos + ): + self._setattr_param(module, param_name, param) + if not shard_param_info.in_shard: + # Allow the original data to be freed via garbage collection + param.data = size_0_empty_tensor + else: + offset = shard_param_info.offset_in_shard + numel_in_shard = shard_param_info.numel_in_shard + param.data = flat_param[offset : offset + numel_in_shard] + if self.flat_param._shared_params is None: + raise AssertionError("Expected _shared_params to be not None") + for param, (param_name, module, _, prim_param_name, prim_module, _) in zip( + self.flat_param._shared_params, self.flat_param._shared_param_infos + ): + self._setattr_param(module, param_name, param) + prim_param = getattr(prim_module, prim_param_name) + param.data = prim_param # could be both empty and non-empty + if self._training_state == HandleTrainingState.BACKWARD_POST: + # Clear the saved `Tensor`s since they are unneeded now + for i in range(len(self.flat_param._tensors)): + self.flat_param._tensors[i] = None + + @no_type_check + @torch.no_grad() + def _use_sharded_grad_views(self) -> None: + """ + Set the original parameter variables' gradients to be flattened views into the sharded flat parameter's gradient. + + This is a no-op if there is no gradient. + + Parameters whose data is not present in the sharded flat parameter and + parameters with ``requires_grad=False`` have their gradients set to + ``None``. Since the gradient variables do not need to be preserved, + this method does not manipulate existing ``Tensor`` data directly and + creates new ``Tensor`` variables instead. + """ + flat_param = self.flat_param + self._check_sharded(flat_param) + grad = self.sharded_grad + if grad is None: + for param in chain(flat_param._params, flat_param._shared_params): + param.grad = None + return + self._check_sharded(grad) + for param, shard_param_info, is_grad_none in zip( + flat_param._params, + flat_param._shard_param_infos, + flat_param._is_grad_none_mask, + ): + if not shard_param_info.in_shard: + param.grad = None + else: + numel_in_shard = shard_param_info.numel_in_shard + if param.requires_grad and not is_grad_none: + offset = shard_param_info.offset_in_shard + if self._keep_low_precision_grads or param.dtype != grad.dtype: + # NOTE: This is a hack using `.data` to side step the + # check that parameter/gradient dtypes match. Here, + # `param` has full precision; `grad` has low precision. + if param.grad is None: + # `.grad` must have the same shape as `param` + param.grad = torch.empty_like(param) + param.grad.data = grad[ + offset : offset + numel_in_shard + ].reshape(param.shape) + else: + param.grad = grad[offset : offset + numel_in_shard].reshape( + param.shape + ) + else: + param.grad = None + if flat_param._shared_params is None: + raise AssertionError("Expected _shared_params to be not None") + for param, (_, _, _, prim_param_name, prim_module, _) in zip( + flat_param._shared_params, flat_param._shared_param_infos + ): + in_sharded_flat_param = hasattr(prim_module, prim_param_name) + if in_sharded_flat_param and param.requires_grad: + prim_param = getattr(prim_module, prim_param_name) + param.grad = prim_param.grad # share the same reference + else: + param.grad = None + + @no_type_check + @torch.no_grad() + def _writeback_orig_params(self) -> bool: + """ + Write back any parameters that changed storage to the handle's ``FlatParameter``. + + Iterates over the original parameters and writes back any parameters + that changed storages (due to a non-inplace operator) to the handle's + ``FlatParameter``. This method preserves the ``FlatParameter` 's + device even if an original parameter's device changes. + + Raises: + RuntimeError: If an original parameter or gradient changes storages + but no longer has the expected flattened shape. + Returns: ``True`` if some writeback happened, and ``False`` otherwise. + """ + if ( + self.uses_sharded_strategy + and not self.is_sharded(self.flat_param) + and not self._skipped_use_sharded_views + ): + # For `NO_SHARD`, we may still need to writeback + return False + flat_param = self.flat_param + wroteback = False + if self._skipped_use_sharded_views and self.uses_sharded_strategy: + # NOTE: We must use the unsharded flat parameter from which the + # unsharded views were computed, not the one from the current + # calling context (`_get_padded_unsharded_flat_param()`) since that + # may be different (e.g. the model changed from train to eval). + flat_param_tensor = self._unsharded_flat_param_for_skipped_views + _p_assert( + _data_ptr_allocated(flat_param_tensor), + "If skipped using sharded views, the unsharded flat parameter " + "should be allocated", + ) + else: + flat_param_tensor = flat_param + # NOTE: Since this method is called in the pre-unshard, which is only + # called during computation in the pre-forward or pre-backward, the + # sharded gradient should be guaranteed to be in `.grad`, not in + # `._saved_grad_shard`. + flat_param_grad = ( + flat_param.grad + if self.uses_sharded_strategy or not self._offload_params + else flat_param._cpu_grad + ) + for i, ( + param, + (in_shard, offset_in_shard, numel_in_shard, _, _), + (param_name, module, _), + ) in enumerate( + zip( + flat_param._params, + flat_param._shard_param_infos, + flat_param._param_infos, + ) + ): + if not in_shard: + continue + if not hasattr(module, param_name): + # Do not writeback if original parameters are deregistered + # (e.g. during model checkpointing) + continue + + # Check for parameter writeback + if self._skipped_use_sharded_views: + param = flat_param._tensors[i] + _p_assert( + param is not None, + f"Expects to have saved tensor for {flat_param._fqns[i]}", + ) + param_changed = getattr(module, param_name) is not param + needs_param_writeback = ( + param_changed # changed parameter variable itself + or not _same_storage(param, flat_param_tensor) + ) + if self._skipped_use_sharded_views and ( + param_changed or needs_param_writeback + ): + raise AssertionError( + "FSDP does not support changing the parameters between " + f"forward and backward for {self._sharding_strategy}" + ) + if param_changed: + # NOTE: The gradient is not preserved after a parameter change. + param = getattr(module, param_name) + flat_param._params[i] = param + if needs_param_writeback: + expected_shape = torch.Size([numel_in_shard]) + src = param if self.uses_sharded_strategy else param.view(-1) + self._writeback_tensor( + src, flat_param, i, expected_shape, offset_in_shard, True + ) + wroteback = True + + # Check for gradient writeback + if self._skipped_use_sharded_views: + # Skip the writeback check because we do not expose gradients + # when we skipped using sharded views + continue + if param.grad is None and flat_param.grad is not None: + expected_shape = torch.Size([numel_in_shard]) + self._writeback_tensor( + None, flat_param.grad, i, expected_shape, offset_in_shard, False + ) + elif param.grad is not None: + # For `NO_SHARD` + CPU offloading, `_cpu_grad` is always in + # memory and owns the gradient storage, so it will never + # require gradient writeback. + if not self.uses_sharded_strategy and self._offload_params: + # Explicitly continue to handle the case of `no_sync()`, + # where `param.grad` is a view into the GPU gradient + # referenced by `flat_param.grad`, while `flat_param_grad` + # is `flat_param._cpu_grad`, which is on CPU + continue + + needs_grad_writeback = flat_param_grad is None or not _same_storage( + param.grad, flat_param_grad + ) + if needs_grad_writeback: + if flat_param_grad is None: + flat_param_grad = torch.zeros_like(flat_param) + expected_shape = torch.Size([numel_in_shard]) + src = ( + param.grad + if self.uses_sharded_strategy + else param.grad.view(-1) + ) + self._writeback_tensor( + src, + flat_param_grad, + i, + expected_shape, + offset_in_shard, + False, + ) + flat_param.grad = flat_param_grad + flat_param_grad = flat_param.grad + + # TODO: If we want to handle shared parameters, we need to re-generate + # the shared parameter data structures in case sharedness changed. + for ( + param_name, + module, + _, + prim_param_name, + prim_module, + _, + ) in flat_param._shared_param_infos: + if getattr(module, param_name) is not getattr(prim_module, prim_param_name): + raise NotImplementedError( + "Changing shared parameters is not supported yet" + ) + return wroteback + + def _writeback_tensor( + self, + src_tensor: Tensor | None, + dst_tensor: Tensor, + tensor_index: int, + expected_shape: torch.Size, + offset: int, + is_param: bool, # else gradient + ) -> None: + """ + Write back ``src_tensor`` to ``dst_tensor`` at offset ``offset``, where ``src_tensor`` should have shape ``expected_shape``. + + ``is_param`` indicates if the tensor is the parameter (if ``True``) or gradient (if + ``False``). If ``src_tensor`` is ``None``, then the effect is zeroing + instead of copying. ``tensor_index`` gives the index of ``src_tensor`` + in the metadata structures. + + Raises: + RuntimeError: If the ``src_tensor`` does not have the expected + shape. + """ + _p_assert( + len(expected_shape) == 1, + f"Expects a 1D expected shape but got {expected_shape}", + ) + if self._debug_level == dist.DebugLevel.INFO: + rank = self.rank if hasattr(self, "rank") else dist.get_rank() + src_shape = src_tensor.shape if src_tensor is not None else None + src_device = src_tensor.device if src_tensor is not None else None + warnings.warn( + f"[Rank {rank}] {'Parameter' if is_param else 'Gradient'} needs " + f"writeback in {self._training_state}\n" + f"expected shape={expected_shape} shape={src_shape} " + f"expected device={dst_tensor.device} device={src_device}", + stacklevel=2, + ) + if src_tensor is not None and src_tensor.shape != expected_shape: + # NOTE: Gradient shape mismatch is not possible in practice since + # the gradient shape is enforced to match that of the parameter and + # we already check for parameter shape mismatch. + raise RuntimeError( + f"Cannot writeback when the {'parameter' if is_param else 'gradient'} " + f"shape changes\nExpects {expected_shape} but got {src_tensor.shape}" + ) + if src_tensor is not None: + dst_tensor[offset : offset + expected_shape.numel()].copy_(src_tensor) + else: + dst_tensor[offset : offset + expected_shape.numel()].zero_() + if self.flat_param._is_grad_none_mask is None: + raise AssertionError("Expected _is_grad_none_mask to be not None") + self.flat_param._is_grad_none_mask[tensor_index] = True + + def _reset_flat_param_grad_info_if_needed(self): + """ + Reset ``flat_param.grad`` if needed. + + When ``use_orig_params=True``: + (1) sets the underlying ``flat_param.grad`` to ``None`` if *all* of the + original parameters' ``.grad`` are ``None``, and + (2) sets ``flat_param.requires_grad=False`` if *none* of the original + parameters require gradient. + For (1), this is targeting ``optim.zero_grad(set_to_none=True)``, in + which case we want to free the gradients as soon after the + ``zero_grad()`` call as possible. + """ + if not self._use_orig_params: + return + flat_param = self.flat_param + if flat_param._params is None: + raise AssertionError("Expected _params to be not None") # mypy + all_grad_none = True + requires_grad = False + for param in flat_param._params: + all_grad_none &= param.grad is None + requires_grad |= param.requires_grad + if all_grad_none: + flat_param.grad = None + # As long as one parameter requires gradient, then the flat parameter + # must require gradient + flat_param.requires_grad = requires_grad + + def _deregister_orig_params(self): + for param_info in self.flat_param._param_infos: + param_name, module, _ = param_info + if hasattr(module, param_name): + delattr(module, param_name) + for param_name, module, _, _, _, _ in self.flat_param._shared_param_infos: + if hasattr(module, param_name): + delattr(module, param_name) + + ########### + # HELPERS # + ########### + def flat_param_to(self, *args, **kwargs): + """Wrap an in-place call to ``.to()`` for ``self.flat_param``.""" + + self.flat_param.data = self.flat_param.to(*args, **kwargs) + if self._use_orig_params: + # Refresh the views because their storage may have changed + if self.is_sharded(self.flat_param): + self._use_sharded_views() + else: + self._use_unsharded_views(as_params=True) + + def _get_modules(self) -> set[nn.Module]: + """Return a :class:`set` of the modules whose parameters are included in this handle's flat parameter.""" + return {pi.module for pi in self.flat_param._param_infos}.union( + {spi.module for spi in self.flat_param._shared_param_infos} + ) + + def is_sharded(self, tensor: Tensor) -> bool: + """ + Return whether ``tensor`` is *currently* sharded. + + For ``NO_SHARD``, we choose to have this always return ``False`` for clarity. + """ + if ( + not hasattr(self.flat_param, "_sharded_size") + or not self.uses_sharded_strategy + ): + # `_sharded_size` is defined iff `handle.shard()` has been called + return False + sharded_size = self.flat_param._sharded_size # type: ignore[attr-defined] + return tensor.size() == sharded_size + + def param_module_names(self) -> Iterator[tuple[str, str]]: + shared_param_infos = [ + ParamInfo(param_name, module, module_name) + for ( + param_name, + module, + module_name, + _, + _, + _, + ) in self.flat_param._shared_param_infos + ] + for param_info in chain(self.flat_param._param_infos, shared_param_infos): + param_name, _, module_name = param_info # type: ignore[misc] + yield (param_name, module_name) + + def shared_param_module_names(self) -> Iterator[tuple[str, str]]: + for param_name, _, module_name in [ + ParamInfo(param_name, module, module_name) + for ( + param_name, + module, + module_name, + _, + _, + _, + ) in self.flat_param._shared_param_infos + ]: + yield (param_name, module_name) + + @property + def _fqns_in_shard(self) -> list[str]: + """Return the FQNs of the parameters present in this rank's shard.""" + fqns_in_shard: list[str] = [] + for fqn, shard_param_info in zip( + self.flat_param._fqns, + self.flat_param._shard_param_infos, # type: ignore[attr-defined] + ): + if shard_param_info.in_shard: + fqns_in_shard.append(fqn) + return fqns_in_shard + + @property + def sharded_grad(self) -> Tensor | None: + """Return the handle's sharded gradient.""" + flat_param = self.flat_param + # Priority for non-`None`: `_cpu_grad` > `_saved_grad_shard` > `grad` + # - CPU offloading: `_cpu_grad` + # - No CPU offloading + sharded strategies: `_saved_grad_shard` + # - No CPU offloading + `NO_SHARD`: `grad` + grad: Tensor | None + if hasattr(flat_param, "_cpu_grad"): + grad = flat_param._cpu_grad # type: ignore[attr-defined] + elif hasattr(flat_param, "_saved_grad_shard"): + # In the post-backward hook, the sharded gradient is still in + # `_saved_grad_shard`. + grad = flat_param._saved_grad_shard # type: ignore[attr-defined] + else: + # If in IDLE or in FORWARD states, then there may be an + # (accumulated) gradient. If accessed in IDLE, then this should + # be due to re-registering the original parameters (e.g. in state + # dict load). + _p_assert( + flat_param.grad is None + or not self.uses_sharded_strategy + or self._training_state + in (HandleTrainingState.FORWARD, HandleTrainingState.IDLE), + "Sharded strategies should use `_cpu_grad` or `_saved_grad_shard` " + "unless in IDLE or FORWARD", + ) + grad = flat_param.grad + return grad + + def _reset_is_grad_none(self) -> None: + """ + Reset ``_is_grad_none_mask`` as needed. + + This method should only be + called in the post-backward after gradient computation, in which case + if a parameter requires gradient, then it will surely receive a + gradient and we may reset its mask entry to ``False``. + """ + if not self._use_orig_params: + return + _p_assert( + self._training_state == HandleTrainingState.BACKWARD_POST, + "Expects to only be called in the post-backward after gradient computation", + ) + flat_param = self.flat_param + if flat_param._params is None: + raise AssertionError("Expected _params to be not None") # mypy + for i, param in enumerate(flat_param._params): # type: ignore[arg-type] + # As long as the parameter requires gradient, it should receive a + # meaningful gradient (even if the gradient happens to be zeros) + if param.requires_grad: + if flat_param._is_grad_none_mask is None: + raise AssertionError( + "Expected _is_grad_none_mask to be not None" + ) # mypy + flat_param._is_grad_none_mask[i] = False + + ####################### + # CHECKS & INVARIANTS # + ####################### + def _check_sharded_strategy(self): + _p_assert(self.uses_sharded_strategy, "Expects sharded strategy") + + def _check_on_compute_device(self, tensor: Tensor): + _p_assert( + tensor.device == self.device, + f"Expects tensor to be on the compute device {self.device}, was on {tensor.device}", + ) + + def _check_on_cpu(self, tensor: Tensor): + _p_assert( + tensor.device == torch.device("cpu"), + f"Expects tensor to be on CPU but got {tensor.device}", + ) + + @staticmethod + def _check_storage_freed(tensor: Tensor): + # Compile does not resize during trace + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + _p_assert( + _same_storage_size(tensor, 0), + "Expects storage to be freed but got storage with size > 0", + ) + + @staticmethod + def _check_storage_allocated(tensor: Tensor): + _p_assert(_storage_size_allocated(tensor), "Expects storage to be allocated") + + def _check_low_precision_shard(self): + _p_assert( + self._uses_param_mixed_precision, + "Not using low precision for parameters", + ) + _p_assert( + getattr(self.flat_param, "_mp_shard", None) is not None, + "Expects `_mp_shard` to exist", + ) + device = self.flat_param._mp_shard.device # type: ignore[attr-defined] + _p_assert( + device == self.device, + f"Expects the low precision shard to be on {self.device} but got {device}", + ) + + def _check_unsharded(self, tensor: Tensor): + msg_prefix = "Expects tensor to be unsharded " + _p_assert(tensor is not None, msg_prefix + "but got `None`") + unsharded_size = self.flat_param._unpadded_unsharded_size + _p_assert( + tensor.size() == unsharded_size, + msg_prefix + f"with size {unsharded_size} but got {tensor.size()}", + ) + + def _check_sharded(self, tensor: Tensor): + msg_prefix = "Expects tensor to be sharded " + _p_assert(tensor is not None, msg_prefix + "but got `None`") + sharded_size = self.flat_param._sharded_size # type: ignore[attr-defined] + _p_assert( + tensor.size() == sharded_size, + msg_prefix + f"with size {sharded_size} but got {tensor.size()}", + ) + + ############## + # PROPERTIES # + ############## + @property + def uses_sharded_strategy(self) -> bool: + return self._sharding_strategy != HandleShardingStrategy.NO_SHARD + + @property + def _uses_param_mixed_precision(self) -> bool: + return self._fwd_bwd_param_dtype != self._orig_param_dtype + + @property + def _uses_reduce_mixed_precision(self) -> bool: + return self._reduce_dtype != self._orig_param_dtype + + @property + def _force_full_precision(self) -> bool: + return ( + self._uses_param_mixed_precision or self._uses_reduce_mixed_precision + ) and ( + self._training_state == HandleTrainingState.SUMMON_FULL_PARAMS + or + # Also disable mixed precision in model eval mode, if configured + (not self._fully_sharded_module.training and self._use_full_prec_in_eval) + ) + + @property + def _skipped_use_sharded_views(self) -> bool: + """ + This property is used for sharding strategies that do not free after forward with ``use_orig_params=True``. + + This returns if this handle is + currently in a state where it has skipped using sharded views, in which + case it can restore view invariants via ``_use_sharded_views()``. + """ + return self._unsharded_flat_param_for_skipped_views is not None + + +# NOTE: These are hacks to bypass `nn.Module.__setattr__` checks. +def _unsafe_setattr_param( + module: nn.Module, param_name: str, param: nn.Parameter +) -> None: + module._parameters[param_name] = param + # This bypasses any overrides in case `module` is an instance of an + # `nn.Module` subclass + super(nn.Module, module).__setattr__(param_name, param) + + +def _unsafe_setattr_tensor(module: nn.Module, param_name: str, tensor: Tensor) -> None: + module._parameters.pop(param_name, None) + # This bypasses any overrides in case `module` is an instance of an + # `nn.Module` subclass + super(nn.Module, module).__setattr__(param_name, tensor) + + +def _safe_setattr_tensor_or_param( + module: nn.Module, param_name: str, tensor_or_param: Tensor | nn.Parameter +): + # Call `delattr()` and `setattr()` to go through `nn.Module` checks + if hasattr(module, param_name): + delattr(module, param_name) + setattr(module, param_name, tensor_or_param) + + +def _convert_to_params( + tensors: list[torch.Tensor | nn.Parameter], +) -> list[nn.Parameter]: + return [t if isinstance(t, nn.Parameter) else nn.Parameter(t) for t in tensors] + + +def _is_truly_contiguous(x: Tensor) -> bool: + # Special case: Pytorch thinks that 1x1 channels_last convolution weights are + # both contiguous and channels_last contiguous at the same time. + # CuDNN does not agree though and refuses to select faster kernels. + # It is the reason of having the extra check here. + return x.stride(-1) == 1 and x.is_contiguous() + + +def _detach_if_needed(param_or_tensor: nn.Parameter | Tensor) -> Tensor: + return ( + param_or_tensor.detach() + if isinstance(param_or_tensor, nn.Parameter) + else param_or_tensor + ) + + +def _get_aligned_numel(unsharded_dtype: torch.dtype): + # NOTE: This alignment constraint comes from TorchInductor. + ALIGNMENT = 16 # bytes + unsharded_dtype_size = _get_dtype_size(unsharded_dtype) + aligned_numel = ALIGNMENT // unsharded_dtype_size + return aligned_numel + + +@functools.lru_cache(8) +def _get_dtype_size(dtype): + return torch.empty((), dtype=dtype).element_size() + + +def _construct_padding_tensor( + padding_numel: int, dtype: torch.dtype, requires_grad: bool, device: torch.device +): + # NOTE: Set the padding value as a magic number for debuggability. The + # value itself should never be used in any user-facing computation. + return ( + torch.ones( + (padding_numel,), dtype=dtype, requires_grad=requires_grad, device=device + ) + * _FLAT_PARAM_PADDING_VALUE + ) + + +# Use `lru_cache(1)` to only log the warning once (assuming the fixed warning +# message is passed in) +@functools.lru_cache(1) +def _warn_skip_writeback_check(log: logging.Logger, warning: str): + logger.warning(warning) + + +# Use `lru_cache(1)` to only log the warning once +@functools.lru_cache(1) +def _warn_use_fake_all_gather(log: logging.Logger, warning: str): + logger.warning(warning) + + +# Use `lru_cache(1)` to only log the warning once +@functools.lru_cache(1) +def _warn_use_fake_reduce(log: logging.Logger, warning: str): + logger.warning(warning) + + +def _same_storage(a, b): + # Params are DTensors in backward + # with SHARD_GRAD_OP + TP + from torch.distributed.tensor import DTensor + + if isinstance(a, DTensor): + a = a._local_tensor + if isinstance(b, DTensor): + b = b._local_tensor + return a.untyped_storage().data_ptr() == b.untyped_storage().data_ptr() + + +def _same_storage_size(a: torch.Tensor, b: int): + return a.untyped_storage().size() // a.element_size() == b + + +def _storage_size_allocated(tensor: Tensor): + storage_size: int = tensor.untyped_storage().size() + return storage_size > 0 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fsdp_extensions.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fsdp_extensions.py new file mode 100644 index 0000000000000000000000000000000000000000..d21e7ac2d059a441da99374a91477703c7562d59 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fsdp_extensions.py @@ -0,0 +1,180 @@ +from abc import ABC, abstractmethod +from typing import Any + +import torch +import torch.distributed as dist +from torch.distributed._shard.sharded_tensor.api import ShardedTensor +from torch.distributed._shard.sharded_tensor.shard import Shard +from torch.distributed.fsdp._shard_utils import ( + _all_gather_dtensor, + _create_chunk_dtensor, + _create_chunk_sharded_tensor, +) +from torch.distributed.tensor import DeviceMesh, DTensor + + +class FSDPExtensions(ABC): + """ + This enables some customizable hooks to enable composability with tensor + parallelism. To activate these hooks, use :func:`_set_fsdp_extensions` to + set a custom :class:`FSDPExtensions` that implements the hooks. + """ + + @abstractmethod + def pre_flatten_transform( + self, + tensor: torch.Tensor, + ) -> tuple[torch.Tensor, Any | None]: + """E.g. converting ``DistributedTensor`` to local tensor.""" + ... + + @abstractmethod + def post_unflatten_transform( + self, + tensor: torch.Tensor, + param_extension: Any, + ) -> torch.Tensor: + """E.g. converting local tensor to ``DistributedTensor``.""" + ... + + @abstractmethod + def chunk_tensor( + self, + tensor: torch.Tensor, + rank: int, + world_size: int, + num_devices_per_node: int, + pg: dist.ProcessGroup, + device: torch.device | None = None, + ) -> torch.Tensor: + """Shards a tensor to chunks and returns the local chunk.""" + ... + + @abstractmethod + def chunk_dtensor( + self, + tensor: torch.Tensor, + rank: int, + device_mesh: DeviceMesh, + ) -> torch.Tensor: + """Shards a tensor/DTensor to DTensor and returns the local DTensor.""" + ... + + @abstractmethod + def pre_load_state_dict_transform( + self, + tensor: torch.Tensor, + ) -> tuple[torch.Tensor, list[Shard]]: + """ + This is to be called before loading a *sharded* model state dict and + should return the tensor and list of shards from which to load data. + """ + ... + + @abstractmethod + def all_gather_dtensor( + self, + tensor: DTensor, + parent_mesh: DeviceMesh | None, + ) -> torch.Tensor: + """ + This is to be called before loading a *sharded* DTensor state dict. + This gathers tensor in FSDP dimension and returns local tensor of + TP DTensor. + """ + ... + + +_extensions: FSDPExtensions | None = None + + +def _set_fsdp_extensions(flattener: FSDPExtensions) -> None: + global _extensions + _extensions = flattener + + +def _ext_pre_flatten_transform( + tensor: torch.Tensor, + fsdp_extension: FSDPExtensions | None = None, +) -> tuple[torch.Tensor, Any | None]: + if fsdp_extension is not None: + new_tensor, param_extension = fsdp_extension.pre_flatten_transform(tensor) + if param_extension is not None: + return new_tensor, param_extension + return tensor, None + + +def _ext_post_unflatten_transform( + tensor: torch.Tensor, + param_extension: Any, + fsdp_extension: FSDPExtensions | None = None, +) -> torch.Tensor: + if fsdp_extension is not None and param_extension is not None: + return fsdp_extension.post_unflatten_transform(tensor, param_extension) + return tensor + + +def _ext_chunk_tensor( + tensor: torch.Tensor, + rank: int, + world_size: int, + num_devices_per_node: int, + pg: dist.ProcessGroup, + fsdp_extension: FSDPExtensions | None = None, +) -> torch.Tensor: + chunk_tensor_fn = ( + fsdp_extension.chunk_tensor + if fsdp_extension is not None + else _create_chunk_sharded_tensor + ) + return chunk_tensor_fn( + tensor, + rank, + world_size, + num_devices_per_node, + pg, + ) + + +def _ext_chunk_dtensor( + tensor: torch.Tensor, + rank: int, + device_mesh: DeviceMesh, + fsdp_extension: FSDPExtensions | None = None, +) -> torch.Tensor: + chunk_dtensor_fn = ( + fsdp_extension.chunk_dtensor + if fsdp_extension is not None + else _create_chunk_dtensor + ) + return chunk_dtensor_fn( + tensor, + rank, + device_mesh, + ) + + +def _ext_pre_load_state_dict_transform( + tensor: torch.Tensor, + fsdp_extension: FSDPExtensions | None = None, +) -> tuple[torch.Tensor, list[Shard]]: + if fsdp_extension is not None: + return fsdp_extension.pre_load_state_dict_transform(tensor) + + if type(tensor) is not ShardedTensor: + raise AssertionError(f"Expected ShardedTensor, got {type(tensor)}") + shards = tensor.local_shards() + return (tensor, shards) + + +def _ext_all_gather_dtensor( + tensor: DTensor, + parent_mesh: DeviceMesh | None, + fsdp_extension: FSDPExtensions | None = None, +) -> torch.Tensor: + all_gather_dtensor_fn = ( + fsdp_extension.all_gather_dtensor + if fsdp_extension is not None + else _all_gather_dtensor + ) + return all_gather_dtensor_fn(tensor, parent_mesh) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f6aab00fc32b7acdb76dc0d656b8841ef8f62955 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__init__.py @@ -0,0 +1,26 @@ +from ._fsdp_api import ( + CPUOffloadPolicy, + DataParallelMeshDims, + MixedPrecisionPolicy, + OffloadPolicy, +) +from ._fully_shard import ( + FSDPModule, + fully_shard, + register_fsdp_forward_method, + share_comm_ctx, + UnshardHandle, +) + + +__all__ = [ + "CPUOffloadPolicy", + "DataParallelMeshDims", + "FSDPModule", + "fully_shard", + "MixedPrecisionPolicy", + "OffloadPolicy", + "register_fsdp_forward_method", + "UnshardHandle", + "share_comm_ctx", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6925ead161a3fd4b9f06c6571b17e26abb5f4295 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bae9b299285deea4fffe263be4b3fc17c70a60ac Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_collectives.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_collectives.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..44dccc8070b47ea85a2acc2e1e4483c35e99edb6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_collectives.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_common.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_common.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b80c401148ee64973d93009d093b5a038270a8dd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_common.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_init.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_init.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1189d4f9f4c4a3ccabd6344ed6c9dda93e5403a8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_init.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3a2ff7f2284ff9dd0dfa9194e544e81059d89216 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param_group.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param_group.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6901e17639dc6b5aabe239baf538e0586b7fc8be Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_param_group.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_state.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_state.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..600aaf94e5e277265010a20d2e5b167130dedae6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fsdp_state.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fully_shard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fully_shard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..14140eb0dfc54fbfcad9eb9fc626aa01077c2ea5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/__pycache__/_fully_shard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_api.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_api.py new file mode 100644 index 0000000000000000000000000000000000000000..d69415f376a9f27807f29aefe82febd0186b9377 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_api.py @@ -0,0 +1,199 @@ +# mypy: allow-untyped-defs +from abc import ABC, abstractmethod +from collections.abc import Sequence +from dataclasses import dataclass + +import torch +import torch.distributed as dist + + +_ReduceOp = dist.ReduceOp | dist.ReduceOp.RedOpType + + +@dataclass(frozen=True) +class MixedPrecisionPolicy: + """ + This configures FSDP's mixed precision. Unlike autocast, this applies mixed + precision at the module level, not op level, which means low-precision + activations are saved for backward and high-to-low-precision casts are + incurred only at module boundaries. + + FSDP works well with module-level mixed precision since it keeps the + high-precision sharded parameters in memory anyway. In other words, FSDP + does not require any extra memory to keep a high-precision copy of the + parameters for the optimizer step. + + Attributes: + param_dtype (Optional[torch.dtype]): This specifies the dtype for + the unsharded parameter and hence the dtype for forward/backward + computation and the parameter all-gather. If this is ``None``, then + the unsharded parameter uses the original dtype. The optimizer step + uses the sharded parameter in the original dtype. (Default: + ``None``) + reduce_dtype (Optional[torch.dtype]): This specifies the dtype for + gradient reduction (i.e. reduce-scatter or all-reduce). If this is + ``None`` but ``param_dtype`` is not ``None``, then the reduction + uses the compute dtype. This can be used to run gradient reduction + in full precision while using low precision for compute. If also + gradient reduction is disabled via :meth:`set_requires_gradient_sync`, + then FSDP will accumulate gradients using ``reduce_dtype``. + (Default: ``None``) + output_dtype (Optional[torch.dtype]): This specifies the dtype for + casting floating-point forward outputs. This can be used to + help implement cases where different modules have different mixed + precision policies. (Default: ``None``) + cast_forward_inputs (bool): This specifies whether FSDP should cast the + forward's floating-point input tensors to ``param_dtype`` or not. + For grouped ``fully_shard([a, b, ...])``, the cast is applied per + module, before each module's forward. + """ + + param_dtype: torch.dtype | None = None + reduce_dtype: torch.dtype | None = None + output_dtype: torch.dtype | None = None + cast_forward_inputs: bool = True + + +class Comm(ABC): + """ + Interface for communication primitives. + A primitive primarily needs to handle 3 tasks, namely: + + 1. How to allocate memory for communication + Depending on the goal, an implementation can choose to: + a. associate each call to a temporary buffer + (best for flexibility and simplicity) + b. reuse a persistent buffer for efficiency reasons + + 2. Where to allocate memory + (e.g. NCCL mem pool or regular cuda caching allocator) + + 3. What to do/call upon the comm is called + (see `AllGather` interface as an example) + """ + + @abstractmethod + def allocate( + self, + size: Sequence[int | torch.SymInt], + *, + dtype: torch.dtype, + device: torch.device, + ) -> torch.Tensor: + """ + This handles the "how to allocate memory" part. + + A default implementation could be simply: + + .. code-block:: python + with self.mem_pool: + torch.empty(...) + + Args: + size (Sequence[Union[int, torch.SymInt]]): size of the tensor buffer + dtype (torch.dtype): dtype of the tensor buffer + device (torch.device): which device to allocate the tensor onto + """ + ... + + +class AllGather(Comm): + """ + Interface for all_gather comm primitive + """ + + @abstractmethod + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + async_op: bool = False, + ) -> dist.Work | None: ... + + +class ReduceScatter(Comm): + """ + Interface for reduce_scatter comm primitive + """ + + @abstractmethod + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + op: _ReduceOp, + async_op: bool = False, + ) -> dist.Work | None: ... + + +@dataclass +class DataParallelMeshDims: + """ + Specifies which dimensions of a full SPMD :class:`DeviceMesh` correspond to + data parallelism when using :func:`fully_shard` whose parameters are already + DTensors on that mesh. + + Attributes: + shard (Optional[Union[str, tuple[str, ...]]]): Mesh dimension name(s) + that FSDP shards parameters on. If a tuple of names, those dims + are flattened into a single shard dimension. At least one of + ``shard`` and ``replicate`` must be set. + replicate (Optional[Union[str, tuple[str, ...]]]): Mesh dimension + name(s) for HSDP or DDP replication. If a tuple of names, those + dims are flattened into a single replicate dimension. + """ + + shard: str | tuple[str, ...] | None = None + replicate: str | tuple[str, ...] | None = None + + def __post_init__(self): + if self.shard is None and self.replicate is None: + raise ValueError( + "At least one of shard or replicate must be set in DataParallelMeshDims" + ) + + @property + def shard_names(self) -> tuple[str, ...]: + if self.shard is None: + return () + if isinstance(self.shard, str): + return (self.shard,) + return tuple(self.shard) + + @property + def replicate_names(self) -> tuple[str, ...]: + if self.replicate is None: + return () + if isinstance(self.replicate, str): + return (self.replicate,) + return tuple(self.replicate) + + +@dataclass +class OffloadPolicy: + """ + This base class represents the policy of no offloading and is only used as + the default value for the ``offload_policy`` arg. + """ + + +@dataclass +class CPUOffloadPolicy(OffloadPolicy): + """ + This offload policy offloads parameters, gradients, and optimizer states to + CPU. Sharded parameters are copied host-to-device before all-gather. The + all-gathered parameters are freed according to ``reshard_after_forward``. + Sharded gradients are copied device-to-host in backward, and the optimizer + step runs on CPU with CPU optimizer states. + + Attributes: + pin_memory (bool): Whether to pin sharded parameter and gradient + memory. Pinning memory allows both more efficient H2D/D2H copies + and for the copies to overlap with compute. However, the pinned + memory cannot be used by other processes. Set this to ``False`` if + you have insufficient CPU memory. (Default: ``True``) + """ + + pin_memory: bool = True diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_collectives.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_collectives.py new file mode 100644 index 0000000000000000000000000000000000000000..7d7da080f31d5f8025816398fdde0746d5e57ac1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_collectives.py @@ -0,0 +1,874 @@ +import math +from collections.abc import Callable, Sequence +from itertools import chain +from typing import Any, cast, Literal, NamedTuple + +import torch +import torch.distributed as dist +import torch.distributed._symmetric_memory as symm_mem +from torch.distributed.device_mesh import _get_device_handle +from torch.distributed.distributed_c10d import ReduceOp +from torch.distributed.fsdp._fully_shard._fsdp_api import AllGather, ReduceScatter +from torch.distributed.tensor import DTensor + +from ._fsdp_api import _ReduceOp +from ._fsdp_common import ( + _get_dim0_padded_size, + _raise_assert_with_print, + _to_dtype_if_needed, +) +from ._fsdp_param import FSDPParam, ShardedState + + +class AllGatherResult(NamedTuple): + all_gather_output: torch.Tensor + all_gather_event: torch.Event | None + all_gather_work: dist.distributed_c10d.Work | None + # For each parameter, the all-gather input dtype for each input + param_all_gather_input_dtypes: list[list[torch.dtype]] + # For each parameter, the all-gather input numel for each input + param_all_gather_input_numels: list[list[int]] + # 1D flattened version of `param_all_gather_input_numels` saved to avoid + # CPU overhead from recomputing + all_gather_input_split_sizes: list[int] + + +lib = torch.library.Library("fsdp", "FRAGMENT") + +lib.define( + """ + all_gather_copy_in( + Tensor[] all_gather_inputs, + Tensor all_gather_output, + SymInt[] inp_split_sizes, + SymInt all_gather_input_numel, + SymInt rank + ) -> (Tensor, Tensor) + """ +) + + +class DefaultAllocMixin: + def allocate( + self, + size: Sequence[int | torch.SymInt], + *, + dtype: torch.dtype, + device: torch.device, + ) -> torch.Tensor: + return torch.empty(*size, dtype=dtype, device=device) + + +class ProcessGroupAllocMixin: + def __init__(self, group: dist.ProcessGroup, *args: Any, **kwargs: Any): + self._group = group + super().__init__(*args, **kwargs) + + def allocate( + self, + size: Sequence[int | torch.SymInt], + *, + dtype: torch.dtype, + device: torch.device, + ) -> torch.Tensor: + backend = self._group._get_backend(device) + if backend.supports_tensor_alloc(device): + size_1d = math.prod(int(s) for s in size) + return backend.allocate_tensor(size_1d, dtype=dtype, device=device) + return torch.empty(*size, dtype=dtype, device=device) + + +class SymmMemAllocMixin: + def __init__( + self, + group: dist.ProcessGroup, + backend: Literal["NCCL"] = "NCCL", + *args: Any, + **kwargs: Any, + ): + self._group = group + symm_mem.set_backend(backend) + # Force initialization of communicator; otherwise, the rendezvous may + # see empty communicator. + # TODO: Remove this, maybe by warning user to perform eager dist init. + # For now, it is okay since it is just a one-time cost at init. + dist.barrier(group=group) + + def allocate( + self, + size: Sequence[int | torch.SymInt], + *, + dtype: torch.dtype, + device: torch.device, + ) -> torch.Tensor: + # Leverage MemPool to reuse the symmetric buffer, avoiding allocation + # and rendezvous overhead + mempool = symm_mem.get_mem_pool(device) + with torch.cuda.use_mem_pool(mempool): + return torch.empty(size, dtype=dtype, device=device) + + +class DefaultAllGather(DefaultAllocMixin, AllGather): + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + async_op: bool = False, + ) -> dist.Work | None: + return dist.all_gather_single( + output_tensor, + input_tensor, + group=group, + async_op=async_op, + ) + + +class ProcessGroupAllocAllGather(ProcessGroupAllocMixin, AllGather): + def __init__(self, group: dist.ProcessGroup) -> None: + super().__init__(group) + + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + async_op: bool = False, + ) -> dist.Work | None: + return dist.all_gather_single( + output_tensor, + input_tensor, + group=group, + async_op=async_op, + ) + + +class SymmMemAllGather(SymmMemAllocMixin, AllGather): + def __init__( + self, + group: dist.ProcessGroup, + backend: Literal["NCCL"] = "NCCL", + ) -> None: + super().__init__(group, backend) + + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + async_op: bool = False, + ) -> dist.Work | None: + # We are doing inplace all-gather, so we need to rendezvous the output tensor only + symm_mem.rendezvous(output_tensor, group=group.group_name) + # Calling regular all-gather would already cause libraries like NCCL to + # use its optimized all-gather implementation for symmetric memory: + # - Copy Engine All-Gather (when zero-CTA policy is enabled) + # - Symmetric Kernel All-Gather (when zero-CTA policy is not enabled) + return dist.all_gather_single( + output_tensor, + input_tensor, + group=group, + async_op=async_op, + ) + + +class DefaultReduceScatter(DefaultAllocMixin, ReduceScatter): + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + op: _ReduceOp, + async_op: bool = False, + ) -> dist.Work | None: + return dist.reduce_scatter_single( + output=output_tensor, + input=input_tensor, + group=group, + op=op, + async_op=async_op, + ) + + +class ProcessGroupAllocReduceScatter(ProcessGroupAllocMixin, ReduceScatter): + def __init__(self, group: dist.ProcessGroup) -> None: + super().__init__(group) + + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + op: _ReduceOp, + async_op: bool = False, + ) -> dist.Work | None: + return dist.reduce_scatter_single( + output=output_tensor, + input=input_tensor, + group=group, + op=op, + async_op=async_op, + ) + + +class SymmMemReduceScatter(SymmMemAllocMixin, ReduceScatter): + def __init__( + self, + group: dist.ProcessGroup, + backend: Literal["NCCL"] = "NCCL", + ) -> None: + super().__init__(group, backend) + + def __call__( + self, + output_tensor: torch.Tensor, + input_tensor: torch.Tensor, + group: dist.ProcessGroup, + op: _ReduceOp, + async_op: bool = False, + ) -> dist.Work | None: + symm_mem.rendezvous(input_tensor, group=group.group_name) + symm_mem.rendezvous(output_tensor, group=group.group_name) + # Calling regular reduce-scatter would already cause libraries like NCCL to + # use its optimized reduce-scatter implementation for symmetric memory + return dist.reduce_scatter_single( + output=output_tensor, + input=input_tensor, + group=group, + op=op, + async_op=async_op, + ) + + +@torch.library.impl(lib, "all_gather_copy_in", "Meta") +def all_gather_copy_in_meta( + all_gather_inputs: list[torch.Tensor], + all_gather_output: torch.Tensor, + inp_split_sizes: list[int], + all_gather_input_numel: int, + rank: int, +) -> tuple[torch.Tensor, torch.Tensor]: + all_gather_input = all_gather_output.narrow( + 0, all_gather_input_numel * rank, all_gather_input_numel + ) + return all_gather_input, all_gather_output + + +@torch.library.impl(lib, "all_gather_copy_in", "CUDA") +@torch.library.impl(lib, "all_gather_copy_in", "XPU") +@torch.library.impl(lib, "all_gather_copy_in", "HPU") +@torch.library.impl(lib, "all_gather_copy_in", "CPU") +@torch.library.impl(lib, "all_gather_copy_in", "MTIA") +@torch.library.impl(lib, "all_gather_copy_in", "PrivateUse1") +def all_gather_copy_in_cuda( + all_gather_inputs: list[torch.Tensor], + all_gather_output: torch.Tensor, + inp_split_sizes: list[int], + all_gather_input_numel: int, + rank: int, +) -> tuple[torch.Tensor, torch.Tensor]: + all_gather_input = all_gather_output.narrow( + 0, all_gather_input_numel * rank, all_gather_input_numel + ) + foreach_copy_dsts = torch.split(all_gather_input, inp_split_sizes) + with torch.no_grad(): + torch._foreach_copy_(foreach_copy_dsts, all_gather_inputs) + return all_gather_input, all_gather_output + + +lib.define( + "split_with_sizes_copy(Tensor all_gather_output, SymInt[] all_gather_input_split_sizes, int dim=0, *, Tensor(a!)[] out) -> ()" +) + + +@torch.library.impl(lib, "split_with_sizes_copy", "Meta") +@torch.library.impl(lib, "split_with_sizes_copy", "CUDA") +@torch.library.impl(lib, "split_with_sizes_copy", "XPU") +@torch.library.impl(lib, "split_with_sizes_copy", "HPU") +@torch.library.impl(lib, "split_with_sizes_copy", "CPU") +@torch.library.impl(lib, "split_with_sizes_copy", "MTIA") +@torch.library.impl(lib, "split_with_sizes_copy", "PrivateUse1") +def split_with_sizes_copy( + all_gather_output: torch.Tensor, + all_gather_input_split_sizes: list[int], + dim: int = 0, + *, + out: list[torch.Tensor], +) -> None: + torch.split_with_sizes_copy( + all_gather_output, all_gather_input_split_sizes, dim=dim, out=out + ) + + +lib.define( + "chunk_cat(Tensor[] tensors, int dim, int num_chunks, *, Tensor(a!) out) -> ()" +) + + +@torch.library.impl(lib, "chunk_cat", "Meta") +@torch.library.impl(lib, "chunk_cat", "CUDA") +@torch.library.impl(lib, "chunk_cat", "XPU") +@torch.library.impl(lib, "chunk_cat", "HPU") +@torch.library.impl(lib, "chunk_cat", "CPU") +@torch.library.impl(lib, "chunk_cat", "MTIA") +@torch.library.impl(lib, "chunk_cat", "PrivateUse1") +def chunk_cat( + tensors: list[torch.Tensor], + dim: int, + num_chunks: int, + out: torch.Tensor, +) -> None: + torch._chunk_cat(tensors, dim, num_chunks, out=out) + + +@torch.no_grad() +def foreach_all_gather( + fsdp_params: list[FSDPParam], + group: dist.ProcessGroup, + async_op: bool, + all_gather_copy_in_stream: torch.Stream, + all_gather_stream: torch.Stream, + device: torch.device, + all_gather_comm: AllGather, +) -> AllGatherResult | None: + world_size, rank = group.size(), group.rank() + device_handle = _get_device_handle(device.type) + with device_handle.stream(all_gather_copy_in_stream): + param_all_gather_inputs = _get_param_all_gather_inputs(fsdp_params) + ( + param_all_gather_input_dtypes, + param_all_gather_input_numels, + dtype, + ) = _get_all_gather_input_metadatas(param_all_gather_inputs) + if dtype == torch.uint8: + all_gather_inputs = [ + t.view(torch.uint8) for ts in param_all_gather_inputs for t in ts + ] + else: + all_gather_inputs = [*chain.from_iterable(param_all_gather_inputs)] + inp_split_sizes = [t.numel() for t in all_gather_inputs] + all_gather_input_numel = sum(inp_split_sizes) + all_gather_output = all_gather_comm.allocate( + (all_gather_input_numel * world_size,), dtype=dtype, device=device + ) + all_gather_input, all_gather_output = torch.ops.fsdp.all_gather_copy_in( + all_gather_inputs, + all_gather_output, + inp_split_sizes, + all_gather_input_numel, + rank, + ) + del param_all_gather_inputs + all_gather_stream.wait_stream(all_gather_copy_in_stream) + with device_handle.stream(all_gather_stream): + all_gather_work = all_gather_comm( + output_tensor=all_gather_output, + input_tensor=all_gather_input, + group=group, + async_op=async_op, + ) + all_gather_event = all_gather_stream.record_event() + return AllGatherResult( + all_gather_output, + all_gather_event, + all_gather_work, + param_all_gather_input_dtypes, + param_all_gather_input_numels, + inp_split_sizes, + ) + + +@torch.no_grad() +def _get_param_all_gather_inputs( + fsdp_params: list[FSDPParam], +) -> list[list[torch.Tensor]]: + # Intentionally try to run a fast-path that bypasses abstractions for the + # common FSDP case of bf16/fp32 mixed precision in order to use foreach + # copy for lower CPU overhead and more efficient copying in eager + def use_foreach_copy(fsdp_param: FSDPParam) -> bool: + return ( + fsdp_param.param_dtype is not None + and not fsdp_param.offload_to_cpu + and not hasattr(fsdp_param._sharded_local_tensor, "fsdp_pre_all_gather") + ) + + param_all_gather_inputs: list[list[torch.Tensor]] = [[] for _ in fsdp_params] + foreach_copy_indices: list[int] = [] + foreach_copy_inputs: list[torch.Tensor] = [] + foreach_copy_input_numels: list[int] = [] + + # 1st pass: for foreach-copy parameters, get inputs and metadata for the + # foreach copy, and for the others, actually get their all-gather inputs + for i, fsdp_param in enumerate(fsdp_params): + if use_foreach_copy(fsdp_param): + foreach_copy_indices.append(i) + all_gather_input = ( + fsdp_param._sharded_param_data + if fsdp_param.sharded_state == ShardedState.SHARDED + else cast(torch.Tensor, fsdp_param._sharded_post_forward_param_data) + ) + foreach_copy_inputs.append(all_gather_input) + foreach_copy_input_numels.append(all_gather_input.numel()) + else: + param_all_gather_inputs[i] = fsdp_param.all_gather_inputs + + # 2nd pass: use foreach copy to compute the remaining all-gather inputs + if foreach_copy_inputs: + fsdp_param_0 = fsdp_params[foreach_copy_indices[0]] + param_dtype, device = fsdp_param_0.param_dtype, fsdp_param_0.device + flat_foreach_copy_input = torch.empty( + (sum(foreach_copy_input_numels),), device=device, dtype=param_dtype + ) + splits = torch.split(flat_foreach_copy_input, foreach_copy_input_numels) + torch._foreach_copy_(splits, foreach_copy_inputs) + for i, split in zip(foreach_copy_indices, splits): + param_all_gather_inputs[i] = [split] + + return param_all_gather_inputs + + +@torch.no_grad() +def foreach_all_gather_copy_out( + all_gather_result: AllGatherResult, + fsdp_params: list[FSDPParam], + group: dist.ProcessGroup, +) -> None: + ( + all_gather_output, + all_gather_event, + all_gather_work, + param_all_gather_input_dtypes, + param_all_gather_input_numels, + all_gather_input_split_sizes, + ) = all_gather_result + _dtype, device = all_gather_output.dtype, all_gather_output.device + device_handle = _get_device_handle(device.type) + if all_gather_event is not None: # sync op + device_handle.current_stream().wait_event(all_gather_event) + if isinstance(all_gather_work, dist.distributed_c10d.Work): # async op + all_gather_work.wait() + world_size, device = group.size(), all_gather_output.device + + split_with_sizes_out: list[torch.Tensor] = [] + shard_i_copy_infos: list[tuple[FSDPParam, list[torch.Tensor]]] = [] + for all_gather_input_numels, all_gather_input_dtypes, fsdp_param in zip( + param_all_gather_input_numels, param_all_gather_input_dtypes, fsdp_params + ): + fsdp_param.init_all_gather_outputs( + all_gather_input_numels, + all_gather_input_dtypes, + world_size, + device, + ) + fsdp_param.alloc_all_gather_outputs() + param_all_gather_outputs = fsdp_param.all_gather_outputs + if fsdp_param.fsdp_placement.dim != 0: + # Copy to a temporary and then chunk-cat into the final all-gather + # output tensors + param_all_gather_outputs = [ + torch.empty_like(t) for t in param_all_gather_outputs + ] + shard_i_copy_infos.append((fsdp_param, param_all_gather_outputs)) + split_with_sizes_out.extend(param_all_gather_outputs) + + all_gather_output = all_gather_output.view(world_size, -1) + if all_gather_output.dtype == torch.uint8: + out = [t.view(world_size, -1).view(torch.uint8) for t in split_with_sizes_out] + else: + out = [t.view(world_size, -1) for t in split_with_sizes_out] + + # only avoid VC bump if we are not in inference mode + non_inference_outs = [o for o in out if not o.is_inference()] + + if len(non_inference_outs) > 0: + with torch.autograd._unsafe_preserve_version_counter(tuple(non_inference_outs)): + torch.ops.fsdp.split_with_sizes_copy( + all_gather_output, all_gather_input_split_sizes, dim=1, out=out + ) + else: + torch.ops.fsdp.split_with_sizes_copy( + all_gather_output, all_gather_input_split_sizes, dim=1, out=out + ) + + for fsdp_param, param_all_gather_outputs in shard_i_copy_infos: + # Chunk-cat from the temporary to the final all-gather output tensors + shard_dim = fsdp_param.fsdp_placement.dim + + with torch.autograd._unsafe_preserve_version_counter( + tuple(fsdp_param.all_gather_outputs) + ): + for param_all_gather_output, target_all_gather_output in zip( + param_all_gather_outputs, fsdp_param.all_gather_outputs + ): + padded_sharded_size = ( + fsdp_param.padded_sharded_param_size + if fsdp_param.sharded_state == ShardedState.SHARDED + else cast( + torch.Tensor, fsdp_param._sharded_post_forward_param_data + ).size() + ) + pre_param_size = list(padded_sharded_size) + pre_param_size[0] *= world_size + chunks = torch.chunk( + param_all_gather_output.view(pre_param_size), world_size, dim=0 + ) + post_param_size = list(padded_sharded_size) + post_param_size[shard_dim] *= world_size + cat_out = target_all_gather_output.view(post_param_size) + torch.cat(chunks, dim=shard_dim, out=cat_out) + + +@torch.no_grad() +def foreach_reduce( + fsdp_params: list[FSDPParam], + unsharded_grads: list[torch.Tensor], + reduce_scatter_group: dist.ProcessGroup, + reduce_scatter_stream: torch.Stream, + reduce_scatter_comm: ReduceScatter, + orig_dtype: torch.dtype | None, + reduce_dtype: torch.dtype | None, + device: torch.device, + gradient_divide_factor: float | None, + all_reduce_group: dist.ProcessGroup | None, # not `None` iff HSDP + all_reduce_stream: torch.Stream, + all_reduce_grads: bool, + partial_reduce_output: torch.Tensor | None, # only used for HSDP + all_reduce_hook: Callable[[torch.Tensor], None] | None, + force_sum_reduction_for_comms: bool = False, +) -> tuple[ + torch.Tensor, + torch.Event, + torch.Stream, + torch.Event, + torch.Tensor | None, + torch.Event | None, + torch.Tensor | None, +]: + """ + ``unsharded_grads`` owns the references to the gradients computed by + autograd, so clearing the list frees the gradients. + """ + + grad_dtypes = {grad.dtype for grad in unsharded_grads} + if len(grad_dtypes) != 1: + # Check this at runtime since it could be a real runtime error if e.g. + # fp8 weights do not produce the correct higher precision gradients + _raise_assert_with_print( + f"FSDP reduce-scatter expects uniform gradient dtype but got {grad_dtypes}" + ) + grad_dtype = unsharded_grads[0].dtype + reduce_dtype = reduce_dtype or grad_dtype + (predivide_factor, postdivide_factor, reduce_scatter_op, all_reduce_op) = ( + _get_gradient_divide_factors( + reduce_scatter_group, + all_reduce_group, + reduce_dtype, + device.type, + gradient_divide_factor, + force_sum_reduction_for_comms, + ) + ) + + if reduce_scatter_group is None: + world_size = 1 + else: + world_size = reduce_scatter_group.size() + device_handle = _get_device_handle(device.type) + current_stream = device_handle.current_stream() + + if world_size > 1: + for i, (fsdp_param, unsharded_grad) in enumerate( + zip(fsdp_params, unsharded_grads) + ): + if (shard_dim := fsdp_param.fsdp_placement.dim) == 0: + continue + if unsharded_grad.size(shard_dim) % world_size != 0: + raise AssertionError( + f"Shard({shard_dim}) requires even sharding: {unsharded_grad.size()=} {world_size=}" + ) + chunks = torch.chunk(unsharded_grad, world_size, dim=shard_dim) + unsharded_grads[i] = torch.cat(chunks, dim=0) + + padded_unsharded_sizes = tuple( + _get_dim0_padded_size(grad.size(), world_size) for grad in unsharded_grads + ) + reduce_scatter_input_numel = sum(s.numel() for s in padded_unsharded_sizes) + reduce_scatter_output_numel = reduce_scatter_input_numel // world_size + reduce_scatter_input = reduce_scatter_comm.allocate( + (reduce_scatter_input_numel,), + dtype=reduce_dtype, + device=device, + ) + + foreach_reduce_scatter_copy_in(unsharded_grads, reduce_scatter_input, world_size) + + # Only after the copy-in finishes can we free the gradients + unsharded_grads.clear() + reduce_scatter_stream.wait_stream(current_stream) + all_reduce_input = None + all_reduce_event = None + + with device_handle.stream(reduce_scatter_stream): + reduce_output = reduce_scatter_comm.allocate( + (reduce_scatter_output_numel,), + dtype=reduce_dtype, + device=device, + ) + _div_if_needed(reduce_scatter_input, predivide_factor) + if world_size > 1: + reduce_scatter_comm( + output_tensor=reduce_output, + input_tensor=reduce_scatter_input, + group=reduce_scatter_group, + op=reduce_scatter_op, + ) + else: + # For single GPU, just copy the input to output (no actual reduce-scatter needed), and + # account for a possible gradient_divide_factor. + if gradient_divide_factor is not None: + reduce_output.copy_(reduce_scatter_input / gradient_divide_factor) + else: + reduce_output.copy_(reduce_scatter_input) + reduce_scatter_event = reduce_scatter_stream.record_event() + post_reduce_stream = reduce_scatter_stream + if all_reduce_group is not None: # HSDP or DDP/replicate + # Accumulations must run in the reduce-scatter stream + if not all_reduce_grads: + if partial_reduce_output is not None: + partial_reduce_output += reduce_output + else: + partial_reduce_output = reduce_output + return ( + reduce_scatter_input, + reduce_scatter_event, + post_reduce_stream, + post_reduce_stream.record_event(), + all_reduce_input, + all_reduce_event, + partial_reduce_output, + ) + if partial_reduce_output is not None: + reduce_output += partial_reduce_output + post_reduce_stream = all_reduce_stream + if world_size >= 1: + all_reduce_stream.wait_stream(reduce_scatter_stream) + else: + all_reduce_stream.wait_stream(current_stream) + with device_handle.stream(all_reduce_stream): + dist.all_reduce( + reduce_output, + group=all_reduce_group, + op=all_reduce_op, + ) + # Keep refs to the reduce-dtype AR buffer + completion + # event so FSDPParamGroup._all_reduce_state can hold them + # across layers. This keeps the buffer off the caching + # allocator's free list; otherwise the next layer's + # reduce-scatter can reuse the same physical block while + # this layer's AR is still in flight, causing cross-layer + # gradient aliasing under slow AR. See PR #140044, + # regression test PR #180900. + all_reduce_input = reduce_output + all_reduce_event = all_reduce_stream.record_event() + # -- END: ops in reduce_scatter stream + + if all_reduce_hook is not None: + # Execute user-specified all reduce hook. + # If native HSDP is used, this is executed after the HSDP all reduce. + # If 1-d FSDP is used, this is executed post reduce-scatter. + post_reduce_stream = all_reduce_stream + all_reduce_stream.wait_stream(reduce_scatter_stream) + with device_handle.stream(all_reduce_stream): + all_reduce_hook(reduce_output) + # -- END: ops post reduce_scatter + + with device_handle.stream(post_reduce_stream): + _div_if_needed(reduce_output, postdivide_factor) + # Rebinds to a new orig_dtype tensor when reduce_dtype != + # orig_dtype. Do NOT rely on this stream-scoped rebind to manage + # the old reduce-dtype buffer's lifetime: the rebind orders the + # cast before the free-event on AR stream, but the freed block + # lands on the caching allocator's free list and the next layer's + # RS on RS stream can reuse it without waiting for this layer's + # AR to finish. The reduce-dtype buffer is held across layers by + # FSDPParamGroup._all_reduce_state (captured above) to prevent + # this. See PR #140044, regression test PR #180900. + reduce_output = _to_dtype_if_needed(reduce_output, orig_dtype) + # View out and accumulate sharded gradients + flat_grad_offset = 0 # [0, reduce_scatter_output_numel - 1] + for padded_unsharded_size, fsdp_param in zip( + padded_unsharded_sizes, fsdp_params + ): + # Assume even sharding for Shard(i), i > 0; otherwise would require + # copy-out for contiguous strides + new_sharded_grad = torch.as_strided( + reduce_output, + size=fsdp_param.sharded_size, + stride=fsdp_param.contiguous_sharded_stride, + storage_offset=flat_grad_offset, + ) + to_accumulate_grad = fsdp_param.sharded_param.grad is not None + if fsdp_param.offload_to_cpu: + # Only overlap the D2H copy (copying to pinned memory) when no + # in-backward CPU consumer of the grad exists. Two such + # consumers suppress the overlap: + # - Accumulating grads: the CPU add kernel depends on the + # copy result and we cannot run the add as a callback. + # - Post-accumulate-grad hooks: user code (e.g. + # optimizer-in-backward) reads ``param.grad`` on CPU + # synchronously. With ``non_blocking=True`` the hook would + # observe in-flight pinned memory — silently wrong + # optimizer updates. + has_post_acc_grad_hook = bool( + getattr( + fsdp_param.sharded_param, + "_post_accumulate_grad_hooks", + None, + ) + ) + non_blocking = ( + fsdp_param.pin_memory + and not to_accumulate_grad + and not has_post_acc_grad_hook + ) + # Since the GPU sharded gradient is allocated in the RS stream, + # we can free it here by not keeping a ref without waiting for + # the D2H copy since future RS-stream ops run after the copy + new_sharded_grad = new_sharded_grad.to( + torch.device("cpu"), non_blocking=non_blocking + ) + if non_blocking: + # Record an event on which to block the CPU thread to + # ensure that the D2H copy finishes before the optimizer + fsdp_param.grad_offload_event = post_reduce_stream.record_event() + if to_accumulate_grad: + if not isinstance(fsdp_param.sharded_param.grad, DTensor): + raise AssertionError( + f"Expected fsdp_param.sharded_param.grad to be DTensor, got {type(fsdp_param.sharded_param.grad)}" + ) + fsdp_param.sharded_param.grad._local_tensor += new_sharded_grad + else: + new_sharded_dtensor_grad = fsdp_param.to_sharded_dtensor( + new_sharded_grad + ) + fsdp_param.sharded_param.grad = new_sharded_dtensor_grad + for hook in ( + getattr(fsdp_param.sharded_param, "_post_accumulate_grad_hooks", {}) + or {} + ).values(): + hook(fsdp_param.sharded_param) + padded_sharded_numel = padded_unsharded_size.numel() // world_size + flat_grad_offset += padded_sharded_numel + post_reduce_event = post_reduce_stream.record_event() + # The RS output is allocated in the RS stream and used in the default + # stream (for optimizer). To ensure its memory is not reused for later + # RSs, we do not need extra synchronization since the sharded parameters + # hold refs through the end of backward. + return ( + reduce_scatter_input, + reduce_scatter_event, + post_reduce_stream, + post_reduce_event, + all_reduce_input, + all_reduce_event, + None, + ) + + +def foreach_reduce_scatter_copy_in( + unsharded_grads: list[torch.Tensor], + reduce_scatter_input: torch.Tensor, + world_size: int, +) -> None: + reduce_scatter_input = reduce_scatter_input.view(world_size, -1) + torch.ops.fsdp.chunk_cat( + unsharded_grads, dim=0, num_chunks=world_size, out=reduce_scatter_input + ) + + +def _get_all_gather_input_metadatas( + param_all_gather_inputs: list[list[torch.Tensor]], +) -> tuple[list[list[torch.dtype]], list[list[int]], torch.dtype]: + param_all_gather_input_dtypes: list[list[torch.dtype]] = [] + param_all_gather_input_numels: list[list[int]] = [] + all_gather_dtype = param_all_gather_inputs[0][0].dtype + for all_gather_inputs in param_all_gather_inputs: + input_dtypes: list[torch.dtype] = [] + input_numels: list[int] = [] + for all_gather_input in all_gather_inputs: + if all_gather_input.dtype != all_gather_dtype: + all_gather_dtype = torch.uint8 + input_dtypes.append(all_gather_input.dtype) + input_numels.append(all_gather_input.numel()) + param_all_gather_input_dtypes.append(input_dtypes) + param_all_gather_input_numels.append(input_numels) + return ( + param_all_gather_input_dtypes, + param_all_gather_input_numels, + all_gather_dtype, + ) + + +def _get_gradient_divide_factors( + reduce_scatter_group: dist.ProcessGroup | None, + all_reduce_group: dist.ProcessGroup | None, + reduce_dtype: torch.dtype, + device_type: str = "", + factor: float | None = None, + force_sum_reduction_for_comms: bool = False, +) -> tuple[ + float | None, + float | None, + dist.ReduceOp | dist.ReduceOp.RedOpType, + dist.ReduceOp | dist.ReduceOp.RedOpType, +]: + # MTIA appears to only support SUM reduction, hence we force it implicitly + if device_type == "mtia": + force_sum_reduction_for_comms = True + + # For fp32/bf16, we do not need to worry about overflow/underflow, so we + # use NCCL's built-in division to avoid separate div kernels + overflow_risk = reduce_dtype not in (torch.float32, torch.bfloat16) + if reduce_scatter_group is not None: + data_parallel_size = reduce_scatter_group.size() + else: + data_parallel_size = 1 + + if all_reduce_group is not None: + data_parallel_size *= all_reduce_group.size() + + if not overflow_risk and not force_sum_reduction_for_comms: + if factor is None: + # Warning: NCCL ReduceOp.AVG may produce incorrect results with + # world size 1. + if data_parallel_size == 1: + return None, None, ReduceOp.SUM, ReduceOp.SUM + return None, None, ReduceOp.AVG, ReduceOp.AVG + if reduce_scatter_group is not None and factor == reduce_scatter_group.size(): + reduce_scatter_op = ReduceOp.AVG + else: + reduce_scatter_op = ReduceOp.PREMUL_SUM(1 / factor) + return None, None, reduce_scatter_op, ReduceOp.SUM + + if factor is None: + factor = float(data_parallel_size) + pre_factor: float | None + if overflow_risk: + # Since fp16 has smaller dynamic range than fp32/bf16, we want to avoid + # overflow/underflow. For N data parallel workers, each worker computes + # g_i, and they collectively reduce (g_1 + ... + g_N) / N. To avoid + # overflow/underflow, we divide by ~sqrt(N) before/after the reduction. + pre_factor = 1 + while factor % pre_factor == 0 and factor / pre_factor > pre_factor: + pre_factor *= 2 + post_factor = factor / pre_factor + else: + # Prefer post-multiplying as it operates on less data and is thus faster + pre_factor, post_factor = None, factor + + return pre_factor, post_factor, ReduceOp.SUM, ReduceOp.SUM + + +def _div_if_needed(tensor: torch.Tensor, div_factor: float | None) -> None: + if div_factor is not None and div_factor != 1: + tensor.div_(div_factor) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_common.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_common.py new file mode 100644 index 0000000000000000000000000000000000000000..8fc08ff7b3279989d947a4f2c1be2d26b90371e7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_common.py @@ -0,0 +1,228 @@ +# mypy: allow-untyped-defs +import functools +import math +import traceback +from collections.abc import Callable +from dataclasses import dataclass, field +from enum import auto, Enum +from typing import Any, TypeVar +from typing_extensions import ParamSpec + +import torch +import torch.distributed as dist +import torch.nn as nn +from torch.distributed._composable.contract import _get_registry +from torch.distributed.tensor import DeviceMesh, DTensor, Shard +from torch.distributed.tensor._dtensor_spec import DTensorSpec + +from ._fsdp_api import DataParallelMeshDims + + +_P = ParamSpec("_P") +_R = TypeVar("_R") + + +def _dynamo_disable(func: Callable[_P, _R]) -> Callable[_P, _R]: + """Disable dynamo tracing for FSDP hooks.""" + + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs): + return torch._dynamo.disable( + func, recursive=True, reason="skipping FSDP hooks" + )(*args, **kwargs) + + return wrapper + + +def _disable_functorch_if_active(func: Callable[_P, _R]) -> Callable[_P, _R]: + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _R: + if torch._C._are_functorch_transforms_active(): + with torch._C._DisableFuncTorch(): + return func(*args, **kwargs) + return func(*args, **kwargs) + + return wrapper + + +@dataclass +class DataParallelMeshInfo: + mesh: DeviceMesh + shard_mesh_dim: int | None = None + replicate_mesh_dim: int | None = None + dp_mesh_dims: DataParallelMeshDims | None = None + # The full SPMD mesh (excluding PP dims) that params are distributed on. + # Must include all non-PP SPMD dims (e.g. DP + TP); passing a submesh + # that omits dims like TP will lead to incorrect behavior. + spmd_mesh: DeviceMesh | None = field(default=None, repr=False) + is_spmd_mesh: bool = field(default=False, init=False, repr=False) + + def __post_init__(self): + if self.shard_mesh_dim is None and self.replicate_mesh_dim is None: + raise AssertionError( + "At least one of shard_mesh_dim and replicate_mesh_dim must not be None" + ) + self.is_spmd_mesh = self.dp_mesh_dims is not None + + +@dataclass +class FSDPMeshInfo(DataParallelMeshInfo): + def __post_init__(self): + super().__post_init__() + if self.shard_mesh_dim is None: + raise AssertionError("Expects non-None shard_mesh_dim") + self.shard_mesh_size: int = self.mesh.size(self.shard_mesh_dim) + self.shard_process_group = self.mesh.get_group(self.shard_mesh_dim) + self.shard_mesh_rank: int = self.shard_process_group.rank() + # Reduce-scatter shares the shard PG (one NCCL communicator) by default; + # FSDPModule.set_separate_reduce_scatter_group assigns a dedicated PG + # here so reduce-scatter can overlap all-gather in backward. + self.reduce_scatter_process_group: dist.ProcessGroup | None = None + + +@dataclass +class DDPMeshInfo(DataParallelMeshInfo): + def __post_init__(self): + super().__post_init__() + if self.replicate_mesh_dim is None: + raise AssertionError("Expects non-None replicate_mesh_dim") + self.replicate_mesh_size: int = self.mesh.size(self.replicate_mesh_dim) + self.replicate_process_group = self.mesh.get_group(self.replicate_mesh_dim) + self.replicate_mesh_rank: int = self.replicate_process_group.rank() + + +@dataclass +class HSDPMeshInfo(FSDPMeshInfo, DDPMeshInfo): + def __post_init__(self): # pylint:disable=useless-parent-delegation + # Calls `FSDPMeshInfo` -> `DDPMeshInfo` -> `DataParallelMeshInfo` + super().__post_init__() + + +class TrainingState(Enum): + """Describes the training state of one FSDP state / parameter group.""" + + # Transition to forward starting pre-forward until post-forward + FORWARD = auto() + # Transition to pre-backward when unsharding in backward + PRE_BACKWARD = auto() + # Transition to post-backward when resharding and reducing gradients + POST_BACKWARD = auto() + # Idle before/after forward or before pre-backward/after post-backward + IDLE = auto() + + +def _raise_assert_with_print(*args: Any, **kwargs: Any): + print(f"[Rank {dist.get_rank()}] ", end="") + print(*args, **kwargs) + traceback.print_stack() + raise AssertionError(*args, **kwargs) + + +def _is_composable_with_fsdp(module: nn.Module) -> bool: + registry = _get_registry(module) + if registry is None: + return True + # Registry keys by function name + return "replicate" not in registry + + +def _get_dim0_padded_size(tensor_size: torch.Size, dim0_factor: int) -> torch.Size: + padded_dim0 = math.ceil(tensor_size[0] / dim0_factor) * dim0_factor + return torch.Size([padded_dim0]) + tensor_size[1:] + + +def _chunk_with_empty( + tensor: torch.Tensor, num_chunks: int, dim: int +) -> list[torch.Tensor]: + chunks = list(torch.chunk(tensor, num_chunks, dim=dim)) + while len(chunks) < num_chunks: + chunks.append(chunks[0].new_empty(0)) + return chunks + + +def _get_dim_chunked_size( + chunk: torch.Tensor, unchunked_size: torch.Size, dim: int +) -> torch.Size: + if chunk.numel() > 0: + return chunk.size() + # For 0 numel, we need to preserve nonzero-sized dims for DTensor APIs + # pyrefly: ignore [bad-return] + return unchunked_size[:dim] + torch.Size([0]) + unchunked_size[dim + 1 :] + + +def _from_local_no_grad( + local_tensor: torch.Tensor, + sharding_spec: DTensorSpec, +) -> DTensor: + """ + This method is similar to ``DTensor.from_local()`` except that in eager mode + it avoids some CPU overhead by avoiding default args and not being differentiable. + """ + # pyrefly: ignore [bad-argument-type] + return DTensor( + # Use the local tensor directly instead of constructing a new tensor + # variable, e.g. with `view_as()`, since this is not differentiable + # pyrefly: ignore [bad-argument-count] + local_tensor, + sharding_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=local_tensor.requires_grad, + ) + + +def _to_dtype_if_needed( + tensor: torch.Tensor, dtype: torch.dtype | None +) -> torch.Tensor: + if dtype is not None and tensor.dtype != dtype: + return tensor.to(dtype) + return tensor + + +def _cast_fp_tensor(dtype: torch.dtype, x: torch.Tensor) -> torch.Tensor: + if ( + not isinstance(x, torch.Tensor) + or not torch.is_floating_point(x) + or x.dtype == dtype + ): + return x + return x.to(dtype) + + +def is_bw() -> bool: + return torch._C._current_graph_task_id() != -1 + + +@dataclass +class ShardPlacementResult: + placement: Shard | None + mesh_info: FSDPMeshInfo + + +ShardPlacementFnResult = Shard | ShardPlacementResult | None + + +def resolve_shard_placement( + result: ShardPlacementFnResult, + default_mesh_info: FSDPMeshInfo, +) -> ShardPlacementResult: + """Resolve the shard_placement_fn result to a ShardPlacementResult. + + Handles different input types and applies defaults: + - None: Use default sharding (Shard(0)) on default mesh + - Shard: Use specified shard dimension on default mesh + - ShardPlacementResult: Use as-is + + Args: + result: The return value from shard_placement_fn, or None if no fn provided. + default_mesh_info: The default FSDPMeshInfo to use if not specified. + + Returns: + A ShardPlacementResult with placement and mesh_info. + """ + if result is None: + return ShardPlacementResult(placement=None, mesh_info=default_mesh_info) + if isinstance(result, Shard): + return ShardPlacementResult(placement=result, mesh_info=default_mesh_info) + if isinstance(result, ShardPlacementResult): + return result + raise ValueError(f"Invalid shard_placement_fn result: {result}") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_init.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_init.py new file mode 100644 index 0000000000000000000000000000000000000000..87efcb3f3344787700e3e39ab6a0cd8a138f94b7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_init.py @@ -0,0 +1,569 @@ +import itertools +import logging +from typing import TYPE_CHECKING + +import torch +import torch.distributed as dist +import torch.nn as nn +from torch._logging import warning_once +from torch.distributed.device_mesh import _get_device_handle +from torch.distributed.tensor import DeviceMesh, DTensor, init_device_mesh +from torch.utils._python_dispatch import is_traceable_wrapper_subclass + +from ._fsdp_common import ( + _is_composable_with_fsdp, + DataParallelMeshInfo, + DDPMeshInfo, + FSDPMeshInfo, + HSDPMeshInfo, +) +from ._fsdp_state import _get_module_fsdp_state + + +if TYPE_CHECKING: + from collections.abc import Callable + from typing import Any + + from ._fsdp_api import DataParallelMeshDims, MixedPrecisionPolicy, OffloadPolicy + from ._fsdp_common import ShardPlacementFnResult + from ._fsdp_state import FSDPState + + +logger = logging.getLogger("torch.distributed.fsdp.fully_shard") + + +def _validate_module(module: nn.Module, func_name: str) -> None: + """ + Validate that the module can be used with fully_shard or replicate. + + Raises ValueError if the module is a container that doesn't implement forward. + """ + if ( + isinstance(module, (nn.ModuleList, nn.ModuleDict)) + and module.__class__.forward is nn.Module.forward + ): + raise ValueError( + f"{func_name} does not support containers that do not implement forward: {module}" + ) + + +def _validate_mesh( + mesh: "DeviceMesh", + dp_mesh_dims: "DataParallelMeshDims | None" = None, +) -> None: + """ + Validate that the mesh can be used with fully_shard. + + When ``dp_mesh_dims`` is provided, validates that the named dims + exist in the mesh and at least one of shard/replicate is set. + Otherwise raises ValueError if the mesh is not 1D or 2D. + """ + if dp_mesh_dims is not None: + if dp_mesh_dims.shard is None and dp_mesh_dims.replicate is None: + raise ValueError( + "At least one of shard or replicate must be set in dp_mesh_dims" + ) + if mesh.mesh_dim_names is None: + raise ValueError( + "mesh must have mesh_dim_names when dp_mesh_dims is provided" + ) + names_to_check: list[str] = list(dp_mesh_dims.shard_names) + names_to_check.extend(dp_mesh_dims.replicate_names) + for name in names_to_check: + if name not in mesh.mesh_dim_names: + raise ValueError( + f"Mesh dim name '{name}' not found in mesh.mesh_dim_names " + f"{mesh.mesh_dim_names}" + ) + return + if mesh.ndim not in (1, 2): + raise ValueError(f"fully_shard expects a 1D or 2D DeviceMesh but got {mesh}") + if mesh.ndim == 2 and mesh.mesh_dim_names is None: + raise AssertionError( + "Please init the 2D mesh for HSDP with mesh_dim_names specified" + ) + + +def _get_mesh_info( + mesh: "DeviceMesh", + dp_mesh_dims: "DataParallelMeshDims | None" = None, +) -> "DataParallelMeshInfo": + """ + Get the appropriate mesh info for the given mesh. + + When ``dp_mesh_dims`` is provided, extracts the DP submesh from the + full SPMD mesh and returns FSDPMeshInfo, HSDPMeshInfo, or DDPMeshInfo + with ``dp_mesh_dims`` set and ``is_spmd_mesh`` as True. + + Returns FSDPMeshInfo for 1D mesh, HSDPMeshInfo for 2D mesh. + """ + if dp_mesh_dims is not None: + return _get_mesh_info_from_named_dims(mesh, dp_mesh_dims) + if mesh.ndim == 1: + return FSDPMeshInfo(mesh, shard_mesh_dim=0) + else: + return HSDPMeshInfo(mesh, shard_mesh_dim=1, replicate_mesh_dim=0) + + +def _get_mesh_info_from_named_dims( + mesh: "DeviceMesh", + dp_mesh_dims: "DataParallelMeshDims", +) -> "DataParallelMeshInfo": + shard_names = dp_mesh_dims.shard_names + replicate_names = dp_mesh_dims.replicate_names + + def _get_submesh(names: tuple[str, ...]) -> "DeviceMesh": + if len(names) == 1: + return mesh[names[0]] + # Flatten multi-dim submesh into a single dim so FSDP's internal + # logic (which expects one shard and/or one replicate dim) works + # unchanged. This creates a new 1D DeviceMesh and ProcessGroup. + return mesh[names]._flatten("_".join(names)) + + if len(shard_names) == 0: # DDP + dp_mesh = _get_submesh(replicate_names) + return DDPMeshInfo( + dp_mesh, + replicate_mesh_dim=0, + dp_mesh_dims=dp_mesh_dims, + spmd_mesh=mesh, + ) + if len(replicate_names) == 0: # FSDP + dp_mesh = _get_submesh(shard_names) + return FSDPMeshInfo( + dp_mesh, + shard_mesh_dim=0, + dp_mesh_dims=dp_mesh_dims, + spmd_mesh=mesh, + ) + # HSDP + shard_mesh = _get_submesh(shard_names) + replicate_mesh = _get_submesh(replicate_names) + dp_mesh = DeviceMesh._concatenate([replicate_mesh, shard_mesh]) + return HSDPMeshInfo( + dp_mesh, + shard_mesh_dim=1, + replicate_mesh_dim=0, + dp_mesh_dims=dp_mesh_dims, + spmd_mesh=mesh, + ) + + +def _get_post_forward_mesh_info( + reshard_after_forward: bool | int, mesh_info: FSDPMeshInfo +) -> FSDPMeshInfo | None: + shard_mesh_size = mesh_info.shard_mesh_size + if not isinstance(reshard_after_forward, (bool, int)): + raise ValueError( + "reshard_after_forward should be a bool or an int representing the " + f"group size to reshard to, not {reshard_after_forward}" + ) + # NOTE: `isinstance(False, int)` returns `True`. + if not isinstance(reshard_after_forward, bool) and isinstance( + reshard_after_forward, int + ): + if ( + reshard_after_forward < 1 + or reshard_after_forward > shard_mesh_size + or shard_mesh_size % reshard_after_forward != 0 + ): + raise ValueError( + "If passing reshard_after_forward as an int, it should be a " + f"factor of {shard_mesh_size}, not {reshard_after_forward}" + ) + elif reshard_after_forward == 1: + msg = ( + "reshard_after_forward=1 (int) means resharding parameters to world size 1, " + "instead of reshard_after_forward=True (bool)" + ) + warning_once(logger, msg, stacklevel=2) + reshard_after_forward = False + elif reshard_after_forward == shard_mesh_size: + reshard_after_forward = True + post_forward_mesh_info = None + if reshard_after_forward is True: + post_forward_mesh_info = mesh_info + elif reshard_after_forward is not False: # int case + # For HSDP, we can flatten the two replicate dims into the 0th dim + post_forward_mesh_tensor = mesh_info.mesh.mesh.view(-1, reshard_after_forward) + post_forward_mesh = DeviceMesh( + mesh_info.mesh.device_type, post_forward_mesh_tensor + ) + post_forward_mesh_info = HSDPMeshInfo( + post_forward_mesh, shard_mesh_dim=1, replicate_mesh_dim=0 + ) + return post_forward_mesh_info + + +def _init_default_mesh( + mesh_dim_names: tuple[str, ...] | None = None, +) -> DeviceMesh: + """Default to global CUDA mesh if possible else global CPU mesh.""" + if not dist.distributed_c10d.is_initialized(): + dist.distributed_c10d.init_process_group() + default_pg = dist.distributed_c10d._get_default_group() + device = torch._C._get_accelerator() + mesh = init_device_mesh( + device.type, + mesh_shape=(default_pg.size(),), + mesh_dim_names=mesh_dim_names, + ) + return mesh + + +def _init_default_fully_shard_mesh() -> DeviceMesh: + """Default to global CUDA mesh if possible else global CPU mesh.""" + return _init_default_mesh() + + +def _get_device_from_mesh(mesh: DeviceMesh) -> torch.device: + if mesh.device_type == "cpu": + return torch.device("cpu") + device_handle = _get_device_handle(mesh.device_type) + return torch.device(mesh.device_type, device_handle.current_device()) + + +def _ignore_module( + module: nn.Module, + ignored_params: set[nn.Parameter], + ignore_decision: dict[nn.Module, bool], +) -> bool: + """ + Decide if it is safe to ignore a module for applying fully_shard. + """ + if module in ignore_decision: + return ignore_decision[module] + + if len(list(module.buffers(recurse=False))) > 0: + # Cannot ignore a module with any buffer + ignore_decision[module] = False + return False + + for _, param in module.named_parameters(recurse=False): + if param not in ignored_params: + # at least one param is not ignored. So this module shouldn't be. + ignore_decision[module] = False + return False + + # Need to consider descendants of module + for child in list(module.children()): + ignore_child = _ignore_module(child, ignored_params, ignore_decision) + if not ignore_child: + # Cannot ignore module if one of its children is not ignored + ignore_decision[module] = False + return False + + # Safe to ignore module + ignore_decision[module] = True + return True + + +def _adjust_managed_modules( + modules: list[nn.Module], ignored_params: set[nn.Parameter] +) -> list[nn.Module]: + """ + Adjust the given list of managed modules by removing those with all parameters ignored. + """ + ignore_decision: dict[nn.Module, bool] = {} + new_modules = [] + for module in modules: + ignored = _ignore_module(module, ignored_params, ignore_decision) + if not ignored: + new_modules.append(module) + return new_modules + + +def _get_managed_modules( + root_modules: tuple[nn.Module, ...], + ignored_params: set[nn.Parameter] | None = None, + is_composable_fn: "Callable[[nn.Module], bool] | None" = None, + get_state_fn: "Callable[[nn.Module], Any] | None" = None, +) -> list[nn.Module]: + """ + Get the list of managed modules for FSDP/replicate. + + Args: + root_modules: The root modules to start the search from. + ignored_params: Parameters to ignore. + is_composable_fn: Callable to check if a module is composable. + Defaults to ``_is_composable_with_fsdp``. + get_state_fn: Callable to get the state of a module. + Defaults to ``_get_module_fsdp_state``. + """ + if is_composable_fn is None: + is_composable_fn = _is_composable_with_fsdp + if get_state_fn is None: + get_state_fn = _get_module_fsdp_state + + modules: list[nn.Module] = [] + root_modules_set = set(root_modules) + # Track visited modules to avoid visiting shared modules multiple times + visited_modules: set[nn.Module] = set() + + def dfs(module: nn.Module) -> None: + """ + Runs a DFS to collect managed modules, not recursing into modules with + a non-composable API or ``fully_shard`` already applied. + """ + if not is_composable_fn(module): + return + elif module not in root_modules_set and get_state_fn(module) is not None: + return # nested `fully_shard` module + visited_modules.add(module) + for submodule in module.children(): + if submodule not in visited_modules: + dfs(submodule) + modules.append(module) + + for root_module in root_modules: + dfs(root_module) + + if ignored_params is None: + return modules + + adjusted_modules = _adjust_managed_modules(modules, ignored_params) + return adjusted_modules + + +def _verify_managed_param(name: str, param: nn.Parameter) -> None: + """ + Verify if the parameter is accepted by fully_shard. The only restriction now + is that the parameter cannot be a scalar tensor (param.numel == 0) since we + need at least one dim to shard. + """ + if len(param.shape) == 0: + raise ValueError( + "fully_shard doesn't support scalar parameters. " + f"Change {name} to a 1D tensor with numel equal to 1." + ) + + +def _get_managed_states( + modules: list[nn.Module], ignored_params: set[nn.Parameter] | None = None +) -> tuple[list[nn.Parameter], list[torch.Tensor]]: + params: list[nn.Parameter] = [] + buffers: list[torch.Tensor] = [] + # Track visited parameters/buffers to avoid visiting shared parameters and + # buffers multiple times + visited_params: set[nn.Parameter] = set() + visited_buffers: set[torch.Tensor] = set() + if ignored_params is None: + ignored_params = set() + + for module in modules: + for name, param in module.named_parameters(recurse=False): + if param in ignored_params: + # do not include an ignored parameters + continue + if param not in visited_params: + _verify_managed_param(name, param) + params.append(param) + visited_params.add(param) + for buffer in module.buffers(recurse=False): + if buffer not in visited_buffers: + buffers.append(buffer) + visited_buffers.add(buffer) + return params, buffers + + +def _move_states_to_device( + params: list[nn.Parameter], + buffers: list[torch.Tensor], + device: torch.device, +) -> None: + """ + We have FSDP move states to device for simpler and faster initialization + since FSDP almost always uses CUDA for training. We move parameters/buffers + rather than modules since modules to support ignoring parameters/buffers in + the future. + """ + # Follow the logic in `nn.Module._apply` + # pyrefly: ignore [bad-argument-type] + for tensor in itertools.chain(params, buffers): + if tensor.device == device or tensor.device.type == "meta": + # Keep meta-device tensors on meta device for deferred init + continue + if isinstance(tensor, DTensor): + if (dtensor_mesh_type := tensor.device_mesh.device_type) != device.type: + raise ValueError( + "Requires DTensor to have mesh of the same type as the FSDP mesh " + f"but got {dtensor_mesh_type} for DTensor and {device.type} for FSDP" + ) + raise AssertionError( + f"Expects DTensor to be moved to {dtensor_mesh_type} but got {tensor.device}" + ) + tensor_ = tensor + if is_traceable_wrapper_subclass(tensor_): + with torch.no_grad(): # avoid autograd increasing C++ refcount by 1 + tensor_on_device = nn.Parameter(tensor.to(device)) + torch.utils.swap_tensors(tensor, tensor_on_device) + else: + tensor.data = tensor.to(device) + + +def _apply_to_module( + modules: tuple[nn.Module, ...], + cls_to_wrapper_cls: dict[type, type], + wrapper_module_cls: type, + wrapper_cls_prefix: str, + unimplemented_deepcopy: "Callable", +) -> None: + """ + Modify module classes to include the wrapper class in their MRO. + + Args: + modules: The modules to apply the wrapper to. + cls_to_wrapper_cls: Cache dict mapping original class to wrapper class. + wrapper_module_cls: The wrapper module class (e.g., FSDPModule, ReplicateModule). + wrapper_cls_prefix: Prefix for the dynamically created class name (e.g., "FSDP", "Replicate"). + unimplemented_deepcopy: The deepcopy function to use for the wrapper class. + """ + for module in modules: + cls = module.__class__ + new_cls = cls_to_wrapper_cls.get(cls) + if not new_cls: + dct = {"__deepcopy__": unimplemented_deepcopy} + new_cls = type( + f"{wrapper_cls_prefix}{cls.__name__}", (wrapper_module_cls, cls), dct + ) + cls_to_wrapper_cls[cls] = new_cls + module.__class__ = new_cls + + +def _init_param_group( + state: "FSDPState", + params: list[nn.Parameter], + modules: tuple[nn.Module, ...], + mesh_info: DataParallelMeshInfo, + post_forward_mesh_info: FSDPMeshInfo | None, + device: torch.device, + shard_placement_fn: "Callable[[nn.Parameter], ShardPlacementFnResult] | None", + mp_policy: "MixedPrecisionPolicy", + offload_policy: "OffloadPolicy", + reshard_after_forward: bool | int = True, +) -> None: + """ + Initialize FSDP param groups for the given state. + + Params are grouped by their process group (derived from ``mesh_info`` via + ``shard_placement_fn``). Each group becomes a separate ``FSDPParamGroup``. + When ``shard_placement_fn`` is ``None`` or returns the same mesh for all + params, this creates a single group. + """ + # Import here to avoid circular imports + from ._fsdp_common import FSDPMeshInfo, resolve_shard_placement + from ._fsdp_param_group import FSDPParamGroup + + if not params: + return + + if shard_placement_fn is None: + # No shard_placement_fn means all params use the same mesh_info, + # so no grouping is needed. This also handles DDPMeshInfo from + # replicate_with_fsdp, which doesn't have shard_process_group. + state._fsdp_param_groups.append( + FSDPParamGroup( + params, + modules, + mesh_info, + post_forward_mesh_info, + device, + shard_placement_fn, + mp_policy, + offload_policy, + ) + ) + return + + # Group params by their process group to support per-param mesh, + # e.g., expert params using ep_mesh vs regular params using dp_mesh. + # For HSDP, also key by replicate_process_group to avoid grouping + # FSDPMeshInfo params with HSDPMeshInfo params that share the same + # shard_process_group but require different gradient reduction behavior. + if not isinstance(mesh_info, FSDPMeshInfo): + raise ValueError( + "Per-param mesh via shard_placement_fn is not supported with " + f"{type(mesh_info).__name__}; it requires FSDPMeshInfo (or subclass)" + ) + pg_to_group: dict[ + tuple[dist.ProcessGroup, dist.ProcessGroup | None], + tuple[FSDPMeshInfo, list[nn.Parameter]], + ] = {} + for param in params: + param_mesh_info = resolve_shard_placement( + shard_placement_fn(param), + mesh_info, + ).mesh_info + shard_pg = param_mesh_info.shard_process_group + replicate_pg: dist.ProcessGroup | None = None + if isinstance(param_mesh_info, HSDPMeshInfo): + replicate_pg = param_mesh_info.replicate_process_group + key = (shard_pg, replicate_pg) + if key not in pg_to_group: + pg_to_group[key] = (param_mesh_info, [param]) + else: + existing_mesh_info = pg_to_group[key][0] + if existing_mesh_info is not param_mesh_info: + raise ValueError( + f"Params sharing the same process group must use the same " + f"FSDPMeshInfo object, but got different objects: " + f"{existing_mesh_info} vs {param_mesh_info}" + ) + pg_to_group[key][1].append(param) + + # Create a FSDPParamGroup per process group + for group_mesh_info, group_params in pg_to_group.values(): + if group_mesh_info is not mesh_info: + group_post_forward = _get_post_forward_mesh_info( + reshard_after_forward, group_mesh_info + ) + else: + group_post_forward = post_forward_mesh_info + state._fsdp_param_groups.append( + FSDPParamGroup( + group_params, + modules, + group_mesh_info, + group_post_forward, + device, + shard_placement_fn, + mp_policy, + offload_policy, + ) + ) + + +def _get_modules_and_states( + module: nn.Module, + device: torch.device, + ignored_params: set[nn.Parameter] | None, + is_composable_fn: "Callable[[nn.Module], bool] | None" = None, + get_state_fn: "Callable[[nn.Module], Any] | None" = None, +) -> tuple[ + nn.Module, + tuple[nn.Module, ...], + list[nn.Module], + list[nn.Parameter], + list[torch.Tensor], +]: + """ + Get modules tuple, managed modules, params, and buffers for FSDP/replicate initialization. + + Returns: + Tuple of (arg_module, modules, managed_modules, params, buffers) + """ + from torch.distributed.utils import _get_root_modules + + arg_module = module + modules = ( + (module,) if isinstance(module, nn.Module) else tuple(_get_root_modules(module)) + ) + + managed_modules = _get_managed_modules( + modules, ignored_params, is_composable_fn, get_state_fn + ) + params, buffers = _get_managed_states(managed_modules, ignored_params) + + _move_states_to_device(params, buffers, device) + + return arg_module, modules, managed_modules, params, buffers diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param.py new file mode 100644 index 0000000000000000000000000000000000000000..239e39e2bf317363d0e515525cbb141722b6d6c4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param.py @@ -0,0 +1,1329 @@ +# mypy: allow-untyped-defs +import inspect +import itertools +from collections.abc import Callable, Sequence +from dataclasses import dataclass, field, replace +from enum import auto, Enum +from typing import Any, cast, TYPE_CHECKING + + +if TYPE_CHECKING: + from ._fsdp_api import DataParallelMeshDims + +import torch +import torch.distributed as dist +import torch.nn as nn + + +if dist._is_spmd_types_available(): + import spmd_types as spmd + from spmd_types._mesh_axis import flatten_axes + from spmd_types.runtime import get_partition_spec + from spmd_types.types import partition_spec_to_shard_types + +from torch._prims_common import make_contiguous_strides_for +from torch.distributed._functional_collectives import AsyncCollectiveTensor +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.fsdp._fully_shard._fsdp_common import DDPMeshInfo +from torch.distributed.tensor import DTensor, Replicate, Shard +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor.placement_types import _StridedShard, Placement + +from ._fsdp_api import CPUOffloadPolicy, MixedPrecisionPolicy, OffloadPolicy +from ._fsdp_common import ( + _chunk_with_empty, + _from_local_no_grad, + _get_dim_chunked_size, + _raise_assert_with_print, + _to_dtype_if_needed, + DataParallelMeshInfo, + FSDPMeshInfo, + HSDPMeshInfo, + resolve_shard_placement, + ShardPlacementFnResult, +) + + +_orig_param_uid_counter = itertools.count() + + +def _get_orig_param_uid(param: nn.Parameter) -> int: + if not hasattr(param, "_fsdp_orig_uid"): + uid = next(_orig_param_uid_counter) + param._fsdp_orig_uid = uid # pyrefly: ignore[missing-attribute] + return param._fsdp_orig_uid # pyrefly: ignore[missing-attribute] + + +def _spans_same_mesh(lhs_axes: tuple[Any, ...], rhs_axes: tuple[Any, ...]) -> bool: + """Return whether two axis sets span the same normalized rank mesh.""" + lhs_mesh = spmd.normalize_mesh(frozenset(lhs_axes)) + rhs_mesh = spmd.normalize_mesh(frozenset(rhs_axes)) + if not lhs_mesh or not rhs_mesh: + return lhs_mesh == rhs_mesh + return flatten_axes(tuple(lhs_mesh)) == flatten_axes(tuple(rhs_mesh)) + + +""" +[Note: FSDP tensors] +FSDP considers the following tensors: +- Original parameter: parameter passed to :class:`FSDPParam`, i.e. the one + on the module when applying FSDP +- Sharded parameter: sharding the original parameter on dim-0 (or a + user-specified dim) as a DTensor over the main mesh +- All-gather inputs: the ``torch.Tensor`` or ``Tensor`` s passed to all-gather, + derived from the sharded parameter +- All-gather output: the ``torch.Tensor`` or ``Tensor`` s resulting from + all-gathering the all-gather inputs +- Unsharded parameter: parameter used for forward/backward computation, derived + from the all-gather output; autograd leaf + +We define these tensors to describe the general framework that can accommodate +extensions, where: +- all-gather-inputs = pre-all-gather-transform(sharded-parameter) +- unsharded-parameter = post-all-gather-transform(all-gather-outputs) + +For the default ``torch.Tensor`` case, there is only one all-gather input, and +it shares the same underlying tensor data as the sharded parameter, meaning +that they can be thought of as the same tensors. The same applies for the +all-gather output and unsharded parameter. For non-``torch.Tensor`` extensions, +these equivalences may no longer hold due to the pre/post-all-gather +transforms, and some may have multiple all-gather inputs/outputs (e.g. +quantized data and scales). + +[Note: FSDP and autograd] +FSDP dynamically frees and allocates the unsharded parameter. Since autograd +can pack a reference to it or a view to save for backward, we use storage +resizing to implement the freeing/allocation since that preserves the aliasing. +This implies that we construct the unsharded parameter object once and write to +it in-place thereafter. For the default ``torch.Tensor` original parameter +case, the all-gather output and unsharded parameter share the same +data, so we use storage resizing on the all-gather output. + +[Note: FSDP and spmd_types] +Refer to https://github.com/meta-pytorch/spmd_types/blob/main/docs/local_spmd_types.md +""" + +lib = torch.library.Library("fsdp", "FRAGMENT") + +lib.define("copy_(Tensor(a!) tensor, Tensor data) -> ()") + + +@torch.library.impl(lib, "copy_", "Meta") +@torch.library.impl(lib, "copy_", "CUDA") +@torch.library.impl(lib, "copy_", "XPU") +@torch.library.impl(lib, "copy_", "HPU") +@torch.library.impl(lib, "copy_", "CPU") +@torch.library.impl(lib, "copy_", "MTIA") +def copy_(tensor, data): + tensor.copy_(data) + + +@torch.library.impl(lib, "copy_", "Functionalize") +def copy__functionalize(tensor, data): + torch._sync(tensor) + torch._sync(data) + tensor_inner = torch._from_functional_tensor(tensor) + data_inner = torch._from_functional_tensor(data) + with torch._C._ExcludeDispatchKeyGuard( + torch._C.DispatchKeySet(torch._C.DispatchKey.Functionalize) + ): + torch.ops.fsdp.copy_.default(tensor_inner, data_inner) + + +torch.fx.node.has_side_effect(torch.ops.fsdp.copy_.default) + + +class ShardedState(Enum): + """ + - ``SHARDED``: The sharded parameter is registered to the module. It is the + only contributor to parameter memory. + - ``SHARDED_POST_FORWARD``: The unsharded parameter is resharded to a + smaller world size. Since this data should not be used for computation, + we do not register it to the module. Users should reshard the module + before any in-place modifications. Both it and the sharded parameter + contribute to parameter memory. + - ``UNSHARDED``: The unsharded parameter is registered to the module. Both + it and the sharded parameter contribute to parameter memory. + """ + + SHARDED = auto() + SHARDED_POST_FORWARD = auto() + UNSHARDED = auto() + + +@dataclass +class ParamModuleInfo: + """ + For a parameter, this stores the module and the parameter name to be able + to do a parameter swap via ``setattr(module, param_name, ...)`` or to get + the parameter via ``getattr(module, param_name)``. We additionally save + shared modules and shared parameter names to update them accordingly. + """ + + # Parameter names are unprefixed, e.g. "weight", not "lin.weight" + module: nn.Module + param_name: str + shared_modules: list[nn.Module] = field(default_factory=list) + shared_param_names: list[str] = field(default_factory=list) + + +@dataclass +class ExtensionsData: + # User-defined metadata passed from pre to post-all-gather + all_gather_metadata: Any | None = None + # Save the all-gather input sizes to unflatten the all-gather outputs to ND + all_gather_input_sizes: Sequence[torch.Size] = () # ND + + def clear(self): + self.all_gather_metadata = None + self.all_gather_input_sizes = () + + +class FSDPParam: + """ + This class manages a parameter with FSDP or FSDP variants applied, + implementing dim-0 per-parameter sharding. + """ + + orig_dtype: torch.dtype + param_dtype: torch.dtype | None + reduce_dtype: torch.dtype | None + _orig_size: torch.Size # ND + sharded_size: torch.Size # ND + contiguous_sharded_stride: tuple[int, ...] + padded_sharded_param_size: torch.Size # ND + sharded_post_forward_size: torch.Size # ND + contiguous_sharded_post_forward_stride: tuple[int, ...] + _sharded_param_data: torch.Tensor # 1D + sharded_param: nn.Parameter # ND + _sharded_post_forward_param_data: torch.Tensor | None # 1D + _sharded_post_forward_param: nn.Parameter | None # ND + _unsharded_param: nn.Parameter # ND + unsharded_accumulated_grad: torch.Tensor | None # ND + _sharding_spec: DTensorSpec + _unsharded_dtensor_spec: ( + DTensorSpec | None + ) # set for DTensor params (SPMD or TP/EP) + all_gather_outputs: list[torch.Tensor] # 1D + # All-gather extension attributes + _extensions_data: ExtensionsData + _unsharded_inner_tensors: list[torch.Tensor] + _orig_param_uid: int + + def __init__( + self, + param: nn.Parameter, + module_info: ParamModuleInfo, + mesh_info: DataParallelMeshInfo, + post_forward_mesh_info: FSDPMeshInfo | None, + device: torch.device, + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None, + mp_policy: MixedPrecisionPolicy, + offload_policy: OffloadPolicy, + ): + self._module_info: ParamModuleInfo = module_info + self.post_forward_mesh_info = post_forward_mesh_info + self.device = device + self.mp_policy = mp_policy + self.offload_to_cpu: bool = isinstance(offload_policy, CPUOffloadPolicy) + self.pin_memory = ( + self.offload_to_cpu and cast(CPUOffloadPolicy, offload_policy).pin_memory + ) + self.grad_offload_event: torch.Event | None = None + self._init_sharded_param(param, device, shard_placement_fn, mesh_info) + if self.post_forward_mesh_info: + self._init_sharded_post_forward_param_metadata(param) + self._init_extensions() + self.all_gather_outputs: list[torch.Tensor] = [] + self.unsharded_accumulated_grad = None + self._param_fqn: str | None = None # prefixed from root module + # TODO: Remove this padding logic once DTensor pads the local tensor: + # https://github.com/pytorch/pytorch/issues/113045 + self._post_load_hook_handle = ( + module_info.module.register_load_state_dict_post_hook( + lambda *args, **kwargs: self.reset_sharded_param() + ) + ) + + @torch.no_grad() + def _init_sharded_param( + self, + param: nn.Parameter, + device: torch.device, + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None, + mesh_info: DataParallelMeshInfo, + ): + shard_placement_result = None + if callable(shard_placement_fn): + shard_placement_result = resolve_shard_placement( + shard_placement_fn(param), + cast(FSDPMeshInfo, mesh_info), + ) + self.mesh_info = shard_placement_result.mesh_info + fsdp_placement = shard_placement_result.placement + else: + self.mesh_info = mesh_info # pyrefly: ignore[bad-assignment] + fsdp_placement = None + self._shard_mesh = self._init_shard_mesh() + if param.device != device and param.device.type != "meta": + raise AssertionError( + f"Expects the parameter to already be moved to device {device} but got {param.device}" + ) + if not param.is_contiguous(): + raise NotImplementedError( + f"FSDP does not support non-contiguous parameters yet: {param.shape=} {param.stride()=}" + ) + if fsdp_placement is None: + fsdp_placement = Shard(0) + elif fsdp_placement.dim < 0: + fsdp_placement = Shard(fsdp_placement.dim + param.ndim) + if not isinstance(fsdp_placement, Shard): + raise AssertionError( + f"Expected Shard, got {type(fsdp_placement)}: {fsdp_placement}" + ) + self.fsdp_placement = fsdp_placement + shard_dim = fsdp_placement.dim + # TODO: Replace the sharded DTensor parameter construction logic with + # `distribute_tensor` after https://github.com/pytorch/pytorch/issues/116101 + # TODO: Simplify the following sharded parameter padding logic after + # https://github.com/pytorch/pytorch/issues/113045 + # Tracks that this parameter contained FSDP-init-time spmd_types annotations; + # this remains true even if spmd typechecking is disabled at runtime. + self.is_spmd_types = ( + dist._is_spmd_types_available() + and bool(spmd_local_type := spmd.get_local_type(param)) + and not isinstance(param, DTensor) + ) + if self.is_spmd_types: + param = self._resolve_spmd_types_for_storage( + param, get_partition_spec(param), spmd_local_type, self.mesh_info + ) + self.is_dtensor = isinstance(param, DTensor) + self._orig_param_uid = _get_orig_param_uid(param) + param_data = self._init_sharding_spec(param, fsdp_placement, shard_dim) + if not param_data.is_contiguous(): + raise AssertionError( + f"Expected contiguous tensor, got {param_data.shape=} {param_data.stride()=}" + ) + shard_dim = fsdp_placement.dim + if shard_dim >= param_data.ndim: + raise AssertionError( + f"Shard dim {shard_dim} is invalid for {param_data.ndim}D tensor: {param.shape}" + ) + self._orig_size = param_data.size() + self._contiguous_orig_stride = make_contiguous_strides_for(self._orig_size) + if isinstance(self.mesh_info, FSDPMeshInfo): # FSDP or HSDP + shard_rank = self.mesh_info.shard_mesh_rank + shard_world_size = self.mesh_info.shard_mesh_size + else: # DDP + shard_rank = 0 + shard_world_size = 1 + + if shard_dim > 0 and param_data.size(shard_dim) % shard_world_size != 0: + # If sharding on nonzero dim, require even sharding for now because + # the uneven sharding (1) requires extra copies before/after FSDP + # collectives and (2) introduces extra complexity to handle padding + # and unpadding + raise NotImplementedError( + f"FSDP does not support uneven sharding on dim {shard_dim}: " + f"{param_data.size()} (world size: {shard_world_size})" + ) + chunks = _chunk_with_empty(param_data, shard_world_size, dim=shard_dim) + sharded_param = chunks[shard_rank] + self.sharded_size = _get_dim_chunked_size( + sharded_param, param_data.size(), dim=shard_dim + ) + self.contiguous_sharded_stride = make_contiguous_strides_for(self.sharded_size) + padded_sharded_size = chunks[0].size() # 0th always padded + self.padded_sharded_param_size = padded_sharded_size + # Pre-pad the sharded parameter to avoid padding before all-gather + padded_sharded_param = param_data.new_zeros(padded_sharded_size) + if sharded_param.numel() > 0: + padded_sharded_param.narrow( + dim=shard_dim, start=0, length=sharded_param.size(shard_dim) + ).copy_(sharded_param) + if self.offload_to_cpu and not padded_sharded_param.is_meta: + padded_sharded_param = padded_sharded_param.cpu() + if self.pin_memory: + padded_sharded_param = padded_sharded_param.pin_memory() + self._sharded_param_data = padded_sharded_param.view(-1) + length = sharded_param.size(shard_dim) if sharded_param.numel() > 0 else 0 + sharded_param = padded_sharded_param.narrow( + dim=shard_dim, start=0, length=length + ) + if not sharded_param.is_contiguous(): + raise AssertionError( + f"Expected contiguous tensor with {self.fsdp_placement=}" + ) + self.sharded_param = nn.Parameter( + self.to_sharded_dtensor(sharded_param), + requires_grad=param.requires_grad, + ) + # Let `param_data` be freed normally when its ref count reaches 0 when + # the `fully_shard` call returns to allow provided parameters to alias + self._setattr_on_modules(self.sharded_param) + self.sharded_state = ShardedState.SHARDED + + def _resolve_spmd_types_for_storage( + self, + param: nn.Parameter, + partition_spec: "spmd.PartitionSpec | None", + init_local_type: "spmd.LocalSpmdType", + mesh_info: DataParallelMeshInfo, + ) -> nn.Parameter: + """ + Resolve spmd_types metadata and translate it into FSDP storage. + + The original parameter is a plain tensor annotated for module compute + on the typechecking mesh. FSDP stores parameters as DTensors on + ``mesh_info.spmd_mesh``, so init time only needs enough type information + to translate into DTensor placements on the storage mesh axes. Axes + named by ``dp_mesh_dims`` are FSDP-managed DP axes and are treated as + ``spmd.R``; non-FSDP storage axes must have a direct spmd_types + annotation. + + The original compute annotations and PartitionSpec are saved. At + compute time, FSDP all-gathers the DTensor storage, exposes a plain + tensor parameter, and restores spmd_types metadata on the init-time + compute mesh. Missing compute axes are filled in as ``spmd.R``. + + Gradient storage placements are derived from the backward spmd_type. + FSDP-managed "DP" axes require forward ``spmd.R`` since FSDP owns + their gradient reduction. Consequently, the backward type is ``spmd.P``, + so the incoming plain grad is wrapped as a ``Partial()`` DTensor on those + axes, pending the gradient all-reduce/reduce-scatter to storage-time + placement. + + For non-FSDP axes (e.g. TP), annotations are flexible. For a parameter + annotated with ``spmd.I`` on TP, the grad is wrapped as ``Replicate()`` and no + reduction is pending. ``spmd.S(dim)`` stays ``Shard(dim)``. + + This follows tied FWD/BWD typing for spmd_types: ``R`` -> ``P``, + ``P`` -> ``R``, ``I`` -> ``I``, and ``S(dim)`` -> ``S(dim)``. + """ + storage_mesh = mesh_info.spmd_mesh + if storage_mesh is None or storage_mesh.mesh_dim_names is None: + raise ValueError( + "spmd_types parameters require fully_shard() to be called with both " + "a named full DeviceMesh for FSDP storage and " + "dp_mesh_dims=DataParallelMeshDims(shard=..., replicate=...), " + "where shard names the mesh axis or axes FSDP shards parameters " + "across and replicate names the HSDP/DDP mesh axis or axes FSDP " + "replicates across." + ) + + storage_mesh_dim_names = storage_mesh.mesh_dim_names + storage_mesh_axes = tuple( + (name, spmd.MeshAxis.of(storage_mesh.get_group(name))) + for name in storage_mesh_dim_names + ) + storage_axes = tuple(axis for _, axis in storage_mesh_axes) + local_type: spmd.PerMeshAxisSpmdTypes = dict(init_local_type) + if partition_spec is not None: + local_type.update( # pyrefly: ignore [no-matching-overload] + partition_spec_to_shard_types(partition_spec) + ) + annotated_axes = tuple(local_type.keys()) + + # resolve typechecking mesh on which parameters are annotated + if _spans_same_mesh(annotated_axes, storage_axes): + restore_mesh = annotated_axes + elif (current_mesh := spmd.current_mesh()) is not None: + restore_mesh = tuple(current_mesh) + else: + raise ValueError( + f"Parameter '{self._module_info.param_name}' has partial " + "spmd_types annotations that cannot be restored from its FSDP " + "storage mesh. Fully annotate the parameter on its storage " + "mesh, or wrap fully_shard() in spmd.set_current_mesh(...) " + "when all parameters in the FSDP unit share one typechecking " + "mesh. Mixed per-parameter typechecking meshes with partial " + "annotations are not supported. " + f"Annotated axes: {annotated_axes}. " + f"Storage mesh axes: {storage_axes}." + ) + + # Validate the typechecking mesh with FSDP storage mesh. + unknown_axes = tuple( + axis for axis in init_local_type if axis not in restore_mesh + ) + if unknown_axes: + raise ValueError( + f"FSDP found spmd_types annotations on parameter " + f"'{self._module_info.param_name}' and will restore runtime " + f"annotations on compute mesh axes {restore_mesh}, but the " + f"parameter is annotated on axes {unknown_axes} that are not in " + "that mesh. Annotate only axes in the compute mesh." + ) + if not _spans_same_mesh(restore_mesh, storage_axes): + raise ValueError( + f"Parameter '{self._module_info.param_name}' uses spmd_types " + "annotations on a typechecking mesh that does not span the " + "same ranks as its FSDP storage mesh. FSDP can fill omitted " + "FSDP-managed axes only when these meshes span the same " + f"rank set. Typechecking mesh axes: {restore_mesh}. " + f"Storage mesh axes: {storage_axes}." + ) + dp_names = set( + itertools.chain( + mesh_info.dp_mesh_dims.shard_names, # pyrefly: ignore [missing-attribute] + mesh_info.dp_mesh_dims.replicate_names, # pyrefly: ignore [missing-attribute] + ) + ) + fsdp_axis = flatten_axes( + tuple(axis for name, axis in storage_mesh_axes if name in dp_names) + ) + non_fsdp_storage_mesh_axes = { + axis for name, axis in storage_mesh_axes if name not in dp_names + } + storage_axis_types: spmd.PerMeshAxisSpmdTypes = { + axis: spmd.R for name, axis in storage_mesh_axes if name in dp_names + } + # Param annotations are either optional spmd.R on FSDP axes or explicit + # annotations on non-FSDP storage axes. + for axis, axis_type in local_type.items(): + # MeshAxis ordering implements the FSDP-submesh containment check. + if axis <= fsdp_axis: # pyrefly: ignore [unsupported-operation] + if axis_type is not spmd.R: + raise ValueError( + f"Expected spmd.R on FSDP DP axis {axis} for parameter " + f"'{self._module_info.param_name}' but got {axis_type}. " + "FSDP requires DP parameters to be R since it handles " + "the DP gradient reduction." + ) + else: + if axis not in non_fsdp_storage_mesh_axes: + raise ValueError( + f"Parameter '{self._module_info.param_name}' has spmd_types " + f"annotation on axis {axis}, which is neither a non-FSDP " + "storage-mesh axis nor contained in the FSDP DP mesh." + ) + storage_axis_types[axis] = axis_type + + if set(storage_axes) != set(storage_axis_types.keys()): + raise ValueError( + f"Parameter '{self._module_info.param_name}' has incomplete " + "spmd_types annotations for FSDP storage. " + f"Annotated axes: {tuple(local_type.keys())}. " + f"Storage mesh axes: {storage_axes}. " + f"FSDP mesh dims: {mesh_info.dp_mesh_dims}. " + "Missing non-FSDP storage axes: " + f"{tuple(axis for _, axis in storage_mesh_axes if axis not in storage_axis_types)}." + ) + + # Translate to DTensor FWD/BWD placements; R@FSDP axes are optional. + restore_type: spmd.PerMeshAxisSpmdTypes = { + axis: init_local_type.get(axis, spmd.R) for axis in restore_mesh + } + placements = [] + grad_placements = [] + for axis_type in storage_axis_types.values(): + placements.append(spmd.spmd_type_to_dtensor_placement(axis_type)) + grad_placements.append( + spmd.spmd_type_to_dtensor_placement(axis_type.backward_type()) + ) + dtensor_param = nn.Parameter( + DTensor.from_local(param.data, storage_mesh, placements, run_check=False), + requires_grad=param.requires_grad, + ) + + # Assign spmd_types state after validation succeeds. + self._spmd_partition_spec: spmd.PartitionSpec | None = partition_spec + self._spmd_init_local_type: spmd.LocalSpmdType = init_local_type + self._spmd_restore_mesh: tuple[Any, ...] = restore_mesh + self._spmd_restore_type: spmd.PerMeshAxisSpmdTypes = restore_type + self._spmd_grad_placements: tuple[Placement, ...] = tuple(grad_placements) + return dtensor_param + + def _restore_spmd_types(self, tensor: torch.Tensor) -> None: + """Restore the saved spmd_types annotation onto a tensor.""" + if not self.is_spmd_types or not spmd.is_type_checking(): + return + if self._spmd_restore_type: + spmd.assert_type( + tensor, + self._spmd_restore_type, + partition_spec=self._spmd_partition_spec, + ) + + def _init_sharding_spec( + self, + param: nn.Parameter, + fsdp_placement: Shard, + shard_dim: int, + ) -> torch.Tensor: + """ + Build ``_sharding_spec``, ``_spmd_mesh``, and ``_spmd_placements`` and + return the local tensor data to be sharded. + """ + self._unsharded_dtensor_spec = None + if self.mesh_info.is_spmd_mesh and not self.is_dtensor: + raise ValueError( + "When dp_mesh_dims is provided, all parameters must be " + "DTensors on the full SPMD mesh (e.g. via distribute_module). " + f"Got plain tensor for parameter '{self._module_info.param_name}'." + ) + if self.is_dtensor and self.mesh_info.is_spmd_mesh: + return self._init_sharding_spec_spmd(param, fsdp_placement, shard_dim) + if self.is_dtensor: + return self._init_sharding_spec_tp(param, fsdp_placement, shard_dim) + return self._init_sharding_spec_plain(param, fsdp_placement) + + def _init_sharding_spec_spmd( + self, + param: nn.Parameter, + fsdp_placement: Shard, + shard_dim: int, + ) -> torch.Tensor: + """SPMD path: param is a DTensor on the full SPMD mesh.""" + self._unsharded_dtensor_spec = cast(DTensor, param)._spec + spmd_mesh = self._unsharded_dtensor_spec.mesh + dp_dim_names = self.mesh_info.dp_mesh_dims + if dp_dim_names is None: + raise AssertionError("dp_dim_names must not be None for SPMD mesh") + if spmd_mesh.mesh_dim_names is None: + raise AssertionError("spmd_mesh.mesh_dim_names must not be None") + if ( + self.mesh_info.spmd_mesh is not None + and spmd_mesh is not self.mesh_info.spmd_mesh + ): + raise ValueError( + "Expected param's DTensor mesh to be the same mesh passed " + "to fully_shard, but got different mesh objects" + ) + + dp_shard_indices = [ + spmd_mesh.mesh_dim_names.index(n) for n in dp_dim_names.shard_names + ] + + orig_placements = self._unsharded_dtensor_spec.placements + for idx in dp_shard_indices: + if not isinstance(orig_placements[idx], Replicate): + raise ValueError( + f"Expected Replicate() on DP shard dim " + f"'{spmd_mesh.mesh_dim_names[idx]}' (index {idx}) " + f"but got {orig_placements[idx]}" + ) + dp_replicate_indices = [] + for rep_name in dp_dim_names.replicate_names: + rep_idx = spmd_mesh.mesh_dim_names.index(rep_name) + dp_replicate_indices.append(rep_idx) + if not isinstance(orig_placements[rep_idx], Replicate): + raise ValueError( + f"Expected Replicate() on DP replicate dim " + f"'{spmd_mesh.mesh_dim_names[rep_idx]}' (index {rep_idx}) " + f"but got {orig_placements[rep_idx]}" + ) + + # Cache DP dim indices so _get_grad_inner_tensor can skip + # redistribution on DP dims and let FSDP's reduce-scatter handle them. + self._dp_dim_indices: frozenset[int] = frozenset( + dp_shard_indices + dp_replicate_indices + ) + + new_placements = list(orig_placements) + for dp_idx in dp_shard_indices: + # split_factor = number of non-DP shards on shard_dim from + # mesh dims with higher index (the "right-side" dims that + # _StridedShard needs to interleave with) + sf = 1 + for j in range(dp_idx + 1, spmd_mesh.ndim): + p = orig_placements[j] + if isinstance(p, (Shard, _StridedShard)) and p.dim == shard_dim: + sf *= spmd_mesh.size(j) + new_placements[dp_idx] = ( + _StridedShard(shard_dim, split_factor=sf) if sf > 1 else fsdp_placement + ) + + self._spmd_mesh = spmd_mesh + self._spmd_placements: tuple[Placement, ...] = tuple(new_placements) + self._sharding_spec = self._build_spmd_sharding_spec( + dp_dim_names, + dp_shard_indices, + fsdp_placement, + ) + return cast(DTensor, param)._local_tensor + + def _build_spmd_sharding_spec( + self, + dp_dim_names: "DataParallelMeshDims", + dp_shard_indices: list[int], + fsdp_placement: Shard, + ) -> DTensorSpec: + """Build the DTensorSpec for the sharded parameter/gradient. + + When multiple DP shard dims exist (e.g. dp_shard + cp), flatten + them into one axis so the parameter and gradient DTensor have one + Shard axis for DP. + """ + spmd_mesh = self._spmd_mesh + spmd_placements = self._spmd_placements + if self._unsharded_dtensor_spec is None: + raise AssertionError("_unsharded_dtensor_spec cannot be None") + tensor_meta = self._unsharded_dtensor_spec.tensor_meta + + if len(dp_shard_indices) <= 1: + return DTensorSpec(spmd_mesh, spmd_placements, tensor_meta=tensor_meta) + + shard_names_set = set(dp_dim_names.shard_names) + replicate_names_set = set(dp_dim_names.replicate_names) + + # Walk spmd_mesh.mesh_dim_names in order. Replace consecutive DP + # shard dims with the flattened DP mesh; keep non-DP dims as-is. + if spmd_mesh.mesh_dim_names is None: + raise AssertionError("mesh_dim_names cannot be None") + submeshes: list[DeviceMesh] = [] + spec_placements: list[Placement] = [] + skip = 0 + for i, name in enumerate(spmd_mesh.mesh_dim_names): + if skip > 0: + skip -= 1 + continue + if name in shard_names_set: + submeshes.append(self.mesh_info.mesh) + if isinstance(self.mesh_info, HSDPMeshInfo): + spec_placements.append(Replicate()) + # Reuse the placement already computed for this DP shard dim + # so that we don't loss _StridedShard. + spec_placements.append(spmd_placements[i]) + skip = len(dp_dim_names.shard_names) - 1 + elif name in replicate_names_set and isinstance( + self.mesh_info, HSDPMeshInfo + ): + # HSDP replicate is inserted by the shard branch above. + continue + else: + submeshes.append(spmd_mesh[name]) + spec_placements.append(spmd_placements[i]) + + spec_mesh = DeviceMesh._concatenate(submeshes) + return DTensorSpec(spec_mesh, tuple(spec_placements), tensor_meta=tensor_meta) + + def _init_sharding_spec_tp( + self, + param: nn.Parameter, + fsdp_placement: Shard, + shard_dim: int, + ) -> torch.Tensor: + """TP/EP path: param is a DTensor, DP mesh is separate from TP mesh.""" + self._unsharded_dtensor_spec = cast(DTensor, param)._spec + dp_mesh, tp_mesh = (self.mesh_info.mesh, self._unsharded_dtensor_spec.mesh) + if dp_mesh is None or tp_mesh is None: + raise AssertionError( + "FSDP requires the DP and model parallel TP/EP mesh to be not None but got: \n" + f"DP's mesh: {dp_mesh}\nTP/EP's mesh: {tp_mesh}" + ) + self._spmd_mesh = DeviceMesh._concatenate([dp_mesh, tp_mesh]) + if len(self._unsharded_dtensor_spec.placements) > 2: + raise NotImplementedError( + f"FSDP only supports 1D TP/EP or 2D EP+TP, not {self._unsharded_dtensor_spec.placements}" + ) + split_factor = self._unsharded_dtensor_spec.num_shards_map[shard_dim] + if not (2 <= self._spmd_mesh.ndim <= 4): + raise AssertionError( + "_spmd_mesh.ndim can only be 2 (FSDP+TP/EP), 3 (FSDP+EP+TP, HSDP+TP/EP), " + f"or 4 (HSDP+EP+TP) but got {self._spmd_mesh.ndim}." + ) + if isinstance(self.mesh_info, FSDPMeshInfo): + dp_shard_tp_placement = ( + ( + _StridedShard(shard_dim, split_factor=split_factor) + if split_factor > 1 + else fsdp_placement + ), + *self._unsharded_dtensor_spec.placements, + ) + else: # DDP + dp_shard_tp_placement = ( + Replicate(), + *self._unsharded_dtensor_spec.placements, + ) + self._spmd_placements: tuple[Placement, ...] + if isinstance(self.mesh_info, HSDPMeshInfo): + if self.mesh_info.replicate_mesh_dim != 0: + raise AssertionError( + f"Expected replicate_mesh_dim to be 0, got {self.mesh_info.replicate_mesh_dim}" + ) + self._spmd_placements = (Replicate(),) + dp_shard_tp_placement + else: + self._spmd_placements = dp_shard_tp_placement + + self._sharding_spec = DTensorSpec( + self._spmd_mesh, + self._spmd_placements, + tensor_meta=self._unsharded_dtensor_spec.tensor_meta, + ) + return cast(DTensor, param)._local_tensor + + def _init_sharding_spec_plain( + self, + param: nn.Parameter, + fsdp_placement: Shard, + ) -> torch.Tensor: + """Plain tensor path: param is not a DTensor.""" + self._spmd_mesh = self.mesh_info.mesh + if isinstance(self.mesh_info, HSDPMeshInfo): + self._spmd_placements = (Replicate(), fsdp_placement) + elif isinstance(self.mesh_info, FSDPMeshInfo): + self._spmd_placements = (fsdp_placement,) + elif isinstance(self.mesh_info, DDPMeshInfo): + self._spmd_placements = (Replicate(),) + self._sharding_spec = DTensorSpec( + self._spmd_mesh, + self._spmd_placements, + tensor_meta=TensorMeta(param.size(), param.stride(), param.dtype), + ) + return param + + def _init_sharded_post_forward_param_metadata(self, param: torch.Tensor) -> None: + mesh_info = self.post_forward_mesh_info + if mesh_info is None: + raise AssertionError("Expected post_forward_mesh_info to not be None") + param_data = param._local_tensor if isinstance(param, DTensor) else param + if isinstance(mesh_info, FSDPMeshInfo): + chunks = _chunk_with_empty(param_data, mesh_info.shard_mesh_size, dim=0) + self.sharded_post_forward_size = _get_dim_chunked_size( + chunks[mesh_info.shard_mesh_rank], + param_data.size(), + dim=self.fsdp_placement.dim, + ) + else: # DDP + chunks = _chunk_with_empty(param_data, 1, dim=0) + self.sharded_post_forward_size = _get_dim_chunked_size( + chunks[0], + param_data.size(), + dim=self.fsdp_placement.dim, + ) + self.contiguous_sharded_post_forward_stride = make_contiguous_strides_for( + self.sharded_post_forward_size + ) + + def init_dtype_attrs(self, mp_policy: MixedPrecisionPolicy): + param_dtype, reduce_dtype = (mp_policy.param_dtype, mp_policy.reduce_dtype) + self.orig_dtype = self.sharded_param.dtype + # Clamp `reduce_dtype` to `None` if no casting is required: since + # gradients are computed in `param_dtype`, if `reduce_dtype` matches, + # then we do not need extra casting + if reduce_dtype == param_dtype: + reduce_dtype = None + # Clamp `param_dtype` to `None` if no casting is required or if the + # parameter is non-floating-point (mixed precision is only meaningful + # for floating-point parameters) + if param_dtype == self.orig_dtype or not self.orig_dtype.is_floating_point: + param_dtype = None + self.param_dtype = param_dtype + self.reduce_dtype = reduce_dtype + # None indicates that the mixed precision is not enabled + + def _init_extensions(self) -> None: + inner_tensor = self._sharded_local_tensor + has_fsdp_pre_all_gather = hasattr(inner_tensor, "fsdp_pre_all_gather") + has_fsdp_post_all_gather = hasattr(inner_tensor, "fsdp_post_all_gather") + if has_fsdp_pre_all_gather != has_fsdp_post_all_gather: + raise AssertionError( + "Both fsdp_pre_all_gather and fsdp_post_all_gather should be defined " + f"if using all-gather extensions: {inner_tensor}" + ) + if has_fsdp_pre_all_gather: + self._extensions_data = ExtensionsData() + self._unsharded_inner_tensors: list[torch.Tensor] = [] + + def init_all_gather_outputs( + self, + all_gather_input_numels: list[int], + all_gather_input_dtypes: list[torch.dtype], + world_size: int, + device: torch.device, + ): + if len(self.all_gather_outputs) > 0: + return # already initialized + self.all_gather_outputs = [ + torch.empty(torch.Size([numel * world_size]), dtype=dtype, device=device) + for numel, dtype in zip(all_gather_input_numels, all_gather_input_dtypes) + ] + + def init_unsharded_param(self): + if hasattr(self, "_unsharded_param"): # after the 1st all-gather + inner_tensor = self._sharded_local_tensor + if not hasattr(inner_tensor, "fsdp_post_all_gather"): + return # already initialized + for tensor in self._unsharded_inner_tensors: + alloc_storage(tensor) + all_gather_outputs = self._unflatten_all_gather_outputs() + inner_tensor.fsdp_post_all_gather( + all_gather_outputs, + self._extensions_data.all_gather_metadata, + self.param_dtype or self.orig_dtype, + out=self._unsharded_param, + ) + self._extensions_data.clear() + return + inner_tensor = self._sharded_local_tensor + if hasattr(inner_tensor, "fsdp_post_all_gather"): + all_gather_outputs = self._unflatten_all_gather_outputs() + ( + unsharded_tensor, + self._unsharded_inner_tensors, + ) = inner_tensor.fsdp_post_all_gather( + all_gather_outputs, + self._extensions_data.all_gather_metadata, + self.param_dtype or self.orig_dtype, + ) + self._extensions_data.clear() + else: + # For the default path (no post-all-gather), the all-gather output + # gives the unsharded parameter data directly + if len(self.all_gather_outputs) != 1: + raise AssertionError( + f"Expected 1 all_gather_output, got {len(self.all_gather_outputs)}" + ) + unsharded_tensor = self.all_gather_outputs[0] + unsharded_param = torch.as_strided( + unsharded_tensor, + self._orig_size, + self._contiguous_orig_stride, + storage_offset=0, + ) + if self.is_spmd_types: + pass # keep as plain tensor; spmd_types restored before module compute + elif self._unsharded_dtensor_spec is not None: + unsharded_dtensor_spec = self._get_unsharded_dtensor_spec(unsharded_param) + unsharded_param = _from_local_no_grad( + unsharded_param, unsharded_dtensor_spec + ) + self._unsharded_param = nn.Parameter( + unsharded_param, requires_grad=self.sharded_param.requires_grad + ) + + def _get_unsharded_dtensor_spec(self, unsharded_param: torch.Tensor) -> DTensorSpec: + if self._unsharded_dtensor_spec is None: + raise AssertionError("Expected _unsharded_dtensor_spec for DTensor param") + tensor_meta = self._unsharded_dtensor_spec.tensor_meta + if tensor_meta is None or tensor_meta.dtype == unsharded_param.dtype: + return self._unsharded_dtensor_spec + return replace( + self._unsharded_dtensor_spec, + tensor_meta=TensorMeta( + tensor_meta.shape, + tensor_meta.stride, + unsharded_param.dtype, + ), + ) + + def _unflatten_all_gather_outputs(self) -> tuple[torch.Tensor, ...]: + return tuple( + t.view(-1, *s[1:]) + for t, s in zip( + self.all_gather_outputs, self._extensions_data.all_gather_input_sizes + ) + ) + + def to_sharded(self) -> None: + self._setattr_on_modules(self.sharded_param) + self.free_unsharded_param() + self.sharded_state = ShardedState.SHARDED + + def to_sharded_post_forward(self) -> None: + if self.is_dtensor: + raise NotImplementedError( + "Resharding to smaller mesh is not supported for DTensor parameters yet" + ) + self._assert_in_states(ShardedState.UNSHARDED) + if self.post_forward_mesh_info is None: + raise AssertionError("Expected post_forward_mesh_info to not be None") + if len(self.all_gather_outputs) != 1: + raise AssertionError( + f"Expected 1 all_gather_output, got {len(self.all_gather_outputs)}" + ) + shard_world_size = self.post_forward_mesh_info.shard_mesh_size + if (numel := self.all_gather_outputs[0].numel()) % shard_world_size != 0: + _raise_assert_with_print( + f"All-gather output size ({numel}) must be divisible by the shard " + f"world size ({shard_world_size})" + ) + shard_rank = self.post_forward_mesh_info.shard_mesh_rank + sharded_numel = numel // shard_world_size + self._sharded_post_forward_param_data = ( + self.all_gather_outputs[0].narrow( + 0, sharded_numel * shard_rank, sharded_numel + ) + ).clone() # clone to be able to free all-gather output + sharded_post_forward_tensor = torch.as_strided( + self._sharded_post_forward_param_data, + size=self.sharded_post_forward_size, + stride=self.contiguous_sharded_post_forward_stride, + storage_offset=0, + ) + self._sharded_post_forward_param = nn.Parameter( + self.to_sharded_post_forward_dtensor(sharded_post_forward_tensor), + requires_grad=self.sharded_param.requires_grad, + ) + self._setattr_on_modules(self._sharded_post_forward_param) + self.free_unsharded_param() + self.sharded_state = ShardedState.SHARDED_POST_FORWARD + + def to_unsharded(self) -> None: + # Assume that the data has been allocated and all-gathered + set_requires_grad_if_needed(self.sharded_param, self._unsharded_param) + self._setattr_on_modules(self._unsharded_param) + if self.sharded_state == ShardedState.SHARDED_POST_FORWARD: + # The data is allocated in the default stream via the post-forward + # reshard and must be kept alive for the next all-gather copy-in. + # Since we call this method after the copy-out, the data's lifetime + # is ensured without further synchronization. + self._sharded_post_forward_param = None + self._sharded_post_forward_param_data = None # free + self.sharded_state = ShardedState.UNSHARDED + + def _setattr_on_modules(self, param: nn.Parameter) -> None: + unsafe_setattr_param( + self._module_info.module, self._module_info.param_name, param + ) + for shared_module, shared_param_name in zip( + self._module_info.shared_modules, self._module_info.shared_param_names + ): + unsafe_setattr_param(shared_module, shared_param_name, param) + + def to_sharded_dtensor(self, tensor: torch.Tensor) -> DTensor: + """ + Converts a local tensor representing either the sharded parameter or + sharded gradient to DTensor. + """ + if tensor.shape != self.sharded_size: + _raise_assert_with_print( + f"Expects size {self.sharded_size} but got {tensor.shape}" + ) + return _from_local_no_grad( + tensor, + self._sharding_spec, + ) + + def to_sharded_post_forward_dtensor(self, tensor: torch.Tensor) -> DTensor: + if tensor.shape != self.sharded_post_forward_size: + _raise_assert_with_print( + f"Expects size {self.sharded_post_forward_size} but got {tensor.shape}" + ) + if not isinstance(self.post_forward_mesh_info, HSDPMeshInfo): + raise AssertionError( + f"Expected HSDPMeshInfo, got {type(self.post_forward_mesh_info)}" + ) + # TODO: Prefer this DTensor to be read-only and generalize the + # placement once we support TP. + post_forward_sharding_spec = DTensorSpec( + self.post_forward_mesh_info.mesh, + (Replicate(), Shard(0)), + tensor_meta=self._sharding_spec.tensor_meta, + ) + return _from_local_no_grad(tensor, post_forward_sharding_spec) + + def to_accumulated_grad_if_needed(self) -> None: + # Access `_unsharded_param` to bypass the sharded state check since we + # prefer to reshard before upcasting the gradient to save memory + if ( + self.reduce_dtype is None + or self._unsharded_param.grad is None + or self._unsharded_param.grad.dtype == self.reduce_dtype + ): + return + unsharded_grad = self._unsharded_param.grad + self._unsharded_param.grad = None + self.unsharded_accumulated_grad = unsharded_grad.to(self.reduce_dtype) + + def accumulate_unsharded_grad_if_needed(self) -> None: + if ( + self.unsharded_accumulated_grad is not None + and self.unsharded_param.grad is not None + ): + self.unsharded_accumulated_grad += self.unsharded_param.grad + self.unsharded_param.grad = None + + def alloc_all_gather_outputs(self) -> None: + for tensor in self.all_gather_outputs: + alloc_storage(tensor) + + def free_unsharded_param(self) -> None: + for tensor in itertools.chain( + self.all_gather_outputs, self._unsharded_inner_tensors + ): + free_storage(tensor) + + @property + def all_gather_inputs(self) -> list[torch.Tensor]: # 1D + self._assert_in_states(ShardedState.SHARDED, ShardedState.SHARDED_POST_FORWARD) + if self.sharded_state == ShardedState.SHARDED: + if hasattr(self._sharded_local_tensor, "fsdp_pre_all_gather"): + sharded_local_tensor = self._sharded_local_tensor + if self.offload_to_cpu: + sharded_local_tensor = sharded_local_tensor.to( + self.device, non_blocking=True + ) + pre_all_gather_signature = inspect.signature( + # pyrefly: ignore [missing-attribute] + sharded_local_tensor.fsdp_pre_all_gather + ) + num_fn_params = len(pre_all_gather_signature.parameters) + # Old signature only passes mesh; keep for BC for now + if num_fn_params not in (1, 5): + raise AssertionError( + f"Invalid fsdp_pre_all_gather: {pre_all_gather_signature}\n" + "Expects fsdp_pre_all_gather(self, mesh: DeviceMesh, " + "outer_size: torch.Size, outer_stride: tuple[int, ...], " + "module: nn.Module, mp_policy: MixedPrecisionPolicy)" + ) + if num_fn_params == 1: + ( + all_gather_inputs, + self._extensions_data.all_gather_metadata, + # pyrefly: ignore [missing-attribute] + ) = sharded_local_tensor.fsdp_pre_all_gather( + self.shard_mesh_from_root + ) + else: + ( + all_gather_inputs, + self._extensions_data.all_gather_metadata, + # pyrefly: ignore [missing-attribute] + ) = sharded_local_tensor.fsdp_pre_all_gather( + self.shard_mesh_from_root, + self._orig_size, + self._contiguous_orig_stride, + self._module_info.module, + self.mp_policy, + ) + if ( + sharded_local_tensor.size() != self.padded_sharded_param_size + and any( + all_gather_input.size() != self.padded_sharded_param_size + for all_gather_input in all_gather_inputs + ) + ): + # NOTE: Since this error can only be raised on the + # ranks that have padding, this can manifest as a NCCL + # watchdog timeout, as the other ranks will not error. + raise AssertionError( + "When a parameter is unevenly sharded by FSDP " + f"(orig size={self._orig_size}, FSDP world size={self.mesh_info.mesh.size()}), " + "fsdp_pre_all_gather must return all-gather inputs with the padded sharded size " + f"{self.padded_sharded_param_size} but got {[t.size() for t in all_gather_inputs]}" + ) + self._extensions_data.all_gather_input_sizes = [ + t.size() for t in all_gather_inputs + ] + return [t.view(-1) for t in all_gather_inputs] + sharded_param_data = self._sharded_param_data + if self.offload_to_cpu: + sharded_param_data = sharded_param_data.to( + self.device, non_blocking=True + ) + return [_to_dtype_if_needed(sharded_param_data, self.param_dtype)] + elif self.sharded_state == ShardedState.SHARDED_POST_FORWARD: + if hasattr(self._sharded_local_tensor, "fsdp_pre_all_gather"): + raise NotImplementedError + all_gather_input = _to_dtype_if_needed( + cast(torch.Tensor, self._sharded_post_forward_param_data), + self.param_dtype, + ) + return [all_gather_input] + return [torch.empty(0)] # mypy + + @property + def unsharded_param(self) -> nn.Parameter: # ND + return self._unsharded_param + + @property + def unsharded_grad_data(self) -> torch.Tensor: + grad = self.unsharded_param.grad + if grad is None: + raise AssertionError("Expects unsharded_param.grad to not be None") + return self._get_grad_inner_tensor(grad) + + @property + def unsharded_accumulated_grad_data(self) -> torch.Tensor: + grad = self.unsharded_accumulated_grad + if grad is None: + raise AssertionError("Expects unsharded_accumulated_grad to not be None") + return self._get_grad_inner_tensor(grad) + + @property + def unsharded_zero_grad_data(self) -> torch.Tensor: + return self._get_grad_inner_tensor(torch.zeros_like(self.unsharded_param)) + + def _get_grad_inner_tensor(self, grad: torch.Tensor) -> torch.Tensor: + if self.is_spmd_types: + if self._unsharded_dtensor_spec is None: + raise AssertionError( + "Expected _unsharded_dtensor_spec for spmd_types param" + ) + grad = DTensor.from_local( + grad, + self._unsharded_dtensor_spec.mesh, + tuple(self._spmd_grad_placements), + run_check=False, + ) + if not self.is_dtensor: + raise AssertionError( + "Expected spmd_types -> DTensor wrapping to use the DTensor gradient " + "path for correct gradient redistribution." + ) + + if self.is_dtensor: + if isinstance(grad, AsyncCollectiveTensor): + grad = grad.wait() + if not isinstance(grad, DTensor): + raise AssertionError(f"Expected DTensor, got {type(grad)}") + if self._unsharded_dtensor_spec is None: + raise AssertionError( + "Expected _unsharded_dtensor_spec for DTensor param" + ) + placements = self._unsharded_dtensor_spec.placements + if self.mesh_info.is_spmd_mesh: + # Only redistribute non-DP dims; keep Partial on DP dims + # so FSDP's reduce-scatter handles them directly, avoiding + # a redundant all-reduce on the DP dimensions. + target_placements = tuple( + grad.placements[i] if i in self._dp_dim_indices else placements[i] + for i in range(len(placements)) + ) + if target_placements != grad.placements: + if len(placements) != len(grad.placements): + raise AssertionError( + f"Expected same placement length: {placements=} {grad.placements=}" + ) + grad = grad.redistribute(placements=target_placements) + else: + if placements != grad.placements: + if len(placements) != len(grad.placements): + raise AssertionError( + f"Expected same placement length: {placements=} {grad.placements=}" + ) + grad = grad.redistribute(placements=placements) + grad = grad._local_tensor + return grad + + @property + def _sharded_local_tensor(self) -> torch.Tensor: + return cast(DTensor, self.sharded_param)._local_tensor + + def _init_shard_mesh(self) -> DeviceMesh: + mesh = self.mesh_info.mesh + if mesh.ndim == 1: + return mesh + if mesh.mesh_dim_names is None: + raise AssertionError("Expected mesh_dim_names to not be None") + return mesh[mesh.mesh_dim_names[-1]] + + @property + def shard_mesh(self): + return self._shard_mesh + + @property + def shard_mesh_from_root(self): + return self.shard_mesh + + def _assert_in_states(self, *states: ShardedState) -> None: + if self.sharded_state not in states: + _raise_assert_with_print( + f"Expects to be in one of {states}, not {self.sharded_state}" + ) + + def reset_sharded_param(self): + # For ops like `nn.Module._apply` or `load_state_dict(assign=True)` + # that change the sharded parameter tensor, we may need to re-pad the + # sharded local tensor and re-save the reference. + module_info = self._module_info + new_param = getattr(module_info.module, module_info.param_name) + if new_param is not self.sharded_param: + if torch.__future__.get_swap_module_params_on_conversion(): + raise AssertionError( + f"Expects swap_tensors to preserve object but got {new_param} " + f"instead of {self.sharded_param}" + ) + self.sharded_param = new_param + + local_tensor = new_param._local_tensor + if local_tensor.is_meta: + return + updated_local_tensor = False + # local_tensor can be padded twice + # 1st time in fully_shard(model) + # 2nd time in model(input) lazy_init + # 2nd time should be no-op if parameters remain unchanged + # 2nd time shouldn't be no-op if people call model.load_state_dict(...) before lazy_init + # this makes it possible for trainer to call `sd = model.state_dict()` before the training loop + # and use `sd` without calling .state_dict() per iteration + same_local_tensor = False + # TODO: need to support tensor subclass + if type(self._sharded_param_data) is torch.Tensor: + same_local_tensor = ( + # when sharding param with shape (1, ...) over 2 ranks + # local_tensor on rank 1 can be size 0, data_ptr() can be 0 + self._sharded_param_data.untyped_storage().data_ptr() > 0 + and self._sharded_param_data.untyped_storage().data_ptr() + == local_tensor.untyped_storage().data_ptr() + ) + padded_sharded_size = self.padded_sharded_param_size + shard_dim = self.fsdp_placement.dim + length = local_tensor.size(shard_dim) if local_tensor.numel() > 0 else 0 + if local_tensor.size() != padded_sharded_size and not same_local_tensor: + if shard_dim != 0: + raise AssertionError( + f"Shard({shard_dim}) requires even sharding: {local_tensor.size()=}" + ) + padded_local_tensor = local_tensor.new_zeros(padded_sharded_size) + padded_local_tensor.narrow(dim=shard_dim, start=0, length=length).copy_( + local_tensor + ) + local_tensor = padded_local_tensor + updated_local_tensor = True + if self.pin_memory and not local_tensor.is_pinned(): + local_tensor = local_tensor.cpu().pin_memory() + updated_local_tensor = True + if not same_local_tensor: + self._sharded_param_data = local_tensor.view(-1) + if not isinstance(self.sharded_param, DTensor): + raise AssertionError(f"Expected DTensor, got {type(self.sharded_param)}") + if updated_local_tensor: + # Only change the local tensor object if needed + self.sharded_param._local_tensor = local_tensor.narrow( + dim=shard_dim, start=0, length=length + ) + if not self.sharded_param._local_tensor.is_contiguous(): + raise AssertionError( + "Expected sharded_param._local_tensor to be contiguous" + ) + self._sharding_spec = self.sharded_param._spec + + def __repr__(self): + return f"FSDPParam(fqn={self._param_fqn}, orig_size={self._orig_size})" + + +def alloc_storage(tensor: torch.Tensor) -> None: + size = tensor.numel() * tensor.itemsize + if (storage := tensor.untyped_storage()).size() != size: + storage.resize_(size) + + +def free_storage(tensor: torch.Tensor) -> None: + if (storage := tensor.untyped_storage()).size() != 0: + storage.resize_(0) + + +# NOTE: These bypass `nn.Module.__setattr__` checks, which incur non-trivial +# CPU overhead, if the module did not override it. For FSDP, we know we do not +# need those checks when transitioning between sharded/unsharded parameters. +def unsafe_setattr_param( + module: nn.Module, param_name: str, param: nn.Parameter +) -> None: + if getattr(module.__setattr__, "__func__", None) is nn.Module.__setattr__: + module._parameters[param_name] = param + else: # slow path + setattr(module, param_name, param) + + +def set_requires_grad_if_needed( + src_tensor: torch.Tensor, dst_tensor: torch.Tensor +) -> None: + # Only call `requires_grad_` if needed to avoid the Python <> C++ context + # switch overhead + if src_tensor.requires_grad != dst_tensor.requires_grad: + dst_tensor.requires_grad_(src_tensor.requires_grad) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param_group.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param_group.py new file mode 100644 index 0000000000000000000000000000000000000000..c719d1250586bf691abebd5f09dc751a6a9c4fcf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_param_group.py @@ -0,0 +1,1175 @@ +# mypy: allow-untyped-defs +from __future__ import annotations + +import contextlib +import logging +from typing import Any, cast, Literal, NamedTuple, TYPE_CHECKING +from typing_extensions import TypeVarTuple, Unpack + +import torch +import torch.distributed as dist +import torch.nn as nn +from torch.distributed import _spmd_no_typecheck +from torch.distributed.device_mesh import _get_device_handle +from torch.distributed.fsdp._common_utils import ( + _named_parameters_with_duplicates, + collect_grad_tensors, + replace_grad_tensors, +) +from torch.profiler import record_function +from torch.utils.hooks import RemovableHandle + +from ._fsdp_api import CPUOffloadPolicy, MixedPrecisionPolicy, OffloadPolicy +from ._fsdp_collectives import ( + AllGather, + AllGatherResult, + DefaultAllGather, + DefaultReduceScatter, + foreach_all_gather, + foreach_all_gather_copy_out, + foreach_reduce, + ProcessGroupAllocAllGather, + ProcessGroupAllocReduceScatter, + ReduceScatter, + SymmMemAllGather, + SymmMemReduceScatter, +) +from ._fsdp_common import ( + _disable_functorch_if_active, + _dynamo_disable, + DataParallelMeshInfo, + DDPMeshInfo, + FSDPMeshInfo, + HSDPMeshInfo, + is_bw, + ShardPlacementFnResult, + TrainingState, +) +from ._fsdp_param import alloc_storage, FSDPParam, ParamModuleInfo, ShardedState + + +if TYPE_CHECKING: + from collections.abc import Callable + + +logger = logging.getLogger("torch.distributed.fsdp.fully_shard") + +_ModuleToHandleDict = dict[nn.Module, RemovableHandle] # for state dict +_GradInputs = TypeVarTuple("_GradInputs") + + +""" +[Note: Overlapping all-gather copy-in and all-gather] +For implicit forward prefetching, we want to overlap the next copy-in with the +current all-gather. We do so using a separate copy-in stream. However, since +we have the all-gather input as a view into the output, we must make sure to +copy into different memory from the current all-gather's output. Thus, we keep +a reference to the current all-gather's output and have the next FSDP parameter +group free it after its copy-in. Finally, we have the last FSDP state flush the +reference to avoid holding onto memory after forward. +""" + + +class FSDPCommContext: + """This has the communication state shared across FSDP states/parameter groups.""" + + def lazy_init(self, device: torch.device): + self.device_handle = _get_device_handle(device.type) + # Setting the all-gather/reduce-scatter streams to be higher priority + # can help avoid some issues where their copies in/out are delayed and + # block computation (this is different from high-pri NCCL streams) + high_priority = -1 + # All-gather state and copy-in stream allow overlapping the next + # copy-in with the current all-gather in forward; copy-in overlaps with + # reduce-scatter in backward without the separate copy-in stream + self.all_gather_copy_in_stream = self.device_handle.Stream( + priority=high_priority + ) + # All-gather stream allows overlapping next all-gather with current + # forward compute + self.all_gather_stream = self.device_handle.Stream(priority=high_priority) + # Reduce-scatter stream gives separate execution "thread" for post- + # backward logic like pre/post-gradient division and reduce-scatter + self.reduce_scatter_stream = self.device_handle.Stream(priority=high_priority) + # The most recent post-reduce event across all param groups, recorded + # on reduce_scatter_stream (FSDP) or all_reduce_stream (HSDP). Since + # later events subsume earlier ones on the same stream, a single wait + # on the last event recorded on each stream used in the post backward + # hook covers all groups. + self._last_post_reduce_events: dict[torch.Stream, torch.Event] = dict() + # Run the HSDP all-reduces concurrently with all-gather/reduce-scatter + # since collectives use different network resources and can overlap + # in the typical intra-node sharding / inter-node replication case + self.all_reduce_stream = self.device_handle.Stream() + # All-gather/reduce-scatter states keep references to collective + # tensors produced in one stream and used in another and accompanying + # CUDA events for synchronization + self.all_gather_state: AllGatherState | None = None + self.reduce_scatter_states: list[ReduceScatterState] = [] + # Effective cap on retained reduce_scatter_states, resolved from the + # per-group reduce_scatter_max_input_buffers in + # FSDPState._init_shared_state. It lives here, not per group, because + # reduce_scatter_states is a single shared list governed by one cap. + self.reduce_scatter_max_input_buffers: int = 1 + # Post-forward order for explicit backward prefetching + self.post_forward_order: list[FSDPParamGroup] = [] # will cause ref cycles + + def get_all_gather_streams( + self, async_op: bool, training_state: TrainingState + ) -> tuple[torch.Stream, torch.Stream]: + if not async_op and training_state in ( + TrainingState.FORWARD, + TrainingState.PRE_BACKWARD, + ): + # Use separate streams for implicit prefetching + return self.all_gather_copy_in_stream, self.all_gather_stream + current_stream = self.device_handle.current_stream() + return current_stream, current_stream + + +# See [Note: Overlapping all-gather copy-in and all-gather] +class AllGatherState(NamedTuple): + all_gather_result: AllGatherResult + event: torch.Event | None # all-gather copy-out + + +class ReduceScatterState(NamedTuple): + reduce_scatter_input: torch.Tensor + event: torch.Event | None # reduce-scatter event + + +class AllReduceState(NamedTuple): + # Holding all_reduce_input (the reduce-dtype AR buffer) keeps the + # caching allocator from reusing the block across layers. This is a + # structural invariant, not bookkeeping: without it, the next layer's + # RS can reuse the same physical block before this layer's AR finishes + # under slow AR, causing gradient aliasing. See PR #140044, PR #180900. + all_reduce_input: torch.Tensor + event: torch.Event | None # all-reduce event + + +class FSDPParamGroup: + """This class represents a parameter group to communicate together.""" + + _orig_dtype: torch.dtype | None + _reduce_dtype: torch.dtype | None + + def __init__( + self, + params: list[nn.Parameter], + modules: tuple[nn.Module, ...], + mesh_info: DataParallelMeshInfo, + post_forward_mesh_info: FSDPMeshInfo | None, + device: torch.device, + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None, + mp_policy: MixedPrecisionPolicy, + offload_policy: OffloadPolicy, + ): + self.modules = modules # permit ref cycle because 1:1 lifetime + param_module_infos = _get_param_module_infos(params, modules) + + self.fsdp_params = [ + FSDPParam( + param, + module_info, + mesh_info, + post_forward_mesh_info, + device, + shard_placement_fn, + mp_policy, + offload_policy, + ) + for param, module_info in zip(params, param_module_infos) + ] + self.mesh_info = mesh_info + self.post_forward_mesh_info = post_forward_mesh_info + self.device = device + self.device_handle = _get_device_handle(device.type) + self.mp_policy = mp_policy + self.offload_policy = offload_policy + self._training_state = TrainingState.IDLE + # Group's sharded state always matches its parameters' sharded states + self._sharded_state = ShardedState.SHARDED + self._module_fqn: str | None = None # prefixed from root module + # Only consider resetting sharded parameters once in lazy init since it + # can incur nontrivial overhead to reset them + self._reset_sharded_params: bool = False + + # - Hook state + self._module_to_pre_save_state_dict_hook_handle: _ModuleToHandleDict = {} + self._module_to_pre_load_state_dict_hook_handle: _ModuleToHandleDict = {} + self._all_reduce_hook: Callable[[torch.Tensor], None] | None = None + self._all_gather_comm: AllGather = DefaultAllGather() + self._all_gather_output = torch.empty(0, device=self.device) + self._reduce_scatter_comm: ReduceScatter = DefaultReduceScatter() + # Optional stream to run the user-defined all-reduce hook in + # Saved here and not in the comm. context because we allow the user to + # specify it, possibly at construction time before lazy init + self._all_reduce_hook_stream: torch.cuda.Stream | None = None + + # - Communication and communication/computation overlap + self.comm_ctx = FSDPCommContext() + self._param_group_index: int = 0 + self._num_param_groups: int = 1 + # Group's indices in the shared post-forward order + self._post_forward_indices: list[int] = [] + # Whether to reduce gradients at all (whether for FSDP or HSDP) + self.reduce_grads: bool = True + # Whether to all-reduce gradients for HSDP; only used if + # `self.reduce_grads` is true, in which case setting this to false + # means reduce-scatter but no all-reduce + self.all_reduce_grads: bool = True + # Whether to reshard parameters after backward (only useful for + # gradient accumulation) + self.reshard_after_backward: bool = True + # Per-group input for the reduce-scatter copy-in (chunk_cat) buffer + # cap-K, set via FSDPModule.set_reduce_scatter_max_input_buffers (see + # its docstring). FSDP keeps 1 by default. These per-group values are + # resolved into the single shared + # comm_ctx.reduce_scatter_max_input_buffers that post_backward reads. + self.reduce_scatter_max_input_buffers: int = 1 + # Optional custom factor for the gradient reduction op (e.g. to divide + # by a factor other than the world size) + self.gradient_divide_factor: float | None = None + # Whether to include zero gradients for parameters that did not + # receive a gradient in backward (e.g. due to conditional parameter + # usage across ranks). This ensures all ranks participate in the same + # reduce-scatter collectives, avoiding collective mismatch errors. + self.reduce_scatter_unused_params: bool = False + # Whether reduce-scatter and all-reduce should be issued using only + # summations, potentially with separate pre-/post-scaling. + self.force_sum_reduction_for_comms: bool = False + # `async_op` arg used for pre-forward/pre-backward unshard; can be + # overridden to only do explicit prefetching and avoid inter-stream + # fragmentation from using separate unshard streams + self.unshard_async_op: bool = False + # Whether to unshard in backward: can be overridden by the user if the + # parameters in this group are not needed for backward (e.g. embedding) + self.unshard_in_backward: bool = True + + # - CUDA events for stream synchronization + # Holds the all-gather output buffer, sync objects, and metadata + self._all_gather_result: AllGatherResult | None = None + # Holds the reduce-scatter/all-reduce view-out CUDA event that marks the end of + # the group's post-backward (e.g. reduce-scatter, all-reduce and div), which + # should be waited on at the end of backward + self._post_reduce_event: torch.Event | None = None + # Holds the reshard-after-forward CUDA event when resharding to a + # different world size, which should be waited on in the next unshard + self._reshard_after_forward_event: torch.Event | None = None + + # Only for HSDP, if accumulating gradients without all-reduce, save the + # partial reduce output (only reduce-scattered but not all-reduced) + self._partial_reduce_output: torch.Tensor | None = None + # Holds the reduce-dtype AR buffer + completion event across + # layers in HSDP+AR with reduce_dtype != orig_dtype (e.g., bf16 + # reduce + fp32 params). Structural invariant: the live Python + # ref keeps the buffer off the caching allocator's free list, + # preventing the next layer's RS from reusing the same physical + # block while this layer's AR is still in flight. See + # AllReduceState docstring and regression test PR #180900. + self._all_reduce_state: AllReduceState | None = None + + # Initialization # + def _init_mp_dtypes(self) -> None: + for fsdp_param in self.fsdp_params: + fsdp_param.init_dtype_attrs(self.mp_policy) + trainable_params: list[FSDPParam] = [ + p for p in self.fsdp_params if p.sharded_param.requires_grad + ] + if trainable_params: + params_for_dtype = trainable_params + else: + params_for_dtype = [ + p for p in self.fsdp_params if p.orig_dtype.is_floating_point + ] + orig_dtypes = {p.orig_dtype for p in params_for_dtype} + reduce_dtypes = {p.reduce_dtype for p in params_for_dtype} + if len(trainable_params) > 0 and len(orig_dtypes) != 1: + # Models may have no grad params + raise AssertionError( + f"FSDP expects uniform original parameter dtype but got {orig_dtypes}" + ) + if len(trainable_params) > 0 and len(reduce_dtypes) != 1: + # This can be relaxed if we issue one reduce-scatter per reduce + # dtype (but we would need a way for users to specify multiple + # reduce dtypes) + raise AssertionError( + f"FSDP expects uniform reduce dtype but got {reduce_dtypes}" + ) + dtype_sets_are_uniform = len(orig_dtypes) == 1 and len(reduce_dtypes) == 1 + self._orig_dtype = next(iter(orig_dtypes)) if dtype_sets_are_uniform else None + self._reduce_dtype = ( + next(iter(reduce_dtypes)) if dtype_sets_are_uniform else None + ) + + def lazy_init(self): + # Lazy init should be idempotent + # Users may change or register parameters after construction time. + # For example, DoRA (https://arxiv.org/abs/2402.09353) initializes linear magnitudes based on + # other parameters (e.g. loaded from the state dict). + if not hasattr(self.comm_ctx, "device_handle"): + self.comm_ctx.device_handle = _get_device_handle(self.device.type) + if self.is_sharded and not self._reset_sharded_params: + for fsdp_param in self.fsdp_params: + fsdp_param.reset_sharded_param() + fsdp_param._init_extensions() # allow monkey patch after init + self._reset_sharded_params = True + self._validate_no_meta_params() + self._validate_cpu_offload_params() + self._validate_reduce_scatter_max_input_buffers() + # Initialize mixed precision attributes lazily in case the user changes + # the parameter dtypes after construction time but before forward + self._init_mp_dtypes() + self._register_state_dict_hooks() + + def set_symm_mem(self, backend: Literal["NCCL"] = "NCCL") -> None: + if not isinstance(self._all_gather_comm, (DefaultAllGather | SymmMemAllGather)): + raise AssertionError( + "cannot call set_symm_mem() " + f"when all gather comm is custom: {self._all_gather_comm.__class__.__name__}" + ) + self._all_gather_comm = SymmMemAllGather( + self._all_gather_process_group, backend + ) + if not isinstance( + self._reduce_scatter_comm, (DefaultReduceScatter | SymmMemReduceScatter) + ): + raise AssertionError( + "cannot call set_symm_mem() " + f"when reduce scatter comm is custom: {self._reduce_scatter_comm.__class__.__name__}" + ) + if self.force_sum_reduction_for_comms: + # As of NCCL 2.29.3, NCCL symmetric reduce-scatter only supports SUM reduction + self._reduce_scatter_comm = SymmMemReduceScatter( + self._reduce_scatter_process_group, backend + ) + + def set_allocate_memory_from_process_group(self, enable: bool) -> None: + """ + Whether to (try to) use the ProcessGroup's allocate_tensor method for + the staging buffers for collective comms. + """ + if not isinstance( + self._all_gather_comm, (DefaultAllGather | ProcessGroupAllocAllGather) + ): + raise AssertionError( + "cannot call set_allocate_memory_from_process_group() " + f"when all gather comm is custom: {self._all_gather_comm.__class__.__name__}" + ) + self._all_gather_comm = ( + ProcessGroupAllocAllGather(self._all_gather_process_group) + if enable + else DefaultAllGather() + ) + + if not isinstance( + self._reduce_scatter_comm, + (DefaultReduceScatter | ProcessGroupAllocReduceScatter), + ): + raise AssertionError( + "cannot call set_allocate_memory_from_process_group() " + f"when reduce scatter comm is custom: {self._reduce_scatter_comm.__class__.__name__}" + ) + self._reduce_scatter_comm = ( + ProcessGroupAllocReduceScatter(self._reduce_scatter_process_group) + if enable + else DefaultReduceScatter() + ) + + # Runtime # + @_disable_functorch_if_active + def unshard(self, async_op: bool = False): + if self._all_gather_result is not None: # already called, pending wait + return + if self.is_unsharded: + return # no-op + if ( + not self.unshard_in_backward + and self._training_state == TrainingState.PRE_BACKWARD + ): + return + if self._reshard_after_forward_event is not None: + # Resharded parameter data is allocated in the default stream and + # used in the all-gather streams + self._wait_all_gather_streams_on_event(self._reshard_after_forward_event) + self._reshard_after_forward_event = None + + if isinstance(self.mesh_info, FSDPMeshInfo): + world_size = self._all_gather_process_group.size() + else: + world_size = 1 + if world_size == 1: + # can't skip due to early return in wait_for_unshard if + # no self._all_gather_result + self._all_gather_result = AllGatherResult( + all_gather_output=self._all_gather_output, + all_gather_event=self.device_handle.Event().record(), + all_gather_work=None, + param_all_gather_input_dtypes=[], + param_all_gather_input_numels=[], + all_gather_input_split_sizes=[], + ) + + return + + with record_function(self._with_fqn("FSDP::all_gather")): + self._all_gather_result = foreach_all_gather( + self.fsdp_params, + self._all_gather_process_group, + async_op, + *self.comm_ctx.get_all_gather_streams(async_op, self._training_state), + self.device, + self._all_gather_comm, + ) + + @_disable_functorch_if_active + def wait_for_unshard(self): + """ + 1. In forward with implicit prefetching, to overlap the current copy-out + with the next all-gather, we save a reference to the current all-gather + result to free after the next copy-out. + 2. Otherwise (explicit prefetching or in backward), we free the + all-gather result immediately after the current copy-out since we can + already overlap the current copy-out with the previous reduce-scatter. + """ + if not self._all_gather_result: + return # no preceding unshard + async_op = self._all_gather_result.all_gather_work is not None + if self._training_state == TrainingState.FORWARD: # implicit prefetch + if prev_all_gather_state := self.comm_ctx.all_gather_state: + self._wait_all_gather_streams_on_event(prev_all_gather_state.event) + self.comm_ctx.all_gather_state = None # free the all-gather result + if isinstance(self.mesh_info, FSDPMeshInfo): + world_size = self._all_gather_process_group.size() + else: + world_size = 1 + if world_size == 1: + # directly initialize unsharded parameters from sharded parameters + + for fsdp_param in self.fsdp_params: + # Use all_gather_inputs which already handles conversion to param_dtype + # This is consistent with the world_size > 1 path + all_gather_input = fsdp_param.all_gather_inputs[0] + + # Make sure the all_gather_outputs has proper storage size before using it + # First ensure we have at least one tensor in all_gather_outputs + fsdp_param.init_all_gather_outputs( + [all_gather_input.numel()], + [all_gather_input.dtype], + world_size, + self.device, + ) + + tensor = fsdp_param.all_gather_outputs[0] + alloc_storage(tensor) + + # find alternative way to check if tensor.is_inference + with torch.autograd._unsafe_preserve_version_counter(tensor): + tensor.copy_(all_gather_input) + + else: + with record_function(self._with_fqn("FSDP::all_gather_copy_out")): + foreach_all_gather_copy_out( + self._all_gather_result, + self.fsdp_params, + self._all_gather_process_group, + ) + + for fsdp_param in self.fsdp_params: + fsdp_param.init_unsharded_param() + + self._to_unsharded() + all_gather_copy_out_event = self.device_handle.Event() + all_gather_copy_out_event.record() + + if ( + not async_op + and self._training_state == TrainingState.FORWARD + and world_size > 1 + ): + # Defer free to allow for overlap of this copy-out with next + # all-gather collective + self.comm_ctx.all_gather_state = AllGatherState( + self._all_gather_result, all_gather_copy_out_event + ) + else: + self._wait_all_gather_streams_on_event(all_gather_copy_out_event) + + self._all_gather_result = None # free unless saved in `all_gather_state` + + def _wait_all_gather_streams_on_event(self, event: torch.Event | None): + # Calling `unshard` before lazy init means streams are not initialized + if hasattr(self.comm_ctx, "all_gather_copy_in_stream") and event is not None: + self.comm_ctx.all_gather_copy_in_stream.wait_event(event) + if hasattr(self.comm_ctx, "all_gather_stream") and event is not None: + self.comm_ctx.all_gather_stream.wait_event(event) + + @_disable_functorch_if_active + def reshard(self): + if self._training_state == TrainingState.FORWARD: + if not self._reshard_after_forward: + return + if self._use_post_forward_mesh: + self._to_sharded_post_forward() + self._reshard_after_forward_event = self.device_handle.Event() + if self._reshard_after_forward_event is not None: + self._reshard_after_forward_event.record() + return + self._to_sharded() + + def _reset_iter_state(self) -> None: + # See FSDPState._reset_iter_state for semantics. Waits on any + # in-flight collective events owned by this group, discards + # accumulated grad-reduction state, and restores sharded params. + current_stream = self.device_handle.current_stream() + if self._all_gather_result is not None: + if (event := self._all_gather_result.all_gather_event) is not None: + current_stream.wait_event(event) + work = self._all_gather_result.all_gather_work + if isinstance(work, dist.distributed_c10d.Work): + work.wait() + self._all_gather_result = None + if self._post_reduce_event is not None: + current_stream.wait_event(self._post_reduce_event) + self._post_reduce_event = None + if ( + self._all_reduce_state is not None + and self._all_reduce_state.event is not None + ): + current_stream.wait_event(self._all_reduce_state.event) + self._all_reduce_state = None + if self._reshard_after_forward_event is not None: + self._wait_all_gather_streams_on_event(self._reshard_after_forward_event) + self._reshard_after_forward_event = None + self._partial_reduce_output = None + self._post_forward_indices.clear() + self._training_state = TrainingState.IDLE + self._to_sharded() + + def pre_forward( + self, module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + logger.debug("%s", self._with_fqn("FSDP::pre_forward")) + with record_function(self._with_fqn("FSDP::pre_forward")): + # FORWARD at entry means another module in the grouped + # ``fully_shard([a, b])`` already registered post_backward this + # pass; skip to avoid duplicate ``RegisterPostBackwardFunction`` + # autograd nodes. + with _spmd_no_typecheck(): + entering_forward_pass = self._training_state != TrainingState.FORWARD + self._training_state = TrainingState.FORWARD + self.unshard(self.unshard_async_op) + self.wait_for_unshard() + for fsdp_param in self.fsdp_params: + fsdp_param._restore_spmd_types(fsdp_param.unsharded_param) + if entering_forward_pass: + args, kwargs = self._register_post_backward_hook(args, kwargs) + return args, kwargs + + def post_forward(self, module: nn.Module, input: Any, output: Any): + logger.debug("%s", self._with_fqn("FSDP::post_forward")) + with ( + _spmd_no_typecheck(), + record_function(self._with_fqn("FSDP::post_forward")), + ): + # for AC(fully_shard(model)), AC runs fsdp's _pre_forward + # it shouldn't change post_forward_order + if not is_bw(): + self.reshard() + self._record_post_forward() + self._training_state = TrainingState.IDLE + return output + + def _record_post_forward(self) -> None: + # Since a group has one pre-backward unshard for each forward call + # before the backward, we record each usage (with multiplicity) + post_forward_index = len(self.comm_ctx.post_forward_order) + self.comm_ctx.post_forward_order.append(self) + self._post_forward_indices.append(post_forward_index) + + @_dynamo_disable + def pre_backward(self, default_prefetch: bool, *unused: Any): + if self._training_state == TrainingState.PRE_BACKWARD: + return + logger.debug("%s", self._with_fqn("FSDP::pre_backward")) + with ( + _spmd_no_typecheck(), + record_function(self._with_fqn("FSDP::pre_backward")), + ): + self._training_state = TrainingState.PRE_BACKWARD + self.unshard(self.unshard_async_op) # no-op if prefetched + self.wait_for_unshard() + if default_prefetch: + self._backward_prefetch() + + @_dynamo_disable + def post_backward(self, *unused: Any): + with _spmd_no_typecheck(): + # This method should be idempotent and safe to call even when this + # FSDP parameter group was not used in backward (should be a no-op) + logger.debug("%s", self._with_fqn("FSDP::post_backward")) + # Partial-group-forward detection: grouped ``fully_shard([a, b, ...])`` + # where the forward ran only a subset of the group's modules (chunked + # loss, 1F1B, etc.). The wrapped post-hook took its partial path, so + # ``state._post_forward`` never ran and no ``_pre_backward`` hook was + # registered — ``pre_backward`` therefore never fired and state stays + # at FORWARD instead of transitioning IDLE → PRE_BACKWARD. + is_partial_group_backward = ( + len(self.modules) > 1 # grouped (structural) + and self._training_state == TrainingState.FORWARD # partial path taken + ) + self._training_state = TrainingState.POST_BACKWARD + with record_function(self._with_fqn("FSDP::post_backward_accumulate")): + for fsdp_param in self.fsdp_params: + fsdp_param.accumulate_unsharded_grad_if_needed() + with record_function(self._with_fqn("FSDP::post_backward_reshard")): + if not self.reduce_grads: + if self.reshard_after_backward: + self.reshard() + for fsdp_param in self.fsdp_params: + fsdp_param.to_accumulated_grad_if_needed() + return + # Save the autograd-computed gradients before resharding to only + # access the unsharded parameters when their data is present + fsdp_params_with_grad: list[FSDPParam] = [] + unsharded_grads: list[torch.Tensor] = [] + + for fsdp_param in self.fsdp_params: + if not hasattr(fsdp_param, "_unsharded_param"): + continue + # May have an accumulated gradient of the reduce dtype if the + # previous backward did not reduce-scatter + if fsdp_param.unsharded_accumulated_grad is not None: + fsdp_params_with_grad.append(fsdp_param) + unsharded_grads.append( + fsdp_param.unsharded_accumulated_grad_data + ) + fsdp_param.unsharded_accumulated_grad = None + elif fsdp_param.unsharded_param.grad is not None: + fsdp_params_with_grad.append(fsdp_param) + unsharded_grads.append(fsdp_param.unsharded_grad_data) + fsdp_param.unsharded_param.grad = None + elif ( + self.reduce_scatter_unused_params + and fsdp_param.unsharded_param.requires_grad + ): + fsdp_params_with_grad.append(fsdp_param) + unsharded_grads.append(fsdp_param.unsharded_zero_grad_data) + if self.reshard_after_backward: + self.reshard() + # Recycle prior modules' reduce-scatter input buffers, keeping at most + # `max_input_buffers` in flight: reclaim the oldest (wait on its + # reduce-scatter, then drop the keepalive ref that was deferring the + # allocator's reuse of its memory) until fewer than that remain, before + # this module's reduce appends one more. See + # set_reduce_scatter_max_input_buffers for the memory/overlap tradeoff. + # (Assumes backward fires groups N-1 first; if not, overlap degrades but + # correctness is preserved.) + max_input_buffers = self.comm_ctx.reduce_scatter_max_input_buffers + states = self.comm_ctx.reduce_scatter_states + if ( + self._param_group_index == self._num_param_groups - 1 + and len(states) >= max_input_buffers + ): + with record_function( + f"FSDP::post_backward_rs_wait ({self._module_fqn})" + ): + while len(states) >= max_input_buffers: + oldest = states.pop(0) + if oldest.event is not None: + self.device_handle.current_stream().wait_event(oldest.event) + del oldest + if len(fsdp_params_with_grad) == 0: + return + with record_function(self._with_fqn("FSDP::post_backward_reduce")): + all_reduce_pg = ( + self._all_reduce_process_group + if isinstance(self.mesh_info, DDPMeshInfo) + else None + ) + all_reduce_stream: torch.cuda.Stream + if all_reduce_pg is None and self._all_reduce_hook_stream is not None: + # this means the native HSDP is not enabled, + # but user may want to have a custom HSDP setup + if self._all_reduce_hook is None: + raise AssertionError( + "all reduce hook stream is specified but hook itself is missing." + ) + all_reduce_stream = self._all_reduce_hook_stream + else: + all_reduce_stream = self.comm_ctx.all_reduce_stream + + self._wait_for_post_backward() + ( + reduce_scatter_input, + reduce_scatter_event, + post_reduce_stream, + self._post_reduce_event, + all_reduce_input, + all_reduce_event, + self._partial_reduce_output, + ) = foreach_reduce( + fsdp_params_with_grad, + unsharded_grads, + ( + # pyrefly: ignore [bad-argument-type] + self._reduce_scatter_process_group + if isinstance(self.mesh_info, FSDPMeshInfo) + else None # pyre-fixme[6] + ), + self.comm_ctx.reduce_scatter_stream, + self._reduce_scatter_comm, + self._orig_dtype, + self._reduce_dtype, + self.device, + self.gradient_divide_factor, + ( + self._all_reduce_process_group + if isinstance(self.mesh_info, DDPMeshInfo) + else None + ), + all_reduce_stream, + self.all_reduce_grads, + self._partial_reduce_output, + self._all_reduce_hook, + self.force_sum_reduction_for_comms, + ) + self.comm_ctx._last_post_reduce_events[post_reduce_stream] = ( + self._post_reduce_event + ) + self.comm_ctx.reduce_scatter_states.append( + ReduceScatterState(reduce_scatter_input, reduce_scatter_event) + ) + if is_partial_group_backward: + # Serialize the default stream on this invocation's + # post-accumulate event before returning to autograd. + # Otherwise the next partial-group invocation's autograd + # kernels queue on the default stream concurrently with + # this one's accumulate on the RS stream; the caching + # allocator then hands an autograd kernel a block that is + # a live input/output of this accumulate, corrupting grads + # on non-zero ranks. The post-accumulate event is load- + # bearing (pre-accumulate reduce_scatter_event is NOT, per + # MI350X); waiting at the next post_backward's entry is + # too late because autograd runs between post_backwards. + # No CPU sync. + # + # TODO(#181218): open questions on scope. + # 1. Conditional vs unconditional. The same cross-stream + # hazard exists structurally in regular FSDP but has + # not been observed to fire, plausibly because + # ``reduce_scatter_states.append(...)`` ref-holds + # ``reduce_scatter_input`` and cross-layer allocator + # pressure is looser than within a group. Dropping + # the gate needs a real-model overlap-loss measurement. + # 2. ROCm-specific vs cross-platform. Only observed on + # ROCm/RCCL/MI350X; CUDA FSDP passes without this fix. + # The standalone repro shows the stream-ordering + # hazard (vector 1) is cross-platform, but FSDP's + # ref-hold closes it on both. The residual FSDP-side + # race on ROCm (vector 2) is on a non-Python-reachable + # buffer — RCCL workspace or allocator fragment. + # Whether vector 2 exists on CUDA FSDP but is timing- + # masked is unresolved. See + # ``fsdp2_chunked_loss_rocm_race.md``. + self.device_handle.current_stream().wait_event( + self._post_reduce_event + ) + if all_reduce_input is not None: + if self.device.type != "cpu": + if all_reduce_event is None: + raise AssertionError( + "Expected all_reduce_event to be set for non-CPU device" + ) + self._all_reduce_state = AllReduceState( + all_reduce_input, all_reduce_event + ) + + def finalize_backward(self): + for event in self.comm_ctx._last_post_reduce_events.values(): + self.device_handle.current_stream().wait_event(event) + self.comm_ctx._last_post_reduce_events = dict() + self._post_reduce_event = None + self._all_reduce_state = None + for fsdp_param in self.fsdp_params: + if fsdp_param.grad_offload_event is not None: + fsdp_param.grad_offload_event.synchronize() + fsdp_param.grad_offload_event = None + if self._all_gather_result is not None: + # If there was a mistargeted unshard without a corresponding wait, + # then we wait here and clear the unshard + if (event := self._all_gather_result.all_gather_event) is not None: + torch.accelerator.current_stream().wait_event(event) + work = self._all_gather_result.all_gather_work + if isinstance(work, dist.distributed_c10d.Work): + work.wait() + self._all_gather_result = None + self._post_forward_indices.clear() + + def _wait_for_post_backward(self): + if self._post_reduce_event is not None: + self.device_handle.current_stream().wait_event(self._post_reduce_event) + self._post_reduce_event = None + if ( + self._all_reduce_state is not None + and self._all_reduce_state.event is not None + ): + self.device_handle.current_stream().wait_event(self._all_reduce_state.event) + self._all_reduce_state = None + + def _backward_prefetch(self) -> None: + if self._training_state == TrainingState.PRE_BACKWARD: + if not self._post_forward_indices: + # Can be cleared if running multiple `backward`s + return + curr_index = self._post_forward_indices.pop() + if self._num_param_groups > 1: + # Backward fires groups in reverse forward order: + # N-1, N-2, ..., 1, 0. Index 1 is always the + # penultimate group regardless of N. Prefetching here + # lets the next module's AG overlap with group 0's RS + # without holding unsharded params too long (as would + # happen if we prefetched from N-1). + if self._param_group_index != 1: + return + # E.g. fully_shard(block, shard_placement_fn=...) creates two + # param groups per block (dense + moe), giving + # post_forward_order = [block0, block0.moe, block1, block1.moe]. + # block1.moe walks back past block1 to prefetch block0.moe then block0. + curr_modules = self.modules + target_modules: tuple[nn.Module, ...] | None = None + for step in range(1, curr_index + 1): + target = self.comm_ctx.post_forward_order[curr_index - step] + if target.modules is curr_modules: + continue + if target_modules is None: + target_modules = target.modules + elif target.modules is not target_modules: + break + # Prefetch all groups of the target module in + # reverse forward order (highest index first), + # matching the explicit path in _pre_backward. + self._prefetch_unshard(target, "backward") + elif curr_index > 0: + target = self.comm_ctx.post_forward_order[curr_index - 1] + self._prefetch_unshard(target, "backward") + + @staticmethod + def _prefetch_unshard( + target_fsdp_param_group: FSDPParamGroup, pass_type: str + ) -> None: + if pass_type == "backward": + training_state = TrainingState.PRE_BACKWARD + elif pass_type == "forward": + training_state = TrainingState.FORWARD + else: + raise ValueError(f"Unknown pass type: {pass_type}") + target_fqn = target_fsdp_param_group._module_fqn + with ( + record_function(f"FSDP::{pass_type}_prefetch for {target_fqn}"), + target_fsdp_param_group.use_training_state(training_state), + ): + async_op = target_fsdp_param_group.unshard_async_op + target_fsdp_param_group.unshard(async_op) + + # Utilities # + def _to_sharded(self): + if not self.is_sharded: + for fsdp_param in self.fsdp_params: + fsdp_param.to_sharded() + self._sharded_state = ShardedState.SHARDED + + def _to_sharded_post_forward(self): + if not self.is_sharded_post_forward: + for fsdp_param in self.fsdp_params: + fsdp_param.to_sharded_post_forward() + self._sharded_state = ShardedState.SHARDED_POST_FORWARD + + def _to_unsharded(self): + if not self.is_unsharded: + for fsdp_param in self.fsdp_params: + fsdp_param.to_unsharded() + self._sharded_state = ShardedState.UNSHARDED + + @property + def is_sharded(self) -> bool: + return self._sharded_state == ShardedState.SHARDED + + @property + def is_sharded_post_forward(self) -> bool: + return self._sharded_state == ShardedState.SHARDED_POST_FORWARD + + @property + def is_unsharded(self) -> bool: + return self._sharded_state == ShardedState.UNSHARDED + + @contextlib.contextmanager + def use_training_state(self, training_state: TrainingState): + old_training_state = self._training_state + self._training_state = training_state + try: + yield + finally: + self._training_state = old_training_state + + # Hook Registration # + def _register_post_backward_hook( + self, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + if not torch.is_grad_enabled(): + return args, kwargs + + # Collect all tensors that require gradients (including from dataclasses) + inp_tensors = collect_grad_tensors((args, kwargs)) + if not inp_tensors: + return args, kwargs + + # Apply RegisterPostBackwardFunction to all tensors at once + out_tensors = RegisterPostBackwardFunction.apply(self, *inp_tensors) + + # Replace tensors in the structure (iterator order matches collect order) + new_args, new_kwargs = replace_grad_tensors((args, kwargs), iter(out_tensors)) + return new_args, new_kwargs + + def _register_state_dict_hooks(self) -> None: + num_pre_save_hooks = len(self._module_to_pre_save_state_dict_hook_handle) + num_pre_load_hooks = len(self._module_to_pre_load_state_dict_hook_handle) + if num_pre_save_hooks != num_pre_load_hooks: + raise AssertionError( + f"Pre-save: {num_pre_save_hooks} pre-load: {num_pre_load_hooks}" + ) + if num_pre_save_hooks > 0: + return # already registered + modules_with_fsdp_params: set[nn.Module] = { + fsdp_param._module_info.module for fsdp_param in self.fsdp_params + } + + def to_sharded_hook(*args: Any, **kwargs: Any) -> None: + self._to_sharded() + + for module in modules_with_fsdp_params: + self._module_to_pre_save_state_dict_hook_handle[module] = ( + module.register_state_dict_pre_hook(to_sharded_hook) + ) + self._module_to_pre_load_state_dict_hook_handle[module] = ( + module._register_load_state_dict_pre_hook(to_sharded_hook) + ) + + # Properties # + @property + def _reshard_after_forward(self) -> bool: + return self.post_forward_mesh_info is not None + + @property + def _use_post_forward_mesh(self) -> bool: + return ( + self._reshard_after_forward + and self.mesh_info != self.post_forward_mesh_info + ) + + @property + def _is_hsdp(self) -> bool: + return isinstance(self.mesh_info, HSDPMeshInfo) + + @property + def _all_gather_process_group(self) -> dist.ProcessGroup: + mesh_info = ( + cast(FSDPMeshInfo, self.post_forward_mesh_info) + if self.is_sharded_post_forward + else self.mesh_info + ) + if not isinstance(mesh_info, FSDPMeshInfo): + raise AssertionError( + f"Expected mesh_info to be FSDPMeshInfo, got {type(mesh_info)}" + ) + return mesh_info.shard_process_group + + @property + def _reduce_scatter_process_group(self) -> dist.ProcessGroup: + if not isinstance(self.mesh_info, FSDPMeshInfo): + raise AssertionError( + f"Expected mesh_info to be FSDPMeshInfo, got {type(self.mesh_info)}" + ) + # Falls back to the shard PG (shared with all-gather) unless + # set_separate_reduce_scatter_group opted in to a dedicated PG. + return ( + self.mesh_info.reduce_scatter_process_group + or self.mesh_info.shard_process_group + ) + + def _set_separate_reduce_scatter_group( + self, enable: bool, new_groups: dict[tuple[int, ...], dist.ProcessGroup] + ) -> None: + # Give reduce-scatter its own process group (separate from the shared + # all-gather/shard PG) so the two can overlap in backward, or reset to + # the shared group. ``new_groups`` caches groups by shard ranks within + # one set_separate_reduce_scatter_group call so meshes sharding over the + # same ranks share one communicator (typically one total) rather than one + # per mesh. HSDP ranks may create only their local shard subgroup, so + # use local synchronization to avoid waiting for nonmember ranks in + # new_group's post-init barrier. + mesh_info = self.mesh_info + if not isinstance(mesh_info, FSDPMeshInfo): + raise AssertionError( + f"Expected mesh_info to be FSDPMeshInfo, got {type(mesh_info)}" + ) + if not enable: + mesh_info.reduce_scatter_process_group = None + return + ranks = tuple(dist.get_process_group_ranks(mesh_info.shard_process_group)) + if ranks not in new_groups: + # Reuse the mesh's existing group if already enabled (idempotent), + # else create one for this rank set. + existing = mesh_info.reduce_scatter_process_group + if existing is not None: + new_groups[ranks] = existing + else: + new_group = dist.new_group( + list(ranks), + use_local_synchronization=True, + group_desc="fsdp_reduce_scatter", + ) + if new_group == dist.GroupMember.NON_GROUP_MEMBER: + raise AssertionError( + f"Current rank was not included in process group {ranks}" + ) + new_groups[ranks] = new_group + mesh_info.reduce_scatter_process_group = new_groups[ranks] + + @property + def _all_reduce_process_group(self) -> dist.ProcessGroup: + if not isinstance(self.mesh_info, DDPMeshInfo): + raise AssertionError( + f"Expected mesh_info to be DDPMeshInfo or HSDPMeshInfo, got {type(self.mesh_info)}" + ) + return self.mesh_info.replicate_process_group + + def _with_fqn(self, label: str) -> str: + if self._module_fqn: + label = f"{label} ({self._module_fqn})" + if self._num_param_groups > 1 and isinstance(self.mesh_info, FSDPMeshInfo): + label = f"{label} [pg={self.mesh_info.shard_mesh_size}]" + return label + + def __repr__(self): + return f"FSDPParamGroup(fqn={self._module_fqn})" + + def _validate_no_meta_params(self): + param_names_on_meta = [ + fsdp_param._param_fqn + for fsdp_param in self.fsdp_params + if fsdp_param.sharded_param.device.type == "meta" + ] + if param_names_on_meta: + raise RuntimeError( + "FSDP parameters should be materialized from meta device before training, " + f"but the following were still on meta device: {param_names_on_meta}\n" + "For example, call module.to_empty(device) to materialize to device and " + "call module.reset_parameters() on each module to initialize values." + ) + + def _validate_cpu_offload_params(self): + if not isinstance(self.offload_policy, CPUOffloadPolicy): + return + fsdp_params_not_on_cpu = [ + fsdp_param + for fsdp_param in self.fsdp_params + if fsdp_param.sharded_param.device.type != "cpu" + ] + if fsdp_params_not_on_cpu: + raise RuntimeError( + "FSDP parameters should be materialized on CPU when enabling CPU offloading. " + 'For example, load a CPU state dict or call module.to_empty(device="cpu"). ' + "Found following parameters on non-CPU device: " + f"{[(fsdp_param._param_fqn, fsdp_param.sharded_param.device) for fsdp_param in fsdp_params_not_on_cpu]}\n" + ) + + def _validate_reduce_scatter_max_input_buffers(self): + # Reject max_input_buffers > 1 with the symmetric-memory reduce-scatter + # comm: retaining >1 input buffers relies on the allocator handing out a + # fresh buffer while prior ones are ref-held, but the symmetric-memory + # pool reuses a single buffer, so retaining K would force K symmetric + # segments and can exhaust symmetric memory. Read the per-group cap (set + # in __init__) so this is safe to call before the comm context is lazily + # initialized (e.g. unshard before the first forward). + if self.reduce_scatter_max_input_buffers > 1 and isinstance( + self._reduce_scatter_comm, SymmMemReduceScatter + ): + raise ValueError( + "set_reduce_scatter_max_input_buffers(>1) is not supported with " + "the symmetric-memory reduce-scatter comm, but got " + f"{self.reduce_scatter_max_input_buffers} for module " + f"'{self._module_fqn}' using " + f"{self._reduce_scatter_comm.__class__.__name__}. Keep " + "max_input_buffers=1 with symmetric memory, or use the default " + "reduce-scatter comm." + ) + + +def _get_param_module_infos( + params: list[nn.Parameter], modules: tuple[nn.Module, ...] +) -> list[ParamModuleInfo]: + """ + Shared parameter: lin1.weight = lin2.weight + Shared module: mlp.lin1 = mlp.lin2 + We do not remove duplicates when traversing both modules and parameters to + find shared modules' parameters and shared parameters within a module. + """ + params_set = set(params) + param_to_module_info: dict[nn.Parameter, ParamModuleInfo] = {} + for module in modules: + for _, submodule in module.named_modules(remove_duplicate=False): + for param_name, param in _named_parameters_with_duplicates( + submodule, recurse=False + ): + if param in params_set: + if param not in param_to_module_info: + param_to_module_info[param] = ParamModuleInfo( + submodule, param_name + ) + else: + param_to_module_info[param].shared_modules.append(submodule) + param_to_module_info[param].shared_param_names.append( + param_name + ) + if len(param_to_module_info) != len(params): + raise AssertionError(f"Some parameters are not in the module tree of {modules}") + return [param_to_module_info[param] for param in params] + + +class RegisterPostBackwardFunction(torch.autograd.Function): + generate_vmap_rule = True + + @staticmethod + # pyrefly: ignore [bad-override] + def forward(param_group: FSDPParamGroup, *inputs: torch.Tensor): + # All tensors in `inputs` should require gradient + return inputs + + @staticmethod + def setup_context(ctx, inputs: tuple[Any, ...], output: Any) -> None: + param_group, *_ = inputs + ctx.param_group = param_group + + @staticmethod + def backward(ctx, *grads: torch.Tensor): + ctx.param_group.post_backward() + return (None,) + grads + + @staticmethod + # pyrefly: ignore [bad-override] + def jvp( + ctx: Any, + param_group_tangent: object, + *grad_inputs: Unpack[_GradInputs], + ) -> tuple[Unpack[_GradInputs]]: + # Drop the non-tensor param_group tangent. The output pre-backward hook + # queues final post-backward after all primal/tangent paths finish. + return grad_inputs + + +if dist._is_spmd_types_available(): + import spmd_types + + spmd_types.register_local_autograd_function(RegisterPostBackwardFunction) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_state.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_state.py new file mode 100644 index 0000000000000000000000000000000000000000..2ec2a9e578d56528667c165c74bac04c7e91ff3b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fsdp_state.py @@ -0,0 +1,607 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs +import functools +import logging +import warnings +from collections.abc import Callable, Sequence +from typing import Any, Generic, TYPE_CHECKING, TypeVar + +import torch +import torch.nn as nn +from torch.autograd import Variable +from torch.autograd.graph import _MultiHandle +from torch.distributed import _spmd_no_typecheck +from torch.distributed._composable_state import ( + _get_module_state, + _insert_module_state, + _State, +) +from torch.distributed.device_mesh import _get_device_handle +from torch.distributed.fsdp._common_utils import collect_grad_tensors +from torch.distributed.utils import _apply_to_tensors, _to_kwargs + +from ._fsdp_api import MixedPrecisionPolicy +from ._fsdp_common import _cast_fp_tensor, _dynamo_disable, TrainingState +from ._fsdp_param_group import FSDPCommContext, FSDPParamGroup + + +if TYPE_CHECKING: + from ._fsdp_param import FSDPParam + +logger = logging.getLogger("torch.distributed.fsdp.fully_shard") + +_StateType = TypeVar("_StateType", bound="FSDPState") + + +class FSDPStateContext(Generic[_StateType]): + """This has state shared across FSDP states.""" + + def __init__(self) -> None: + # All FSDP states in the root state's module tree, in + # ``named_modules()`` pre-order. ``_force_complete_incomplete_states`` + # iterates ``reversed`` so within a nesting chain the deepest + # state's ``output_dtype`` cast applies first, matching the order + # ``_post_forward`` would unwind in the non-partial case. + self.all_states: list[_StateType] = [] + # Iteration's forward root runs the once-per-forward logic; this root + # may not be the overall root set by lazy initialization in cases where + # only a submodule runs forward (e.g. encoder-only for eval) + self.iter_forward_root: _StateType | None = None + # Final callback should only be queued once per backward + self.post_backward_final_callback_queued: bool = False + # Whether to finalize backward in this backward's final callback + self.is_last_backward: bool = True + # Optional user-provided event recorded after optimizer for the + # all-gather streams to wait on in the root pre-forward + self.post_optim_event: torch.Event | None = None + + +class FSDPState(_State): + # Name used in error messages; subclasses can override + _state_name: str = "FSDP" + + def __init__(self) -> None: + super().__init__() + # Support multiple param groups for per-param mesh support. + # Each group has params with the same mesh_info. + self._fsdp_param_groups: list[FSDPParamGroup] = [] + self._is_root: bool | None = None # root set during lazy init + self._state_ctx = FSDPStateContext() + self._comm_ctx = FSDPCommContext() + self._training_state: TrainingState = TrainingState.IDLE + self._states_to_forward_prefetch: list[FSDPState] = [] + self._states_to_backward_prefetch: list[FSDPState] = [] + self._modules_to_run_forward: set[nn.Module] = set() + # ``False`` when user set reshard_after_forward + # through ``fully_shard`` or ``set_reshard_after_forward`` + self._auto_reshard_after_forward: bool | None = True + + def _get_state_for_module(self, module: nn.Module) -> "FSDPState | None": + """Get the state for a module. Subclasses can override to use different state getters.""" + return _get_module_fsdp_state(module) + + @property + def _fsdp_param_group(self) -> FSDPParamGroup | None: + """ + Returns the param group for backward compatibility. + This property is only valid when there is at most one param group. + For per-param mesh support with multiple param groups, use + ``_fsdp_param_groups`` instead. + """ + if len(self._fsdp_param_groups) > 1: + group_fqns = [g._module_fqn for g in self._fsdp_param_groups] + raise AssertionError( + f"Expected at most 1 param group for backward compatibility, " + f"but got {len(self._fsdp_param_groups)} (fqns: {group_fqns}). " + f"Use `_fsdp_param_groups` (plural) to access all param groups " + f"when using per-param mesh via shard_placement_fn returning " + f"ShardPlacementResult." + ) + if self._fsdp_param_groups: + return self._fsdp_param_groups[0] + return None + + # Define a separate init since `__init__` is called in the contract + def init( + self, + modules: tuple[nn.Module, ...], + device: torch.device, + mp_policy: MixedPrecisionPolicy, + auto_reshard_after_forward: bool, + ) -> None: + for module in modules: + _insert_module_state(module, self) + self._modules = modules + self._device = device + self._device_handle = _get_device_handle(device.type) + self._mp_policy = mp_policy + self._auto_reshard_after_forward = auto_reshard_after_forward + if len(modules) == 1: + self._pre_forward_hook_handle = modules[0].register_forward_pre_hook( + self._pre_forward, prepend=True, with_kwargs=True + ) + self._post_forward_hook_handle = modules[0].register_forward_hook( + self._post_forward, prepend=False + ) + else: + hook_handle = _register_group_forward_hooks( + modules, + self._pre_forward, + self._post_forward, + self._modules_to_run_forward, + self._cast_output_dtype, + ) + self._pre_forward_hook_handle = hook_handle + self._post_forward_hook_handle = hook_handle + + def _root_pre_forward( + self, module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + self._lazy_init() + if self._state_ctx.iter_forward_root is not None: + return args, kwargs + logger.debug("FSDP::root_pre_forward") + self._state_ctx.iter_forward_root = self + with torch.profiler.record_function("FSDP::root_pre_forward"): + # Wait for optimizer before implicitly prefetched all-gathers + if (event := self._state_ctx.post_optim_event) is not None: + self._comm_ctx.all_gather_copy_in_stream.wait_event(event) + self._comm_ctx.all_gather_stream.wait_event(event) + self._state_ctx.post_optim_event = None + else: + current_stream = self._device_handle.current_stream() + self._comm_ctx.all_gather_copy_in_stream.wait_stream(current_stream) + self._comm_ctx.all_gather_stream.wait_stream(current_stream) + if self._device.type in [ + "cuda", + "hpu", + "xpu", + "mtia", + torch._C._get_privateuse1_backend_name(), + ]: + if args or kwargs: + with torch.profiler.record_function("FSDP::inputs_to_device"): + args_tuple, kwargs_tuple = _to_kwargs( + args, kwargs, self._device, False + ) # same as DDP + args = args_tuple[0] + kwargs = kwargs_tuple[0] + return args, kwargs + + def _lazy_init(self) -> None: + """ + Lazy initialization represents when all modules' parallelisms have + finalized (e.g. FSDP has been applied to all desired modules). This + means that we can determine which state is the root, and we do so by + the 1st state to run forward. + """ + if self._is_root is not None: + return # no-op: already initialized + self._is_root = True + if len(self._modules) > 1: + raise RuntimeError( + f"{self._state_name} requires a single root module but got {self._modules}" + ) + root_module = self._modules[0] + visited_states: set[FSDPState] = set() + for module_name, module in root_module.named_modules(): + if (state := self._get_state_for_module(module)) is None: + continue + if module is not root_module: + if state not in visited_states and state._is_root is not None: + raise RuntimeError( + f"{self._state_name} state has already been lazily initialized for " + f"{module_name}\n{self._state_name} requires running forward through " + "the root module first" + ) + state._is_root = False + # A single state can map to multiple modules (e.g. + # fully_shard([mod_a, mod_b, mod_c])), so dedup here. + if state not in visited_states: + self._state_ctx.all_states.append(state) + visited_states.add(state) + # For the root, do not reshard after forward since for training, + # the parameters would be freed and all-gathered immediately + if self._auto_reshard_after_forward: + for fsdp_param_group in self._fsdp_param_groups: + fsdp_param_group.post_forward_mesh_info = None + self._init_fqns() + self._init_shared_state() + self._validate_no_duplicate_params() + # Run parameter group lazy inits after initializing FQNs for improved + # error messages + for state in self._state_ctx.all_states: + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.lazy_init() + + def _validate_no_duplicate_params(self) -> None: + seen: set[int] = set() + for state in self._state_ctx.all_states: + for fsdp_param_group in state._fsdp_param_groups: + for fsdp_param in fsdp_param_group.fsdp_params: + if fsdp_param._orig_param_uid in seen: + raise ValueError( + f"Parameter '{fsdp_param._param_fqn}' is shared with a " + f"parameter already managed by another FSDP group. " + f"For shared/tied parameters, use " + f"fully_shard([module_a, module_b]) to place them in " + f"the same FSDP group." + ) + seen.add(fsdp_param._orig_param_uid) + + def _init_shared_state(self) -> None: + self._comm_ctx.lazy_init(self._device) + for state in self._state_ctx.all_states: + state._state_ctx = self._state_ctx + state._comm_ctx = self._comm_ctx + num_groups = len(state._fsdp_param_groups) + for i, fsdp_param_group in enumerate(state._fsdp_param_groups): + fsdp_param_group.comm_ctx = self._comm_ctx + fsdp_param_group._param_group_index = i + fsdp_param_group._num_param_groups = num_groups + # The reduce-scatter input-buffer cap governs the single shared + # reduce_scatter_states list, so resolve the per-group caps into one + # effective value (the most aggressive); a lower-cap group must not + # silently undo a higher-cap group's retention. + self._comm_ctx.reduce_scatter_max_input_buffers = max( + ( + fsdp_param_group.reduce_scatter_max_input_buffers + for state in self._state_ctx.all_states + for fsdp_param_group in state._fsdp_param_groups + ), + default=1, + ) + + def _init_fqns(self) -> None: + """Sets module and parameter FQN attributes for debugging.""" + if not self._is_root: + raise AssertionError("Expected _is_root to be True") + root_module = self._modules[0] + param_to_fsdp_param: dict[nn.Parameter, FSDPParam] = {} + # Build a mapping from module to all its FSDPParamGroups (not just one) + module_to_fsdp_param_groups: dict[nn.Module, list[FSDPParamGroup]] = {} + for state in self._state_ctx.all_states: + for fsdp_param_group in state._fsdp_param_groups: + for fsdp_param in fsdp_param_group.fsdp_params: + param_to_fsdp_param[fsdp_param.sharded_param] = fsdp_param + for module in fsdp_param_group.modules: + if module not in module_to_fsdp_param_groups: + module_to_fsdp_param_groups[module] = [] + module_to_fsdp_param_groups[module].append(fsdp_param_group) + for param_name, param in root_module.named_parameters(): + if param in param_to_fsdp_param: + param_to_fsdp_param[param]._param_fqn = param_name + for module_name, module in root_module.named_modules(): + if module in module_to_fsdp_param_groups: + # Set FQN for all param groups associated with this module + for fsdp_param_group in module_to_fsdp_param_groups[module]: + module_fqn = fsdp_param_group._module_fqn + if module_fqn is None: + fsdp_param_group._module_fqn = module_name + else: + if not isinstance(module_fqn, str): + raise AssertionError( + f"Expected module_fqn to be str, got {type(module_fqn)}: {module_fqn}" + ) + module_fqn += f", {module_name}" + fsdp_param_group._module_fqn = module_fqn + + @_dynamo_disable + def _pre_forward( + self, module: nn.Module, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + # When composing with module-hook-based activation checkpointing, the + # pre-backward hook is responsible for the unshard + if self._training_state == TrainingState.PRE_BACKWARD: + with _spmd_no_typecheck(): + # With nested FSDP and multiple forward passes before backward, + # the params might have been resharded by a previous post_backward. + # We need to ensure params are unsharded for AC recomputation. + for fsdp_param_group in self._fsdp_param_groups: + if not fsdp_param_group.is_unsharded: + fsdp_param_group.unshard() + fsdp_param_group.wait_for_unshard() + return self._cast_forward_inputs(args, kwargs) + + # With grouped ``fully_shard([a, b, ...])`` the pre-hook fires per + # module (so ``cast_forward_inputs`` and ``fsdp_param_group.pre_forward`` + # run for each). Root setup and forward prefetch are one-shot, gated + # on the first module's entry. + with _spmd_no_typecheck(): + state_first_in_pass = self._training_state != TrainingState.FORWARD + self._training_state = TrainingState.FORWARD + if state_first_in_pass: + args, kwargs = self._root_pre_forward(module, args, kwargs) + args, kwargs = self._cast_forward_inputs(args, kwargs) + for fsdp_param_group in self._fsdp_param_groups: + args, kwargs = fsdp_param_group.pre_forward(module, args, kwargs) + with _spmd_no_typecheck(): + if state_first_in_pass: + for fsdp_state in self._states_to_forward_prefetch: + # Forward order (not reversed) to match forward execution + # order; contrast with reversed() in _pre_backward. + for target_param_group in fsdp_state._fsdp_param_groups: + FSDPParamGroup._prefetch_unshard(target_param_group, "forward") + return args, kwargs + + @_dynamo_disable + def _post_forward(self, module: nn.Module, input: Any, output: Any) -> Any: + with _spmd_no_typecheck(): + # When composing with module-hook-based activation checkpointing, the + # post-backward hook is responsible for the reshard + if self._training_state == TrainingState.PRE_BACKWARD: + return self._cast_output_dtype(output) + for fsdp_param_group in self._fsdp_param_groups: + output = fsdp_param_group.post_forward(module, input, output) + output = self._register_pre_backward_hook(output) + self._training_state = TrainingState.IDLE + if self._state_ctx.iter_forward_root is self: + output = self._force_complete_incomplete_states(output) + if all_gather_state := self._comm_ctx.all_gather_state: + # Free the last all-gather result if needed; refer to + # [Note: Overlapping all-gather copy-in and all-gather] + self._comm_ctx.all_gather_copy_in_stream.wait_event( + all_gather_state.event + ) + self._comm_ctx.all_gather_stream.wait_event(all_gather_state.event) + self._comm_ctx.all_gather_state = None # free the all-gather result + self._state_ctx.iter_forward_root = None + return self._cast_output_dtype(output) + + def _cast_forward_inputs( + self, args: tuple[Any, ...], kwargs: dict[str, Any] + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + if not (self._mp_policy.cast_forward_inputs and self._mp_policy.param_dtype): + return args, kwargs + with torch.profiler.record_function("FSDP::cast_forward_inputs"): + cast_fn = functools.partial(_cast_fp_tensor, self._mp_policy.param_dtype) + return _apply_to_tensors(cast_fn, args), _apply_to_tensors(cast_fn, kwargs) + + def _cast_output_dtype(self, output: Any) -> Any: + if self._mp_policy.output_dtype is None: + return output + with torch.profiler.record_function("FSDP::cast_forward_outputs"): + return _apply_to_tensors( + functools.partial(_cast_fp_tensor, self._mp_policy.output_dtype), + output, + ) + + def _force_complete_incomplete_states(self, output: Any) -> Any: + # Complete post-forward for any state whose group forward did not + # run all modules (e.g. chunked loss where model.forward skips + # head). See ``all_states`` init for why we iterate ``reversed``. + for state in reversed(self._state_ctx.all_states): + if state is self or not state._modules_to_run_forward: + continue + logger.debug("FSDP::force_complete_post_forward") + for fsdp_param_group in state._fsdp_param_groups: + output = fsdp_param_group.post_forward(None, None, output) + output = state._register_pre_backward_hook(output) + state._training_state = TrainingState.IDLE + output = state._cast_output_dtype(output) + state._modules_to_run_forward.clear() + return output + + @_dynamo_disable + def _pre_backward(self, grad: torch.Tensor) -> torch.Tensor: + with _spmd_no_typecheck(): + self._training_state = TrainingState.PRE_BACKWARD + self._register_root_post_backward_final_callback() + default_prefetch = len(self._states_to_backward_prefetch) == 0 + for fsdp_param_group in self._fsdp_param_groups: + fsdp_param_group.pre_backward(default_prefetch) + for fsdp_state in self._states_to_backward_prefetch: + # Reverse so higher-indexed groups are prefetched first, + # matching backward execution order (reverse of forward). + for target_param_group in reversed(fsdp_state._fsdp_param_groups): + FSDPParamGroup._prefetch_unshard(target_param_group, "backward") + return grad + + @_dynamo_disable + def _root_post_backward_final_callback(self) -> None: + logger.debug("FSDP::root_post_backward") + with torch.profiler.record_function("FSDP::root_post_backward_callback"): + # Reset per-iteration state. With chunked loss, each standalone + # per-chunk call repopulates an inner state's + # ``_modules_to_run_forward`` (and claims ``iter_forward_root``) + # but the group post-hook never empties it because not all + # modules run; clear both here so the next iteration starts + # fresh. + self._state_ctx.iter_forward_root = None + for state in self._state_ctx.all_states: + state._modules_to_run_forward.clear() + # Reverse so that the last param group (which gates the + # reduce-scatter wait/clear) fires first, matching the + # autograd backward order and preserving RS overlap for + # per-param-mesh modules whose inputs lack gradients. + for fsdp_param_group in reversed(state._fsdp_param_groups): + if fsdp_param_group._training_state != TrainingState.POST_BACKWARD: + # Run post-backward in case forward inputs did not require + # gradient so the autograd backward did not run + fsdp_param_group.post_backward() + fsdp_param_group._training_state = TrainingState.IDLE + state._training_state = TrainingState.IDLE + if self._state_ctx.is_last_backward: + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.finalize_backward() + if self._state_ctx.is_last_backward: + self._comm_ctx.post_forward_order.clear() + # Wait on and release any retained reduce-scatter input buffers: + # the last module's (which no later module's rs_wait clears) and, + # when set_reduce_scatter_max_input_buffers retains more than + # one in flight, the rest. The compute stream (which reuses the + # memory) is ordered past each reduce-scatter first. + for rs_state in self._comm_ctx.reduce_scatter_states: + if rs_state.event is not None: + self._device_handle.current_stream().wait_event(rs_state.event) + self._comm_ctx.reduce_scatter_states.clear() + self._state_ctx.post_backward_final_callback_queued = False + + def _register_pre_backward_hook(self, output: Any) -> Any: + if not torch.is_grad_enabled(): + return output + # In eager, register hooks directly on grad-requiring outputs. + tensors = collect_grad_tensors(output) + for t in tensors: + if t._base is not None: + cls = ", ".join(type(m).__name__ for m in self._modules) or "?" + warnings.warn( + f"FSDP2-wrapped module ({cls}) returned a view tensor. " + "An in-place op on this view (e.g., `x += y`) will silently " + "drop the pre-backward hook and skip the all-gather, which " + "can cause backward to fail or produce wrong gradients. " + "Use out-of-place ops (`out = out + y`, not `out += y`) or " + "`.clone()` the output before any in-place op.", + UserWarning, + stacklevel=2, + ) + t.register_hook(self._pre_backward) + if torch._C._are_functorch_transforms_active(): + # Under functorch, some differentiable outputs report + # requires_grad=False inside the transform even when returned + # primals participate in outer autograd. + return _apply_to_tensors( + lambda t: RegisterPreBackwardFunction.apply(self, t) + if not t.requires_grad + and (torch.is_floating_point(t) or torch.is_complex(t)) + else t, + output, + ) + return output + + def _register_root_post_backward_final_callback(self): + if self._state_ctx.post_backward_final_callback_queued: + return + self._state_ctx.post_backward_final_callback_queued = True + Variable._execution_engine.queue_callback( + self._root_post_backward_final_callback + ) + + def _reset_iter_state(self) -> None: + # Iteration-wide recovery after a mid-forward or mid-backward + # exception. Waits on in-flight collectives, reshards every param + # group, and clears per-iteration trackers so the next forward can + # start from a clean state. Any in-flight gradients (reduce-scatter + # results, HSDP partial reduce outputs, grad-accum state) are + # discarded: the failed iteration is treated as lost. + if self._is_root is False: + raise RuntimeError( + "reset_iter_state must be called on the root FSDP module" + ) + current_stream = self._device_handle.current_stream() + if ag_state := self._comm_ctx.all_gather_state: + if ag_state.event is not None: + current_stream.wait_event(ag_state.event) + self._comm_ctx.all_gather_state = None + for rs_state in self._comm_ctx.reduce_scatter_states: + if rs_state.event is not None: + current_stream.wait_event(rs_state.event) + self._comm_ctx.reduce_scatter_states.clear() + for event in self._comm_ctx._last_post_reduce_events.values(): + current_stream.wait_event(event) + self._comm_ctx._last_post_reduce_events.clear() + self._comm_ctx.post_forward_order.clear() + for state in self._state_ctx.all_states: + state._modules_to_run_forward.clear() + state._training_state = TrainingState.IDLE + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group._reset_iter_state() + self._state_ctx.iter_forward_root = None + self._state_ctx.post_backward_final_callback_queued = False + + +def _get_module_fsdp_state(module: nn.Module) -> FSDPState | None: + state = _get_module_state(module) + if isinstance(state, FSDPState): + return state + return None + + +class RegisterPreBackwardFunction(torch.autograd.Function): + generate_vmap_rule = True + + @staticmethod + # pyrefly: ignore [bad-override] + def forward(state: FSDPState, output: torch.Tensor): + return output + + @staticmethod + def setup_context(ctx, inputs: tuple[Any, ...], output: Any) -> None: + state, _ = inputs + ctx.state = state + + @staticmethod + def backward(ctx: Any, *grad_outputs: Any) -> Any: + (grad,) = grad_outputs + return None, ctx.state._pre_backward(grad) + + @staticmethod + def jvp(ctx: Any, *grad_inputs: Any) -> Any: + # Drop the non-tensor FSDP state tangent; keep tangent backward on the + # FSDP pre-backward path. + tangent = grad_inputs[1] + if tangent is None: + return None + return RegisterPreBackwardFunction.apply(ctx.state, tangent) + + +def _register_group_forward_hooks( + modules: Sequence[nn.Module], + pre_hook: Callable, + post_hook: Callable, + modules_to_run: set[nn.Module], + cast_output_dtype: Callable[[Any], Any], +) -> _MultiHandle: + """ + Registers group forward pre and post-hooks. The pre-hook runs on every + module pre-forward; downstream state gating ensures one-shot work + (root setup, post_backward hook registration) fires once per group + pass. The post-hook runs on every module's post-forward: on the + partial path (group not yet complete) it applies only the + ``mp_policy.output_dtype`` cast so standalone per-module callers + observe the same output dtype semantics as the non-grouped case; on + the last module it runs the full ``post_hook`` (which reshards, + registers the pre-backward hook, and itself casts output_dtype). If + a module never runs forward, the post-hook does not fire for it and + ``_force_complete_incomplete_states`` finishes the group from the + root's post-forward. + """ + modules_set = set(modules) + + @_dynamo_disable + @functools.wraps(pre_hook) + def wrapped_pre_hook(*args: Any, **kwargs: Any): + if len(modules_to_run) == 0: + modules_to_run.update(modules_set) + return pre_hook(*args, **kwargs) + + def get_wrapped_post_hook(module: nn.Module): + @_dynamo_disable + @functools.wraps(post_hook) + def wrapped_post_hook(hook_module: nn.Module, input: Any, output: Any) -> Any: + # Full path fires once, when this invocation completes the + # group. Otherwise apply only the output_dtype cast so every + # module in the group (including repeat invocations such as + # per-chunk standalone head calls) produces output in the + # mp_policy's output_dtype. + if module in modules_to_run: + modules_to_run.remove(module) + if len(modules_to_run) == 0: + return post_hook(hook_module, input, output) + return cast_output_dtype(output) + + return wrapped_post_hook + + pre_handles = [ + module.register_forward_pre_hook( + wrapped_pre_hook, prepend=True, with_kwargs=True + ) + for module in modules + ] + post_handles = [ + module.register_forward_hook( + get_wrapped_post_hook(module), prepend=False, always_call=True + ) + for module in modules + ] + return _MultiHandle(tuple(pre_handles + post_handles)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fully_shard.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fully_shard.py new file mode 100644 index 0000000000000000000000000000000000000000..d29ea46ed9e7bbc3206ee7aae878fa794735fcdf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_fully_shard/_fully_shard.py @@ -0,0 +1,996 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs + +from __future__ import annotations + +import functools +from contextlib import contextmanager +from typing import Any, cast, Literal, NoReturn, overload, TYPE_CHECKING +from typing_extensions import deprecated + +import torch +import torch.nn as nn +from torch.distributed._composable import contract + +from ._fsdp_api import ( + AllGather, + DataParallelMeshDims, + MixedPrecisionPolicy, + OffloadPolicy, + ReduceScatter, +) +from ._fsdp_common import _dynamo_disable, FSDPMeshInfo, ShardPlacementFnResult +from ._fsdp_init import ( + _apply_to_module, + _get_device_from_mesh, + _get_mesh_info, + _get_modules_and_states, + _get_post_forward_mesh_info, + _init_default_mesh, + _init_param_group, + _validate_mesh, + _validate_module, +) +from ._fsdp_state import _get_module_fsdp_state, FSDPState + + +if TYPE_CHECKING: + from collections.abc import Callable, Iterable, Iterator + + from torch.distributed.tensor import DeviceMesh + + from ._fsdp_param_group import FSDPParamGroup + +__all__ = [ + "fully_shard", + "FSDPModule", + "UnshardHandle", + "register_fsdp_forward_method", + "get_cls_to_fsdp_cls", + "disable_fsdp_module_new_init", + "share_comm_ctx", +] + + +cls_to_fsdp_cls: dict[type, type] = {} + + +def get_cls_to_fsdp_cls() -> dict[type, type]: + return cls_to_fsdp_cls + + +@overload +# pyrefly: ignore [inconsistent-overload] +def fully_shard( + module: nn.Module, + *, + mesh: DeviceMesh | None = ..., + reshard_after_forward: bool | int | None = ..., + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None = ..., + mp_policy: MixedPrecisionPolicy = ..., + offload_policy: OffloadPolicy = ..., + ignored_params: set[nn.Parameter] | None = ..., + dp_mesh_dims: DataParallelMeshDims | None = ..., +) -> FSDPModule: ... + + +@overload +# pyrefly: ignore [inconsistent-overload] +def fully_shard( + module: list[nn.Module], + *, + mesh: DeviceMesh | None = ..., + reshard_after_forward: bool | int | None = ..., + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None = ..., + mp_policy: MixedPrecisionPolicy = ..., + offload_policy: OffloadPolicy = ..., + ignored_params: set[nn.Parameter] | None = ..., + dp_mesh_dims: DataParallelMeshDims | None = ..., +) -> list[FSDPModule]: ... + + +# The decorator adds a state object to `module` that can be accessed via +# `fully_shard.state(module)`. The state object and module are 1:1. +# [1] Python runtime decorator does not play well with static type checking +# so suppressing some type checks to support type overloads +# such that caller can still get correct return types based on input type +@contract(state_cls=FSDPState) # type: ignore[misc] # see [1] +def fully_shard( + module, + *, + mesh: DeviceMesh | None = None, + reshard_after_forward: bool | int | None = None, + shard_placement_fn: Callable[[nn.Parameter], ShardPlacementFnResult] | None = None, + mp_policy: MixedPrecisionPolicy = MixedPrecisionPolicy(), + offload_policy: OffloadPolicy = OffloadPolicy(), + ignored_params: set[nn.Parameter] | None = None, + dp_mesh_dims: DataParallelMeshDims | None = None, +): + """ + Apply fully sharded data parallelism (FSDP) to ``module``, where FSDP + shards module parameters, gradients, and optimizer states across data + parallel workers to save memory at the cost of communication. + + At initialization, FSDP shards the module's parameters across the data + parallel workers given by ``mesh``. Before forward, FSDP all-gathers the + sharded parameters across the data-parallel workers to get the unsharded + parameters for forward computation. If ``reshard_after_forward`` is + ``True``, then FSDP frees the unsharded parameters after forward and + re-all-gathers them in backward before gradient computation. After gradient + computation, FSDP frees the unsharded parameters and reduce-scatters the + unsharded gradients across data-parallel workers. + + This implementation represents the sharded parameters as :class:`DTensor` s + sharded on dim-0, while the unsharded parameters will be like the original + parameters on ``module`` (e.g. :class:`torch.Tensor` if originally + :class:`torch.Tensor`). A module + `forward pre-hook `_ + on ``module`` all-gathers the parameters, and a module + `forward hook `_ + on ``module`` frees them (if needed). Similar backward hooks all-gather + parameters and later free parameters and reduce-scatter gradients. + + Since grouping multiple tensors together for one collective is critical for + communication efficiency, this implementation makes this grouping first + class. Calling :meth:`fully_shard` on ``module`` constructs one group that + includes the parameters in ``module.parameters()`` except those already + assigned to a group from an earlier call on a submodule. This means that + :meth:`fully_shard` should be called bottom-up on your model. Each group's + parameters are all-gathered in one collective, and its gradients are + reduce-scattered in one collective. Partitioning the model into multiple + groups ("layer by layer") allows for peak memory savings and communication/computation + overlap. Users generally should *not* call :meth:`fully_shard` only on the + topmost root module. + + When called with a list (``fully_shard([a, b, ...])``), the model's + forward may run only a subset of the grouped modules, with the rest + called later in the same iteration. Chunked-loss training with + ``fully_shard([norm, head])`` is the motivating case: the main forward + runs norm only, then head is invoked per chunk. Caveats: + + - Each standalone per-chunk invocation registers its own post_backward + autograd node, so N chunk calls produce N reduce-scatters for that + group. + - ``mp_policy.cast_forward_inputs`` and ``mp_policy.output_dtype`` + both apply per module in the group — every invocation (including + each standalone per-chunk call) casts its inputs to ``param_dtype`` + and its output to ``output_dtype``. + + .. note:: + If ``forward()`` or ``backward()`` raises, FSDP's per-iteration + state (iteration forward-root marker, grouped-module run + trackers, in-flight collective state, per-group training states) + is left in an undefined condition. To recover and run another + iteration, call :meth:`FSDPModule.reset_iter_state` on the root + FSDP module. The failed iteration's gradients are discarded, + including any ``no_sync`` / HSDP partial-reduce accumulation + state. + + Args: + module (Union[nn.Module, List[nn.Module]]): The module or modules to + shard with FSDP and group together for communication. + mesh (Optional[DeviceMesh]): This data parallel mesh defines the + sharding and device. If 1D, then parameters are fully sharded + across the 1D mesh (FSDP) with ``(Shard(0),)`` placement. If 2D, + then parameters are sharded across the 1st dim and replicated + across the 0th dim (HSDP) with ``(Replicate(), Shard(0))`` + placement. The mesh's device type gives the device type used for + communication; if a CUDA or CUDA-like device type, then we use the + current device. + reshard_after_forward (Optional[Union[bool, int]]): This controls the parameter + behavior after forward and can trade off memory and communication: + + - If ``True``, then this reshards parameters after forward and + re-all-gathers in backward. + - If ``False``, then this keeps the unsharded parameters in memory + after forward and avoids the all-gather in backward. For best performance, + we usually set ``False`` for the root module, because the root module + is typically required immediately when the backward pass begins. + - If ``None``, it is set to ``True`` for non-root modules and ``False`` + for root modules. + - If an ``int``, then this represents the world size to reshard to + after forward. It should be a non-trivial divisor of the ``mesh`` + shard dim size (i.e. excluding 1 and the dim size itself). A + choice may be the intra-node size (e.g. ``torch.cuda.device_count()``). + This allows the all-gather in backward to be over a smaller world + size at the cost of higher memory usage than setting to ``True``. + - After forward, the parameters registered to the module depend on + to this: The registered parameters are the sharded parameters if + ``True``; unsharded parameters if ``False``; and the parameters + resharded to the smaller mesh otherwise. To modify the parameters + between forward and backward, the registered parameters must be + the sharded parameters. For ``False`` or an ``int``, this can be + done by manually resharding via :meth:`reshard`. + shard_placement_fn (Optional[Callable[[nn.Parameter], Optional[Shard | ShardPlacementResult]]]): + This callable can be used to override the sharding placement and/or + mesh for a parameter. It can return: + + - ``None``: Use default sharding (Shard(0)) on the mesh passed to + ``fully_shard``. + - :class:`Shard`: Shard the parameter on the specified dimension + using the mesh passed to ``fully_shard``. + - :class:`ShardPlacementResult`: Specify both the shard placement + and a custom :class:`FSDPMeshInfo`. This allows different + parameters to be sharded across different process groups, enabling + use cases like Mixture of Experts where expert params use a + different mesh than regular params. + + If sharding on a nonzero dim, we currently require even sharding, + i.e. the tensor dim size on that dim must be divisible by the FSDP + shard mesh size. + mp_policy (MixedPrecisionPolicy): This controls the mixed precision + policy, which offers parameter/reduction mixed precision for this + module. See :class:`MixedPrecisionPolicy` for details. + offload_policy (OffloadPolicy): This controls the offloading policy, + which offers parameter/gradient/optimizer state offloading. See + :class:`OffloadPolicy` and its subclasses for details. + ignored_params: Optional(Set[nn.Parameter]): The set of parameters to be + ignored by FSDP. They will not be sharded, nor moved to the device + during init, nor have their gradients reduced in backward. + dp_mesh_dims (Optional[DataParallelMeshDims]): When provided, + ``mesh`` is treated as the full SPMD mesh, and parameters should be + DTensors on this mesh with ``Replicate()`` on all DP dimensions. + The ``shard`` field names which dim(s) FSDP shards on (multiple + dims are flattened). The ``replicate`` field names the HSDP + replication dim(s) (multiple dims are flattened). + + Returns: + FSDPModule: The module with FSDP applied (in-place). + """ + torch._C._log_api_usage_once("torch.distributed.fsdp.fully_shard") + _validate_module(module, "fully_shard") + mesh = mesh or _init_default_mesh() + _validate_mesh(mesh, dp_mesh_dims) + mesh_info = _get_mesh_info(mesh, dp_mesh_dims) + device = _get_device_from_mesh(mesh) + auto_reshard_after_forward = reshard_after_forward is None + # If the user does not provide ``reshard_after_forward``, we set it to True. + # During lazy_init, we identify which module is the root and override its value to False + if isinstance(mesh_info, FSDPMeshInfo): + if ( + mesh_info.is_spmd_mesh + and not isinstance(reshard_after_forward, bool) + and isinstance(reshard_after_forward, int) + ): + raise NotImplementedError( + "reshard_after_forward as int is not yet supported with " + "SPMD mesh (dp_mesh_dims)" + ) + post_forward_mesh_info = _get_post_forward_mesh_info( + reshard_after_forward if not auto_reshard_after_forward else True, # type: ignore[arg-type] + mesh_info, + ) + else: + # DDPMeshInfo: no sharding, so no post-forward resharding needed + post_forward_mesh_info = None + arg_module, modules, managed_modules, params, buffers = _get_modules_and_states( + module, device, ignored_params + ) + state = fully_shard.state(modules[0]) # type: ignore[attr-defined] + state.init(modules, device, mp_policy, auto_reshard_after_forward) + + _init_param_group( + state, + params, + modules, + mesh_info, + post_forward_mesh_info, + device, + shard_placement_fn, + mp_policy, + offload_policy, + reshard_after_forward=reshard_after_forward + if not auto_reshard_after_forward + else True, + ) + + # For Dynamo + for managed_module in managed_modules: + managed_module._is_fsdp_managed_module = True # type: ignore[assignment] + managed_module._fsdp_use_orig_params = True # type: ignore[assignment] + + # Place FSDP leftmost for highest priority in the method resolution order + _apply_to_module( + modules, cls_to_fsdp_cls, FSDPModule, "FSDP", _unimplemented_deepcopy + ) + return arg_module + + +def _unimplemented_deepcopy(*args: Any, **kwargs: Any) -> NoReturn: + raise AssertionError( + "FSDP does not support deepcopy. Please use state dict for serialization." + ) + + +_enable_fsdp_module_new_init: bool = True + + +@contextmanager +def disable_fsdp_module_new_init() -> Iterator[None]: + global _enable_fsdp_module_new_init + prev, _enable_fsdp_module_new_init = _enable_fsdp_module_new_init, False + try: + yield + finally: + _enable_fsdp_module_new_init = prev + + +class FSDPModule: + # Index in MRO where the original class is found. + # For FSDP: [FSDP, FSDPModule, Orig, ...] -> index 2 + # Subclasses like ReplicateModule override this. + _orig_cls_mro_index: int = 2 + + def __new__(cls, *args, **kwargs): + """ + Override ``__new__`` to remove the FSDP class and directly construct + the original class for cases like indexing into a container module. + """ + orig_cls = cls.__mro__[cls._orig_cls_mro_index] + self = orig_cls.__new__(orig_cls, *args, **kwargs) + if _enable_fsdp_module_new_init: + self.__init__(*args, **kwargs) + return self + + def reshard(self) -> None: + """ + Reshards the module's parameters, freeing the unsharded parameters if + they are allocated and registering the sharded parameters to the + module. This method is *not* recursive. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.reshard() + + def unshard(self, async_op: bool = False) -> UnshardHandle | None: + """ + Unshards the module's parameters by allocating memory and all-gathering + the parameters. This method is *not* recursive. The unshard follows the + :class:`MixedPrecisionPolicy`, so it will all-gather following + ``param_dtype`` if set. + + Args: + async_op (bool): If ``True``, then returns a :class:`UnshardHandle` + that has a :meth:`wait` method to wait on the unshard op. If + ``False``, then returns ``None`` and waits on the handle inside + this function. + + .. note:: If ``async_op=True``, then FSDP will wait on the pending + unshard in the module's pre-forward for the user. The user only + needs to call :meth:`wait` explicitly if the wait should happen + before pre-forward. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.lazy_init() + fsdp_param_group.unshard(async_op=async_op) + handle = _UnshardHandleImpl( + list(state._fsdp_param_groups) if state._fsdp_param_groups else None + ) + if async_op: + return handle + handle.wait() + return None + + @_dynamo_disable + def reset_iter_state(self) -> None: + """ + Resets FSDP's per-iteration state after an exception aborted a + forward or backward mid-flight. The supported recovery workflow is: + + 1. Catch the exception from ``forward()`` or ``backward()``. + 2. Call ``reset_iter_state()`` on the *root* FSDP module. + 3. Run the next iteration normally. + + The reset waits on any in-flight all-gather/reduce-scatter events, + reshards every parameter group, and clears iteration trackers + (``iter_forward_root``, ``_modules_to_run_forward``, post-forward + order, per-group training states). Any in-flight gradient + reductions are discarded: the failed iteration's gradients are + lost, including HSDP partial-reduce-accumulation state and + ``no_sync`` grad-accumulation state. Callers doing gradient + accumulation should treat the microbatch sequence as invalidated + and restart it. + + Must be called on the root FSDP module — i.e. the module the + top-level ``fully_shard`` was applied to, equivalently the + module first forwarded. Calling on a non-root module raises + ``RuntimeError``. + """ + state = self._get_fsdp_state() + state._reset_iter_state() + + def set_is_last_backward(self, is_last_backward: bool) -> None: + """ + Sets whether the next backward is the last one. On the last backward, + FSDP waits on pending gradient reduction and clears internal data + data structures for backward prefetching. This can be useful for + microbatching. + """ + state = self._get_fsdp_state() + state._state_ctx.is_last_backward = is_last_backward + + def set_requires_gradient_sync( + self, requires_gradient_sync: bool, *, recurse: bool = True + ) -> None: + """ + Sets if the module should sync gradients. This can be used to implement + gradient accumulation *without communication*. For HSDP, this controls + both reduce-scatter and all-reduce together. This is the equivalence of + `no_sync` in FSDP1. + + Args: + requires_gradient_sync (bool): Whether to reduce gradients for the + module's parameters. + recurse (bool): Whether to set for all FSDP submodules or just the + passed-in module. + """ + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.reduce_grads = requires_gradient_sync + fsdp_param_group.all_reduce_grads = requires_gradient_sync + + def set_requires_all_reduce( + self, requires_all_reduce: bool, *, recurse: bool = True + ) -> None: + """ + Sets if the module should all-reduce gradients. This can be used to + implement gradient accumulation with only reduce-scatter but not + all-reduce for HSDP. + """ + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.all_reduce_grads = requires_all_reduce + + def set_reshard_after_forward( + self, reshard_after_forward: bool, recurse: bool = True + ) -> None: + """ + Sets if the module should reshard parameters after forward. This can be + used to change the ``reshard_after_forward`` FSDP arg at runtime. For + example, this can be used to set the FSDP root module's value to + ``True`` (since it is otherwise specially set to ``False``), or it can + set an FSDP module's value to ``False`` for running evals and set back + to ``True`` for training. + + Args: + reshard_after_forward (bool): Whether to reshard parameters after + forward. + recurse (bool): Whether to set for all FSDP submodules or just the + passed-in module. + """ + if not isinstance(reshard_after_forward, bool): + raise ValueError( + f"reshard_after_forward should be a bool, got {type(reshard_after_forward)}" + ) + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + state._auto_reshard_after_forward = False + for fsdp_param_group in state._fsdp_param_groups: + if not isinstance(fsdp_param_group.mesh_info, FSDPMeshInfo): + raise AssertionError + fsdp_param_group.post_forward_mesh_info = ( + _get_post_forward_mesh_info( + reshard_after_forward, + fsdp_param_group.mesh_info, + ) + ) + + def set_reshard_after_backward( + self, reshard_after_backward: bool, *, recurse: bool = True + ) -> None: + """ + Sets if the module should reshard parameters after backward. This can + be used during gradient accumulation to trade off higher memory for + reduced communication since the unsharded parameters do not need to be + re-all-gathered before the next forward. + + Args: + reshard_after_backward (bool): Whether to reshard parameters after + backward. + recurse (bool): Whether to set for all FSDP submodules or just the + passed-in module. + """ + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.reshard_after_backward = reshard_after_backward + + def set_modules_to_forward_prefetch(self, modules: list[FSDPModule]) -> None: + """ + Sets the FSDP modules for which this FSDP module should explicitly + prefetch all-gathers in forward. The prefetching runs after this + module's all-gather copy-out. + + Passing a singleton list containing the next FSDP module gives the same + all-gather overlap behavior as the default overlap behavior, except the + prefetched all-gather is issued earlier from the CPU. Passing a list + with at least length two is required for more aggressive overlap and + will use more reserved memory. + + Args: + modules (List[FSDPModule]): FSDP modules to prefetch. + """ + _assert_all_fsdp_modules(modules) + self._get_fsdp_state()._states_to_forward_prefetch = [ + module._get_fsdp_state() for module in modules + ] + + def set_modules_to_backward_prefetch(self, modules: list[FSDPModule]) -> None: + """ + Sets the FSDP modules for which this FSDP module should explicitly + prefetch all-gathers in backward. This overrides the default backward + pretching implementation that prefetches the next FSDP module based on + the reverse post-forward order. + + Passing a singleton list containing the previous FSDP module gives the + same all-gather overlap behavior as the default overlap behavior. + Passing a list with at least length two is required for more aggressive + overlap and will use more reserved memory. + + Args: + modules (List[FSDPModule]): FSDP modules to prefetch. + """ + _assert_all_fsdp_modules(modules) + self._get_fsdp_state()._states_to_backward_prefetch = [ + module._get_fsdp_state() for module in modules + ] + + def set_custom_all_gather(self, comm: AllGather) -> None: + """ + Overrides the default ``all_gather`` communication behavior, + to have better control over the communication and memory usage. + See `Comm` and `ReduceScatter` for details. + + Args: + comm (AllGather): Custom all-gather communication. + """ + state = self._get_fsdp_state() + if len(state._fsdp_param_groups) > 1: + raise ValueError( + "set_custom_all_gather is not supported with multiple param " + "groups (from per-param mesh via shard_placement_fn). " + "The custom comm would be ambiguous across groups with different meshes." + ) + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group._all_gather_comm = comm + + def set_custom_reduce_scatter(self, comm: ReduceScatter) -> None: + """ + Overrides the default ``reduce_scatter`` communication behavior, + to have better control over the communication and memory usage. + See `Comm` and `ReduceScatter` for details. + + Args: + comm (ReduceScatter): Custom reduce_scatter communication. + """ + state = self._get_fsdp_state() + if len(state._fsdp_param_groups) > 1: + raise ValueError( + "set_custom_reduce_scatter is not supported with multiple param " + "groups (from per-param mesh via shard_placement_fn). " + "The custom comm would be ambiguous across groups with different meshes." + ) + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group._reduce_scatter_comm = comm + + def set_all_reduce_hook( + self, + hook: Callable[[torch.Tensor], None], + *, + stream: torch.cuda.Stream | None = None, + ): + """ + Args: + hook (Callable[[torch.Tensor], None]): User-defined all-reduce hook + with expected signature ``hook(reduce_output: torch.Tensor) -> None`` + where ``reduce_output`` is the reduce-scatter output if only + using FSDP or the all-reduce output if using native HSDP. + stream (Optional[torch.cuda.Stream]): Stream to run the all-reduce + hook in. This should only be set if not using native HSDP. If + using native HSDP, the hook will run in the internally defined + all-reduce stream used by the native HSDP all-reduce. + """ + state = self._get_fsdp_state() + if len(state._fsdp_param_groups) > 1: + raise ValueError( + "set_all_reduce_hook is not supported with multiple param " + "groups (from per-param mesh via shard_placement_fn). " + "The hook would be ambiguous across groups with different meshes." + ) + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group._all_reduce_hook = hook + if stream is not None: + if fsdp_param_group._is_hsdp: + raise ValueError("stream cannot be set when using native HSDP") + fsdp_param_group._all_reduce_hook_stream = stream + + def set_post_optim_event(self, event: torch.Event) -> None: + """ + Sets a post-optimizer-step event for the root FSDP module to wait the + all-gather streams on. + + By default, the root FSDP module waits the all-gather streams on the + current stream to ensure that the optimizer step has finished before + all-gathering. However, this may introduce false dependencies if + there is unrelated computation after the optimizer step. This API + allows the user to provide their own event to wait on. After the root + waits on the event, the event is discarded, so this API should be + called with a new event each iteration. + + Args: + event (torch.Event): Event recorded after the optimizer step + to wait all-gather streams on. + """ + self._get_fsdp_state()._state_ctx.post_optim_event = event + + @deprecated("Use `set_gradient_divide_factor` instead") + def set_reduce_scatter_divide_factor(self, factor: float) -> None: + """Use :py:meth:`set_gradient_divide_factor` instead""" + self.set_gradient_divide_factor(factor) + + def set_gradient_divide_factor(self, factor: float) -> None: + """ + Sets a custom divide factor for the gradient reduction. This might use + a custom reduce op using NCCL's PreMulSum, which allows multiplying by + the factor before reduction. + + Args: + factor (float): Custom divide factor. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.gradient_divide_factor = factor + + def set_force_sum_reduction_for_comms(self, enable: bool) -> None: + """ + Sets whether to require the low-level collective communication + primitives to exclusively use "sum"-type reductions, even if it comes + at the cost of separate additional pre- or post-scaling operations. + This is needed for example because NCCL currently supports zero-copy + transfers only for this kind of collectives. + + NB: for MTIA devices, this is always implicitly enabled. + + NB: if `set_all_reduce_hook` is used under FSDP setup, the caller needs + to ensure the custom all-reduce across FSDP units follow this strategy + as well, as FSDP can no longer automatically handle that. + + Args: + enable (bool): Whether to only ever use ReduceOp.SUM for comms. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.force_sum_reduction_for_comms = enable + + def set_reduce_scatter_unused_params( + self, reduce_scatter_unused_params: bool, *, recurse: bool = True + ) -> None: + """ + Sets whether to include zero gradients for parameters that did not + receive a gradient in backward. This is needed when different ranks + use different parameters due to conditional control flow (e.g. + multi-modal models, mixture of experts), causing mismatched + reduce-scatter collectives. Similar to DDP's + ``find_unused_parameters``. + + Args: + reduce_scatter_unused_params (bool): Whether to include zero + gradients for unused parameters in gradient reduction. + recurse (bool): Whether to set for all FSDP submodules or just + the passed-in module. + """ + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.reduce_scatter_unused_params = ( + reduce_scatter_unused_params + ) + + def set_reduce_scatter_max_input_buffers( + self, max_input_buffers: int, *, recurse: bool = True + ) -> None: + """ + Sets how many gradient reduce-scatter input buffers may be in flight at + once -- the copy-in (``chunk_cat``) buffer cap-K (experimental). + + FSDP keeps **1** such buffer in flight by default, so the compute stream + must wait on the previous reduce-scatter before the next copy-in can + reuse that buffer. When the reduce-scatter is exposed (communication + slower than the backward compute meant to hide it), that recycle wait + stalls the compute stream. Raising the cap lets the next copy-in write a + **fresh** buffer instead of waiting -- removing the stall -- at the cost + of extra peak memory for the retained buffers. The copy-in stays on the + compute stream; there is no extra stream and no ``record_stream``. This + helps only when the reduce-scatter is exposed. + + Args: + max_input_buffers (int): Max reduce-scatter input buffers retained in + flight (the memory<->overlap dial); must be ``>= 1``. ``1`` is + FSDP's default behavior (single buffer; the exposed-RS stall). A + small value (e.g. ``2``) bounds peak memory and adds no stall as + long as it is ``>=`` the reduce-scatter pipeline depth (otherwise + an exposed reduce-scatter trades back a tail stall); a larger + value retains more buffers for deeper overlap at higher peak + memory. + recurse (bool): Whether to set for all FSDP submodules or just the + passed-in module. + """ + # bool is an int subclass; reject it so True does not silently mean 1. + if isinstance(max_input_buffers, bool) or not isinstance( + max_input_buffers, int + ): + raise TypeError( + "max_input_buffers must be an int, got " + f"{type(max_input_buffers).__name__}" + ) + if max_input_buffers < 1: + raise ValueError( + f"max_input_buffers must be a positive int, got {max_input_buffers}" + ) + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.reduce_scatter_max_input_buffers = ( + max_input_buffers + ) + + def set_separate_reduce_scatter_group( + self, enable: bool = True, *, recurse: bool = True + ) -> None: + """ + Enables (or disables) running gradient reduce-scatter on its own process + group so it can overlap with all-gather in the backward pass + (experimental). + + By default FSDP runs all-gather and reduce-scatter on separate CUDA + streams but through the **same** process group -- one NCCL communicator, + which processes one collective at a time and so serializes them on the + wire. When enabled, FSDP creates a dedicated process group over the shard + ranks (``dist.new_group(..., use_local_synchronization=True)``) -- one + per distinct set of shard ranks, typically a single communicator -- so + the two collectives can progress concurrently when the network can + sustain it. This is collective for each shard rank set: like other FSDP + comm setup, call it consistently across ranks using this FSDP mesh. + + Args: + enable (bool): ``True`` (default) gives reduce-scatter its own + process group; ``False`` resets it to the shared shard/all-gather + group. + recurse (bool): Whether to set for all FSDP submodules or just the + passed-in module. + """ + self_module = cast(nn.Module, self) + modules = list(self_module.modules()) if recurse else [self_module] + # Cache created groups by shard ranks so meshes that shard over the same + # ranks share one communicator (typically one total), not one per mesh. + new_groups: dict = {} + for module in modules: + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group._set_separate_reduce_scatter_group( + enable, new_groups + ) + + def set_unshard_in_backward(self, unshard_in_backward: bool) -> None: + """ + Sets whether the FSDP module's parameters need to be unsharded in + backward. This can be used in expert cases when the user knows that all + parameters in this FSDP module's parameter group are not needed for + backward computation (e.g. embedding). + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.unshard_in_backward = unshard_in_backward + + def set_allocate_memory_from_process_group_for_comm(self, enable: bool) -> None: + """ + Sets whether the temporary staging buffers used to send and receive data + over collective communications should be allocated using the custom + optimized allocator provided by the ProcessGroup itself (if any). This + might allow the ProcessGroup to be more efficient. For example, when + using NCCL, this enables it to leverage zero-copy transfers over SHARP + (for NVLink and/or InfiniBand). + + This cannot be used together with :meth:`set_custom_all_gather` or + :meth:`set_custom_reduce_scatter` as those APIs allow for + finer-grained control over each communication, and this method cannot + determine their staging buffer allocation strategy. + + Args: + enable (bool): Whether to turn on ProcessGroup allocation. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.set_allocate_memory_from_process_group(enable) + + def set_symm_mem_for_comm(self, backend: Literal["NCCL"] = "NCCL") -> None: + """ + Sets the symmetric memory (``symm_mem``) backend for allocating the + staging buffers used in all-gather collectives. This allows NCCL to use + optimized all-gather implementations via symmetric memory. Such + optimization may depend on the topology of the system. For single node, + Copy Engine All-Gather may be used. For multi-node, Symmetric Kernel + All-Gather may be used. + + To enable Copy Engine All-Gather, you need to set the NCCL process group + with the zero-CTA policy. + ```python + opts = dist.ProcessGroupNCCL.Options() + opts.config.cta_policy = dist.ProcessGroupNCCL.NCCL_CTA_POLICY_ZERO + dist.init_process_group(backend="nccl", pg_options=opts, device_id=device) + ``` + Alternatively, you can set the environment variable `NCCL_CTA_POLICY` to 2. + ```bash + export NCCL_CTA_POLICY=2 + ``` + For more details, see [Copy Engine + Collectives](https://docs.pytorch.org/docs/2.11/symmetric_memory.html#copy-engine-collectives). + + This cannot be used together with :meth:`set_custom_all_gather` or + :meth:`set_custom_reduce_scatter`. + + Args: + backend (str): The symmetric memory backend to use. Defaults to + ``"NCCL"``. Currently, only ``"NCCL"`` is supported. + """ + state = self._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.set_symm_mem(backend) + + def _set_unshard_async_op(self, async_op: bool): + """ + Sets whether to use ``async_op=True`` or ``False`` for the pre-forward + and pre-backward unshard op. This defaults to ``False`` but can be set + to ``True`` with this method. + + Setting this to ``True`` allows the all-gather allocations to happen in + the default stream, avoiding inter-stream memory fragmentation. + However, you must use explicit prefetching (e.g. via :meth:`unshard`) + in forward to still get overlap, and the pre-all-gather ops like dtype + casting and copy-in will not overlap with compute. + """ + self_module = cast(nn.Module, self) + for module in self_module.modules(): + if isinstance(module, FSDPModule): + state = module._get_fsdp_state() + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.unshard_async_op = async_op + + def _get_fsdp_state(self) -> FSDPState: + if (state := _get_module_fsdp_state(cast(nn.Module, self))) is None: + raise AssertionError(f"No FSDP state found on {self}") + return state + + def _apply(self, *args: Any, **kwargs: Any) -> Any: + # Reshard to ensure that sharded parameters are registered + self.reshard() + ret = super()._apply(*args, **kwargs) # type: ignore[misc] + state = self._get_fsdp_state() + if not state._fsdp_param_groups: + return ret + # TODO: Remove this padding logic once DTensor pads the local tensor: + # https://github.com/pytorch/pytorch/issues/113045 + with torch.no_grad(): + for fsdp_param_group in state._fsdp_param_groups: + for fsdp_param in fsdp_param_group.fsdp_params: + fsdp_param.reset_sharded_param() + return ret + + +class UnshardHandle: + """ + A handle to wait on a :meth:`FSDPModule.unshard` op. + """ + + def wait(self) -> None: + """ + Waits on the unshard op. This ensures that the current stream can use + the unsharded parameters, which are now registered to the module. + """ + return + + +class _UnshardHandleImpl(UnshardHandle): + def __init__(self, fsdp_param_groups: list[FSDPParamGroup] | None): + self._fsdp_param_groups = fsdp_param_groups + + def wait(self): + if self._fsdp_param_groups is not None: + for fsdp_param_group in self._fsdp_param_groups: + fsdp_param_group.wait_for_unshard() + # Avoid keeping a reference + self._fsdp_param_groups = None + + +def register_fsdp_forward_method(module: nn.Module, method_name: str) -> None: + """ + Registers a method on ``module`` to be considered a forward method for + FSDP. + + FSDP all-gathers parameters pre-forward and optionally frees parameters + post-forward (depending on ``reshard_after_forward``). FSDP only knows to + do this for :meth:`nn.Module.forward` by default. This function patches a + user-specified method to run the pre/post-forward hooks before/after the + method, respectively. If ``module`` is not an :class:`FSDPModule`, then + this is a no-op. + + Args: + module (nn.Module): Module to register the forward method on. + method_name (str): Name of the forward method. + """ + if not isinstance(module, FSDPModule): + # Make no-op to allow including both when using/not using FSDP + return + if not hasattr(module, method_name): + raise ValueError(f"{type(module)} does not have a method {method_name}") + orig_method = getattr(module, method_name) + + @functools.wraps(orig_method) + def wrapped_method(self, *args, **kwargs): + fsdp_state = self._get_fsdp_state() + args, kwargs = fsdp_state._pre_forward(self, args, kwargs) + out = orig_method(*args, **kwargs) + return fsdp_state._post_forward(self, args, out) + + # Use `__get__` to make `wrapped_method` an instance method + setattr( + module, + method_name, + wrapped_method.__get__(module, type(module)), # type:ignore[attr-defined] + ) + + +def share_comm_ctx(modules: list[FSDPModule]) -> None: + """ + Share cuda streams for multiple FSDPModules + + Example usage: + from torch.distributed.fsdp import share_comm_ctx + share_comm_ctx([fsdp_model_1, fsdp_model_2, ...]) + + For Pipeline Parallelism (PP), each model chunk is a FSDP root. We want + to share cuda streams for all-gather, reduce-scatter, and all-reduce. + This avoids allocating inter-stream memory framgmentation + + Args: + modules (List[FSDPModule]): modules to share cuda streams + """ + if len(modules) == 0: + return + for module in modules: + if not isinstance(module, FSDPModule): + raise ValueError(f"Expects list of FSDPModules but got {module}") + fsdp_states = [module._get_fsdp_state() for module in modules] + comm_ctx = fsdp_states[0]._comm_ctx + for fsdp_state in fsdp_states[1:]: + fsdp_state._comm_ctx = comm_ctx + for fsdp_param_group in fsdp_state._fsdp_param_groups: + fsdp_param_group.comm_ctx = comm_ctx + + +def _assert_all_fsdp_modules(modules: Iterable[Any]) -> None: + for module in modules: + if not isinstance(module, FSDPModule): + raise ValueError(f"Expects FSDPModule but got {type(module)}: {module}") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_init_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_init_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..d7803282943bca9857291ad75ae60f00c818460c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_init_utils.py @@ -0,0 +1,1236 @@ +# mypy: allow-untyped-defs +import collections +import itertools +import os +import warnings +from collections.abc import Callable, Generator, Iterable, Iterator +from typing import Any, no_type_check, TYPE_CHECKING + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._exec_order_utils as exec_order_utils +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.distributed.fsdp.fully_sharded_data_parallel as fsdp_file +import torch.nn as nn +from torch._custom_class_base import CustomClassBase +from torch.distributed.algorithms._comm_hooks import default_hooks +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.distributed_c10d import _get_default_group +from torch.distributed.fsdp._common_utils import ( + _FSDPDeviceHandle, + _FSDPState, + _get_module_fsdp_state, + _is_fsdp_flattened, + _named_parameters_with_duplicates, + clean_tensor_name, + TrainingState, +) +from torch.distributed.fsdp._flat_param import ( + _FSDP_USE_FULL_PREC_IN_EVAL, + FlatParameter, + FlatParamHandle, + HandleShardingStrategy, +) +from torch.distributed.fsdp._limiter_utils import _FreeEventQueue +from torch.distributed.fsdp.api import ( + BackwardPrefetch, + CPUOffload, + FullOptimStateDictConfig, + FullStateDictConfig, + MixedPrecision, + ShardingStrategy, + StateDictConfig, + StateDictType, +) +from torch.distributed.fsdp.wrap import _Policy +from torch.distributed.tensor.parallel.fsdp import DTensorExtensions +from torch.distributed.utils import _sync_params_and_buffers +from torch.utils._python_dispatch import is_traceable_wrapper_subclass + + +if TYPE_CHECKING: + from torch.utils.hooks import RemovableHandle + +_TORCHDISTX_AVAIL = True +try: + from torchdistx import deferred_init, fake # type: ignore[import] +except ImportError: + _TORCHDISTX_AVAIL = False + +PARAM_BROADCAST_BUCKET_SIZE = 250 * 1024 * 1024 +FSDP_SYNCED = "_fsdp_synced" +# Specification of process groups for hybrid sharding strategies. +HybridShardProcessGroupType = tuple[dist.ProcessGroup, dist.ProcessGroup] +# Overall specification of process group. +ProcessGroupType = dist.ProcessGroup | HybridShardProcessGroupType | None + + +# TODO (awgu): Refactor this later +SHARDING_STRATEGY_MAP = { + ShardingStrategy.NO_SHARD: HandleShardingStrategy.NO_SHARD, + ShardingStrategy.FULL_SHARD: HandleShardingStrategy.FULL_SHARD, + ShardingStrategy.SHARD_GRAD_OP: HandleShardingStrategy.SHARD_GRAD_OP, + ShardingStrategy.HYBRID_SHARD: HandleShardingStrategy.HYBRID_SHARD, + ShardingStrategy._HYBRID_SHARD_ZERO2: HandleShardingStrategy._HYBRID_SHARD_ZERO2, +} +HYBRID_SHARDING_STRATEGIES = [ + ShardingStrategy.HYBRID_SHARD, + ShardingStrategy._HYBRID_SHARD_ZERO2, +] +NO_RESHARD_AFTER_FORWARD_STRATEGIES = ( + ShardingStrategy.SHARD_GRAD_OP, + ShardingStrategy._HYBRID_SHARD_ZERO2, +) + + +# NOTE: Since non-self attributes cannot be type annotated, several attributes +# on `state` are defined first as local variables before being assigned. + + +@no_type_check +def _init_process_group_state( + state: _FSDPState, + process_group: ProcessGroupType, + sharding_strategy: ShardingStrategy, + policy: _Policy | None, + device_mesh: DeviceMesh | None = None, +) -> _FSDPState: + if process_group is not None and device_mesh is not None: + raise ValueError( + "Cannot pass both process_group and device_mesh at the " + "same time. Please just pass only one of them." + ) + is_hybrid_strategy = sharding_strategy in HYBRID_SHARDING_STRATEGIES + if is_hybrid_strategy: + if process_group is None and policy is None and device_mesh is None: + # Raise an error here, since this is manual wrapping with no process group + # passed in, there is no way to ensure all wrapped FSDP instances use the same + # process groups. + raise ValueError( + f"Manual wrapping with {sharding_strategy} " + "requires explicit specification of process group or device_mesh." + ) + else: + state = _init_process_group_state_for_hybrid_shard( + state, process_group, device_mesh + ) + else: + if device_mesh: + state._device_mesh = device_mesh + state.process_group = device_mesh.get_group(mesh_dim=0) + else: + state.process_group = ( + process_group if process_group is not None else _get_default_group() + ) + + state.rank = state.process_group.rank() + state.world_size = state.process_group.size() + data_parallel_world_size = state.world_size + if is_hybrid_strategy: + data_parallel_world_size *= state._inter_node_pg.size() + state._gradient_predivide_factor = ( + default_hooks.DefaultState._get_gradient_predivide_factor( + data_parallel_world_size + ) + ) + state._gradient_postdivide_factor = ( + data_parallel_world_size / state._gradient_predivide_factor + ) + return state + + +@no_type_check +def _init_process_group_state_for_hybrid_shard( + state: _FSDPState, + process_group: ProcessGroupType, + device_mesh: DeviceMesh, +) -> _FSDPState: + if device_mesh: + if _is_valid_hybrid_shard_device_mesh(device_mesh): + state._device_mesh = device_mesh + # We currently only allow _inter_node_pg to be the outermost dimension, and the + # process_group(intra_node) to be the innermost dimension. + state._inter_node_pg = device_mesh.get_group(mesh_dim=0) + state.process_group = device_mesh.get_group(mesh_dim=1) + else: + raise ValueError( + f"Expected device_mesh to have ndim=2 but got {device_mesh.ndim}" + ) + elif process_group is None: + default_group = _get_default_group() + intra_node_group, inter_node_group = _init_intra_and_inter_node_groups( + default_group, state._device_handle.device_count() + ) + # we shard across intra-node + state.process_group = intra_node_group + # save _inter_node_pg to allreduce across. + state._inter_node_pg = inter_node_group + else: + # Check type and assign state.process_group and state._inter_node_pg. + if _is_valid_hybrid_shard_pg_type(process_group): + # Assuming that user passed in as intra node group and inter node group + # as documented. + state.process_group, state._inter_node_pg = process_group + else: + raise ValueError( + "Expected process_group to be passed in as either None or " + f"Tuple[dist.ProcessGroup, dist.ProcessGroup] but got {type(process_group)}" + ) + # Create state for allreduce + state._inter_node_state = _get_default_comm_hook_state( + process_group=state._inter_node_pg, + ) + return state + + +@no_type_check +def _is_valid_hybrid_shard_pg_type(process_group: Any) -> bool: + return ( + isinstance(process_group, tuple) + and len(process_group) == 2 + and all(isinstance(pg, dist.ProcessGroup) for pg in process_group) + ) + + +@no_type_check +def _is_valid_hybrid_shard_device_mesh(device_mesh: DeviceMesh) -> bool: + return isinstance(device_mesh, DeviceMesh) and device_mesh.ndim == 2 + + +@no_type_check +def _init_intra_node_process_group(num_devices_per_node: int) -> dist.ProcessGroup: + """ + Return a process group across the current node. + + For example, given each row is a distinct node: + 0 1 2 3 4 5 6 7 + 8 9 10 11 12 13 14 15 + This API would return an intra-node subgroup across + [0, 1, ..., 7] or [8, 9, ..., 15] depending on the process's rank. + For example, rank 3 would get [0, 1, ..., 7]. + """ + intra_node_subgroup, _ = dist.new_subgroups(num_devices_per_node) + return intra_node_subgroup + + +@no_type_check +def _init_inter_node_process_group( + global_process_group: dist.ProcessGroup, + num_devices_per_node: int, +) -> dist.ProcessGroup: + """ + Return an inter-node process group where each contained rank has the same local rank. + + For example, given each row is a distinct node: + 0 1 2 3 4 5 6 7 + 8 9 10 11 12 13 14 15 + This API would return inter-node process group [0, 8], [1, 9], [2, 10], and so forth + depending on the process's rank. For example, rank 1 would get [1, 9], rank 5 + would get [5, 13]. + """ + # the inter-node pg that is returned + inter_node_pg = None + sharding_backend = dist.get_backend(global_process_group) + world_size = dist.get_world_size(global_process_group) + # Assuming fully homogeneous setup + num_nodes = world_size // num_devices_per_node + my_local_rank = dist.get_rank(global_process_group) % num_devices_per_node + for local_rank in range(num_devices_per_node): + ranks_for_inter_group = [ + local_rank + (i * num_devices_per_node) for i in range(num_nodes) + ] + # every rank always needs to call dist.new_group + grp = dist.new_group(ranks=ranks_for_inter_group, backend=sharding_backend) + if local_rank == my_local_rank: + inter_node_pg = grp + + if inter_node_pg is None: + raise AssertionError( + f"{my_local_rank} expected to assign inter-node pg, but did not" + ) + return inter_node_pg + + +def _init_intra_and_inter_node_groups( + global_process_group: dist.ProcessGroup, + num_devices_per_node: int, +) -> tuple[dist.ProcessGroup, dist.ProcessGroup]: + """ + Initialize intra and inter-node process groups and return the ones corresponding to this process's rank. + + This function can be used to initialize process groups for ``HYBRID_SHARD`` or + ``_HYBRID_SHARD_ZERO2`` in FSDP. + This function assumes each node has an equal number of CUDA-enabled devices. + Returns: + Tuple[dist.ProcessGroup, dist.ProcessGroup]: Intra and inter-node process group. + """ + return ( + _init_intra_node_process_group(num_devices_per_node), + _init_inter_node_process_group(global_process_group, num_devices_per_node), + ) + + +@no_type_check +def _init_ignored_module_states( + state: _FSDPState, + module: nn.Module, + ignored_modules: Iterable[torch.nn.Module] | None, + ignored_states: Iterable[torch.nn.Parameter] + | Iterable[torch.nn.Module] + | None = None, +) -> _FSDPState: + if ignored_modules is not None and ignored_states is not None: + raise ValueError( + "Cannot pass both ignored_modules and ignored_states at the " + "same time. Please just pass ignored_states." + ) + ignored_parameters = None + passed_as_ignored_states = ignored_states is not None + if passed_as_ignored_states: + ignored_states_list = list(ignored_states) + _check_ignored_states(ignored_states_list, True) + else: + ignored_states_list = [] + _check_ignored_states( + list(ignored_modules) if ignored_modules is not None else [], False + ) + if len(ignored_states_list) > 0: + if isinstance(ignored_states_list[0], nn.Parameter): + ignored_parameters = ignored_states_list + else: + ignored_modules = ignored_states_list + state._ignored_modules = _get_ignored_modules(module, ignored_modules) + state._ignored_params = _get_ignored_params( + module, + state._ignored_modules, + ignored_parameters, + ) + state._ignored_buffer_names = _get_ignored_buffer_names( + module, + state._ignored_modules, + ) + # TODO: FSDP's contract for buffers is not well-defined. They are + # implicitly ignored for most functionality since they are not sharded; + # however, FSDP still imposes some semantics on buffers (e.g. buffer mixed + # precision). We should formalize this contract and decide if we need to + # compute and store `_ignored_buffers`. + return state + + +def _check_ignored_states( + ignored_states: list[Any], passed_as_ignored_states: bool +) -> None: + """ + Check that the ignored states are uniformly parameters or uniformly modules. + + We may remove this check in the future if we permit mixing. + """ + if len(ignored_states) == 0: + return + if passed_as_ignored_states: + all_params = all(isinstance(state, nn.Parameter) for state in ignored_states) + all_modules = all(isinstance(state, nn.Module) for state in ignored_states) + if not all_params and not all_modules: + # Sort for consistent ordering for unit test regex matching + sorted_types = sorted({type(state) for state in ignored_states}, key=repr) + raise ValueError( + "ignored_states expects all nn.Parameter or all nn.Module list " + f"elements but got types {sorted_types}" + ) + else: + if not all(isinstance(state, nn.Module) for state in ignored_states): + sorted_types = sorted({type(state) for state in ignored_states}, key=repr) + raise ValueError( + "ignored_modules expects nn.Module list elements but got " + f"types {sorted_types}" + ) + + +@no_type_check +def _init_device_handle( + state: _FSDPState, + module: nn.Module, + ignored_params: set[nn.Parameter], + device_id: int | torch.device | None, +) -> _FSDPState: + """ + Determine device handle used for initializing FSDP. + + If a device is specified by ``device_id``, + then returns device handle corresponds to that device type. Otherwise, If the + module is already on a non-CPU device, then the device type is that non-CPU device type. + If the module is on CPU or meta, then the device type is the current accelerator device. + See the :ref:`Accelerators` for details. + + + This method will be called once ignored parameters was determined, as the device handle maybe needed + for other initialization. + """ + determined_device = None + if device_id is not None: + determined_device = ( + device_id + if isinstance(device_id, torch.device) + else torch.device(device_id) + ) + if determined_device is None: + for param in _get_orig_params(module, ignored_params): + if param.device.type in {"cpu", "meta"}: + continue + if determined_device is None: + determined_device = param.device + else: + if param.device.type != determined_device.type: + raise RuntimeError( + f"FSDP does not support modules with different device types " + f"but got params on {determined_device.type} and {param.device.type}" + ) + determined_device = determined_device or torch._C._get_accelerator() + if determined_device.type == "cpu": + raise RuntimeError( + "FSDP needs a non-CPU accelerator device, but no accelerator device is detected." + ) + + state._device_handle = _FSDPDeviceHandle.from_device(determined_device) + return state + + +@no_type_check +def _init_buffer_state( + state: _FSDPState, + module: nn.Module, +) -> _FSDPState: + state._buffer_names = _get_buffer_names(module) + # Save a mapping from clean fully-qualified buffer name (starting from + # `module`) to its original dtype for restoring that dtype during model + # checkpointing when buffer mixed precision is enabled. The names should + # be clean since the casting happens in a `summon_full_params()` context. + _buffer_name_to_orig_dtype: dict[str, torch.dtype] = {} + for buffer_name, buffer in module.named_buffers(): + buffer_name = clean_tensor_name(buffer_name) + _buffer_name_to_orig_dtype[buffer_name] = buffer.dtype + state._buffer_name_to_orig_dtype = _buffer_name_to_orig_dtype + return state + + +@no_type_check +def _init_core_state( + state: _FSDPState, + sharding_strategy: ShardingStrategy | None, + mixed_precision: MixedPrecision | None, + cpu_offload: CPUOffload | None, + limit_all_gathers: bool, + use_orig_params: bool, + backward_prefetch_limit: int, + forward_prefetch_limit: int, +) -> _FSDPState: + # We clamp the strategy to `NO_SHARD` for world size of 1 since they are + # currently functionally equivalent. This may change if/when we integrate + # FSDP with MoE. + if state.world_size == 1: + if sharding_strategy != ShardingStrategy.NO_SHARD: + warnings.warn( + "FSDP is switching to use `NO_SHARD` instead of " + f"{sharding_strategy or ShardingStrategy.FULL_SHARD} since " + "the world size is 1.", + stacklevel=2, + ) + sharding_strategy = ShardingStrategy.NO_SHARD + elif sharding_strategy == ShardingStrategy.NO_SHARD: + warnings.warn( + "The `NO_SHARD` sharding strategy is deprecated. If having issues, " + "please use `DistributedDataParallel` instead.", + FutureWarning, + # Level 1 is here, level 2 is from `FullyShardedDataParallel`, and + # level 3 is from the true caller + stacklevel=3, + ) + state.sharding_strategy = sharding_strategy or ShardingStrategy.FULL_SHARD + state.mixed_precision = mixed_precision or MixedPrecision() + if mixed_precision is not None: + torch._C._log_api_usage_once( + f"torch.distributed.fsdp.mixed_precision.{str(state.mixed_precision)}" + ) + state._use_full_prec_in_eval = ( + os.environ.get(_FSDP_USE_FULL_PREC_IN_EVAL, "") == "1" + ) + state.cpu_offload = cpu_offload or CPUOffload() + state.limit_all_gathers = limit_all_gathers + state._use_orig_params = use_orig_params + state.training_state = TrainingState.IDLE + state._is_root = None + state._free_event_queue = _FreeEventQueue() + state._debug_level = dist.get_debug_level() + state._exec_order_data = exec_order_utils._ExecOrderData( + state._debug_level, + backward_prefetch_limit, + forward_prefetch_limit, + ) + state._unshard_event = None + # Mapping from fully sharded module to the handles it is responsible to + # unshard and reshard (see [Note: Fully Sharded Module]) + _fully_sharded_module_to_handle: dict[nn.Module, FlatParamHandle] = {} + state._fully_sharded_module_to_handle = _fully_sharded_module_to_handle + # Invariant: `state.params` contains exactly the `FlatParameter`s of the + # handles in `state._handle` + _handle: FlatParamHandle | None = None + state._handle = _handle + params: list[FlatParameter] = [] + state.params = params + return state + + +@no_type_check +def _init_runtime_state( + state: _FSDPState, +) -> _FSDPState: + _root_pre_forward_handles: list[RemovableHandle] = [] + state._root_pre_forward_handles = _root_pre_forward_handles + _pre_forward_handles: list[RemovableHandle] = [] + state._pre_forward_handles = _pre_forward_handles + _post_forward_handles: list[RemovableHandle] = [] + state._post_forward_handles = _post_forward_handles + state._sync_gradients = True + state._comm_hook = None + state._comm_hook_state = None + # Used to prevent running the pre-backward hook multiple times + return state + + +@no_type_check +def _init_prefetching_state( + state: _FSDPState, + backward_prefetch: BackwardPrefetch, + forward_prefetch: bool, +) -> _FSDPState: + state.backward_prefetch = backward_prefetch + state.forward_prefetch = forward_prefetch + # The data structures use tuples of handles to generalize over the case + # where a module's forward involves multiple handles. + return state + + +@no_type_check +# pyrefly: ignore [bad-function-definition] +def _init_extension(state: _FSDPState, device_mesh: DeviceMesh = None) -> _FSDPState: + # TODO: we need to add additional check once we support FSDP + PiPPy. + # This check is currently sufficient, since we only support FSDP + TP. + root_mesh = device_mesh._get_root_mesh() if device_mesh is not None else None + # if a root mesh is not the same as device_mesh, + # meaning the device_mesh is sliced out from the root mesh. + if device_mesh and root_mesh != state._device_mesh: + state._fsdp_extension = DTensorExtensions(state._device_handle) + else: + # We need to explicitly set _fsdp_extension to None. + # Otherwise, we will run into an infinite recursion when getting the attribute. + state._fsdp_extension = None + return state + + +@no_type_check +def _init_state_dict_state(state: _FSDPState) -> _FSDPState: + state._state_dict_type = StateDictType.FULL_STATE_DICT + state_dict_config: StateDictConfig = FullStateDictConfig() + state._optim_state_dict_config = FullOptimStateDictConfig() + state._state_dict_config = state_dict_config + unshard_params_ctx: dict[nn.Module, Generator] = {} + state._unshard_params_ctx = unshard_params_ctx + + return state + + +def _verify_managed_params(module: nn.Module, params: list[nn.Parameter]) -> None: + """ + Verify if the parameters are accepted by FSDP. The only restriction now + is that the parameter cannot be a scalar tensor (param.shape == []). + """ + for param in params: + if len(param.shape) == 0: + param_name = "" + for name, param_ in module.named_parameters(): + if param is param_: + param_name = name + break + if not param_name: + raise AssertionError("Expected param_name to be set") + raise ValueError( + "FSDP doesn't support scalar parameters. " + f"Change {param_name} to a 1D tensor with numel equal to 1." + ) + + +@no_type_check +def _init_param_handle_from_module( + state: _FSDPState, + fully_sharded_module: nn.Module, + device_id: int | torch.device | None, + param_init_fn: Callable[[nn.Module], None] | None, + sync_module_states: bool, +) -> _FSDPState: + """Initialize a ``FlatParamHandle`` from a module ``fully_sharded_module``.""" + _check_single_device_module(fully_sharded_module, state._ignored_params, device_id) + device_from_device_id = _get_device_from_device_id( + device_id, state.rank, state._device_handle + ) + is_meta_module, is_torchdistX_deferred_init = _need_to_materialize_module( + fully_sharded_module, state._ignored_params, state._ignored_modules + ) + # Materialize the module if needed + if (is_meta_module or is_torchdistX_deferred_init) and param_init_fn is not None: + _materialize_with_param_init_fn( + fully_sharded_module, param_init_fn, state._ignored_modules + ) + elif is_meta_module: + _materialize_meta_module( + fully_sharded_module, + device_id, + state._ignored_modules, + state._device_handle, + ) + elif is_torchdistX_deferred_init: + deferred_init.materialize_module( + fully_sharded_module, + check_fn=lambda submodule: _get_module_fsdp_state(submodule) is None + and submodule not in state._ignored_modules, + ) + + ignored_buffers = { + buffer + for ignored_module in state._ignored_modules + for buffer in ignored_module.buffers() + } + + _move_module_to_device( + fully_sharded_module, + state._ignored_params, + ignored_buffers, + device_from_device_id, + ) + state.compute_device = _get_compute_device( + fully_sharded_module, + state._ignored_params, + device_from_device_id, + state.rank, + state._device_handle, + ) + + managed_params = list(_get_orig_params(fully_sharded_module, state._ignored_params)) + _verify_managed_params(fully_sharded_module, managed_params) + if sync_module_states: + if state.sharding_strategy in HYBRID_SHARDING_STRATEGIES: + # Broadcast inter-node first, then intra-node. The inter-node + # broadcast propagates rank 0's states to each node's local + # rank 0, so the subsequent intra-node broadcast has the + # correct source values on every node. Reversing this order + # causes local rank 0 on non-source nodes to broadcast + # uninitialized states (e.g. from meta-device materialization). + _sync_module_params_and_buffers( + fully_sharded_module, managed_params, state._inter_node_pg + ) + # _sync_module_params_and_buffers marks each buffer with + # FSDP_SYNCED=True to avoid redundant syncs in nested + # wrapping. Reset the flag here so the intra-node broadcast + # below also includes buffers. + for buffer in fully_sharded_module.buffers(): + if hasattr(buffer, FSDP_SYNCED): + setattr(buffer, FSDP_SYNCED, False) + _sync_module_params_and_buffers( + fully_sharded_module, managed_params, state.process_group + ) + _init_param_handle_from_params(state, managed_params, fully_sharded_module) + return state + + +@no_type_check +def _init_param_handle_from_params( + state: _FSDPState, + params: list[nn.Parameter], + fully_sharded_module: nn.Module, +): + if len(params) == 0: + return + handle = FlatParamHandle( + params, + fully_sharded_module, + state.compute_device, + SHARDING_STRATEGY_MAP[state.sharding_strategy], + state.cpu_offload.offload_params, + state.mixed_precision.param_dtype, + state.mixed_precision.reduce_dtype, + state.mixed_precision.keep_low_precision_grads, + state.process_group, + state._use_orig_params, + fsdp_extension=state._fsdp_extension, + ) + handle.shard() + if state._handle: + raise AssertionError("Expected state._handle to be None") + state.params.append(handle.flat_param) + state._handle = handle + state._fully_sharded_module_to_handle[handle._fully_sharded_module] = handle + cpu_device = torch.device("cpu") + if state.cpu_offload.offload_params and handle.flat_param.device != cpu_device: + handle.flat_param_to(cpu_device) + + +def _get_ignored_modules( + root_module: nn.Module, + _ignored_modules: Iterable[torch.nn.Module] | None, +) -> set[nn.Module]: + """ + Check that ``_ignored_modules`` is an iterable of ``nn.Module`` s without any FSDP instances. + + Return the modules contained in their module + subtrees as a :class:`set`. Nested FSDP instances are excluded, but their + already-computed ignored modules are included. + + ``_ignored_modules`` represents the argument passed by the user to FSDP. + """ + msg_prefix = "`ignored_modules` should be an iterable of `torch.nn.Module`s " + try: + ignored_root_modules = ( + set(_ignored_modules) if _ignored_modules is not None else set() + ) + except TypeError as e: + raise TypeError(msg_prefix + f"but got {type(_ignored_modules)}") from e + for module in ignored_root_modules: + if not isinstance(module, torch.nn.Module): + raise TypeError(msg_prefix + f"but got an iterable with {type(module)}") + if _get_module_fsdp_state(module): + # TODO: We may relax this by taking the FSDP instance's wrapped + # module to provide more flexibility to the user. + raise ValueError("`ignored_modules` should not include FSDP modules") + # Treat modules that cannot compose with `fully_shard` as ignored modules, + # meaning that their subtrees are ignored + for module in root_module.modules(): + if not traversal_utils._composable(module): + ignored_root_modules.add(module) + # NOTE: Even if `ignored_root_modules` is empty, do not return early so + # that this FSDP instance can get any ignored modules from its children. + + # Include child modules and exclude nested FSDP modules themselves + ignored_modules = { + child + for module in ignored_root_modules + for child in module.modules() + if not isinstance(child, fsdp_file.FullyShardedDataParallel) + } + if root_module in ignored_modules: + warnings.warn( + "Trying to ignore the top-level module passed into the FSDP " + "constructor itself will result in all parameters being " + f"ignored and is not well-supported: {module}", + stacklevel=2, + ) + # Include nested FSDP modules' ignored modules + for submodule in root_module.modules(): + optional_fsdp_state = _get_module_fsdp_state(submodule) + if optional_fsdp_state is not None: + if not hasattr(optional_fsdp_state, "_ignored_modules"): + raise AssertionError( + "Expected optional_fsdp_state to have _ignored_modules attribute" + ) + ignored_modules.update(optional_fsdp_state._ignored_modules) + return ignored_modules + + +def _get_ignored_params( + root_module: torch.nn.Module, + ignored_modules: set[torch.nn.Module], + ignored_parameters: Iterable[torch.nn.Parameter] | None = None, +) -> set[torch.nn.Parameter]: + """ + Return the parameters of the modules in ``ignored_modules`` and the parameters in ``ignored_parameters``. + + :class:`FlatParameter` s are excluded from the result. + """ + all_ignored_params: set[torch.nn.Parameter] = set() + + params_in_ignored_modules = { + p for m in ignored_modules for p in m.parameters() if not _is_fsdp_flattened(p) + } + + all_ignored_params.update(params_in_ignored_modules) + + if ignored_parameters is not None: + params_in_ignored_parameters = { + p for p in ignored_parameters if not _is_fsdp_flattened(p) + } + all_ignored_params.update(params_in_ignored_parameters) + + # Always include nested FSDP modules' ignored parameters + for submodule in root_module.modules(): + optional_fsdp_state = _get_module_fsdp_state(submodule) + if optional_fsdp_state is not None: + if not hasattr(optional_fsdp_state, "_ignored_params"): + raise AssertionError( + "Expected optional_fsdp_state to have _ignored_params attribute" + ) + all_ignored_params.update(optional_fsdp_state._ignored_params) + + return all_ignored_params + + +def _get_ignored_buffer_names( + root_module: torch.nn.Module, + ignored_modules: set[torch.nn.Module], +) -> set[str]: + """Return the cleaned buffer FQNs in ``ignored_modules``.""" + all_ignored_buffer_names: set[str] = set() + + buffers_in_ignored_modules = { + buffer for m in ignored_modules for buffer in m.buffers() + } + + all_ignored_buffer_names.update( + { + clean_tensor_name(buffer_name) + for buffer_name, buffer in root_module.named_buffers() + if buffer in buffers_in_ignored_modules + } + ) + + # Always include nested FSDP modules' ignored buffer names + for submodule in root_module.modules(): + optional_fsdp_state = _get_module_fsdp_state(submodule) + if optional_fsdp_state is not None: + if not hasattr(optional_fsdp_state, "_ignored_buffer_names"): + raise AssertionError( + "Expected optional_fsdp_state to have _ignored_buffer_names attribute" + ) + all_ignored_buffer_names.update(optional_fsdp_state._ignored_buffer_names) + + return all_ignored_buffer_names + + +def _get_buffer_names(root_module: nn.Module) -> set[str]: + """Return the fully prefixed names of all buffers in the module hierarchy rooted at ``root_module`` as a class:`set`.""" + return { + clean_tensor_name(buffer_name) for buffer_name, _ in root_module.named_buffers() + } + + +def _check_single_device_module( + module: nn.Module, + ignored_params: set[nn.Parameter], + device_id: int | torch.device | None, +) -> None: + """ + Raise an error if ``module`` has original parameters on multiple devices, ignoring the parameters in ``ignored_params``. + + Thus, after this method, the + module must be either fully on the CPU or fully on a non-CPU device. + """ + devices = {param.device for param in _get_orig_params(module, ignored_params)} + # We allow module to be partially on CPU and partially on GPU if device_id is not + # None, since the device_id arg will result in the CPU portion being moved to + # GPU. This is useful in cases where part of the module may be parallelized + # by another algorithm and may already be on GPU. We'd like to enforce device_id + # to not be None, otherwise we'd flatten parameters in a mixed module which is + # not supported. + if len(devices) == 2 and torch.device("cpu") in devices: + if device_id is None: + raise RuntimeError( + "To support a module with both CPU and GPU params, " + "please pass in device_id argument." + ) + elif len(devices) > 1: + raise RuntimeError( + f"FSDP only supports single device modules but got params on {devices}" + ) + + +def _get_device_from_device_id( + device_id: int | torch.device | None, + rank: int, + device_handle: _FSDPDeviceHandle, +) -> torch.device | None: + """ + Return a ``torch.device`` for the specified ``device_id``. + + Processes ``device_id`` and returns either the corresponding device or + ``None`` if ``device_id`` is ``None``. + """ + if device_id is None: + return None + device = ( + device_id if isinstance(device_id, torch.device) else torch.device(device_id) + ) + if device.type != "cpu" and device.index is None: + warnings.warn( + f"FSDP got the argument `device_id` {device_id} on rank " + f"{rank}, which does not have an explicit index. " + f"FSDP will use the current device {device_handle.current_device()}. " + f"If this is incorrect, please explicitly call `torch.{device.type}.set_device()` " + "before FSDP initialization or pass in the explicit device " + "index as the `device_id` argument.", + stacklevel=2, + ) + device = torch.device(device_handle.current_device()) + return device + + +def _need_to_materialize_module( + module: nn.Module, + ignored_params: set[nn.Parameter], + ignored_modules: set[nn.Module], +) -> tuple[bool, bool]: + """ + Return if ``module`` has parameters on meta device and if ``module`` is using torchdistX deferred initialization. + + At most of the returned bools can + be ``True``. If either is ``True``, then ``module`` needs to be + materialized. + """ + managed_params = list(_get_orig_params(module, ignored_params)) + is_meta_module = any(param.is_meta for param in managed_params) + # TODO: We need to establish a contract for FSDP and buffers. For now, we + # skip checking for meta buffers from ignored modules. We should consider + # refactoring the initialization holistically to avoid so many traversals. + for submodule in module.modules(): + if submodule in ignored_modules: + continue + for buf in submodule.buffers(recurse=False): + is_meta_module |= buf.is_meta + is_torchdistX_deferred_init = ( + not is_meta_module + and _TORCHDISTX_AVAIL + and any(fake.is_fake(param) for param in managed_params) + ) + return is_meta_module, is_torchdistX_deferred_init + + +def _materialize_with_param_init_fn( + root_module: nn.Module, + param_init_fn: Callable[[nn.Module], None], + ignored_modules: set[nn.Module], +) -> None: + if not callable(param_init_fn): + raise ValueError( + f"Expected {param_init_fn} to be callable but got {type(param_init_fn)}" + ) + modules_to_materialize = _get_modules_to_materialize(root_module, ignored_modules) + for module in modules_to_materialize: + param_init_fn(module) + + +def _materialize_meta_module( + root_module: nn.Module, + device_from_device_id: torch.device | None, + ignored_modules: set[nn.Module], + device_handle: _FSDPDeviceHandle, +): + # Run default meta device initialization + materialization_device = device_from_device_id or torch.device( + device_handle.current_device() + ) + modules_to_materialize = _get_modules_to_materialize(root_module, ignored_modules) + module = None + try: + # Assume that each module's `reset_parameters()` only initializes its + # own parameters and not those of its children + with torch.no_grad(): + for module in modules_to_materialize: + # As a contract to the user, only call `reset_parameters()` if + # the module has directly managed parameters/buffers + module_state_iter = itertools.chain( + module.parameters(recurse=False), + # pyrefly: ignore [bad-argument-type] + module.buffers(recurse=False), + ) + has_module_states = len(list(module_state_iter)) > 0 + if has_module_states: + module.to_empty(device=materialization_device, recurse=False) + module.reset_parameters() # type: ignore[operator] + except BaseException as e: + warnings.warn( + "Unable to call `reset_parameters()` for module on meta " + f"device with error {str(e)}. Please ensure that your module of" + f"type {type(module)} implements a `reset_parameters()` method.", + stacklevel=2, # type: ignore[possibly-undefined] + ) + raise e + + +def _get_modules_to_materialize( + root_module: nn.Module, ignored_modules: set[nn.Module] +) -> list[nn.Module]: + # Run BFS to collect the modules to materialize via `reset_parameters()`, + # stopping at any module with FSDP already applied or at ignored modules. + modules_to_materialize: list[nn.Module] = [] + queue = collections.deque([root_module]) + visited_modules: set[nn.Module] = {root_module} + while queue: + module = queue.popleft() + modules_to_materialize.append(module) + for child_module in module.children(): + if ( + child_module not in visited_modules + and _get_module_fsdp_state(child_module) is None + and child_module not in ignored_modules + ): + visited_modules.add(child_module) + queue.append(child_module) + return modules_to_materialize + + +def _move_module_to_device( + module: nn.Module, + ignored_params: set[nn.Parameter], + ignored_buffers: set[torch.Tensor], + device_from_device_id: torch.device | None, +) -> None: + """ + Move ``module`` depending on ``device_from_device_id`` and its current device. + + This includes moving ignored modules' parameters. + + - If ``device_from_device_id`` is not ``None``, then this moves + ``module`` to the device. + - If ``device_from_device_id`` is ``None``, then this does not move + ``module`` but warns the user if it is on CPU. + + Precondition: ``_check_single_device_module()``. + """ + cpu_device = torch.device("cpu") + if device_from_device_id is not None: + # BFS from `module` without traversing any nested FSDP instances to + # collect the parameters/buffers that have not yet been managed + queue: collections.deque[nn.Module] = collections.deque() + queue.append(module) + params: list[nn.Parameter] = [] + buffers: list[torch.Tensor] = [] + while queue: + curr_module = queue.popleft() + # NOTE: We include a check to only move parameters/buffers that are + # on CPU device. If they are on a CUDA device different from the + # one specified by `device_id`, then this does NOT move them. This + # is so that we can raise an error in `_get_compute_device()`. + params.extend( + param + for param in curr_module.parameters(recurse=False) + if param.device == cpu_device + ) + buffers.extend( + buffer + for buffer in curr_module.buffers(recurse=False) + if buffer.device == cpu_device + ) + for submodule in curr_module.children(): + if not isinstance(submodule, fsdp_file.FullyShardedDataParallel): + queue.append(submodule) + params_to_move = [p for p in params if p not in ignored_params] + bufs_to_move = [p for p in buffers if p not in ignored_buffers] + _move_states_to_device(params_to_move, bufs_to_move, device_from_device_id) + return + param = next(_get_orig_params(module, ignored_params), None) + if param is not None and param.device == cpu_device: + _warn_cpu_init() + + +def _move_states_to_device( + params: list[nn.Parameter], + buffers: list[torch.Tensor], + device_from_device_id: torch.device | None, +) -> None: + """ + Move states to the specified device. + + Precondition: ``_check_single_device_module()`` and module's parameters and + buffers have been materialized if needed. + """ + if len(params) == 0 and len(buffers) == 0: + return + if len(params) > 0: + current_device = params[0].device + elif len(buffers) > 0: + current_device = buffers[0].device + cpu_device = torch.device("cpu") + if device_from_device_id is not None: + # Move the parameters and buffers like the `.data` code path in + # `nn.Module._apply()`, which underlies `nn.Module.to()` + for param in params: + with torch.no_grad(): + param.data = param.to(device_from_device_id) + if param.grad is not None: + param.grad.data = param.grad.to(device_from_device_id) + for buffer in buffers: + buffer.data = buffer.to(device_from_device_id) + elif current_device == cpu_device: # type: ignore[possibly-undefined] + _warn_cpu_init() + + +def _warn_cpu_init(): + warnings.warn( + "The passed-in `module` is on CPU and will thus have FSDP's sharding " + "initialization run on CPU, which may be slower than on GPU. We " + "recommend passing in the `device_id` argument for FSDP to move " + "`module` to GPU for the sharding initialization. `module` must also " + "be on GPU device to work with the `sync_module_states=True` flag " + "since that requires GPU communication.", + stacklevel=2, + ) + + +def _get_compute_device( + module: nn.Module, + ignored_params: set[nn.Parameter], + device_from_device_id: torch.device | None, + rank: int, + device_handle: _FSDPDeviceHandle, +) -> torch.device: + """ + Determine and return this FSDP instance's compute device. + + If the module is already on a non-CPU device, then the compute device is that non-CPU + device. If the module is on CPU, then the compute device is the current + device. + + Since this method should be called after materializing the module, any + non-CPU device should not be meta device. For now, the compute device is + always a CUDA or CUDA-like device with its explicit index. + + Precondition: ``_check_single_device_module()`` and + ``_move_module_to_device()``. + """ + param = next(_get_orig_params(module, ignored_params), None) + if param is not None and param.device.type != "cpu": + compute_device = param.device # Determined by model param placement + else: + compute_device = torch.device(device_handle.current_device()) + if device_from_device_id is not None and compute_device != device_from_device_id: + raise ValueError( + f"Inconsistent compute device and `device_id` on rank {rank}: " + f"{compute_device} vs {device_from_device_id}" + ) + return compute_device + + +# TODO: See how to deprecate! +def _sync_module_params_and_buffers( + module: nn.Module, + params: list[nn.Parameter], + process_group: dist.ProcessGroup, +) -> None: + """ + Synchronize module states (i.e. parameters ``params`` and all not-yet-synced buffers) by broadcasting from rank 0 to all ranks. + + Precondition: ``sync_module_states == True`` and ``self.process_group`` has + been set. + """ + module_states: list[torch.Tensor] = [] + for buffer in module.buffers(): + # Avoid re-synchronizing buffers in case of nested wrapping + if not getattr(buffer, FSDP_SYNCED, False): + setattr(buffer, FSDP_SYNCED, True) + detached_buffer = buffer.detach() + if is_traceable_wrapper_subclass(detached_buffer): + # NOTE: Here we assume no nested subclasses, at most one level of subclass + # in both model's buffers and params + attrs, _ = detached_buffer.__tensor_flatten__() # type: ignore[attr-defined] + for attr in attrs: + match getattr(detached_buffer, attr): + case torch.Tensor() as v: + module_states.append(v) + case CustomClassBase(): + pass + case unexpected: + raise AssertionError( + f"expected Tensor or CustomClassBase, got {type(unexpected)}" + ) + else: + module_states.append(detached_buffer) + + for param in params: + detached_param = param.detach() + if is_traceable_wrapper_subclass(detached_param): + attrs, _ = detached_param.__tensor_flatten__() # type: ignore[attr-defined] + for attr in attrs: + match getattr(detached_param, attr): + case torch.Tensor() as v: + module_states.append(v) + case CustomClassBase(): + pass + case unexpected: + raise AssertionError( + f"expected Tensor or CustomClassBase, got {type(unexpected)}" + ) + else: + module_states.append(detached_param) + + _check_module_states_for_sync_module_states(module_states) + _sync_params_and_buffers( + process_group, + module_states, + PARAM_BROADCAST_BUCKET_SIZE, + src=0, + ) + + +def _check_module_states_for_sync_module_states( + module_states: list[torch.Tensor], +) -> None: + if module_states and any( + tensor.device == torch.device("cpu") for tensor in module_states + ): + raise ValueError( + "The module has CPU parameters or buffers when `sync_module_states=True`, " + "which requires them to be on GPU. Please specify the `device_id` argument " + "or move the module to GPU before passing it to FSDP." + ) + + +def _get_orig_params( + module: nn.Module, + ignored_params: set[nn.Parameter], +) -> Iterator[nn.Parameter]: + """ + Return an iterator over the original parameters in ``module``. + + The iterator does not return + the parameters in ``ignored_params``, any ``FlatParameter`` s (which may be + present due to nested FSDP wrapping), or any original parameters already + flattened (only relevant when ``use_orig_params=True``). + """ + param_gen = module.parameters() + try: + while True: + param = next(param_gen) + if param not in ignored_params and not _is_fsdp_flattened(param): + yield param + except StopIteration: + pass + + +def _check_orig_params_flattened( + fsdp_module, + ignored_params: set[nn.Parameter], +) -> None: + """ + Check that original parameters in ``fsdp_module`` have been flattened. + + The flattened parameters are made + invisible to ``named_parameters()`` for the module hierarchy rooted at + ``fsdp_module``. This should be called as a sanity check after flattening + the wrapped module's parameters. + """ + for param_name, param in _named_parameters_with_duplicates(fsdp_module): + if param not in ignored_params and not _is_fsdp_flattened(param): + raise RuntimeError( + f"Found an unflattened parameter: {param_name}; " + f"{param.size()} {param.__class__}" + ) + + +def _get_default_comm_hook(sharding_strategy: ShardingStrategy): + return ( + default_hooks.allreduce_hook + if sharding_strategy == ShardingStrategy.NO_SHARD + else default_hooks.reduce_scatter_hook + ) + + +def _get_default_comm_hook_state( + process_group: dist.ProcessGroup, +) -> default_hooks.DefaultState: + return default_hooks.DefaultState(process_group=process_group) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_limiter_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_limiter_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..cac9c9108b3955b5bcb3b253363b13ec59efbd54 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_limiter_utils.py @@ -0,0 +1,32 @@ +import collections + +import torch + + +class _FreeEventQueue: + """ + This tracks all pending frees corresponding to inflight all-gathers. The + queueing pattern is iterative enqueues with a single dequeue per iteration + once the limit ``_max_num_inflight_all_gathers`` is reached. + """ + + def __init__(self) -> None: + self._queue: collections.deque[torch.Event] = collections.deque() + self._max_num_inflight_all_gathers = 2 # empirically chosen + + def enqueue(self, free_event: torch.Event) -> None: + """Enqueues a free event.""" + self._queue.append(free_event) + + def dequeue_if_needed(self) -> torch.Event | None: + """Dequeues a single event if the limit is reached.""" + if len(self._queue) >= self._max_num_inflight_all_gathers: + return self._dequeue() + return None + + def _dequeue(self) -> torch.Event | None: + """Dequeues a free event if possible.""" + if self._queue: + event = self._queue.popleft() + return event + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_optim_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_optim_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..aa56041f8b1522794ded35d3945c778ac3376d7d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_optim_utils.py @@ -0,0 +1,2110 @@ +# mypy: allow-untyped-defs +import copy +import functools +import logging +import warnings +from collections.abc import Iterable, Iterator, Sequence +from contextlib import ExitStack +from dataclasses import dataclass, field +from itertools import chain +from typing import Any, cast, NamedTuple, no_type_check, TYPE_CHECKING + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.nn as nn +from torch.distributed._state_dict_utils import _gather_state_dict +from torch.distributed.distributed_c10d import _get_pg_default_device +from torch.distributed.fsdp._common_utils import ( + _apply_to_modules, + _FSDPState, + _get_module_fsdp_state_if_fully_sharded_module, + _get_param_to_fqns, + _module_handle, + _named_parameters_with_duplicates, + clean_tensor_name, +) +from torch.distributed.fsdp._debug_utils import SimpleProfiler +from torch.distributed.fsdp._flat_param import FlatParameter, FlatParamHandle +from torch.distributed.fsdp._fsdp_extensions import ( + _ext_chunk_dtensor, + _ext_chunk_tensor, +) +from torch.distributed.fsdp._runtime_utils import ( + _lazy_init, + _reset_flat_param_grad_info_if_needed, +) +from torch.distributed.fsdp.api import ( + ShardingStrategy, + StateDictSettings, + StateDictType, +) +from torch.distributed.tensor import DTensor, Replicate +from torch.utils._pytree import tree_map_only + + +if TYPE_CHECKING: + from torch.distributed._shard.sharded_tensor import ShardedTensor + + +logger = logging.getLogger(__name__) + + +@dataclass +class FSDPParamInfo: + state: _FSDPState + handle: FlatParamHandle + param_indices: dict[str, int] + param_requires_grad: list[bool] + + +def sorted_items(dictionary: dict[str, Any]) -> Iterator[tuple[str, Any]]: + keys = sorted(dictionary.keys()) + for k in keys: + yield k, dictionary[k] + + +@dataclass +class _ConsolidatedOptimState: + """ + This holds the consolidated optimizer state on the target rank. Positive- + dimension tensor state is communicated across ranks, while zero-dimension + tensor state and non-tensor state is taken directly from the target rank. + + PyTorch version 1.12 moved to using zero-dimension tensors for scalar + values, but user implemented optimizers may still use float (i.e. a + non-tensor). Thus, we support both and handle them identically. + + Attributes: + tensor_state (Dict[str, torch.Tensor]): Mapping from positive-dimension + tensor state name to the unsharded flat tensor representing the + state. + zero_dim_tensor_state (Dict[str, torch.Tensor]): Mapping from zero- + dimension tensor state name to its value. + non_tensor_state (Dict[str, Any]): Mapping from non-tensor state + name to its value. + """ + + tensor_state: dict[str, torch.Tensor] = field(default_factory=dict) + zero_dim_tensor_state: dict[str, torch.Tensor] = field(default_factory=dict) + non_tensor_state: dict[str, Any] = field(default_factory=dict) + + +class _PosDimTensorInfo(NamedTuple): + """ + Metadata for positive-dimension tensors used internally for + :meth:`scatter_full_optim_state_dict`. + + Attributes: + shape (torch.Size): Sharded tensor shape (which is equal to the + unsharded tensor shape if the tensor is optimizer state for a + non-FSDP parameter and is hence not sharded). + dtype (torch.dtype): Data type of the tensor. + """ + + shape: torch.Size + dtype: torch.dtype + + +class _OptimStateKey(NamedTuple): + """ + This represents an optimizer state key that may be used commonly across + ranks. It is based on the unflattened parameter names rather than parameter + IDs to make it independent of each rank's own optimizer construction. + """ + + unflat_param_names: tuple[str, ...] + is_fsdp_managed: bool + + +def _unflatten_optim_state( + fsdp_param_info: FSDPParamInfo, + flat_param_state: dict[str, Any], + to_save: bool, + shard_state: bool, + cpu_offload: bool, +) -> list[dict[str, Any]]: + """ + Unflattens the optimizer state, consisting of the "state" part and the + "param_groups" part. Unflattening the "state" part involves consolidating + the state on the target rank and remapping from flattened to unflattened + parameter IDs, and the "param_groups" part only involves remapping from + flattened to unflattened parameter IDs. + + Args: + fsdp_param_info (FSDPParamInfo): The FSDP state, the handle, and a + mapping from FQN to original parameter index. + flat_param_state (Dict[str, Any]): Entry for the flat parameter in the + "state" part of the optimizer state dict. + to_save (bool): Whether to save the state on this rank. + + Returns: + List[Dict[str, Any]]: A :class:`list` holding the entries in the + "state" part of the optimizer state dict corresponding to the + unflattened parameters comprising the flat parameter if on the target + rank or an empty :class:`list` otherwise. The final optimizer state + dict will need to map these entries using the proper unflattened + parameter IDs. + """ + if shard_state and not to_save: + raise AssertionError("If ``shard_state`` is True, ``to_save`` has to be True.") + consolidated_state = _communicate_optim_state( + fsdp_param_info, + flat_param_state, + ) + if to_save: + unflat_param_state = _unflatten_communicated_optim_state( + fsdp_param_info, + consolidated_state, + shard_state, + ) + for optim_state in unflat_param_state: + # We can't use .items() below cuz we'd run into a concurrent modification error + if cpu_offload: + for key in list(optim_state.keys()): + state = optim_state[key] + if not isinstance(state, torch.Tensor): + continue + optim_state[key] = state.cpu() + return unflat_param_state + else: + return [] + + +def _is_zero_dim_tensor(x: Any) -> bool: + return torch.is_tensor(x) and x.dim() == 0 + + +def _communicate_optim_state( + fsdp_param_info: FSDPParamInfo, + flat_param_state: dict[str, Any], +) -> _ConsolidatedOptimState: + """ + Communicates the optimizer state for a flat parameter across ranks. All + ranks will hold the entire non-sharded optimizer state on GPU. + + If ``N`` is the number of tensor optimizer states in the optimizer state + dict, then the communication complexity is 0 if ``N = 0`` and ``N + 1`` + otherwise (where the plus 1 comes from all-gathering the padding per rank). + + Args: + fsdp_param_info (FSDPParamInfo): The FSDP state, the handle, and a + mapping from FQN to original parameter index. + flat_param_state (Dict[str, Any]): The entry in the "state" part of the + optimizer state dict corresponding to the flat parameter. + + Returns: + ConsolidatedOptimState: Consolidated optimizer state for the target + flat parameter. + """ + fsdp_state = fsdp_param_info.state + flat_param = fsdp_param_info.handle.flat_param + state = _ConsolidatedOptimState() + tensor_state, zero_dim_tensor_state, non_tensor_state = ( + state.tensor_state, + state.zero_dim_tensor_state, + state.non_tensor_state, + ) + + for state_name, value in sorted_items(flat_param_state): + # Positive-dimension tensor state: communicate across ranks + if torch.is_tensor(value) and value.dim() > 0: + # If the parameter is not sharded, then neither is the + # positive-dimension tensor state, so no need to communicate it -- + # we take the target rank's value + if ( + fsdp_state.world_size == 1 + or fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD + ): + tensor_state[state_name] = value + continue + if fsdp_state.compute_device is None: + raise AssertionError("compute_device has not been initialized") + if value.device.type != fsdp_state.compute_device.type: + value = value.to(fsdp_state.compute_device) + # Assume that positive-dimension tensor optimizer state + # has the same shape as the sharded flat parameter + buffer_size = flat_param._full_param_padded.size() # type: ignore[attr-defined] + tensor_buffer = value.new_zeros(*buffer_size) + dist.all_gather_single(tensor_buffer, value, group=fsdp_state.process_group) + fsdp_state._device_handle.synchronize() + unpadded_numel = cast( + nn.Parameter, flat_param._unpadded_unsharded_size + ).numel() + tensor_state[state_name] = tensor_buffer[:unpadded_numel] + # Zero-dimension tensor state and non-tensor state: take this rank's + # value directly + else: + if _is_zero_dim_tensor(value): + zero_dim_tensor_state[state_name] = value.detach().clone() + else: + non_tensor_state[state_name] = value + return state + + +def _unflatten_communicated_optim_state( + fsdp_param_info: FSDPParamInfo, + state: _ConsolidatedOptimState, + shard_state: bool, +) -> list[dict[str, Any]]: + """ + Unflattens the communicated optimizer state (given by ``tensor_state``, + ``non_tensor_state``, and ``zero_dim_tensor_state``) for a single flat + parameter. This should only be called on the target rank. + + Args: + fsdp_param_info (FSDPParamInfo): The FSDP state, the handle, and a + mapping from FQN to original parameter index. + state (_ConsolidatedOptimState): Consolidated optimizer state. + + Returns: + List[Dict[str, Any]]: A :class:`list` holding the entries in the + "state" part of the optimizer state dict corresponding to the + unflattened parameters comprising the flat parameter. The final + optimizer state dict will need to map these entries using the proper + unflattened parameter IDs. + """ + fsdp_state = fsdp_param_info.state + handle = fsdp_param_info.handle + flat_param = handle.flat_param + unflat_param_state: list[dict[str, Any]] = [] + flat_param_views: dict[str, Iterator] = {} + num_unflat_params = flat_param._num_params + tensor_state, zero_dim_tensor_state, non_tensor_state = ( + state.tensor_state, + state.zero_dim_tensor_state, + state.non_tensor_state, + ) + + for _ in range(num_unflat_params): + unflat_state_param = {} + # Add positive-dimension tensor state: unflatten with views + for state_name, flat_tensor in sorted_items(tensor_state): + views_generated = state_name in flat_param_views + if not views_generated: + views = handle._get_unflat_views(flat_tensor) + flat_param_views[state_name] = views + else: + views = flat_param_views[state_name] + optim_state: torch.Tensor | ShardedTensor | DTensor = next(views) + if shard_state: + osd_config = fsdp_state._optim_state_dict_config + if getattr(osd_config, "_use_dtensor", False): + if fsdp_state._device_mesh is None: + raise AssertionError( + f"Expected _device_mesh to be not None, got {fsdp_state._device_mesh}" + ) + optim_state = _ext_chunk_dtensor( + optim_state, + fsdp_state.rank, + fsdp_state._device_mesh, + fsdp_state._fsdp_extension, + ) + else: + if fsdp_state.process_group is None: + raise AssertionError( + f"Expected process_group to be not None, got {fsdp_state.process_group}" + ) + optim_state = _ext_chunk_tensor( + optim_state, + fsdp_state.rank, + fsdp_state.world_size, + fsdp_state._device_handle.device_count(), + fsdp_state.process_group, + fsdp_state._fsdp_extension, + ) + unflat_state_param[state_name] = optim_state + + # Add zero-dimension tensor state: take the target rank's value + unflat_state_param.update(sorted_items(zero_dim_tensor_state)) + # Add non-tensor state: take the target rank's value + unflat_state_param.update(sorted_items(non_tensor_state)) + unflat_param_state.append(unflat_state_param) + return unflat_param_state + + +def _broadcast_processed_state( + fsdp_state: _FSDPState, + optim_state: dict[str, Any], + group: dist.ProcessGroup | None, +) -> dict[str, Any]: + objects: list[Any] = [None] + if dist.get_rank(group) == 0: + objects[0] = tree_map_only( + torch.Tensor, + lambda v: v.cpu() if v.dim() == 0 else _PosDimTensorInfo(v.shape, v.dtype), # type: ignore[union-attr] + optim_state, + ) + dist.broadcast_object_list(objects, src=0, group=group) + if dist.get_rank(group) == 0: + return optim_state + else: + return objects[0] + + +def _broadcast_state( + fsdp_state: _FSDPState, state: Any, group: dist.ProcessGroup | None +) -> Any: + if dist.get_rank(group) == 0: + if not isinstance(state, torch.Tensor) or state.dim() == 0: + return state + tensor = state.to(fsdp_state.compute_device) + else: + if isinstance(state, torch.Tensor): + if state.dim() != 0: + raise AssertionError( + "For non-zero ranks, a tensor state should have zero dimension, " + f"but got the state with shape {state.shape}." + ) + return state + elif not isinstance(state, _PosDimTensorInfo): + return state + tensor = torch.zeros( + state.shape, dtype=state.dtype, device=fsdp_state.compute_device + ) + dist.broadcast(tensor, src=0, group=group) + return tensor + + +def _shard_orig_param_state( + fsdp_param_info: FSDPParamInfo, + fqn: str, + optim_state: dict[str, Any], +) -> dict[str, Any]: + """ + Shard the optimizer state for the original parameter with the name ``fqn``. + This API should only be used when ``use_orig_params`` is True. + """ + if not optim_state: + return {} + fsdp_state = fsdp_param_info.state + flat_param = fsdp_param_info.handle.flat_param + param_idx = fsdp_param_info.param_indices[fqn] + shard_param_info = flat_param._shard_param_infos[param_idx] # type: ignore[attr-defined] + optim_state = _gather_state_dict( + optim_state, pg=fsdp_state.process_group, device=fsdp_state.compute_device + ) + if not shard_param_info.in_shard: + return {} + # Flatten and shard the state. + new_optim_state: dict[str, Any] = {} + intra_param_start_idx = shard_param_info.intra_param_start_idx + intra_param_end_idx = shard_param_info.intra_param_end_idx + for state_name, value in optim_state.items(): + if ( + torch.is_tensor(value) + and value.dim() > 0 + and fsdp_state.sharding_strategy != ShardingStrategy.NO_SHARD + ): + value = value.flatten()[ + intra_param_start_idx : intra_param_end_idx # type: ignore[operator] + + 1 + ].clone() + new_optim_state[state_name] = value + return new_optim_state + + +def _flatten_optim_state_dict( + optim_state_dict: dict[str, Any], + model: nn.Module, + use_orig_params: bool = False, + optim: torch.optim.Optimizer | None = None, + rank0_only: bool = False, + group: dist.ProcessGroup | None = None, +) -> dict[str, Any]: + """ + Flattens the full optimizer state dict, still keying by unflattened parameter + names. + + If ``use_orig_params`` is True, each rank will have all FSDP-managed + parameters but some of these parameters may be empty due to the sharding. + For a regular optim.Optimizer, states for those empty parameters will + not be initialized. So, when aggregating the FQNs across ranks, no assert + will be raised on a rank even if it does not have all the states -- it is + valid and FSDP know how to aggregate them. However, FSDP has to ignore + handling those parameters that are not managed by FSDP and do not exist on + the local rank -- it is managed by other parallelism and FSDP does not + know ho to handle/aggregate them. + + Note that ``_flatten_tensor_optim_state`` does not need ``optim`` to + flatten/shard the state. However, NamedOptimizer and KeyedOptimizer require + all the states even if the corresponding parameters are empty. To this end, + ``optim`` will be used to get the initial state of the empty parameters. + ``optim`` should only be non-None if the ``optim` is KeyedOptimizer or + NamedOptimizer. + + Returns: + Dict[str, Any]: The flattened optimizer state dict. + """ + SimpleProfiler.reset() + + unflat_osd = optim_state_dict + if "state" not in unflat_osd and not rank0_only: + raise ValueError( + '`optim_state_dict` must have the keys "state"' + "to be a valid optimizer state dict" + ) + param_to_fqns = _get_param_to_fqns(model) + fqn_to_fsdp_param_info = _get_fqn_to_fsdp_param_info(model) + fsdp_state = next(iter(fqn_to_fsdp_param_info.values())).state + + # Broadcast unflat_osd without non-scalar tensor if rank0_only is True. + if rank0_only: + unflat_osd = _broadcast_processed_state(fsdp_state, unflat_osd, group=group) + + # Construct the "state" part + flat_osd_state: dict[_OptimStateKey | str, Any] = {} + unflat_osd_state = unflat_osd["state"] + all_state_keys = set(unflat_osd_state.keys()) + + for param, fqns in param_to_fqns.items(): + fqn = fqns[0] + if fqn not in unflat_osd_state: + continue + all_state_keys.difference_update(fqns) + + if rank0_only: + for fqn in fqns: + if not unflat_osd_state[fqn]: + continue + for state_name in unflat_osd_state[fqn]: + unflat_osd_state[fqn][state_name] = _broadcast_state( + fsdp_state, unflat_osd_state[fqn][state_name], group=group + ) + fqn = fqns[0] + if fqn in fqn_to_fsdp_param_info: + fsdp_param_info = fqn_to_fsdp_param_info[fqn] + if use_orig_params: + with SimpleProfiler.profile(SimpleProfiler.Type.RESHARDING): + flat_state = _shard_orig_param_state( + fsdp_param_info, + fqn, + unflat_osd_state[fqn], + ) + else: + flat_state = _flatten_optim_state( + fsdp_param_info, + unflat_osd_state, + fqns, + ) + key = _OptimStateKey(tuple(fqns), True) + # Only include non-empty states since as expected by + # `torch.optim.Optimizer` s unless the optimizer is KeyedOptimizer + # or NamedOptimizer. + if flat_state: + flat_osd_state[key] = flat_state + elif use_orig_params: + if len(fqns) != 1: + raise AssertionError( + f"use_orig_params is True but there are multiple FQNs, {fqns}." + ) + if optim is not None: # NamedOptimizer or KeyedOptimizer case. + state = optim.state.get(param, None) # type: ignore[call-overload] + if state is not None: + flat_osd_state[key] = copy.deepcopy(state) + else: + warnings.warn( + f"optim_state[{key}] is not on rank{fsdp_state.rank}.", + stacklevel=2, + ) + + else: + raise RuntimeError( + f"The state of {key} is empty. This should happen when " + "use_orig_params=True." + ) + else: # do not flatten non-FSDP parameters' states + if len(fqns) != 1: + raise AssertionError(f"Expected len(fqns) == 1, got {len(fqns)}") + key = _OptimStateKey(tuple(fqns), False) + flat_osd_state[key] = copy.copy(unflat_osd_state[fqn]) + + if rank0_only: + for fqn in fqns: + if not unflat_osd_state[fqn]: + continue + for state_name, param_state in list(unflat_osd_state[fqn].items()): + if fsdp_state.rank > 0: + # Deference the tensor so that PyTorch can collect the memory. + del unflat_osd_state[fqn][state_name] + else: + # Move the tensor in the original osd back to CPU to make the + # original osd unaffected. + unflat_osd_state[fqn][state_name] = param_state.cpu() + + # Handle user-defined state, states that are not associated with parameters. + for key in all_state_keys: + user_state = unflat_osd_state[key] + if isinstance(user_state, torch.Tensor) and rank0_only and use_orig_params: + user_state = _broadcast_state(fsdp_state, user_state, group=group) + flat_osd_state[key] = copy.copy(user_state) + + SimpleProfiler.dump_and_reset("FSDP _flatten_optim_state_dict() profiling: ") + # Construct the "param_groups" part -- copy as is since it will be + # rekeyed later according to the target rank's optimizer + # Only copy param_groups if it exists in unflat_osd + if "param_groups" in unflat_osd: + flat_osd_param_groups = copy.deepcopy(unflat_osd["param_groups"]) + return {"state": flat_osd_state, "param_groups": flat_osd_param_groups} + else: + return {"state": flat_osd_state} + + +def _flatten_optim_state( + fsdp_param_info: FSDPParamInfo, + unflat_osd_state: dict[str, dict[str, Any]], + unflat_param_names: list[str], +) -> dict[str, Any]: + """ + Flattens the optimizer state in ``full_optim_state_dict`` for a single + flat parameter in ``fsdp_param_info`` corresponding to the unflattened + parameter names in ``unflat_param_names``. + + Args: + fsdp_param_info (FSDPParamInfo): The FSDP state, the handle, and a + mapping from FQN to original parameter index. + unflat_osd_state (Dict[str, Dict[str, Any]]): The "state" part of the + optimizer state dict corresponding to the unflattened parameters. + unflat_param_names (List[str]): A :class:`list` of unflattened + parameter names corresponding to the flat parameter ``flat_param``. + + Returns: + Dict[str, Any]: A :class:`dict` mapping state names to their values for + a particular flat parameter. The sharded optimizer state dict's "state" + part will map a key to this returned value. + """ + fsdp_state = fsdp_param_info.state + handle = fsdp_param_info.handle + flat_param = handle.flat_param + num_unflat_params = len(unflat_param_names) + if num_unflat_params <= 0: + raise AssertionError( + "Expects at least one unflattened parameter corresponding to the flat parameter" + ) + unflat_param_shapes = flat_param._shapes + num_unflat_param_shapes = len(unflat_param_shapes) + if num_unflat_params != num_unflat_param_shapes: + raise AssertionError( + f"Expects {num_unflat_params} shapes but got {num_unflat_param_shapes}" + ) + + # Check if these unflattened parameters have any optimizer state + has_state = [ + bool(unflat_param_name in unflat_osd_state) + for unflat_param_name in unflat_param_names + ] + # If none of the unflattened parameters comprising this flat parameter have + # any state, then we do not want an entry in the optimizer state dict + if not any(has_state): + return {} # no need to flatten any state + # There may still be some unflattened parameters with state and some + # without + unflat_param_states = [ + _gather_state_dict( + unflat_osd_state[unflat_param_name], + pg=fsdp_state.process_group, + device=fsdp_state.compute_device, + ) + if unflat_param_name in unflat_osd_state + else None + for unflat_param_name in unflat_param_names + ] + # Check that the unflattened parameters have the same state names + state_names = None + # pyrefly: ignore [bad-assignment] + for unflat_param_state in unflat_param_states: + if unflat_param_state is None: + continue + if state_names is None: + state_names = set(unflat_param_state.keys()) + else: + if state_names != set(unflat_param_state.keys()): + raise ValueError( + "Differing optimizer state names for the unflattened " + f"parameters: {unflat_param_names}" + ) + if state_names is None: + raise AssertionError(f"Expected state_names to be not None, got {state_names}") + + # Flatten the state + flat_state: dict[str, torch.Tensor | None] = {} + for state_name in state_names: + state_values = [ + unflat_param_state[state_name] if unflat_param_state is not None else None + for unflat_param_state in unflat_param_states + ] + non_none_state_values = [v for v in state_values if v is not None] + # If all ranks have None, this is a None value + if not non_none_state_values: + flat_state[state_name] = None + continue + are_pos_dim_tensors = are_zero_dim_tensors = are_non_tensors = True + for v in non_none_state_values: + are_pos_dim_tensors &= torch.is_tensor(v) and v.dim() > 0 + are_zero_dim_tensors &= _is_zero_dim_tensor(v) + are_non_tensors &= not torch.is_tensor(v) + types = {type(v) for v in non_none_state_values} + if len(types) != 1 or not ( + are_pos_dim_tensors or are_zero_dim_tensors or are_non_tensors + ): + raise ValueError( + f"Differing optimizer state types for state {state_name}, " + f"values {non_none_state_values}, and unflattened parameter " + f"names {unflat_param_names}" + ) + if are_pos_dim_tensors: + flat_tensor = _flatten_tensor_optim_state( + state_name, + state_values, # type: ignore[arg-type] + unflat_param_names, + unflat_param_shapes, + handle, + ) + # Shard the flattened tensor immediately to minimize max memory + # usage + if ( + fsdp_state.world_size != 1 + and fsdp_state.sharding_strategy != ShardingStrategy.NO_SHARD + ): + sharded_flat_tensor, _ = FlatParamHandle._get_shard( + flat_tensor, + fsdp_state.rank, + fsdp_state.world_size, + ) + else: + sharded_flat_tensor = flat_tensor + flat_state[state_name] = sharded_flat_tensor + elif are_zero_dim_tensors: + flat_state[state_name] = _flatten_zero_dim_tensor_optim_state( + state_name, + state_values, # type: ignore[arg-type] + unflat_param_names, + ) + else: + if not are_non_tensors: + raise AssertionError( + f"Expected are_non_tensors to be True, got {are_non_tensors}" + ) + flat_state[state_name] = _flatten_non_tensor_optim_state( + state_name, + state_values, + unflat_param_names, + ) + + return flat_state + + +def _flatten_tensor_optim_state( + state_name: str, + pos_dim_tensors: list[torch.Tensor], + unflat_param_names: list[str], + unflat_param_shapes: Sequence[torch.Size], + handle: FlatParamHandle, +) -> torch.Tensor: + """ + Flattens the positive-dimension tensor optimizer state given by the values + ``tensors`` for the state ``state_name`` for a single flat parameter + from ``handle`` corresponding to the unflattened parameter names + ``unflat_param_names`` and unflatted parameter shapes + ``unflat_param_shapes``. This flattens each unflattened parameter's tensor + state into one tensor. + + NOTE: We use zero tensors for any unflattened parameters without state + since some value is required to fill those entries. This assumes that the + zero tensor is mathematically equivalent to having no state, which is true + for Adam's "exp_avg" and "exp_avg_sq" but may not be true for all + optimizers. + + Args: + state_name (str): Optimizer state name. + pos_dim_tensors (List[torch.Tensor]): Positive-dimension tensor + optimizer state values for the unflattened parameters corresponding + to the single flat parameter. + unflat_param_names (List[str]): A :class:`list` of unflattened + parameter names corresponding to the single flat parameter. + unflat_param_shapes (List[torch.Size]): Unflattened parameter shapes + corresponding to the single flat parameter. + handle (FlatParamHandle): The flat parameter's handle. + + Returns: + torch.Tensor: A flat tensor containing the optimizer state + corresponding to ``state_name`` constructed by concatenating the + unflattened parameter tensor states in ``pos_dim_tensors`` (using zero + tensors for any unflattened parameters without the state). + """ + flat_param = handle.flat_param + non_none_tensors = [t for t in pos_dim_tensors if t is not None] + # Check that all are tensors with the same dtype + dtypes = {t.dtype for t in non_none_tensors} + if len(dtypes) != 1: + raise ValueError( + "All unflattened parameters comprising a single flat " + "parameter must have positive-dimension tensor state with the " + f"same dtype but got dtypes {dtypes} for state {state_name} and " + f"unflattened parameter names {unflat_param_names}" + ) + dtype = next(iter(dtypes)) + # Check that each tensor state matches its parameter's shape + for tensor, shape in zip(pos_dim_tensors, unflat_param_shapes): + if tensor is None and len(shape) == 0: + raise ValueError("Flattening a zero-dimension parameter is not supported") + elif tensor is not None and tensor.shape != shape: + raise ValueError( + "Tensor optimizer state does not have same shape as its " + f"parameter: {tensor.shape} {shape}" + ) + # Flatten the tensor states: we do not need to add any right-hand-side + # padding since the flat optimizer state tensor is sharded via + # `_get_shard()`, which pads the shard as needed (just like for the flat + # parameter) + cpu_device = torch.device("cpu") + tensors_to_flatten = [ + torch.flatten(state_value.to(cpu_device)) + if state_value is not None + else torch.flatten( + torch.zeros( + size=shape, + dtype=dtype, + device=cpu_device, + ) + ) + for state_value, shape in zip(pos_dim_tensors, unflat_param_shapes) + ] + flat_tensor = handle.flatten_tensors(tensors_to_flatten, handle._aligned_numel) + flat_param_shape = flat_param._unpadded_unsharded_size # type: ignore[attr-defined] + if flat_tensor.shape != flat_param_shape: + raise AssertionError( + f"tensor optim state: {flat_tensor.shape} flat parameter: {flat_param_shape}" + ) + return flat_tensor + + +def _flatten_zero_dim_tensor_optim_state( + state_name: str, + zero_dim_tensors: list[torch.Tensor], + unflat_param_names: list[str], +) -> torch.Tensor: + """ + Flattens the zero-dimension tensor optimizer state given by the values + ``zero_dim_tensors`` for the state ``state_name`` for a single flat + parameter corresponding to the unflattened parameter names + ``unflat_param_names`` by enforcing that all tensors are the same and using + that common value. + + NOTE: The requirement that the tensors are the same across all unflattened + parameters comprising the flat parameter is needed to maintain the + invariant that FSDP performs the same computation as its non-sharded + equivalent. This means that none of the unflattened parameters can be + missing this state since imposing a value may differ from having no value. + For example, for Adam's "step", no value means maximum bias correction, + while having some positive value means less bias correction. + + Args: + state_name (str): Optimizer state name. + zero_dim_tensors (List[torch.Tensor]): Zero-dimension optimizer state + for the unflattened parameters corresponding to the single + flat parameter. + unflat_param_names (List[str]): A :class:`list` of unflattened + parameter names corresponding to the single flat parameter. + + Returns: + torch.Tensor: A zero-dimensional tensor giving the value of the state + ``state_name`` for all unflattened parameters corresponding to the + names ``unflat_param_names``. + """ + non_none_tensors = [t for t in zero_dim_tensors if t is not None] + # Enforce that all have the same value and dtype + values_set = {t.item() if t is not None else None for t in zero_dim_tensors} + dtypes = {t.dtype if t is not None else None for t in zero_dim_tensors} + if ( + len(non_none_tensors) != len(zero_dim_tensors) + or len(values_set) != 1 + or len(dtypes) != 1 + ): + raise ValueError( + "All unflattened parameters comprising a single flat " + "parameter must have scalar state with the same value and dtype " + f"but got values {values_set} and dtypes {dtypes} for state " + f"{state_name} and unflattened parameter names " + f"{unflat_param_names}" + ) + value = next(iter(values_set)) + dtype = next(iter(dtypes)) + return torch.tensor(value, dtype=dtype, device=torch.device("cpu")) + + +def _flatten_non_tensor_optim_state( + state_name: str, + non_tensors: list[Any], + unflat_param_names: list[str], +) -> Any: + """ + Flattens the non-tensor optimizer state given by the values ``non_tensors`` + for the state ``state_name`` for a single flat parameter corresponding + to the unflattened parameter names ``unflat_param_names`` by enforcing that + all values are the same and using that common value. + + See the note in :func:`_flatten_zero_dim_tensor_optim_state`. + + Args: + state_name (str): Optimizer state name. + non_tensors (List[Any]): Non-tensor optimizer state for the unflattened + parameters corresponding to the single flat parameter. + unflat_param_names (List[str]): A :class:`list` of unflattened + parameter names corresponding to the single flat parameter. + + Returns: + Any: A non-tensor giving the value of the state ``state_name`` for all + unflattened parameters corresponding to the names + ``unflat_param_names``. + """ + non_none_non_tensors = [nt for nt in non_tensors if nt is not None] + # Enforce that all have the same value (same type already checked) + non_tensor_set = set(non_tensors) + if len(non_none_non_tensors) != len(non_tensors) or len(non_tensor_set) != 1: + raise ValueError( + "All unflattened parameters comprising a single flat " + "parameter must have scalar state with the same value and dtype " + f"but got values {non_tensor_set} for state {state_name} and " + f"unflattened parameter names {unflat_param_names}" + ) + non_tensor = next(iter(non_tensor_set)) + return non_tensor + + +def _rekey_sharded_optim_state_dict( + sharded_osd: dict[str, Any], + model: nn.Module, + optim: torch.optim.Optimizer, + optim_input: list[dict[str, Any]] | Iterable[nn.Parameter] | None, + using_optim_input: bool, + is_named_optimizer: bool = False, +) -> dict[str, Any]: + """ + Rekeys the optimizer state dict from unflattened parameter names to flat + parameter IDs according to the calling rank's ``optim``, which may be + different across ranks. In particular, the unflattened parameter names are + represented as :class:`_OptimStateKey` s. + """ + param_to_fqns = _get_param_to_fqns(model) + flat_param_to_fqn = _get_flat_param_to_fqn(model) + param_to_param_key: dict[nn.Parameter, int | str] = cast( + dict[nn.Parameter, int | str], + ( + _get_param_to_param_id_from_optim_input(model, optim_input) + if using_optim_input + else _get_param_to_param_key( + optim, model, is_named_optimizer, param_to_fqns, flat_param_to_fqn + ) + ), + ) + # All parameter keys in `param_to_param_key` should be in + # `param_to_fqns` -- strict inequality follows when not all parameters are + # passed to the optimizer + if len(param_to_param_key) > len(param_to_fqns): + raise AssertionError( + f"Expected len(param_to_param_key) <= len(param_to_fqns), got {len(param_to_param_key)} > {len(param_to_fqns)}" + ) + + unflat_param_names_to_flat_param_key: dict[ + tuple[str, ...], int | str + ] = {} # for "state" + unflat_param_name_to_flat_param_key: dict[str, int | str] = {} # for "param_groups" + for param, unflat_param_names in param_to_fqns.items(): + if param not in param_to_param_key: + # This parameter was not passed to the optimizer + continue + flat_param_key = param_to_param_key[param] + unflat_param_names_to_flat_param_key[tuple(unflat_param_names)] = flat_param_key + for unflat_param_name in unflat_param_names: + unflat_param_name_to_flat_param_key[unflat_param_name] = flat_param_key + + sharded_osd_state = sharded_osd["state"] + rekeyed_osd_state: dict[str | int, Any] = {} + for key, param_state in sharded_osd_state.items(): + if isinstance(key, str): + rekeyed_osd_state[key] = param_state + continue + flat_param_key = unflat_param_names_to_flat_param_key.get( + key.unflat_param_names, key.unflat_param_names + ) + + rekeyed_osd_state[flat_param_key] = param_state + + # Only process param_groups if it exists in sharded_osd + if "param_groups" in sharded_osd: + rekeyed_osd_param_groups: list[dict[str, Any]] = [] + for unflat_param_group in sharded_osd["param_groups"]: + flat_param_group = copy.deepcopy(unflat_param_group) + flat_param_keys = sorted( + { + unflat_param_name_to_flat_param_key[unflat_param_name] + for unflat_param_name in unflat_param_group["params"] + } + ) + flat_param_group["params"] = flat_param_keys + rekeyed_osd_param_groups.append(flat_param_group) + return {"state": rekeyed_osd_state, "param_groups": rekeyed_osd_param_groups} + else: + return {"state": rekeyed_osd_state} + + +def _get_param_id_to_param_from_optim_input( + model: nn.Module, + optim_input: list[dict[str, Any]] | Iterable[nn.Parameter] | None = None, +) -> dict[int, nn.Parameter]: + """ + Constructs a mapping from parameter IDs to parameters. This may be used + both for models with ``FlatParameter`` s and without. + + NOTE: This method is only preserved for backward compatibility. The method + :meth:`_get_param_key_to_param` is the preferred code path that does not + rely on ``optim_input``. + + NOTE: We critically assume that, whether the optimizer input is a list of + parameters or a list of parameter groups, :class:`torch.optim.Optimizer` + enumerates the parameter IDs in order. In other words, for a parameter list + input, the parameter IDs should be in that list order, and for a parameter + groups input, the parameter IDs should be in order within each parameter + group and in order across parameter groups. + + Args: + model (nn.Module): Model whose parameters are passed into the + optimizer. + optim_input (Optional[Union[List[Dict[str, Any]], + Iterable[nn.Parameter]]]): Input passed into the optimizer + representing either a :class:`list` of parameter groups or an + iterable of parameters; if ``None``, then this method assumes the + input was ``model.parameters()``. (Default: ``None``) + + Returns: + List[nn.Parameter]: Mapping from parameter IDs to parameters, + where the parameter ID is implicitly the index in the :class:`list`. + """ + # Assume the standard case of passing `model.parameters()` to the optimizer + # if `optim_input` is not specified + if optim_input is None: + return dict(enumerate(model.parameters())) + try: + # pyrefly: ignore [redundant-cast] + params = cast(list[nn.Parameter], list(optim_input)) + except TypeError as e: + raise TypeError( + "Optimizer input should be an iterable of Tensors or dicts, " + f"but got {optim_input}" + ) from e + if len(params) == 0: + raise ValueError("Optimizer input should not be empty") + + # Check if the optimizer input represents tensors or parameter groups + all_tensors = True + all_dicts = True + for param in params: + all_tensors &= isinstance(param, torch.Tensor) + all_dicts &= isinstance(param, dict) + if not all_tensors and not all_dicts: + raise TypeError("Optimizer input should be an iterable of Tensors or dicts") + if all_tensors: + return dict(enumerate(params)) + if not all_dicts: + raise AssertionError(f"Expected all_dicts to be True, got {all_dicts}") + param_id_to_param: list[nn.Parameter] = [] + for param_group in params: + has_params_key = "params" in param_group # type: ignore[operator] + if not has_params_key: + raise AssertionError( + 'A parameter group should map "params" to a list of the parameters in the group' + ) + # Implicitly map `flat_param_id` (current length of the list) to + # `param` + param_id_to_param.extend(param_group["params"]) # type: ignore[index] + return dict(enumerate(param_id_to_param)) + + +def _get_flat_param_to_fqn(model: torch.nn.Module) -> dict[FlatParameter, str]: + """ + Constructs a mapping from ``FlatParameter`` to a cleaned (devoid of prefixes + from wrappers) fully qualified name (FQN). Note that this FQN is "non-canonical" + because ``FlatParameter`` s do not come from the original module but are + registered only after FSDP has been applied. This function returns the FSDP-given + name for the ``FlatParameter`` (usually module._flat_param) as opposed to the + canonical FQNs returned for ``FlatParameter`` s in ``_common_utils._get_param_to_fqns(...)``). + + Consequently, this function will only return a non-empty mapping if FSDP was + applied with ``use_orig_params=False`` as, otherwise, the original parameters + are used within the module and there would be no ``FlatParameter`` s in the module. + + """ + + def module_fn(module, prefix, tree_level, flat_param_to_fqn): + for param_name, param in _named_parameters_with_duplicates( + module, recurse=False + ): + if not isinstance(param, FlatParameter): + continue + fqn = clean_tensor_name(prefix + param_name) + flat_param_to_fqn[param] = fqn + + def return_fn(flat_param_to_fqn): + return flat_param_to_fqn + + flat_param_to_fqn_ret: dict[FlatParameter, str] = {} + return _apply_to_modules( + model, + module_fn, + return_fn, + [fqn for fqn, _ in _named_parameters_with_duplicates(model)], + flat_param_to_fqn_ret, + ) + + +def _get_param_key_to_param( + optim: torch.optim.Optimizer, + model: nn.Module | None = None, + is_named_optimizer: bool = False, + param_to_fqns: dict[nn.Parameter, list[str]] | None = None, + flat_param_to_fqn: dict[FlatParameter, str] | None = None, +) -> dict[int | str, nn.Parameter]: + """ + Constructs a mapping from parameter keys to parameters. For the regular + optimizers, the keys are parameter IDs. For NamedOptimizer, the keys + are FQNs. This API may be used both for models with ``FlatParameter`` s and + without. + """ + clean_fqn_to_curr_fqn: dict[str, str] = {} + if is_named_optimizer: + if param_to_fqns is None or flat_param_to_fqn is None: + raise AssertionError( + "The optimizer is a NamedOptimizer, `param_to_fqns` must not be None." + ) + if model is None: + raise AssertionError(f"Expected model to be not None, got {model}") + for key, _ in _named_parameters_with_duplicates(model): + clean_fqn_to_curr_fqn[clean_tensor_name(key)] = key + + param_key_to_param: dict[str | int, nn.Parameter] = {} + pid = 0 + for param_group in optim.param_groups: + if is_named_optimizer: + for param in param_group["params"]: + if flat_param_to_fqn is None: + raise AssertionError( + f"Expected flat_param_to_fqn to be not None, got {flat_param_to_fqn}" + ) + if param in flat_param_to_fqn: + # FlatParameter case + key = flat_param_to_fqn[param] + else: + if param_to_fqns is None: + raise AssertionError( + f"Expected param_to_fqns to be not None, got {param_to_fqns}" + ) + # use_orig_params case + if len(param_to_fqns[param]) != 1: + raise AssertionError( + f"Expected len(param_to_fqns[param]) == 1, got {len(param_to_fqns[param])}" + ) + key = param_to_fqns[param][0] + try: + key = clean_fqn_to_curr_fqn[key] + except KeyError as e: + raise KeyError( + f"Can't find {key} from {list(clean_fqn_to_curr_fqn.keys())}." + ) from e + param_key_to_param[key] = param + else: + for param in param_group["params"]: + param_key_to_param[pid] = param + pid += 1 + + return param_key_to_param + + +def _get_param_to_param_key( + optim: torch.optim.Optimizer, + model: nn.Module | None = None, + is_named_optimizer: bool = False, + param_to_fqns: dict[nn.Parameter, list[str]] | None = None, + flat_param_to_fqn: dict[FlatParameter, str] | None = None, +) -> dict[nn.Parameter, int | str]: + """ + Constructs the inverse mapping of :func:`_get_param_key_to_param`. This API + only supports the case where `optim` is a regular optimizer, not NamedOptimizer. + So the parameter keys will be parameter ids. + """ + param_id_to_param = _get_param_key_to_param( + optim, model, is_named_optimizer, param_to_fqns, flat_param_to_fqn + ) + return {param: param_id for param_id, param in param_id_to_param.items()} + + +def _get_param_to_param_id_from_optim_input( + model: nn.Module, + optim_input: list[dict[str, Any]] | Iterable[nn.Parameter] | None = None, +) -> dict[nn.Parameter, int]: + """Constructs the inverse mapping of :func:`_get_param_id_to_param_from_optim_input`.""" + param_id_to_param = _get_param_id_to_param_from_optim_input(model, optim_input) + return {param: param_id for param_id, param in param_id_to_param.items()} + + +def _check_missing_keys_on_rank( + r0_optim_state_keys: list[_OptimStateKey], + optim_state_key_to_param_key: dict[_OptimStateKey, str | int], + param_key_to_param: dict[str | int, nn.Parameter], + group: dist.ProcessGroup | None, +) -> None: + # Ensure that all ranks have at least the optimizer states needed by + # rank 0's optimizer + missing_keys: list[_OptimStateKey] = [] + for r0_optim_state_key in r0_optim_state_keys: + if r0_optim_state_key not in optim_state_key_to_param_key: + # A parameter from rank 0's optimizer does not exist for this + # rank's optimizer + missing_keys.append(r0_optim_state_key) + continue + param_key = optim_state_key_to_param_key[r0_optim_state_key] + if isinstance(param_key, int): + if not (param_key >= 0 and param_key < len(param_key_to_param)): + raise AssertionError("Check the `param_key_to_param` construction") + # We cannot use FSDPState.compute_device as this API is a global view. + device = _get_pg_default_device(group) + num_missing = torch.tensor([len(missing_keys)], dtype=torch.int32, device=device) + dist.all_reduce(num_missing, group=group) + if num_missing.item() > 0: + obj_list = [None for _ in range(dist.get_world_size(group))] + dist.all_gather_object(obj_list, missing_keys, group=group) + error_msg = ( + "FSDP currently requires each rank to have at least the " + "optimizer states needed by rank 0's optimizer but some ranks " + "are missing some of those states" + ) + for rank, keys in enumerate(obj_list): + keys = cast(list[_OptimStateKey], keys) + if len(keys) > 0: + error_msg += ( + f"\nRank {rank} is missing states for the parameters: " + f"{[key.unflat_param_names for key in keys]}" + ) + raise RuntimeError(error_msg) + + +def _map_param_key_to_optim_keys( + optim_state_dict: dict[str, Any], + group: dist.ProcessGroup | None, + param_key_to_param: dict[int | str, nn.Parameter], + param_to_fqns: dict[nn.Parameter, list[str]], + fqn_to_fsdp_param_info: dict[str, FSDPParamInfo], + merge_keys: bool = False, +) -> tuple[list[_OptimStateKey], dict[_OptimStateKey, int | str]]: + """ + Construct the local mapping between the ``_OptimStateKey`` and parameter keys + and all the ``_OptimStateKey`` across ranks. If ``merge_keys`` is False, rank0 + must contain all the ``_OptimStateKey``, an exception will be raised otherwise. + Note that ``merge_keys`` should equal to ``use_orig_params``. + """ + rank = dist.get_rank(group) + optim_state_key_to_param_key: dict[_OptimStateKey, int | str] = {} # local + all_optim_state_keys: list[_OptimStateKey] = [] + + for param_key, param in param_key_to_param.items(): + # Do not include parameters without state to avoid empty mappings + # just like in normal `torch.optim.Optimizer.state_dict()` + if param_key not in optim_state_dict["state"]: + continue + fqns = param_to_fqns[param] + is_fsdp_managed = isinstance(param, FlatParameter) + if is_fsdp_managed: + if fqns[0] not in fqn_to_fsdp_param_info: + raise AssertionError( + f"Expected {fqns[0]} to be in fqn_to_fsdp_param_info, got keys: {list(fqn_to_fsdp_param_info.keys())}" + ) + is_fsdp_managed = fqns[0] in fqn_to_fsdp_param_info + optim_state_key = _OptimStateKey( + unflat_param_names=tuple(fqns), + is_fsdp_managed=is_fsdp_managed, + ) + if rank == 0 or merge_keys: + all_optim_state_keys.append(optim_state_key) + optim_state_key_to_param_key[optim_state_key] = param_key + + if merge_keys: + all_keys: list[list[_OptimStateKey]] = [ + [] for _ in range(dist.get_world_size(group)) + ] + dist.all_gather_object(all_keys, all_optim_state_keys, group=group) + merge_all_optim_state_keys = [*chain.from_iterable(all_keys)] + all_optim_state_keys = sorted(set(merge_all_optim_state_keys)) + else: + key_obj_list: list[list[_OptimStateKey] | None] = ( + [all_optim_state_keys] if rank == 0 else [None] + ) + dist.broadcast_object_list(key_obj_list, src=0, group=group) + if key_obj_list[0] is None: + raise AssertionError( + f"Expected key_obj_list[0] to be not None, got {key_obj_list[0]}" + ) + all_optim_state_keys = key_obj_list[0] + _check_missing_keys_on_rank( + all_optim_state_keys, + optim_state_key_to_param_key, + param_key_to_param, + group, + ) + + return all_optim_state_keys, optim_state_key_to_param_key + + +def _unflatten_param_groups( + state_dict: dict[str, Any], + param_key_to_param: dict[int | str, nn.Parameter], + param_to_fqns: dict[nn.Parameter, list[str]], +) -> list[dict[str, Any]]: + param_groups: list[dict[str, Any]] = [] + for flat_param_group in state_dict["param_groups"]: + unflat_param_group = copy.deepcopy(flat_param_group) + param_group_params = [ + param_key_to_param[flat_param_key] + for flat_param_key in flat_param_group["params"] + ] + nested_unflat_param_names = [ + param_to_fqns[param] for param in param_group_params + ] + unflat_param_group["params"] = [ + *chain.from_iterable(nested_unflat_param_names) + ] # flatten the list of lists + param_groups.append(unflat_param_group) + return param_groups + + +def _is_named_optimizer(optim_state_dict: dict[str, Any]) -> bool: + """ + Returns whether the state_dict is from a NamedOptimizer. + This function checks that the keys in the state_dict['state'] are strings + (which usually are FQNs) versus integers (which usually refer to param_ids + from a vanilla torch.optim.Optimizer). + """ + state = optim_state_dict.get("state") + if not state: + # If we cannot find a state, assume it is not NamedOptimizer as + # NamedOptimizer has eager initialization. + return False + try: + key = next(iter(state.keys())) + except Exception as e: + raise Exception(optim_state_dict) from e # noqa: TRY002 + return isinstance(key, str) + + +@dataclass +class StateInfo: + # The key of these dictionaries are the state name, e.g., `exp_avg`. + tensors: dict[str, _PosDimTensorInfo] + scalar_tensors: dict[str, torch.Tensor] + non_tensors: dict[str, Any] + + +def _allgather_state_info( + fsdp_state: _FSDPState, + input_states: dict[str, Any], +) -> list[dict[str, StateInfo]]: + """ + Given the ``input_states``, allgather StateInfo for each state. The function + uses all_gather_object to gather StateInfo so no GPU tensors are sent. + """ + + processed_state_dict: dict[str, StateInfo] = {} + gathered_state_info: list[dict[str, StateInfo]] = [ + {} for _ in range(fsdp_state.world_size) + ] + + for fqn, optim_state in input_states.items(): + # Allgather the scalar tensor state, non-tensor states and tensors metadata. + processed_state = StateInfo({}, {}, {}) + for state_name, value in sorted_items(optim_state): + if torch.is_tensor(value): + if value.dim() == 0: + # Ensure that `step` is on CPU. + processed_state.scalar_tensors[state_name] = value.cpu() + else: + processed_state.tensors[state_name] = _PosDimTensorInfo( + value.shape, value.dtype + ) + else: + processed_state.non_tensors[state_name] = value + processed_state_dict[fqn] = processed_state + dist.all_gather_object( + gathered_state_info, + processed_state_dict, + group=fsdp_state.process_group, + ) + return gathered_state_info + + +def _convert_all_state_info( + fsdp_param_info: FSDPParamInfo, + gathered_state_info: list[dict[str, StateInfo]], + input_states: dict[str, Any], + output_states: dict[str, dict[str, Any]], +) -> tuple[torch.dtype | None, dict[str, list[torch.Tensor | None]]]: + """ + Given the ``gathered_state_info`` and ``input_states``, the API converted + the StateInfo into the original state if the state is not a non-scalar + tensor. For a multi-dimensional tensor, the local state will be stored in + ``state_buffer`` in a correct order for later allgather purpose. + """ + + state_buffers: dict[str, list[torch.Tensor | None]] = {} + + for fqn, gathered_state in output_states.items(): + state_info = [s[fqn] for s in gathered_state_info] + all_tensor_states = sorted({n for state in state_info for n in state.tensors}) + empty_ranks: set[int] = set() + dtype: torch.dtype | None = None + # First check all the non-scalar states and get the information of + # states on each rank. + for state_name in all_tensor_states: + numels = [] + _empty_ranks: set[int] = set() + for rank, object_state in enumerate(state_info): + numels.append(0) + info = object_state.tensors.get(state_name, None) + if info is not None: + numels[-1] = info.shape.numel() + if not dtype: + dtype = info.dtype + else: + if dtype != info.dtype: + raise AssertionError( + f"Expected dtype == info.dtype, got {dtype} != {info.dtype}" + ) + if numels[-1] == 0: + _empty_ranks.add(rank) + + if not (not empty_ranks or empty_ranks == _empty_ranks): + raise AssertionError( + f"Expected empty_ranks to be empty or equal to _empty_ranks, got {empty_ranks} vs {_empty_ranks}" + ) + empty_ranks = _empty_ranks + if state_name not in state_buffers: + state_buffers[state_name] = [ + None for _ in fsdp_param_info.param_indices + ] + local_state = input_states[fqn].get(state_name, None) + # N.B. We need to move the state to compute_device. The reason is + # not yet clear and we need to figure out why the state may be on a + # different device. + if local_state is not None: + local_state = local_state.to(fsdp_param_info.state.compute_device) + state_buffers[state_name][fsdp_param_info.param_indices[fqn]] = local_state + + # Restoring the scalar and non-tensor states. If the corresponding + # non-scalar states do not exist on the rank, we also skip the scalar + # non-tensor states on that rank. + for rank, object_state in enumerate(state_info): + if rank in empty_ranks: + continue + for name, non_tensor_value in object_state.non_tensors.items(): + curr_non_tensor_value = gathered_state.get(name, None) + if not ( + curr_non_tensor_value is None + or curr_non_tensor_value == non_tensor_value + ): + raise AssertionError( + f"Rank {rank} has different values for {name}: {non_tensor_value}." + + f" Other ranks: {curr_non_tensor_value}" + ) + gathered_state[name] = non_tensor_value + + for name, scalar_tensor_value in object_state.scalar_tensors.items(): + curr_scalar_tensor_value = gathered_state.get(name, None) + if not ( + curr_scalar_tensor_value is None + or torch.equal(scalar_tensor_value, curr_scalar_tensor_value) + ): + raise AssertionError( + f"Rank {rank} has different values for {name}: {scalar_tensor_value}." + + f" Other ranks: {curr_scalar_tensor_value}" + ) + gathered_state[name] = scalar_tensor_value + + return dtype, state_buffers # type: ignore[possibly-undefined] + + +def _unflatten_orig_param_states( + fsdp_param_info: FSDPParamInfo, + output_states: dict[str, dict[str, Any]], + state_name: str, + shard_state: bool, + to_save: bool, + cpu_offload: bool, +) -> None: + """ + Given an output state dict, ``output_states``, which the keys are FQNs to the + original parameters (not FlatParameters nor parameter ID), and the values + are gathered states, unflatten the states to the original dimensions. + + This function performs the unflattening process in-place. + """ + if not to_save: + return + flat_param = fsdp_param_info.handle.flat_param + fsdp_state = fsdp_param_info.state + for fqn, gathered_state in output_states.items(): + value = gathered_state[state_name] + param_idx = fsdp_param_info.param_indices[fqn] + + # TODO: This solution is not general and only apply to PTD TP solution. + if isinstance(value, DTensor): + placement = value.placements[0] + # If gathered state is a DTensor and its TP placement is not Replicate(), we need to + # gather the tensor on its TP dimension before chunking them into DTensor again. + if placement != Replicate(): + placement_dim = placement.dim # type: ignore[attr-defined] + value.redistribute(placements=(Replicate(),)) + reshape_size = list(flat_param._shapes[param_idx]) + reshape_size[placement_dim] *= value.device_mesh.size(0) + reshape_size = torch.Size(reshape_size) + value = value.reshape(reshape_size) + # If gathered state is a replicate DTensor, we directly reshape it. + else: + value = value.reshape(flat_param._shapes[param_idx]) + else: + # If gathered state is a tensor, we directly reshape it into unflatten state. + value = value.reshape(flat_param._shapes[param_idx]) + + if shard_state: + osd_config = fsdp_state._optim_state_dict_config + if getattr(osd_config, "_use_dtensor", False): + if fsdp_state._device_mesh is None: + raise AssertionError( + f"Expected _device_mesh to be not None, got {fsdp_state._device_mesh}" + ) + value = _ext_chunk_dtensor( + value, + fsdp_state.rank, + fsdp_state._device_mesh, + fsdp_state._fsdp_extension, + ) + else: + if fsdp_state.process_group is None: + raise AssertionError( + f"Expected process_group to be not None, got {fsdp_state.process_group}" + ) + value = _ext_chunk_tensor( + value, + fsdp_state.rank, + fsdp_state.world_size, + fsdp_state._device_handle.device_count(), + fsdp_state.process_group, + fsdp_state._fsdp_extension, + ) + elif not cpu_offload: + with SimpleProfiler.profile("clone"): + value = value.detach().clone() + + if cpu_offload: + with SimpleProfiler.profile(SimpleProfiler.Type.D2H): + value = value.cpu() + gathered_state[state_name] = value + + +def _allgather_orig_param_states( + fsdp_param_info: FSDPParamInfo, + gathered_state_info: list[dict[str, StateInfo]], + input_states: dict[str, Any], + shard_state: bool, + to_save: bool, + cpu_offload: bool, +) -> dict[str, dict[str, Any]]: + """ + Given the ``gathered_state_info`` and ``input_states``, the API allgathers + all tensor states and restore non-tensor states from ``gathered_state_info``. + """ + fsdp_state = fsdp_param_info.state + if fsdp_state.rank == 0 and dist.get_debug_level() == dist.DebugLevel.DETAIL: + logger.info( + "Memory Summary before calling to _allgather_orig_param_states %s", + fsdp_state._device_handle.memory_summary(), + ) + + output_states: dict[str, dict[str, Any]] = {fqn: {} for fqn in input_states} + + dtype, state_buffers = _convert_all_state_info( + fsdp_param_info, gathered_state_info, input_states, output_states + ) + + if len(state_buffers) == 0: + return output_states + + has_state_params: list[bool] = [ + fqn in output_states for fqn in fsdp_param_info.param_indices + ] + + # Loop through the ``state_buffers`` and construct the flattened, concatenated, + # sharded states. The size of the constructed state will be the same size as + # flat_param (also sharded). + # Then we perform an allgather_into_tensor to get the full flat_param state. + # The full flat_param state is the result of concatenation of multiple states + # the order of flat_param._fqns. + # The final step is to split the flat_param state into original param states + # and return the result. + flat_param = fsdp_param_info.handle.flat_param + empty_func = functools.partial( + torch.empty, dtype=dtype, device=fsdp_state.compute_device + ) + gathered_tensor = empty_func(flat_param._padded_unsharded_size) + # Synchronize can be slow but this will be easier for us to debug. + fsdp_state._device_handle.synchronize() + for state_name, buffers in state_buffers.items(): + local_buffers: list[torch.Tensor] = [] + begin = fsdp_state.rank * flat_param._sharded_size.numel() + # End is inclusive. + end = begin + flat_param._sharded_size.numel() - 1 + # param_idx corresponds to the parameter index in the FlatParameter. + mem_offset, param_idx = 0, 0 + for numel, is_padding in zip( + flat_param._numels_with_padding, flat_param._is_padding_mask + ): + frozen_and_no_state = not is_padding and ( + not fsdp_param_info.param_requires_grad[param_idx] + and not has_state_params[param_idx] + ) + + if is_padding or frozen_and_no_state: + # This memory range is a padding or the param is frozen and does + # not require gradient. For the later case, we treat it as a + # padding and add empty values to the local_buffers. + + padding_begin, padding_end = mem_offset, mem_offset + numel - 1 + if padding_begin <= begin <= padding_end: + # The range is an align padding before the first parameter in + # the shard. The shard includes parts of this align padding. + padding_len = ( + padding_end - begin + 1 + if end >= padding_end + else end - begin + 1 + ) + elif padding_begin <= end <= padding_end: + # The range is an align padding after the last parameter in + # the shard. The shard includes parts of this align padding. + padding_len = ( + end - padding_begin + 1 + if begin <= padding_begin + else end - begin + 1 + ) + elif begin < padding_begin <= padding_end < end: + # The range is an align padding that is completely in the + # shard. + padding_len = numel + else: + padding_len = 0 + if padding_len: + local_buffers.append(empty_func(padding_len)) + + if not is_padding: + # This memory range is a parameter in FlatParameter. So there + # should be an corresponding state in the optimizer unless the + # parameter is frozen, which we treat it as a padding above. + + # We need to check if this rank owns the buffer. If this is None: + # 1.) the rank does not own any part of the original parameter. + # As a result, there is no corresponding optimizer state on + # the rank as well. + # 2.) the parameter is frozen AND no optimizer state for the + # parameter. If a parameter is frozen, there can still be + # optimizer state if the parameter is not frozen in the + # previous steps. + if buffers[param_idx] is not None: + local_buffers.append(cast(torch.Tensor, buffers[param_idx])) + param_idx += 1 + + mem_offset += numel + + shard_numel_padded = flat_param._sharded_size.numel() - ( + sum(t.numel() for t in local_buffers) + ) + + if flat_param._shard_numel_padded != shard_numel_padded: + raise AssertionError( + "Manually calculated _sharded_numel_padded is incorrect. " + f"_shard_numel_padded={flat_param._shard_numel_padded}, " + f"shard_numel_padded={shard_numel_padded}, " + f"_sharded_size.numel={flat_param._sharded_size.numel()}, " + f"_numels_with_padding={flat_param._numels_with_padding}, " + f"begin={begin}, end={end}," + ) + if shard_numel_padded > 0: + # Add right-handed padding. + local_buffers.append(empty_func(shard_numel_padded)) + local_shard = torch.cat(local_buffers) + if local_shard.numel() * fsdp_state.world_size != gathered_tensor.numel(): + raise AssertionError( + "The size of local shard times the world size should equal to the " + "gathered tensor size. The inconsistency may be from a bug of " + "FlatParameter's metadata or the reconstruction logic in optimizer " + "state dict." + ) + fsdp_state._device_handle.synchronize() + with SimpleProfiler.profile(SimpleProfiler.Type.ALLGATHER): + dist.all_gather_single( + gathered_tensor, local_shard, group=fsdp_state.process_group + ) + # Synchronize can be slow but this will be easier for us to debug. + fsdp_state._device_handle.synchronize() + + unpadded_tensor = gathered_tensor[: flat_param._unpadded_unsharded_size.numel()] + flat_param_handle = fsdp_param_info.handle + orig_states = flat_param_handle._get_unflat_views_aligned(unpadded_tensor) + if len(orig_states) != len(fsdp_param_info.param_indices): + raise AssertionError( + "The number of parameters from FlatParameter is not consistent to " + "the number of states used by optimizer state dict reconstruction " + "logic." + ) + for fqn, idx in fsdp_param_info.param_indices.items(): + if fsdp_param_info.param_requires_grad[idx] or fqn in output_states: + output_states[fqn][state_name] = orig_states[idx] + + _unflatten_orig_param_states( + fsdp_param_info, + output_states, + state_name, + shard_state, + to_save, + cpu_offload, + ) + + del gathered_tensor + return output_states + + +def _gather_all_orig_param_state( + fsdp_param_info: FSDPParamInfo, + input_states: dict[str, Any], + shard_state: bool, + to_save: bool, + cpu_offload: bool, +) -> dict[str, Any]: + """ + Given a optimizer state dict, ``input_states``, which the keys are FQNs to the + original parameters (not FlatParameters nor parameter ID), gather all the + states and unflatten them to the original dimensions. Note that all the + params referred by the ``input_states`` must be managed by FSDP. + """ + fsdp_state = fsdp_param_info.state + if ( + fsdp_state.world_size == 1 + or fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD + ): + return input_states if to_save else {} + + with SimpleProfiler.profile(SimpleProfiler.Type.RESHARDING): + with SimpleProfiler.profile(SimpleProfiler.Type.ALLGATHER_OBJ): + gathered_state_info = _allgather_state_info(fsdp_state, input_states) + output_states = _allgather_orig_param_states( + fsdp_param_info, + gathered_state_info, + input_states, + shard_state, + to_save, + cpu_offload, + ) + if to_save: + for key, idx in fsdp_param_info.param_indices.items(): + if key in output_states: + continue + if not fsdp_param_info.param_requires_grad[idx]: + continue + + raise RuntimeError( + f"{key} is not in the output state. " + "The FSDPParamInfo has the param keys " + f"{sorted(fsdp_param_info.param_indices.keys())} while " + "the output_states has the param keys " + f"{sorted(output_states.keys())}." + ) + return output_states + else: + return {} + + +def _convert_state_with_orig_params( + all_optim_state_keys: list[_OptimStateKey], + optim_state_key_to_param_key: dict[_OptimStateKey, int | str], + fqn_to_fsdp_param_info: dict[str, FSDPParamInfo], + optim_state_dict: dict[str | int, Any], + to_save: bool, + shard_state: bool, + cpu_offload: bool = True, +) -> dict[str, Any]: + fsdp_osd_state: dict[str, Any] = {} + # This variable is used to deduplicate the FSDPParamInfo as one FSDPParamInfo + # usually corresponds to multiple parameters. We could not use FSDPParamInfo + # as the key because FSDPParamInfo is not hashable. As a result, we fall back + # to `id(FSDPParamInfo)`, which the type is an integer. + all_states: dict[int, dict[str, Any]] = {} + # Iterate in rank 0's flat parameter ID order to ensure aligned all-gathers + # across ranks + for optim_state_key in all_optim_state_keys: + param_key: str | int | None = optim_state_key_to_param_key.get(optim_state_key) + + if param_key is None and not optim_state_key.is_fsdp_managed: + continue + + if optim_state_key.is_fsdp_managed: + fqn = optim_state_key.unflat_param_names[0] + fsdp_param_info = fqn_to_fsdp_param_info.get(fqn) + if fsdp_param_info is None: + # This can happen if the not all FSDP instances have all the + # parameters. This can happen with FSDP + some MPMD style + # parallelism. + + # TODO: it is unclear if we need to do the same check with + # non-FSDP managed keys. + continue + state = {} if param_key is None else optim_state_dict[param_key] + if id(fsdp_param_info) not in all_states: + all_states[id(fsdp_param_info)] = {} + all_states[id(fsdp_param_info)][fqn] = state + + elif to_save: + if len(optim_state_key.unflat_param_names) != 1: + raise AssertionError( + f"Expected len(optim_state_key.unflat_param_names) == 1, got {len(optim_state_key.unflat_param_names)}" + ) + unflat_param_name = optim_state_key.unflat_param_names[0] + with SimpleProfiler.profile("none_fsdp_managed_copy"): + param_key = cast(str | int, param_key) + fsdp_osd_state[unflat_param_name] = copy.copy( + optim_state_dict[param_key] + ) + if cpu_offload: + for state_name, value in sorted_items( + fsdp_osd_state[unflat_param_name] + ): + if not torch.is_tensor(value): + continue + fsdp_osd_state[unflat_param_name][state_name] = value.cpu() + + # Instead of gathering the state of each parameter individually, we perform + # the gathering all at once to speed up the process. + for _all_states in all_states.values(): + fqn = next(iter(_all_states.keys())) + fsdp_param_info = fqn_to_fsdp_param_info[fqn] + if len(fsdp_param_info.param_requires_grad) <= 0: + raise AssertionError( + "With use_orig_params, FSDPParamInfo should have requires_grad " + "information. However, the length is zero." + ) + for key, idx in fsdp_param_info.param_indices.items(): + if key in _all_states: + continue + if not fsdp_param_info.param_requires_grad[idx]: + continue + raise RuntimeError( + f"{key} is not in the optimizer state. " + "The FSDPParamInfo has the param keys " + f"{sorted(fsdp_param_info.param_indices.keys())} while " + "the optimizer has the param keys " + f"{sorted(_all_states.keys())}." + ) + fsdp_osd_state.update( + _gather_all_orig_param_state( + fsdp_param_info, + _all_states, + shard_state, + to_save, + cpu_offload, + ) + ) + + return fsdp_osd_state + + +def _convert_state_with_flat_params( + all_optim_state_keys: list[_OptimStateKey], + optim_state_key_to_param_key: dict[_OptimStateKey, int | str], + fqn_to_fsdp_param_info: dict[str, FSDPParamInfo], + optim_state_dict: dict[str | int, Any], + to_save: bool, + shard_state: bool, + cpu_offload: bool = True, +) -> dict[str, Any]: + fsdp_osd_state: dict[str, Any] = {} + # Iterate in rank 0's flat parameter ID order to ensure aligned all-gathers + # across ranks + for optim_state_key in all_optim_state_keys: + param_key: str | int | None = optim_state_key_to_param_key.get(optim_state_key) + + if param_key is None: + raise AssertionError( + "If use_orig_params is False, we must be able to find the " + f"corresponding param id. {optim_state_key} {param_key}" + ) + + if optim_state_key.is_fsdp_managed: + # If there are multiple unflat_param_names (not use_orig_params), + # they share the same FSDPParamInfo. So the first unflat_param_name + # is sufficient to fetch the FSDPParamInfo. + fqn = optim_state_key.unflat_param_names[0] + fsdp_param_info = fqn_to_fsdp_param_info[fqn] + unflat_state = _unflatten_optim_state( + fsdp_param_info, + optim_state_dict[param_key], + to_save, + shard_state, + cpu_offload, + ) + if to_save: + if len(unflat_state) != len(optim_state_key.unflat_param_names): + raise AssertionError( + f"Expected len(unflat_state) == len(optim_state_key.unflat_param_names), " + f"got {len(unflat_state)} != {len(optim_state_key.unflat_param_names)}" + ) + fsdp_osd_state.update( + zip( + optim_state_key.unflat_param_names, + unflat_state, + ) + ) + elif to_save: + if len(optim_state_key.unflat_param_names) != 1: + raise AssertionError( + f"Expected len(optim_state_key.unflat_param_names) == 1, got {len(optim_state_key.unflat_param_names)}" + ) + unflat_param_name = optim_state_key.unflat_param_names[0] + fsdp_osd_state[unflat_param_name] = copy.copy(optim_state_dict[param_key]) + if cpu_offload: + for state_name, value in sorted_items( + fsdp_osd_state[unflat_param_name] + ): + if not torch.is_tensor(value): + continue + fsdp_osd_state[unflat_param_name][state_name] = value.cpu() + + return fsdp_osd_state + + +@torch.no_grad() +def _optim_state_dict( + model: nn.Module, + optim: torch.optim.Optimizer, + optim_state_dict: dict[str, Any], + optim_input: list[dict[str, Any]] | Iterable[nn.Parameter] | None, + rank0_only: bool, + shard_state: bool, + group: dist.ProcessGroup | None, + using_optim_input: bool, + use_orig_params: bool = False, + cpu_offload: bool = True, +) -> dict[str, Any]: + """ + Consolidates the optimizer state and returns it as a :class:`dict` + following the convention of :meth:`torch.optim.Optimizer.state_dict`, + i.e. with keys ``"state"`` and ``"param_groups"``. + The flat parameters in ``FSDP`` modules contained in ``model`` are mapped + back to their unflattened parameters. + + Parameter keys are not well-defined. For a regular optimizer, the optimizer + state_dict contains a mapping from parameter IDs to parameter states. + Parameter IDs are the order of parameters in ``optim.param_groups()`` across + all the groups. This API also allows user to pass ``optim_input`` for the + mapping between parameters and parameter IDs. Using ``optim_input`` is being + deprecated. + + If the optimizer is a ``NamedOptimizer``, the optimizer state_dict does not + contain parameter IDs mapping but a mapping from parameter FQNs to parameter + states. This API finds the mapping from FQNs to parameters if the optimizer + is a ``NamedOptimizer``. + + If ``use_orig_params`` is True, each rank will have all FSDP-managed + parameters but some of these parameters may be empty due to the sharding. + For a regular optim.Optimizer, states for those empty parameters will + not be initialized. So, when aggregating the FQNs across ranks, no assert + will be raised on a rank even if it does not have all the states -- it is + valid and FSDP knows how to aggregate them. However, FSDP has to ignore + handling those parameters that are not managed by FSDP and do not exist on + the local rank -- those are managed by other parallelisms and FSDP does not + know how to handle/aggregate them. + + Args: + model (nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + were passed into the optimizer ``optim``. + optim (torch.optim.Optimizer): Optimizer for ``model`` 's + parameters. + rank0_only (bool): If ``True``, saves the populated :class:`dict` + only on rank 0; if ``False``, saves it on all ranks. (Default: + ``True``) + shard_state (bool): If ``True``, shard and distribute all + non-zero-dimension states. + + Returns: + Dict[str, Any]: A :class:`dict` containing the optimizer state for + ``model`` 's original unflattened parameters and including keys + "state" and "param_groups" following the convention of + :meth:`torch.optim.Optimizer.state_dict`. If ``rank0_only=False``, + then nonzero ranks return an empty :class:`dict`. + """ + SimpleProfiler.reset() + cm = ExitStack() + cm.enter_context(SimpleProfiler.profile(SimpleProfiler.Type.ALL)) + _reset_flat_param_grad_info_if_needed(traversal_utils._get_fsdp_handles(model)) + to_save = not rank0_only or dist.get_rank(group) == 0 or shard_state + + with SimpleProfiler.profile("preprocessing"): + param_to_fqns = _get_param_to_fqns(model) + flat_param_to_fqn = _get_flat_param_to_fqn(model) + is_named_optimizer = _is_named_optimizer(optim_state_dict) + + param_key_to_param = cast( + dict[int | str, nn.Parameter], + ( + _get_param_id_to_param_from_optim_input(model, optim_input) + if using_optim_input + else _get_param_key_to_param( + optim, model, is_named_optimizer, param_to_fqns, flat_param_to_fqn + ) + ), + ) + fqn_to_fsdp_param_info = _get_fqn_to_fsdp_param_info(model) + + with SimpleProfiler.profile("preprocessing_with_comm"): + ( + all_optim_state_keys, + optim_state_key_to_param_key, + ) = _map_param_key_to_optim_keys( + optim_state_dict, + group, + param_key_to_param, + param_to_fqns, + fqn_to_fsdp_param_info, + merge_keys=use_orig_params, + ) + + with SimpleProfiler.profile("state_converting"): + convert_fn = ( + _convert_state_with_orig_params + if use_orig_params + else _convert_state_with_flat_params + ) + fsdp_osd_state = convert_fn( + all_optim_state_keys, + optim_state_key_to_param_key, + fqn_to_fsdp_param_info, + optim_state_dict["state"], + to_save, + shard_state, + cpu_offload, + ) + + # At this point, communication is complete and ranks can return early if nothing + # will be saved on that rank. + if not to_save: + return {} + + fsdp_osd: dict[str, Any] = {"state": fsdp_osd_state} + + flat_param_fqns = set(flat_param_to_fqn.values()) + for key, value in optim_state_dict["state"].items(): + if key in fsdp_osd_state: + continue + if key in flat_param_fqns: + continue + if key in param_key_to_param: + continue + # This key is not recognized by FSDP. It may be a user-defined state + # or some parameters state that FSDP is unable to map from + # ``optim.param_groups``. + warnings.warn( + f"Found a optim state, {key}, that FSDP cannot process. FSDP " + "will directly copy everything to the returned state_dict. In " + "most cases, this is a user-defined state that is not " + "associated with any particular parameter. Another possible " + "case is this state is managed by TorchRec. Otherwise, there may " + " be a mismatched assumption of optim_state_dict of this mode.", + stacklevel=2, + ) + fsdp_osd_state[key] = value + + if "param_groups" in optim_state_dict: + fsdp_osd["param_groups"] = _unflatten_param_groups( + optim_state_dict, param_key_to_param, param_to_fqns + ) + + cm.close() + SimpleProfiler.dump_and_reset("FSDP _optim_state_dict() profiling: ") + + return fsdp_osd + + +def _get_fqn_to_fsdp_param_info(model: nn.Module) -> dict[str, FSDPParamInfo]: + """ + Construct the mapping from a param's fqn to its corresponding ``FSDPParamInfo`` + if the param is managed by FSDP. Shared parameters, or original parameters that + are shared across multiple nn.Modules, are required to belong to one and only + one FSDP instance and thus correspond to one ``FlatParameter``. Within the one + ``FlatParameter``, ``FlatParameter._fqns`` only stores the first FQN of a shared + parameter. Thus, the keys in the mapping are guaranteed to map to unique parameters. + """ + + def module_fn(module, prefix, tree_level, fqn_to_param_info): + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state is None: + return + _lazy_init(fsdp_state, module) + handle = _module_handle(fsdp_state, module) + if not handle: + return + flat_param = handle.flat_param + fsdp_param_info = FSDPParamInfo(fsdp_state, handle, {}, []) + # NOTE: `idx` indexes into the data structures *without* padding + # elements + for idx, local_fqn in enumerate(flat_param._fqns): + fqn = clean_tensor_name(prefix + local_fqn) + if fqn in fqn_to_param_info: + if fqn_to_param_info[fqn].handle.flat_param is not flat_param: + raise AssertionError( + f"Expected fqn_to_param_info[fqn].handle.flat_param is flat_param for {fqn}" + ) + fqn_to_param_info[fqn] = fsdp_param_info + fsdp_param_info.param_indices[fqn] = idx + if flat_param._params is not None: + fsdp_param_info.param_requires_grad.append( + flat_param._params[idx].requires_grad + ) + + def return_fn(fqn_to_param_info): + return fqn_to_param_info + + fqn_to_param_info: dict[str, FSDPParamInfo] = {} + # FlatParameter._fqns stores the local fqn, starting from the root of the + # FSDP. Using _apply_to_modules() with model (may not be the FSDP root + # module) allows us to construct the global fqn. + return _apply_to_modules( + model, + module_fn, + return_fn, + [fqn for fqn, _ in _named_parameters_with_duplicates(model)], + fqn_to_param_info, + ) + + +@no_type_check +def _set_optim_use_dtensor( + fsdp_state: _FSDPState, + state_dict_settings: StateDictSettings, +) -> None: + # If device_mesh is passed in when initializing FSDP, we automatically turn the + # _use_dtensor flag to be true for ShardedOptimStateDictConfig() if state_dict_type + # has to be set to SHARDED_STATE_DICT. + if getattr(fsdp_state, "_device_mesh", None): + state_dict_type = state_dict_settings.state_dict_type + if state_dict_type == StateDictType.LOCAL_STATE_DICT: + raise RuntimeError( + "Found state_dict_type LOCAL_STATE_DICT. " + "DeviceMesh is not compatible with LOCAL_STATE_DICT. " + "Please set state_dict_type to SHARDED_STATE_DICT to get DTensor state_dict." + ) + else: + state_dict_settings.optim_state_dict_config._use_dtensor = True diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_runtime_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_runtime_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..e25bc75e12770df26c2a61b546976e0933885783 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_runtime_utils.py @@ -0,0 +1,1654 @@ +# mypy: allow-untyped-defs +import functools +import logging +from collections.abc import Callable +from enum import auto, Enum +from typing import Any, no_type_check + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.nn as nn +import torch.nn.functional as F +from torch.autograd import Variable +from torch.autograd.graph import register_multi_grad_hook +from torch.distributed.algorithms._comm_hooks import LOW_PRECISION_HOOKS +from torch.distributed.fsdp._common_utils import ( + _assert_in_training_states, + _FSDPState, + _get_module_fsdp_state, + _is_composable, + _log_post_backward_hook, + _no_dispatch_record_stream, + clean_tensor_name, + TrainingState, +) +from torch.distributed.fsdp._flat_param import ( + FlatParameter, + FlatParamHandle, + HandleShardingStrategy, + HandleTrainingState, + RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES, +) +from torch.distributed.fsdp._init_utils import HYBRID_SHARDING_STRATEGIES +from torch.distributed.fsdp.api import BackwardPrefetch +from torch.distributed.utils import ( + _apply_to_tensors, + _cast_forward_inputs, + _p_assert, + _to_kwargs, +) +from torch.utils import _pytree as pytree + + +logger = logging.getLogger(__name__) + +# Do not include "process_group" to enable hybrid shard and MoE cases +HOMOGENEOUS_ATTR_NAMES = ( + "_use_orig_params", + "limit_all_gathers", + "_use_full_prec_in_eval", +) + + +class _PrefetchMode(Enum): + BACKWARD = auto() + FORWARD = auto() + + +def _get_fsdp_root_states_with_modules( + module: nn.Module, +) -> tuple[list[_FSDPState], list[nn.Module]]: + """ + Returns a tuple containing: + 1. A list of the root ``_FSDPState`` instances in the module tree rooted at + ``module`` without any duplicates and following the ``module.modules()`` + traversal order (which is assumed to be depth-first). + 2. A corresponding list of the root modules owning the states in the first + list. + + This is similar to :func:`_get_fsdp_states_with_modules` except that we + must call :func:`_is_fsdp_root` to force a lazy initialization to determine + the FSDP root in case lazy initialization has not yet happened. + """ + fsdp_root_states: list[_FSDPState] = [] + fsdp_root_modules: list[nn.Module] = [] + visited_fsdp_states: set[_FSDPState] = set() + # NOTE: This function assumes that `module.modules()` proceeds top-down. + for submodule in module.modules(): + optional_state = _get_module_fsdp_state(submodule) + if ( + optional_state is not None + and optional_state not in visited_fsdp_states + and _is_fsdp_root(optional_state, submodule) + ): + visited_fsdp_states.add(optional_state) + fsdp_root_states.append(optional_state) + fsdp_root_modules.append(submodule) + return fsdp_root_states, fsdp_root_modules + + +def _get_fsdp_root_states(module: nn.Module) -> list[_FSDPState]: + """See :func:`_get_fsdp_root_states_with_modules`.""" + fsdp_root_states, _ = _get_fsdp_root_states_with_modules(module) + return fsdp_root_states + + +def _is_fsdp_root(state: _FSDPState, module: nn.Module) -> bool: + """ + Returns if ``state`` corresponds to that of an FSDP root. + + For the wrapper code path, ``state`` and ``module`` should be the same. For + the non-wrapper code path, ``state`` should be ``module`` 's state. + """ + # Force a lazy initialization to determine the FSDP root + _lazy_init(state, module) + if state._is_root is None: + raise AssertionError("Expected _is_root to be set after lazy init") + return state._is_root + + +@no_type_check +def _lazy_init( + state: _FSDPState, + root_module: nn.Module, +) -> _FSDPState: + """ + Performs initialization lazily, typically right before the first forward + pass. The laziness is needed to ensure that the parameter device/dtype and + the FSDP hierarchy have finalized. This method's actual logic only runs on + the root FSDP instance, which performs initialization for all non-root FSDP + instances to avoid partial initialization. + + For the non-composable code path, ``state`` and ``root_module`` should be + the same, namely the FSDP instance itself. + """ + if state._is_root is not None: + return # no-op: already lazily initialized + if not state._device_handle.is_available(): + # Allow the FSDP constructor to run even without CUDA but check this + # once we start real execution + raise RuntimeError("FSDP does not support CPU only execution") + # The following logic is only run on the root FSDP instance since it will + # set `_is_root=False` for the non-root instances + state._is_root = True + _assert_in_training_states(state, [TrainingState.IDLE]) + _check_flat_params_on_expected_device(state, root_module) + state._all_fsdp_states = traversal_utils._get_fsdp_states(root_module) + _init_streams(state) + buffers, buffer_dtypes = _get_buffers_and_dtypes_for_computation(state, root_module) + _cast_buffers_to_dtype_and_device(buffers, buffer_dtypes, state.compute_device) + state._exec_order_data.init(state, root_module, state.process_group) + _share_state_and_init_handle_attrs(state, root_module) + return state + + +def _check_flat_params_on_expected_device(state: _FSDPState, module: nn.Module): + """ + Checks that all ``FlatParameter``s in ``module`` 's tree managed by + ``state`` are on the expected device for *lazy initialization*. + """ + cpu_device = torch.device("cpu") + for handle in traversal_utils._get_fsdp_handles(module): + if ( + not handle._offload_params + and handle.flat_param.device != state.compute_device + ): + raise RuntimeError( + "An FSDP-managed module unexpectedly has parameters on " + f"{handle.flat_param.device}. Make sure to move the module to " + f"{state.compute_device} before training." + ) + elif handle._offload_params and handle.flat_param.device != cpu_device: + raise RuntimeError( + "An FSDP-managed module with parameter CPU offloading enabled " + f"has parameters on {handle.flat_param.device}. Make sure to " + f"not move the module from CPU when offloading parameters." + ) + + +@no_type_check +def _share_state_and_init_handle_attrs( + root_state: _FSDPState, + root_module: nn.Module, +) -> None: + """ + Shares data structure state from the ``root_state`` to all FSDP states in + ``root_module`` 's module tree, and initializes handle attributes. These + are done together to require a single loop over the states. + """ + handle = root_state._handle + if handle: + handle.init_flat_param_attributes() + attr_name_to_values: dict[str, set[Any]] = {} + for attr_name in HOMOGENEOUS_ATTR_NAMES: + attr_name_to_values[attr_name] = set() + root_state._all_handles = root_state._exec_order_data.all_handles # share reference + # Update _has_optim_in_backward for each handle. + for handle in root_state._all_handles: + flat_param = handle.flat_param + if hasattr(flat_param, "_in_backward_optimizers"): + raise RuntimeError( + "FSDP optimizer in backward only supported with use_orig_params=True!" + ) + handle._has_optim_in_backward = flat_param._params is not None and any( + hasattr(param, "_in_backward_optimizers") for param in flat_param._params + ) + if handle._has_optim_in_backward: + torch._C._log_api_usage_once("fsdp.optimizer_in_backward") + for fsdp_state in root_state._all_fsdp_states: + for attr_name in HOMOGENEOUS_ATTR_NAMES: + _p_assert( + hasattr(fsdp_state, attr_name), + f"FSDP state missing attribute {attr_name}", + ) + attr_name_to_values[attr_name].add(getattr(fsdp_state, attr_name)) + if fsdp_state is root_state: + continue + # Relax the assert for non-root FSDP instances in case the nested + # initialized module is wrapped again in FSDP later (e.g. after + # training to run inference) + _p_assert( + fsdp_state._is_root is None or not fsdp_state._is_root, + "Non-root FSDP instance's `_is_root` should not have been " + "set yet or should have been set to `False`", + ) + fsdp_state._is_root = False + fsdp_state._unshard_stream = root_state._unshard_stream + fsdp_state._post_backward_stream = root_state._post_backward_stream + fsdp_state._pre_unshard_stream = root_state._pre_unshard_stream + fsdp_state._all_reduce_stream = root_state._all_reduce_stream + fsdp_state._default_stream = root_state._default_stream + fsdp_state._exec_order_data = root_state._exec_order_data + fsdp_state._free_event_queue = root_state._free_event_queue + if fsdp_state._fsdp_extension is not None: + fsdp_state._fsdp_extension.compute_stream = root_state._default_stream + handle = fsdp_state._handle + if handle: + handle.init_flat_param_attributes() + for attr_name, attr_values in attr_name_to_values.items(): + if len(attr_values) != 1: + raise ValueError( + f"Expects one homogeneous value for {attr_name} but got {attr_values}" + ) + + +@no_type_check +def _init_streams( + state: _FSDPState, +) -> None: + """ + Initializes CUDA streams for overlapping communication, computation, and + data transfers. The streams should be shared across FSDP instances. + """ + if not state._is_root: + raise AssertionError("Expected state to be root") + if not state._device_handle.is_available(): + raise AssertionError("Expected device handle to be available") + uses_hybrid_sharding = any( + fsdp_state.sharding_strategy in HYBRID_SHARDING_STRATEGIES + for fsdp_state in state._all_fsdp_states + ) + # Prioritize all-gathers/reduce-scatters over async all-reduce for HSDP and + # preserve the default priority of 0 otherwise + high_priority = -1 if state.limit_all_gathers and uses_hybrid_sharding else 0 + # Default stream for computation + state._default_stream = state._device_handle.current_stream() + if state._fsdp_extension is not None: + # set the compute stream to the FSDP extension + state._fsdp_extension.compute_stream = state._default_stream + + # Stream for unshard logic, including allocating the all-gather destination + # tensors and the all-gathers themselves + state._unshard_stream = state._device_handle.Stream(priority=high_priority) + # Stream for overlapping gradient reduction with the backward pass gradient + # computation + state._post_backward_stream = state._device_handle.Stream(priority=high_priority) + # Stream for pre-unshard logic, namely allocations and writes for CPU + # offloading (H2D copy) and mixed precision (low precision cast) + state._pre_unshard_stream = state._device_handle.Stream(priority=high_priority) + # Stream to run HSDP's all-reduce as async (if using HSDP) + state._all_reduce_stream = ( + state._device_handle.Stream() if uses_hybrid_sharding else state._default_stream + ) + + +@no_type_check +def _unshard( + state: _FSDPState, + handle: FlatParamHandle, + unshard_stream: torch.Stream, + pre_unshard_stream: torch.Stream, +) -> None: + """ + Unshards the handles in ``handles``. If the handles are in + :meth:`summon_full_params` and are using mixed precision, then they are + forced to full precision. + + Postcondition: handle's ``FlatParameter`` 's data is the padded + unsharded flat parameter on the compute device. + """ + if not handle: + return + with state._device_handle.stream(pre_unshard_stream): + ran_pre_unshard = handle.pre_unshard() + if ran_pre_unshard: + unshard_stream.wait_stream(pre_unshard_stream) + if state.limit_all_gathers: + event = state._free_event_queue.dequeue_if_needed() + if event: + with torch.profiler.record_function( + "FullyShardedDataParallel.rate_limiter" + ): + event.synchronize() + with state._device_handle.stream(unshard_stream): + handle.unshard() + handle.post_unshard() + + +@no_type_check +def _reshard( + state: _FSDPState, + handle: FlatParamHandle, + free_unsharded_flat_param: bool, +): + """ + Reshards the handle. ``free_unsharded_flat_param`` indicates whether to + free the handle's padded unsharded flat parameter. + """ + handle.reshard(free_unsharded_flat_param) + if state.limit_all_gathers and free_unsharded_flat_param: + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + # We don't run an event queue for freeing under torch compile atm + # But maybe we need to? TODO(voz): Look into this + free_event = state._device_handle.Event() + free_event.record() + state._free_event_queue.enqueue(free_event) + handle.post_reshard() + # Flat parameter freed or not, we always have to "unshard" the parameter + # upon next access to get its shape correct. + handle._prefetched = False + + +def _unshard_grads( + handle: FlatParamHandle | None, +) -> None: + if handle: + handle.unshard_grad() + + +def _reshard_grads( + handle: FlatParamHandle | None, +) -> None: + if handle: + handle.reshard_grad() + + +@no_type_check +def _pre_forward( + state: _FSDPState, + handle: FlatParamHandle | None, + unshard_fn: Callable, + module: nn.Module, + args: tuple[Any, ...], + kwargs: dict[str, Any], +) -> tuple[tuple[Any, ...], dict[str, Any]]: + """ + Runs the pre-forward logic. This includes an opportunity to unshard + currently sharded parameters such as those for the current forward and + registering post-backward hooks for these current parameters. This function + also converts forward ``args`` and ``kwargs`` to the given precision. + + Args: + handles (List[FlatParamHandle]): Handles giving the parameters used in + the current forward. + unshard_fn (Optional[Callable]): A callable to unshard any currently + sharded parameters or ``None`` to not do any unsharding. + module (nn.Module): Module whose forward this method runs right before; + expected by the hook signature. + args (Tuple[Any, ...]): Module forward ``args``. + kwargs (Dict[str, Any]): Module forward ``kwargs``. + """ + with torch.profiler.record_function("FullyShardedDataParallel._pre_forward"): + # For `fully_shard` + `checkpoint`, skip pre-forward logic in the + # recomputed forward + if handle and handle._training_state == HandleTrainingState.BACKWARD_PRE: + # For both checkpoint implementations, we do not need to re-cast + # inputs here since they will be checkpointed in the low precision + # either by AC or normally by autograd as long as the AC region is + # nested within FSDP + return args, kwargs + state.training_state = TrainingState.FORWARD_BACKWARD + state._exec_order_data.record_pre_forward(handle, module.training) + if handle: + handle._training_state = HandleTrainingState.FORWARD + if unshard_fn is not None: + unshard_fn(state, handle) + # Register post-backward hooks to reshard the parameters and reduce-scatter + # their gradients. They must be re-registered every forward pass in case + # the `grad_fn` is mutated. + _register_post_backward_hook(state, handle) + # We have to reallocate the _cpu_grad if optimizer overlap + # set the grad to None in the backward pass. + if handle and handle._offload_params and handle.flat_param._cpu_grad is None: + handle.flat_param._cpu_grad = torch.zeros_like( + handle.flat_param._local_shard, device=torch.device("cpu") + ).pin_memory() + + should_cast_forward_inputs = ( + state._handle and not state._handle._force_full_precision + ) + + if should_cast_forward_inputs and state.mixed_precision.cast_forward_inputs: + # Recursively convert args and kwargs to specified precision. + input_dtype: torch.dtype | None = state.mixed_precision.param_dtype + args, kwargs = _cast_forward_inputs(input_dtype, *args, **kwargs) + _register_post_backward_reshard_only_hook(state, handle, args, kwargs) + return args, kwargs + + +@no_type_check +def _pre_forward_unshard( + state: _FSDPState, + handle: FlatParamHandle | None, +) -> None: + """Unshards parameters in the pre-forward.""" + if not handle: + return + # If the handles have been prefetched, then there is no need to call + # `_unshard()` again + if not handle._prefetched: + _unshard(state, handle, state._unshard_stream, state._pre_unshard_stream) + handle._needs_pre_forward_unshard = False + # Don't wait during trace + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + current_stream = state._device_handle.current_stream() + if state._unshard_event is not None: + current_stream.wait_event(state._unshard_event) + state._unshard_event = None + else: + current_stream.wait_stream(state._unshard_stream) + with torch.profiler.record_function( + "FullyShardedDataParallel._pre_forward_prefetch" + ): + _prefetch_handle(state, handle, _PrefetchMode.FORWARD) + + +@no_type_check +def _post_forward( + state: _FSDPState, + handle: FlatParamHandle | None, + reshard_fn: Callable, + module: nn.Module, + input: Any, + output: Any, +) -> Any: + """ + Runs the post-forward logic. This includes an opportunity to reshard + currently unsharded parameters such as those used in the current forward + and registering pre-backward hooks on the forward outputs. + + Args: + handles (List[FlatParamHandle]): Handles giving the parameters used in + the current forward. + reshard_fn (Optional[Callable]): A callable to reshard any currently + unsharded parameters (e.g. from the current forward) or ``None`` to + not do any resharding. + module (nn.Module): Module whose forward just ran, which should be a + fully sharded module (see [Note: Fully Sharded Module]); expected + by the hook signature. + input (Any): Unused; expected by the hook signature. + output (Any): Forward pass output; pre-backward hooks are registered on + the tensors that require gradients in this output. + + Postcondition: Each ``FlatParameter`` 's data points to the sharded flat + parameter. + """ + with torch.profiler.record_function("FullyShardedDataParallel._post_forward"): + # For `fully_shard` + `checkpoint`, skip post-forward logic in the + # recomputed forward + if handle and handle._training_state == HandleTrainingState.BACKWARD_PRE: + return output + + state._exec_order_data.record_post_forward(handle) + if reshard_fn is not None: + reshard_fn(state, handle) + # Register pre-backward hooks to unshard the flat parameters for the + # gradient computation (if needed) + output = _register_pre_backward_hooks(state, module, output, handle) + state.training_state = TrainingState.IDLE + if handle: + handle._training_state = HandleTrainingState.IDLE + return output + + +@no_type_check +def _post_forward_reshard( + state: _FSDPState, + handle: FlatParamHandle, +) -> None: + """Reshards parameters in the post-forward.""" + if not handle: + return + # Do not free the root's parameters in the post-forward for `FULL_SHARD` + # with the intention that they are immediately used for backward + # computation (though this may not be true) + free_unsharded_flat_param = ( + not state._is_root + and handle._sharding_strategy in RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES + ) + _reshard(state, handle, free_unsharded_flat_param) + + +@no_type_check +def _root_pre_forward( + state: _FSDPState, + module: nn.Module, + args, + kwargs, +) -> None: + """ + Runs pre-forward logic specific to the root FSDP instance, which should run + before any individual module's pre-forward. This starts with an attempt at + lazy initialization (which only runs non-vacuously once). Otherwise, if + this is called on a non-root FSDP instance, then it returns directly. + + Args: + module (nn.Module): Module for which this logic tries to run. It may or + may not be the root. If not, then this method does not do anything. + """ + with torch.profiler.record_function("FullyShardedDataParallel._root_pre_forward"): + _lazy_init(state, module) + _p_assert(state._is_root is not None, "Expects a root FSDP to have been set") + if not state._is_root: + # Always cast forward inputs in the root of this local FSDP unit for mixed + # precision, as this is where mixed precision could be configured. + # This is more useful for auto wrapping that is recommended in composable path. + # For manual wrapping, cast forward inputs on each local FSDP unit root will + # increase some overhead, so not turned on for model wrapper path right now where + # manual wrapping is more broadly used. + if _is_composable(state): + return _root_cast_forward_input(state, module, args, kwargs) + return args, kwargs + + # We cast buffers back to full precision if we're forcing full precision. Disjointly, we check if buffers + # are in full precision and if we should cast them back to lower precision, which happens when + # exiting eval() mode. + handle = state._handle + if handle: + should_cast_buffers_to_full_prec = handle._force_full_precision + else: + # If the root has no handle (no managed parameters), then we fall + # back to checking if any child wants to force full precision as a + # workaround + handles = traversal_utils._get_fsdp_handles(module) + should_cast_buffers_to_full_prec = any( + handle._force_full_precision for handle in handles + ) + + if should_cast_buffers_to_full_prec: + _cast_buffers_to_dtype_and_device( + buffers=dict(module.named_buffers()).values(), + buffer_dtypes=list(state._buffer_name_to_orig_dtype.values()), + device=state.compute_device, + ) + # This flag is only set when we cast buffers to full precision, to avoid the + # CPU overhead that can stem from retrieving all buffers and their types in the + # following else branch. + state._needs_buffer_dtype_restore_check = True + elif getattr(state, "_needs_buffer_dtype_restore_check", False): + # Check if buffers are in full precision and we need to cast them + # back down. + ( + buffers, + buffer_dtypes_for_computation, + ) = _get_buffers_and_dtypes_for_computation(state, module) + if len(buffers) > 0 and len(buffer_dtypes_for_computation) > 0: + if any( + buffer.dtype != buffer_dtype_for_computation + for buffer, buffer_dtype_for_computation in zip( + buffers, buffer_dtypes_for_computation + ) + ): + # Assume we have to cast everything if there is one mismatch + _cast_buffers_to_dtype_and_device( + buffers, buffer_dtypes_for_computation, state.compute_device + ) + # We don't have to check this again until we cast buffers to full precision again. + state._needs_buffer_dtype_restore_check = False + + if state.forward_prefetch: + handles = [ + fsdp_state._handle + for fsdp_state in state._all_fsdp_states + if fsdp_state._handle + ] + for handle in handles: + handle._needs_pre_forward_unshard = True + handle._prefetched = False + _wait_for_computation_stream( + state._device_handle.current_stream(), + state._unshard_stream, + state._pre_unshard_stream, + ) + _reset_flat_param_grad_info_if_needed(state._all_handles) + + # Prepares the forward inputs by moving them to ``compute_device`` + # TODO: Do not use the side stream for tensor copies for now; investigate + # the perf with/without it. + with torch.profiler.record_function("FullyShardedDataParallel._to_kwargs"): + args_tuple, kwargs_tuple = _to_kwargs( + args, kwargs, state.compute_device, False + ) + args = args_tuple[0] if args_tuple else () + kwargs = kwargs_tuple[0] if kwargs_tuple else {} + + return _root_cast_forward_input(state, module, args, kwargs) + + +@no_type_check +def _root_cast_forward_input( + state: _FSDPState, module: torch.nn.Module, args, kwargs +) -> tuple[Any, Any]: + if state._handle: + force_full_precision = not state._handle._force_full_precision + else: + force_full_precision = True + + should_cast_forward_inputs = ( + (module.training or not state._use_full_prec_in_eval) and force_full_precision + ) and state.mixed_precision.cast_root_forward_inputs + + if should_cast_forward_inputs: + input_dtype: torch.dtype | None = state.mixed_precision.param_dtype + args, kwargs = _cast_forward_inputs(input_dtype, *args, **kwargs) + + return args, kwargs + + +@no_type_check +def _pre_backward_hook( + state: _FSDPState, + module: nn.Module, + handle: FlatParamHandle, + grad, + *unused: Any, +) -> Any: + """ + Prepares ``_handle`` 's ``FlatParameter`` s for gradient computation. + + Args: + module (nn.Module): Fully sharded module (see [Note: Fully Sharded + Module]). + """ + # Only run the pre-backward hook once per group of handles involved in the + # same module forward computation + if ( + handle + and hasattr(handle, "_ran_pre_backward_hook") + and handle._ran_pre_backward_hook + ): + return grad + + with torch.profiler.record_function("FullyShardedDataParallel._pre_backward_hook"): + # Queue the post-backward callback once for the root FSDP instance to + # attach it to the outermost backward graph task so that it is called + # after all backward calls complete + if state._is_root and not state._post_backward_callback_queued: + _register_post_backward_final_callback(state, module) + _reset_flat_param_grad_info_if_needed(state._all_handles) + elif handle: + allowed_states = [TrainingState.IDLE] + if _is_composable(state): + allowed_states.append(TrainingState.FORWARD_BACKWARD) + _assert_in_training_states(state, allowed_states) + state.training_state = TrainingState.FORWARD_BACKWARD + # Queueing the post-backward callback is the only logic that is not + # per-handle in the pre-backward hook, so we can return early here if + # there are no handles. + if not handle: + return grad + handle._training_state = HandleTrainingState.BACKWARD_PRE + + if handle._needs_pre_backward_unshard: + # If the handles have been prefetched, then there is no need to + # call `_unshard()` again + if not handle._prefetched: + _unshard( + state, + handle, + state._unshard_stream, + state._pre_unshard_stream, + ) + # Don't wait during trace + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + state._device_handle.current_stream().wait_stream(state._unshard_stream) + + # Set this to `False` to ensure that a mistargeted prefetch does not + # actually unshard these handles + handle._needs_pre_backward_unshard = False + with torch.profiler.record_function( + "FullyShardedDataParallel._pre_backward_prefetch" + ): + _prefetch_handle(state, handle, _PrefetchMode.BACKWARD) + handle.prepare_gradient_for_backward() + handle._ran_pre_backward_hook = True + return grad + + +@no_type_check +@torch.no_grad() +def _post_backward_hook( + state: _FSDPState, + handle: FlatParamHandle, + flat_param, + *unused: Any, +): + """ + Reduce-scatters the gradient of ``handle`` 's ``FlatParameter``. + + Precondition: The ``FlatParameter`` 's ``.grad`` attribute contains the + unsharded gradient for the local batch. + + Postcondition: + - If using ``NO_SHARD``, then the ``.grad`` attribute is the reduced + unsharded gradient. + - Otherwise, the ``_saved_grad_shard`` attribute is the reduced sharded + gradient (accumulating with any existing gradient). + """ + _log_post_backward_hook(state, handle, logger) + flat_param = handle.flat_param + flat_param._post_backward_called = True + with torch.autograd.profiler.record_function( + "FullyShardedDataParallel._post_backward_hook" + ): + _assert_in_training_states(state, [TrainingState.FORWARD_BACKWARD]) + # For multiple applications of reentrant AC across submodules sharing + # the same `FlatParameter`, the post-backward hook may run multiple + # times in one backward, in which case we permit the state to already + # be in `BACKWARD_POST`. + _p_assert( + handle._training_state + in (HandleTrainingState.BACKWARD_PRE, HandleTrainingState.BACKWARD_POST), + f"Expects `BACKWARD_PRE` or `BACKWARD_POST` state but got {handle._training_state}", + ) + handle._training_state = HandleTrainingState.BACKWARD_POST + + if flat_param.grad is None: + return + if flat_param.grad.requires_grad: + raise RuntimeError("FSDP does not support gradients of gradients") + + _post_backward_reshard(state, handle) + if not state._sync_gradients: + if handle._use_orig_params: + handle._use_unsharded_grad_views() + return + + # Wait for all ops in the current stream (e.g. gradient computation) to + # finish before reduce-scattering the gradient + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + state._post_backward_stream.wait_stream( + state._device_handle.current_stream() + ) + + with state._device_handle.stream(state._post_backward_stream): + autograd_computed_grad = flat_param.grad.data + if ( + not _low_precision_hook_enabled(state) + and flat_param.grad.dtype != handle._reduce_dtype + # If we are forcing full precision but communicating grads + # (i.e. model.eval() + full precision in eval was configured), don't downcast gradient. + and not handle._force_full_precision + ): + flat_param.grad.data = flat_param.grad.to(handle._reduce_dtype) + if handle.uses_sharded_strategy: + _reduce_grad(state, handle) + else: + _reduce_grad_no_shard(state, handle) + # Since the unsharded gradient is produced in the computation + # stream and consumed in the post-backward stream, inform the + # caching allocator (before it goes out of scope) + _no_dispatch_record_stream( + autograd_computed_grad, state._post_backward_stream + ) + + +def _post_backward_reshard_only_hook( + state: _FSDPState, + handle: FlatParamHandle, + *unused: Any, +) -> None: + with torch.profiler.record_function( + "FullyShardedDataParallel._post_backward_hook_reshard_only" + ): + # `_pre_backward_hook` may not get executed + # if forward output does not require grad + # overwrite IDLE state for post-backward prefetching + state.training_state = TrainingState.FORWARD_BACKWARD + handle._training_state = HandleTrainingState.BACKWARD_POST + _post_backward_reshard(state, handle) + + +def _post_backward_reshard( + state: _FSDPState, + handle: FlatParamHandle, + *unused: Any, +) -> None: + free_unsharded_flat_param = _should_free_in_backward(state, handle) + _reshard(state, handle, free_unsharded_flat_param) + + # TODO: Post-backward prefetching does not support the multiple handles + # per module case since the post-backward hook runs per handle, not per + # group of handles. + with torch.profiler.record_function( + "FullyShardedDataParallel._post_backward_prefetch" + ): + _prefetch_handle(state, handle, _PrefetchMode.BACKWARD) + + +@no_type_check +def _should_free_in_backward( + state: _FSDPState, + handle: FlatParamHandle, +) -> bool: + """ + Returns whether FSDP should free the unsharded flat parameter in the + post-backward or not. + """ + if not handle.uses_sharded_strategy: + return False + # If not syncing gradients, then we do not free for strategies that do not + # reshard after forward as a *heuristic* to tradeoff higher memory for + # higher throughput. + return ( + state._sync_gradients + or handle._sharding_strategy in RESHARD_AFTER_FORWARD_HANDLE_STRATEGIES + ) + + +@no_type_check +def _reduce_grad(state: _FSDPState, handle: FlatParamHandle) -> None: + """ + For sharded strategies, this runs gradient reduction, sharded gradient + accumulation if needed, and the post-reduction callback. + """ + flat_param = handle.flat_param + uses_hybrid_sharded_strategy = handle._sharding_strategy in ( + HandleShardingStrategy.HYBRID_SHARD, + HandleShardingStrategy._HYBRID_SHARD_ZERO2, + ) + # We clear `.grad` to permit multiple backwards. This avoids a race where + # the second backward pass computation precedes ahead of the first backward + # pass reduction, which is possible since the reduction is issued in a + # separate stream and is async and would result in reducing the wrong + # gradient. + unsharded_grad = flat_param.grad.data + flat_param.grad = None + padded_unsharded_grad, new_sharded_grad = _get_reduce_scatter_tensors( + state, unsharded_grad + ) + if state._comm_hook is None: # default path + _div_if_needed(padded_unsharded_grad, state._gradient_predivide_factor) + pg = ( + handle._fake_process_group + if handle._use_fake_reduce + else state.process_group + ) + dist.reduce_scatter_single( + new_sharded_grad, + padded_unsharded_grad, + group=pg, + ) + if uses_hybrid_sharded_strategy: + # Don't wait during trace + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + state._all_reduce_stream.wait_stream(state._post_backward_stream) + with state._device_handle.stream(state._all_reduce_stream): + # Since the new sharded gradient is produced in the post- + # backward stream and consumed in the all-reduce stream, + # inform the caching allocator + _no_dispatch_record_stream(new_sharded_grad, state._all_reduce_stream) + dist.all_reduce(new_sharded_grad, group=state._inter_node_pg) + _div_if_needed(new_sharded_grad, state._gradient_postdivide_factor) + grad_to_offload = _accumulate_sharded_grad( + state, handle, new_sharded_grad + ) + _post_reduce_grad_callback(state, handle, grad_to_offload) + return + _div_if_needed(new_sharded_grad, state._gradient_postdivide_factor) + else: + state._comm_hook( + state._comm_hook_state, padded_unsharded_grad, new_sharded_grad + ) + # NOTE: HSDP variants do not support communication hook. + grad_to_offload = _accumulate_sharded_grad(state, handle, new_sharded_grad) + _post_reduce_grad_callback(state, handle, grad_to_offload) + + +@no_type_check +def _get_reduce_scatter_tensors( + state: _FSDPState, unsharded_grad: torch.Tensor +) -> tuple[torch.Tensor, torch.Tensor]: + """ + Returns the input and output tensors to reduce-scatter, respectively. + """ + chunks = list(unsharded_grad.chunk(state.world_size)) + numel_to_pad = state.world_size * chunks[0].numel() - unsharded_grad.numel() + padded_unsharded_grad = ( + F.pad(unsharded_grad, [0, numel_to_pad]) if numel_to_pad > 0 else unsharded_grad + ) + new_sharded_grad = torch.empty_like(chunks[0]) # padded + return padded_unsharded_grad, new_sharded_grad + + +@no_type_check +def _accumulate_sharded_grad( + state: _FSDPState, + handle: FlatParamHandle, + sharded_grad: torch.Tensor, +) -> torch.Tensor: + """ + Accumulates the reduce-scattered sharded gradient with any existing sharded + gradient if needed, returning the gradient to offload (if CPU offloading is + enabled). + """ + flat_param = handle.flat_param + _cast_grad_to_param_dtype(state, sharded_grad, flat_param) + # Save the sharded gradient in `_saved_grad_shard` to support gradient + # accumulation -- for multiple backwards, the gradient reductions may + # happen in arbitrary order + accumulate_grad = hasattr(flat_param, "_saved_grad_shard") + if accumulate_grad: + _check_grad_to_accumulate(sharded_grad, flat_param._saved_grad_shard) + flat_param._saved_grad_shard += sharded_grad + else: + flat_param._saved_grad_shard = sharded_grad + grad_to_offload = flat_param._saved_grad_shard + return grad_to_offload + + +@no_type_check +def _reduce_grad_no_shard(state: _FSDPState, handle: FlatParamHandle) -> None: + """ + For no-shard, this runs gradient reduction (which directly covers any + gradient accumulation implicitly) and the post-reduction callback. + """ + flat_param = handle.flat_param + if state._comm_hook is None: # default path + _div_if_needed(flat_param.grad, state._gradient_predivide_factor) + dist.all_reduce(flat_param.grad, group=state.process_group) + _div_if_needed(flat_param.grad, state._gradient_postdivide_factor) + else: + state._comm_hook(state._comm_hook_state, flat_param.grad) + # For `NO_SHARD`, we can keep the low precision gradients by simply + # omitting the cast altogether + if not handle._keep_low_precision_grads: + _cast_grad_to_param_dtype(state, flat_param.grad, flat_param) + grad_to_offload = flat_param.grad.data + _post_reduce_grad_callback(state, handle, grad_to_offload) + + +@no_type_check +def _post_reduce_grad_callback( + state: _FSDPState, + handle: FlatParamHandle, + # Additional arguments needed for the callback logic + grad_to_offload: torch.Tensor, +): + """ + This callback captures any logic to run after the gradient reduction + finishes. Currently, this offloads the gradient to CPU if CPU offloading is + enabled and uses sharded gradient views if ``use_orig_params=True``. + """ + _offload_grad(state, handle, grad_to_offload) + _post_backward_use_sharded_grad_views(handle) + + +@no_type_check +def _offload_grad( + state: _FSDPState, + handle: FlatParamHandle, + grad_to_offload: torch.Tensor, +): + if not handle._offload_params: + return + # Offload the gradient to CPU to ensure parameters and gradients are on the + # same device as required by the optimizer + # TODO: Investigate why `NO_SHARD` breaks correctness when using + # `non_blocking=True` here. + # TODO (rohan-varma): When CPU offload and optimizer overlap, + # non_blocking=True won't work since the copy may have not finished before + # the optimizer step executes on CPU. If we want to use non-blocking=True + # here, we'll have to synchronize before using result on CPU. + non_blocking = handle.uses_sharded_strategy and not handle._has_optim_in_backward + handle.flat_param._cpu_grad.copy_( + grad_to_offload.detach(), non_blocking=non_blocking + ) # synchronized in the post-backward callback + # Since the gradient being offloaded may have been produced in the + # computation stream and is being consumed here in the post-backward + # stream, inform the caching allocator + _no_dispatch_record_stream(grad_to_offload.data, state._post_backward_stream) + + +@no_type_check +def _post_backward_use_sharded_grad_views(handle: FlatParamHandle): + if not handle._use_orig_params: + return + # Since the handle's `FlatParameter` completed its gradient computation, we + # should reset the gradient noneness mask + handle._reset_is_grad_none() + # Delay using sharded gradient views until after the reduce-scatter instead + # of immediately after resharding + handle._use_sharded_grad_views() + if handle._has_optim_in_backward: + handle.prepare_gradient_for_optim() + for orig_param in handle.flat_param._params: + # Check for `None` gradient to filter parameters not in the rank + if orig_param.grad is not None and hasattr( + orig_param, "_in_backward_optimizers" + ): + # TODO (rohan-varma): For CPU offload, this unfortunately + # operates on CPU because the parameters and gradients have + # already been offloaded. We should run this on GPU after + # refactoring. + for optim in orig_param._in_backward_optimizers: + optim.step() + + optim.zero_grad(set_to_none=True) + handle._reset_flat_param_grad_info_if_needed() + if handle._offload_params: + handle.flat_param._cpu_grad = None + + +def _div_if_needed(tensor: torch.Tensor, div_factor: float) -> None: + if div_factor > 1: + tensor.div_(div_factor) + + +@no_type_check +def _cast_grad_to_param_dtype( + state: _FSDPState, + sharded_grad: torch.Tensor, + param: FlatParameter, +): + """ + Casts ``sharded_grad`` back to the full parameter dtype so that the + optimizer step runs with that dtype. This performs an actual cast if + 1. parameters were in reduced precision during the forward since then + gradients would be in that reduced precision, or + 2. parameters were not in reduced precision but gradients were in + reduced precision for communication. + However, if a low precision communication hook is registered, then this + dtype cast happens in the hook instead. + """ + _assert_in_training_states(state, [TrainingState.FORWARD_BACKWARD]) + if not _low_precision_hook_enabled(state) and sharded_grad.dtype != param.dtype: + low_prec_grad_data = sharded_grad.data + sharded_grad.data = sharded_grad.data.to(dtype=param.dtype) + # Since for `NO_SHARD`, the gradient is produced in the computation + # stream and consumed here in the post-backward stream, inform the + # caching allocator; for the sharded strategies, the gradient is + # produced in the post-backward stream, so this `record_stream()` + # should be a no-op + _no_dispatch_record_stream( + low_prec_grad_data, state._device_handle.current_stream() + ) + + +def _check_grad_to_accumulate( + new_sharded_grad: torch.Tensor, + accumulated_grad: torch.Tensor, +) -> None: + _p_assert( + accumulated_grad.shape == new_sharded_grad.shape, + "Shape mismatch when accumulating gradients: " + f"existing gradient shape={accumulated_grad.shape} " + f"new gradient shape={new_sharded_grad.shape}", + ) + _p_assert( + accumulated_grad.device == new_sharded_grad.device, + "Device mismatch when accumulating gradients: " + f"existing gradient device={accumulated_grad.device} " + f"new gradient device={new_sharded_grad.device}", + ) + + +@no_type_check +def _low_precision_hook_enabled(state: _FSDPState) -> bool: + return state._comm_hook in LOW_PRECISION_HOOKS + + +@no_type_check +@torch.no_grad() +def _post_backward_final_callback( + state: _FSDPState, + module: nn.Module, +): + """ + This waits for the post-backward to finish and performs some final cleanup. + This runs at the end of the entire backward pass and should only be called + on the root FSDP instance. + """ + _p_assert( + state._is_root, + "The post-backward callback should only be called on the root FSDP instance", + ) + root_state = state + + if root_state._sync_gradients: + current_stream = state._device_handle.current_stream() + # TODO (rohan-varma): this also waits for the overlapped optimizer step to finish + # since it currently runs in the post-backward stream. That can be + # pushed to the next forward if run in a different stream + current_stream.wait_stream(root_state._post_backward_stream) + if root_state._all_reduce_stream is not current_stream: # uses HSDP + current_stream.wait_stream(root_state._all_reduce_stream) + if root_state.cpu_offload.offload_params: + # Wait for non-blocking GPU -> CPU sharded gradient copies from the + # post-backward hooks to finish explicitly since CPU gradients do + # not automatically synchronize with the GPU + state._device_handle.current_stream().synchronize() + root_state._exec_order_data.next_iter() + + for fsdp_state in state._all_fsdp_states: + _catch_all_reshard(fsdp_state) + _finalize_params(fsdp_state) + fsdp_state.training_state = TrainingState.IDLE + handle = fsdp_state._handle + if handle: + handle._ran_pre_backward_hook = False + handle._needs_pre_backward_unshard = False + handle._post_forward_index = None + handle._training_state = HandleTrainingState.IDLE + handle._prefetched = False + # Reset for cases like one forward and multiple backwards + root_state._post_backward_callback_queued = False + + +@no_type_check +def _catch_all_reshard( + state: _FSDPState, +) -> None: + """ + Reshards the parameters that may not have been resharded in the + post-backward hook. This can happen when a module's output is used in the + forward pass, meaning that its pre-backward hook runs (unsharding the + parameter), but the post-backward hook does not run because the output was + not used in the loss computation corresponding to this backward pass. + """ + # Wrap with a try-except to provide a more informative traceback if an + # error is raised + try: + if state._handle: + # TODO: This already-resharded check is brittle: + # https://github.com/pytorch/pytorch/issues/83956 + already_resharded = ( + state._handle.flat_param.data_ptr() + == state._handle.flat_param._local_shard.data_ptr() + # If FSDP skipped using sharded views, then the flat parameter + # still points to the sharded data, so we need to reshard to + # use sharded views + and not state._handle._skipped_use_sharded_views + ) + if already_resharded: + return + free_unsharded_flat_param = _should_free_in_backward(state, state._handle) + _reshard(state, state._handle, free_unsharded_flat_param) + except Exception as e: + _p_assert( + False, + f"Got exception in the catch-all reshard for {state}: {str(e)}", + raise_assertion_error=False, + ) + raise e + + +@no_type_check +def _finalize_params( + state: _FSDPState, +) -> None: + """Finalizes the parameters before the next iteration.""" + handle = state._handle + if not handle: + return + flat_param = handle.flat_param + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + if hasattr(flat_param, "_post_backward_hook_handle"): + pbhs_handle = flat_param._post_backward_hook_handle + pbhs_handle.remove() + del flat_param._post_backward_hook_handle + else: + if hasattr(flat_param, "_post_backward_hook_state"): + post_backward_hook_state_len = len(flat_param._post_backward_hook_state) + expected_post_backward_hook_state_len = int(flat_param.requires_grad) + 1 + _p_assert( + post_backward_hook_state_len == expected_post_backward_hook_state_len, + f"Invalid: ``_post_backward_hook_state``: {flat_param._post_backward_hook_state}", + ) + flat_param._post_backward_hook_state[-1].remove() + delattr(flat_param, "_post_backward_hook_state") + if flat_param.requires_grad: + if not state._sync_gradients: + # Preserve the gradient accumulation state if not synchronizing + # gradients: `.grad` remains the unsharded gradient from prior + # `no_sync()` iterations, and `_saved_grad_shard` remains the + # sharded gradient from the last synchronized iteration + return + if not handle._has_optim_in_backward: + handle.prepare_gradient_for_optim() + _p_assert( + hasattr(flat_param, "_post_backward_called"), + "Expects `_post_backward_called` to be set on the `FlatParameter`", + ) + flat_param._post_backward_called = False + + +@no_type_check +def _prefetch_handle( + state: _FSDPState, + current_handle: FlatParamHandle | None, + prefetch_mode: _PrefetchMode, +) -> None: + """ + Prefetches the next handles if needed (without synchronization). An empty + handles key cannot prefetch. + """ + if not current_handle: + return + handle = _get_handle_to_prefetch(state, current_handle) + if not handle: + return + # Temporarily emulate the training state while calling `_unshard` to + # ensure the correct `as_params` for `_use_unsharded_views()` + prev_training_state = handle._training_state + if prefetch_mode == _PrefetchMode.BACKWARD: + handle._training_state = HandleTrainingState.BACKWARD_PRE + elif prefetch_mode == _PrefetchMode.FORWARD: + handle._training_state = HandleTrainingState.FORWARD + else: + raise ValueError(f"Invalid prefetch mode on rank {state.rank}: {prefetch_mode}") + # Prefetch the next set of handles without synchronizing to allow + # the sync to happen as late as possible to maximize overlap + _unshard(state, handle, state._unshard_stream, state._pre_unshard_stream) + handle._training_state = prev_training_state + handle._prefetched = True + + +@no_type_check +def _get_handle_to_prefetch( + state: _FSDPState, + current_handle: FlatParamHandle, +) -> FlatParamHandle: + """ + Returns a :class:`list` of the handles keys to prefetch for the next + module(s), where ``current_handle`` represents the current module. + + "Prefetching" refers to running the unshard logic early (without + synchronization), and the "next" modules depend on the recorded execution + order and the current training state. + """ + training_state = _get_training_state(current_handle) + valid_training_states = ( + HandleTrainingState.BACKWARD_PRE, + HandleTrainingState.BACKWARD_POST, + HandleTrainingState.FORWARD, + ) + _p_assert( + training_state in valid_training_states, + f"Prefetching is only supported in {valid_training_states} but " + f"currently in {training_state}", + ) + eod = state._exec_order_data + target_handle: FlatParamHandle | None = None + if ( + training_state == HandleTrainingState.BACKWARD_PRE + and state.backward_prefetch == BackwardPrefetch.BACKWARD_PRE + ) or ( + training_state == HandleTrainingState.BACKWARD_POST + and state.backward_prefetch == BackwardPrefetch.BACKWARD_POST + ): + target_handle_candidate = eod.get_handle_to_backward_prefetch(current_handle) + if ( + target_handle_candidate + and target_handle_candidate._needs_pre_backward_unshard + and not target_handle_candidate._prefetched + ): + target_handle = target_handle_candidate + else: + target_handle = None + elif training_state == HandleTrainingState.FORWARD and state.forward_prefetch: + target_handle_candidate = eod.get_handle_to_forward_prefetch(current_handle) + if ( + target_handle_candidate + and target_handle_candidate._needs_pre_forward_unshard + and not target_handle_candidate._prefetched + ): + target_handle = target_handle_candidate + else: + target_handle = None + + return target_handle + + +def _get_training_state( + handle: FlatParamHandle, +) -> HandleTrainingState: + """Returns the training state of the handles in ``handle``.""" + _p_assert(handle, "Expects a non-empty handle") + return handle._training_state + + +@no_type_check +def _register_pre_forward_hook( + state: _FSDPState, + module: nn.Module, +) -> None: + """ + Registers a pre-forward hook on ``module``. + """ + for forward_handle in state._pre_forward_handles: + forward_handle.remove() + state._pre_forward_handles.clear() + module_param_handle = state._fully_sharded_module_to_handle.get(module, None) + hook = functools.partial( + _pre_forward, state, module_param_handle, _pre_forward_unshard + ) + state._pre_forward_handles.append( + module.register_forward_pre_hook(hook, prepend=True, with_kwargs=True) + ) + + +@no_type_check +def _register_post_forward_hook( + state: _FSDPState, + module: nn.Module, +) -> None: + """ + Registers a post-forward hook on ``module``. Even if the module has no + handles, we should register the hook since it will register the module's + pre-backward hook. + """ + for forward_handle in state._post_forward_handles: + forward_handle.remove() + state._post_forward_handles.clear() + module_param_handle = state._fully_sharded_module_to_handle.get(module, None) + hook = functools.partial( + _post_forward, + state, + module_param_handle, + _post_forward_reshard, + ) + state._post_forward_handles.append(module.register_forward_hook(hook)) + + +@no_type_check +def _register_root_pre_forward_hook( + state: _FSDPState, + module: nn.Module, +): + """ + Registers root pre-forward hook on ``module``, which should be the local + FSDP root. + + NOTE: For the current composable FSDP design, we have each application of + ``fully_shard()`` to a module to indicate that that module is the local + FSDP root. We may remove this assumption in the future, in which case we + will need to register this root pre-forward hook on any candidate module + that may be the local FSDP root. + """ + for forward_handle in state._root_pre_forward_handles: + forward_handle.remove() + state._root_pre_forward_handles.clear() + hook = functools.partial(_root_pre_forward, state) + state._root_pre_forward_handles.append( + module.register_forward_pre_hook(hook, prepend=True, with_kwargs=True) + ) + + +@no_type_check +def _register_pre_backward_hooks( + state: _FSDPState, + module: nn.Module, + outputs: Any, + handle: FlatParamHandle, +) -> None: + """ + Registers pre-backward hooks on the tensors that require gradients in the + forward pass outputs ``outputs``, which were computed using the + ``FlatParameter`` s of ``handles``. + + Args: + module (nn.Module): Fully sharded module (see [Note: Fully Sharded + Module]). + + Returns: + Forward pass outputs with pre-backward hooks registered to tensors that + require gradients. + """ + # If there is no gradient computation, then there is no need for + # pre-backward logic + if not torch.is_grad_enabled(): + return outputs + if state._is_root: + state._post_backward_callback_queued = False # only defined on the root + + if handle: + handle._needs_pre_backward_unshard = False + # Since these handles' `FlatParameter`s participated in a forward, we + # conservatively assume that they will be used in the backward + handle._ran_pre_backward_hook = False + + def _register_hook(t: torch.Tensor) -> torch.Tensor: + if t.requires_grad: + t.register_hook( + torch.utils.hooks.unserializable_hook( + functools.partial(_pre_backward_hook, state, module, handle) + ) + ) + if handle: + handle._needs_pre_backward_unshard = True + return t + + return _apply_to_tensors(_register_hook, outputs) + + +def _register_post_backward_hook( + state: _FSDPState, + handle: FlatParamHandle | None, +) -> None: + """ + Registers post-backward hooks on the ``FlatParameter`` s' + ``AccumulateGrad`` objects to reshard and to reduce-scatter gradients. + + The ``AccumulateGrad`` object represents the last function that finalizes + the ``FlatParameter`` 's gradient, so it only runs after its entire + gradient computation has finished. + + We register the post-backward hook only once in the *first* forward that a + ``FlatParameter`` participates in. This relies on the ``AccumulateGrad`` + object being preserved through multiple forwards. + + NOTE: We follow this heuristic to prefer the *first* forward to target the + parameter mixed precision case, where there are *separate* + ``AccumulateGrad`` objects across the different forwards. (Without + parameter mixed precision, the ``AccumulateGrad`` objects are the same.) If + we instead prefer the *last* forward, then the hook runs early. + """ + # If there is no gradient computation, then there is no need for + # post-backward logic + if not torch.is_grad_enabled(): + return + if not handle: + return + flat_param = handle.flat_param + + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + already_registered = hasattr(flat_param, "_post_backward_hook_handle") + if already_registered or not flat_param.requires_grad: + return + hook = functools.partial(_post_backward_hook, state, handle) + hook_handle = flat_param.register_post_accumulate_grad_hook(hook) + flat_param._post_backward_hook_handle = hook_handle # type: ignore[attr-defined] + else: + already_registered = hasattr(flat_param, "_post_backward_hook_state") + if already_registered or not flat_param.requires_grad: + return + # Get the `AccumulateGrad` object + temp_flat_param = flat_param.expand_as(flat_param) + _p_assert( + temp_flat_param.grad_fn is not None, + "The `grad_fn` is needed to access the `AccumulateGrad` and " + "register the post-backward hook", + ) + acc_grad = temp_flat_param.grad_fn.next_functions[0][0] # type: ignore[union-attr] + if acc_grad is None: + raise AssertionError("Expected acc_grad to be set") + hook_handle = acc_grad.register_hook( + functools.partial(_post_backward_hook, state, handle) + ) + flat_param._post_backward_hook_state = (acc_grad, hook_handle) # type: ignore[attr-defined] + + +def _register_post_backward_reshard_only_hook( + state: _FSDPState, + handle: FlatParamHandle | None, + args: tuple[Any, ...], + kwargs: dict[str, Any], +) -> None: + """ + Registers post-backward hooks to reshard flat parameters that do not + require gradient. We register these using multi-post-grad hooks on the + input activations to ensure that all gradients that may depend on the + parameters have been computed before resharding. + """ + # If there is no gradient computation, then there is no need for + # post-backward logic + if not torch.is_grad_enabled(): + return + # Construct `inp_tensors` lazily to avoid CPU overhead in typical case + # where each flat parameter requires gradient + inp_tensors: list[torch.Tensor] | None = None + if not handle: + return + flat_param = handle.flat_param + + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + already_registered = hasattr(flat_param, "_post_backward_hook_handle") + else: + already_registered = hasattr(flat_param, "_post_backward_hook_state") + + if already_registered or flat_param.requires_grad: + return + if inp_tensors is None: + args_flat = pytree.arg_tree_leaves(*args, **kwargs) + inp_tensors = [ + obj for obj in args_flat if torch.is_tensor(obj) and obj.requires_grad + ] + if inp_tensors is None: + raise AssertionError("Expected inp_tensors to be set") + hook_handle = register_multi_grad_hook( + inp_tensors, functools.partial(_post_backward_reshard_only_hook, state, handle) + ) + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + flat_param._post_backward_hook_handle = hook_handle # type: ignore[attr-defined, assignment] + else: + flat_param._post_backward_hook_state = (hook_handle,) # type: ignore[attr-defined, assignment] + + +@no_type_check +def _register_post_backward_final_callback( + state: _FSDPState, module: nn.Module +) -> None: + """ + Registers the post-backward final callback that runs at the end of the + backward pass. This should be called from the root FSDP instance at the + beginning of the pre-backward. + """ + _p_assert( + state._is_root, + "Only the root FSDP instance should register the post-backward callback", + ) + if state._post_backward_callback_queued: + return + _assert_in_training_states(state, [TrainingState.IDLE]) + # Trace does not need this callback + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + state._post_backward_callback_queued = True + Variable._execution_engine.queue_callback( + functools.partial(_post_backward_final_callback, state, module) + ) + + +def _wait_for_computation_stream( + computation_stream: torch.Stream, + unshard_stream: torch.Stream, + pre_unshard_stream: torch.Stream, +): + """ + Has the unshard and pre-unshard streams wait for the computation stream. + For example, this should be called in the FSDP root's pre-forward to + respect optimizer step computation. + """ + # Tracing does not need to wait + if torch.distributed._functional_collectives.is_torchdynamo_compiling(): + return + unshard_stream.wait_stream(computation_stream) # type: ignore[attr-defined] + # Having the pre-all-gather stream wait for the current stream even if we + # do not leverage the pre-all-gather stream is tolerable since this only + # runs once per iteration + pre_unshard_stream.wait_stream(computation_stream) # type: ignore[attr-defined] + + +def _reset_flat_param_grad_info_if_needed( + handles: list[FlatParamHandle], +): + """ + Clears the original parameters' gradients if needed. This method's CPU + overhead is minimal, so we may call it throughout FSDP methods, which serve + as callsites to free the gradient memory earlier. + """ + if not isinstance(handles, list): + handles = [handles] + for handle in handles: + if handle._use_orig_params: + handle._reset_flat_param_grad_info_if_needed() + + +@no_type_check +def _get_buffers_and_dtypes_for_computation( + state: _FSDPState, + root_module: nn.Module, +) -> tuple[list[torch.Tensor], list[torch.dtype | None]]: + """ + Returns all buffers in the module tree rooted at ``root_module`` and a + corresponding list of the buffer dtypes for computation. Each buffer dtype + is either ``None`` if buffer mixed precision is not enabled or the buffer + low precision dtype otherwise. + """ + _p_assert(state._is_root, "Expects the root to cast buffers") + buffers: list[torch.Tensor] = [] + buffer_dtypes: list[torch.dtype | None] = [] + visited_buffers: set[torch.Tensor] = set() + # Traverse the FSDP states bottom-up so that we prefer the owning FSDP + # instance's mixed precision setting for each buffer + fsdp_states, fsdp_modules = traversal_utils._get_fsdp_states_with_modules( + root_module + ) + for fsdp_state, fsdp_module in zip(reversed(fsdp_states), reversed(fsdp_modules)): + for buffer_name, buffer in fsdp_module.named_buffers(): + if buffer in visited_buffers: + continue + visited_buffers.add(buffer) + if clean_tensor_name(buffer_name) in fsdp_state._ignored_buffer_names: + continue + buffers.append(buffer) + buffer_dtypes.append(fsdp_state.mixed_precision.buffer_dtype) + if len(buffers) != len(buffer_dtypes): + raise AssertionError( + f"Expected buffers and buffer_dtypes to have the same length, got {len(buffers)} and {len(buffer_dtypes)}" + ) + return buffers, buffer_dtypes + + +@no_type_check +def _get_orig_buffer_dtypes( + state: _FSDPState, + buffer_names: list[str], +) -> list[torch.dtype]: + """ + Returns the original buffer types of the given buffer names. + """ + buffer_dtypes: list[torch.dtype] = [] + for buffer_name in buffer_names: + _p_assert( + buffer_name in state._buffer_name_to_orig_dtype, + f"{buffer_name} is missing from pre-computed dict on rank " + f"{state.rank}, which only has keys " + f"{state._buffer_name_to_orig_dtype.keys()}", + ) + buffer_dtypes.append(state._buffer_name_to_orig_dtype[buffer_name]) + return buffer_dtypes + + +def _cast_buffers_to_dtype_and_device( + buffers: list[torch.Tensor], + buffer_dtypes: list[torch.dtype | None], + device: torch.device, +) -> None: + """ + Casts ``buffers`` to the dtypes given by ``buffer_dtypes`` and moves them + to ``device``. If an element in ``buffer_dtypes`` is ``None``, then the + corresponding buffer is only moved to ``device``. + """ + _p_assert( + buffer_dtypes is None or len(buffers) == len(buffer_dtypes), + f"Expects `buffers` and `buffer_dtypes` to have the same length if " + f"`buffer_dtypes` is specified but got {len(buffers)} and " + f"{len(buffer_dtypes)}", + ) + for buffer, buffer_dtype in zip(buffers, buffer_dtypes): + if not torch.is_floating_point(buffer) or buffer_dtype is None: + buffer.data = buffer.to(device=device) + else: + buffer.data = buffer.to(device=device, dtype=buffer_dtype) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_shard_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_shard_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..5168aa54c134c3fb5eea2fd04d6ce932a8b30c4b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_shard_utils.py @@ -0,0 +1,139 @@ +# mypy: allow-untyped-defs +import copy +import itertools +import math + +import torch +import torch.distributed as dist +from torch._utils import _get_device_module +from torch.distributed import distributed_c10d +from torch.distributed._shard.sharded_tensor import ( + Shard, + ShardedTensor, + ShardedTensorMetadata, + TensorProperties, +) +from torch.distributed._shard.sharding_spec import ShardMetadata +from torch.distributed.tensor import DeviceMesh, DTensor, Replicate, Shard as DShard + + +def _get_remote_device_str(rank, device_type, num_devices_per_node): + if device_type.lower() == "cpu": + return f"rank:{rank}/{device_type}" + elif device_type.lower() == "hpu": + return f"rank:{rank}/{device_type}:{_get_device_module(device_type).current_device()}" + else: + return f"rank:{rank}/{device_type}:{rank % num_devices_per_node}" + + +def _create_chunk_sharded_tensor( + tensor: torch.Tensor, + rank: int, + world_size: int, + num_devices_per_node: int, + pg: dist.ProcessGroup, + device: torch.device | None = None, +) -> ShardedTensor: + """ + Shard a tensor to chunks along the first dimension. The local rank will get its + corresponding chunk as the local shard to create a ShardedTensor. + """ + chunks = tensor.chunk(world_size, dim=0) + if len(chunks) > rank: + local_shard = chunks[rank].clone() + offsets = [0 for _ in tensor.size()] + offsets[0] = math.ceil(tensor.size()[0] / world_size) * rank + local_shards = [Shard.from_tensor_and_offsets(local_shard, offsets, rank)] + else: + local_shards = [] + + # Create a ShardedTensor without invoking communication. + chunk_sizes = [list(chunk.size()) for chunk in chunks] + dim0_offsets = [0] + list( + itertools.accumulate([chunk_size[0] for chunk_size in chunk_sizes]) + )[:-1] + offsets = [0] * (len(chunk_sizes[0]) - 1) + chunk_offsets = [[d0] + offsets for d0 in dim0_offsets] + device_type = ( + distributed_c10d._get_pg_default_device(pg).type + if device is None + else device.type + ) + placements = [ + _get_remote_device_str( + dist.get_global_rank(pg, r), + device_type, + num_devices_per_node, + ) + for r in range(len(chunk_sizes)) + ] + if len(chunk_sizes) != len(chunk_offsets) or len(chunk_sizes) != len(placements): + raise AssertionError( + f"Expected chunk_sizes, chunk_offsets, and placements to have the same length, " + f"got {len(chunk_sizes)}, {len(chunk_offsets)}, {len(placements)}" + ) + shard_metadata = [ + ShardMetadata(offset, size, placement) + for offset, size, placement in zip(chunk_offsets, chunk_sizes, placements) + ] + sharded_tensor_metadata = ShardedTensorMetadata( + shards_metadata=shard_metadata, + size=tensor.size(), + tensor_properties=TensorProperties( + dtype=tensor.dtype, + layout=tensor.layout, + requires_grad=False, + memory_format=torch.contiguous_format, + pin_memory=tensor.is_pinned(), + ), + ) + return ShardedTensor._init_from_local_shards_and_global_metadata( + local_shards, sharded_tensor_metadata=sharded_tensor_metadata, process_group=pg + ) + + +def _create_chunk_dtensor( + tensor: torch.Tensor, + rank: int, + device_mesh: DeviceMesh, +) -> DTensor: + """ + Shard a tensor to chunks along the first dimension. The local rank will get its + corresponding chunk as the local tensor to create a DTensor. + """ + # We need to explicitly call .detach() to return a new tensor detached from the current graph. + tensor = tensor.detach().clone() + + # FSDP placements: [Shard(0)] + # HSDP placements: [Replicate(), Shard(0)] + replicate_placements = [Replicate() for _ in range(device_mesh.ndim)] + shard_placements = [Replicate() for _ in range(device_mesh.ndim)] + shard_placements[-1] = DShard(0) # type: ignore[call-overload] + + return DTensor.from_local( + tensor, device_mesh, replicate_placements, run_check=False + ).redistribute( + placements=shard_placements, + ) + + +def _all_gather_dtensor( + tensor: DTensor, + root_mesh: DeviceMesh | None, +) -> torch.Tensor: + """ + All gather a DTensor in its sharded dimension and return the local tensor. + """ + if root_mesh != tensor.device_mesh: + raise AssertionError("The device mesh of a tensor should be a root mesh.") + + placements = list(copy.deepcopy(tensor.placements)) + # FSDP placements: [Shard(0)] -> [Replicate()] + # HSDP placements: [Replicate(), Shard(0)] -> [Replicate(), Replicate()] + placements[-1] = Replicate() + tensor = tensor.redistribute( + device_mesh=tensor.device_mesh, + placements=placements, + ) + + return tensor.to_local() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_state_dict_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_state_dict_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..39484524260c26055be0e305c561b88b446a873d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_state_dict_utils.py @@ -0,0 +1,932 @@ +# mypy: allow-untyped-defs +import contextlib +import logging +import math +import warnings +from collections.abc import Callable, Generator, Iterator +from typing import Any, cast, no_type_check + +import torch +import torch.distributed as dist +import torch.distributed.algorithms._checkpoint.checkpoint_wrapper as checkpoint_wrapper +import torch.nn as nn +import torch.nn.functional as F +from torch.distributed._shard.sharded_tensor import ( + init_from_local_shards, + Shard, + ShardedTensor, +) +from torch.distributed.fsdp._common_utils import ( + _FSDPState, + _get_module_fsdp_state_if_fully_sharded_module, + _has_fsdp_params, + _is_composable, + _module_handle, + clean_tensor_name, + FSDP_PREFIX, + FSDP_WRAPPED_MODULE, +) +from torch.distributed.fsdp._debug_utils import SimpleProfiler +from torch.distributed.fsdp._runtime_utils import ( + _cast_buffers_to_dtype_and_device, + _get_orig_buffer_dtypes, + _lazy_init, + _reset_flat_param_grad_info_if_needed, +) +from torch.distributed.fsdp.api import ( + FullStateDictConfig, + ShardingStrategy, + StateDictType, +) +from torch.distributed.tensor import DTensor +from torch.distributed.utils import _replace_by_prefix + +from ._fsdp_extensions import ( + _ext_all_gather_dtensor, + _ext_chunk_dtensor, + _ext_chunk_tensor, + _ext_post_unflatten_transform, + _ext_pre_load_state_dict_transform, +) +from ._unshard_param_utils import _unshard_fsdp_state_params, FLAT_PARAM + + +logger = logging.getLogger(__name__) + + +def _should_unshard_params(fsdp_state: _FSDPState) -> bool: + return not ( + fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD + and (_is_composable(fsdp_state) or fsdp_state._use_orig_params) + ) + + +def _convert_to_wrapped_module_name(module_name: str) -> str: + module_name = module_name.replace(f"{FSDP_PREFIX}", "") + module_name = module_name.replace(f"{FSDP_WRAPPED_MODULE}", "") + if module_name: + module_name = f"{module_name}." + # `CheckpointWrapper` adds a prefix that has to be removed as well. + module_name = module_name.replace(checkpoint_wrapper._CHECKPOINT_PREFIX, "") + return module_name + + +def _param_name_infos( + module: nn.Module, fsdp_state: _FSDPState +) -> Iterator[tuple[str, str, str]]: + if not _has_fsdp_params(fsdp_state, module): + return + for param_name, module_name in _module_handle( + fsdp_state, module + ).param_module_names(): + module_name = _convert_to_wrapped_module_name(module_name) + fqn = f"{module_name}{param_name}" + yield fqn, param_name, module_name + + +def _shared_param_name_infos( + module: nn.Module, fsdp_state +) -> Iterator[tuple[str, str, str]]: + for param_name, module_name in _module_handle( + fsdp_state, module + ).shared_param_module_names(): + module_name = _convert_to_wrapped_module_name(module_name) + fqn = f"{module_name}{param_name}" + yield fqn, param_name, module_name + + +@no_type_check +def _enter_unshard_params_ctx( + module: nn.Module, + fsdp_state: _FSDPState, + writeback: bool = False, + rank0_only: bool = False, + offload_to_cpu: bool = False, + with_grads: bool = False, +) -> None: + """ + state_dict hooks cannot use the pure context call as the checkpoint flow + requires to enter the context in the pre-hook but leave the context in the + post-hook. This API enters the context of ``_unshard_fsdp_state_params``. + """ + if module in fsdp_state._unshard_params_ctx: + raise AssertionError( + "Entering the ``_unshard_fsdp_state_params`` context but _unshard_params_ctx[module] " + "is not None." + ) + fsdp_state._unshard_params_ctx[module] = _unshard_fsdp_state_params( + module, + fsdp_state, + writeback=writeback, + rank0_only=rank0_only, + offload_to_cpu=offload_to_cpu, + with_grads=with_grads, + ) + fsdp_state._unshard_params_ctx[module].__enter__() + + +@no_type_check +def _exit_unshard_params_ctx(module: nn.Module, fsdp_state: _FSDPState) -> None: + """A helper function to exit ``_unshard_fsdp_state_params`` context.""" + fsdp_state._unshard_params_ctx[module].__exit__(None, None, None) + fsdp_state._unshard_params_ctx.pop(module) + + +def _common_pre_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, +) -> None: + """Performs the pre-state_dict tasks shared by all state_dict types.""" + if fsdp_state._device_handle.is_available(): + fsdp_state._device_handle.synchronize() + # TODO: need to check if this is always correct for composable FSDP. + _lazy_init(fsdp_state, module) + if fsdp_state._is_root: + _reset_flat_param_grad_info_if_needed(fsdp_state._all_handles) + + +def _common_unshard_pre_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + offload_to_cpu: bool, + rank0_only: bool, +) -> None: + """ + Performs the pre-state_dict tasks shared by all state_dict types that require + ``_unshard_fsdp_state_params()``. FULL_STATE_DICT and SHARDED_STATE_DICT use this hook. + """ + # For composable `fully_shard`, it does not need to unshard parameters for `NO_SHARD` cases. + if not _should_unshard_params(fsdp_state): + return + _enter_unshard_params_ctx( + module, + fsdp_state, + writeback=False, + offload_to_cpu=offload_to_cpu, + rank0_only=rank0_only, + ) + + +@no_type_check +def _common_unshard_post_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, + param_hook: Callable, +) -> dict[str, Any]: + """ + The post-state_dict flow that shared by all state_dict types that require + ``_unshard_fsdp_state_params()``. FULL_STATE_DICT and SHARDED_STATE_DICT use this + hook. + """ + _replace_by_prefix(state_dict, prefix + f"{FSDP_PREFIX}", prefix) + # Return early for trivial cases + if not state_dict or not _has_fsdp_params(fsdp_state, module): + if _should_unshard_params(fsdp_state): + _exit_unshard_params_ctx(module, fsdp_state) + return state_dict + + # If a rank does not have unsharded parameters(when `rank0_only=True` + # and `rank != 0`), then the rank only needed to participate in the + # all-gather and does not need to save the # state dict. We simply check + # rank0_only to ensure this issue. + rank0_only = ( + fsdp_state._state_dict_type == StateDictType.FULL_STATE_DICT + and cast(FullStateDictConfig, fsdp_state._state_dict_config).rank0_only + ) + # no_fsdp_return means the state_dict returned by this rank should contain + # only non-FSDP controlled parameters and buffers. + no_fsdp_return = rank0_only and fsdp_state.rank != 0 + if no_fsdp_return and not fsdp_state._use_orig_params: + for clean_key in fsdp_state._buffer_names: + # This is a hack to support activation checkpoint. + clean_key = clean_key.replace( + f"{checkpoint_wrapper._CHECKPOINT_PREFIX}.", "" + ) + state_dict.pop(f"{prefix}{clean_key}", None) + # Non-zero ranks have flat_param key when rank0_only=True, because rank0_only=True is + # passed in to unshard context, but nonzero ranks reshard early, causing this flat_param + # to appear in state_dict. + state_dict.pop(f"{prefix}{FLAT_PARAM}") + _exit_unshard_params_ctx(module, fsdp_state) + return state_dict + + # Loop only the parameters saved in this instance's wrapped module to + # avoid processing buffers. + for fqn, param_name, module_name in _param_name_infos(module, fsdp_state): + fqn = f"{prefix}{fqn}" + if no_fsdp_return: + state_dict.pop(fqn) + continue + if fqn not in state_dict: + raise AssertionError( + f"FSDP assumes {fqn} is in the state_dict but the state_dict only " + f"has {state_dict.keys()}. " + f"prefix={prefix}, module_name={module_name}, " + f"param_name={param_name} rank={fsdp_state.rank}." + ) + + param_hook(state_dict, prefix, fqn) + + if _should_unshard_params(fsdp_state): + _exit_unshard_params_ctx(module, fsdp_state) + + cpu_device = torch.device("cpu") + buffer_clean_fqns = [] + buffers = [] + for clean_key in fsdp_state._buffer_names: + # This is a hack to support activation checkpoint. + clean_key = clean_tensor_name(clean_key) + fqn = f"{prefix}{clean_key}" + if fqn not in state_dict: + # A buffer can be registered as non-persistent. + continue + if no_fsdp_return: + state_dict.pop(fqn) + else: + buffer = state_dict[fqn] + if ( + fsdp_state._state_dict_config.offload_to_cpu + and buffer.device != cpu_device + ): + state_dict[fqn] = buffer.to(cpu_device) + # skip upcasting for ignored buffers + if clean_key not in fsdp_state._ignored_buffer_names: + buffer_clean_fqns.append(clean_key) + buffers.append(state_dict[fqn]) + + if buffers: + mixed_precision_enabled_for_buffers = ( + fsdp_state._mixed_precision_enabled_for_buffers() + if not _is_composable(fsdp_state) + else (fsdp_state.mixed_precision.buffer_dtype is not None) + ) + if mixed_precision_enabled_for_buffers: + buffer_dtypes = _get_orig_buffer_dtypes(fsdp_state, buffer_clean_fqns) + _cast_buffers_to_dtype_and_device( + buffers, buffer_dtypes, fsdp_state.compute_device + ) + for buffer, clean_fqn in zip(buffers, buffer_clean_fqns): + fqn = f"{prefix}{clean_fqn}" + logger.info("FSDP is casting the dtype of %s to %s", fqn, buffer.dtype) + state_dict[fqn] = buffer.clone() + return state_dict + + +@no_type_check +def _full_pre_state_dict_hook( + fsdp_state: _FSDPState, + module: nn.Module, + *args, + **kwargs, +) -> None: + """ + Hook that runs before model.state_dict() is called. pre-state_dict hook is + not actually supported by ``nn.Module``. As a result, this API is called + from ``_full_post_state_dict_hook()`` to simulate the case. Once pre-state_dict + is supported in ``nn.Module``, this hook will be registered as a hook in + ``nn.Module``. + """ + if getattr(fsdp_state, "_device_mesh", False): + fsdp_state._device_mesh._get_root_mesh() + + _common_pre_state_dict_hook(module, fsdp_state) + _common_unshard_pre_state_dict_hook( + module, + fsdp_state, + offload_to_cpu=fsdp_state._state_dict_config.offload_to_cpu, + rank0_only=cast(FullStateDictConfig, fsdp_state._state_dict_config).rank0_only, + ) + + +@no_type_check +def _full_post_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> dict[str, Any]: + """ + Hook that runs after model.state_dict() is called before returning result to + user. For FSDP, we may have to clone the tensors in state_dict as params go + back to sharded version after _unshard_fsdp_state_params ends, and also remove + the ``FSDP_WRAPPED_MODULE`` prefix. + """ + + def param_hook( + state_dict: dict[str, Any], + prefix: str, + fqn: str, + ) -> None: + clean_key = fqn + clean_prefix = clean_tensor_name(prefix) + # Strip prefix out of key if needed as buffer names and param names + # do not have prefix considered as they are not computed in `state_dict` + # call. + clean_key = clean_key.removeprefix(clean_prefix) + + # Clone parameters before exiting the `_unshard_fsdp_state_params()` context. + if not getattr(state_dict[fqn], "_has_been_cloned", False): + try: + state_dict[fqn] = state_dict[fqn].detach().clone() + state_dict[fqn]._has_been_cloned = True # type: ignore[attr-defined] + except BaseException as e: + warnings.warn( + f"Failed to clone() tensor with name {fqn} on rank {fsdp_state.rank}. " + "This may mean that this state_dict entry could point to invalid " + "memory regions after returning from state_dict() call if this " + "parameter is managed by FSDP. Please check clone " + f"implementation of {fqn}. Error: {str(e)}", + stacklevel=2, + ) + + return _common_unshard_post_state_dict_hook( + module, fsdp_state, state_dict, prefix, param_hook + ) + + +def _full_pre_load_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> None: + _lazy_init(fsdp_state, module) + if _should_unshard_params(fsdp_state): + with SimpleProfiler.profile("_enter_unshard_params_ctx"): + _enter_unshard_params_ctx(module, fsdp_state, writeback=True) + # Add FSDP_PREFIX only for wrapper-based FSDP. + if not _is_composable(fsdp_state): + _replace_by_prefix(state_dict, prefix, prefix + f"{FSDP_PREFIX}") + + +def _full_post_load_state_dict_hook( + module: nn.Module, fsdp_state: _FSDPState, *args, **kwargs +) -> None: + if _should_unshard_params(fsdp_state): + with SimpleProfiler.profile("_exit_unshard_params_ctx"): + _exit_unshard_params_ctx(module, fsdp_state) + + +def _local_pre_state_dict_hook( + fsdp_state: _FSDPState, + module: nn.Module, + *args, + **kwargs, +) -> None: + """ + Hook that runs before model.state_dict() is called. Right now, pre-state_dict + hook is not supported by the PyTorch core. So this API is called from + `_local_post_state_dict_hook()` to simulate the case. + """ + if ( + _has_fsdp_params(fsdp_state, module) + and not _module_handle(fsdp_state, module).uses_sharded_strategy + ): + raise RuntimeError( + "``local_state_dict`` can only be used when parameters are flatten " + "and sharded." + ) + _common_pre_state_dict_hook(module, fsdp_state) + + +@no_type_check +def _local_post_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> dict[str, Any]: + """ + This hook create a ShardedTensor from the local flat_param and replace + the state_dict[f"{prefix}{FLAT_PARAM}] with the ShardedTensor. No copy + will happen. The underlying storage is the same. + """ + + _replace_by_prefix(state_dict, f"{prefix}{FSDP_PREFIX}", prefix) + if not _has_fsdp_params(fsdp_state, module): + return state_dict + + # state_dict[f"{prefix}{FLAT_PARAM}"] exists and has the same tensor + # value as the flat_param but it is a pure Tensor because + # nn.Module.state_dict() will detach the parameter. Therefore, we need + # to get flat_param to get the metadata. + if not _module_handle(fsdp_state, module): + raise AssertionError("Should have returned early") + flat_param = _module_handle(fsdp_state, module).flat_param + # Constructs a ShardedTensor from the flat_param "without" padding. + # Removing the padding allows users to change the number of ranks + # when loading the local_state_dict. + full_numel = flat_param._unpadded_unsharded_size.numel() # type: ignore[attr-defined] + shard_offset = flat_param.numel() * fsdp_state.rank + valid_data_size = flat_param.numel() - flat_param._shard_numel_padded + if valid_data_size > 0: + # If FlatParameter is returned, FlatParameter._local_shard cause a + # pickling issue (can be torch.save but not torch.load). Since there + # is no benefit for state_dict to return the actual FlatParameter class, + # a view (which is a tensor) of the FlatParameter will be returned. + flat_param = flat_param[:valid_data_size].view(valid_data_size) + local_shards = [ + Shard.from_tensor_and_offsets(flat_param, [shard_offset], fsdp_state.rank) + ] + else: + local_shards = [] + sharded_tensor = init_from_local_shards( + local_shards, full_numel, process_group=fsdp_state.process_group + ) # type: ignore[assignment] + # TODO: Add DTensor state_dict support for LOCAL_STATE_DICT. + if fsdp_state._state_dict_config.offload_to_cpu: + sharded_tensor = sharded_tensor.cpu() + state_dict[f"{prefix}{FLAT_PARAM}"] = sharded_tensor + return state_dict + + +def _local_post_load_state_dict_hook( + module: nn.Module, fsdp_state: _FSDPState, *args, **kwargs +) -> None: + pass + + +def _local_pre_load_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> None: + """ + This hook finds the local flat_param for this FSDP module from the + state_dict. The flat_param should be a ShardedTensor. This hook converts + the ShardedTensor to a tensor. No copy happen unless padding is required. + """ + _lazy_init(fsdp_state, module) + _replace_by_prefix(state_dict, prefix, f"{prefix}{FSDP_PREFIX}") + fqn = f"{prefix}{FSDP_PREFIX}{FLAT_PARAM}" + if fqn not in state_dict: + if _has_fsdp_params(fsdp_state, module): + raise AssertionError( + "No `FlatParameter` in `state_dict` for this FSDP instance " + "but it has parameters" + ) + return + load_tensor = state_dict[fqn] + if not isinstance(load_tensor, ShardedTensor): + raise AssertionError("Tensors in local_state_dict should be ShardedTensor.") + + # Convert the ShardedTensor to a Tensor. + flat_param = _module_handle(fsdp_state, module).flat_param + if flat_param is None: + raise AssertionError("Expected flat_param to be set") + valid_data_size = flat_param.numel() - flat_param._shard_numel_padded + shards = load_tensor.local_shards() + if valid_data_size > 0: + if not len(shards): + raise AssertionError( + "load_local_state_dict assume one shard per ShardedTensor." + ) + load_tensor = shards[0].tensor + + # Get the metadata of the flat_param to decide whether to pad the loaded + # tensor. + if flat_param._shard_numel_padded > 0: + if load_tensor.numel() >= flat_param.numel(): + raise AssertionError( + f"Local shard size = {flat_param.numel()} and the tensor in " + f"the state_dict is {load_tensor.numel()}." + ) + load_tensor = F.pad(load_tensor, [0, flat_param._shard_numel_padded]) + else: + load_tensor = flat_param + # TODO: Add DTensor state_dict support for LOCAL_STATE_DICT. + state_dict[fqn] = load_tensor + + +def _sharded_pre_state_dict_hook( + fsdp_state: _FSDPState, + module: nn.Module, + *args, + **kwargs, +) -> None: + """ + Hook that runs before model.state_dict() is called. Check + ``_full_pre_load_state_dict_hook`` for the detail. + """ + if ( + _has_fsdp_params(fsdp_state, module) + and not _module_handle(fsdp_state, module).uses_sharded_strategy + ): + raise RuntimeError( + "``sharded_state_dict`` can only be used when parameters are flatten " + "and sharded." + ) + _common_pre_state_dict_hook(module, fsdp_state) + # Setting offload_to_cpu here does not work even if offload_to_cpu is True. + # We have to create ShardedTensor first then move it to CPU. + _common_unshard_pre_state_dict_hook( + module, + fsdp_state, + offload_to_cpu=False, + rank0_only=False, + ) + + +@no_type_check +def _sharded_post_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> dict[str, Any]: + """ + The hook replaces the unflattened, unsharded parameter in the state_dict + with a unflattened, sharded parameter (a ShardedTensor). + """ + + def param_hook(state_dict: dict[str, Any], prefix: str, fqn: str): + param = state_dict[fqn] + if not fsdp_state._state_dict_config._use_dtensor: + sharded_tensor = _ext_chunk_tensor( + tensor=param, + rank=fsdp_state.rank, + world_size=fsdp_state.world_size, + num_devices_per_node=fsdp_state._device_handle.device_count(), + pg=fsdp_state.process_group, + fsdp_extension=fsdp_state._fsdp_extension, + ) + else: + sharded_tensor = _ext_chunk_dtensor( + tensor=param, + rank=fsdp_state.rank, + device_mesh=fsdp_state._device_mesh, + fsdp_extension=fsdp_state._fsdp_extension, + ) + if fsdp_state._state_dict_config.offload_to_cpu: + sharded_tensor = sharded_tensor.cpu() + state_dict[fqn] = sharded_tensor + + return _common_unshard_post_state_dict_hook( + module, fsdp_state, state_dict, prefix, param_hook + ) + + +@no_type_check +def _sharded_post_load_state_dict_hook( + module: nn.Module, fsdp_state: _FSDPState, *args, **kwargs +) -> None: + if _has_fsdp_params(fsdp_state, module): + with SimpleProfiler.profile("_exit_unshard_params_ctx"): + _exit_unshard_params_ctx(module, fsdp_state) + + +@no_type_check +def _sharded_pre_load_state_dict_hook( + module: nn.Module, + fsdp_state: _FSDPState, + state_dict: dict[str, Any], + prefix: str, +) -> None: + """ + The hook combines the unflattened, sharded parameters (ShardedTensor) to + a new FlatParameter and shards the new FlatParameter to the local chunk. + """ + _lazy_init(fsdp_state, module) + if not _is_composable(fsdp_state): + _replace_by_prefix(state_dict, prefix, prefix + f"{FSDP_PREFIX}") + if not _has_fsdp_params(fsdp_state, module): + return + + handle = _module_handle(fsdp_state, module) + if not handle.uses_sharded_strategy: + raise RuntimeError( + "load_sharded_state_dict can only be called when parameters " + "are flattened and sharded." + ) + fqn_to_param_ext = dict( + zip(handle.flat_param._fqns, handle.flat_param._param_extensions) + ) + + for fqn, _, _ in _param_name_infos(module, fsdp_state): + if not _is_composable(fsdp_state): + fqn_from_global_root = f"{prefix}{FSDP_PREFIX}{fqn}" + else: + fqn_from_global_root = f"{prefix}{fqn}" + try: + param = state_dict.pop(fqn_from_global_root) + except KeyError: + logger.warning( + f"Did not find param with FQN {fqn_from_global_root}, skipping it. " # noqa: G004 + "The weight will not be filled if you expect it to be." + ) + continue # TODO: Improve unittesting for state_dict finetuning + # cases: https://github.com/pytorch/pytorch/issues/109134 + + if not fsdp_state._state_dict_config._use_dtensor: + # All-gather the param (ShardedTensor) + param, shards = _ext_pre_load_state_dict_transform( + param, fsdp_state._fsdp_extension + ) + + if len(shards) >= 2: + raise AssertionError( + "Expects 0 or 1 shard per rank " + f"but got {len(shards)} shards on rank {fsdp_state.rank}." + ) + param_numel = param.size().numel() + dim_0_size = param.size()[0] + chunk_size = ( + math.ceil(dim_0_size / fsdp_state.world_size) + * param_numel + // dim_0_size + ) + if len(shards) == 1: + local_tensor = shards[0].tensor.flatten() + with SimpleProfiler.profile(SimpleProfiler.Type.H2D): + local_tensor = local_tensor.to(fsdp_state.compute_device) + num_padding = chunk_size - local_tensor.numel() + if num_padding > 0: + local_tensor = F.pad(local_tensor, [0, num_padding]) + else: + local_tensor = torch.zeros( + chunk_size, dtype=param.dtype, device=fsdp_state.compute_device + ) + tensor = torch.empty( + chunk_size * fsdp_state.world_size, + dtype=local_tensor.dtype, + device=fsdp_state.compute_device, + ) + with SimpleProfiler.profile(SimpleProfiler.Type.ALLGATHER): + dist.all_gather_single( + tensor, local_tensor, group=fsdp_state.process_group + ) + tensor = tensor.narrow(0, 0, param_numel).reshape(param.size()) + state_dict[fqn_from_global_root] = tensor + else: + if param.device != fsdp_state._device_mesh.device_type: + param = param.to(fsdp_state._device_mesh.device_type) + + root_mesh = fsdp_state._device_mesh._get_root_mesh() + local_tensor = _ext_all_gather_dtensor( + param, root_mesh, fsdp_state._fsdp_extension + ) + + if fqn_to_param_ext.get(fqn) is not None: + ext = fqn_to_param_ext[fqn] + local_tensor = _ext_post_unflatten_transform( + local_tensor, ext, fsdp_state._fsdp_extension + ) + state_dict[fqn_from_global_root] = local_tensor + + with SimpleProfiler.profile("_enter_unshard_params_ctx"): + _enter_unshard_params_ctx(module, fsdp_state, writeback=True) + + +@contextlib.contextmanager +def _replace_with_full_state_dict_type(fsdp_state: _FSDPState) -> Generator: + old_state_dict_config = fsdp_state._state_dict_config + old_state_dict_type = fsdp_state._state_dict_type + fsdp_state._state_dict_config = FullStateDictConfig() + fsdp_state._state_dict_type = StateDictType.FULL_STATE_DICT + yield + fsdp_state._state_dict_config = old_state_dict_config + fsdp_state._state_dict_type = old_state_dict_type + + +@no_type_check +@torch.no_grad() +def _post_state_dict_hook( + module: nn.Module, + state_dict: dict[str, Any], + prefix: str, + *args: Any, +) -> dict[str, Any]: + """ + _post_state_dict_hook() is called after the state_dict() of this + FSDP module is executed. ``fsdp_state._state_dict_type`` is used to decide + what postprocessing will be done. + """ + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD: + context = _replace_with_full_state_dict_type(fsdp_state) + warnings.warn( + "When using ``NO_SHARD`` for ``ShardingStrategy``, full_state_dict will " + "be returned.", + stacklevel=2, + ) + else: + context = contextlib.nullcontext() + + with context: + _post_state_dict_hook_fn = { + StateDictType.FULL_STATE_DICT: _full_post_state_dict_hook, + StateDictType.LOCAL_STATE_DICT: _local_post_state_dict_hook, + StateDictType.SHARDED_STATE_DICT: _sharded_post_state_dict_hook, + } + processed_state_dict = _post_state_dict_hook_fn[fsdp_state._state_dict_type]( + module, fsdp_state, state_dict, prefix + ) + + if fsdp_state._is_root: + logger.info("FSDP finished processing state_dict(), prefix=%s", prefix) + for key, tensor in sorted(processed_state_dict.items()): + if key.startswith(prefix) and isinstance(tensor, torch.Tensor): + local_shape = tensor.shape + device = None + if isinstance(tensor, ShardedTensor): + local_shape = None + shards = tensor.local_shards() + if shards: + local_shape = shards[0].tensor.shape + device = shards[0].tensor.device + elif isinstance(tensor, DTensor): + local_shape = tensor.to_local().shape + device = tensor.device + else: + device = tensor.device + logger.info( + "FQN=%s: type=%s, shape=%s, local_shape=%s, dtype=%s, device=%s", + key, + type(tensor), + tensor.shape, + local_shape, + tensor.dtype, + device, + ) + + return processed_state_dict + + +@no_type_check +@torch.no_grad() +def _pre_state_dict_hook( + module: nn.Module, + *args, + **kwargs, +) -> None: + """ + This is called before the core state dict saving logic of ``module``. + ``fsdp_state._state_dict_type`` is used to decide what postprocessing will + be done. + """ + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD: + context = _replace_with_full_state_dict_type(fsdp_state) + warnings.warn( + "When using ``NO_SHARD`` for ``ShardingStrategy``, full_state_dict will " + "be returned.", + stacklevel=2, + ) + else: + _set_use_dtensor(fsdp_state) + context = contextlib.nullcontext() + + with context: + _pre_state_dict_hook_fn = { + StateDictType.FULL_STATE_DICT: _full_pre_state_dict_hook, + StateDictType.LOCAL_STATE_DICT: _local_pre_state_dict_hook, + StateDictType.SHARDED_STATE_DICT: _sharded_pre_state_dict_hook, + } + _pre_state_dict_hook_fn[fsdp_state._state_dict_type]( + fsdp_state, + module, + *args, + **kwargs, + ) + + +@no_type_check +def _set_use_dtensor(fsdp_state: _FSDPState) -> None: + # If device_mesh is passed in when initializing FSDP, we automatically turn the + # _use_dtensor flag to be true for ShardedStateDictConfig(). + if getattr(fsdp_state, "_device_mesh", None): + state_dict_type = fsdp_state._state_dict_type + if state_dict_type == StateDictType.LOCAL_STATE_DICT: + raise RuntimeError( + "Found state_dict_type LOCAL_STATE_DICT. " + "DeviceMesh is not compatible with LOCAL_STATE_DICT. " + "Please set state_dict_type to SHARDED_STATE_DICT to get DTensor state_dict." + ) + else: + fsdp_state._state_dict_config._use_dtensor = True + + +@no_type_check +@torch.no_grad() +def _pre_load_state_dict_hook( + module: nn.Module, + state_dict: dict[str, Any], + prefix: str, + *args: Any, +) -> None: + """ + This is called before ``module._load_from_state_dict()``. + ``fsdp_state._state_dict_type`` is used to decide what preprocessing will + be done. + """ + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD: + context = _replace_with_full_state_dict_type(fsdp_state) + warnings.warn( + "When using ``NO_SHARD`` for ``ShardingStrategy``, full_state_dict will" + "be returned.", + stacklevel=2, + ) + else: + _set_use_dtensor(fsdp_state) + context = contextlib.nullcontext() + + _lazy_init(fsdp_state, module) + if fsdp_state._is_root: + SimpleProfiler.reset() + + with context: + _pre_load_state_dict_hook_fn = { + StateDictType.FULL_STATE_DICT: _full_pre_load_state_dict_hook, + StateDictType.LOCAL_STATE_DICT: _local_pre_load_state_dict_hook, + StateDictType.SHARDED_STATE_DICT: _sharded_pre_load_state_dict_hook, + } + # Code that is common for all state_dict impls + if fsdp_state._device_handle.is_available(): + fsdp_state._device_handle.synchronize() + # Dispatch into state_dict specific implementation of pre-hook. + _pre_load_state_dict_hook_fn[fsdp_state._state_dict_type]( + module, fsdp_state, state_dict, prefix + ) + + +@no_type_check +@torch.no_grad() +def _post_load_state_dict_hook( + module: nn.Module, + incompatible_keys: tuple[list[str], list[str]], + *args: Any, +) -> None: + fsdp_state = _get_module_fsdp_state_if_fully_sharded_module(module) + if fsdp_state.sharding_strategy == ShardingStrategy.NO_SHARD: + context = _replace_with_full_state_dict_type(fsdp_state) + warnings.warn( + "When using ``NO_SHARD`` for ``ShardingStrategy``, full_state_dict will" + "be returned.", + stacklevel=2, + ) + else: + context = contextlib.nullcontext() + + with context: + _post_load_state_dict_hook_fn = { + StateDictType.FULL_STATE_DICT: _full_post_load_state_dict_hook, + StateDictType.LOCAL_STATE_DICT: _local_post_load_state_dict_hook, + StateDictType.SHARDED_STATE_DICT: _sharded_post_load_state_dict_hook, + } + # Code that is common for all state_dict impls + # Dispatch into state_dict type specific implementation of post-hook for + # loading state_dict. + _post_load_state_dict_hook_fn[fsdp_state._state_dict_type](module, fsdp_state) + + # When reporting incompatible keys, trim FSDP prefixes. + missing_keys = incompatible_keys[0] + unexpected_keys = incompatible_keys[1] + for i in range(len(missing_keys)): + missing_keys[i] = clean_tensor_name(missing_keys[i]) + + for i in range(len(unexpected_keys)): + unexpected_keys[i] = clean_tensor_name(unexpected_keys[i]) + + if fsdp_state._is_root: + SimpleProfiler.dump_and_reset("FSDP model load_state_dict profiling: ") + + +def _register_all_state_dict_hooks(state: _FSDPState): + """ + Registers pre-save, post-save, pre-load, and post-load state dict hooks. + """ + for hook_registration_fn_str, hook, hook_registration_fn_kwargs in ( + ("register_state_dict_pre_hook", _pre_state_dict_hook, {}), + ("_register_state_dict_hook", _post_state_dict_hook, {}), + ( + "_register_load_state_dict_pre_hook", + _pre_load_state_dict_hook, + {"with_module": True}, + ), + ("register_load_state_dict_post_hook", _post_load_state_dict_hook, {}), + ): + _register_state_dict_hooks_base( + state, hook_registration_fn_str, hook, hook_registration_fn_kwargs + ) + + +@no_type_check +def _register_state_dict_hooks_base( + state: _FSDPState, + hook_registration_fn_name: str, + hook: Callable, + hook_registration_fn_kwargs: dict[str, Any], +) -> None: + """Registers ``hook`` using ``hook_registration_fn``.""" + if not _is_composable(state): + getattr(state, hook_registration_fn_name)(hook, **hook_registration_fn_kwargs) + else: + handle = state._handle + if handle: + getattr(handle._fully_sharded_module, hook_registration_fn_name)( + hook, **hook_registration_fn_kwargs + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_trace_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_trace_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..930c91d67cec8e442d4e452b94d71b016c34769b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_trace_utils.py @@ -0,0 +1,240 @@ +# mypy: allow-untyped-defs +import functools +from collections.abc import Callable +from contextlib import contextmanager +from dataclasses import dataclass, field +from typing import Any, NamedTuple + +import torch +import torch.nn as nn + + +@dataclass +class TracingConfig: + """ + This represents a symbolic tracing configuration. + + Args: + tracer (torch.fx.Tracer): An instance of :class:`torch.fx.Tracer` to + use for symbolic tracing. The default value is the native + :class:`torch.fx.Tracer` constructed with default arguments. + However, the user may want to pass a different value such as the + ``HFTracer`` for models in the HuggingFace Transformers_ library. + .. _Transformers: https://huggingface.co/docs/transformers/index + concrete_args (Optional[Dict[str, Any]]): Concrete arguments that + should not be treated as ``torch.fx.Proxy`` when tracing the + module ``forward()``. Passing ``concrete_args`` allows partially + specializing the forward, e.g. to remove control flow or data + structures. This ``concrete_args`` here is the same argument used + in :meth:`~torch.fx.Tracer.trace`. + """ + + tracer: torch.fx.Tracer = field(default_factory=torch.fx.Tracer) + concrete_args: dict[str, Any] | None = None + + +class _ParamUsageInfo(NamedTuple): + """ + This is used for ``_ExecutionInfo.module_to_param_usage_infos`` to record + execution information. The ``dict`` maps modules to a list of these + ``_ParamUsageInfo`` instances, where each instance represents a group of + parameters used together. + + Specifically, for each module key in the ``dict``, each instance of this + class represents either: + (1) the module and some sublist of its ``named_parameters()`` used + together in execution (see ``_patched_create_proxy()``), or + (2) a submodule and all of ``submodule.named_parameters()`` (see + ``_patched_call_module()``). + + Type (1) corresponds to directly using parameters in ops without calling + ``forward()``, and type (2) corresponds to calling ``forward()``. The + mapped-to lists in the ``dict`` follow the execution order. + """ + + module: nn.Module + named_params: list[tuple[str, nn.Parameter]] + + +class _ExecutionInfo: + """ + This represents the execution order information from the forward pass. + + Attributes: + curr_module (nn.Module): Current module being traced. + module_forward_order (List[nn.Module]): The modules in (pre-)forward + order, i.e. the order in which their ``forward()`` methods are + called. Each call to a module's ``forward()`` corresponds to one + element in the list. + module_to_param_usage_infos (Dict[nn.Module, List[_ParamUsageInfo]]): + Maps a module to a list of module execution infos. See + :class:`_ParamUsageInfo` for details. + param_forward_order (List[nn.Parameter]): The parameters in forward + execution order, where only a parameter's first participation is + included. + visited_params (Set[nn.Parameter]): The parameters visited so far + during the trace. This is only used during tracing for fast + membership check. Invariant: The parameters in + ``param_forward_order`` are exactly those in ``visited_params``. + """ + + def __init__(self, root_module: nn.Module) -> None: + self.curr_module: nn.Module = root_module + self.module_forward_order: list[nn.Module] = [root_module] + self.module_to_param_usage_infos: dict[nn.Module, list[_ParamUsageInfo]] = { + root_module: [] + } + self.param_forward_order: list[nn.Parameter] = [] + self.visited_params: set[nn.Parameter] = set() + + +class _ExecOrderTracer: + def __init__(self) -> None: + self.exec_info: _ExecutionInfo | None = None + + @contextmanager + def patch_tracer(self, tracer: torch.fx.Tracer, root_module: nn.Module): + self.exec_info = _ExecutionInfo(root_module) + orig_call_module = tracer.call_module + orig_create_proxy = tracer.create_proxy + tracer.call_module = functools.partial( # type: ignore[method-assign] + self._patched_call_module, orig_call_module, self.exec_info + ) + fqn_to_param = dict(root_module.named_parameters()) + tracer.create_proxy = functools.partial( # type: ignore[method-assign] + self._patched_create_proxy, + orig_create_proxy, + self.exec_info, + fqn_to_param, + ) + try: + yield + finally: + tracer.call_module = orig_call_module # type: ignore[method-assign] + tracer.create_proxy = orig_create_proxy # type: ignore[method-assign] + + def _patched_call_module( + self, + call_module: Callable, + exec_info: _ExecutionInfo, + # Below are the expected arguments to `call_module()` + module: nn.Module, + forward: Callable, + args: tuple[Any, ...], + kwargs: dict[str, Any], + ) -> Any: + """ + Overrides ``call_module`` to save execution information to + ``exec_info``. Note that ``call_module`` is called during symbolic + tracing for each non-root module. + + Args: + call_module (Callable): Original ``call_module`` to override. + exec_info (_ExecutionInfo): Used to record execution information. + module (nn.Module): Module corresponding to this ``call_module``. + forward (Callable): ``forward()`` method of ``module`` to be called + for this ``call_module``. + args (Tuple[Any, ...]): Positional arguments for ``forward``. + kwargs (Dict[str, Any]): Keyword arguments for ``forward``. + + Returns: + Same return value as ``call_module``. + """ + exec_info.module_forward_order.append(module) + named_params = list(module.named_parameters()) + curr_module = exec_info.curr_module + if named_params: + if curr_module not in exec_info.module_to_param_usage_infos: + raise AssertionError( + "The current module should have already been processed by a patched `call_module`" + ) + exec_info.module_to_param_usage_infos[exec_info.curr_module].append( + _ParamUsageInfo(module, named_params) + ) + prev_curr_module = curr_module + exec_info.curr_module = module + exec_info.module_to_param_usage_infos[module] = [] + output = call_module(module, forward, args, kwargs) + exec_info.curr_module = prev_curr_module + return output + + def _patched_create_proxy( + self, + create_proxy: Callable, + exec_info: _ExecutionInfo, + fqn_to_param: dict[str, nn.Parameter], + # Below are the expected arguments to `create_proxy()` + kind: str, + target: torch.fx.node.Target, + args: tuple[Any, ...], + kwargs: dict[str, Any], + name: str | None = None, + type_expr: Any | None = None, + proxy_factory_fn: Callable[[torch.fx.Node], torch.fx.Proxy] | None = None, + ) -> torch.fx.Proxy: + """ + Overrides ``create_proxy`` to save execution information to + ``exec_info``. Note that ``create_proxy`` is called during symbolic + tracing for each leaf function/method/module. + + Args: + create_proxy (Callable): Original ``create_proxy`` to override. + exec_info (_ExecutionInfo): Used to record execution information. + fqn_to_param (Dict[str, nn.Parameter]): ``dict`` version of the + root module's ``named_parameters()`` with FQN as key and + parameter as value. + kind (str): Kind of the target method ('call_function', + 'call_method', 'get_attr', 'call_module', 'placeholder', or + 'output'). See :class:`torch.fx.Graph` for details. This is + passed to ``create_proxy``. + target (torch.fx.node.Target): Contains the string name of the + function/method/module. This is passed to ``create_proxy``. + args (Tuple[Any, ...]): Positional arguments for the function/ + method/module. This is passed to ``create_proxy``. + kwargs (Dict[str, Any]): Keyword arguments for the function/method/ + module. This is passed to ``create_proxy`` + name (Optional[str]): An optional string name for the ``Node`` + created in ``create_proxy``. This is passed to + ``create_proxy``. + type_expr (Optional[Any]): An optional type annotation representing + the Python type that the output of the node has. This is passed + to ``create_proxy``. + proxy_factory_fn (Callable[[torch.fx.Node], torch.fx.Proxy]): + An alternative proxy constructor used in ``create_proxy``. This + is passed to ``create_proxy``. + + Returns: + torch.fx.Proxy: Created ``Node`` wrapped in a ``Proxy`` object. + """ + proxy = create_proxy( + kind, target, args, kwargs, name, type_expr, proxy_factory_fn + ) + curr_module = exec_info.curr_module + if kind in ("call_function", "call_method"): + if args is not None: + named_params: list[tuple[str, nn.Parameter]] = [] + for arg in args: + if ( + isinstance(arg, torch.fx.Proxy) + and arg.node.target in fqn_to_param + ): + param = fqn_to_param[arg.node.target] # type: ignore[index] + named_params.append((arg.node.target, param)) # type: ignore[arg-type] + if param not in exec_info.visited_params: + exec_info.visited_params.add(param) + exec_info.param_forward_order.append(param) + if named_params: + exec_info.module_to_param_usage_infos[curr_module].append( + _ParamUsageInfo(curr_module, named_params) + ) + elif kind == "call_module": + named_params = list(curr_module.named_parameters()) + if named_params: + exec_info.module_to_param_usage_infos[curr_module].append( + _ParamUsageInfo(curr_module, named_params) + ) + for _, param in named_params: + if param not in exec_info.visited_params: + exec_info.visited_params.add(param) + exec_info.param_forward_order.append(param) + return proxy diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_traversal_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_traversal_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..51140d3b0a8d3d16ab50226b414e651f22772648 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_traversal_utils.py @@ -0,0 +1,112 @@ +""" +NOTE: This file must be imported like +``import torch.distributed.fsdp._traversal_utils`` and not like +``from torch.distributed.fsdp._traversal_utils import ...`` to avoid circular +imports. For brevity, we may import the file as ``traversal_utils``. +""" + +import collections + +import torch.nn as nn +from torch.distributed._composable.contract import _get_registry +from torch.distributed.fsdp._common_utils import _FSDPState, _get_module_fsdp_state + + +""" +[Note: FSDP State Traversal] +For the wrapper code path, ``_FSDPState`` is the ``FullyShardedDataParallel`` +module wrapping a fully sharded module, and for the non-wrapper code path, +``_FSDPState`` is an object that gets embedded on a fully sharded module. +See [Note: Fully Sharded Module] for the definition. + +There are three common traversal idioms: Given a root module, +- ``_get_fsdp_states()`` returns all ``_FSDPState`` s in the tree. +- ``get_fsdp_root_states()`` returns all local root ``_FSDPState`` s in the +tree (i.e. those with ``_is_root == True``). +- ``_get_fsdp_handles()``returns all ``FlatParamHandle`` s in the tree. + +All of these methods must take in the root module (i.e. an ``nn.Module``) and +not a general ``_FSDPState`` because ``_FSDPState`` does not support a graph +traversal, whereas ``nn.Module`` has ``nn.Module.modules()`` for traversal. +""" + + +def _composable(module: nn.Module) -> bool: + """ + Returns if ``module`` can compose with ``fully_shard``. + """ + # TODO: Add any other composable APIs that are mutually exclusive. + registry = _get_registry(module) + if registry is None: + return True + return "replicate" not in registry + + +# TODO (awgu): We may be able to remove this function if we retired the +# `use_orig_params=False` code path since so far we only need the module for +# `FlatParameter` registration, which is not needed for `use_orig_params=True`. +def _get_fsdp_states_with_modules( + module: nn.Module, +) -> tuple[list[_FSDPState], list[nn.Module]]: + """ + Returns a tuple containing: + 1. A list of the ``_FSDPState`` instances in the module tree rooted at + ``module`` without any duplicates and following the ``module.modules()`` + traversal order (which is assumed to be depth-first). + 2. A corresponding list of the modules owning the states in the first list. + + For the wrapper code path, both returned lists are the same, each + containing all ``FullyShardedDataParallel`` instances. For the composable + code path, this returns a list of all composable state instances and a list + of the corresponding fully sharded modules. See [Note: Fully Sharded + Module]. + + NOTE: The traversal does not proceed into any module annotated by an + incompatible API (e.g. ``replicate``). + """ + fsdp_states: list[_FSDPState] = [] + fsdp_modules: list[nn.Module] = [] + # Track the visited FSDP states since multiple modules may share the same + # one and we want to return a de-duplicated list + visited_fsdp_states: set[_FSDPState] = set() + # Track the visited modules in case of shared modules, which implies the + # module graph is no longer a tree + visited_modules: set[nn.Module] = set() + + # Perform depth-first search from `module` to ensure that we do not + # traverse into an incompatible API's subtree (use DFS instead of BFS to + # match `.modules()` order) + deque: collections.deque[nn.Module] = collections.deque([module]) + while deque: + submodule = deque.popleft() + visited_modules.add(submodule) + if not _composable(submodule): + continue + for child_module in reversed(list(submodule.children())): + if child_module not in visited_modules: + deque.appendleft(child_module) + optional_state = _get_module_fsdp_state(submodule) + if optional_state is not None and optional_state not in visited_fsdp_states: + visited_fsdp_states.add(optional_state) + fsdp_states.append(optional_state) + fsdp_modules.append(submodule) + return fsdp_states, fsdp_modules + + +def _get_fsdp_states(module: nn.Module) -> list[_FSDPState]: + """See :func:`_get_fsdp_states_with_modules`.""" + fsdp_states, _ = _get_fsdp_states_with_modules(module) + return fsdp_states + + +def _get_fsdp_handles(module: nn.Module) -> list: + """ + Returns all ``FlatParamHandle`` s in the module tree rooted at ``module`` + following the rules in :func:`_get_fsdp_state`. + """ + handles = [ + fsdp_state._handle + for fsdp_state in _get_fsdp_states(module) + if fsdp_state._handle is not None + ] + return handles diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_unshard_param_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_unshard_param_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..f006dc22868fbeda3121834a63e08f51d9105ee9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_unshard_param_utils.py @@ -0,0 +1,340 @@ +# mypy: allow-untyped-defs +import contextlib +import warnings +from collections.abc import Generator +from typing import cast + +import torch +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.nn as nn +from torch.distributed.fsdp._common_utils import ( + _FSDPState, + _get_module_fsdp_state, + _has_fsdp_params, + _module_handle, + HandleTrainingState, + TrainingState, +) +from torch.distributed.fsdp._runtime_utils import ( + _lazy_init, + _reset_flat_param_grad_info_if_needed, + _reshard, + _reshard_grads, + _unshard, + _unshard_grads, +) +from torch.distributed.utils import _p_assert + +from ._flat_param import FlatParamHandle + + +FLAT_PARAM = "_flat_param" + + +@torch.no_grad() +def _writeback_to_local_shard( + handle: FlatParamHandle, + writeback_grad: bool, +): + """ + For the handle, writes back this rank's shard of the unsharded + flattened parameter to the sharded flattened parameter. If + ``writeback_grad=True``, then writes back to the sharded gradient as + well. + + Precondition: The handle's ``FlatParameter`` 's data points to the + padded unsharded flattened parameter. + """ + + def _get_shard(flat_param_or_grad: torch.Tensor) -> torch.Tensor: + if handle.uses_sharded_strategy: + # For sharded strategies, get the *unpadded* shard instead of + # the *padded* shard to persist user changes to the padding + # (though FSDP does not explicitly support this) + shard, _ = FlatParamHandle._get_unpadded_shard( + flat_param_or_grad, + handle.rank, + handle.world_size, + ) + return shard + # For `NO_SHARD`, the `flat_param` or its gradient may be modified, + # so we write it back directly + return flat_param_or_grad + + param_shard = _get_shard(handle.flat_param) + handle.flat_param._local_shard[: param_shard.numel()].copy_(param_shard) # type: ignore[attr-defined] + if writeback_grad: + existing_grad = handle.sharded_grad + if existing_grad is not None: + if handle.flat_param.grad is None: + raise AssertionError("Expected handle.flat_param.grad to not be None") + grad_shard = _get_shard(handle.flat_param.grad) + existing_grad[: grad_shard.numel()].copy_(grad_shard) + + +def _deregister_flat_param(state: _FSDPState, module: nn.Module) -> None: + """ + De-registers the flattened parameter from the wrapped module, hiding it + from ``nn.Module`` methods. + + We do not use ``del`` because we want ``FLAT_PARAM`` to always be an + attribute but dynamically change whether it is visible to ``nn.Module`` + methods. + """ + if _has_fsdp_params(state, module): + # TODO: figure out the case for the composable APIs. + cast(nn.Module, module.module)._parameters.pop(FLAT_PARAM, None) + + +def _register_flat_param(state: _FSDPState, module: nn.Module) -> None: + """ + Registers the flattened parameter to the wrapped module, making it + visible to ``nn.Module`` methods. + + We do not use :meth:`nn.Module.register_parameter` because we want + ``FLAT_PARAM`` to always be an attribute but dynamically change whether + it is visible to ``nn.Module`` methods. + """ + handle = _module_handle(state, module) + if _has_fsdp_params(state, module): + # TODO: figure out the case for the composable APIs. + cast(nn.Module, module.module)._parameters[FLAT_PARAM] = handle.flat_param + + +@contextlib.contextmanager +def _unflatten_as_params(state: _FSDPState, module: nn.Module) -> Generator: + """ + Assumes that the flattened parameter is unsharded. When in the context, + de-registers the flattened parameter and unflattens the original + parameters as ``nn.Parameter`` views into the flattened parameter. + After the context, re-registers the flattened parameter and restores + the original parameters as ``Tensor`` views into the flattened + parameter. + """ + handle = _module_handle(state, module) + if not handle: + yield + else: + _deregister_flat_param(state, module) + try: + with handle.unflatten_as_params(): + yield + finally: + if not handle._use_orig_params: + _register_flat_param(state, module) + + +def _validate_unshard_params_args( + state: _FSDPState, + writeback: bool, + rank0_only: bool, + offload_to_cpu: bool, + with_grads: bool, +) -> None: + if with_grads and (offload_to_cpu or not state._use_orig_params): + raise NotImplementedError( + f"with_grads={with_grads}, " + f"use_orig_params={state._use_orig_params}, " + f"offload_to_cpu={offload_to_cpu} " + f"is not supported yet" + ) + if offload_to_cpu and state._handle and (not state._handle.uses_sharded_strategy): + raise NotImplementedError( + "offload_to_cpu=True and NO_SHARD is not supported yet" + ) + if writeback and rank0_only: + # TODO: Rank 0 can broadcast the `FlatParameter` to allow all ranks to + # persist the changes. + raise NotImplementedError( + "writeback=True and rank0_only=True is not supported yet" + ) + if offload_to_cpu and not rank0_only: + warnings.warn( + "offload_to_cpu=True and rank0_only=False may result in the" + "unsharded parameters being redundantly copied to CPU memory for " + "GPUs sharing the same CPU memory, which risks CPU OOM. We " + "recommend using offload_to_cpu=True with rank0_only=True.", + stacklevel=2, + ) + + +@contextlib.contextmanager +def _unshard_fsdp_state_params( + module: nn.Module, + state: _FSDPState, + writeback: bool, + rank0_only: bool, + offload_to_cpu: bool, + with_grads: bool, +): + """ + This unshards the parameters for a single FSDP state ``state`` that + corresponds to ``module``. + """ + _validate_unshard_params_args( + state, writeback, rank0_only, offload_to_cpu, with_grads + ) + state._device_handle.synchronize() + # If handles are shared by other module(s), the handle may be already unsharded. + maybe_handle = _module_handle(state, module) + handle = None + if ( + maybe_handle + and maybe_handle._training_state != HandleTrainingState.SUMMON_FULL_PARAMS + ): + handle = maybe_handle + if not handle: + yield + return + + if handle._training_state != HandleTrainingState.IDLE: + raise AssertionError( + f"Expects the handle training to be IDLE but got {handle._training_state}" + ) + + handle._training_state = HandleTrainingState.SUMMON_FULL_PARAMS + + _reset_flat_param_grad_info_if_needed(handle) + free_unsharded_flat_param = handle.needs_unshard() + # No need to call `wait_stream()` since we unshard in the computation + # stream directly + computation_stream = state._device_handle.current_stream() + _unshard(state, handle, computation_stream, computation_stream) + if with_grads: + _unshard_grads(handle) + + if rank0_only and state.rank != 0: + # Free the unsharded flattened parameter early + _reshard(state, handle, free_unsharded_flat_param) + if with_grads: + _reshard_grads(handle) + try: + yield + finally: + handle._training_state = HandleTrainingState.IDLE + else: + # Unflatten the unsharded flattened parameters + with contextlib.ExitStack() as stack: + # Invariant: rank == 0 or !rank0_only + if offload_to_cpu and handle.uses_sharded_strategy: + stack.enter_context(handle.to_cpu()) + # NOTE: Since PyTorch enforces that a parameter and its + # gradients need to match metadata (e.g. device), we must + # move gradients to CPU *after* we move parameters. + # NOTE: This assumes 1 `FlatParameter` + if not state._use_orig_params: + stack.enter_context(_unflatten_as_params(state, module)) + try: + yield + finally: + stack.close() + if writeback: + _writeback_to_local_shard(handle, with_grads) + _reshard(state, handle, free_unsharded_flat_param) + if with_grads: + _reshard_grads(handle) + handle._training_state = HandleTrainingState.IDLE + + +@contextlib.contextmanager +def _unshard_params_for_summon( + module: nn.Module, + state: _FSDPState, + writeback: bool, + rank0_only: bool, + offload_to_cpu: bool, + with_grads: bool, +): + _validate_unshard_params_args( + state, writeback, rank0_only, offload_to_cpu, with_grads + ) + _lazy_init(state, module) + if state.training_state == TrainingState.FORWARD_BACKWARD: + raise AssertionError( + "Cannot manually unshard parameters during forward/backward" + ) + elif state.training_state == TrainingState.SUMMON_FULL_PARAMS: + raise AssertionError( + "Cannot manually unshard parameters when already unsharding parameters" + ) + with _unshard_fsdp_state_params( + module=module, + state=state, + writeback=writeback, + rank0_only=rank0_only, + offload_to_cpu=offload_to_cpu, + with_grads=with_grads, + ): + try: + state.training_state = TrainingState.SUMMON_FULL_PARAMS + yield + finally: + state.training_state = TrainingState.IDLE + + +@contextlib.contextmanager +def _unshard_params( + module: nn.Module, + recurse: bool, + writeback: bool, + rank0_only: bool, + offload_to_cpu: bool, + with_grads: bool, +): + """ + This unshards FSDP-managed parameters for all modules with FSDP applied in + the module tree rooted at ``module``. + """ + if not recurse: + optional_state = _get_module_fsdp_state(module) + if optional_state is None: + with contextlib.nullcontext(): + yield + return + states_and_modules = ([optional_state], [module]) + else: + states_and_modules = traversal_utils._get_fsdp_states_with_modules(module) + with contextlib.ExitStack() as stack: + for state, module in zip(*states_and_modules): + stack.enter_context( + _unshard_params_for_summon( + module=module, + state=state, + writeback=writeback, + rank0_only=rank0_only, + offload_to_cpu=offload_to_cpu, + with_grads=with_grads, + ) + ) + yield + + +def _deregister_orig_params(state: _FSDPState, module: nn.Module) -> None: + """ + Deregisters the original parameters; registers the ``FlatParameter``. + """ + handle = _module_handle(state, module) + if not handle: + return + _p_assert( + handle._use_orig_params, + f"Inconsistent `_use_orig_params` -- FSDP: {state._use_orig_params} " + f"handle: {handle._use_orig_params}", + ) + handle._deregister_orig_params() + _register_flat_param(state, module) + + +def _register_orig_params(state: _FSDPState, module: nn.Module) -> None: + """ + Deregisters the ``FlatParameter``; registers the original parameters. + """ + handle = _module_handle(state, module) + if not handle: + return + _deregister_flat_param(state, module) + if handle.is_sharded(handle.flat_param): + handle._use_sharded_views() + handle._use_sharded_grad_views() + else: + handle._use_unsharded_views(as_params=True) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_wrap_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_wrap_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..c3be5bfcde1e83020ffcb31eddab10c5a5ed8f03 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/_wrap_utils.py @@ -0,0 +1,264 @@ +# mypy: allow-untyped-defs +import collections +import functools +import inspect +import warnings +from collections.abc import Callable +from functools import partial +from typing import Any + +import torch.nn as nn +from torch.distributed.fsdp._common_utils import ( + _get_module_fsdp_state, + _override_module_mixed_precision, +) +from torch.distributed.fsdp.wrap import ( + _construct_wrap_fn, + _or_policy, + _Policy, + _post_order_apply, + _recursive_wrap, + _run_mixed_precision_override_policy, + _wrap_module_cls_individually, +) + + +def _auto_wrap( + root_module: nn.Module, + policy: Callable | _Policy, + ignored_modules: set[nn.Module], + ignored_params: set[nn.Parameter], + root_kwargs: dict[str, Any], + fsdp_fn: Callable, # e.g. `FullyShardedDataParallel` or `fully_shard` +): + """ + Auto wraps modules in ``root_module`` 's tree according to ``policy`` + following a post-order traversal. + + Precondition: ``root_kwargs`` should contain all arguments except + ``module``. This function accepts the kwargs dict directly since it gets + forwarded into the post-order traversal function. + """ + mixed_precision = root_kwargs["mixed_precision"] + is_wrapper = inspect.isclass(fsdp_fn) + # TODO: We may relax this no-nested-wrapping constraint to support manual + # wrapping followed by auto wrapping. + _check_nested_wrapping(root_module) + + if isinstance(policy, _Policy): + root_kwargs["auto_wrap_policy" if is_wrapper else "policy"] = None + target_module_to_kwargs = policy._run_policy( + root_module, ignored_modules, root_kwargs + ) + if mixed_precision is not None: + target_module_to_kwargs = _run_mixed_precision_override_policy( + root_module, + mixed_precision._module_classes_to_ignore, + ignored_modules, + root_kwargs, + target_module_to_kwargs, + ) + overridden_module_classes = _override_module_mixed_precision( + root_module, mixed_precision._module_classes_to_ignore + ) + _warn_on_overridden_mixed_precision(overridden_module_classes) + use_orig_params = root_kwargs.get("use_orig_params", False) + _validate_frozen_params( + root_module, + set(target_module_to_kwargs.keys()), + ignored_params, + use_orig_params, + ) + wrap_fn = _construct_wrap_fn(root_module, target_module_to_kwargs, fsdp_fn) + _post_order_apply(root_module, wrap_fn) + return + + recursive_wrap_kwargs = { + "module": root_module, + "auto_wrap_policy": policy, + "wrapper_cls": fsdp_fn, + "ignored_modules": ignored_modules, + "ignored_params": ignored_params, + "only_wrap_children": True, + } + if mixed_precision is not None: + # Wrap modules of the ignored types separately and register forward + # hooks to cast to fp32 and back to the original dtype, respectively + overridden_module_classes = _override_module_mixed_precision( + root_module, mixed_precision._module_classes_to_ignore + ) + policy = functools.partial( + _or_policy, + policies=[ + policy, + partial( + _wrap_module_cls_individually, + module_classes=mixed_precision._module_classes_to_ignore, + ), + ], + ) + recursive_wrap_kwargs["auto_wrap_policy"] = policy + _warn_on_overridden_mixed_precision(overridden_module_classes) + _recursive_wrap(**recursive_wrap_kwargs, **root_kwargs) # type: ignore[arg-type] + + +def _check_nested_wrapping(root_module: nn.Module): + for module_name, module in root_module.named_modules(): + if _get_module_fsdp_state(module) is not None: + raise ValueError( + "FSDP auto wrapping requires modules to not already have " + f"FSDP applied but found {module_name} in\n{root_module}" + ) + + +def _warn_on_overridden_mixed_precision( + overridden_module_classes: set[type[nn.Module]], +): + if len(overridden_module_classes) == 0: + return + warnings.warn( + "Both mixed precision and an auto_wrap_policy were specified to FSDP, " + f"where the wrapped module has submodules of type:\n{overridden_module_classes}\n" + "These modules will be wrapped as separate FSDP instances with mixed " + "precision disabled.", + stacklevel=2, + ) + + +def _validate_frozen_params( + root_module: nn.Module, + modules_to_wrap: set[nn.Module], + ignored_params: set[nn.Parameter], + use_orig_params: bool, +): + """ + This checks that, given ``modules_to_wrap``, each module would manage + parameters that are uniformly frozen or non-frozen. This uniformity + requirement is strict for ``use_orig_params=False`` (hard error) and highly + recommended for ``use_orig_params=True`` (user warning). + """ + post_order_named_modules = _get_post_order_named_modules(root_module) + visited_modules: set[nn.Module] = set() + for module_name, module in post_order_named_modules: + if module in modules_to_wrap: + param_to_fqn = _get_managed_param_to_fqn( + module, ignored_params, visited_modules, module_name + ) + frozen_param_fqns: list[str] = [] + frozen_param_numel = 0 + nonfrozen_param_fqns: list[str] = [] + nonfrozen_param_numel = 0 + for param, fqn in param_to_fqn.items(): + if param.requires_grad: + nonfrozen_param_fqns.append(fqn) + nonfrozen_param_numel += param.numel() + else: + frozen_param_fqns.append(fqn) + frozen_param_numel += param.numel() + if len(frozen_param_fqns) > 0 and len(nonfrozen_param_fqns) > 0: + msg = f"{module_name} has both parameters with requires_grad=True and False." + if use_orig_params: + total_param_numel = frozen_param_numel + nonfrozen_param_numel + msg += ( + " We do not recommend wrapping such modules since " + "the gradient memory usage will be higher than expected " + f"({total_param_numel} numel instead of {nonfrozen_param_numel} numel " + "before sharding via reduce-scatter). " + ) + else: + msg += " FSDP does not support wrapping such modules when use_orig_params=False. " + msg += "If possible, wrap the frozen parameters with FSDP separately.\n" + msg += ( + f"The following parameters have requires_grad=True:\n{nonfrozen_param_fqns}\n" + f"The following parameters have requires_grad=False:\n{frozen_param_fqns}" + ) + if use_orig_params: + warnings.warn(msg, stacklevel=2) + else: + raise ValueError(msg) + + +def _get_post_order_named_modules( + root_module: nn.Module, +) -> list[tuple[str, nn.Module]]: + """ + This returns the named modules following a post-order traversal, which is a + valid reverse topological sort. We achieve this using the reverse of a + stack-based DFS order instead of reversing ``root_module.named_modules()`` + since the former gives the modules in registration order at each level in + the module tree (as opposed to the reverse), which allows us to error/warn + on the first registered module that violates the condition. + + For example, consider the following module structure: + M( + S1(), + S2( + SS1(), + SS2(), + ), + S3(), + ) + The reverse DFS order is [S1, SS1, SS2, S2, S3, M], while the reverse + ``named_modules()`` order is [S3, SS2, SS1, S2, S1, M]. + """ + visited_modules = {root_module} + stack = [("", root_module)] + # Append and reverse at the end for linear-time algorithm + reverse_post_order_named_modules: list[tuple[str, nn.Module]] = [] + while stack: + module_name, module = stack.pop() + reverse_post_order_named_modules.append((module_name, module)) + for child_module_name, child_module in module.named_children(): + if child_module is None: # only for overrides of `named_children()` + continue + if child_module not in visited_modules: + visited_modules.add(child_module) + if module_name != "": + child_module_name = module_name + "." + child_module_name + stack.append((child_module_name, child_module)) + post_order_named_modules = list(reversed(reverse_post_order_named_modules)) + return post_order_named_modules + + +def _get_managed_param_to_fqn( + module_to_wrap: nn.Module, + ignored_params: set[nn.Parameter], + visited_modules: set[nn.Module], + root_prefix: str, +) -> dict[nn.Parameter, str]: + """ + This returns a dict that maps managed parameter to its FQN for the given + ``module_to_wrap``. The dict's keys are exactly the parameters that would + be managed by the module, where this is achieved by calling this function + on the modules to wrap in reverse topological order, destructively updating + ``visited_modules``, and not traversing into those modules. The FQNs are + prefixed from the root (via ``root_prefix``) to be more informative. + + NOTE: This function is meant to be called pre-wrapping and iteratively in + reverse topological order to cover the full module tree. This differs from + the ``_get_param_to_fqn()`` function meant to be called post-wrapping and + on the full module tree in one shot. Given those differences, we do not try + to unify the two. + """ + param_to_fqn: dict[nn.Parameter, str] = {} + # Run BFS (or any tree traversal works) + queue = collections.deque([(module_to_wrap, root_prefix)]) + visited_modules.add(module_to_wrap) + while queue: + module, prefix = queue.popleft() + for param_name, param in module.named_parameters(recurse=False): + if param not in ignored_params: + fqn = param_name if prefix == "" else prefix + "." + param_name + param_to_fqn[param] = fqn + for child_module_name, child_module in module.named_children(): + if child_module is None: # only for overrides of `named_children()` + continue + if child_module not in visited_modules: + visited_modules.add(child_module) + child_prefix = ( + child_module_name + if prefix == "" + else prefix + "." + child_module_name + ) + queue.append((child_module, child_prefix)) + return param_to_fqn diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/api.py new file mode 100644 index 0000000000000000000000000000000000000000..f17f3dc2119911fd56512b8b524d0cfee6bd8a84 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/api.py @@ -0,0 +1,416 @@ +""" +This file includes public APIs for FSDP such as the classes used for the +constructor arguments. +""" + +from collections.abc import Sequence +from dataclasses import dataclass +from enum import auto, Enum + +import torch +from torch.nn.modules.batchnorm import _BatchNorm + + +__all__ = [ + "ShardingStrategy", + "BackwardPrefetch", + "MixedPrecision", + "CPUOffload", + "StateDictType", + "StateDictConfig", + "FullStateDictConfig", + "LocalStateDictConfig", + "ShardedStateDictConfig", + "OptimStateDictConfig", + "FullOptimStateDictConfig", + "LocalOptimStateDictConfig", + "ShardedOptimStateDictConfig", + "StateDictSettings", +] + + +class ShardingStrategy(Enum): + """ + This specifies the sharding strategy to be used for distributed training by + :class:`FullyShardedDataParallel`. + + - ``FULL_SHARD``: Parameters, gradients, and optimizer states are sharded. + For the parameters, this strategy unshards (via all-gather) before the + forward, reshards after the forward, unshards before the backward + computation, and reshards after the backward computation. For gradients, + it synchronizes and shards them (via reduce-scatter) after the backward + computation. The sharded optimizer states are updated locally per rank. + - ``SHARD_GRAD_OP``: Gradients and optimizer states are sharded during + computation, and additionally, parameters are sharded outside + computation. For the parameters, this strategy unshards before the + forward, does not reshard them after the forward, and only reshards them + after the backward computation. The sharded optimizer states are updated + locally per rank. Inside ``no_sync()``, the parameters are not resharded + after the backward computation. + - ``NO_SHARD``: Parameters, gradients, and optimizer states are not sharded + but instead replicated across ranks similar to PyTorch's + :class:`DistributedDataParallel` API. For gradients, this strategy + synchronizes them (via all-reduce) after the backward computation. The + unsharded optimizer states are updated locally per rank. + - ``HYBRID_SHARD``: Apply ``FULL_SHARD`` within a node, and replicate parameters across + nodes. This results in reduced communication volume as expensive all-gathers and + reduce-scatters are only done within a node, which can be more performant for medium + -sized models. + - ``_HYBRID_SHARD_ZERO2``: Apply ``SHARD_GRAD_OP`` within a node, and replicate parameters across + nodes. This is like ``HYBRID_SHARD``, except this may provide even higher throughput + since the unsharded parameters are not freed after the forward pass, saving the + all-gathers in the pre-backward. + """ + + FULL_SHARD = auto() + SHARD_GRAD_OP = auto() + NO_SHARD = auto() + HYBRID_SHARD = auto() + _HYBRID_SHARD_ZERO2 = auto() + + +class BackwardPrefetch(Enum): + """ + This configures explicit backward prefetching, which improves throughput by + enabling communication and computation overlap in the backward pass at the + cost of slightly increased memory usage. + + - ``BACKWARD_PRE``: This enables the most overlap but increases memory + usage the most. This prefetches the next set of parameters *before* the + current set of parameters' gradient computation. This overlaps the *next + all-gather* and the *current gradient computation*, and at the peak, it + holds the current set of parameters, next set of parameters, and current + set of gradients in memory. + - ``BACKWARD_POST``: This enables less overlap but requires less memory + usage. This prefetches the next set of parameters *after* the current + set of parameters' gradient computation. This overlaps the *current + reduce-scatter* and the *next gradient computation*, and it frees the + current set of parameters before allocating memory for the next set of + parameters, only holding the next set of parameters and current set of + gradients in memory at the peak. + - FSDP's ``backward_prefetch`` argument accepts ``None``, which disables + the backward prefetching altogether. This has no overlap and does not + increase memory usage. In general, we do not recommend this setting since + it may degrade throughput significantly. + + For more technical context: For a single process group using NCCL backend, + any collectives, even if issued from different streams, contend for the + same per-device NCCL stream, which implies that the relative order in which + the collectives are issued matters for overlapping. The two backward + prefetching values correspond to different issue orders. + """ + + # NOTE: For both modes, the ordering that defines "current" and "next" is + # not always exact in the current implementation. A mistargeted prefetch + # simply means that the parameter memory is allocated earlier than needed, + # possibly increasing peak memory usage, but does not affect correctness. + BACKWARD_PRE = auto() + BACKWARD_POST = auto() + + +@dataclass +class MixedPrecision: + """ + This configures FSDP-native mixed precision training. + + Attributes: + param_dtype (Optional[torch.dtype]): This specifies the dtype for model + parameters during forward and backward and thus the dtype for + forward and backward computation. Outside forward and backward, the + *sharded* parameters are kept in full precision (e.g. for the + optimizer step), and for model checkpointing, the parameters are + always saved in full precision. (Default: ``None``) + reduce_dtype (Optional[torch.dtype]): This specifies the dtype for + gradient reduction (i.e. reduce-scatter or all-reduce). If this is + ``None`` but ``param_dtype`` is not ``None``, then this takes on + the ``param_dtype`` value, still running gradient reduction in low + precision. This is permitted to differ from ``param_dtype``, e.g. + to force gradient reduction to run in full precision. (Default: + ``None``) + buffer_dtype (Optional[torch.dtype]): This specifies the dtype for + buffers. FSDP does not shard buffers. Rather, FSDP casts them to + ``buffer_dtype`` in the first forward pass and keeps them in that + dtype thereafter. For model checkpointing, the buffers are saved + in full precision except for ``LOCAL_STATE_DICT``. (Default: + ``None``) + keep_low_precision_grads (bool): If ``False``, then FSDP upcasts + gradients to full precision after the backward pass in preparation + for the optimizer step. If ``True``, then FSDP keeps the gradients + in the dtype used for gradient reduction, which can save memory if + using a custom optimizer that supports running in low precision. + (Default: ``False``) + cast_forward_inputs (bool): If ``True``, then this FSDP module casts + its forward args and kwargs to ``param_dtype``. This is to ensure + that parameter and input dtypes match for forward computation, as + required by many ops. This may need to be set to ``True`` when only + applying mixed precision to some but not all FSDP modules, in which + case a mixed-precision FSDP submodule needs to recast its inputs. + (Default: ``False``) + cast_root_forward_inputs (bool): If ``True``, then the root FSDP module + casts its forward args and kwargs to ``param_dtype``, overriding + the value of ``cast_forward_inputs``. For non-root FSDP modules, + this does not do anything. (Default: ``True``) + _module_classes_to_ignore: (Sequence[Type[nn.Module]]): This specifies + module classes to ignore for mixed precision when using an + ``auto_wrap_policy``: Modules of these classes will have FSDP + applied to them separately with mixed precision disabled (meaning + that the final FSDP construction would deviate from the specified + policy). If ``auto_wrap_policy`` is not specified, then this does + not do anything. This API is experimental and subject to change. + (Default: ``(_BatchNorm,)``) + + .. note:: This API is experimental and subject to change. + + .. note:: Only floating point tensors are cast to their specified dtypes. + + .. note:: In ``summon_full_params``, parameters are forced to full + precision, but buffers are not. + + .. note:: Layer norm and batch norm accumulate in ``float32`` even when + their inputs are in a low precision like ``float16`` or ``bfloat16``. + Disabling FSDP's mixed precision for those norm modules only means that + the affine parameters are kept in ``float32``. However, this incurs + separate all-gathers and reduce-scatters for those norm modules, which + may be inefficient, so if the workload permits, the user should prefer + to still apply mixed precision to those modules. + + .. note:: By default, if the user passes a model with any ``_BatchNorm`` + modules and specifies an ``auto_wrap_policy``, then the batch norm + modules will have FSDP applied to them separately with mixed precision + disabled. See the ``_module_classes_to_ignore`` argument. + + .. note:: ``MixedPrecision`` has ``cast_root_forward_inputs=True`` and + ``cast_forward_inputs=False`` by default. For the root FSDP instance, + its ``cast_root_forward_inputs`` takes precedence over its + ``cast_forward_inputs``. For non-root FSDP instances, their + ``cast_root_forward_inputs`` values are ignored. The default setting is + sufficient for the typical case where each FSDP instance has the same + ``MixedPrecision`` configuration and only needs to cast inputs to the + ``param_dtype`` at the beginning of the model's forward pass. + + .. note:: For nested FSDP instances with different ``MixedPrecision`` + configurations, we recommend setting individual ``cast_forward_inputs`` + values to configure casting inputs or not before each instance's + forward. In such a case, since the casts happen before each FSDP + instance's forward, a parent FSDP instance should have its non-FSDP + submodules run before its FSDP submodules to avoid the activation dtype + being changed due to a different ``MixedPrecision`` configuration. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> model = nn.Sequential(nn.Linear(3, 3), nn.Linear(3, 3)) + >>> model[1] = FSDP( + >>> model[1], + >>> mixed_precision=MixedPrecision(param_dtype=torch.float16, cast_forward_inputs=True), + >>> ) + >>> model = FSDP( + >>> model, + >>> mixed_precision=MixedPrecision(param_dtype=torch.bfloat16, cast_forward_inputs=True), + >>> ) + + The above shows a working example. On the other hand, if ``model[1]`` + were replaced with ``model[0]``, meaning that the submodule using + different ``MixedPrecision`` ran its forward first, then ``model[1]`` + would incorrectly see ``float16`` activations instead of ``bfloat16`` + ones. + + """ + + param_dtype: torch.dtype | None = None + reduce_dtype: torch.dtype | None = None + buffer_dtype: torch.dtype | None = None + keep_low_precision_grads: bool = False + cast_forward_inputs: bool = False + cast_root_forward_inputs: bool = True + _module_classes_to_ignore: Sequence[type[torch.nn.Module]] = (_BatchNorm,) + + +@dataclass +class CPUOffload: + """ + This configures CPU offloading. + + Attributes: + offload_params (bool): This specifies whether to offload parameters to + CPU when not involved in computation. If ``True``, then this + offloads gradients to CPU as well, meaning that the optimizer step + runs on CPU. + """ + + offload_params: bool = False + + +class StateDictType(Enum): + """ + This enum indicates that which type of ``state_dict`` the FSDP module is + currently processing (returning or loading). + The default value is FULL_STATE_DICT to comply the PyTorch convention. + + .. note:: + FSDP currently supports three types of ``state_dict``: + 1. ``state_dict/load_state_dict`: this pair of APIs return and load + the non-sharded, unflattened parameters. The semantics is the + same as using DDP. + 2. ``_local_state_dict/_load_local_state_dict``: this pair of APIs return + and load local sharded, flattened parameters. The values returned + by ``_local_state_dict`` can be directly used by FSDP and is only + meaningful to FSDP (because parameters are flattened). Note that + these APIs are meant for use via the :func:`state_dict_type` + context manager as follows: + >>> # xdoctest: +SKIP("undefined variables") + >>> with fsdp.state_dict_type(StateDictType.LOCAL_STATE_DICT): + ... state = fsdp.state_dict() # loads local state dict + 3. ``_sharded_state_dict/_load_sharded_state_dict``: this pair of APIs + return and load sharded, unflattened parameters. The ``state_dict`` + return by ``sharded_state_dict`` can be used by all other parallel + schemes (resharding may be required). + """ + + FULL_STATE_DICT = auto() + LOCAL_STATE_DICT = auto() + SHARDED_STATE_DICT = auto() + + +@dataclass +class StateDictConfig: + """ + ``StateDictConfig`` is the base class for all ``state_dict`` configuration + classes. Users should instantiate a child class (e.g. + ``FullStateDictConfig``) in order to configure settings for the + corresponding ``state_dict`` type supported by FSDP. + + Attributes: + offload_to_cpu (bool): If ``True``, then FSDP offloads the state dict + values to CPU, and if ``False``, then FSDP keeps them on GPU. + (Default: ``False``) + """ + + offload_to_cpu: bool = False + + +@dataclass +class FullStateDictConfig(StateDictConfig): + """ + ``FullStateDictConfig`` is a config class meant to be used with + ``StateDictType.FULL_STATE_DICT``. We recommend enabling both + ``offload_to_cpu=True`` and ``rank0_only=True`` when saving full state + dicts to save GPU memory and CPU memory, respectively. This config class + is meant to be used via the :func:`state_dict_type` context manager as + follows: + + >>> # xdoctest: +SKIP("undefined variables") + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> fsdp = FSDP(model, auto_wrap_policy=...) + >>> cfg = FullStateDictConfig(offload_to_cpu=True, rank0_only=True) + >>> with FSDP.state_dict_type(fsdp, StateDictType.FULL_STATE_DICT, cfg): + >>> state = fsdp.state_dict() + >>> # `state` will be empty on non rank 0 and contain CPU tensors on rank 0. + >>> # To reload checkpoint for inference, finetuning, transfer learning, etc: + >>> model = model_fn() # Initialize model in preparation for wrapping with FSDP + >>> if dist.get_rank() == 0: + >>> # Load checkpoint only on rank 0 to avoid memory redundancy + >>> state_dict = torch.load("my_checkpoint.pt") + >>> model.load_state_dict(state_dict) + >>> # All ranks initialize FSDP module as usual. `sync_module_states` argument + >>> # communicates loaded checkpoint states from rank 0 to rest of the world. + >>> fsdp = FSDP( + ... model, + ... device_id=torch.cuda.current_device(), + ... auto_wrap_policy=..., + ... sync_module_states=True, + ... ) + >>> # After this point, all ranks have FSDP model with loaded checkpoint. + + Attributes: + rank0_only (bool): If ``True``, then only rank 0 saves the full state + dict, and nonzero ranks save an empty dict. If ``False``, then all + ranks save the full state dict. (Default: ``False``) + """ + + rank0_only: bool = False + + +@dataclass +class LocalStateDictConfig(StateDictConfig): + pass + + +@dataclass +class ShardedStateDictConfig(StateDictConfig): + """ + ``ShardedStateDictConfig`` is a config class meant to be used with + ``StateDictType.SHARDED_STATE_DICT``. + + Attributes: + _use_dtensor (bool): If ``True``, then FSDP saves the state dict values + as ``DTensor``, and if ``False``, then FSDP saves them as + ``ShardedTensor``. (Default: ``False``) + + .. warning:: ``_use_dtensor`` is a private field of :class:`ShardedStateDictConfig` + and it is used by FSDP to determine the type of state dict values. Users should not + manually modify ``_use_dtensor``. + """ + + _use_dtensor: bool = False + + +@dataclass +class OptimStateDictConfig: + """ + ``OptimStateDictConfig`` is the base class for all ``optim_state_dict`` + configuration classes. Users should instantiate a child class (e.g. + ``FullOptimStateDictConfig``) in order to configure settings for the + corresponding ``optim_state_dict`` type supported by FSDP. + + Attributes: + offload_to_cpu (bool): If ``True``, then FSDP offloads the state dict's + tensor values to CPU, and if ``False``, then FSDP keeps them on the + original device (which is GPU unless parameter CPU offloading is + enabled). (Default: ``True``) + """ + + offload_to_cpu: bool = True + + +@dataclass +class FullOptimStateDictConfig(OptimStateDictConfig): + """ + Attributes: + rank0_only (bool): If ``True``, then only rank 0 saves the full state + dict, and nonzero ranks save an empty dict. If ``False``, then all + ranks save the full state dict. (Default: ``False``) + """ + + rank0_only: bool = False + + +@dataclass +class LocalOptimStateDictConfig(OptimStateDictConfig): + offload_to_cpu: bool = False + + +@dataclass +class ShardedOptimStateDictConfig(OptimStateDictConfig): + """ + ``ShardedOptimStateDictConfig`` is a config class meant to be used with + ``StateDictType.SHARDED_STATE_DICT``. + + Attributes: + _use_dtensor (bool): If ``True``, then FSDP saves the state dict values + as ``DTensor``, and if ``False``, then FSDP saves them as + ``ShardedTensor``. (Default: ``False``) + + .. warning:: ``_use_dtensor`` is a private field of :class:`ShardedOptimStateDictConfig` + and it is used by FSDP to determine the type of state dict values. Users should not + manually modify ``_use_dtensor``. + """ + + _use_dtensor: bool = False + + +@dataclass +class StateDictSettings: + state_dict_type: StateDictType + state_dict_config: StateDictConfig + optim_state_dict_config: OptimStateDictConfig diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/fully_sharded_data_parallel.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/fully_sharded_data_parallel.py new file mode 100644 index 0000000000000000000000000000000000000000..21f1699200827c3303c2f2f00c1f187ee2bef8b3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/fully_sharded_data_parallel.py @@ -0,0 +1,2167 @@ +# mypy: ignore-errors + +import contextlib +import copy +import functools +import math +import traceback +import warnings +from collections.abc import Callable, Generator, Iterable, Iterator +from contextlib import contextmanager +from enum import auto, Enum +from typing import Any + +import torch +import torch.distributed as dist +import torch.distributed.fsdp._traversal_utils as traversal_utils +import torch.nn as nn +from torch.distributed.algorithms._checkpoint.checkpoint_wrapper import ( + _CHECKPOINT_WRAPPED_MODULE, + ActivationWrapper, +) +from torch.distributed.algorithms._comm_hooks import LOW_PRECISION_HOOKS +from torch.distributed.fsdp._common_utils import ( + _FSDPState, + _get_param_to_fqns, + FSDP_PREFIX, + FSDP_WRAPPED_MODULE, + HandleTrainingState, + TrainingState, +) +from torch.distributed.fsdp._dynamo_utils import _annotate_modules_for_dynamo +from torch.distributed.fsdp._init_utils import ( + _check_orig_params_flattened, + _init_buffer_state, + _init_core_state, + _init_device_handle, + _init_extension, + _init_ignored_module_states, + _init_param_handle_from_module, + _init_prefetching_state, + _init_process_group_state, + _init_runtime_state, + _init_state_dict_state, + HYBRID_SHARDING_STRATEGIES, + ProcessGroupType, +) +from torch.distributed.fsdp._runtime_utils import ( + _get_fsdp_root_states, + _is_fsdp_root, + _lazy_init, + _post_forward, + _post_forward_reshard, + _pre_forward, + _pre_forward_unshard, + _root_pre_forward, + _unshard, + _wait_for_computation_stream, +) +from torch.distributed.fsdp._wrap_utils import _auto_wrap +from torch.distributed.fsdp.api import ( + BackwardPrefetch, + CPUOffload, + FullOptimStateDictConfig, + FullStateDictConfig, + LocalOptimStateDictConfig, + LocalStateDictConfig, + MixedPrecision, + OptimStateDictConfig, + ShardedOptimStateDictConfig, + ShardedStateDictConfig, + ShardingStrategy, + StateDictConfig, + StateDictSettings, + StateDictType, +) +from torch.distributed.tensor import DeviceMesh +from torch.distributed.utils import _p_assert +from torch.utils._typing_utils import copy_method_params, copy_method_sig + +from ._flat_param import FlatParameter, FlatParamHandle +from ._optim_utils import ( + _flatten_optim_state_dict, + _get_param_id_to_param_from_optim_input, + _get_param_key_to_param, + _get_param_to_param_id_from_optim_input, + _get_param_to_param_key, + _optim_state_dict, + _rekey_sharded_optim_state_dict, + _set_optim_use_dtensor, +) +from ._state_dict_utils import _register_all_state_dict_hooks +from ._unshard_param_utils import ( + _deregister_orig_params, + _register_flat_param, + _register_orig_params, + _unshard_params, + _unshard_params_for_summon, +) +from .wrap import CustomPolicy, ModuleWrapPolicy + + +__all__ = [ + "FullyShardedDataParallel", + "OptimStateKeyType", +] + + +FLAT_PARAM = "_flat_param" + + +class OptimStateKeyType(Enum): + """Represents the type of key in an optimizer state-dict.""" + + PARAM_NAME = auto() + PARAM_ID = auto() + + +class FullyShardedDataParallel(nn.Module, _FSDPState): + """A wrapper for sharding module parameters across data parallel workers. + + This is inspired by `Xu et al. `_ as + well as the ZeRO Stage 3 from `DeepSpeed `_. + FullyShardedDataParallel is commonly shortened to FSDP. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> import torch + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> torch.cuda.set_device(device_id) + >>> sharded_module = FSDP(my_module) + >>> optim = torch.optim.Adam(sharded_module.parameters(), lr=0.0001) + >>> x = sharded_module(x, y=3, z=torch.Tensor([1])) + >>> loss = x.sum() + >>> loss.backward() + >>> optim.step() + + Using FSDP involves wrapping your module and then initializing your + optimizer after. This is required since FSDP changes the parameter + variables. + + When setting up FSDP, you need to consider the destination CUDA + device. If the device has an ID (``dev_id``), you have three options: + + * Place the module on that device + * Set the device using ``torch.cuda.set_device(dev_id)`` + * Pass ``dev_id`` into the ``device_id`` constructor argument. + + This ensures that the FSDP instance's compute device is the + destination device. For option 1 and 3, the FSDP initialization + always occurs on GPU. For option 2, the FSDP initialization + happens on module's current device, which may be a CPU. + + If you're using the ``sync_module_states=True`` flag, you need to + ensure that the module is on a GPU or use the ``device_id`` + argument to specify a CUDA device that FSDP will move the module + to in the FSDP constructor. This is necessary because + ``sync_module_states=True`` requires GPU communication. + + FSDP also takes care of moving input tensors to the forward method + to the GPU compute device, so you don't need to manually move them + from CPU. + + For ``use_orig_params=True``, + ``ShardingStrategy.SHARD_GRAD_OP`` exposes the unsharded + parameters, not the sharded parameters after forward, unlike + ``ShardingStrategy.FULL_SHARD``. If you want + to inspect the gradients, you can use the ``summon_full_params`` + method with ``with_grads=True``. + + With ``limit_all_gathers=True``, you may see a gap in the FSDP + pre-forward where the CPU thread is not issuing any kernels. This is + intentional and shows the rate limiter in effect. Synchronizing the CPU + thread in that way prevents over-allocating memory for subsequent + all-gathers, and it should not actually delay GPU kernel execution. + + FSDP replaces managed modules' parameters with ``torch.Tensor`` + views during forward and backward computation for autograd-related + reasons. If your module's forward relies on saved references to + the parameters instead of reacquiring the references each + iteration, then it will not see FSDP's newly created views, + and autograd will not work correctly. + + Finally, when using ``sharding_strategy=ShardingStrategy.HYBRID_SHARD`` + with the sharding process group being intra-node and the + replication process group being inter-node, setting + ``NCCL_CROSS_NIC=1`` can help improve the all-reduce times over + the replication process group for some cluster setups. + + **Limitations** + + There are several limitations to be aware of when using FSDP: + + * FSDP currently does not support gradient accumulation outside + ``no_sync()`` when using CPU offloading. This is because FSDP + uses the newly-reduced gradient instead of accumulating with any + existing gradient, which can lead to incorrect results. + + * FSDP does not support running the forward pass of a submodule + that is contained in an FSDP instance. This is because the + submodule's parameters will be sharded, but the submodule itself + is not an FSDP instance, so its forward pass will not all-gather + the full parameters appropriately. + + * FSDP does not work with double backwards due to the way it + registers backward hooks. + + * FSDP has some constraints when freezing parameters. + For ``use_orig_params=False``, each FSDP instance must manage + parameters that are all frozen or all non-frozen. For + ``use_orig_params=True``, FSDP supports mixing frozen and + non-frozen parameters, but it's recommended to avoid doing so to + prevent higher than expected gradient memory usage. + + * As of PyTorch 1.12, FSDP offers limited support for shared + parameters. If enhanced shared parameter support is needed for + your use case, please post in + `this issue `__. + + * You should avoid modifying the parameters between forward and + backward without using the ``summon_full_params`` context, as + the modifications may not persist. + + Args: + module (nn.Module): + This is the module to be wrapped with FSDP. + process_group (Optional[Union[ProcessGroup, Tuple[ProcessGroup, ProcessGroup]]]): + This is the process group over which the model is sharded and thus + the one used for FSDP's all-gather and reduce-scatter collective + communications. If ``None``, then FSDP uses the default process + group. For hybrid sharding strategies such as + ``ShardingStrategy.HYBRID_SHARD``, users can pass in a tuple of + process groups, representing the groups over which to shard and + replicate, respectively. If ``None``, then FSDP constructs process + groups for the user to shard intra-node and replicate inter-node. + (Default: ``None``) + sharding_strategy (Optional[ShardingStrategy]): + This configures the sharding strategy, which may trade off memory + saving and communication overhead. See :class:`ShardingStrategy` + for details. (Default: ``FULL_SHARD``) + cpu_offload (Optional[CPUOffload]): + This configures CPU offloading. If this is set to ``None``, then + no CPU offloading happens. See :class:`CPUOffload` for details. + (Default: ``None``) + auto_wrap_policy (Optional[Union[Callable[[nn.Module, bool, int], bool], ModuleWrapPolicy, CustomPolicy]]): + This specifies a policy to apply FSDP to submodules of ``module``, + which is needed for communication and computation overlap and thus + affects performance. If ``None``, then FSDP only applies to + ``module``, and users should manually apply FSDP to parent modules + themselves (proceeding bottom-up). For convenience, this accepts + ``ModuleWrapPolicy`` directly, which allows users to specify the + module classes to wrap (e.g. the transformer block). Otherwise, + this should be a callable that takes in three arguments + ``module: nn.Module``, ``recurse: bool``, and + ``nonwrapped_numel: int`` and should return a ``bool`` specifying + whether the passed-in ``module`` should have FSDP applied if + ``recurse=False`` or if the traversal should continue into the + module's subtree if ``recurse=True``. Users may add additional + arguments to the callable. The ``size_based_auto_wrap_policy`` in + ``torch.distributed.fsdp.wrap.py`` gives an example callable that + applies FSDP to a module if the parameters in its subtree exceed + 100M numel. We recommend printing the model after applying FSDP + and adjusting as needed. + + Example:: + + >>> def custom_auto_wrap_policy( + >>> module: nn.Module, + >>> recurse: bool, + >>> nonwrapped_numel: int, + >>> # Additional custom arguments + >>> min_num_params: int = int(1e8), + >>> ) -> bool: + >>> return nonwrapped_numel >= min_num_params + >>> # Configure a custom `min_num_params` + >>> my_auto_wrap_policy = functools.partial(custom_auto_wrap_policy, min_num_params=int(1e5)) + + backward_prefetch (Optional[BackwardPrefetch]): + This configures explicit backward prefetching of all-gathers. If + ``None``, then FSDP does not backward prefetch, and there is no + communication and computation overlap in the backward pass. See + :class:`BackwardPrefetch` for details. (Default: ``BACKWARD_PRE``) + mixed_precision (Optional[MixedPrecision]): + This configures native mixed precision for FSDP. If this is set to + ``None``, then no mixed precision is used. Otherwise, parameter, + buffer, and gradient reduction dtypes can be set. See + :class:`MixedPrecision` for details. (Default: ``None``) + ignored_modules (Optional[Iterable[torch.nn.Module]]): Modules whose + own parameters and child modules' parameters and buffers are + ignored by this instance. None of the modules directly in + ``ignored_modules`` should be :class:`FullyShardedDataParallel` + instances, and any child modules that are already-constructed + :class:`FullyShardedDataParallel` instances will not be ignored if + they are nested under this instance. This argument may be used to + avoid sharding specific parameters at module granularity when using an + ``auto_wrap_policy`` or if parameters' sharding is not managed by + FSDP. (Default: ``None``) + param_init_fn (Optional[Callable[[nn.Module], None]]): + A ``Callable[torch.nn.Module] -> None`` that + specifies how modules that are currently on the meta device should + be initialized onto an actual device. As of v1.12, FSDP detects + modules with parameters or buffers on meta device via ``is_meta`` + and either applies ``param_init_fn`` if specified or calls + ``nn.Module.reset_parameters()`` otherwise. For both cases, the + implementation should *only* initialize the parameters/buffers of + the module, not those of its submodules. This is to avoid + re-initialization. In addition, FSDP also supports deferred + initialization via torchdistX's (https://github.com/pytorch/torchdistX) + ``deferred_init()`` API, where the deferred modules are initialized + by calling ``param_init_fn`` if specified or torchdistX's default + ``materialize_module()`` otherwise. If ``param_init_fn`` is + specified, then it is applied to all meta-device modules, meaning + that it should probably case on the module type. FSDP calls the + initialization function before parameter flattening and sharding. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> module = MyModule(device="meta") + >>> def my_init_fn(module: nn.Module): + >>> # E.g. initialize depending on the module type + >>> ... + >>> fsdp_model = FSDP(module, param_init_fn=my_init_fn, auto_wrap_policy=size_based_auto_wrap_policy) + >>> print(next(fsdp_model.parameters()).device) # current CUDA device + >>> # With torchdistX + >>> module = deferred_init.deferred_init(MyModule, device="cuda") + >>> # Will initialize via deferred_init.materialize_module(). + >>> fsdp_model = FSDP(module, auto_wrap_policy=size_based_auto_wrap_policy) + + device_id (Optional[Union[int, torch.device]]): An ``int`` or + ``torch.device`` giving the CUDA device on which FSDP + initialization takes place, including the module initialization + if needed and the parameter sharding. This should be specified to + improve initialization speed if ``module`` is on CPU. If the + default CUDA device was set (e.g. via ``torch.cuda.set_device``), + then the user may pass ``torch.cuda.current_device`` to this. + (Default: ``None``) + sync_module_states (bool): If ``True``, then each FSDP module will + broadcast module parameters and buffers from rank 0 to ensure that + they are replicated across ranks (adding communication overhead to + this constructor). This can help load ``state_dict`` checkpoints + via ``load_state_dict`` in a memory efficient way. See + :class:`FullStateDictConfig` for an example of this. (Default: + ``False``) + forward_prefetch (bool): If ``True``, then FSDP *explicitly* prefetches + the next forward-pass all-gather before the current forward + computation. This is only useful for CPU-bound workloads, in which + case issuing the next all-gather earlier may improve overlap. This + should only be used for static-graph models since the prefetching + follows the first iteration's execution order. (Default: ``False``) + limit_all_gathers (bool): If ``True``, then FSDP explicitly + synchronizes the CPU thread to ensure GPU memory usage from only + *two* consecutive FSDP instances (the current instance running + computation and the next instance whose all-gather is prefetched). + If ``False``, then FSDP allows the CPU thread to issue all-gathers + without any extra synchronization. (Default: ``True``) We often + refer to this feature as the "rate limiter". This flag should only + be set to ``False`` for specific CPU-bound workloads with low + memory pressure in which case the CPU thread can aggressively issue + all kernels without concern for the GPU memory usage. + use_orig_params (bool): Setting this to ``True`` has FSDP use + ``module`` 's original parameters. FSDP exposes those original + parameters to the user via :meth:`nn.Module.named_parameters` + instead of FSDP's internal :class:`FlatParameter` s. This means + that the optimizer step runs on the original parameters, enabling + per-original-parameter hyperparameters. FSDP preserves the original + parameter variables and manipulates their data between unsharded + and sharded forms, where they are always views into the underlying + unsharded or sharded :class:`FlatParameter`, respectively. With the + current algorithm, the sharded form is always 1D, losing the + original tensor structure. An original parameter may have all, + some, or none of its data present for a given rank. In the none + case, its data will be like a size-0 empty tensor. Users should not + author programs relying on what data is present for a given + original parameter in its sharded form. ``True`` is required to + use ``torch.compile()``. Setting this to ``False`` exposes FSDP's + internal :class:`FlatParameter` s to the user via + :meth:`nn.Module.named_parameters`. (Default: ``False``) + ignored_states (Optional[Iterable[torch.nn.Parameter]], Optional[Iterable[torch.nn.Module]]): + Ignored parameters or modules that will not be managed by this FSDP + instance, meaning that the parameters are not sharded and their + gradients are not reduced across ranks. This argument unifies with + the existing ``ignored_modules`` argument, and we may deprecate + ``ignored_modules`` soon. For backward compatibility, we keep both + ``ignored_states`` and `ignored_modules``, but FSDP only allows one + of them to be specified as not ``None``. + device_mesh (Optional[DeviceMesh]): DeviceMesh can be used as an alternative to + process_group. When device_mesh is passed, FSDP will use the underlying process + groups for all-gather and reduce-scatter collective communications. Therefore, + these two args need to be mutually exclusive. For hybrid sharding strategies such as + ``ShardingStrategy.HYBRID_SHARD``, users can pass in a 2D DeviceMesh instead + of a tuple of process groups. For 2D FSDP + TP, users are required to pass in + device_mesh instead of process_group. For more DeviceMesh info, please visit: + https://pytorch.org/tutorials/recipes/distributed_device_mesh.html + """ + + def __init__( + self, + module: nn.Module, + process_group: ProcessGroupType = None, + sharding_strategy: ShardingStrategy | None = None, + cpu_offload: CPUOffload | None = None, + auto_wrap_policy: Callable | ModuleWrapPolicy | CustomPolicy | None = None, + backward_prefetch: BackwardPrefetch | None = BackwardPrefetch.BACKWARD_PRE, + mixed_precision: MixedPrecision | None = None, + ignored_modules: Iterable[torch.nn.Module] | None = None, + param_init_fn: Callable[[nn.Module], None] | None = None, + device_id: int | torch.device | None = None, + sync_module_states: bool = False, + forward_prefetch: bool = False, + limit_all_gathers: bool = True, + use_orig_params: bool = False, + ignored_states: Iterable[torch.nn.Parameter] + | Iterable[torch.nn.Module] + | None = None, + device_mesh: DeviceMesh | None = None, + ): + torch._C._log_api_usage_once("torch.distributed.fsdp") + super().__init__() + if isinstance(module, (nn.ModuleList, nn.ModuleDict)): + warnings.warn( + "FSDP will not all-gather parameters for containers that do " + f"not implement forward: {module}", + stacklevel=2, + ) + _init_ignored_module_states(self, module, ignored_modules, ignored_states) + _init_device_handle(self, module, self._ignored_params, device_id) + + # Add module annotations for Dynamo support (see function for details) + _annotate_modules_for_dynamo(module, self._ignored_modules, use_orig_params) + + # Initializes self.process_group, along with rank and world size. This will + # also set another attribute, _inter_node_pg, to control the process group + # over which sharding occurs, if sharding_strategy is {HYBRID_SHARD, _HYBRID_SHARD_ZERO2}. + # Note that this is done before auto_wrapping, so that child FSDP modules simply pick up + # the same process group state as the root FSDP module. + self._device_mesh = device_mesh + _init_process_group_state( + self, + process_group, + sharding_strategy, + auto_wrap_policy, + device_mesh, + ) + if auto_wrap_policy is not None: + root_kwargs = { + "process_group": process_group, + "sharding_strategy": sharding_strategy, + "cpu_offload": cpu_offload, + "backward_prefetch": backward_prefetch, + "mixed_precision": mixed_precision, + "param_init_fn": param_init_fn, + "device_id": device_id, + "sync_module_states": sync_module_states, + "forward_prefetch": forward_prefetch, + "limit_all_gathers": limit_all_gathers, + "use_orig_params": use_orig_params, + "ignored_states": self._ignored_params, + "device_mesh": device_mesh, + } + if sharding_strategy in HYBRID_SHARDING_STRATEGIES and device_mesh is None: + # Share root process groups with children to maintain + # the invariant that all FSDP modules will have the same + # process groups. + root_kwargs["process_group"] = (self.process_group, self._inter_node_pg) + + _auto_wrap( + module, + auto_wrap_policy, + self._ignored_modules, + self._ignored_params, + root_kwargs, + FullyShardedDataParallel, + ) + + backward_prefetch_limit = 1 + forward_prefetch_limit = 1 + _init_core_state( + self, + sharding_strategy, + mixed_precision, + cpu_offload, + limit_all_gathers, + use_orig_params, + backward_prefetch_limit, + forward_prefetch_limit, + ) + _init_runtime_state(self) + _init_prefetching_state(self, backward_prefetch, forward_prefetch) + _init_buffer_state(self, module) + # extension needs to be set before `_init_param_handle_from_module()` + _init_extension(self, device_mesh) + _init_param_handle_from_module( + self, + module, + device_id, + param_init_fn, + sync_module_states, + ) + self._fsdp_wrapped_module = module + if not use_orig_params: + _check_orig_params_flattened(self, self._ignored_params) + _register_flat_param(self, self) + + # `_state_dict_type` controls the `state_dict()` behavior, which is + # implemented using post-save and pre-load hooks + _init_state_dict_state(self) + _register_all_state_dict_hooks(self) + self._zero_scalar = None + + @property + def module(self) -> nn.Module: + """Return the wrapped module.""" + # FSDP's `.module` must refer to the innermost wrapped module when + # composing with other module wrappers in order for state dict to work + if isinstance(self._fsdp_wrapped_module, ActivationWrapper): + return getattr(self._fsdp_wrapped_module, _CHECKPOINT_WRAPPED_MODULE) + return self._fsdp_wrapped_module + + @property + def _has_params(self) -> bool: + """Returns whether this FSDP instance manages any parameters.""" + return hasattr(self, "_handle") and self._handle is not None + + @property + def _flat_param(self) -> FlatParameter | None: + return self._handle.flat_param if self._handle else None + + def __getattr__(self, name: str) -> Any: + """Forward missing attributes to the wrapped module.""" + try: + return super().__getattr__(name) # defer to nn.Module's logic + except AttributeError: + return getattr(self._fsdp_wrapped_module, name) + + def __getitem__(self, key: int) -> Any: + """Forward indexing calls in case the module is an ``nn.Sequential``.""" + if hasattr(self, FSDP_WRAPPED_MODULE): + return self._fsdp_wrapped_module.__getitem__(key) # type: ignore[operator] + return super().__getitem__(key) + + def check_is_root(self) -> bool: + """Check if this instance is a root FSDP module.""" + return _is_fsdp_root(self, self) + + @staticmethod + def fsdp_modules( + module: nn.Module, + root_only: bool = False, + ) -> list["FullyShardedDataParallel"]: + """Return all nested FSDP instances. + + This possibly includes ``module`` itself and only includes FSDP root modules if ``root_only=True``. + + Args: + module (torch.nn.Module): Root module, which may or may not be an + ``FSDP`` module. + root_only (bool): Whether to return only FSDP root modules. + (Default: ``False``) + + Returns: + List[FullyShardedDataParallel]: FSDP modules that are nested in + the input ``module``. + """ + if root_only: + return _get_fsdp_root_states(module) + return traversal_utils._get_fsdp_states(module) + + def apply(self, fn: Callable[[nn.Module], None]) -> "FullyShardedDataParallel": + r"""Apply ``fn`` recursively to every submodule (as returned by ``.children()``) as well as self. + + Typical use includes initializing the parameters of a model (see also :ref:`nn-init-doc`). + + Compared to ``torch.nn.Module.apply``, this version additionally gathers + the full parameters before applying ``fn``. It should not be called from + within another ``summon_full_params`` context. + + Args: + fn (:class:`Module` -> None): function to be applied to each submodule + + Returns: + Module: self + """ + uninitialized = self._is_root is None + self._assert_state(TrainingState.IDLE) + # Use `_unshard_params_for_summon()` with `recurse=False` instead of + # `_unshard_fsdp_state_params()` directly to perform lazy + # initialization, which is needed to initialize `FlatParameter` + # parameter attributes as required by the unshard logic + with _unshard_params_for_summon( + self, + self, + writeback=True, + rank0_only=False, + offload_to_cpu=False, + with_grads=False, + ): + ret = super().apply(fn) + + # Reset lazy init called in `_unshard_params_for_summon()` since + # `apply()` may have been called on FSDP instance that is not truly a + # root, in which case it will be incorrectly marked as one. + if uninitialized and self._is_root: + for module in traversal_utils._get_fsdp_states(self): + module._reset_lazy_init() + + return ret + + def _mixed_precision_enabled_for_buffers(self) -> bool: + """Return whether the user explicitly enabled buffer mixed precision. + + NOTE: Unlike parameters and gradient reduction, buffer mixed precision + is applied at the FSDP instance level, not the ``FlatParameter`` level, + which may be different for the composable code path. + """ + return self.mixed_precision.buffer_dtype is not None + + def _low_precision_hook_enabled(self) -> bool: + """Whether a low precision hook is registered or not.""" + return self._comm_hook is not None and self._comm_hook in LOW_PRECISION_HOOKS + + def _reset_lazy_init(self) -> None: + """Reset instance so :func:`_lazy_init` will run on the next forward.""" + self._is_root: bool | None = None + + @staticmethod + def set_state_dict_type( + module: nn.Module, + state_dict_type: StateDictType, + state_dict_config: StateDictConfig | None = None, + optim_state_dict_config: OptimStateDictConfig | None = None, + ) -> StateDictSettings: + """Set the ``state_dict_type`` of all the descendant FSDP modules of the target module. + + Also takes (optional) configuration for the model's and optimizer's state dict. + The target module does not have to be a FSDP module. If the target + module is a FSDP module, its ``state_dict_type`` will also be changed. + + .. note:: This API should be called for only the top-level (root) + module. + + .. note:: This API enables users to transparently use the conventional + ``state_dict`` API to take model checkpoints in cases where the + root FSDP module is wrapped by another ``nn.Module``. For example, + the following will ensure ``state_dict`` is called on all non-FSDP + instances, while dispatching into `sharded_state_dict` implementation + for FSDP: + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> model = DDP(FSDP(...)) + >>> FSDP.set_state_dict_type( + >>> model, + >>> StateDictType.SHARDED_STATE_DICT, + >>> state_dict_config = ShardedStateDictConfig(offload_to_cpu=True), + >>> optim_state_dict_config = OptimStateDictConfig(offload_to_cpu=True), + >>> ) + >>> param_state_dict = model.state_dict() + >>> optim_state_dict = FSDP.optim_state_dict(model, optim) + + Args: + module (torch.nn.Module): Root module. + state_dict_type (StateDictType): the desired ``state_dict_type`` to set. + state_dict_config (Optional[StateDictConfig]): the configuration for the + target ``state_dict_type``. + optim_state_dict_config (Optional[OptimStateDictConfig]): the configuration + for the optimizer state dict. + + Returns: + A StateDictSettings that include the previous state_dict type and + configuration for the module. + """ + warnings.warn( + "FSDP.state_dict_type() and FSDP.set_state_dict_type() are being " + "deprecated. Please use APIs, get_state_dict() and set_state_dict(), " + "which can support different parallelisms, FSDP1, FSDP2, DDP. " + "API doc: https://pytorch.org/docs/stable/distributed.checkpoint.html" + "#torch.distributed.checkpoint.state_dict.get_state_dict ." + "Tutorial: https://pytorch.org/tutorials/recipes/distributed_checkpoint_recipe.html .", + FutureWarning, + stacklevel=2, + ) + _state_dict_type_to_config = { + StateDictType.FULL_STATE_DICT: FullStateDictConfig, + StateDictType.LOCAL_STATE_DICT: LocalStateDictConfig, + StateDictType.SHARDED_STATE_DICT: ShardedStateDictConfig, + } + _optim_state_dict_type_to_config = { + StateDictType.FULL_STATE_DICT: FullOptimStateDictConfig, + StateDictType.LOCAL_STATE_DICT: LocalOptimStateDictConfig, + StateDictType.SHARDED_STATE_DICT: ShardedOptimStateDictConfig, + } + + # Use the default config if a state_dict config is not set. + state_dict_config_type = _state_dict_type_to_config[state_dict_type] + optim_state_dict_config_type = _optim_state_dict_type_to_config[state_dict_type] + if state_dict_config is None: + state_dict_config = state_dict_config_type() + if optim_state_dict_config is None: + optim_state_dict_config = optim_state_dict_config_type() + if state_dict_config_type is not type(state_dict_config): + raise RuntimeError( + f"Expected state_dict_config of type {state_dict_config_type} " + f"but got {type(state_dict_config)}" + ) + if optim_state_dict_config_type is not type(optim_state_dict_config): + raise RuntimeError( + f"Expected optim_state_dict_config of type {optim_state_dict_config_type} " + f"but got {type(optim_state_dict_config)}" + ) + + # Set the state_dict type and configurations. + prev_state_dict_type = None + prev_state_dict_config = None + prev_optim_state_dict_config = None + for submodule in traversal_utils._get_fsdp_states(module): + if prev_state_dict_type is None: + prev_state_dict_type = submodule._state_dict_type + else: + if prev_state_dict_type != submodule._state_dict_type: + raise AssertionError( + "All FSDP modules should have the same state_dict_type." + ) + if prev_state_dict_config is None: + prev_state_dict_config = submodule._state_dict_config + else: + if not isinstance( + submodule._state_dict_config, type(prev_state_dict_config) + ): + raise AssertionError( + "All FSDP modules must have the same type of state_dict_config." + ) + if prev_optim_state_dict_config is None: + prev_optim_state_dict_config = submodule._optim_state_dict_config + else: + if not isinstance( + submodule._optim_state_dict_config, + type(prev_optim_state_dict_config), + ): + raise AssertionError( + "All FSDP modules must have the same type of optim_state_dict_config." + ) + + submodule._state_dict_type = state_dict_type + submodule._state_dict_config = state_dict_config + submodule._optim_state_dict_config = optim_state_dict_config + + return StateDictSettings( + prev_state_dict_type, prev_state_dict_config, prev_optim_state_dict_config + ) + + @staticmethod + def get_state_dict_type(module: nn.Module) -> StateDictSettings: + """Get the state_dict_type and the corresponding configurations for the FSDP modules rooted at ``module``. + + The target module does not have to be an FSDP module. + + Returns: + A ``StateDictSettings`` containing the state_dict_type and + state_dict / optim_state_dict configs that are currently set. + + Raises: + ``AssertionError`` if the ``StateDictSettings`` for different + FSDP submodules differ. + """ + state_dict_settings: StateDictSettings | None = None + for submodule in FullyShardedDataParallel.fsdp_modules(module): + if state_dict_settings is None: + state_dict_settings = StateDictSettings( + state_dict_type=submodule._state_dict_type, + state_dict_config=submodule._state_dict_config, + optim_state_dict_config=submodule._optim_state_dict_config, + ) + _set_optim_use_dtensor(submodule, state_dict_settings) + else: + submodule_settings = StateDictSettings( + submodule._state_dict_type, + submodule._state_dict_config, + submodule._optim_state_dict_config, + ) + if state_dict_settings != submodule_settings: + raise AssertionError( + "All FSDP modules must have the same state dict settings." + f"Got {submodule_settings} and {state_dict_settings}." + ) + _set_optim_use_dtensor(submodule, submodule_settings) + return state_dict_settings + + @staticmethod + @contextlib.contextmanager + def state_dict_type( + module: nn.Module, + state_dict_type: StateDictType, + state_dict_config: StateDictConfig | None = None, + optim_state_dict_config: OptimStateDictConfig | None = None, + ) -> Generator: + """Set the ``state_dict_type`` of all the descendant FSDP modules of the target module. + + This context manager has the same functions as :meth:`set_state_dict_type`. Read the document of + :meth:`set_state_dict_type` for the detail. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> model = DDP(FSDP(...)) + >>> with FSDP.state_dict_type( + >>> model, + >>> StateDictType.SHARDED_STATE_DICT, + >>> ): + >>> checkpoint = model.state_dict() + + Args: + module (torch.nn.Module): Root module. + state_dict_type (StateDictType): the desired ``state_dict_type`` to set. + state_dict_config (Optional[StateDictConfig]): the model ``state_dict`` + configuration for the target ``state_dict_type``. + optim_state_dict_config (Optional[OptimStateDictConfig]): the optimizer + ``state_dict`` configuration for the target ``state_dict_type``. + """ + prev_state_dict_settings = FullyShardedDataParallel.set_state_dict_type( + module, + state_dict_type, + state_dict_config, + optim_state_dict_config, + ) + yield + FullyShardedDataParallel.set_state_dict_type( + module, + prev_state_dict_settings.state_dict_type, + prev_state_dict_settings.state_dict_config, + prev_state_dict_settings.optim_state_dict_config, + ) + + def forward(self, *args: Any, **kwargs: Any) -> Any: + """Run the forward pass for the wrapped module, inserting FSDP-specific pre- and post-forward sharding logic.""" + handle = self._handle + with torch.autograd.profiler.record_function( + "FullyShardedDataParallel.forward" + ): + args, kwargs = _root_pre_forward(self, self, args, kwargs) + unused = None + args, kwargs = _pre_forward( + self, + handle, + _pre_forward_unshard, + self._fsdp_wrapped_module, + args, + kwargs, + ) + if handle: + _p_assert( + handle.flat_param.device == self.compute_device, + "Expected `FlatParameter` to be on the compute device " + f"{self.compute_device} but got {handle.flat_param.device}", + ) + output = self._fsdp_wrapped_module(*args, **kwargs) + return _post_forward( + self, handle, _post_forward_reshard, self, unused, output + ) + + @staticmethod + @contextlib.contextmanager + def summon_full_params( + module: nn.Module, + recurse: bool = True, + writeback: bool = True, + rank0_only: bool = False, + offload_to_cpu: bool = False, + with_grads: bool = False, + ) -> Generator: + r"""Expose full params for FSDP instances with this context manager. + + Can be useful *after* forward/backward for a model to get + the params for additional processing or checking. It can take a non-FSDP + module and will summon full params for all contained FSDP modules as + well as their children, depending on the ``recurse`` argument. + + .. note:: This can be used on inner FSDPs. + .. note:: This can *not* be used within a forward or backward pass. Nor + can forward and backward be started from within this context. + .. note:: Parameters will revert to their local shards after the context + manager exits, storage behavior is the same as forward. + .. note:: The full parameters can be modified, but only the portion + corresponding to the local param shard will persist after the + context manager exits (unless ``writeback=False``, in which case + changes will be discarded). In the case where FSDP does not shard + the parameters, currently only when ``world_size == 1``, or ``NO_SHARD`` + config, the modification is persisted regardless of ``writeback``. + .. note:: This method works on modules which are not FSDP themselves but + may contain multiple independent FSDP units. In that case, the given + arguments will apply to all contained FSDP units. + + .. warning:: Note that ``rank0_only=True`` in conjunction with + ``writeback=True`` is not currently supported and will raise an + error. This is because model parameter shapes would be different + across ranks within the context, and writing to them can lead to + inconsistency across ranks when the context is exited. + + .. warning:: Note that ``offload_to_cpu`` and ``rank0_only=False`` will + result in full parameters being redundantly copied to CPU memory for + GPUs that reside on the same machine, which may incur the risk of + CPU OOM. It is recommended to use ``offload_to_cpu`` with + ``rank0_only=True``. + + Args: + recurse (bool, Optional): recursively summon all params for nested + FSDP instances (default: True). + writeback (bool, Optional): if ``False``, modifications to params are + discarded after the context manager exits; + disabling this can be slightly more efficient (default: True) + rank0_only (bool, Optional): if ``True``, full parameters are + materialized on only global rank 0. This means that within the + context, only rank 0 will have full parameters and the other + ranks will have sharded parameters. Note that setting + ``rank0_only=True`` with ``writeback=True`` is not supported, + as model parameter shapes will be different across ranks + within the context, and writing to them can lead to + inconsistency across ranks when the context is exited. + offload_to_cpu (bool, Optional): If ``True``, full parameters are + offloaded to CPU. Note that this offloading currently only + occurs if the parameter is sharded (which is only not the case + for world_size = 1 or ``NO_SHARD`` config). It is recommended + to use ``offload_to_cpu`` with ``rank0_only=True`` to avoid + redundant copies of model parameters being offloaded to the same CPU memory. + with_grads (bool, Optional): If ``True``, gradients are also + unsharded with the parameters. Currently, this is only + supported when passing ``use_orig_params=True`` to the FSDP + constructor and ``offload_to_cpu=False`` to this method. + (Default: ``False``) + """ + with _unshard_params( + module, recurse, writeback, rank0_only, offload_to_cpu, with_grads + ): + yield + + @contextlib.contextmanager + def _deregister_orig_params_ctx(self): + """Deregister the original parameters and expose the :class:`FlatParameter`. + + If a :class:`FlatParameter` is sharded, then + this refreshes the sharded views before exiting. This method should + only be called when using the original parameters. + """ + _p_assert( + self._use_orig_params, + "`_deregister_orig_params_ctx()` should only be called when " + "`_use_orig_params=True`", + ) + for fsdp_module in traversal_utils._get_fsdp_states(self): + _deregister_orig_params(fsdp_module, fsdp_module) + try: + yield + finally: + for fsdp_module in traversal_utils._get_fsdp_states(self): + _register_orig_params(fsdp_module, fsdp_module) + + @copy_method_sig(nn.Module._apply) + def _apply(self, *args, **kwargs): + """Deregister the original parameters and expose the :class:`FlatParameter` s before calling ``_apply()``.""" + # When using the original parameters: Since (1) the `FlatParameter`s + # own the storage and (2) `_apply()` is the subroutine underlying the + # most common storage-changing ops like `to()` and `cuda()`, we + # override `_apply()` to have the storage change directly performed on + # the `FlatParameter`s instead of applying to the original parameters + # and then writing back to the `FlatParameter`s. + context = ( + self._deregister_orig_params_ctx() + if self._use_orig_params + else contextlib.nullcontext() + ) + with context: + return super()._apply(*args, **kwargs) + + @copy_method_params(nn.Module.named_buffers) + def named_buffers( + self, + *args, + **kwargs, + ) -> Iterator[tuple[str, torch.Tensor]]: + """Return an iterator over module buffers, yielding both the name of the buffer and the buffer itself. + + Intercepts buffer names and removes all occurrences of the FSDP-specific flattened buffer prefix + when inside the :meth:`summon_full_params` context manager. + """ + should_clean_name = self.training_state == TrainingState.SUMMON_FULL_PARAMS + for buffer_name, buffer in super().named_buffers(*args, **kwargs): + if should_clean_name: + # Remove any instances of the FSDP-specific prefix; there can + # be multiple in the case of nested FSDP modules + buffer_name = buffer_name.replace(FSDP_PREFIX, "") + yield (buffer_name, buffer) + + @copy_method_params(nn.Module.named_parameters) + def named_parameters( + self, + *args, + **kwargs, + ) -> Iterator[tuple[str, torch.nn.Parameter]]: + """Return an iterator over module parameters, yielding both the name of the parameter and the parameter itself. + + Intercepts parameter names and removes all occurrences of the FSDP-specific flattened parameter prefix + when inside the :meth:`summon_full_params` context manager. + """ + should_clean_name = self.training_state == TrainingState.SUMMON_FULL_PARAMS + for param_name, param in super().named_parameters(*args, **kwargs): + if should_clean_name: + # Remove any instances of the FSDP-specific prefix; there can + # be multiple in the case of nested FSDP modules + param_name = param_name.replace(FSDP_PREFIX, "") + yield (param_name, param) + + def _assert_state(self, state: TrainingState | list[TrainingState]) -> None: + """Assert we are in the given state.""" + # Since assert can be turned off and this error checking + # is really important, we use explicit error checking + # and raise a ValueError if needed. + if isinstance(state, TrainingState): + state = [state] + if self.training_state not in state: + msg = ( + f"expected to be in states {state} but current state " + f"is {self.training_state}" + ) + # In case we are failing in the context of autograd hook, asserting + # may not generate useful msg. So, let's print it to be sure. + if self.rank == 0: + print(f"Asserting FSDP instance is: {self}") + print(f"ERROR: {msg}") + traceback.print_stack() + raise ValueError(msg) + + @contextmanager + def no_sync(self) -> Generator: + """Disable gradient synchronizations across FSDP instances. + + Within this context, gradients will be accumulated in module + variables, which will later be synchronized in the first + forward-backward pass after exiting the context. This should only be + used on the root FSDP instance and will recursively apply to all + children FSDP instances. + + .. note:: This likely results in higher memory usage because FSDP will + accumulate the full model gradients (instead of gradient shards) + until the eventual sync. + + .. note:: When used with CPU offloading, the gradients will not be + offloaded to CPU when inside the context manager. Instead, they + will only be offloaded right after the eventual sync. + """ + _lazy_init(self, self) + if not self._is_root: + raise RuntimeError( + "`no_sync()` on inner FSDP instances is not supported. Please call `no_sync()` on root FSDP module." + ) + self._assert_state(TrainingState.IDLE) + old_flags = [] + for m in self.modules(): + if isinstance(m, FullyShardedDataParallel): + old_flags.append((m, m._sync_gradients)) + m._sync_gradients = False + try: + yield + finally: + for m, old_flag in old_flags: + if m._sync_gradients: + raise AssertionError( + "`_sync_gradients` was incorrectly set to " + "`True` while in the `no_sync()` context manager" + ) + m._sync_gradients = old_flag + + @torch.no_grad() + def clip_grad_norm_( + self, max_norm: float | int, norm_type: float | int = 2.0 + ) -> torch.Tensor: + """Clip the gradient norm of all parameters. + + The norm is computed over all parameters' gradients as viewed as a single vector, and the + gradients are modified in-place. + + Args: + max_norm (float or int): max norm of the gradients + norm_type (float or int): type of the used p-norm. Can be ``'inf'`` + for infinity norm. + + Returns: + Total norm of the parameters (viewed as a single vector). + + If every FSDP instance uses ``NO_SHARD``, meaning that no + gradients are sharded across ranks, then you may directly use + :func:`torch.nn.utils.clip_grad_norm_`. + + If at least some FSDP instance uses a sharded strategy (i.e. + one other than ``NO_SHARD``), then you should use this method + instead of :func:`torch.nn.utils.clip_grad_norm_` since this method + handles the fact that gradients are sharded across ranks. + + The total norm returned will have the "largest" dtype across + all parameters/gradients as defined by PyTorch's type promotion + semantics. For example, if *all* parameters/gradients use a low + precision dtype, then the returned norm's dtype will be that low + precision dtype, but if there exists at least one parameter/ + gradient using FP32, then the returned norm's dtype will be FP32. + + .. warning:: This needs to be called on all ranks since it uses + collective communications. + """ + _lazy_init(self, self) + if not self._is_root: + raise RuntimeError( + "`clip_grad_norm_()` should only be called on the root FSDP instance" + ) + if self._zero_scalar is None: + self._zero_scalar = torch.tensor(0.0, device=self.compute_device) + self._assert_state(TrainingState.IDLE) + # If every FSDP instance uses `NO_SHARD`, then we can directly use + # the normal `nn.utils` one targeting local gradients + all_no_shard = all( + not handle.uses_sharded_strategy for handle in self._all_handles + ) + if all_no_shard: + return torch.nn.utils.clip_grad_norm_( + self.parameters(), max_norm, norm_type + ) + # Otherwise, there exists some FSDP instance using a sharded strategy, + # where sharded and non-sharded parameters must be handled separately + max_norm = float(max_norm) + norm_type = float(norm_type) + sharded_params_set = set() + nonsharded_params_set = set() # `NO_SHARD` or not FSDP-managed + # Make sure to compute the local norm using lists for deterministic + # iteration order and hence deterministic total norm computation + sharded_params = [] + nonsharded_params = [] + grads: list[torch.Tensor] = [] + for handle in self._all_handles: + if handle.uses_sharded_strategy: + target_set = sharded_params_set + target_list = sharded_params + else: + target_set = nonsharded_params_set + target_list = nonsharded_params + if handle._use_orig_params: + for param in handle.flat_param._params: + if param not in target_set: + target_set.add(param) + target_list.append(param) + if param.grad is not None: + grads.append(param.grad) + else: + if handle.flat_param not in target_set: + target_set.add(handle.flat_param) + target_list.append(handle.flat_param) + if handle.flat_param.grad is not None: + grads.append(handle.flat_param.grad) + for param in self.parameters(): + not_fsdp_managed = ( + param not in sharded_params_set and param not in nonsharded_params_set + ) + if not_fsdp_managed: + nonsharded_params_set.add(param) + nonsharded_params.append(param) + if param.grad is not None: + grads.append(param.grad) + # Compute local norms (forced to be in FP32) + local_sharded_norm = _get_grad_norm( + sharded_params, norm_type, self._zero_scalar, self.compute_device + ) + local_nonsharded_norm = ( + _get_grad_norm( + nonsharded_params, norm_type, self._zero_scalar, self.compute_device + ) + if nonsharded_params + else None + ) + # Reconstruct the total gradient norm depending on the norm type + if norm_type == math.inf: + total_norm = ( + torch.maximum(local_sharded_norm, local_nonsharded_norm) + if local_nonsharded_norm is not None + else local_sharded_norm + ) + dist.all_reduce( + total_norm, op=torch.distributed.ReduceOp.MAX, group=self.process_group + ) + else: + total_norm = local_sharded_norm**norm_type + dist.all_reduce(total_norm, group=self.process_group) + # All-reducing the local non-sharded norm would count it an extra + # world-size-many times + if local_nonsharded_norm is not None: + total_norm += local_nonsharded_norm**norm_type + total_norm = total_norm ** (1.0 / norm_type) + if self.cpu_offload.offload_params: + total_norm = total_norm.cpu() + + clip_coef = max_norm / (total_norm + 1e-6) + # Multiplying by the clamped coefficient is meaningless when it is + # equal to 1, but it avoids the host-device sync that would result from + # `if clip_coef < 1` + clip_coef_clamped = torch.clamp(clip_coef, max=1.0) + for grad in grads: + grad.mul_(clip_coef_clamped.to(grad.device, grad.dtype)) + # Use the "largest" dtype by type promotion semantics to use the same + # dtype as if we did not force local norm computation to be in FP32 + if len(grads) == 0: + # If this rank has no gradients, then we must default to FP32 + # unless we use additional communication, which we prefer to avoid + # since `clip_grad_norm_()` is called in the training loop + warnings.warn( + f"Called FSDP.clip_grad_norm_() on rank {self.rank} with no " + "gradients -- returning the total norm in the default dtype " + f"{total_norm.dtype}", + stacklevel=2, + ) # warn since this is generally unexpected + return total_norm + total_norm_dtype = functools.reduce( + torch.promote_types, + [grad.dtype for grad in grads], + ) + return total_norm.to(total_norm_dtype) + + @staticmethod + def _warn_optim_input(optim_input, *, stacklevel: int = 1): + if optim_input is not None: + warnings.warn( + "The `optim_input` argument is deprecated and will be removed after PyTorch 1.13. " + "You may remove it from your code without changing its functionality.", + FutureWarning, + stacklevel=stacklevel + 1, + ) + + @staticmethod + def _is_using_optim_input(optim_input, optim) -> bool: + if optim_input is None and optim is None: + # Use the default behavior of `optim_input`` + return True + if optim_input is not None: + # Use the `optim_input` code path + return True + # Use the `optim` code path + return False + + @staticmethod + def _warn_legacy_optim_state_dict(curr: str, new: str, *, stacklevel: int = 1): + warnings.warn( + f"``FullyShardedDataParallel.{curr}``is being deprecated and is " + f"replaced by ``FullyShardedDataParallel.{new}``. " + f"``FullyShardedDataParallel.{curr}`` may be removed after PyTorch 2.2.", + FutureWarning, + stacklevel=stacklevel + 1, + ) + + @staticmethod + def _optim_state_dict_impl( + model: torch.nn.Module, + optim: torch.optim.Optimizer, + optim_state_dict: dict[str, Any], + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + rank0_only: bool = True, + full_state_dict: bool = True, + group: dist.ProcessGroup | None = None, + cpu_offload: bool = True, + *, + _stacklevel: int = 1, + ) -> dict[str, Any]: + """Transform the state-dict of an optimizer corresponding to a sharded model. + + This is the internal API that is used by all the optim_state_dict implementations. + Given model, optim, the original optim_state_dict, this API removes the + FSDP internal information and internal sharding from the optim_state_dict. + """ + if full_state_dict: + FullyShardedDataParallel._warn_optim_input( + optim_input, stacklevel=_stacklevel + 1 + ) + using_optim_input = FullyShardedDataParallel._is_using_optim_input( + optim_input, + optim, + ) + else: + using_optim_input = False + if optim_input is not None or rank0_only: + raise AssertionError( + f"Expected optim_input to be None and rank0_only to be False, " + f"got optim_input={optim_input}, rank0_only={rank0_only}" + ) + + use_orig_params = FullyShardedDataParallel.fsdp_modules(model)[ + 0 + ]._use_orig_params + if not all( + use_orig_params == m._use_orig_params + for m in FullyShardedDataParallel.fsdp_modules(model) + ): + raise AssertionError( + "Not all FSDP modules have the same _use_orig_params value" + ) + + return _optim_state_dict( + model=model, + optim=optim, + optim_state_dict=optim_state_dict, + optim_input=optim_input, + rank0_only=rank0_only, + shard_state=not full_state_dict, + group=group, + using_optim_input=using_optim_input, + use_orig_params=use_orig_params, + cpu_offload=cpu_offload, + ) + + @staticmethod + def _optim_state_dict_to_load_impl( + optim_state_dict: dict[str, Any], + model: torch.nn.Module, + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + optim: torch.optim.Optimizer | None = None, + full_state_dict: bool = True, + rank0_only: bool = False, + is_named_optimizer: bool = False, + group: dist.ProcessGroup | None = None, + ) -> dict[str, Any]: + """ + Convert an optimizer state-dict so that it can be loaded into the optimizer associated with the FSDP model. + + This is the internal API that is used by all the load optim_state_dict implementations. + Given model, optim, and the saved optim_state_dict, this API adds the FSDP + internal information and internal sharding to the optim_state_dict. + """ + if full_state_dict: + FullyShardedDataParallel._warn_optim_input(optim_input) + using_optim_input = FullyShardedDataParallel._is_using_optim_input( + optim_input, + optim, + ) + else: + using_optim_input = False + if optim_input is not None or rank0_only: + raise AssertionError( + f"Expected optim_input to be None and rank0_only to be False, " + f"got optim_input={optim_input}, rank0_only={rank0_only}" + ) + + use_orig_params = FullyShardedDataParallel.fsdp_modules(model)[ + 0 + ]._use_orig_params + if not all( + use_orig_params == m._use_orig_params + for m in FullyShardedDataParallel.fsdp_modules(model) + ): + raise AssertionError( + "Not all FSDP modules have the same _use_orig_params value" + ) + + if rank0_only and dist.get_rank(group) > 0: + optim_state_dict = {} + sharded_osd = _flatten_optim_state_dict( + optim_state_dict, + model=model, + use_orig_params=use_orig_params, + optim=(optim if is_named_optimizer else None), + rank0_only=rank0_only, + group=group, + ) + return _rekey_sharded_optim_state_dict( + sharded_osd, + model=model, + optim=optim, + optim_input=optim_input, + using_optim_input=using_optim_input, + is_named_optimizer=is_named_optimizer, + ) + + @staticmethod + def full_optim_state_dict( + model: torch.nn.Module, + optim: torch.optim.Optimizer, + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + rank0_only: bool = True, + group: dist.ProcessGroup | None = None, + ) -> dict[str, Any]: + """Return the full optimizer state-dict. + + Consolidates the full optimizer state on rank 0 and returns it + as a :class:`dict` following the convention of + :meth:`torch.optim.Optimizer.state_dict`, i.e. with keys ``"state"`` + and ``"param_groups"``. The flattened parameters in ``FSDP`` modules + contained in ``model`` are mapped back to their unflattened parameters. + + This needs to be called on all ranks since it uses + collective communications. However, if ``rank0_only=True``, then + the state dict is only populated on rank 0, and all other ranks + return an empty :class:`dict`. + + Unlike ``torch.optim.Optimizer.state_dict()``, this method + uses full parameter names as keys instead of parameter IDs. + + Like in :meth:`torch.optim.Optimizer.state_dict`, the tensors + contained in the optimizer state dict are not cloned, so there may + be aliasing surprises. For best practices, consider saving the + returned optimizer state dict immediately, e.g. using + ``torch.save()``. + + Args: + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + were passed into the optimizer ``optim``. + optim (torch.optim.Optimizer): Optimizer for ``model`` 's + parameters. + optim_input (Optional[Union[List[Dict[str, Any]], Iterable[torch.nn.Parameter]]]): + Input passed into the optimizer ``optim`` representing either a + :class:`list` of parameter groups or an iterable of parameters; + if ``None``, then this method assumes the input was + ``model.parameters()``. This argument is deprecated, and there + is no need to pass it in anymore. (Default: ``None``) + rank0_only (bool): If ``True``, saves the populated :class:`dict` + only on rank 0; if ``False``, saves it on all ranks. (Default: + ``True``) + group (dist.ProcessGroup): Model's process group or ``None`` if using + the default process group. (Default: ``None``) + + Returns: + Dict[str, Any]: A :class:`dict` containing the optimizer state for + ``model`` 's original unflattened parameters and including keys + "state" and "param_groups" following the convention of + :meth:`torch.optim.Optimizer.state_dict`. If ``rank0_only=True``, + then nonzero ranks return an empty :class:`dict`. + """ + FullyShardedDataParallel._warn_legacy_optim_state_dict( + "full_optim_state_dict", + "optim_state_dict", + stacklevel=2, + ) + return FullyShardedDataParallel._optim_state_dict_impl( + model=model, + optim=optim, + optim_state_dict=optim.state_dict(), + optim_input=optim_input, + rank0_only=rank0_only, + group=group, + full_state_dict=True, + _stacklevel=2, + ) + + @staticmethod + def sharded_optim_state_dict( + model: torch.nn.Module, + optim: torch.optim.Optimizer, + group: dist.ProcessGroup | None = None, + ) -> dict[str, Any]: + """Return the optimizer state-dict in its sharded form. + + The API is similar to :meth:`full_optim_state_dict` but this API chunks + all non-zero-dimension states to :class:`ShardedTensor` to save memory. + This API should only be used when the model ``state_dict`` is derived + with the context manager ``with state_dict_type(SHARDED_STATE_DICT):``. + + For the detailed usage, refer to :meth:`full_optim_state_dict`. + + .. warning:: The returned state dict contains ``ShardedTensor`` and + cannot be directly used by the regular ``optim.load_state_dict``. + """ + FullyShardedDataParallel._warn_legacy_optim_state_dict( + "sharded_optim_state_dict", + "optim_state_dict", + stacklevel=2, + ) + return FullyShardedDataParallel._optim_state_dict_impl( + model=model, + optim=optim, + optim_state_dict=optim.state_dict(), + optim_input=None, + rank0_only=False, + full_state_dict=False, + group=group, + _stacklevel=2, + ) + + @staticmethod + def shard_full_optim_state_dict( + full_optim_state_dict: dict[str, Any], + model: torch.nn.Module, + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + optim: torch.optim.Optimizer | None = None, + ) -> dict[str, Any]: + """Shard a full optimizer state-dict. + + Remaps the state in ``full_optim_state_dict`` to flattened parameters instead of unflattened + parameters and restricts to only this rank's part of the optimizer state. + The first argument should be the return value of :meth:`full_optim_state_dict`. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> model, optim = ... + >>> full_osd = FSDP.full_optim_state_dict(model, optim) + >>> torch.save(full_osd, PATH) + >>> # Define new model with possibly different world size + >>> new_model, new_optim = ... + >>> full_osd = torch.load(PATH) + >>> sharded_osd = FSDP.shard_full_optim_state_dict(full_osd, new_model) + >>> new_optim.load_state_dict(sharded_osd) + + .. note:: Both :meth:`shard_full_optim_state_dict` and + :meth:`scatter_full_optim_state_dict` may be used to get the + sharded optimizer state dict to load. Assuming that the full + optimizer state dict resides in CPU memory, the former requires + each rank to have the full dict in CPU memory, where each rank + individually shards the dict without any communication, while the + latter requires only rank 0 to have the full dict in CPU memory, + where rank 0 moves each shard to GPU memory (for NCCL) and + communicates it to ranks appropriately. Hence, the former has + higher aggregate CPU memory cost, while the latter has higher + communication cost. + + Args: + full_optim_state_dict (Dict[str, Any]): Optimizer state dict + corresponding to the unflattened parameters and holding the + full non-sharded optimizer state. + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + correspond to the optimizer state in ``full_optim_state_dict``. + optim_input (Optional[Union[List[Dict[str, Any]], Iterable[torch.nn.Parameter]]]): + Input passed into the optimizer representing either a + :class:`list` of parameter groups or an iterable of parameters; + if ``None``, then this method assumes the input was + ``model.parameters()``. This argument is deprecated, and there + is no need to pass it in anymore. (Default: ``None``) + optim (Optional[torch.optim.Optimizer]): Optimizer that will load + the state dict returned by this method. This is the preferred + argument to use over ``optim_input``. (Default: ``None``) + + Returns: + Dict[str, Any]: The full optimizer state dict now remapped to + flattened parameters instead of unflattened parameters and + restricted to only include this rank's part of the optimizer state. + """ + FullyShardedDataParallel._warn_legacy_optim_state_dict( + "shard_full_optim_state_dict", + "optim_state_dict_to_load", + stacklevel=2, + ) + return FullyShardedDataParallel._optim_state_dict_to_load_impl( + optim_state_dict=full_optim_state_dict, + model=model, + optim_input=optim_input, + optim=optim, + full_state_dict=True, + is_named_optimizer=False, + ) + + @staticmethod + def flatten_sharded_optim_state_dict( + sharded_optim_state_dict: dict[str, Any], + model: torch.nn.Module, + optim: torch.optim.Optimizer, + ) -> dict[str, Any]: + """Flatten a sharded optimizer state-dict. + + The API is similar to :meth:`shard_full_optim_state_dict`. The only + difference is that the input ``sharded_optim_state_dict`` should be + returned from :meth:`sharded_optim_state_dict`. Therefore, there will + be all-gather calls on each rank to gather ``ShardedTensor`` s. + + Args: + sharded_optim_state_dict (Dict[str, Any]): Optimizer state dict + corresponding to the unflattened parameters and holding the + sharded optimizer state. + model (torch.nn.Module): + Refer to :meth:`shard_full_optim_state_dict`. + optim (torch.optim.Optimizer): Optimizer for ``model`` 's + parameters. + + Returns: + Refer to :meth:`shard_full_optim_state_dict`. + """ + FullyShardedDataParallel._warn_legacy_optim_state_dict( + "flatten_sharded_optim_state_dict", + "optim_state_dict_to_load", + stacklevel=2, + ) + return FullyShardedDataParallel._optim_state_dict_to_load_impl( + optim_state_dict=sharded_optim_state_dict, + model=model, + optim_input=None, + optim=optim, + full_state_dict=False, + is_named_optimizer=False, + ) + + @staticmethod + def scatter_full_optim_state_dict( + full_optim_state_dict: dict[str, Any] | None, + model: torch.nn.Module, + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + optim: torch.optim.Optimizer | None = None, + group: Any | None = None, + ) -> dict[str, Any]: + """Scatter the full optimizer state dict from rank 0 to all other ranks. + + Returns the sharded optimizer state dict on each rank. + The return value is the same as :meth:`shard_full_optim_state_dict`, and on rank + 0, the first argument should be the return value of + :meth:`full_optim_state_dict`. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> model, optim = ... + >>> full_osd = FSDP.full_optim_state_dict(model, optim) # only non-empty on rank 0 + >>> # Define new model with possibly different world size + >>> new_model, new_optim, new_group = ... + >>> sharded_osd = FSDP.scatter_full_optim_state_dict(full_osd, new_model, group=new_group) + >>> new_optim.load_state_dict(sharded_osd) + + .. note:: Both :meth:`shard_full_optim_state_dict` and + :meth:`scatter_full_optim_state_dict` may be used to get the + sharded optimizer state dict to load. Assuming that the full + optimizer state dict resides in CPU memory, the former requires + each rank to have the full dict in CPU memory, where each rank + individually shards the dict without any communication, while the + latter requires only rank 0 to have the full dict in CPU memory, + where rank 0 moves each shard to GPU memory (for NCCL) and + communicates it to ranks appropriately. Hence, the former has + higher aggregate CPU memory cost, while the latter has higher + communication cost. + + Args: + full_optim_state_dict (Optional[Dict[str, Any]]): Optimizer state + dict corresponding to the unflattened parameters and holding + the full non-sharded optimizer state if on rank 0; the argument + is ignored on nonzero ranks. + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + correspond to the optimizer state in ``full_optim_state_dict``. + optim_input (Optional[Union[List[Dict[str, Any]], Iterable[torch.nn.Parameter]]]): + Input passed into the optimizer representing either a + :class:`list` of parameter groups or an iterable of parameters; + if ``None``, then this method assumes the input was + ``model.parameters()``. This argument is deprecated, and there + is no need to pass it in anymore. (Default: ``None``) + optim (Optional[torch.optim.Optimizer]): Optimizer that will load + the state dict returned by this method. This is the preferred + argument to use over ``optim_input``. (Default: ``None``) + group (dist.ProcessGroup): Model's process group or ``None`` if + using the default process group. (Default: ``None``) + + Returns: + Dict[str, Any]: The full optimizer state dict now remapped to + flattened parameters instead of unflattened parameters and + restricted to only include this rank's part of the optimizer state. + """ + FullyShardedDataParallel._warn_legacy_optim_state_dict( + "scatter_full_optim_state_dict", + "optim_state_dict_to_load", + stacklevel=2, + ) + return FullyShardedDataParallel._optim_state_dict_to_load_impl( + optim_state_dict=full_optim_state_dict, + model=model, + optim_input=optim_input, + optim=optim, + full_state_dict=True, + rank0_only=True, + is_named_optimizer=False, + group=group, + ) + + @staticmethod + def rekey_optim_state_dict( + optim_state_dict: dict[str, Any], + optim_state_key_type: OptimStateKeyType, + model: torch.nn.Module, + optim_input: list[dict[str, Any]] | Iterable[torch.nn.Parameter] | None = None, + optim: torch.optim.Optimizer | None = None, + ) -> dict[str, Any]: + """Re-keys the optimizer state dict ``optim_state_dict`` to use the key type ``optim_state_key_type``. + + This can be used to achieve compatibility between optimizer state dicts from models with FSDP + instances and ones without. + + To re-key an FSDP full optimizer state dict (i.e. from + :meth:`full_optim_state_dict`) to use parameter IDs and be loadable to + a non-wrapped model:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> wrapped_model, wrapped_optim = ... + >>> full_osd = FSDP.full_optim_state_dict(wrapped_model, wrapped_optim) + >>> nonwrapped_model, nonwrapped_optim = ... + >>> rekeyed_osd = FSDP.rekey_optim_state_dict(full_osd, OptimStateKeyType.PARAM_ID, nonwrapped_model) + >>> nonwrapped_optim.load_state_dict(rekeyed_osd) + + To re-key a normal optimizer state dict from a non-wrapped model to be + loadable to a wrapped model:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> nonwrapped_model, nonwrapped_optim = ... + >>> osd = nonwrapped_optim.state_dict() + >>> rekeyed_osd = FSDP.rekey_optim_state_dict(osd, OptimStateKeyType.PARAM_NAME, nonwrapped_model) + >>> wrapped_model, wrapped_optim = ... + >>> sharded_osd = FSDP.shard_full_optim_state_dict(rekeyed_osd, wrapped_model) + >>> wrapped_optim.load_state_dict(sharded_osd) + + Returns: + Dict[str, Any]: The optimizer state dict re-keyed using the + parameter keys specified by ``optim_state_key_type``. + """ + FullyShardedDataParallel._warn_optim_input(optim_input) + using_optim_input = FullyShardedDataParallel._is_using_optim_input( + optim_input, + optim, + ) + if optim_state_key_type not in ( + OptimStateKeyType.PARAM_NAME, + OptimStateKeyType.PARAM_ID, + ): + raise AssertionError( + f"Expected optim_state_key_type to be PARAM_NAME or PARAM_ID, got {optim_state_key_type}" + ) + osd = optim_state_dict # alias + # Validate that the existing parameter keys are uniformly typed + uses_param_name_mask = [type(param_key) is str for param_key in osd["state"]] + uses_param_id_mask = [type(param_key) is int for param_key in osd["state"]] + if (any(uses_param_name_mask) and not all(uses_param_name_mask)) or ( + any(uses_param_id_mask) and not all(uses_param_id_mask) + ): + error_msg = f"Invalid parameter keys: {osd['state'].keys()}" + raise ValueError(error_msg) + # Return directly if the existing key type matches the target key type + if ( + optim_state_key_type == OptimStateKeyType.PARAM_NAME + and all(uses_param_name_mask) + ) or ( + optim_state_key_type == OptimStateKeyType.PARAM_ID + and all(uses_param_id_mask) + ): + return osd + # Otherwise, actually perform the re-keying + new_osd = {} + if optim_state_key_type == OptimStateKeyType.PARAM_NAME: # ID -> name + param_id_to_param = ( + _get_param_id_to_param_from_optim_input(model, optim_input) + if using_optim_input + else _get_param_key_to_param(optim) + ) + param_to_param_name = _get_param_to_fqn(model) + param_id_to_param_name: list[str] = [ + param_to_param_name[param] for param in param_id_to_param.values() + ] + new_osd["state"] = { + param_id_to_param_name[param_id]: param_state + for param_id, param_state in osd["state"].items() + } + new_osd["param_groups"] = copy.deepcopy(osd["param_groups"]) + for param_group in new_osd["param_groups"]: + param_group["params"] = sorted( + [ + param_id_to_param_name[param_id] + for param_id in param_group["params"] + ] + ) + return new_osd + elif optim_state_key_type == OptimStateKeyType.PARAM_ID: # name -> ID + param_name_to_param = _get_fqn_to_param(model) + param_to_param_id = ( + _get_param_to_param_id_from_optim_input(model, optim_input) + if using_optim_input + else _get_param_to_param_key(optim) + ) + # Because not all model parameters may be passed as the optimizer + # input, we may need to drop some parameters from this mapping + param_name_to_param_id = { + param_name: param_to_param_id[param] + for param_name, param in param_name_to_param.items() + if param in param_to_param_id + } + new_osd["state"] = { + param_name_to_param_id[param_name]: param_state + for param_name, param_state in osd["state"].items() + } + new_osd["param_groups"] = copy.deepcopy(osd["param_groups"]) + for param_group in new_osd["param_groups"]: + param_group["params"] = sorted( + [ + param_name_to_param_id[param_name] + for param_name in param_group["params"] + ] + ) + return new_osd + return new_osd # should never reach here + + @staticmethod + def optim_state_dict( + model: torch.nn.Module, + optim: torch.optim.Optimizer, + optim_state_dict: dict[str, Any] | None = None, + group: dist.ProcessGroup | None = None, + ) -> dict[str, Any]: + """ + Transform the state-dict of an optimizer corresponding to a sharded model. + + The given state-dict can be transformed to one of three types: + 1) full optimizer state_dict, 2) sharded optimizer state_dict, 3) local optimizer state_dict. + + For full optimizer state_dict, all states are unflattened and not sharded. + Rank0 only and CPU only can be specified via :meth:`state_dict_type` to + avoid OOM. + + For sharded optimizer state_dict, all states are unflattened but sharded. + CPU only can be specified via :meth:`state_dict_type` to further save + memory. + + For local state_dict, no transformation will be performed. But a state + will be converted from nn.Tensor to ShardedTensor to represent its sharding + nature (this is not supported yet). + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> from torch.distributed.fsdp import StateDictType + >>> from torch.distributed.fsdp import FullStateDictConfig + >>> from torch.distributed.fsdp import FullOptimStateDictConfig + >>> # Save a checkpoint + >>> model, optim = ... + >>> FSDP.set_state_dict_type( + >>> model, + >>> StateDictType.FULL_STATE_DICT, + >>> FullStateDictConfig(rank0_only=False), + >>> FullOptimStateDictConfig(rank0_only=False), + >>> ) + >>> state_dict = model.state_dict() + >>> optim_state_dict = FSDP.optim_state_dict(model, optim) + >>> save_a_checkpoint(state_dict, optim_state_dict) + >>> # Load a checkpoint + >>> model, optim = ... + >>> state_dict, optim_state_dict = load_a_checkpoint() + >>> FSDP.set_state_dict_type( + >>> model, + >>> StateDictType.FULL_STATE_DICT, + >>> FullStateDictConfig(rank0_only=False), + >>> FullOptimStateDictConfig(rank0_only=False), + >>> ) + >>> model.load_state_dict(state_dict) + >>> optim_state_dict = FSDP.optim_state_dict_to_load( + >>> model, optim, optim_state_dict + >>> ) + >>> optim.load_state_dict(optim_state_dict) + + Args: + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + were passed into the optimizer ``optim``. + optim (torch.optim.Optimizer): Optimizer for ``model`` 's + parameters. + optim_state_dict (Dict[str, Any]): the target optimizer state_dict to + transform. If the value is None, optim.state_dict() will be used. ( + Default: ``None``) + group (dist.ProcessGroup): Model's process group across which parameters + are sharded or ``None`` if using the default process group. ( + Default: ``None``) + + Returns: + Dict[str, Any]: A :class:`dict` containing the optimizer state for + ``model``. The sharding of the optimizer state is based on + ``state_dict_type``. + """ + state_dict_settings = FullyShardedDataParallel.get_state_dict_type(model) + if optim_state_dict is None: + optim_state_dict = optim.state_dict() + return FullyShardedDataParallel._optim_state_dict_impl( + model=model, + optim=optim, + optim_state_dict=optim_state_dict, + optim_input=None, + rank0_only=getattr( + state_dict_settings.optim_state_dict_config, "rank0_only", False + ), + full_state_dict=state_dict_settings.state_dict_type + == StateDictType.FULL_STATE_DICT, + group=group, + cpu_offload=getattr( + state_dict_settings.optim_state_dict_config, "offload_to_cpu", True + ), + _stacklevel=2, + ) + + @staticmethod + def optim_state_dict_to_load( + model: torch.nn.Module, + optim: torch.optim.Optimizer, + optim_state_dict: dict[str, Any], + is_named_optimizer: bool = False, + load_directly: bool = False, + group: dist.ProcessGroup | None = None, + ) -> dict[str, Any]: + """ + Convert an optimizer state-dict so that it can be loaded into the optimizer associated with the FSDP model. + + Given a ``optim_state_dict`` that is transformed through + :meth:`optim_state_dict`, it gets converted to the flattened optimizer + state_dict that can be loaded to ``optim`` which is the optimizer for + ``model``. ``model`` must be sharded by FullyShardedDataParallel. + + >>> # xdoctest: +SKIP("undefined variables") + >>> from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + >>> from torch.distributed.fsdp import StateDictType + >>> from torch.distributed.fsdp import FullStateDictConfig + >>> from torch.distributed.fsdp import FullOptimStateDictConfig + >>> # Save a checkpoint + >>> model, optim = ... + >>> FSDP.set_state_dict_type( + >>> model, + >>> StateDictType.FULL_STATE_DICT, + >>> FullStateDictConfig(rank0_only=False), + >>> FullOptimStateDictConfig(rank0_only=False), + >>> ) + >>> state_dict = model.state_dict() + >>> original_osd = optim.state_dict() + >>> optim_state_dict = FSDP.optim_state_dict( + >>> model, + >>> optim, + >>> optim_state_dict=original_osd + >>> ) + >>> save_a_checkpoint(state_dict, optim_state_dict) + >>> # Load a checkpoint + >>> model, optim = ... + >>> state_dict, optim_state_dict = load_a_checkpoint() + >>> FSDP.set_state_dict_type( + >>> model, + >>> StateDictType.FULL_STATE_DICT, + >>> FullStateDictConfig(rank0_only=False), + >>> FullOptimStateDictConfig(rank0_only=False), + >>> ) + >>> model.load_state_dict(state_dict) + >>> optim_state_dict = FSDP.optim_state_dict_to_load( + >>> model, optim, optim_state_dict + >>> ) + >>> optim.load_state_dict(optim_state_dict) + + Args: + model (torch.nn.Module): Root module (which may or may not be a + :class:`FullyShardedDataParallel` instance) whose parameters + were passed into the optimizer ``optim``. + optim (torch.optim.Optimizer): Optimizer for ``model`` 's + parameters. + optim_state_dict (Dict[str, Any]): The optimizer states to be loaded. + is_named_optimizer (bool): Is this optimizer a NamedOptimizer or + KeyedOptimizer. Only set to True if ``optim`` is TorchRec's + KeyedOptimizer or torch.distributed's NamedOptimizer. + load_directly (bool): If this is set to True, this API will also + call optim.load_state_dict(result) before returning the result. + Otherwise, users are responsible to call ``optim.load_state_dict()`` + (Default: ``False``) + group (dist.ProcessGroup): Model's process group across which parameters + are sharded or ``None`` if using the default process group. ( + Default: ``None``) + """ + state_dict_settings = FullyShardedDataParallel.get_state_dict_type(model) + result = FullyShardedDataParallel._optim_state_dict_to_load_impl( + optim_state_dict=optim_state_dict, + model=model, + optim_input=None, + optim=optim, + full_state_dict=( + state_dict_settings.state_dict_type == StateDictType.FULL_STATE_DICT + ), + rank0_only=getattr( + state_dict_settings.optim_state_dict_config, "rank0_only", False + ), + is_named_optimizer=is_named_optimizer, + group=group, + ) + if load_directly: + optim.load_state_dict(result) + return result + + def register_comm_hook(self, state: object, hook: callable) -> None: + """Register a communication hook. + + This is an enhancement that provides a flexible hook to users where they can specify how FSDP aggregates + gradients across multiple workers. + This hook can be used to implement several algorithms like + `GossipGrad `_ and gradient compression + which involve different communication strategies for + parameter syncs while training with :class:`FullyShardedDataParallel`. + + .. warning :: + FSDP communication hook should be registered before running an initial forward pass + and only once. + + Args: + state (object): Passed to the hook to maintain any state information during the training process. + Examples include error feedback in gradient compression, + peers to communicate with next in `GossipGrad `_, etc. + It is locally stored by each worker + and shared by all the gradient tensors on the worker. + hook (Callable): Callable, which has one of the following signatures: + 1) ``hook: Callable[torch.Tensor] -> None``: + This function takes in a Python tensor, which represents + the full, flattened, unsharded gradient with respect to all variables + corresponding to the model this FSDP unit is wrapping + (that are not wrapped by other FSDP sub-units). + It then performs all necessary processing and returns ``None``; + 2) ``hook: Callable[torch.Tensor, torch.Tensor] -> None``: + This function takes in two Python tensors, the first one represents + the full, flattened, unsharded gradient with respect to all variables + corresponding to the model this FSDP unit is wrapping + (that are not wrapped by other FSDP sub-units). The latter + represents a pre-sized tensor to store a chunk of a sharded gradient after + reduction. + In both cases, callable performs all necessary processing and returns ``None``. + Callables with signature 1 are expected to handle gradient communication for a `NO_SHARD` case. + Callables with signature 2 are expected to handle gradient communication for sharded cases. + + """ + if not self.check_is_root(): + raise AssertionError( + "register_comm_hook can only be called on a root instance." + ) + for fsdp_state in traversal_utils._get_fsdp_states(self): + if fsdp_state.sharding_strategy in HYBRID_SHARDING_STRATEGIES: + raise AssertionError( + f"Communication hook is not supported for hybrid strategies: {fsdp_state.sharding_strategy}" + ) + if fsdp_state._comm_hook is not None: + raise AssertionError("A communication hook is already registered") + if not callable(hook): + raise ValueError( + f"The communication hook must be callable but got {hook}" + ) + fsdp_state._comm_hook = hook + fsdp_state._comm_hook_state = state + + def _unshard(self, async_op: bool = False): + class UnshardHandle: + def __init__( + self, + flat_param_handle: FlatParamHandle | None, + unshard_event: torch.Event, + ): + self._flat_param_handle = flat_param_handle + self._unshard_event = unshard_event + + def wait(self): + if self._flat_param_handle is not None: + current_stream = ( + self._flat_param_handle._device_handle.current_stream() + ) + current_stream.wait_event(self._unshard_event) + self._flat_param_handle = None + + if self._handle: + with self._use_training_state( + TrainingState.FORWARD_BACKWARD, HandleTrainingState.FORWARD + ): + _unshard( + self, self._handle, self._unshard_stream, self._pre_unshard_stream + ) + self._unshard_event = self._unshard_stream.record_event() + self._handle._prefetched = True + unshard_handle = UnshardHandle(self._handle, self._unshard_event) + if async_op: + return unshard_handle + unshard_handle.wait() + return None + + def _wait_unshard_streams_on_current_stream(self): + _wait_for_computation_stream( + self._device_handle.current_stream(), + self._unshard_stream, + self._pre_unshard_stream, + ) + + @contextlib.contextmanager + def _use_training_state( + self, training_state: TrainingState, handle_training_state: HandleTrainingState + ): + prev_training_state = self.training_state + self.training_state = training_state + if self._handle: + prev_handle_training_state = self._handle._training_state + self._handle._training_state = handle_training_state + try: + yield + finally: + self.training_state = prev_training_state + if self._handle: + self._handle._training_state = prev_handle_training_state + + +def _get_grad_norm( + params: Iterable[nn.Parameter], + norm_type: float, + zero: torch.Tensor, + device: torch.device, +) -> torch.Tensor: + """ + Return the gradient norm of parameters ``param`` s, where the gradients are viewed as a single vector. + + The returned norm is in FP32 even if parameters/gradients are in a low precision. This is because the downstream + use of this return value is a reduction across ranks. + """ + params_with_grad = [param for param in params if param.grad is not None] + if len(params_with_grad) == 0: + # Reuse a tensor for zero to avoid a GPU sync + return zero + grads = [param.grad for param in params_with_grad] + # Compute the gradient norm in FP32, where we treat the gradients as a + # single vector. This naturally handles mixed-dtype gradients since we + # cast each gradient to FP32 during the norm computation. + grad_norm = torch.linalg.vector_norm( + torch.stack( + [ + torch.linalg.vector_norm(grad.detach(), norm_type, dtype=torch.float32) + for grad in grads + ], + ), + norm_type, + dtype=torch.float32, + ) + return grad_norm.to(device=device) + + +def _get_param_to_fqn( + model: torch.nn.Module, +) -> dict[torch.nn.Parameter, str]: + """ + Construct a mapping from parameters to their parameter names. + + The ``model`` should not contain any :class:`FullyShardedDataParallel` instances, which + means that none of the parameters should be ``FlatParameter`` s. As a + result, compared to :meth:`_get_param_to_fqns`, the mapped + values may be flattened from singleton :class:`list` s to the contained + names themselves. + + Args: + model (torch.nn.Module): Root module, which should not contain any + :class:`FullyShardedDataParallel` instances. + """ + param_to_param_names = _get_param_to_fqns(model) + for param_names in param_to_param_names.values(): + if len(param_names) == 0: + raise AssertionError( + "`_get_param_to_fqns()` should not construct empty lists" + ) + if len(param_names) > 1: + raise RuntimeError( + "Each parameter should only map to one parameter name but got " + f"{len(param_names)}: {param_names}" + ) + param_to_param_name = { + param: param_names[0] for param, param_names in param_to_param_names.items() + } + return param_to_param_name + + +def _get_fqn_to_param( + model: torch.nn.Module, +) -> dict[str, torch.nn.Parameter]: + """Construct the inverse mapping of :meth:`_get_param_to_fqn`.""" + param_to_param_name = _get_param_to_fqn(model) + return dict(zip(param_to_param_name.values(), param_to_param_name.keys())) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/sharded_grad_scaler.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/sharded_grad_scaler.py new file mode 100644 index 0000000000000000000000000000000000000000..4722d5260572985e51955fd6f724b38211f461b9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/sharded_grad_scaler.py @@ -0,0 +1,377 @@ +# mypy: allow-untyped-defs +import logging +from collections import abc, defaultdict +from collections.abc import Iterable +from typing import Any, overload + +import torch +import torch.distributed as dist +from torch.amp.grad_scaler import _MultiDeviceReplicator, GradScaler, OptState +from torch.distributed.distributed_c10d import ProcessGroup + + +logger = logging.getLogger(__name__) + + +def _refresh_per_optimizer_state() -> dict[str, Any]: + return {"stage": OptState.READY, "found_inf_per_device": {}} + + +def _is_supported_device(tensor: torch.Tensor) -> bool: + return tensor.is_cuda or tensor.device.type in ( + "xla", + "cpu", + "hpu", + "mtia", + "xpu", + torch._C._get_privateuse1_backend_name(), + ) + + +class _GeneralMultiDeviceReplicator(_MultiDeviceReplicator): + """ + Lazily serves tensor to request device. This class extends + _MultiDeviceReplicator to allow support for "cpu" as a device. + """ + + def __init__(self, master_tensor: torch.Tensor) -> None: + if not _is_supported_device(master_tensor): + raise AssertionError( + f"Expected supported device, got {master_tensor.device}" + ) + self.master = master_tensor + self._per_device_tensors: dict[torch.device, torch.Tensor] = {} + + +class ShardedGradScaler(GradScaler): + """ + ShardedGradScaler helps perform gradient scaling in a shard aware manner. It extends + functionality from GradScaler: + * Supports PyTorch DDP and FSDP implementations + * Support CPU offloaded tensors (as used in fully sharded data parallel[FSDP]) + * Supports the custom Mixed Precision loss dtype (fp16, bf16) that FSDP returns + * Sync inf/nan for scaled gradient tensors on any torch.device (where tensors are placed) across + nodes + + Example:: + + # Creates a ShardedGradScaler once at the beginning of training. + scaler = ShardedGradScaler() + + for epoch in epochs: + for input, target in data: + optimizer.zero_grad() + output = model(input) + loss = loss_fn(output, target) + + # Scales loss. Calls backward() on scaled loss to create scaled gradients. + scaler.scale(loss).backward() + + # scaler.step() first unscales gradients of the optimizer's params. + # If gradients don't contain infs/NaNs, optimizer.step() is then called, + # otherwise, optimizer.step() is skipped. + scaler.step(optimizer) + + # Updates the scale for next iteration. + scaler.update() + + See :class:`GradScaler` for explanation of scaling/unscaling and more use cases. + + Args: + init_scale (float, optional, default=2.**16): Initial scale factor. + growth_factor (float, optional, default=2.0): Factor by which the scale is multiplied during + :meth:`update` if no inf/NaN gradients occur for ``growth_interval`` consecutive iterations. + backoff_factor (float, optional, default=0.5): Factor by which the scale is multiplied during + :meth:`update` if inf/NaN gradients occur in an iteration. + growth_interval (int, optional, default=2000): Number of consecutive iterations without inf/NaN gradients + that must occur for the scale to be multiplied by ``growth_factor``. + enabled (bool, optional): If ``False``, disables gradient scaling. :meth:`step` simply + invokes the underlying ``optimizer.step()``, and other methods become no-ops. + Default: ``True`` + process_group (ProcessGroup, optional, default=torch.distributed.group.WORLD): + process group for sharding + """ + + def __init__( + self, + device: str = "cuda", + init_scale: float = 2.0**16, + backoff_factor: float = 0.5, + growth_factor: float = 2.0, + growth_interval: int = 2000, + enabled: bool = True, + process_group: ProcessGroup | None = dist.group.WORLD, + ) -> None: + super().__init__( + device, + init_scale=init_scale, + backoff_factor=backoff_factor, + growth_factor=growth_factor, + growth_interval=growth_interval, + enabled=enabled, + ) + if self._enabled: + self.process_group = process_group + self._per_optimizer_states = defaultdict(_refresh_per_optimizer_state) + + @overload + def scale(self, outputs: torch.Tensor) -> torch.Tensor: ... + + @overload + def scale(self, outputs: list[torch.Tensor]) -> list[torch.Tensor]: ... + + @overload + def scale(self, outputs: tuple[torch.Tensor, ...]) -> tuple[torch.Tensor, ...]: ... + + @overload + def scale(self, outputs: Iterable[torch.Tensor]) -> Iterable[torch.Tensor]: ... + + def scale( + self, outputs: torch.Tensor | Iterable[torch.Tensor] + ) -> torch.Tensor | Iterable[torch.Tensor]: + if not self._enabled: + return outputs + + if isinstance(outputs, torch.Tensor): + if not _is_supported_device(outputs): + raise AssertionError(f"Expected supported device, got {outputs.device}") + if self._scale is None: + self._lazy_init_scale_growth_tracker(outputs.device) + if self._scale is None: + raise AssertionError("Expected _scale to be initialized, got None") + scaled_output = outputs * self._scale.to( + device=outputs.device, non_blocking=True + ) + # Here we ensure the return dtype is the same as the outputs dtype. + # For the FSDP + Mixed Precision use case, the loss output is in the Mixed Precision + # format (fp16, bf16) and so the scaled loss should be of the same dtype. + return scaled_output.type(outputs.dtype) + + stash: list[_GeneralMultiDeviceReplicator] = [] + + def apply_scale(val: torch.Tensor | Iterable[torch.Tensor]): + if isinstance(val, torch.Tensor): + if not _is_supported_device(val): + raise AssertionError(f"Expected supported device, got {val.device}") + if len(stash) == 0: + if self._scale is None: + self._lazy_init_scale_growth_tracker(val.device) + if self._scale is None: + raise AssertionError( + "Expected _scale to be initialized, got None" + ) + stash.append(_GeneralMultiDeviceReplicator(self._scale)) + scaled_val = val * stash[0].get(val.device) + # Here we ensure the return dtype is the same as the outputs dtype. + # For the FSDP + Mixed Precision use case, the loss output is in the Mixed Precision + # format (fp16, bf16) and so the scaled loss should be of the same dtype. + return scaled_val.type(val.dtype) + if isinstance(val, abc.Iterable): + iterator = map(apply_scale, val) + if isinstance(val, (list, tuple)): + return type(val)(iterator) + return iterator + raise ValueError("outputs must be a Tensor or an iterable of Tensors") + + return apply_scale(outputs) + + def _unscale_grads_( + self, + optimizer: torch.optim.Optimizer, + inv_scale: torch.Tensor, + found_inf: torch.Tensor, + allow_fp16: bool = True, + ) -> dict[torch.device, torch.Tensor]: + per_device_inv_scale = _GeneralMultiDeviceReplicator(inv_scale) + per_device_found_inf = _GeneralMultiDeviceReplicator(found_inf) + + # To set up _amp_foreach_non_finite_check_and_unscale_, split grads by device and dtype. + # There could be thousands of grads, so we'd like to iterate through them just once. + # However, we don't know their devices or dtypes in advance. + + # https://stackoverflow.com/questions/5029934/defaultdict-of-defaultdict + # Google says mypy struggles with defaultdicts type annotations. + per_device_and_dtype_grads = defaultdict(lambda: defaultdict(list)) # type: ignore[var-annotated] + with torch.no_grad(): + for group in optimizer.param_groups: + for param in group["params"]: + if param.grad is None: + continue + if (not allow_fp16) and param.grad.dtype == torch.float16: + raise ValueError("Attempting to unscale FP16 gradients.") + if param.grad.is_sparse: + # is_coalesced() == False means the sparse grad has values with duplicate indices. + # coalesce() deduplicates indices and adds all values that have the same index. + # For scaled fp16 values, there's a good chance coalescing will cause overflow, + # so we should check the coalesced _values(). + if param.grad.dtype is torch.float16: + # coalesce is not supported in torch.float16 + param_grad_fp32 = param.grad.type(torch.float32).coalesce() + param.grad = param_grad_fp32.type(torch.float16) + to_unscale = param.grad._values() + else: + to_unscale = param.grad + + per_device_and_dtype_grads[to_unscale.device][ + to_unscale.dtype + ].append(to_unscale) + + for device, per_dtype_grads in per_device_and_dtype_grads.items(): + for grads in per_dtype_grads.values(): + torch._amp_foreach_non_finite_check_and_unscale_( + grads, + per_device_found_inf.get(device), + per_device_inv_scale.get(device), + ) + # There exist contexts (e.g. w/ `use_orig_params=True`) wherein some + # ranks may have no (non-zero sized) parameter shards, necessitating the + # initialization of `per_device_found_inf._per_device_tensors` here + if not per_device_found_inf._per_device_tensors: + if self._scale is None: + raise AssertionError("Expected _scale to be initialized, got None") + per_device_found_inf.get(self._scale.device) + return per_device_found_inf._per_device_tensors + + def unscale_(self, optimizer: torch.optim.Optimizer) -> None: + if not self._enabled: + return + + self._check_scale_growth_tracker("unscale_") + + optimizer_state = self._per_optimizer_states[id(optimizer)] + + if optimizer_state["stage"] is OptState.UNSCALED: + raise RuntimeError( + "unscale_() has already been called on this optimizer since the last update()." + ) + elif optimizer_state["stage"] is OptState.STEPPED: + raise RuntimeError("unscale_() is being called after step().") + + # FP32 division can be imprecise for certain compile options, so we carry out the reciprocal in FP64. + if self._scale is None: + raise AssertionError("Expected _scale to be initialized, got None") + inv_scale = self._scale.double().reciprocal().float() + found_inf = torch.full( + (1,), 0.0, dtype=torch.float32, device=self._scale.device + ) + + optimizer_state["found_inf_per_device"] = self._unscale_grads_( + optimizer, inv_scale, found_inf, True + ) + optimizer_state["stage"] = OptState.UNSCALED + + # Synchronize the detected inf across the ranks + optimizer_state = self._per_optimizer_states[id(optimizer)] + works = [] + found_inf_on_cpus = [] + found_inf_on_devices = [] + + for found_inf in optimizer_state["found_inf_per_device"].values(): + if self._device != "cpu" and found_inf.device.type == "cpu": + found_inf_on_cpus.append(found_inf) + found_inf_on_device = found_inf.to(self._device) + found_inf_on_devices.append(found_inf_on_device) + works.append( + dist.all_reduce( + found_inf_on_device, async_op=True, group=self.process_group + ) + ) + else: + works.append( + dist.all_reduce(found_inf, async_op=True, group=self.process_group) + ) + for work in works: + work.wait() + if found_inf_on_cpus: + torch._foreach_copy_(found_inf_on_cpus, found_inf_on_devices) + + def _amp_update_scale_cpu_(self, found_inf: torch.Tensor) -> None: + """ + If found_inf is 1.0 (True), then scale is multiplied by backoff_factor and growth_tracker is set to zero. + Otherwise, scale is multiplied by the growth factor when the growth interval is reached. + """ + if self._scale is None or self._growth_tracker is None: + raise AssertionError( + "Expected _scale and _growth_tracker to be initialized, got None" + ) + + if found_inf.item() >= 1.0: + self._scale *= self._backoff_factor + self._growth_tracker.fill_(0) + else: + successful = self._growth_tracker + 1 + if successful == self._growth_interval: + self._scale *= self._growth_factor + self._growth_tracker.fill_(0) + else: + self._growth_tracker = successful + + def update(self, new_scale: float | torch.Tensor | None = None) -> None: + """ + Updates the scale factor. + If any optimizer steps were skipped the scale is multiplied by ``backoff_factor`` + to reduce it. If ``growth_interval`` unskipped iterations occurred consecutively, + the scale is multiplied by ``growth_factor`` to increase it. + Passing ``new_scale`` sets the new scale value manually. (``new_scale`` is not + used directly, it's used to fill GradScaler's internal scale tensor. So if + ``new_scale`` was a tensor, later in-place changes to that tensor will not further + affect the scale GradScaler uses internally.) + Args: + new_scale (float or :class:`torch.Tensor`, optional, default=None): New scale factor. + .. warning:: + :meth:`update` should only be called at the end of the iteration, after ``scaler.step(optimizer)`` has + been invoked for all optimizers used this iteration. + """ + + if not self._enabled: + return + + _scale, _growth_tracker = self._check_scale_growth_tracker("update") # type: ignore[var-annotated] + + if new_scale is not None: + # Accept a new user-defined scale. + if isinstance(new_scale, float): + self._scale.fill_(new_scale) # type: ignore[union-attr] + else: + reason = ( + "new_scale should be a float or a 1-element torch.cuda.FloatTensor or " + "torch.FloatTensor with requires_grad=False." + ) + if new_scale.device.type != self._device: + raise AssertionError(reason) + if new_scale.numel() != 1: + raise AssertionError(reason) + if new_scale.requires_grad is not False: + raise AssertionError(reason) + self._scale.copy_(new_scale) # type: ignore[union-attr] + else: + # Consume shared inf/nan data collected from optimizers to update the scale. + # If all found_inf tensors are on the same device as self._scale, this operation is asynchronous. + found_infs = [ + found_inf.to(device=_scale.device, non_blocking=True) + for state in self._per_optimizer_states.values() + for found_inf in state["found_inf_per_device"].values() + ] + + if len(found_infs) == 0: + raise AssertionError("No inf checks were recorded prior to update.") + + found_inf_combined = found_infs[0] + if len(found_infs) > 1: + for i in range(1, len(found_infs)): + found_inf_combined += found_infs[i] + + if _scale.device.type == "cpu": + self._amp_update_scale_cpu_(found_inf_combined) + else: + torch._amp_update_scale_( + self._scale, # type: ignore[arg-type] + self._growth_tracker, # type: ignore[arg-type] + found_inf_combined, + self._growth_factor, # type: ignore[arg-type] + self._backoff_factor, # type: ignore[arg-type] + self._growth_interval, # type: ignore[arg-type] + ) + + # To prepare for next iteration, clear the data collected from optimizers this iteration. + self._per_optimizer_states = defaultdict(_refresh_per_optimizer_state) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/wrap.py b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/wrap.py new file mode 100644 index 0000000000000000000000000000000000000000..35efa507994eaddc8b9369859dc6e15f3554e1b6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/fsdp/wrap.py @@ -0,0 +1,608 @@ +# mypy: allow-untyped-defs +# Copyright (c) Facebook, Inc. and its affiliates. +# +# This source code is licensed under the BSD license found in the +# LICENSE file in the root directory of this source tree. + +import contextlib +import copy +from abc import ABC, abstractmethod +from collections.abc import Callable, Generator, Iterable, Sequence +from typing import Any, cast + +import torch.nn as nn + + +__all__ = [ + "always_wrap_policy", + "lambda_auto_wrap_policy", + "transformer_auto_wrap_policy", + "size_based_auto_wrap_policy", + "enable_wrap", + "wrap", + "CustomPolicy", + "ModuleWrapPolicy", +] + + +# NOTE: We intentionally keep this function simple and isolate the complexity +# to `fn` to enable using this function generically. We may move this to a +# non-FSDP-specific folder and/or make it public in the future. +def _post_order_apply( + root_module: nn.Module, + fn: Callable[[nn.Module], nn.Module | None], +): + """ + This applies ``fn`` to every module in the module tree of ``root_module`` + following a post-order traversal. If ``fn`` returns an :class:`nn.Module`, + then this replaces the original module with the newly returned one in the + tree. Otherwise, ``fn`` should return ``None``, in which case the module is + not changed. + """ + # Track visited modules to avoid visiting shared modules multiple times + visited_modules: set[nn.Module] = {root_module} + + def _post_order_apply_inner( + module: nn.Module, + module_name: str, + parent_module: nn.Module | None, + ): + for child_module_name, child_module in module.named_children(): + if child_module not in visited_modules: + visited_modules.add(child_module) + _post_order_apply_inner(child_module, child_module_name, module) + optional_module = fn(module) + if optional_module is not None: + if not isinstance(parent_module, nn.Module): + raise AssertionError( + "Non-root modules should have their parent module set but got " + f"{parent_module} for {module}" + ) + if not module_name: + raise AssertionError( + "Non-root modules should have their module name set but got " + f"an empty module name for {module}" + ) + if not isinstance(optional_module, nn.Module): + raise AssertionError( + f"fn should return None or an nn.Module but got {optional_module}" + ) + setattr(parent_module, module_name, optional_module) + + _post_order_apply_inner(root_module, "", None) + + +def _construct_wrap_fn( + root_module: nn.Module, + target_module_to_kwargs: dict[nn.Module, dict[str, Any]], + fsdp_fn: Callable, +) -> Callable[[nn.Module], nn.Module | None]: + """ + This constructs the "wrap" function to pass to :func:`_post_order_apply` + based on ``target_module_to_kwargs``, which should be constructed from the + wrapping policy. + """ + + def fn(module: nn.Module) -> nn.Module | None: + # Explicitly avoid wrapping the root module since for FSDP, it is + # handled by the caller + if module in target_module_to_kwargs and module is not root_module: + kwargs = target_module_to_kwargs[module] + return fsdp_fn(module, **kwargs) + return None + + return fn + + +def _run_mixed_precision_override_policy( + root_module: nn.Module, + module_classes: Iterable[type[nn.Module]], + ignored_modules: set[nn.Module], + root_kwargs: dict[str, Any], + target_module_to_kwargs: dict[nn.Module, dict[str, Any]], +): + module_classes_tuple = tuple(set(module_classes)) + for module in root_module.modules(): + if module in ignored_modules: + continue + elif isinstance(module, module_classes_tuple): + # This policy overrides any existing policy + if module not in target_module_to_kwargs: + # Only inherit from the root kwargs if not already specified + target_module_to_kwargs[module] = root_kwargs + target_module_to_kwargs[module]["mixed_precision"] = None + return target_module_to_kwargs + + +def always_wrap_policy(*args, **kwargs) -> bool: + """ + A simple recursive wrap policy that always returns ``True``. This means + that every submodule is wrapped by the wrapper class in + :func:`_recursive_wrap`. + """ + return True + + +class _Policy(ABC): + """ + This defines an abstract base class that represents a policy for applying + a module-level API. + """ + + @abstractmethod + def _run_policy( + self, + root_module: nn.Module, + ignored_modules: set[nn.Module], + root_kwargs: dict[str, Any], + ) -> dict[nn.Module, dict[str, Any]]: + """ + This should return a dict ``target_module_to_kwargs`` that maps from + each target module to wrap to its kwargs. + """ + ... + + +def _module_wrap_policy( + module: nn.Module, + recurse: bool, + nonwrapped_numel: int, + module_classes: set[type[nn.Module]], +) -> bool: + """ + This auto wrap policy wraps every module that is an instance of any type in + ``module_classes`` as its own FSDP instance. The root module given by + ``module`` is always wrapped as an FSDP instance regardless. Since the + wrapping proceeds bottom up, each FSDP instance manages the parameters in + its subtree excluding any already managed by a child FSDP instance. + + Args: + module (nn.Module): Current module being considered. + recurse (bool): If ``False``, then this function must decide whether + ``module`` should be wrapped as an FSDP instance or not. If + ``True``, then the function is still recursing down the module + tree as a part of the DFS. + nonwrapped_numel (int): Parameter numel not yet wrapped. + module_classes (Set[Type[nn.Module]]): Set of module classes that are + wrapped as FSDP instances. + + Returns: + ``True`` if ``recurse=True``, and whether ``module`` should be wrapped + if ``recurse=False``. + """ + if recurse: + return True # always recurse + return isinstance(module, tuple(module_classes)) + + +class ModuleWrapPolicy(_Policy): + """ + This policy applies to every module of the specified module classes, + passing in the kwargs given to the root. + """ + + def __init__(self, module_classes: Iterable[type[nn.Module]]): + module_classes_set = set(module_classes) + self._module_classes = module_classes_set + self._module_classes_str = str(module_classes_set) + + def _run_policy( + self, + root_module: nn.Module, + ignored_modules: set[nn.Module], + root_kwargs: dict[str, Any], + ) -> dict[nn.Module, dict[str, Any]]: + module_classes = tuple(self._module_classes) + target_module_to_kwargs: dict[nn.Module, dict[str, Any]] = {} + for module in root_module.modules(): + if module in ignored_modules: + continue + elif isinstance(module, module_classes): + # Shallow copy to avoid coupling changes across modules + target_module_to_kwargs[module] = copy.copy(root_kwargs) + return target_module_to_kwargs + + def __call__(self, module, recurse, *args, **kwargs): + # nonwrapped_numel is not used. + return _module_wrap_policy( + module, recurse, nonwrapped_numel=-1, module_classes=self._module_classes + ) + + def __repr__(self) -> str: + return super().__repr__() + f"({self._module_classes_str})" + + +class CustomPolicy(_Policy): + """ + This policy takes in a lambda function that maps a given ``nn.Module`` to + either ``False``, ``True``, or a kwarg dictionary. + - If the function returns ``False`` or an empty dictionary, then the module + does not have the API applied. + - If the function returns ``True``, then the module has the API applied + with the root's kwargs. + - If the function returns a non-empty dictionary, then the module has the + API applied, and the dictionary overrides the root's kwargs. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> model = init_transformer_model(...) + >>> def lambda_fn(module: nn.Module): + >>> if module is model.lm_head: + >>> return {"sharding_strategy": ShardingStrategy.SHARD_GRAD_OP} + >>> elif isinstance(module, TransformerBlock): + >>> return True + >>> return False + >>> policy = CustomPolicy(lambda_fn) + >>> fsdp_model = FSDP(model, auto_wrap_policy=policy) + """ + + def __init__(self, lambda_fn: Callable[[nn.Module], bool | dict[str, Any]]): + self._lambda_fn = lambda_fn + + def _run_policy( + self, + root_module: nn.Module, + ignored_modules: set[nn.Module], + root_kwargs: dict[str, Any], + ) -> dict[nn.Module, dict[str, Any]]: + target_module_to_kwargs: dict[nn.Module, dict[str, Any]] = {} + for module in root_module.modules(): + if module in ignored_modules: + continue + res = self._lambda_fn(module) + if not isinstance(res, (dict, bool)): + raise ValueError( + "The lambda_fn passed to CustomPolicy should return " + f"False/True or a kwarg dict, but it returned {res}" + ) + if not res: + continue + kwargs = copy.copy(root_kwargs) + if isinstance(res, dict): + # Override the root kwargs with the ones specified by the + # lambda function + kwargs.update(res) + target_module_to_kwargs[module] = kwargs + return target_module_to_kwargs + + +def lambda_auto_wrap_policy( + module: nn.Module, recurse: bool, nonwrapped_numel: int, lambda_fn: Callable +) -> bool: + """ + A convenient auto wrap policy to wrap submodules based on an arbitrary user + function. If `lambda_fn(submodule) == True``, the submodule will be wrapped as + a `wrapper_cls` unit. + + Return if a module should be wrapped during auto wrapping. + + The first three parameters are required by :func:`_recursive_wrap`. + + Args: + module (nn.Module): Current module being considered. + recurse (bool): If ``False``, then this function must decide whether + ``module`` should be wrapped as an FSDP instance or not. If + ``True``, then the function is still recursing down the module + tree as a part of the DFS. + nonwrapped_numel (int): Parameter numel not yet wrapped. + + lambda_fn (Callable[[nn.Module], bool]): If this returns ``True``, then + this module will be wrapped. + """ + if recurse: + return True # always recurse + return lambda_fn(module) + + +def transformer_auto_wrap_policy( + module: nn.Module, + recurse: bool, + nonwrapped_numel: int, + transformer_layer_cls: set[type[nn.Module]], +) -> bool: + """ + See :func:`_module_wrap_policy`, where ``transformer_layer_cls`` is the + same as ``module_classes``. Note that shared parameters must be wrapped in + the same FSDP instance, so this auto wrap policy can help wrap shared + embeddings into the same FSDP instance for transformer models. + """ + return _module_wrap_policy(module, recurse, nonwrapped_numel, transformer_layer_cls) + + +def _wrap_module_cls_individually( + module: nn.Module, module_classes: Sequence[type], recurse: bool, *args, **kwargs +): + if recurse: + # always recurse + return True + else: + # if not recursing, decide whether we should wrap based on whether the type of module + # is in `module_classes`. + return isinstance(module, tuple(module_classes)) + + +def _or_policy( + module: nn.Module, + recurse: bool, + nonwrapped_numel: int, + policies, +) -> bool: + """ + A policy that wraps ``module`` if any policy in the passed in iterable of + ``policies`` returns ``True``. + """ + return any( + policy(module=module, recurse=recurse, nonwrapped_numel=nonwrapped_numel) + for policy in policies + ) + + +def size_based_auto_wrap_policy( + module: nn.Module, + recurse: bool, + nonwrapped_numel: int, + # Additional custom arguments + min_num_params: int = int(1e8), + force_leaf_modules: set[type[nn.Module]] | None = None, + exclude_wrap_modules: set[type[nn.Module]] | None = None, +) -> bool: + """ + A size-based auto wrap policy. + + Args: + module (nn.Module): Current module being considered. + recurse (bool): If ``False``, then this function must decide whether + ``module`` should be wrapped as an FSDP instance or not. If + ``True``, then the function is still recursing down the module + tree as a part of the DFS. + nonwrapped_numel (int): Parameter numel not yet wrapped. + + min_num_params (int): Customizable policy input that controls the size + threshold over which a module is ready to be wrapped. This is in + units of numel. + force_leaf_modules (Optional[set[type[nn.Module]]]): Set of module types to keep + as leaves, i.e. their children will never be wrapped. + exclude_wrap_modules (Optional[set[type[nn.Module]]]): Set of module types to be + excluded in wrapping. + + Returns: + Whether ``module`` should be wrapped. + """ + force_leaf_modules = ( + size_based_auto_wrap_policy.FORCE_LEAF_MODULES # type: ignore[attr-defined] + if force_leaf_modules is None + else force_leaf_modules + ) + exclude_wrap_modules = ( + size_based_auto_wrap_policy.EXCLUDE_WRAP_MODULES # type: ignore[attr-defined] + if exclude_wrap_modules is None + else exclude_wrap_modules + ) + + # Keep the argument `min_num_params` for BC for now, but it represents the + # minimum non-wrapped *numel* before triggering a wrapping + min_nonwrapped_numel = min_num_params + is_large = nonwrapped_numel >= min_nonwrapped_numel + if recurse: + # We should recurse if the module is big enough but not in force_leaf_modules list. + return is_large and not isinstance(module, tuple(force_leaf_modules)) + else: + # If we are not recursing, determine if we should wrap. + return is_large and not isinstance(module, tuple(exclude_wrap_modules)) + + +# Set those defaults to the size_based_auto_wrap_policy function. Make them easy to be imported. +size_based_auto_wrap_policy.EXCLUDE_WRAP_MODULES = {nn.ModuleList, nn.ModuleDict} # type: ignore[attr-defined] +size_based_auto_wrap_policy.FORCE_LEAF_MODULES = {nn.MultiheadAttention} # type: ignore[attr-defined] + + +@contextlib.contextmanager +def enable_wrap( + *, wrapper_cls: Any, **wrapper_kwargs: Any +) -> Generator[None, None, None]: + """ + Context manager to wrap modules using a wrapper. + + Useful for when you'd like to apply the same configuration arguments to all + child modules that you wrap. A particularly important use case is wrapping + large layers so that they get sharded (in-place) during initialization, to + avoid running out of system memory. Large layers can indicate that they + should be sharded via the ``wrap`` annotation and this context manager can + provide the exact configuration for these nested instances. + + Usage:: + + with enable_wrap(wrapper_cls, **params): + # Wraps layer in FSDP by default if within context + self.l1 = wrap(torch.nn.Linear(5, 5)) + + Args: + wrapper_cls: + Class that `wrap` annotation will `wrap` modules with, such as + `FullyShardedDataParallel`. + **wrapper_kwargs: + Configuration settings that will be passed to all ``wrap`` + instances inside the context + """ + kwargs = { + "wrapper_cls": wrapper_cls, + **wrapper_kwargs, + } + with _ConfigAutoWrap(**kwargs): + yield + + +def wrap(module: nn.Module, **wrap_overrides: Any) -> nn.Module: + """ + Annotate that a module should be wrapped. Annotated modules will only be + wrapped if inside of an :func:`enable_wrap` context manager. This allows + a module to be initialized both with and without a wrapper without code + change. + + The class that this function wraps the passed in ``nn.Module`` with is the + passed in ``wrapper_cls`` argument into ``enable_wrap``. Both + ``enable_wrap`` and ``wrap`` can take in kwargs specifying how to construct + the ``wrapper_cls`` instance. In the case of duplicate kwargs in + ``enable_wrap`` and ``wrap``, the argument passed into ``wrap`` will be + respected. + + Usage:: + + with enable_wrap(wrapper_cls=FSDP, **fsdp_config): + # Wraps layer in FSDP by default if within context + self.l1 = wrap(torch.nn.Linear(5, 5)) + + Args: + module (nn.Module): module to wrap (if in :func:`enable_wrap` context) + **wrap_overrides: configuration overrides that will take priority over + the values provided by the :func:`enable_wrap` context + """ + if _ConfigAutoWrap.in_autowrap_context: + if _ConfigAutoWrap.wrapper_cls is None: + raise AssertionError("Expected _ConfigAutoWrap.wrapper_cls to be set") + + wrap_overrides = {**_ConfigAutoWrap.kwargs, **wrap_overrides} + return _wrap( + module, + _ConfigAutoWrap.wrapper_cls, + **wrap_overrides, + ) + return module + + +def _wrap(module: nn.Module, wrapper_cls: Callable, **kwargs) -> nn.Module: + if wrapper_cls is None: + raise AssertionError("Expected wrapper_cls to be set") + if hasattr(module, "_wrap_overrides"): + # If module has a _wrap_overrides attribute, we force overriding the + # FSDP config with these attributes for this module. Currently this + # is only used to disable mixed precision for BatchNorm when + # auto_wrapping. + overrides = {**kwargs, **module._wrap_overrides} # type: ignore[arg-type, dict-item] + return wrapper_cls(module, **overrides) + + return wrapper_cls(module, **kwargs) + + +def _recursive_wrap( + module: nn.Module, + auto_wrap_policy: Callable, + wrapper_cls: Callable, + ignored_modules: set[nn.Module], + ignored_params: set[nn.Parameter], + only_wrap_children: bool = False, + **kwargs: Any, +) -> tuple[nn.Module, int]: + """ + Wraps submodules of ``module`` for which ``auto_wrap_policy`` returns + ``True`` with ``wrapper_cls``. + + Args: + module (nn.Module): Module to recursively wrap. + auto_wrap_policy (Callable): A callable representing a policy that + determines which modules to recursively wrap with ``wrapper_cls``. + ignored_modules (set[torch.nn.Module]): Modules to ignore when + wrapping. + ignored_params (set[torch.nn.Parameter]): Parameters to ignore when + wrapping; these should be the parameters contained in the modules + in ``ignored_modules``. + Returns: + (nn.Module, int): + ``module`` after wrapping and the numel recursively wrapped. + """ + if auto_wrap_policy is None: + raise AssertionError("Must specify auto_wrap_policy.") + if wrapper_cls is None: + raise AssertionError("Must specify wrapper_cls") + # Make sure no child is already wrapped. + for _, child in module.named_modules(): + if child in ignored_modules: + continue + try: + if isinstance(child, cast(type, wrapper_cls)): + raise AssertionError( + f"Child module {child} is already wrapped by {wrapper_cls}" + ) + except TypeError: + # wrapper_cls is a function as opposed to a class type, just bypass above check. + pass + + # We count all params, assuming none of them are already wrapped. + nonwrapped_numel = sum( + p.numel() for p in module.parameters() if p not in ignored_params + ) + + if auto_wrap_policy is None: + raise AssertionError("Expected auto_wrap_policy to be set") + if auto_wrap_policy(module=module, recurse=True, nonwrapped_numel=nonwrapped_numel): + total_wrapped_numel = 0 + # Iterate through the children, recursively wrap if necessary + for name, child in module.named_children(): + if child in ignored_modules: + continue + wrapped_child, num_wrapped_params = _recursive_wrap( + module=child, + auto_wrap_policy=auto_wrap_policy, + wrapper_cls=wrapper_cls, + ignored_modules=ignored_modules, + ignored_params=ignored_params, + **kwargs, + ) + setattr(module, name, wrapped_child) + # Keep track of how many parameters have been wrapped + total_wrapped_numel += num_wrapped_params + # decide if we need to wrap the current module, + # since the left over parameters exceed the number of params to wrap + remainder = nonwrapped_numel - total_wrapped_numel + if not only_wrap_children and auto_wrap_policy( + module=module, recurse=False, nonwrapped_numel=remainder + ): + # Leaf node or final wrapping of the remainder both happen here. + return _wrap(module, wrapper_cls, **kwargs), nonwrapped_numel + else: + return module, total_wrapped_numel + return module, 0 + + +class _ConfigAutoWrap: + """ + Helper class to wrap modules based on default config args via a context manager. + See :func:`enable_wrap` for more information. + """ + + in_autowrap_context: bool = False # Context flag + wrapper_cls: Callable | None = None # The wrapper class + kwargs: dict[str, Any] = {} # Wrapper's args + + def __init__(self, **kwargs: dict[str, Any]): + self.kwargs = kwargs + + @staticmethod + def enable_autowrap_context(kwargs: Any) -> None: + if _ConfigAutoWrap.in_autowrap_context: + raise NotImplementedError( + "You are already within an autowrap context and we currently do not supported nested autowrap." + ) + _ConfigAutoWrap.in_autowrap_context = True + # Get and save the wrapper cls for the context. + if "wrapper_cls" not in kwargs: + raise AssertionError( + "Expected to pass in wrapper_cls arg into _ConfigAutoWrap." + ) + _ConfigAutoWrap.wrapper_cls = cast(Callable, kwargs["wrapper_cls"]) + del kwargs["wrapper_cls"] + # Save the rest. + _ConfigAutoWrap.kwargs = kwargs + + @staticmethod + def disable_autowrap_context() -> None: + _ConfigAutoWrap.in_autowrap_context = False + _ConfigAutoWrap.wrapper_cls = None + _ConfigAutoWrap.kwargs = {} + + def __enter__(self) -> None: + self.enable_autowrap_context(self.kwargs) + + def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None: + self.disable_autowrap_context() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/launch.py b/.venv/lib/python3.12/site-packages/torch/distributed/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..ad3307c13303d0319af710923669d119b4cff30c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/launch.py @@ -0,0 +1,207 @@ +# mypy: allow-untyped-defs +r""" +Module ``torch.distributed.launch``. + +``torch.distributed.launch`` is a module that spawns up multiple distributed +training processes on each of the training nodes. + +.. warning:: + + This module is going to be deprecated in favor of :ref:`torchrun `. + +The utility can be used for single-node distributed training, in which one or +more processes per node will be spawned. The utility can be used for either +CPU training or GPU training. If the utility is used for GPU training, +each distributed process will be operating on a single GPU. This can achieve +well-improved single-node training performance. It can also be used in +multi-node distributed training, by spawning up multiple processes on each node +for well-improved multi-node distributed training performance as well. +This will especially be beneficial for systems with multiple Infiniband +interfaces that have direct-GPU support, since all of them can be utilized for +aggregated communication bandwidth. + +In both cases of single-node distributed training or multi-node distributed +training, this utility will launch the given number of processes per node +(``--nproc-per-node``). If used for GPU training, this number needs to be less +or equal to the number of GPUs on the current system (``nproc_per_node``), +and each process will be operating on a single GPU from *GPU 0 to +GPU (nproc_per_node - 1)*. + +**How to use this module:** + +1. Single-Node multi-process distributed training + +:: + + python -m torch.distributed.launch --nproc-per-node=NUM_GPUS_YOU_HAVE + YOUR_TRAINING_SCRIPT.py (--arg1 --arg2 --arg3 and all other + arguments of your training script) + +2. Multi-Node multi-process distributed training: (e.g. two nodes) + + +Node 1: *(IP: 192.168.1.1, and has a free port: 1234)* + +:: + + python -m torch.distributed.launch --nproc-per-node=NUM_GPUS_YOU_HAVE + --nnodes=2 --node-rank=0 --master-addr="192.168.1.1" + --master-port=1234 YOUR_TRAINING_SCRIPT.py (--arg1 --arg2 --arg3 + and all other arguments of your training script) + +Node 2: + +:: + + python -m torch.distributed.launch --nproc-per-node=NUM_GPUS_YOU_HAVE + --nnodes=2 --node-rank=1 --master-addr="192.168.1.1" + --master-port=1234 YOUR_TRAINING_SCRIPT.py (--arg1 --arg2 --arg3 + and all other arguments of your training script) + +3. To look up what optional arguments this module offers: + +:: + + python -m torch.distributed.launch --help + + +**Important Notices:** + +1. This utility and multi-process distributed (single-node or +multi-node) GPU training currently only achieves the best performance using +the NCCL distributed backend. Thus NCCL backend is the recommended backend to +use for GPU training. + +2. In your training program, you must parse the command-line argument: +``--local-rank=LOCAL_PROCESS_RANK``, which will be provided by this module. +If your training program uses GPUs, you should ensure that your code only +runs on the GPU device of LOCAL_PROCESS_RANK. This can be done by: + +Parsing the local_rank argument + +:: + + >>> # xdoctest: +SKIP + >>> import argparse + >>> parser = argparse.ArgumentParser() + >>> parser.add_argument("--local-rank", "--local_rank", type=int) + >>> args = parser.parse_args() + +Set your device to local rank using either + +:: + + >>> torch.cuda.set_device(args.local_rank) # before your code runs + +or + +:: + + >>> with torch.cuda.device(args.local_rank): + >>> # your code to run + >>> ... + +.. versionchanged:: 2.0.0 + + The launcher will passes the ``--local-rank=`` argument to your script. + From PyTorch 2.0.0 onwards, the dashed ``--local-rank`` is preferred over the + previously used underscored ``--local_rank``. + + For backward compatibility, it may be necessary for users to handle both + cases in their argument parsing code. This means including both ``"--local-rank"`` + and ``"--local_rank"`` in the argument parser. If only ``"--local_rank"`` is + provided, the launcher will trigger an error: "error: unrecognized arguments: + --local-rank=". For training code that only supports PyTorch 2.0.0+, + including ``"--local-rank"`` should be sufficient. + +3. In your training program, you are supposed to call the following function +at the beginning to start the distributed backend. It is strongly recommended +that ``init_method=env://``. Other init methods (e.g. ``tcp://``) may work, +but ``env://`` is the one that is officially supported by this module. + +:: + + >>> torch.distributed.init_process_group(backend='YOUR BACKEND', + >>> init_method='env://') + +4. In your training program, you can either use regular distributed functions +or use :func:`torch.nn.parallel.DistributedDataParallel` module. If your +training program uses GPUs for training and you would like to use +:func:`torch.nn.parallel.DistributedDataParallel` module, +here is how to configure it. + +:: + + >>> model = torch.nn.parallel.DistributedDataParallel(model, + >>> device_ids=[args.local_rank], + >>> output_device=args.local_rank) + +Please ensure that ``device_ids`` argument is set to be the only GPU device id +that your code will be operating on. This is generally the local rank of the +process. In other words, the ``device_ids`` needs to be ``[args.local_rank]``, +and ``output_device`` needs to be ``args.local_rank`` in order to use this +utility + +5. Another way to pass ``local_rank`` to the subprocesses via environment variable +``LOCAL_RANK``. This behavior is enabled when you launch the script with +``--use-env=True``. You must adjust the subprocess example above to replace +``args.local_rank`` with ``os.environ['LOCAL_RANK']``; the launcher +will not pass ``--local-rank`` when you specify this flag. + +.. warning:: + + ``local_rank`` is NOT globally unique: it is only unique per process + on a machine. Thus, don't use it to decide if you should, e.g., + write to a networked filesystem. See + https://github.com/pytorch/pytorch/issues/12042 for an example of + how things can go wrong if you don't do this correctly. + + + +""" + +from typing_extensions import deprecated as _deprecated + +from torch.distributed.run import get_args_parser, run + + +def parse_args(args): + parser = get_args_parser() + parser.add_argument( + "--use-env", + "--use_env", + default=False, + action="store_true", + help="Use environment variable to pass " + "'local rank'. For legacy reasons, the default value is False. " + "If set to True, the script will not pass " + "--local-rank as argument, and will instead set LOCAL_RANK.", + ) + return parser.parse_args(args) + + +def launch(args): + if args.no_python and not args.use_env: + raise ValueError( + "When using the '--no-python' flag, you must also set the '--use-env' flag." + ) + run(args) + + +@_deprecated( + "The module torch.distributed.launch is deprecated\n" + "and will be removed in future. Use torchrun.\n" + "Note that --use-env is set by default in torchrun.\n" + "If your script expects `--local-rank` argument to be set, please\n" + "change it to read from `os.environ['LOCAL_RANK']` instead. See \n" + "https://pytorch.org/docs/stable/distributed.html#launch-utility for \n" + "further instructions\n", + category=FutureWarning, +) +def main(args=None): + args = parse_args(args) + launch(args) + + +if __name__ == "__main__": + main() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..966c0385c50804eb9ff84fa7b155f87919324289 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__init__.py @@ -0,0 +1,10 @@ +#!/usr/bin/env/python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + + +from torch.distributed.launcher.api import elastic_launch, launch_agent, LaunchConfig diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d697b97305876c42e2f4db14400fe4dbc40e5d9e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6fff5fb9d682624af330ff6ebc8581efbafcbc4f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/launcher/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/api.py new file mode 100644 index 0000000000000000000000000000000000000000..452eb26fc749cc63b25875326768f46b43af0848 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/launcher/api.py @@ -0,0 +1,406 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. +import os +import sys +import uuid +from collections.abc import Callable +from dataclasses import dataclass, field +from typing import Any + +import torch +import torch.distributed.elastic.rendezvous.registry as rdzv_registry +from torch._utils_internal import get_default_numa_options, justknobs_check +from torch.distributed.elastic import events, metrics +from torch.distributed.elastic.agent.server.api import WorkerSpec +from torch.distributed.elastic.agent.server.health_check_server import ( + create_healthcheck_server, + HealthCheckServer, +) +from torch.distributed.elastic.agent.server.local_elastic_agent import ( + _AliveCallbackProxy, + LocalElasticAgent, + TORCHELASTIC_HEALTH_CHECK_PORT, +) +from torch.distributed.elastic.multiprocessing import ( + DefaultLogsSpecs, + LogsSpecs, + SignalException, +) +from torch.distributed.elastic.multiprocessing.errors import ChildFailedError +from torch.distributed.elastic.rendezvous import RendezvousParameters +from torch.distributed.elastic.rendezvous.utils import parse_rendezvous_endpoint +from torch.distributed.elastic.utils.logging import get_logger +from torch.numa.binding import NumaOptions + + +__all__ = ["LaunchConfig", "elastic_launch", "launch_agent"] + +logger = get_logger(__name__) + + +@dataclass +class LaunchConfig: + """ + Creates a rendezvous config. + + Args: + min_nodes: Minimum amount of nodes that the user function will + be launched on. Elastic agent ensures that the user + function start only when the min_nodes amount enters + the rendezvous. + max_nodes: Maximum amount of nodes that the user function + will be launched on. + nproc_per_node: On each node the elastic agent will launch + this amount of workers that will execute user + defined function. + rdzv_backend: rdzv_backend to use in the rendezvous (zeus-adapter, etcd). + rdzv_endpoint: The endpoint of the rdzv sync. storage. + rdzv_configs: Key, value pair that specifies rendezvous specific configuration. + rdzv_timeout: Legacy argument that specifies timeout for the rendezvous. It is going + to be removed in future versions, see the note below. The default timeout is 900 seconds. + run_id: The unique run id of the job (if not passed a unique one will be + deduced from run environment - flow workflow id in flow - or auto generated). + role: User defined role of the worker (defaults to "trainer"). + max_restarts: The maximum amount of restarts that elastic agent will conduct + on workers before failure. + monitor_interval: The interval in seconds that is used by the elastic_agent + as a period of monitoring workers. + start_method: The method is used by the elastic agent to start the + workers (spawn, fork, forkserver). + metrics_cfg: configuration to initialize metrics. + local_addr: address of the local node if any. If not set, a lookup on the local + machine's FQDN will be performed. + local_ranks_filter: ranks for which to show logs in console. If not set, show from all. + event_log_handler: name of the event logging handler as registered in + `elastic/events/handlers.py `_. + duplicate_stdout_filters: If non-empty, duplicates stdout to a file containing only lines + that match _any_ of the filter strings. + duplicate_stderr_filters: If non-empty, duplicates stderr to a file containing only lines + that match _any_ of the filter strings. + virtual_local_rank: Enable virtual local rank mode for workers (defaults to False). + When enabled, LOCAL_RANK is set to 0 for all workers and + CUDA_VISIBLE_DEVICES is adjusted so each worker accesses its + assigned GPU at device index 0. + shutdown_timeout: Time in seconds to wait for graceful shutdown of workers before + sending SIGKILL. Can also be set via TORCH_ELASTIC_SHUTDOWN_TIMEOUT + environment variable. Defaults to 30 seconds. + + + .. note:: + `rdzv_timeout` is a legacy argument that will be removed in future. + Set the timeout via `rdzv_configs['timeout']` + + """ + + min_nodes: int + max_nodes: int + nproc_per_node: int + logs_specs: LogsSpecs | None = None + run_id: str = "" + role: str = "default_role" + rdzv_endpoint: str = "" + rdzv_backend: str = "etcd" + rdzv_configs: dict[str, Any] = field(default_factory=dict) + rdzv_timeout: int = -1 + max_restarts: int = 3 + monitor_interval: float = 0.1 + start_method: str = "spawn" + log_line_prefix_template: str | None = None + metrics_cfg: dict[str, str] = field(default_factory=dict) + local_addr: str | None = None + event_log_handler: str = "null" + numa_options: NumaOptions | None = None + signals_to_handle: str = "SIGTERM,SIGINT,SIGHUP,SIGQUIT" + duplicate_stdout_filters: list[str] | None = None + duplicate_stderr_filters: list[str] | None = None + virtual_local_rank: bool = False + shutdown_timeout: int | None = None + + def __post_init__(self): + default_timeout = 900 + if self.rdzv_timeout != -1: + self.rdzv_configs["timeout"] = self.rdzv_timeout + elif "timeout" not in self.rdzv_configs: + self.rdzv_configs["timeout"] = default_timeout + + # Post-processing to enable refactoring to introduce logs_specs due to non-torchrun API usage + if self.logs_specs is None: + self.logs_specs = DefaultLogsSpecs() + + if self.numa_options is None: + try: + should_use_default_numa = ( + torch.accelerator.is_available() + and torch.accelerator.device_count() == self.nproc_per_node + ) + except RuntimeError: + should_use_default_numa = False + if should_use_default_numa: + self.numa_options = get_default_numa_options() + logger.info("Using default numa options = %r", self.numa_options) + + # Set shutdown_timeout from environment variable if not explicitly set + if self.shutdown_timeout is None: + self.shutdown_timeout = int( + os.environ.get("TORCH_ELASTIC_SHUTDOWN_TIMEOUT", "30") + ) + elif self.shutdown_timeout < 0: + raise ValueError( + f"shutdown_timeout must be non-negative, got {self.shutdown_timeout}" + ) + + +class elastic_launch: + """ + Launches a torchelastic agent on the container that invoked the entrypoint. + + 1. Pass the ``entrypoint`` arguments as non ``kwargs`` (e.g. no named parameters)/ + ``entrypoint`` can be a function or a command. + 2. The return value is a map of each worker's output mapped + by their respective global rank. + + Usage + + :: + + def worker_fn(foo): + # ... + + def main(): + # entrypoint is a function. + outputs = elastic_launch(LaunchConfig, worker_fn)(foo) + # return rank 0's output + return outputs[0] + + # entrypoint is a command and ``script.py`` is the python module. + outputs = elastic_launch(LaunchConfig, "script.py")(args) + outputs = elastic_launch(LaunchConfig, "python")("script.py") + """ + + def __init__( + self, + config: LaunchConfig, + entrypoint: Callable | str | None, + health_check_server: HealthCheckServer | None = None, + ): + self._config = config + self._entrypoint = entrypoint + self._health_check_server = health_check_server + + def __call__(self, *args): + return launch_agent( + self._config, + self._entrypoint, + list(args), + health_check_server=self._health_check_server, + ) + + +def _get_entrypoint_name(entrypoint: Callable | str | None, args: list[Any]) -> str: + """Retrieve entrypoint name with the rule: + 1. If entrypoint is a function, use ``entrypoint.__qualname__``. + 2. If entrypoint is a string, check its value: + 2.1 if entrypoint equals to ``sys.executable`` (like "python"), use the first element from ``args`` + which does not start with hyphen letter (for example, "-u" will be skipped). + 2.2 otherwise, use ``entrypoint`` value. + 3. Otherwise, return empty string. + """ + if isinstance(entrypoint, Callable): # type: ignore[arg-type] + return entrypoint.__name__ # type: ignore[union-attr] + elif isinstance(entrypoint, str): + if entrypoint == sys.executable: + return next((arg for arg in args if arg[0] != "-"), "") + else: + return entrypoint + else: + return "" + + +def _get_addr_and_port( + rdzv_parameters: RendezvousParameters, +) -> tuple[str | None, int | None]: + if rdzv_parameters.backend != "static": + return (None, None) + endpoint = rdzv_parameters.endpoint + endpoint = endpoint.strip() + if not endpoint: + raise ValueError( + "Endpoint is missing in endpoint. Try to add --master-addr and --master-port" + ) + master_addr, master_port = parse_rendezvous_endpoint(endpoint, default_port=-1) + if master_port == -1: + raise ValueError( + f"port is missing in endpoint: {endpoint}. Try to specify --master-port" + ) + return (master_addr, master_port) + + +def launch_agent( + config: LaunchConfig, + entrypoint: Callable | str | None, + args: list[Any], + health_check_server: HealthCheckServer | None = None, +) -> dict[int, Any]: + if not config.run_id: + run_id = str(uuid.uuid4().int) + logger.warning("config has no run_id, generated a random run_id: %s", run_id) + config.run_id = run_id + + entrypoint_name = _get_entrypoint_name(entrypoint, args) + + logger.info( + "Starting elastic_operator with launch configs:\n" + " entrypoint : %(entrypoint)s\n" + " min_nodes : %(min_nodes)s\n" + " max_nodes : %(max_nodes)s\n" + " nproc_per_node : %(nproc_per_node)s\n" + " run_id : %(run_id)s\n" + " rdzv_backend : %(rdzv_backend)s\n" + " rdzv_endpoint : %(rdzv_endpoint)s\n" + " rdzv_configs : %(rdzv_configs)s\n" + " max_restarts : %(max_restarts)s\n" + " monitor_interval : %(monitor_interval)s\n" + " log_dir : %(log_dir)s\n" + " metrics_cfg : %(metrics_cfg)s\n" + " event_log_handler : %(event_log_handler)s\n" + " numa_options : %(numa_options)s\n" + " signals_to_handle : %(signals_to_handle)s\n" + " duplicate_stdout_filters : %(duplicate_stdout_filters)s\n" + " duplicate_stderr_filters : %(duplicate_stderr_filters)s\n", + { + "entrypoint": entrypoint_name, + "min_nodes": config.min_nodes, + "max_nodes": config.max_nodes, + "nproc_per_node": config.nproc_per_node, + "run_id": config.run_id, + "rdzv_backend": config.rdzv_backend, + "rdzv_endpoint": config.rdzv_endpoint, + "rdzv_configs": config.rdzv_configs, + "max_restarts": config.max_restarts, + "monitor_interval": config.monitor_interval, + "log_dir": config.logs_specs.root_log_dir, # type: ignore[union-attr] + "metrics_cfg": config.metrics_cfg, + "event_log_handler": config.event_log_handler, + "numa_options": config.numa_options, + "signals_to_handle": config.signals_to_handle, + "duplicate_stdout_filters": config.duplicate_stdout_filters, + "duplicate_stderr_filters": config.duplicate_stderr_filters, + }, + ) + + rdzv_parameters = RendezvousParameters( + backend=config.rdzv_backend, + endpoint=config.rdzv_endpoint, + run_id=config.run_id, + min_nodes=config.min_nodes, + max_nodes=config.max_nodes, + local_addr=config.local_addr, + **config.rdzv_configs, + ) + + master_addr, master_port = _get_addr_and_port(rdzv_parameters) + + # Set the signals to handle in the environment variable + os.environ["TORCHELASTIC_SIGNALS_TO_HANDLE"] = config.signals_to_handle + + # Start health check server before rendezvous so TW sees a healthy + # thrift port during the potentially long MAST rendezvous store barrier + # (10-22+ min for large jobs). The _AliveCallbackProxy returns + # time.time() until wired to the agent after construction. + # Skip if a server was already provided by the caller (e.g. started + # before remote_pre_launch in the APF executor). + if health_check_server is None: + healthcheck_port = os.getenv(TORCHELASTIC_HEALTH_CHECK_PORT) + if healthcheck_port is not None and justknobs_check( + "ai_infra/pytorch_distributed:torchelastic_enable_healthcheck_before_rendezvous", + default=False, + ): + try: + health_check_server = create_healthcheck_server( + alive_callback=_AliveCallbackProxy(), + port=int(healthcheck_port), + timeout=60, + ) + health_check_server.start() + logger.info( + "Started early health check server on port %s before rendezvous", + healthcheck_port, + ) + except Exception: + logger.warning( + "Failed to start early health check server", exc_info=True + ) + health_check_server = None + + spec = WorkerSpec( + role=config.role, + local_world_size=config.nproc_per_node, + entrypoint=entrypoint, + args=tuple(args), + rdzv_handler=rdzv_registry.get_rendezvous_handler(rdzv_parameters), + max_restarts=config.max_restarts, + monitor_interval=config.monitor_interval, + master_addr=master_addr, + master_port=master_port, + local_addr=config.local_addr, + event_log_handler=config.event_log_handler, + numa_options=config.numa_options, + duplicate_stdout_filters=config.duplicate_stdout_filters, + duplicate_stderr_filters=config.duplicate_stderr_filters, + virtual_local_rank=config.virtual_local_rank, + ) + + agent = LocalElasticAgent( + spec=spec, + logs_specs=config.logs_specs, # type: ignore[arg-type] + start_method=config.start_method, + log_line_prefix_template=config.log_line_prefix_template, + shutdown_timeout=config.shutdown_timeout, # type: ignore[arg-type] + health_check_server=health_check_server, + ) + + if health_check_server is not None: + cb = health_check_server.alive_callback + if isinstance(cb, _AliveCallbackProxy): + cb.set_delegate(agent._get_alive_time) + + shutdown_rdzv = True + try: + metrics.initialize_metrics(metrics.MetricsConfig(config.metrics_cfg)) + + result = agent.run() + # records that agent.run() has succeeded NOT that workers have succeeded + events.record(agent.get_event_succeeded(), config.event_log_handler) + + if result.is_failed(): + # ChildFailedError is treated specially by @record + # if the error files for the failed children exist + # @record will copy the first error (root cause) + # to the error file of the launcher process. + raise ChildFailedError( + name=entrypoint_name, + failures=result.failures, + ) + + return result.return_values + except ChildFailedError: + raise + except SignalException: + # when the agent dies with a signal do NOT shutdown the rdzv_handler + # since this closes the rendezvous on this rdzv_id permanently and + # prevents any additional scaling events + shutdown_rdzv = False + events.record(agent.get_event_failed(), config.event_log_handler) + raise + except Exception: + events.record(agent.get_event_failed(), config.event_log_handler) + raise + finally: + if shutdown_rdzv: + spec.rdzv_handler.shutdown() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/logging_handlers.py b/.venv/lib/python3.12/site-packages/torch/distributed/logging_handlers.py new file mode 100644 index 0000000000000000000000000000000000000000..ed6832fd1ae834b6365a6b005b07bbbfffe90726 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/logging_handlers.py @@ -0,0 +1,16 @@ +#!/usr/bin/env python3 + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +import logging + + +__all__: list[str] = [] + +_log_handlers: dict[str, logging.Handler] = { + "default": logging.NullHandler(), +} diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e15fb517052e4aefeb7377d1f0ca63cf2b2da753 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__init__.py @@ -0,0 +1,7 @@ +import torch + +from .functional import * # noqa: F403 + + +if torch.distributed.rpc.is_available(): + from .api.remote_module import RemoteModule diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1d974de3dfbb8e827ddaa5dc1ac1a69b25f5f544 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/functional.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/functional.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fb2a241f2b4a5deaedfa8b8d5df9dbb8ffdd9bed Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/__pycache__/functional.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dca4631ff5557be7ca29daacda712ef3a43d5bc8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/remote_module.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/remote_module.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fe5682526955dca882a543a7052d7a5f290ef14f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/__pycache__/remote_module.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/remote_module.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/remote_module.py new file mode 100644 index 0000000000000000000000000000000000000000..15dfbb8d3e03ae571a47d4f540b2e80aba00abcc --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/nn/api/remote_module.py @@ -0,0 +1,762 @@ +#!/usr/bin/python3 +# mypy: allow-untyped-defs +import collections +import io +import sys +import types +from collections.abc import Callable, Iterator, Mapping +from typing import Any, TypeVar +from typing_extensions import Self + +import torch +import torch.distributed.rpc as rpc +from torch import device, dtype, nn, Tensor +from torch.distributed import _remote_device +from torch.distributed.nn.jit import instantiator +from torch.distributed.rpc.internal import _internal_rpc_pickler +from torch.nn import Module +from torch.nn.parameter import Parameter +from torch.utils.hooks import RemovableHandle + + +__all__ = ["RemoteModule"] + +_grad_t = tuple[Tensor, ...] | Tensor +# See https://mypy.readthedocs.io/en/latest/generics.html#generic-methods-and-generic-self for the use +# of `T` to annotate `self`. Many methods of `Module` return `self` and we want those return values to be +# the type of the subclass, not the looser type of `Module`. +T = TypeVar("T", bound="Module") + +_NON_SCRIPTABLE_REMOTE_MODULE_MODULE = ( + instantiator.instantiate_non_scriptable_remote_module_template() +) + +_REMOTE_MODULE_PICKLED_ATTRIBUTES = ( + "on", + "device", + "is_device_map_set", + "is_scriptable", + "generated_methods", + "module_rref", +) + +_SerializedRemoteModule = collections.namedtuple( # type: ignore[misc] + "_SerializedRemoteModule", + _REMOTE_MODULE_PICKLED_ATTRIBUTES, +) + +# These attributes are mostly from RemoteModule's parent class and are intentionally not pickled. +# A new attribute of RemoteModule should be either in _REMOTE_MODULE_PICKLED_ATTRIBUTES +# or _REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING. +# Otherwise, it will not be pickled. +_REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING = ( + "training", + "_parameters", + "_buffers", + "_non_persistent_buffers_set", + "_backward_hooks", + "_backward_pre_hooks", + "_is_full_backward_hook", + "_forward_hooks", + "_forward_hooks_with_kwargs", + "_forward_hooks_always_called", + "_forward_pre_hooks", + "_forward_pre_hooks_with_kwargs", + "_state_dict_hooks", + "_state_dict_pre_hooks", + "_load_state_dict_pre_hooks", + "_load_state_dict_post_hooks", + "_state_dict_pre_hooks", + "_modules", + # The two attributes below are generated methods, not available at pickling time. + "forward_async", + "forward", +) + + +# RPC handler. +def _instantiate_template(module_interface_cls, enable_moving_cpu_tensors_to_cuda): + instantiator.instantiate_scriptable_remote_module_template( + module_interface_cls, enable_moving_cpu_tensors_to_cuda + ) + + +def _create_module(module_cls, args, kwargs, device): + module = module_cls(*args, **kwargs) + if not isinstance(module, nn.Module): + raise ValueError( + "Expect `module_cls(*args, **kwargs)` returns an instance of , " + f"but it returns an instance of {type(module)}." + ) + module.to(device) + return module + + +def _create_module_with_interface( + module_cls, args, kwargs, device, module_interface_cls +): + module = _create_module(module_cls, args, kwargs, device) + if module_interface_cls is not None: + module = torch.jit.script(module) + return rpc.RRef(module, module_interface_cls) + + +def _param_rrefs(module_rref, recurse) -> list[rpc.RRef[Parameter]]: + ret: list[rpc.RRef[Parameter]] = [ + rpc.RRef(param) for param in module_rref.local_value().parameters(recurse) + ] + return ret + + +def _raise_not_supported(name: str) -> None: + raise ValueError(f"Method ``{name}`` not supported for RemoteModule") + + +class _RemoteModule(nn.Module): + def __new__(cls, *args, **kwargs): + # Use __new__ for logging purposes. + torch._C._log_api_usage_once("torch.distributed.nn.api.remote_module") + return super().__new__(cls) + + def __init__( + self, + remote_device: str, + module_cls: type[nn.Module], + args: tuple | None = None, + kwargs: dict[str, Any] | None = None, + _module_interface_cls: Any = None, + ): + """ + RemoteModule instance can only be created after RPC initialization. + + It creates a user-specified module on a specified remote node. + It behaves like a regular ``nn.Module`` except that the ``forward`` method is + executed on the remote node. + It takes care of autograd recording to ensure the backward pass propagates + gradients back to the corresponding remote module. + It can be shared across processors using `RPC framework `__, + without incurring any overheads of copying the actual module, + which is equivalent to an :class:`~torch.distributed.rpc.RRef` + pointing to the remote module. + + The arguments of ``forward_async`` and ``forward`` are the same as + the ``forward`` method of the module returned by the ``module_cls``. + + Apart from ``forward_async`` and ``forward``, no other methods are supported from nn.Module for now. + + Particularly, to create a hybrid model, typically the local modules should be + created outside of remote modules, rather than as submodules of any remote module (by calling ``add_module``). + Hybrid Example: + >>> class HybridModel(nn.Module): + >>> def __init__(self) -> None: + >>> nn.Module.__init__(self) + >>> self.remote_embedding = RemoteModule(...) + >>> self.local_linear = nn.Linear(...) + + For example, if ``module_cls`` returns an instance of ``nn.Linear``, + that has ``forward`` method signature, ``def forward(input: Tensor) -> Tensor:``, + the generated ``RemoteModule`` will have 2 methods in signature of + ``def forward(input: Tensor) -> Tensor:`` and + ``def forward_async(input: Tensor) -> Future[Tensor]:``. + + .. note:: + If the remote module is placed on a cuda device, + any input CPU tensors will be automatically moved to the same cuda device, + and GPU tensors are returned over the wire according to the device map of the remote worker on TensorPipe RPC backend. + + Args: + remote_device (str): Device on the destination worker where we'd like to place this module. + The device can be a local device or a remote device specified by one of the following remote + formats: + + 1. "rank:/" (ex: "rank:0/cuda:0"). + 2. "/" (ex: "trainer0/cuda:0"). + + In addition, the device field can be optional and the default value is "cpu". + module_cls (nn.Module): For example, + >>> class MyModule(nn.Module): + >>> def forward(input): + >>> return input + 1 + >>> + >>> module_cls = MyModule + args (Sequence, optional): args to be passed to ``module_cls``. + kwargs (Dict, optional): kwargs to be passed to ``module_cls``. + _module_interface_cls (type, optional): The TorchScript interface type for the module + to be created. The type object should be decorated by @torch.jit.interface. + If not provided, the generated RemoteModule is not torchscript-able. + Warning, this is an experimental API and susceptible to frequent changes. + + Returns: + A remote module instance which wraps the :class:`~nn.Module` created by the + user-provided ``module_cls``, it has a blocking ``forward`` method and an + asynchronous ``forward_async`` method that returns a future of the ``forward`` call + on the user-provided module on the remote side. + + Example:: + Run the following code in two different processes: + + >>> # xdoctest: +SKIP("distributed") + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> from torch import nn, Tensor + >>> from torch.distributed.nn.api.remote_module import RemoteModule + >>> + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> remote_linear_module = RemoteModule( + >>> "worker1/cpu", nn.Linear, args=(20, 30), + >>> ) + >>> input = torch.randn(128, 20) + >>> ret_fut = remote_linear_module.forward_async(input) + >>> ret = ret_fut.wait() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + """ + super().__init__() + + enable_moving_cpu_tensors_to_cuda = self._prepare_init(remote_device) + + # Default arguments preparation. + args = args if args is not None else () + kwargs = kwargs if kwargs is not None else {} + + if _module_interface_cls is not None: + # Users reply on this field to know if this generated RemoteModule is TorchScript-able. + self.is_scriptable = True + + # Instantiate template on remote side. + fut = rpc.rpc_async( + self.on, + _instantiate_template, + (_module_interface_cls, enable_moving_cpu_tensors_to_cuda), + ) + + self._init_template( + _module_interface_cls, enable_moving_cpu_tensors_to_cuda + ) + + # Instantiate template on remote side. + fut = rpc.rpc_async( + self.on, + _instantiate_template, + (_module_interface_cls, enable_moving_cpu_tensors_to_cuda), + ) + + # Create the module on the remote side. + fut.wait() # Ensure remote_module_cls is available on remote side. + + # TODO: We need to change this to rpc.remote, and make it async (see the else branch below). + # For that we need to be able to apply _module_interface_cls to the RRef returned by rpc.remote + # See https://github.com/pytorch/pytorch/issues/58098 for more context. + self.module_rref = rpc.rpc_sync( + self.on, + _create_module_with_interface, + (module_cls, args, kwargs, self.device, _module_interface_cls), + ) + else: + self.is_scriptable = False + self.generated_methods = ( + _NON_SCRIPTABLE_REMOTE_MODULE_MODULE._generated_methods + ) + # Create the module on the remote side. + self.module_rref = rpc.remote( + self.on, + _create_module, + (module_cls, args, kwargs, self.device), + ) + + self._install_generated_methods() + self._check_attribute_picklability() + + def remote_parameters(self, recurse: bool = True) -> list[rpc.RRef[Parameter]]: + """ + Return a list of :class:`~torch.distributed.rpc.RRef` pointing to the remote module's parameters. + + This can typically be used in conjunction + with :class:`~torch.distributed.optim.DistributedOptimizer`. + + Args: + recurse (bool): if True, then returns parameters of the remote + module and all submodules of the remote module. Otherwise, + returns only parameters that are direct members of the + remote module. + + Returns: + A list of :class:`~torch.distributed.rpc.RRef` (``List[RRef[nn.Parameter]]``) + to remote module's parameters. + """ + return rpc.rpc_sync(self.on, _param_rrefs, args=(self.module_rref, recurse)) + + def get_module_rref(self) -> rpc.RRef[nn.Module]: + """Return an :class:`~torch.distributed.rpc.RRef` (``RRef[nn.Module]``) pointing to the remote module.""" + return self.module_rref + + @torch.jit.export + def __getstate__(self): + raise RuntimeError( + "Cannot pickle RemoteModule in python pickler. RemoteModule can only be pickled when using RPC" + ) + + @torch.jit.export + def __setstate__(self, state): + raise RuntimeError( + "Cannot unpickle RemoteModule in python pickler. RemoteModule can only be unpickled when using RPC" + ) + + def register_buffer( + self, name: str, tensor: Tensor | None, persistent: bool = True + ) -> None: + _raise_not_supported(self.register_buffer.__name__) + + def register_parameter(self, name: str, param: Parameter | None) -> None: + _raise_not_supported(self.register_parameter.__name__) + + def add_module(self, name: str, module: Module | None) -> None: + _raise_not_supported(self.add_module.__name__) + + def apply(self, fn: Callable[[Module], None]) -> Self: # type: ignore[return] + _raise_not_supported(self.apply.__name__) + + def cuda(self, device: int | device | None = None) -> Self: # type: ignore[return] + _raise_not_supported(self.cuda.__name__) + + def ipu(self, device: int | device | None = None) -> Self: # type: ignore[return] + _raise_not_supported(self.ipu.__name__) + + def xpu(self, device: int | device | None = None) -> Self: # type: ignore[return] + _raise_not_supported(self.xpu.__name__) + + def cpu(self) -> Self: # type: ignore[return] + _raise_not_supported(self.cpu.__name__) + + def type(self, dst_type: dtype | str) -> Self: # type: ignore[return] + _raise_not_supported(self.type.__name__) + + def float(self) -> Self: # type: ignore[return] + _raise_not_supported(self.float.__name__) + + def double(self) -> Self: # type: ignore[return] + _raise_not_supported(self.double.__name__) + + def half(self) -> Self: # type: ignore[return] + _raise_not_supported(self.half.__name__) + + def bfloat16(self) -> Self: # type: ignore[return] + _raise_not_supported(self.bfloat16.__name__) + + def to(self, *args, **kwargs) -> T: # type: ignore[misc, return, type-var] + _raise_not_supported(self.to.__name__) + + def register_backward_hook( # type: ignore[return] + self, + hook: Callable[[Module, _grad_t, _grad_t], _grad_t | None], + # pyrefly: ignore [bad-return] + ) -> RemovableHandle: + _raise_not_supported(self.register_backward_hook.__name__) + + def register_forward_pre_hook( # type: ignore[return] + self, + hook: Callable[[T, tuple[Any, ...]], Any | None] + | Callable[ + [T, tuple[Any, ...], dict[str, Any]], tuple[Any, dict[str, Any]] | None + ], + prepend: bool = False, + with_kwargs: bool = False, + # pyrefly: ignore [bad-return] + ) -> RemovableHandle: + _raise_not_supported(self.register_forward_pre_hook.__name__) + + def register_forward_hook( # type: ignore[return, override] + self, + hook: Callable[[T, tuple[Any, ...], Any], Any | None] + | Callable[[T, tuple[Any, ...], dict[str, Any], Any], Any | None], + prepend: bool = False, + with_kwargs: bool = False, + # pyrefly: ignore [bad-return] + ) -> RemovableHandle: + _raise_not_supported(self.register_forward_hook.__name__) + + def state_dict(self, *args, **kwargs): + _raise_not_supported(self.state_dict.__name__) + + def load_state_dict( + self, + state_dict: Mapping[str, Any], + strict: bool = True, + assign: bool = False, + ): + _raise_not_supported(self.load_state_dict.__name__) + + def parameters(self, recurse: bool = True) -> Iterator[Parameter]: + raise ValueError( + "Method ``parameters`` not supported for RemoteModule. Please use ``remote_parameters`` instead." + ) + + def named_parameters( # type: ignore[return] + self, + prefix: str = "", + recurse: bool = True, + remove_duplicate: bool = True, + # pyrefly: ignore [bad-return] + ) -> Iterator[tuple[str, Parameter]]: + _raise_not_supported(self.named_parameters.__name__) + + def buffers(self, recurse: bool = True) -> Iterator[Tensor]: # type: ignore[return] + _raise_not_supported(self.buffers.__name__) + + def named_buffers( # type: ignore[return] + self, + prefix: str = "", + recurse: bool = True, + remove_duplicate: bool = True, + # pyrefly: ignore [bad-return] + ) -> Iterator[tuple[str, Tensor]]: + _raise_not_supported(self.named_buffers.__name__) + + def children(self) -> Iterator[Module]: # type: ignore[return] + _raise_not_supported(self.children.__name__) + + def named_children(self) -> Iterator[tuple[str, Module]]: # type: ignore[return] + _raise_not_supported(self.named_children.__name__) + + def modules(self) -> Iterator[Module]: # type: ignore[return] + _raise_not_supported(self.modules.__name__) + + def named_modules( + self, + memo: set[Module] | None = None, + prefix: str = "", + remove_duplicate: bool = True, + ): + _raise_not_supported(self.named_modules.__name__) + + def train(self, mode: bool = True) -> Self: + return self.module_rref.rpc_sync().train() # type: ignore[operator, union-attr] + + def eval(self) -> Self: + return self.module_rref.rpc_sync().eval() # type: ignore[operator, union-attr] + + def requires_grad_(self, requires_grad: bool = True) -> Self: # type: ignore[return] + _raise_not_supported(self.requires_grad_.__name__) + + def zero_grad(self, set_to_none: bool = True) -> None: + _raise_not_supported(self.zero_grad.__name__) + + def share_memory(self) -> Self: # type: ignore[return] + _raise_not_supported(self.share_memory.__name__) + + def extra_repr(self) -> str: # type: ignore[return] + _raise_not_supported(self.extra_repr.__name__) + + def _prepare_init(self, remote_device_str: str) -> bool: + """Prepare the initialization and returns whether to enable automatically moving CPU tensors to CUDA devices.""" + # Sanity check. + if not rpc._is_current_rpc_agent_set(): + raise AssertionError("RemoteModule only works in RPC.") + + remote_device = _remote_device(remote_device_str) + self.on = ( + remote_device.worker_name() + if remote_device.worker_name() is not None + else remote_device.rank() + ) + self.device = str(remote_device.device()) + agent = rpc._get_current_rpc_agent() + # If the device map of the remote worker is set, + # then enable moving any input CPU tensors to the same cuda device. + self.is_device_map_set = bool( + agent._get_device_map(agent.get_worker_info(self.on)) # type: ignore[arg-type] + ) + # ``enable_moving_cpu_tensors_to_cuda`` is less strict than ``is_device_map_set``: + # If ``enable_moving_cpu_tensors_to_cuda`` is true, but the device map is not set, + # then any CPU tensors can still be moved to a cuda device to run forward, + # but the output must be moved back to CPU before being sent over the wire. + enable_moving_cpu_tensors_to_cuda = torch.device(self.device).type == "cuda" + return enable_moving_cpu_tensors_to_cuda + + def _init_template(self, module_interface_cls, enable_moving_cpu_tensors_to_cuda): + """Instantiate template on local side.""" + generated_module = instantiator.instantiate_scriptable_remote_module_template( + module_interface_cls, enable_moving_cpu_tensors_to_cuda + ) + self.generated_methods = generated_module._generated_methods + + def _check_attribute_picklability(self): + """Check if all the attribute has explicitly defined whether to be pickled (i.e., picklability).""" + for k in self.__dict__: + if ( + k not in _REMOTE_MODULE_PICKLED_ATTRIBUTES + and k not in _REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING + ): + raise AttributeError( + f"Attribute {k} must be either in ``_REMOTE_MODULE_PICKLED_ATTRIBUTES`` or " + "``_REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING``." + ) + + def _install_generated_methods(self): + for method in self.generated_methods: + method_name = method.__name__ + method = torch.jit.export(method) + setattr(self, method_name, types.MethodType(method, self)) + + @staticmethod + def init_from_module_rref( + remote_device: str, + module_rref: rpc.RRef[nn.Module], + _module_interface_cls: Any = None, + ): + """ + Besides the constructor, a RemoteModule instance can also be initialized given a module RRef. + + This alternate initialization method can be particularly useful if we want to create multiple + RemoteModule instances that share the same underlying module and reduce memory consumption. + + Moreover, this also provides a workaround for passing script RemoteModule over RPC, + which is not supported. The recommended way is as follows: + + 1. the sender creates a RemoteModule; + 2. the sender sends its ``module_rref`` over RPC; + 3. the receiver calls this method to initialize another RemoteModule using the same ``module_rref``. + + Example:: + Run the following code in two different processes: + + >>> # xdoctest: +SKIP("distributed") + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> from torch import nn, Tensor + >>> from torch.distributed.nn.api.remote_module import RemoteModule + >>> + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> remote_module = RemoteModule( + >>> "worker1/cpu", nn.Linear, args=(20, 30), + >>> ) + >>> + >>> remote_module1 = rpc.rpc_sync( + >>> "worker1/cpu", + >>> RemoteModule.init_from_module_rref, + >>> ("worker1/cpu", remote_module1.get_module_rref()), + >>> ) + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + Args: + remote_device (str): Device on the destination worker where we'd like to place this module. + The device can be a local device or a remote device specified by one of the following remote + formats: + + 1. "rank:/" (ex: "rank:0/cuda:0"). + 2. "/" (ex: "trainer0/cuda:0"). + + In addition, the device field can be optional and the default value is "cpu". + module_rref (RRef[nn.Module]): The module reference shared by both the caller and + the created remote module. + _module_interface_cls (type, optional): The TorchScript interface type for the module + to be created. The type object should be decorated by @torch.jit.interface. + If not provided, the generated RemoteModule is not torchscript-able. + Warning, this is an experimental API and susceptible to frequent changes. + + Returns: + A remote module instance which wraps the :class:`~nn.Module` created by the + user-provided ``module_rref``, it has a blocking ``forward`` method and an + asynchronous ``forward_async`` method that returns a future of the ``forward`` call + on the user-provided module on the remote side. + """ + # NOTE: if a new attribute is added to this class, also need to add it + # to ``_REMOTE_MODULE_PICKLED_ATTRIBUTES`` for pickling/unpickling. + + remote_module = object.__new__(RemoteModule) + + enable_moving_cpu_tensors_to_cuda = remote_module._prepare_init(remote_device) + + if _module_interface_cls is not None: + # Users reply on this field to know if this generated RemoteModule is TorchScript-able. + remote_module.is_scriptable = True + + remote_module._init_template( + _module_interface_cls, enable_moving_cpu_tensors_to_cuda + ) + else: + remote_module.is_scriptable = False + remote_module.generated_methods = ( + _NON_SCRIPTABLE_REMOTE_MODULE_MODULE._generated_methods + ) + remote_module.module_rref = module_rref + + remote_module._install_generated_methods() + remote_module._check_attribute_picklability() + + return remote_module + + +class RemoteModule(_RemoteModule): + """ + A RemoteModule instance can only be created after RPC initialization. + + It creates a user-specified module on a specified remote node. + It behaves like a regular ``nn.Module`` except that the ``forward`` method is + executed on the remote node. + It takes care of autograd recording to ensure the backward pass propagates + gradients back to the corresponding remote module. + + It generates two methods ``forward_async`` and ``forward`` based on the + signature of the ``forward`` method of ``module_cls``. ``forward_async`` + runs asynchronously and returns a Future. The arguments of ``forward_async`` + and ``forward`` are the same as the ``forward`` method of the module + returned by the ``module_cls``. + + For example, if ``module_cls`` returns an instance of ``nn.Linear``, + that has ``forward`` method signature: ``def forward(input: Tensor) -> Tensor:``, + the generated ``RemoteModule`` will have 2 methods with the signatures: + + | ``def forward(input: Tensor) -> Tensor:`` + | ``def forward_async(input: Tensor) -> Future[Tensor]:`` + + Args: + remote_device (str): Device on the destination worker where we'd like to place this module. + The format should be "/", where the device field can be parsed as torch.device type. + E.g., "trainer0/cpu", "trainer0", "ps0/cuda:0". + In addition, the device field can be optional and the default value is "cpu". + module_cls (nn.Module): Class for the module to be created remotely. For example, + + >>> class MyModule(nn.Module): + >>> def forward(input): + >>> return input + 1 + >>> + >>> module_cls = MyModule + + args (Sequence, optional): args to be passed to ``module_cls``. + kwargs (Dict, optional): kwargs to be passed to ``module_cls``. + + Returns: + A remote module instance which wraps the :class:`~nn.Module` created by the + user-provided ``module_cls``, it has a blocking ``forward`` method and an + asynchronous ``forward_async`` method that returns a future of the ``forward`` call + on the user-provided module on the remote side. + + Example:: + Run the following code in two different processes: + + >>> # xdoctest: +SKIP("distributed") + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> from torch import nn, Tensor + >>> from torch.distributed.nn.api.remote_module import RemoteModule + >>> + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> remote_linear_module = RemoteModule( + >>> "worker1/cpu", nn.Linear, args=(20, 30), + >>> ) + >>> input = torch.randn(128, 20) + >>> ret_fut = remote_linear_module.forward_async(input) + >>> ret = ret_fut.wait() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + Furthermore, a more practical example that is combined with + `DistributedDataParallel `__ (DDP) + can be found in this `tutorial `__. + """ + + def __init__( + self, + remote_device: str, + module_cls: type[nn.Module], + args: tuple | None = None, + kwargs: dict[str, Any] | None = None, + ): + super().__init__(remote_device, module_cls, args, kwargs) + + +def _remote_module_receiver( + *remote_module_pickled_attrs, +): + """Deserializes a RemoteModule.""" + serialized_remote_module = _SerializedRemoteModule._make( + remote_module_pickled_attrs + ) + m = object.__new__(RemoteModule) + m.__dict__.update(serialized_remote_module._asdict()) + + # Unpickling the attribute `module_rref` must invoke RRef's `_deserialize()` method. + m.module_rref = rpc.PyRRef._deserialize(m.module_rref) + + # Install generated methods when unpickled. + for method in m.generated_methods: + method_name = method.__name__ + method = torch.jit.export(method) + setattr(m, method_name, types.MethodType(method, m)) + + return m + + +def _remote_module_reducer(remote_module): + """Serialize a RemoteModule.""" + pickled_attrs = {} + for k, v in remote_module.__dict__.items(): + # Pickling the attribute `module_rref` must invoke RRef's `_serialize()` method. + if k == "module_rref": + pickled_attrs[k] = v._serialize() + elif k in _REMOTE_MODULE_PICKLED_ATTRIBUTES: + pickled_attrs[k] = v + # Check if unpickled attributes are all in _REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING. + elif k not in _REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING: + print( + f"The new attribute ``{k}`` of RemoteModule is ignored during RPC pickling. " + "To pickle this attribute, please add it to ``_REMOTE_MODULE_PICKLED_ATTRIBUTES``. " + "Otherwise, please explicitly add it to ``_REMOTE_MODULE_ATTRIBUTES_IGNORE_FOR_PICKLING``.", + file=sys.stderr, + ) + + return ( + _remote_module_receiver, + tuple(pickled_attrs.values()), + ) + + +def _recursive_script_module_receiver( + recursive_script_module_serialized, +): + """Deserializes a RecursiveScriptModule that does not contain a script RemoteModule.""" + f = io.BytesIO(recursive_script_module_serialized) + m = torch.jit.load(f) + return m + + +def _recursive_script_module_reducer(recursive_script_module): + """Serialize a RecursiveScriptModule that does not contain a script RemoteModule, and raises an error otherwise.""" + if hasattr(recursive_script_module._c, "module_rref"): + raise RuntimeError( + "Passing a script RemoteModule over RPC is not supported. Please create a RemoteModule in the sender, " + "send the `module_rref` to the receiver, and create a new instance on the receiver end by passing this `module_rref`." + ) + + f = io.BytesIO() + torch.jit.save(recursive_script_module, f) + return (_recursive_script_module_receiver, (f.getvalue(),)) + + +_internal_rpc_pickler._register_reducer(RemoteModule, _remote_module_reducer) +_internal_rpc_pickler._register_reducer( + torch.jit.RecursiveScriptModule, _recursive_script_module_reducer +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/functional.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/functional.py new file mode 100644 index 0000000000000000000000000000000000000000..e49230c5d2d9ba4cbac1bd4afa234c8b3ff1d7c7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/nn/functional.py @@ -0,0 +1,538 @@ +# mypy: allow-untyped-defs +import warnings + +import torch +import torch.distributed as dist +from torch.autograd import Function + +# The two imports below are not always available depending on the +# USE_DISTRIBUTED compile flag. Make sure they raise import error +# if we're trying to use them. +from torch.distributed import group, ReduceOp + + +def _not_supported_under_compile(name, *, suggestion=None): + msg = ( + f"torch.distributed.nn.functional.{name} is not supported under torch.compile." + ) + if suggestion: + msg += f" Use {suggestion} instead." + raise RuntimeError(msg) + + +def _deprecated(name, suggestion): + warnings.warn( + f"torch.distributed.nn.functional.{name} is deprecated, " + f"use {suggestion} instead.", + category=FutureWarning, + stacklevel=3, + ) + + +def broadcast(tensor, src, group=group.WORLD): + """ + Broadcasts the tensor to the whole group. + + ``tensor`` must have the same number of elements in all processes + participating in the collective. + + Arguments: + tensor (Tensor): Data to be sent if ``src`` is the rank of current + process. + src (int): Source rank. + group (ProcessGroup, optional): The process group to work on. + + Returns: + Tensor: Received tensor from the broadcast op. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile( + "broadcast", + suggestion="torch.distributed._functional_collectives.broadcast", + ) + _deprecated("broadcast", "torch.distributed._functional_collectives.broadcast") + return _Broadcast.apply(src, group, tensor) + + +def gather(tensor, dst=0, group=group.WORLD): + """ + Gathers a list of tensors in a single process. + + Arguments: + tensor (Tensor): Input tensor. + dst (int, optional): Destination rank (default is 0). + group (ProcessGroup, optional): The process group to work on. + + Returns: + tuple[Tensor]: List of appropriately-sized tensors with the gathered data. + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile("gather") + return _Gather.apply(dst, group, tensor) + + +def scatter(tensors, src=0, group=group.WORLD): + """ + Scatters a list of tensors to all processes in a group. + + Each process will receive exactly one tensor and store its data in the + ``tensor`` argument. + + Arguments: + tensors (list[Tensor]): List of tensors to scatter on the source rank. + Receivers must pass ``None`. + src (int, optional): Source rank (default is 0). + group (ProcessGroup, optional): The process group to work on. + + Returns: + Tensor: Output tensor from the scatter operation. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile("scatter") + return _Scatter.apply(src, group, *tensors) + + +def reduce(tensor, dst, op=ReduceOp.SUM, group=group.WORLD): + """ + Reduces the tensor data across all machines. + + Only the process with rank ``dst`` is going to receive the final result. + + Arguments: + tensor (Tensor): Input of the collective. + dst (int): Destination rank. + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. + + Returns: + Tensor: Output of the collective. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile("reduce") + return _Reduce.apply(dst, op, group, tensor) + + +def reduce_scatter(output, input_list, op=ReduceOp.SUM, group=group.WORLD): + """ + Reduces, then scatters a list of tensors to all processes in a group. + + Arguments: + output (Tensor): Output tensor. + input_list (list[Tensor]): List of tensors to reduce and scatter. + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. + + Returns: + Tensor: Output of the collective. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile( + "reduce_scatter", + suggestion="torch.distributed._functional_collectives.reduce_scatter_single", + ) + _deprecated( + "reduce_scatter", + "torch.distributed._functional_collectives.reduce_scatter_single", + ) + return _Reduce_Scatter.apply(op, group, output, *input_list) + + +def all_gather(tensor, group=group.WORLD): + """ + Gathers tensors from the whole group in a list. + + Arguments: + tensor (Tensor): Tensor to be broadcast from current process. + group (ProcessGroup, optional): The process group to work on. + + Returns: + tuple([Tensor]): Output of the collective. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile( + "all_gather", + suggestion="torch.distributed._functional_collectives.all_gather_single", + ) + _deprecated( + "all_gather", "torch.distributed._functional_collectives.all_gather_single" + ) + return _AllGather.apply(group, tensor) + + +def _all_gather_base(output_tensor, input_tensor, group=group.WORLD): + """ + Single tensor all gather. Gathers a single tensor from all ranks, and puts them in a single output tensor. + + Args: + output_tensor (Tensor): Output tensor. It should contain + correctly-sized tensors to be used for output of the collective. + input_tensor (Tensor): Tensor to be broadcast from current process. + group (ProcessGroup, optional): The process group to work on. If None, + the default process group will be used. + + Examples: + >>> # All tensors below are of torch.int64 dtype. + >>> # We have 2 process groups, 2 ranks. + >>> # xdoctest: +SKIP("incorrect want text") + >>> output_tensor = torch.zeros(2, dtype=torch.int64) + >>> output_tensor + [tensor([0, 0])] # Rank 0 and 1 + >>> tensor = torch.arange(1, dtype=torch.int64) + 1 + rank + >>> tensor + tensor([1]) # Rank 0 + tensor([2]) # Rank 1 + >>> dist.all_gather_base(output_tensor, tensor) + >>> output_tensor + tensor([1,2]) # Rank 0 + tensor([1,2]) # Rank 1 + + .. warning:: + `_all_gather_base` is experimental and subject to change. + It is the caller's responsibility to ensure the output_tensor + is correctly sized. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile("_all_gather_base") + return _AllGatherBase.apply(output_tensor, input_tensor, group) + + +def all_to_all(output_tensor_list, input_tensor_list, group=group.WORLD): + """ + Each process scatters list of input tensors to all processes in a group and return gathered list of tensors in output list. + + Arguments: + output_tensor_list (list[Tensor]): list of tensors to gather one per rank. + input_tensor_list (list[Tensor]): List of tensors to scatter one per rank. + group (ProcessGroup, optional): The process group to work on. + + Returns: + tuple([Tensor]): Output of the collective. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile("all_to_all") + return _AlltoAll.apply(group, output_tensor_list, *input_tensor_list) + + +def all_to_all_single( + output, + input, + output_split_sizes=None, + input_split_sizes=None, + group=group.WORLD, +): + """ + Each process splits input tensor and then scatters the split list to all processes in a group. + + Then concatenate the received tensors from all the processes in the group and return single output tensor. + + Arguments: + output (Tensor): Gathered concatenated output tensor. + input (Tensor): Input tensor to scatter. + output_split_sizes: (list[Int], optional): Output split sizes for dim 0 + if specified None or empty, dim 0 of ``output`` tensor must divide + equally by ``world_size``. + input_split_sizes: (list[Int], optional): Input split sizes for dim 0 + if specified None or empty, dim 0 of ``input`` tensor must divide + equally by ``world_size``. + + Returns: + Tensor: Output of the collective. + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile( + "all_to_all_single", + suggestion="torch.distributed._functional_collectives.all_to_all_single", + ) + _deprecated( + "all_to_all_single", + "torch.distributed._functional_collectives.all_to_all_single", + ) + return _AlltoAllSingle.apply( + group, output, output_split_sizes, input_split_sizes, input + ) + + +def all_reduce(tensor, op=ReduceOp.SUM, group=group.WORLD): + """ + Reduces the tensor data across all machines in such a way that all get the final result. + + After the call the returned tensor is going to be bitwise + identical in all processes. + + Arguments: + tensor (Tensor): Input of the collective. + op (optional): One of the values from + ``torch.distributed.ReduceOp`` + enum. Specifies an operation used for element-wise reductions. + group (ProcessGroup, optional): The process group to work on. + + Returns: + Tensor: Output of the collective + + """ + if torch.compiler.is_compiling(): + _not_supported_under_compile( + "all_reduce", + suggestion="torch.distributed._functional_collectives.all_reduce", + ) + _deprecated("all_reduce", "torch.distributed._functional_collectives.all_reduce") + return _AllReduce.apply(op, group, tensor) + + +class _Broadcast(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, src, group, tensor): + ctx.src = src + ctx.group = group + ctx.global_rank = dist.get_rank() + # torch.distributed makes all the calls in place + # we allocate new tensors to avoid this + tensor = tensor.clone() + dist.broadcast(tensor, src, group=group) + return tensor + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + gx = _Reduce.apply(ctx.src, ReduceOp.SUM, ctx.group, grad_output) + if ctx.src != ctx.global_rank: + gx.zero_() + return (None, None, gx) + + +class _Gather(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, dst, group, tensor): + ctx.dst = dst + ctx.group = group + # Need to create a list of tensors here to do the + # aggregation, get it from the group size + # tensor should be correctly sized for the method + # gathering + tensor_list = [ + torch.zeros_like(tensor) for i in range(dist.get_world_size(group=group)) + ] + + tensor = tensor.contiguous() + if dist.get_rank(group=group) == dst: + dist.gather(tensor, tensor_list, dst, group=group) + else: + dist.gather(tensor, None, dst, group=group) + return tuple(tensor_list) + + @staticmethod + def backward(ctx, *grad_outputs): + return (None, None) + (_Scatter.apply(ctx.dst, ctx.group, *grad_outputs),) + + +class _Scatter(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, src, group, *tensors): + ctx.src = src + ctx.group = group + if not all(t.size() == tensors[0].size() for t in tensors): + raise AssertionError + output = torch.zeros_like(tensors[0]) + if dist.get_rank(group=group) == src: + dist.scatter(output, list(tensors), src, group=group) + else: + dist.scatter(output, None, src, group=group) + return output + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + return (None, None) + _Gather.apply(ctx.src, ctx.group, grad_output) + + +class _Reduce(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, src, op, group, tensor): + ctx.src = src + ctx.group = group + tensor = tensor.clone() + dist.reduce(tensor, src, op=op, group=group) + return tensor + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + return (None, None, None) + (_Broadcast.apply(ctx.src, ctx.group, grad_output),) + + +class _Reduce_Scatter(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, op, group, tensor, *input_tensor_list): + ctx.group = group + # Need contiguous tensors for collectives. + tensor = tensor.contiguous() + input_tensor_list = tuple(t.contiguous() for t in input_tensor_list) + dist.reduce_scatter(tensor, list(input_tensor_list), op=op, group=group) + return tensor + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + return (None, None, None) + _AllGather.apply(ctx.group, grad_output) + + +class _AllGather(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, group, tensor): + # Need contiguous tensors for collectives. + tensor = tensor.contiguous() + + ctx.group = group + out_tensor_list = [ + torch.empty_like(tensor) for _ in range(dist.get_world_size(group=group)) + ] + + dist.all_gather(out_tensor_list, tensor, group=group) + return tuple(out_tensor_list) + + @staticmethod + def backward(ctx, *grad_outputs): + if dist.get_backend(group=ctx.group) in (dist.Backend.NCCL, dist.Backend.XCCL): + rank = dist.get_rank(group=ctx.group) + gx = torch.empty_like(grad_outputs[rank]) + gx = _Reduce_Scatter.apply(ReduceOp.SUM, ctx.group, gx, *grad_outputs) + else: + # As many backends doesn't support ReduceScatter, we use AlltoAll with .sum() + # to emulate the ReduceScatter behavior + tensor_list = [torch.empty_like(tensor) for tensor in grad_outputs] + gxs = _AlltoAll.apply(ctx.group, tensor_list, *grad_outputs) + gx = torch.sum(torch.stack(gxs), dim=0) + return (None, gx) + + +class _AllGatherBase(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, output_tensor, input_tensor, group): + ctx.group = group + dist._all_gather_base(output_tensor, input_tensor.contiguous(), group=group) + return output_tensor + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + if dist.get_backend(group=ctx.group) in (dist.Backend.NCCL, dist.Backend.XCCL): + world_size = dist.get_world_size(group=ctx.group) + out_size = list(grad_output.size()) + if out_size[0] % world_size != 0: + raise RuntimeError( + f"Tensor with dimensions: {out_size} does " + f"not have first dimension divisible by world_size: {world_size}" + ) + out_size[0] = out_size[0] // dist.get_world_size(group=ctx.group) + gx = torch.empty( + out_size, device=grad_output.device, dtype=grad_output.dtype + ) + dist._reduce_scatter_base(gx, grad_output, ReduceOp.SUM, ctx.group) + else: + raise RuntimeError("Backend not supported!") + return (None, gx, None) + + +class _AlltoAll(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, group, out_tensor_list, *tensors): + ctx.group = group + ctx.input_tensor_size_list = [ + tensors[i].size() for i in range(dist.get_world_size(group=group)) + ] + my_rank = dist.get_rank(group=group) + tensors = tuple(t.contiguous() for t in tensors) + # Implement it on means of scatter/gather, send/recv async operations have issues + if dist.get_backend(group=group) is dist.Backend.GLOO: + for i in range(dist.get_world_size(group=group)): + to_send = None + if i == my_rank: + to_send = list(tensors) + dist.scatter(out_tensor_list[i], to_send, i, group=group) + else: + dist.all_to_all( + out_tensor_list, + list(tensors), + group=group, + ) + return tuple(out_tensor_list) + + @staticmethod + def backward(ctx, *grad_outputs): + tensor_list = [ + torch.empty( + size, device=grad_outputs[0].device, dtype=grad_outputs[0].dtype + ) + for size in ctx.input_tensor_size_list + ] + return (None, None) + _AlltoAll.apply(ctx.group, tensor_list, *grad_outputs) + + +class _AlltoAllSingle(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, group, output, output_split_sizes, input_split_sizes, input): + ctx.group = group + ctx.input_size = input.size() + ctx.output_split_sizes = input_split_sizes + ctx.input_split_sizes = output_split_sizes + dist.all_to_all_single( + output, + input, + output_split_sizes=output_split_sizes, + input_split_sizes=input_split_sizes, + group=group, + ) + return output + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + tensor = torch.empty( + ctx.input_size, device=grad_output.device, dtype=grad_output.dtype + ) + return (None, None, None, None) + ( + _AlltoAllSingle.apply( + ctx.group, + tensor, + ctx.output_split_sizes, + ctx.input_split_sizes, + grad_output.contiguous(), + ), + ) + + +class _AllReduce(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, op, group, tensor): + ctx.group = group + ctx.op = op + tensor = tensor.clone(memory_format=torch.contiguous_format) + dist.all_reduce(tensor, op=op, group=group) + return tensor + + @staticmethod + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + return (None, None) + (_AllReduce.apply(ctx.op, ctx.group, grad_output),) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d7e0a7c4ae329a8b5e4f5506e4af015949eefabd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/instantiator.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/instantiator.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..27dfd4c49157cd54facf8f40208ffb531faaada1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/__pycache__/instantiator.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/instantiator.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/instantiator.py new file mode 100644 index 0000000000000000000000000000000000000000..3fe5526ba415e17164ff85a4aa6e0e85eb239ead --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/instantiator.py @@ -0,0 +1,171 @@ +#!/usr/bin/python3 +# mypy: allow-untyped-defs +import importlib.abc +import importlib.util +import sys + +import torch +from torch.distributed.nn.jit.templates.remote_module_template import ( + get_remote_module_template, +) + + +_FILE_PREFIX = "_remote_module_" + + +def get_arg_return_types_from_interface(module_interface): + if not getattr(module_interface, "__torch_script_interface__", False): + raise AssertionError( + "Expect a TorchScript class interface decorated by @torch.jit.interface." + ) + qualified_name = torch._jit_internal._qualified_name(module_interface) + cu = torch.jit._state._python_cu + module_interface_c = cu.get_interface(qualified_name) + if "forward" not in module_interface_c.getMethodNames(): + raise AssertionError( + f"Expect forward in interface methods, while it has {module_interface_c.getMethodNames()}" + ) + method_schema = module_interface_c.getMethod("forward") + + arg_str_list = [] + arg_type_str_list = [] + if method_schema is None: + raise AssertionError + for argument in method_schema.arguments: + arg_str_list.append(argument.name) + + if argument.has_default_value(): + default_value_str = f" = {argument.default_value}" + else: + default_value_str = "" + arg_type_str = f"{argument.name}: {argument.type}{default_value_str}" + arg_type_str_list.append(arg_type_str) + + arg_str_list = arg_str_list[1:] # Remove "self". + args_str = ", ".join(arg_str_list) + + arg_type_str_list = arg_type_str_list[1:] # Remove "self". + arg_types_str = ", ".join(arg_type_str_list) + + if len(method_schema.returns) != 1: + raise AssertionError + argument = method_schema.returns[0] + return_type_str = str(argument.type) + + return args_str, arg_types_str, return_type_str + + +class _StringLoader(importlib.abc.SourceLoader): + """ + A custom loader for dynamically generated Python source code. + + Inherits from SourceLoader for API compatibility but overrides exec_module() + to avoid bytecode caching issues. The default SourceLoader.exec_module() calls + cache_from_source() which fails with IndexError when the filename doesn't + correspond to a real filesystem path with a .py extension. + """ + + def __init__(self, data: str) -> None: + self.data = data + + def get_source(self, fullname: str) -> str: + return self.data + + def get_data(self, path: str) -> bytes: + return self.data.encode("utf-8") + + def get_filename(self, fullname: str) -> str: + return f"<{fullname}>.py" + + def path_stats(self, path: str) -> dict: + # Raise OSError since source is dynamically generated (no filesystem stats) + raise OSError("dynamically generated module has no filesystem stats") + + def exec_module(self, module) -> None: + """ + Execute the module by compiling and running the source directly. + + This overrides SourceLoader.exec_module() to bypass the problematic + get_code() -> cache_from_source() code path that fails on dynamic modules. + """ + source = self.get_source(module.__name__) + filename = self.get_filename(module.__name__) + code = compile(source, filename, "exec", dont_inherit=True) + exec(code, module.__dict__) + + +def _do_instantiate_remote_module_template( + generated_module_name, str_dict, enable_moving_cpu_tensors_to_cuda +): + if generated_module_name in sys.modules: + return sys.modules[generated_module_name] + + loader = _StringLoader( + get_remote_module_template(enable_moving_cpu_tensors_to_cuda).format(**str_dict) + ) + spec = importlib.util.spec_from_loader( + generated_module_name, loader, origin="torch-git" + ) + if spec is None: + raise AssertionError + module = importlib.util.module_from_spec(spec) + sys.modules[generated_module_name] = module + loader.exec_module(module) + return module + + +def instantiate_scriptable_remote_module_template( + module_interface_cls, enable_moving_cpu_tensors_to_cuda=True +): + if not getattr(module_interface_cls, "__torch_script_interface__", False): + raise ValueError( + f"module_interface_cls {module_interface_cls} must be a type object decorated by " + "@torch.jit.interface" + ) + + # Generate the template instance name. + module_interface_cls_name = torch._jit_internal._qualified_name( + module_interface_cls + ).replace(".", "_") + generated_module_name = f"{_FILE_PREFIX}{module_interface_cls_name}" + + # Generate type annotation strs. + assign_module_interface_cls_str = ( + f"from {module_interface_cls.__module__} import " + f"{module_interface_cls.__name__} as module_interface_cls" + ) + args_str, arg_types_str, return_type_str = get_arg_return_types_from_interface( + module_interface_cls + ) + kwargs_str = "" + arrow_and_return_type_str = f" -> {return_type_str}" + arrow_and_future_return_type_str = f" -> Future[{return_type_str}]" + + str_dict = dict( + assign_module_interface_cls=assign_module_interface_cls_str, + arg_types=arg_types_str, + arrow_and_return_type=arrow_and_return_type_str, + arrow_and_future_return_type=arrow_and_future_return_type_str, + args=args_str, + kwargs=kwargs_str, + jit_script_decorator="@torch.jit.script", + ) + return _do_instantiate_remote_module_template( + generated_module_name, str_dict, enable_moving_cpu_tensors_to_cuda + ) + + +def instantiate_non_scriptable_remote_module_template(): + generated_module_name = f"{_FILE_PREFIX}non_scriptable" + str_dict = dict( + assign_module_interface_cls="module_interface_cls = None", + args="*args", + kwargs="**kwargs", + arg_types="*args, **kwargs", + arrow_and_return_type="", + arrow_and_future_return_type="", + jit_script_decorator="", + ) + # For a non-scriptable template, always enable moving CPU tensors to a cuda device, + # because there is no syntax limitation on the extra handling caused by the script. + return _do_instantiate_remote_module_template(generated_module_name, str_dict, True) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6d7c74cde6be39e4302cc2a6972b3d5e03c44c98 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/remote_module_template.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/remote_module_template.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4d1022583b5b65b29d3c6b8fc6d4de9439c60a62 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/__pycache__/remote_module_template.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/remote_module_template.py b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/remote_module_template.py new file mode 100644 index 0000000000000000000000000000000000000000..07b055774b36af4835e308c8a1f85afd0ab35f0f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/nn/jit/templates/remote_module_template.py @@ -0,0 +1,108 @@ +#!/usr/bin/python3 +# mypy: allow-untyped-defs + + +def get_remote_module_template(enable_moving_cpu_tensors_to_cuda: bool): + return _TEMPLATE_PREFIX + ( + _REMOTE_FORWARD_TEMPLATE_ENABLE_MOVING_CPU_TENSORS_TO_CUDA + if enable_moving_cpu_tensors_to_cuda + else _REMOTE_FORWARD_TEMPLATE + ) + + +_TEMPLATE_PREFIX = """from typing import * + +import torch +import torch.distributed.rpc as rpc +from torch import Tensor +from torch._jit_internal import Future +from torch.distributed.rpc import RRef +from typing import Tuple # pyre-ignore: unused import + + +{assign_module_interface_cls} + + +def forward_async(self, {arg_types}){arrow_and_future_return_type}: + args = (self.module_rref, self.device, self.is_device_map_set, {args}) + kwargs = {{{kwargs}}} + return rpc.rpc_async( + self.module_rref.owner(), + _remote_forward, + args, + kwargs, + ) + + +def forward(self, {arg_types}){arrow_and_return_type}: + args = (self.module_rref, self.device, self.is_device_map_set, {args}) + kwargs = {{{kwargs}}} + ret_fut = rpc.rpc_async( + self.module_rref.owner(), + _remote_forward, + args, + kwargs, + ) + return ret_fut.wait() + + +_generated_methods = [ + forward_async, + forward, +] + + +{jit_script_decorator} +""" + +# This template may cause typing error (the mismatch between ``Tuple[()]`` and ``Tuple[Any]``) +# even if the code is only used for instantiation but not execution. +# Therefore, only include handling moving CPU tensors to a cuda device if necessary. +# TODO: Merge these two templates together in the future once TorchScript syntax is improved. +_REMOTE_FORWARD_TEMPLATE_ENABLE_MOVING_CPU_TENSORS_TO_CUDA = """ +def _remote_forward( + module_rref: RRef[module_interface_cls], device: str, is_device_map_set: bool, {arg_types}){arrow_and_return_type}: + module = module_rref.local_value() + device = torch.device(device) + + if device.type != "cuda": + return module.forward({args}, {kwargs}) + + # If the module is on a cuda device, + # move any CPU tensor in args or kwargs to the same cuda device. + # Since torch script does not support generator expression, + # have to use concatenation instead of + # ``tuple(i.to(device) if isinstance(i, Tensor) else i for i in *args)``. + args = ({args},) + out_args: Tuple[()] = () + for arg in args: + arg = (arg.to(device),) if isinstance(arg, Tensor) else (arg,) + out_args = out_args + arg + + kwargs = {{{kwargs}}} + for k, v in kwargs.items(): + if isinstance(v, Tensor): + kwargs[k] = kwargs[k].to(device) + + if is_device_map_set: + return module.forward(*out_args, {kwargs}) + + # If the device map is empty, then only CPU tensors are allowed to send over wire, + # so have to move any GPU tensor to CPU in the output. + # Since torch script does not support generator expression, + # have to use concatenation instead of + # ``tuple(i.cpu() if isinstance(i, Tensor) else i for i in module.forward(*out_args, {kwargs}))``. + ret: Tuple[()] = () + for i in module.forward(*out_args, {kwargs}): + i = (i.cpu(),) if isinstance(i, Tensor) else (i,) + ret = ret + i + return ret +""" + +_REMOTE_FORWARD_TEMPLATE = """ +def _remote_forward( + module_rref: RRef[module_interface_cls], device: str, is_device_map_set: bool, {arg_types}){arrow_and_return_type}: + module = module_rref.local_value() + + return module.forward({args}, {kwargs}) +""" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..faac68bb632934ba730ba7c5ce3cf7fe934a58cf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__init__.py @@ -0,0 +1,44 @@ +""" +:mod:`torch.distributed.optim` exposes DistributedOptimizer, which takes a list +of remote parameters (:class:`~torch.distributed.rpc.RRef`) and runs the +optimizer locally on the workers where the parameters live. The distributed +optimizer can use any of the local optimizer :ref:`optimizer-algorithms` to +apply the gradients on each worker. +""" + +import warnings + +import torch +from torch import optim + +from .apply_optimizer_in_backward import ( + _apply_optimizer_in_backward, + _get_in_backward_optimizers, +) +from .functional_adadelta import _FunctionalAdadelta +from .functional_adagrad import _FunctionalAdagrad +from .functional_adam import _FunctionalAdam +from .functional_adamax import _FunctionalAdamax +from .functional_adamw import _FunctionalAdamW +from .functional_rmsprop import _FunctionalRMSprop +from .functional_rprop import _FunctionalRprop +from .functional_sgd import _FunctionalSGD +from .named_optimizer import _NamedOptimizer +from .utils import as_functional_optim + + +# DistributedOptimizer imports torch.distributed.rpc names, so gate availability +# based on RPC being available. +if hasattr(torch._C, "_rpc_init"): + from .optimizer import DistributedOptimizer + +from .post_localSGD_optimizer import PostLocalSGDOptimizer +from .zero_redundancy_optimizer import ZeroRedundancyOptimizer + + +__all__ = [ + "as_functional_optim", + "DistributedOptimizer", + "PostLocalSGDOptimizer", + "ZeroRedundancyOptimizer", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0356513faf19aa179c5530e158ccc4dca304af65 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/_deprecation_warning.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/_deprecation_warning.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0e27e26a827c33eb8fae406b6de028b83b0fd0ad Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/_deprecation_warning.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/apply_optimizer_in_backward.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/apply_optimizer_in_backward.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..58c25c696715c952ab513c3f7d7ba23476be6624 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/apply_optimizer_in_backward.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adadelta.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adadelta.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..63617000dbbf9cac139ec6d665f3c9047a8949af Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adadelta.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adagrad.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adagrad.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eb580646fe804253bf8bf2298a750eec404639de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adagrad.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adam.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adam.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7bb7e01a17b732a4a0023ba5b08c4999a6f7b338 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adam.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamax.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamax.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6d630630da3d23715a25250d4454fac3989f0c76 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamax.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamw.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamw.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0256aaca61c519a67bf76e798058e65d89f8b70b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_adamw.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rmsprop.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rmsprop.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7d20450bf8afa10042f7945bc27bb0fe01442f90 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rmsprop.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rprop.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rprop.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..99d0cc20a59499c13d06a01ec8587a518aa4cad4 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_rprop.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_sgd.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_sgd.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..36b10f97eb4ce4c2f30db0ed717c0fd8ee71154b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/functional_sgd.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/named_optimizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/named_optimizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..be77e5feb8014ab10ffaf08aab61d0426897b4a2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/named_optimizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/optimizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/optimizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..56a886f3cc2f1d0c43ab84f5eb1b1bea30e8cf7e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/optimizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/post_localSGD_optimizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/post_localSGD_optimizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ce0ff44de066a68ddd9084275c142f44edb2febb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/post_localSGD_optimizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4eeab50f764c2db3dd268c3a24321f37380c5351 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/zero_redundancy_optimizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/zero_redundancy_optimizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..54b8505e39e6007aa2559899c7e8f53ea0ac276c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/optim/__pycache__/zero_redundancy_optimizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/_deprecation_warning.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/_deprecation_warning.py new file mode 100644 index 0000000000000000000000000000000000000000..c3434a4cd4f081843295e488c18a67a5c297fcbf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/_deprecation_warning.py @@ -0,0 +1,16 @@ +import warnings + +import torch + + +@torch.jit.ignore # type: ignore[misc] +def _scripted_functional_optimizer_deprecation_warning(stacklevel: int = 0) -> None: + with warnings.catch_warnings(): + warnings.simplefilter("always") + warnings.warn( + "`TorchScript` support for functional optimizers is deprecated " + "and will be removed in a future PyTorch release. " + "Consider using the `torch.compile` optimizer instead.", + DeprecationWarning, + stacklevel=stacklevel + 2, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/apply_optimizer_in_backward.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/apply_optimizer_in_backward.py new file mode 100644 index 0000000000000000000000000000000000000000..1ff9854793df1aa96a27cb105a1afd1190df942a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/apply_optimizer_in_backward.py @@ -0,0 +1,121 @@ +from collections.abc import Iterable +from typing import Any, no_type_check + +import torch + + +__all__: list[str] = [] + +# WeakTensorKeyDictionary to store relevant meta-data for the Tensor/Parameter +# without changing it's life-time. +# NOTE: Alternative is to add the meta-data as an attribute to the tensor, +# but that will serialize the meta-data if Tensor is serialized. +param_to_optim_hook_handle_map = torch.utils.weak.WeakTensorKeyDictionary() +param_to_acc_grad_map = torch.utils.weak.WeakTensorKeyDictionary() + + +@no_type_check +def _apply_optimizer_in_backward( + optimizer_class: type[torch.optim.Optimizer], + params: Iterable[torch.nn.Parameter], + optimizer_kwargs: dict[str, Any], + register_hook: bool = True, +) -> None: + """ + Upon ``backward()``, the optimizer specified for each parameter will fire after + the gradient has been accumulated into the parameter. + + Note - gradients for these parameters will be set to None after ``backward()``. + This means that any other optimizer not specified via `_apply_optimizer_in_backward` + over this parameter will be a no-op. + + Args: + optimizer_class: (Type[torch.optim.Optimizer]): Optimizer to apply to parameter + params: (Iterator[nn.Parameter]): parameters to apply optimizer state to + optimizer_kwargs: (Dict[str, Any]): kwargs to pass to optimizer constructor + register_hook: (bool): whether to register a hook that runs the optimizer + after gradient for this parameter is accumulated. This is the default + way that optimizer in backward is implemented, but specific use cases + (such as DDP) may wish to override this to implement custom behavior. + (Default = True) + + Example:: + params_generator = model.parameters() + param_1 = next(params_generator) + remainder_params = list(params_generator) + + apply_optimizer_in_backward(torch.optim.SGD, [param_1], {"lr": 0.02}) + apply_optimizer_in_backward(torch.optim.Adam, remainder_params, {"lr": 0.04}) + + model(...).sum().backward() # after backward, parameters will already + # have their registered optimizer(s) applied. + + """ + torch._C._log_api_usage_once("torch.distributed.optim.apply_optimizer_in_backward") + + @no_type_check + def _apply_optimizer_in_backward_to_param(param: torch.nn.Parameter) -> None: + # view_as creates a node in autograd graph that allows us access to the + # parameter's AccumulateGrad autograd function object. We register a + # hook on this object to fire the optimizer when the gradient for + # this parameter is ready (has been accumulated into .grad field) + + # Don't create a new acc_grad if we already have one + # i.e. for shared parameters or attaching multiple optimizers to a param. + if param not in param_to_acc_grad_map: + param_to_acc_grad_map[param] = param.view_as(param).grad_fn.next_functions[ + 0 + ][0] + + optimizer = optimizer_class([param], **optimizer_kwargs) + + if not hasattr(param, "_in_backward_optimizers"): + param._in_backward_optimizers = [] # type: ignore[attr-defined] + # TODO: Remove these attributes once we have a better way of accessing + # optimizer classes and kwargs for a parameter. + param._optimizer_classes = [] # type: ignore[attr-defined] + param._optimizer_kwargs = [] # type: ignore[attr-defined] + + param._in_backward_optimizers.append(optimizer) # type: ignore[attr-defined] + param._optimizer_classes.append(optimizer_class) # type: ignore[attr-defined] + param._optimizer_kwargs.append(optimizer_kwargs) # type: ignore[attr-defined] + + if not register_hook: + return + + def optimizer_hook(*_unused) -> None: + for opt in param._in_backward_optimizers: # type: ignore[attr-defined] + opt.step() + + param.grad = None + + handle = param_to_acc_grad_map[param].register_hook(optimizer_hook) # type: ignore[attr-defined] + if param not in param_to_optim_hook_handle_map: + param_to_optim_hook_handle_map[param] = [] + param_to_optim_hook_handle_map[param].append(handle) + + for param in params: + _apply_optimizer_in_backward_to_param(param) + + +def _get_in_backward_optimizers(module: torch.nn.Module) -> list[torch.optim.Optimizer]: + """ + Return a list of in-backward optimizers applied to ``module``'s parameters. Note that these + optimizers are not intended to directly have their ``step`` or ``zero_grad`` methods called + by the user and are intended to be used for things like checkpointing. + + Args: + module: (torch.nn.Module): model to retrieve in-backward optimizers for + + Returns: + List[torch.optim.Optimizer]: the in-backward optimizers. + + Example:: + _apply_optimizer_in_backward(torch.optim.SGD, model.parameters(), {"lr": 0.01}) + optims = _get_optimizers_in_backward(model) + """ + optims: list[torch.optim.Optimizer] = [] + for param in module.parameters(): + optims.extend(getattr(param, "_in_backward_optimizers", [])) + + return optims diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adadelta.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adadelta.py new file mode 100644 index 0000000000000000000000000000000000000000..4020d117f282463bf6423e93df88cc2f6e6d675d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adadelta.py @@ -0,0 +1,109 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional Adadelta Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalAdadelta: + def __init__( + self, + params: list[Tensor], + lr: float = 1.0, + rho: float = 0.9, + eps: float = 1e-6, + weight_decay: float = 0.0, + foreach: bool = False, + maximize: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + self.defaults = { + "lr": lr, + "rho": rho, + "eps": eps, + "weight_decay": weight_decay, + } + self.foreach = foreach + self.maximize = maximize + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + square_avgs = [] + acc_deltas = [] + state_steps = [] + lr = self.defaults["lr"] + rho = self.defaults["rho"] + eps = self.defaults["eps"] + weight_decay = self.defaults["weight_decay"] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + has_complex = False + for param, gradient in zip(params, gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + state["square_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + state["acc_delta"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + square_avgs.append(state["square_avg"]) + acc_deltas.append(state["acc_delta"]) + state_steps.append(state["step"]) + + with torch.no_grad(): + F.adadelta( + params_with_grad, + grads, + square_avgs, + acc_deltas, + state_steps, + lr=lr, + rho=rho, + eps=eps, + weight_decay=weight_decay, + foreach=self.foreach, + maximize=self.maximize, + has_complex=has_complex, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adagrad.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adagrad.py new file mode 100644 index 0000000000000000000000000000000000000000..e4f0dcaf1f9676444904c0d47014b6d6fcf18871 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adagrad.py @@ -0,0 +1,113 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional Adagrad Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly let the user pass gradients to the `step` function +# this is so that we could separate the gradients and parameters +# and allow multithreaded trainer to update the parameters +# without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalAdagrad: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-2, + lr_decay: float = 0.0, + weight_decay: float = 0.0, + initial_accumulator_value: float = 0.0, + warmup_lr_multiplier: float = 1.0, + warmup_num_iters: float = 0.0, + eps: float = 1e-10, + coalesce_grad: bool = True, + foreach: bool = False, + fused: bool = False, + maximize: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + self.defaults = { + "lr": lr, + "lr_decay": lr_decay, + "eps": eps, + "weight_decay": weight_decay, + "initial_accumulator_value": initial_accumulator_value, + "warmup_lr_multiplier": warmup_lr_multiplier, + "warmup_num_iters": warmup_num_iters, + } + self.coalesce_grad = coalesce_grad + self.foreach = foreach + self.fused = fused + self.maximize = maximize + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + # TODO: no union or any types in TorchScript, make step a scalar tensor instead + # This is also needed by if we want to share_memory on the step across processes + for p in self.param_group["params"]: + self.state[p] = { + "sum": torch.full_like(p.data, initial_accumulator_value), + "step": torch.tensor(0.0), + } + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + state_sums = [] + state_steps: list[Tensor] = [] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_sparse_grad, has_complex = False, False + for param, gradient in zip(self.param_group["params"], gradients): + if gradient is not None: + has_sparse_grad |= gradient.is_sparse + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + state = self.state[param] + state_sums.append(state["sum"]) + state_steps.append(state["step"]) + + with torch.no_grad(): + F.adagrad( + params, + grads, + state_sums, + state_steps, + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + lr_decay=self.defaults["lr_decay"], + eps=self.defaults["eps"], + has_sparse_grad=has_sparse_grad, + foreach=self.foreach, + maximize=self.maximize, + has_complex=has_complex, + fused=self.fused, + grad_scale=None, + found_inf=None, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adam.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adam.py new file mode 100644 index 0000000000000000000000000000000000000000..709d8c8af71994d79093d7e5ef725c70118dbbda --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adam.py @@ -0,0 +1,200 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional Adam Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalAdam: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-3, + betas: tuple[float, float] = (0.9, 0.999), + eps: float = 1e-8, + weight_decay: float = 0.0, + amsgrad: bool = False, + maximize: bool = False, + foreach: bool = False, + fused: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + if not 0.0 <= lr: + raise ValueError(f"Invalid learning rate: {lr}") + if not 0.0 <= eps: + raise ValueError(f"Invalid epsilon value: {eps}") + if not 0.0 <= betas[0] < 1.0: + raise ValueError(f"Invalid beta parameter at index 0: {betas[0]}") + if not 0.0 <= betas[1] < 1.0: + raise ValueError(f"Invalid beta parameter at index 1: {betas[1]}") + if not 0.0 <= weight_decay: + raise ValueError(f"Invalid weight_decay value: {weight_decay}") + + self.defaults = { + "lr": lr, + "eps": eps, + "beta1": betas[0], + "beta2": betas[1], + "weight_decay": weight_decay, + } + self.amsgrad = amsgrad + self.maximize = maximize + self.foreach = foreach + self.fused = fused + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + def step_param(self, param: Tensor, grad: Tensor | None): + """ + Similar to step, but operates on a single parameter and optionally a + gradient tensor. + """ + params_with_grad = [] + grads = [] + exp_avgs = [] + exp_avg_sqs = [] + max_exp_avg_sqs = [] + state_steps: list[Tensor] = [] + has_complex = torch.is_complex(param) + if grad is not None: + params_with_grad.append(param) + grads.append(grad) + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + state["exp_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + state["exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if self.amsgrad: + state["max_exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + exp_avgs.append(state["exp_avg"]) + exp_avg_sqs.append(state["exp_avg_sq"]) + + if self.amsgrad: + max_exp_avg_sqs.append(state["max_exp_avg_sq"]) + + state_steps.append(state["step"]) + with torch.no_grad(): + F.adam( + params_with_grad, + grads, + exp_avgs, + exp_avg_sqs, + max_exp_avg_sqs, + state_steps, + amsgrad=self.amsgrad, + has_complex=has_complex, + maximize=self.maximize, + beta1=self.defaults["beta1"], + beta2=self.defaults["beta2"], + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + eps=self.defaults["eps"], + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + ) + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + exp_avgs = [] + exp_avg_sqs = [] + max_exp_avg_sqs = [] + state_steps: list[Tensor] = [] + has_complex = False + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + for param, gradient in zip(self.param_group["params"], gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + # Exponential moving average of gradient values + state["exp_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + # Exponential moving average of squared gradient values + state["exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if self.amsgrad: + # Maintains max of all exp. moving avg. of sq. grad. values + state["max_exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + + exp_avgs.append(state["exp_avg"]) + exp_avg_sqs.append(state["exp_avg_sq"]) + + if self.amsgrad: + max_exp_avg_sqs.append(state["max_exp_avg_sq"]) + + state_steps.append(state["step"]) + + with torch.no_grad(): + F.adam( + params_with_grad, + grads, + exp_avgs, + exp_avg_sqs, + max_exp_avg_sqs, + state_steps, + amsgrad=self.amsgrad, + has_complex=has_complex, + maximize=self.maximize, + beta1=self.defaults["beta1"], + beta2=self.defaults["beta2"], + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + eps=self.defaults["eps"], + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamax.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamax.py new file mode 100644 index 0000000000000000000000000000000000000000..b17480d024c5d4b6a509d909af3f549b7acedcb7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamax.py @@ -0,0 +1,121 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional Adamax Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalAdamax: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-3, + betas: tuple[float, float] = (0.9, 0.999), + eps: float = 1e-8, + weight_decay: float = 0.0, + foreach: bool = False, + maximize: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + if not 0.0 <= lr: + raise ValueError(f"Invalid learning rate: {lr}") + if not 0.0 <= eps: + raise ValueError(f"Invalid epsilon value: {eps}") + if not 0.0 <= betas[0] < 1.0: + raise ValueError(f"Invalid beta parameter at index 0: {betas[0]}") + if not 0.0 <= betas[1] < 1.0: + raise ValueError(f"Invalid beta parameter at index 1: {betas[1]}") + if not 0.0 <= weight_decay: + raise ValueError(f"Invalid weight_decay value: {weight_decay}") + + self.defaults = { + "lr": lr, + "eps": eps, + "beta1": betas[0], + "beta2": betas[1], + "weight_decay": weight_decay, + } + self.foreach = foreach + self.maximize = maximize + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + exp_avgs = [] + exp_infs = [] + state_steps: list[Tensor] = [] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_complex = False + for param, gradient in zip(self.param_group["params"], gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + # Exponential moving average of gradient values + state["exp_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + # Exponential moving average of squared gradient values + state["exp_inf"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + + exp_avgs.append(state["exp_avg"]) + exp_infs.append(state["exp_inf"]) + state_steps.append(state["step"]) + + with torch.no_grad(): + F.adamax( + params_with_grad, + grads, + exp_avgs, + exp_infs, + state_steps, + eps=self.defaults["eps"], + beta1=self.defaults["beta1"], + beta2=self.defaults["beta2"], + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + foreach=self.foreach, + maximize=self.maximize, + has_complex=has_complex, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamw.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamw.py new file mode 100644 index 0000000000000000000000000000000000000000..f70844681187b7e3814bf1c862af35e352823568 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adamw.py @@ -0,0 +1,201 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional AdamW Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalAdamW: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-3, + betas: tuple[float, float] = (0.9, 0.999), + eps: float = 1e-8, + weight_decay: float = 1e-2, + amsgrad: bool = False, + maximize: bool = False, + foreach: bool = False, + fused: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + if not 0.0 <= lr: + raise ValueError(f"Invalid learning rate: {lr}") + if not 0.0 <= eps: + raise ValueError(f"Invalid epsilon value: {eps}") + if not 0.0 <= betas[0] < 1.0: + raise ValueError(f"Invalid beta parameter at index 0: {betas[0]}") + if not 0.0 <= betas[1] < 1.0: + raise ValueError(f"Invalid beta parameter at index 1: {betas[1]}") + if not 0.0 <= weight_decay: + raise ValueError(f"Invalid weight_decay value: {weight_decay}") + + self.defaults = { + "lr": lr, + "eps": eps, + "beta1": betas[0], + "beta2": betas[1], + "weight_decay": weight_decay, + } + self.amsgrad = amsgrad + self.maximize = maximize + self.foreach = foreach + self.fused = fused + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + def step_param(self, param: Tensor, grad: Tensor | None): + params_with_grad = [] + grads = [] + exp_avgs = [] + exp_avg_sqs = [] + max_exp_avg_sqs = [] + state_steps: list[Tensor] = [] + has_complex = torch.is_complex(param) + if grad is not None: + params_with_grad.append(param) + grads.append(grad) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + # Exponential moving average of gradient values + state["exp_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + # Exponential moving average of squared gradient values + state["exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if self.amsgrad: + # Maintains max of all exp. moving avg. of sq. grad. values + state["max_exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + + exp_avgs.append(state["exp_avg"]) + exp_avg_sqs.append(state["exp_avg_sq"]) + + if self.amsgrad: + max_exp_avg_sqs.append(state["max_exp_avg_sq"]) + + state_steps.append(state["step"]) + with torch.no_grad(): + F.adamw( + params_with_grad, + grads, + exp_avgs, + exp_avg_sqs, + max_exp_avg_sqs, + state_steps, + amsgrad=self.amsgrad, + maximize=self.maximize, + beta1=self.defaults["beta1"], + beta2=self.defaults["beta2"], + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + eps=self.defaults["eps"], + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + has_complex=has_complex, + ) + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + exp_avgs = [] + exp_avg_sqs = [] + max_exp_avg_sqs = [] + state_steps: list[Tensor] = [] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_complex = False + for param, gradient in zip(self.param_group["params"], gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + # Exponential moving average of gradient values + state["exp_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + # Exponential moving average of squared gradient values + state["exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if self.amsgrad: + # Maintains max of all exp. moving avg. of sq. grad. values + state["max_exp_avg_sq"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + + exp_avgs.append(state["exp_avg"]) + exp_avg_sqs.append(state["exp_avg_sq"]) + + if self.amsgrad: + max_exp_avg_sqs.append(state["max_exp_avg_sq"]) + + state_steps.append(state["step"]) + + with torch.no_grad(): + F.adamw( + params_with_grad, + grads, + exp_avgs, + exp_avg_sqs, + max_exp_avg_sqs, + state_steps, + amsgrad=self.amsgrad, + maximize=self.maximize, + beta1=self.defaults["beta1"], + beta2=self.defaults["beta2"], + lr=self.defaults["lr"], + weight_decay=self.defaults["weight_decay"], + eps=self.defaults["eps"], + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + has_complex=has_complex, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rmsprop.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rmsprop.py new file mode 100644 index 0000000000000000000000000000000000000000..55368f253baf744e30802549bff666fcef95e6d1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rmsprop.py @@ -0,0 +1,128 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional RMSprop Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalRMSprop: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-2, + alpha: float = 0.99, + eps: float = 1e-8, + weight_decay: float = 0.0, + momentum: float = 0.0, + centered: bool = False, + foreach: bool = False, + maximize: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + self.defaults = { + "lr": lr, + "alpha": alpha, + "eps": eps, + "weight_decay": weight_decay, + "momentum": momentum, + } + self.centered = centered + self.foreach = foreach + self.maximize = maximize + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + square_avgs = [] + grad_avgs = [] + momentum_buffer_list = [] + state_steps = [] + lr = self.defaults["lr"] + alpha = self.defaults["alpha"] + eps = self.defaults["eps"] + momentum = self.defaults["momentum"] + weight_decay = self.defaults["weight_decay"] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_complex = False + for param, gradient in zip(params, gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + state["square_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if momentum > 0: + state["momentum_buffer"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + if self.centered: + state["grad_avg"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + + state = self.state[param] + square_avgs.append(state["square_avg"]) + if momentum > 0: + momentum_buffer_list.append(state["momentum_buffer"]) + if self.centered: + grad_avgs.append(state["grad_avg"]) + + state_steps.append(state["step"]) + + with torch.no_grad(): + F.rmsprop( + params_with_grad, + grads, + square_avgs, + grad_avgs, + momentum_buffer_list, + state_steps, + lr=lr, + alpha=alpha, + eps=eps, + weight_decay=weight_decay, + momentum=momentum, + centered=self.centered, + foreach=self.foreach, + maximize=self.maximize, + has_complex=has_complex, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rprop.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rprop.py new file mode 100644 index 0000000000000000000000000000000000000000..9896012c196abccd060e390a0b5c70c972f13e53 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_rprop.py @@ -0,0 +1,105 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional Rprop Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalRprop: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-2, + etas: tuple[float, float] = (0.5, 1.2), + step_sizes: tuple[float, float] = (1e-6, 50), + foreach: bool = False, + maximize: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + self.defaults = { + "lr": lr, + } + self.etas = etas + self.step_sizes = step_sizes + self.foreach = foreach + self.maximize = maximize + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + prevs = [] + step_sizes = [] + state_steps = [] + lr = self.defaults["lr"] + etaminus, etaplus = self.etas + step_size_min, step_size_max = self.step_sizes + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_complex = False + for param, gradient in zip(params, gradients): + if gradient is not None: + has_complex |= torch.is_complex(param) + params_with_grad.append(param) + grads.append(gradient) + # Lazy state initialization + if param not in self.state: + self.state[param] = {} + state = self.state[param] + state["step"] = torch.tensor(0.0) + state["prev"] = torch.zeros_like( + param, memory_format=torch.preserve_format + ) + state["step_size"] = torch.full_like(gradient, lr) + + state = self.state[param] + prevs.append(state["prev"]) + step_sizes.append(state["step_size"]) + state_steps.append(state["step"]) + + with torch.no_grad(): + F.rprop( + params_with_grad, + grads, + prevs, + step_sizes, + state_steps, + step_size_min=step_size_min, + step_size_max=step_size_max, + etaminus=etaminus, + etaplus=etaplus, + foreach=self.foreach, + maximize=self.maximize, + has_complex=has_complex, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_sgd.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_sgd.py new file mode 100644 index 0000000000000000000000000000000000000000..d9d2755da6bc41c74d7caae5cbb57f703c5605a5 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/functional_sgd.py @@ -0,0 +1,164 @@ +# mypy: allow-untyped-defs + +import torch +import torch.optim._functional as F +from torch import Tensor +from torch.distributed.optim._deprecation_warning import ( + _scripted_functional_optimizer_deprecation_warning, +) + + +__all__: list[str] = [] + + +# Define a Functional SGD Optimizer +# where we use these optimizer in a functional way. +# Instead of using the `param.grad` when updating parameters, +# we explicitly allow the distributed optimizer pass gradients to +# the `step` function. In this way, we could separate the gradients +# and parameters and allow multithreaded trainer to update the +# parameters without data traces on accumulating to the same .grad. +# NOTE: This should be only used by distributed optimizer internals +# and not meant to expose to the user. +class _FunctionalSGD: + def __init__( + self, + params: list[Tensor], + lr: float = 1e-2, + momentum: float = 0.0, + dampening: float = 0.0, + weight_decay: float = 0.0, + nesterov: bool = False, + maximize: bool = False, + foreach: bool = False, + fused: bool = False, + _allow_empty_param_list: bool = False, + ): + _scripted_functional_optimizer_deprecation_warning(stacklevel=2) + self.defaults = { + "lr": lr, + "momentum": momentum, + "dampening": dampening, + "weight_decay": weight_decay, + } + self.nesterov = nesterov + self.maximize = maximize + self.foreach = foreach + self.fused = fused + self.state = torch.jit.annotate(dict[torch.Tensor, dict[str, torch.Tensor]], {}) + + if len(params) == 0 and not _allow_empty_param_list: + raise ValueError("optimizer got an empty parameter list") + + # NOTE: we only have one param_group and don't allow user to add additional + # param group as it's not a common use case. + self.param_group = {"params": params} + + def step_param(self, param: Tensor, grad: Tensor | None): + """Similar to self.step, but operates on a single parameter and + its gradient. + """ + # TODO: Once step_param interface is robust, refactor step to call + # step param on each param. + weight_decay = self.defaults["weight_decay"] + momentum = self.defaults["momentum"] + dampening = self.defaults["dampening"] + lr = self.defaults["lr"] + params = [param] + momentum_buffer_list: list[Tensor | None] = [] + grads = [] + + has_sparse_grad = False + if grad is not None: + grads.append(grad) + if grad.is_sparse: + has_sparse_grad = True + if param not in self.state: + self.state[param] = {} + state = self.state[param] + if "momentum_buffer" not in state: + momentum_buffer_list.append(None) + else: + momentum_buffer_list.append(state["momentum_buffer"]) + + with torch.no_grad(): + F.sgd( + params, + grads, + momentum_buffer_list, + weight_decay=weight_decay, + momentum=momentum, + lr=lr, + dampening=dampening, + nesterov=self.nesterov, + maximize=self.maximize, + has_sparse_grad=has_sparse_grad, + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + ) + # update momentum_buffer in state + state = self.state[param] + momentum_buffer = momentum_buffer_list[0] + if momentum_buffer is not None: + state["momentum_buffer"] = momentum_buffer + + def step(self, gradients: list[Tensor | None]): + params = self.param_group["params"] + params_with_grad = [] + grads = [] + momentum_buffer_list: list[Tensor | None] = [] + lr = self.defaults["lr"] + weight_decay = self.defaults["weight_decay"] + momentum = self.defaults["momentum"] + dampening = self.defaults["dampening"] + + if len(params) != len(gradients): + raise ValueError( + "the gradients passed in does not equal to the size of the parameters!" + + f"Params length: {len(params)}. " + + f"Gradients length: {len(gradients)}" + ) + + has_sparse_grad = False + for param, gradient in zip(params, gradients): + if gradient is not None: + params_with_grad.append(param) + grads.append(gradient) + if gradient.is_sparse: + has_sparse_grad = True + + if param not in self.state: + self.state[param] = {} + + state = self.state[param] + if "momentum_buffer" not in state: + momentum_buffer_list.append(None) + else: + momentum_buffer_list.append(state["momentum_buffer"]) + + with torch.no_grad(): + F.sgd( + params_with_grad, + grads, + momentum_buffer_list, + weight_decay=weight_decay, + momentum=momentum, + lr=lr, + dampening=dampening, + nesterov=self.nesterov, + maximize=self.maximize, + has_sparse_grad=has_sparse_grad, + foreach=self.foreach, + fused=self.fused, + grad_scale=None, + found_inf=None, + ) + + # update momentum_buffers in state + for i, p in enumerate(params_with_grad): + state = self.state[p] + momentum_buffer = momentum_buffer_list[i] + if momentum_buffer is not None: + state["momentum_buffer"] = momentum_buffer diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/named_optimizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/named_optimizer.py new file mode 100644 index 0000000000000000000000000000000000000000..d434e94abf42773ea0d00676efb253b1d7377034 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/named_optimizer.py @@ -0,0 +1,333 @@ +import logging +import warnings +from collections.abc import Callable, Collection, Mapping +from copy import deepcopy +from typing import Any, overload + +import torch +import torch.nn as nn +from torch import optim +from torch.distributed._shard.sharded_tensor import ShardedTensor +from torch.distributed.fsdp import FullyShardedDataParallel as FSDP + + +__all__: list[str] = [] + +logger = logging.getLogger(__name__) + + +class _NamedOptimizer(optim.Optimizer): + """ + ``_NamedOptimizer`` takes a dict of parameters and exposes ``state_dict`` by parameter key. + + We replace the original key (number) in an optim to the + fully qualified name (FQN) string. User can initialize the optim as they + initialize a PyTorch optim, the only difference is that they also need to + pass in the FQN of each parameters. + + Args: + named_parameters (Mapping[str, Union[torch.Tensor, ShardedTensor]]): + Mapping from FQN to parameter. + optimizer_class (optim.Optimizer): + The class of optimizer to instantiate. + param_groups (Collection[Mapping[str, Any]]): + `param_groups` to pass to optimizer if specified. + The key of the inner map needs to be FQNs. + Default: None + module (nn.Module): the module whose parameters to be updated + by the optimizer. + args: arguments to pass to the optimizer constructor. + kwargs: arguments to pass to the optimizer constructor. + + Example:: + >>> # xdoctest: +SKIP("distributed") + >>> from torch import optim + >>> from torch.distributed.optim import _NamedOptimizer + >>> + >>> # Define the named optimizer. + >>> m = Model(...) + >>> named_optim = _NamedOptimizer(m.named_parameters(), optim.SGD) + >>> # Forward pass + backward pass. + >>> named_optim.step() + >>> ... + >>> # Call state_dict for the named optimizer returns an FQN state_dict. + >>> named_optim.state_dict() + + Warning: This API is still in development and subject to change. + + TODO: Add tutorial for _NamedOptimizer. + TODO: Add documentation in the docstring for the public attributes + like self.param_groups and self.named_parameters. + """ + + def __init__( + self, + named_parameters: Mapping[str, torch.Tensor | ShardedTensor], + optimizer_class: optim.Optimizer, + param_groups: Collection[Mapping[str, Any]] | None = None, + module: nn.Module | None = None, + *args: tuple[Any, ...], + **kwargs: dict[str, Any], + ) -> None: + torch._C._log_api_usage_once("torch.distributed.optim._NamedOptimizer") + self.param_groups: Collection[Mapping[str, Any]] = param_groups # type: ignore[assignment] + self._param_groups_check() + self.named_parameters = dict(named_parameters) + params_for_optimizer = ( + self.named_parameters.values() if param_groups is None else param_groups + ) + self._optimizer = optimizer_class( # type: ignore[operator] + params_for_optimizer, + *args, + **kwargs, + ) + self.module = module + if param_groups is None: + self.ordered_param_keys = list(self.named_parameters.keys()) + else: + warnings.warn( + "Since we pass in param_groups, we will use param_groups to " + "initialize the optimizer, not all parameters of the module.", + stacklevel=2, + ) + param_to_key = {param: key for key, param in self.named_parameters.items()} # type: ignore[misc, has-type] + ordered_param_keys = [] + for group in param_groups: + for param in group["params"]: + if param not in param_to_key: + raise ValueError( + f"Expect param name {param} found in param group but is missing." + ) + ordered_param_keys.append(param_to_key[param]) + self.ordered_param_keys = ordered_param_keys + # Update param_groups from optimizer. + self.param_groups = self._optimizer.param_groups + + def _param_groups_check(self) -> None: + if self.param_groups is not None: + for param_group in self.param_groups: + if not isinstance(param_group, dict): + raise AssertionError("param group must be a dict") + if "params" not in param_group: + raise AssertionError("param group must contain key params") + params = param_group["params"] + if isinstance(params, torch.Tensor): + params = [params] + params = list(params) + for param in params: + if not isinstance(param, torch.Tensor): + raise TypeError( + "optimizer can only optimize Tensors, " + "but one of the params is " + torch.typename(param) + ) + param_group["params"] = params + + def state_dict(self) -> dict[str, Any]: + """ + Return the ``state_dict`` of the optimizer. + + Instead of using number to index + parameters, we will use module fully qualified name (FQN) as the key. + """ + state_dict = self._optimizer.state_dict() + param_groups = state_dict["param_groups"] + + ret_state = { + self.ordered_param_keys[st_key]: state_val + for st_key, state_val in state_dict["state"].items() + } + + ret_groups = [] + for group in param_groups: + param_keys = [self.ordered_param_keys[param] for param in group["params"]] + ret_group = {"params": sorted(param_keys)} + for k, v in group.items(): + if k != "params": + ret_group[k] = deepcopy(v) + ret_groups.append(ret_group) + + return self._post_state_dict({"state": ret_state, "param_groups": ret_groups}) + + @overload + def step(self, closure: None = None) -> None: ... + + @overload + def step(self, closure: Callable[[], float]) -> float: ... + + def step(self, closure: Callable[[], float] | None = None) -> float | None: + """ + Perform a single optimization step. + + This will call :meth:`torch.optim.Optimizer.step` on the wrapped + optimizer. + """ + return self._optimizer.step(closure=closure) + + @property + def state(self) -> Mapping[torch.Tensor, Any]: # type: ignore[override] + return self._optimizer.state + + def load_state_dict(self, state_dict: dict[str, Any]) -> None: + """ + Define the default behavior to load a state_dict for ``_NamedOptimizer``. + + Sample Code + ``` + my_model = MyModule() + optimizer = _NamedOptimizer(my_model.named_parameters(), Adagrad) + ... + + optim_state_dict = optimizer.state_dict() + ... + ... + + optimizer.load_state_dict(optim_state_dict) + ... + ``` + Args: + state_dict (dict[str, Any]) : A ``state_dict`` to load into the optimizer. + Note that this state dict update is performed in place. + + .. note:: PyTorch is using lazy init to initialize the optim states. + So it is possible that there is no optim state when user call + ``load_state_dict`` and for ``_NamedOptimizer`` we make it stricter + that users can only call ``load_state_dict`` after the state is initialized. + By doing this, we can validate the optim ``state_dict`` to be loaded. + """ + new_state_dict = self._optimizer.state_dict() + state_dict = self._pre_load_state_dict(state_dict) + state = state_dict["state"] + new_state = new_state_dict["state"] + if len(new_state) == 0: + raise ValueError( + "Expects the optim to be initialized before load but found not initialized." + ) + + for idx, param_key in enumerate(self.ordered_param_keys): + # When the conditional training is performed, not all parameters are updated in the optim. + if param_key not in state: + continue + if len(state[param_key]) != len(new_state[idx]): + raise ValueError( + f"Expects equal length as {len(new_state[idx])} for parameter {param_key} but found: {len(state[param_key])}" + ) + # Iterate through all optimizer states. + for state_key, state_val in new_state[idx].items(): + if state_key not in state[param_key]: + raise ValueError( + f"Expects state {state_key} for parameter {param_key} but not found." + ) + + src_state_val = state[param_key][state_key] + if isinstance(state_val, ShardedTensor): + if not isinstance(src_state_val, ShardedTensor): + raise AssertionError + num_shards = len(state_val.local_shards()) + num_new_shards = len(src_state_val.local_shards()) + if num_shards != num_new_shards: + raise ValueError( + f"Expects equal number of shards as {num_new_shards} but found {num_shards} for {param_key}/{state_key}" + ) + for shard, src_shard in zip( + state_val.local_shards(), src_state_val.local_shards() + ): + shard.tensor.detach().copy_(src_shard.tensor) + elif isinstance(state_val, torch.Tensor): + if not isinstance(src_state_val, torch.Tensor): + raise AssertionError + state_val.detach().copy_(src_state_val) + else: + new_state[idx][state_key] = deepcopy(src_state_val) + + # Load param_groups of state_dict + src_param_groups = state_dict["param_groups"] + new_param_groups = new_state_dict["param_groups"] + + src_group_map = {} + for group in src_param_groups: + param_keys = list(group["params"]) + src_group_map[_gen_param_group_key(param_keys)] = group + new_group_map = {} + for new_group in new_param_groups: + param_keys = [] + for param_key in new_group["params"]: + param_keys.append(self.ordered_param_keys[param_key]) # type: ignore[call-overload] + new_group_map[_gen_param_group_key(param_keys)] = new_group + for group_key, new_group in new_group_map.items(): + # When not all parameters are used in training or receive gradient, aka., not all parameters + # would be in the param_group. Thus we skip the group_key here. + if group_key not in src_group_map: + continue + src_group = src_group_map[group_key] + if len(src_group) != len(new_group): + raise ValueError( + f"Expects equal param_group size as {len(new_group)} for group {group_key} but found {len(src_group)}." + ) + for k in src_group: + if k not in new_group: + raise ValueError( + f"Expects group key {k} to be in group {group_key} in `state_dict` but is missing." + ) + if k != "params": + new_group[k] = deepcopy(src_group[k]) + + self._optimizer.load_state_dict(new_state_dict) + + def add_param_group(self, param_group: Mapping[str, Any]) -> None: + """ + Add a param group to the :class:`_NamedOptimizer` s `param_groups`. + + Warning: This API is still in development and subject to change. + """ + if not isinstance(param_group, dict): + raise AssertionError("param group must be a dict") + + params = param_group["params"] + if isinstance(params, torch.Tensor): + param_group["params"] = [params] + else: + param_group["params"] = list(params) + + param_to_key = {param: key for key, param in self.named_parameters.items()} # type: ignore[misc, has-type] + for param in param_group["params"]: + if param not in param_to_key: + raise ValueError("some parameters are not in the module") + self.ordered_param_keys.append(param_to_key[param]) + + self._optimizer.add_param_group(param_group) + # Update param_groups from optimizer. + self.param_groups = self._optimizer.param_groups + + def init_state(self) -> None: + """ + Run a dummy optimizer step, which allows us to initialize optimizer state because we do lazy init for most optimizers. + + This allows doing in-place loading of optimizer state from a checkpoint. + """ + for param in self.named_parameters.values(): + if param.requires_grad: + t = torch.zeros_like(param) + param.grad = torch.autograd.Variable(t) + # Calling ``step`` will load the initial state for optimizer states. + self.step(closure=None) + + def _pre_load_state_dict(self, state_dict: dict[str, Any]) -> dict[str, Any]: + # TODO(chienchin): This API should be FSDP agnostic and should support + # general user hooks. + if isinstance(self.module, FSDP): + return FSDP.optim_state_dict_to_load( + self.module, self._optimizer, state_dict, is_named_optimizer=True + ) + return state_dict + + def _post_state_dict(self, state_dict: dict[str, Any]) -> dict[str, Any]: + # TODO(chienchin): This API should be FSDP agnostic and should support + # general user hooks. + if isinstance(self.module, FSDP): + FSDP.optim_state_dict(self.module, self._optimizer, state_dict) + return state_dict + + +def _gen_param_group_key(param_keys: list[str]) -> str: + """Concatenate all param keys as a unique identifier for one param group.""" + return "/".join(sorted(param_keys)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/optimizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/optimizer.py new file mode 100644 index 0000000000000000000000000000000000000000..ec0cfd0dd770c5529b0061572472d48709e87c03 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/optimizer.py @@ -0,0 +1,222 @@ +# mypy: allow-untyped-defs +import logging +from collections import defaultdict +from threading import Lock + +import torch +import torch.distributed.autograd as dist_autograd +import torch.distributed.rpc as rpc +import torch.jit as jit +from torch import Tensor +from torch.distributed.rpc import RRef + +from .utils import functional_optim_map + + +__all__ = ["DistributedOptimizer"] + +logger = logging.getLogger(__name__) + + +class _FunctionalLocalOptimizer: + def __init__(self, optim_cls, local_params_rref, *args, **kwargs): + self._local_params = [rref.local_value() for rref in local_params_rref] + self.optim = optim_cls(self._local_params, *args, **kwargs) + + def step(self, autograd_ctx_id: int): + all_local_grads = dist_autograd.get_gradients(autograd_ctx_id) + # apply functional optimizer step with a list of gradients + grads: list[Tensor | None] = [ + all_local_grads[p] if p in all_local_grads else None # noqa: SIM401 + for p in self._local_params + ] + + self.optim.step(grads) + + +# TODO (wanchaol): remove/merge this with _FunctionalLocalOptimizer once +# we have converted all to functional optimizer in distributed.optim +class _LocalOptimizer: + # Ideally we would only need to share a lock for instances of + # _LocalOptimizer that deal with the same parameters. We are + # making a simplifying assumption here that if there is more + # than one instance of _LocalOptimizer per worker, they will + # be optimizing the same parameters (e.g. each data parallel + # trainer will create its own instance of _LocalOptimizer but + # they will all optimize the same parameters on each worker) + global_lock = Lock() + + def __init__(self, optim_cls, local_params_rref, *args, **kwargs): + self._local_params = [rref.local_value() for rref in local_params_rref] + self.optim = optim_cls(self._local_params, *args, **kwargs) + + def step(self, autograd_ctx_id): + all_local_grads = dist_autograd.get_gradients(autograd_ctx_id) + + with _LocalOptimizer.global_lock: + for param, grad in all_local_grads.items(): + param.grad = grad + self.optim.step() + + +def _new_local_optimizer(optim_cls, local_params_rref, *args, **kwargs): + return rpc.RRef(_LocalOptimizer(optim_cls, local_params_rref, *args, **kwargs)) + + +def _local_optimizer_step(local_optim_rref, autograd_ctx_id): + local_optim = local_optim_rref.local_value() + local_optim.step(autograd_ctx_id) + + +def _new_functional_local_optimizer(optim_cls, local_params_rref, *args, **kwargs): + optim = _FunctionalLocalOptimizer(optim_cls, local_params_rref, *args, **kwargs) + return rpc.RRef(optim) + + +def _functional_local_optimizer_step( + local_optim_rref: RRef, autograd_ctx_id: int +) -> None: + local_optim = local_optim_rref.local_value() + local_optim.step(autograd_ctx_id) + + +def _wait_for_all(rpc_futs): + # TODO: improve error propagation + exception = None + results = [] + for fut in rpc_futs: + try: + results.append(fut.wait()) + except Exception as e: + results.append(e) + exception = e + if exception is not None: + raise exception + return results + + +class DistributedOptimizer: + """ + DistributedOptimizer takes remote references to parameters scattered + across workers and applies the given optimizer locally for each parameter. + + This class uses :meth:`~torch.distributed.autograd.get_gradients` in order + to retrieve the gradients for specific parameters. + + Concurrent calls to + :meth:`~torch.distributed.optim.DistributedOptimizer.step`, + either from the same or different clients, will + be serialized on each worker -- as each worker's optimizer can only work + on one set of gradients at a time. However, there is no guarantee that + the full forward-backward-optimizer sequence will execute for one client + at a time. This means that the gradients being applied may not correspond + to the latest forward pass executed on a given worker. Also, there is no + guaranteed ordering across workers. + + `DistributedOptimizer` uses a functional optimizer internally when one is + available for the given ``optimizer_class``, so that optimizer updates are + not blocked by the Python Global Interpreter Lock (GIL) in the case of + multithreaded training (e.g. Distributed Model Parallel). This feature is + currently enabled for most optimizers. + + Args: + optimizer_class (optim.Optimizer): the class of optimizer to + instantiate on each worker. + params_rref (list[RRef]): list of RRefs to local or remote parameters + to optimize. + args: arguments to pass to the optimizer constructor on each worker. + kwargs: arguments to pass to the optimizer constructor on each worker. + + Example:: + >>> # xdoctest: +SKIP("distributed") + >>> import torch.distributed.autograd as dist_autograd + >>> import torch.distributed.rpc as rpc + >>> from torch import optim + >>> from torch.distributed.optim import DistributedOptimizer + >>> + >>> with dist_autograd.context() as context_id: + >>> # Forward pass. + >>> rref1 = rpc.remote("worker1", torch.add, args=(torch.ones(2), 3)) + >>> rref2 = rpc.remote("worker1", torch.add, args=(torch.ones(2), 1)) + >>> loss = rref1.to_here() + rref2.to_here() + >>> + >>> # Backward pass. + >>> dist_autograd.backward(context_id, [loss.sum()]) + >>> + >>> # Optimizer. + >>> dist_optim = DistributedOptimizer( + >>> optim.SGD, + >>> [rref1, rref2], + >>> lr=0.05, + >>> ) + >>> dist_optim.step(context_id) + """ + + def __init__(self, optimizer_class, params_rref, *args, **kwargs): + torch._C._log_api_usage_once("torch.distributed.optim.DistributedOptimizer") + per_worker_params_rref = defaultdict(list) + for param in params_rref: + per_worker_params_rref[param.owner()].append(param) + + if optimizer_class in functional_optim_map and jit._state._enabled: + optim_ctor = functional_optim_map.get(optimizer_class) + else: + optim_ctor = optimizer_class + self.is_functional_optim = optim_ctor != optimizer_class + + if self.is_functional_optim: + optimizer_new_func = _new_functional_local_optimizer + else: + logger.warning( + "Creating the optimizer %s without a functional optimizer, " + "this might result in slow computation time in multithreading environment" + "(i.e. Distributed Model Parallel training on CPU) due to the Python's " + "Global Interpreter Lock (GIL). Please file an issue if you need a " + "functional optimizer for this optimizer. ", + optimizer_class, + ) + optimizer_new_func = _new_local_optimizer + + remote_optim_futs = [] + for worker, param_rrefs in per_worker_params_rref.items(): + remote_optim_rref_fut = rpc.rpc_async( + worker, + optimizer_new_func, + args=(optim_ctor, param_rrefs) + args, + kwargs=kwargs, + ) + remote_optim_futs.append(remote_optim_rref_fut) + + self.remote_optimizers = _wait_for_all(remote_optim_futs) + + def step(self, context_id): + """ + Performs a single optimization step. + + This will call :meth:`torch.optim.Optimizer.step` on each worker + containing parameters to be optimized, and will block until all workers + return. The provided ``context_id`` will be used to retrieve the + corresponding :class:`~torch.distributed.autograd.context` that + contains the gradients that should be applied to the parameters. + + Args: + context_id: the autograd context id for which we should run the + optimizer step. + """ + dist_autograd._is_valid_context(context_id) + + optimizer_step_func = ( + _functional_local_optimizer_step + if self.is_functional_optim + else _local_optimizer_step + ) + + rpc_futs = [ + rpc.rpc_async( + optimizer.owner(), + optimizer_step_func, + args=(optimizer, context_id), + ) + for optimizer in self.remote_optimizers + ] + _wait_for_all(rpc_futs) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/post_localSGD_optimizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/post_localSGD_optimizer.py new file mode 100644 index 0000000000000000000000000000000000000000..c7b78510ed1a111998a4eda21546b003eedbcce7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/post_localSGD_optimizer.py @@ -0,0 +1,111 @@ +# mypy: allow-untyped-defs +import warnings + +import torch +import torch.distributed.algorithms.model_averaging.averagers as averagers + + +class PostLocalSGDOptimizer(torch.optim.Optimizer): + r""" + Wraps an arbitrary :class:`torch.optim.Optimizer` and runs `post-local SGD `_, + This optimizer runs local optimizer at every step. + After the warm-up stage, it averages parameters periodically after the local optimizer is applied. + + Args: + optim: The local optimizer. + averager: A model averager instance to run post-localSGD algorithm. + + Example:: + + >>> # xdoctest: +SKIP("undefined variables") + >>> import torch + >>> import torch.distributed as dist + >>> import torch.distributed.algorithms.model_averaging.averagers as averagers + >>> import torch.nn as nn + >>> from torch.distributed.optim import PostLocalSGDOptimizer + >>> from torch.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook import ( + >>> PostLocalSGDState, + >>> post_localSGD_hook, + >>> ) + >>> + >>> model = nn.parallel.DistributedDataParallel( + >>> module, device_ids=[rank], output_device=rank + >>> ) + >>> + >>> # Register a post-localSGD communication hook. + >>> state = PostLocalSGDState(process_group=None, subgroup=None, start_localSGD_iter=100) + >>> model.register_comm_hook(state, post_localSGD_hook) + >>> + >>> # Create a post-localSGD optimizer that wraps a local optimizer. + >>> # Note that ``warmup_steps`` used in ``PostLocalSGDOptimizer`` must be the same as + >>> # ``start_localSGD_iter`` used in ``PostLocalSGDState``. + >>> local_optim = torch.optim.SGD(params=model.parameters(), lr=0.01) + >>> opt = PostLocalSGDOptimizer( + >>> optim=local_optim, + >>> averager=averagers.PeriodicModelAverager(period=4, warmup_steps=100) + >>> ) + >>> + >>> # In the first 100 steps, DDP runs global gradient averaging at every step. + >>> # After 100 steps, DDP runs gradient averaging within each subgroup (intra-node by default), + >>> # and post-localSGD optimizer runs global model averaging every 4 steps after applying the local optimizer. + >>> for step in range(0, 200): + >>> opt.zero_grad() + >>> loss = loss_fn(output, labels) + >>> loss.backward() + >>> opt.step() + """ + + def __init__(self, optim: torch.optim.Optimizer, averager: averagers.ModelAverager): + self.optim = optim + self.param_groups = self.optim.param_groups + self.averager = averager + + @property + def state(self): # type: ignore[override] + return self.optim.state + + def __repr__(self): + return self.optim.__repr__() + + def state_dict(self): + r""" + This is the same as :class:`torch.optim.Optimizer` :meth:`state_dict`, + but adds an extra entry to record model averager's step to the checkpoint + to ensure reload does not cause unnecessary warm up again. + """ + optim_state_dict = self.optim.state_dict() + optim_state_dict["step"] = self.averager.step + return optim_state_dict + + def load_state_dict(self, state_dict): + r""" + This is the same as :class:`torch.optim.Optimizer` :meth:`load_state_dict`, + but also restores model averager's step value to the one + saved in the provided ``state_dict``. + + If there is no ``"step"`` entry in ``state_dict``, + it will raise a warning and initialize the model averager's step to 0. + """ + self.optim.load_state_dict(state_dict) + if "step" in state_dict: + self.averager.step = state_dict["step"] + else: + warnings.warn( + "Loaded state dict does not contain a step counter for an averager. " + "Setting step counter to 0.", + stacklevel=2, + ) + self.averager.step = 0 + + def step(self): # type: ignore[override] + r""" + Performs a single optimization step (parameter update). + """ + self.optim.step() + self.averager.average_parameters(params=self.param_groups) + + def zero_grad(self, set_to_none: bool = True): # type: ignore[override] + self.optim.zero_grad(set_to_none=set_to_none) + + def add_param_group(self, param_group): + self.optim.add_param_group(param_group) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..867462d187ac55fc916b16fb74c50eb459fbed5d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/utils.py @@ -0,0 +1,67 @@ +# mypy: allow-untyped-defs + +from torch import optim + +from .functional_adadelta import _FunctionalAdadelta +from .functional_adagrad import _FunctionalAdagrad +from .functional_adam import _FunctionalAdam +from .functional_adamax import _FunctionalAdamax +from .functional_adamw import _FunctionalAdamW +from .functional_rmsprop import _FunctionalRMSprop +from .functional_rprop import _FunctionalRprop +from .functional_sgd import _FunctionalSGD + + +# dict to map a user passed in optimizer_class to a functional +# optimizer class if we have already defined inside the +# distributed.optim package, this is so that we hide the +# functional optimizer to user and still provide the same API. +functional_optim_map = { + optim.Adagrad: _FunctionalAdagrad, + optim.Adam: _FunctionalAdam, + optim.AdamW: _FunctionalAdamW, + optim.SGD: _FunctionalSGD, + optim.Adadelta: _FunctionalAdadelta, + optim.RMSprop: _FunctionalRMSprop, + optim.Rprop: _FunctionalRprop, + optim.Adamax: _FunctionalAdamax, +} + + +def register_functional_optim(key, optim): + """ + Interface to insert a new functional optimizer to functional_optim_map. + ``fn_optim_key`` and ``fn_optimizer`` are user defined. The optimizer and key + need not be of :class:`torch.optim.Optimizer` (e.g. for custom optimizers). + + Example:: + + >>> # import the new functional optimizer + >>> # xdoctest: +SKIP + >>> from xyz import fn_optimizer + >>> from torch.distributed.optim.utils import register_functional_optim + >>> fn_optim_key = "XYZ_optim" + >>> register_functional_optim(fn_optim_key, fn_optimizer) + """ + if key not in functional_optim_map: + functional_optim_map[key] = optim + + +def as_functional_optim(optim_cls: type, *args, **kwargs): + try: + functional_cls = functional_optim_map[optim_cls] + except KeyError as e: + raise ValueError( + f"Optimizer {optim_cls} does not have a functional counterpart!" + ) from e + + return _create_functional_optim(functional_cls, *args, **kwargs) + + +def _create_functional_optim(functional_optim_cls: type, *args, **kwargs): + return functional_optim_cls( + [], + *args, + **kwargs, + _allow_empty_param_list=True, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.py new file mode 100644 index 0000000000000000000000000000000000000000..0ad335d10d7d5a4afe94e4ec72b54aca419beaca --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.py @@ -0,0 +1,1700 @@ +# Copyright (c) Facebook, Inc. and its affiliates. All rights reserved. +# +# This source code is licensed under the BSD license found in the +# LICENSE file in the root directory of this source tree. + +r"""Zero Redundancy Optimizer.""" + +import collections +import copy +import enum +import inspect +import io +import logging +from collections.abc import Callable +from itertools import chain +from typing import Any + +import torch +import torch.distributed as dist +from torch.distributed.algorithms.join import Join, Joinable, JoinHook +from torch.distributed.optim.utils import functional_optim_map +from torch.optim import Optimizer + + +__all__ = ["ZeroRedundancyOptimizer"] + + +logger = logging.getLogger(__name__) + + +# Credits: classy_vision/generic/distributed_util.py +def _recursive_copy_to_device( + value: Any, + non_blocking: bool, + device: torch.device, +) -> Any: + r""" + Recursively searches lists, tuples, dicts and copies tensors to device if possible. + + Non-tensor values are passed as-is in the result. + + .. note:: + These are all copies, so if there are two objects that reference + the same object, then after this call, there will be two different objects + referenced on the device. + """ + if isinstance(value, torch.Tensor): + return value.to(device, non_blocking=non_blocking) + + if isinstance(value, (list, tuple)): + values = [ + _recursive_copy_to_device(val, non_blocking=non_blocking, device=device) + for val in value + ] + return values if isinstance(value, list) else tuple(values) + + if isinstance(value, collections.abc.Mapping): + return { + key: _recursive_copy_to_device( + val, non_blocking=non_blocking, device=device + ) + for key, val in value.items() + } + + return value + + +def _is_trainable(param: torch.Tensor) -> bool: + r"""Return if a parameter is trainable, where trainability is equivalent to requiring a gradient.""" + return param.requires_grad + + +def _broadcast_object( + obj: Any, + src_rank: int, + group: object = dist.group.WORLD, + device: torch.device = torch.device("cpu"), +) -> Any: + r""" + Broadcasts an object to the given group. + + It will be sending the object if called from the source rank and receiving + the object otherwise. + + Arguments: + obj: object to broadcast; only used if called on the source rank. + src_rank (int): source rank. + group (``ProcessGroup``, optional): group used for the broadcast + (default: ``dist.group.WORLD``). + device (``torch.device``, optional): device to send from or receive + to (default: ``torch.device("cpu")``). + + Returns: + The broadcasted object. + """ + if dist.get_rank() == src_rank: + # Send the object + buffer = io.BytesIO() + torch.save(obj, buffer) + data = bytearray(buffer.getbuffer()) + length_tensor = torch.LongTensor([len(data)]).to(device) + data_send_tensor = torch.ByteTensor(data).to(device) + # pyrefly: ignore [bad-argument-type] + dist.broadcast(length_tensor, src=src_rank, group=group, async_op=False) + # pyrefly: ignore [bad-argument-type] + dist.broadcast(data_send_tensor, src=src_rank, group=group, async_op=False) + else: + # Receive the object + length_tensor = torch.LongTensor([0]).to(device) + # pyrefly: ignore [bad-argument-type] + dist.broadcast(length_tensor, src=src_rank, group=group, async_op=False) + data_recv_tensor = torch.empty( + [int(length_tensor.item())], dtype=torch.uint8, device=device + ) + # pyrefly: ignore [bad-argument-type] + dist.broadcast(data_recv_tensor, src=src_rank, group=group, async_op=False) + buffer = io.BytesIO(data_recv_tensor.cpu().numpy()) + obj = torch.load(buffer, map_location=device, weights_only=False) + return obj + + +class _ZeROJoinHook(JoinHook): + def __init__(self, zero): + if not isinstance(zero, ZeroRedundancyOptimizer): + raise AssertionError( + "ZeRO join hook requires passing in a " + "ZeroRedundancyOptimizer instance as the state" + ) + self.zero = zero + super().__init__() + + def main_hook(self): + """ + Perform an optimizer step. + + This step updates the joined process's shard of + the parameters and broadcasts those parameters. + """ + self.zero.step() + + +class _DDPBucketAssignment: + r""" + Represent a :class:`DistributedDataParallel` bucket assignment. + + This means that a (possibly non-strict) subset of the parameters corresponding to + a DDP bucket assigned to a rank to update. + + Attributes: + bucket_index (int): index of the bucket determined by the DDP gradient + bucket all-reduce order. + parameters (List[torch.Tensor]): model parameters in the bucket + assigned to this rank. + offset (int): offset into the :class:`GradBucket` 's :meth:`parameters` + giving the index of the first element in the passed-in + ``parameters``; this equivalently indexes into the + :class:`GradBucket` 's :meth:`gradients`. + device (torch.device): device on which the parameters are stored. + tensor (torch.Tensor): flattened tensor giving the data of the + parameter subset assigned to the rank. + """ + + def __init__( + self, + bucket_index: int, + parameters: list[torch.Tensor], + offset: int, + ): + self.bucket_index = bucket_index + self.parameters = parameters + self.offset = offset + if len(self.parameters) == 0: + raise ValueError("Empty bucket assignment") + # DDP guarantees all parameters in the bucket have the same device + self.device: torch.device = self.parameters[0].device + self.tensor: torch.Tensor | None = None + + +class _OverlapStatus(enum.IntEnum): + r""" + Define possible statuses that :class:`ZeroRedundancyOptimizer` can be in when overlapping with :class:`DistributedDataParallel`. + + Attributes: + ``UNINITIALIZED``: The ZeRO instance is effectively uninitialized and + is waiting for DDP to finalize its bucketing. + ``DDP_HAS_REBUILT_BUCKETS``: DDP has rebuilt its buckets, meaning that + its bucketing is finalized. The ZeRO instance can now collect the + necessary information about the DDP bucketing. + ``INITIALIZED``: The ZeRO instance is fully initialized and can now + optimize parameters. + """ + + UNINITIALIZED = 0 + DDP_HAS_REBUILT_BUCKETS = 1 + INITIALIZED = 2 + + +class _OverlapInfo: + r""" + Information needed by :class:`ZeroRedundancyOptimizer` to overlap with :class:`DistributedDataParallel`. + + Arguments: + world_size (int): world size of the process group being used. + + Attributes: + shard_buckets (bool): if ``True``, then the assignment of each + :class:`DistributedDataParallel` bucket is partitioned across + possibly multiple :class:`ZeroRedundancyOptimizer` instances (i.e. + across possibly multiple ranks) to approximate uniformity following + a threshold given by the total parameter size divided by the world + size; if ``False``, then each bucket is wholly assigned to a single + :class:`ZeroRedundancyOptimizer` instance (i.e. to a single rank); + this should be set to the value passed into the hook constructor. + status (_OverlapStatus): current status; see :class:`_OverlapStatus` + for more information. + params_per_bucket (List[List[torch.Tensor]]): ``params_per_bucket[i]`` + gives the model parameters in the ``i``th bucket. + params_per_rank (List[List[torch.Tensor]]): ``params_per_rank[i]`` + gives the model parameters assigned to the ``i``th rank, where the + parameters are grouped by increasing bucket indices. + offsets (Dict[int, int]): maps from bucket index to the offset in + ``self.params_per_rank[rank]`` giving the index of the first + parameter in that bucket, where ``rank`` is this process's own + rank; the keys of this :class:`dict` are the bucket indices + assigned to this rank. + num_bucket_assignments (int): total number of bucket assignments across + all ranks; this is equal to the number of + :class:`DistributedDataParallel` gradient buckets if + ``shard_buckets=False`` and possibly greater otherwise. + total_size (int, optional): total size of all buckets (i.e. sum of + ``param.numel()`` for all ``param`` across all buckets) if + ``shard_buckets=True``; otherwise, ``None``. + broadcast_handles (List[Work]): :class:`list` of async work handles for + the parameter broadcasts. + bucket_index_to_future (Dict[int, torch.futures.Future]): + :class:`dict` mapping bucket index to the corresponding all-reduce + future. + bucket_index_to_bucket (Dict[int, dist.GradBucket]): :class:`dict` + mapping bucket index to the corresponding bucket. + bucket_indices_seen (List[int]): :class:`list` of the bucket indices + seen on this iteration. + """ + + def __init__(self, world_size) -> None: + self.status: _OverlapStatus = _OverlapStatus.UNINITIALIZED + self.shard_buckets: bool = False + + # Modified per bucket reconstruction + self.params_per_bucket: list[list[torch.Tensor]] = [] + self.params_per_rank: list[list[torch.Tensor]] = [[] for _ in range(world_size)] + self.offsets: dict[int, int] = {} + # Group Ranks + self.assigned_ranks_per_bucket: list[set[int]] = [] + self.num_bucket_assignments: int = 0 + self.total_size: int | None = None + + # Modified per iteration + self.broadcast_handles: list[Any] = [] + self.bucket_indices_seen: list[int] = [] + # Used by `hook_with_zero_step()` + self.bucket_index_to_future: dict[int, torch.futures.Future] = {} + self.bucket_index_to_bucket: dict[int, dist.GradBucket] = {} + + def wait_for_broadcasts(self) -> None: + r""" + Wait for all parameter broadcasts. + + This function should be called once all broadcasts have been scheduled, + meaning ``self.broadcast_handles`` is filled. This clears ``self.broadcast_handles`` + in preparation for the next iteration. + """ + if len(self.broadcast_handles) != self.num_bucket_assignments: + raise AssertionError( + f"Missing at least one broadcast handle on rank {dist.get_rank()}" + ) + _ = [x.wait() for x in self.broadcast_handles] + self.broadcast_handles.clear() + + def clear_per_iter_info(self) -> None: + r""" + Clear the data structures that are modified per-iteration. + + This function should be called at the end of an iteration. + """ + self.bucket_indices_seen.clear() + self.bucket_index_to_future.clear() + self.bucket_index_to_bucket.clear() + + +class ZeroRedundancyOptimizer(Optimizer, Joinable): + r""" + Wrap an arbitrary :class:`optim.Optimizer ` and shards its states across ranks in the group. + + The sharing is done as described by `ZeRO `_. + + The local optimizer instance in each rank is only + responsible for updating approximately ``1 / world_size`` parameters and + hence only needs to keep ``1 / world_size`` optimizer states. After + parameters are updated locally, each rank will broadcast its parameters to + all other peers to keep all model replicas in the same state. + ``ZeroRedundancyOptimizer`` can be used in conjunction with + :class:`torch.nn.parallel.DistributedDataParallel` to reduce per-rank peak + memory consumption. + + ``ZeroRedundancyOptimizer`` uses a sorted-greedy algorithm to pack a number + of parameters at each rank. Each parameter belongs to a single rank and is + not divided among ranks. The partition is arbitrary and might not match the + parameter registration or usage order. + + Arguments: + params (``Iterable``): an ``Iterable`` of :class:`torch.Tensor` s + or :class:`dict` s giving all parameters, which will be sharded + across ranks. + + Keyword Args: + optimizer_class (:class:`torch.nn.Optimizer`): the class of the local + optimizer. + process_group (``ProcessGroup``, optional): ``torch.distributed`` + ``ProcessGroup`` (default: ``dist.group.WORLD`` initialized by + :meth:`torch.distributed.init_process_group`). + parameters_as_bucket_view (bool, optional): if ``True``, parameters are + packed into buckets to speed up communication, and ``param.data`` + fields point to bucket views at different offsets; if ``False``, + each individual parameter is communicated separately, and each + ``params.data`` stays intact (default: ``False``). + overlap_with_ddp (bool, optional): if ``True``, :meth:`step` is + overlapped with :class:`DistributedDataParallel` 's gradient + synchronization; this requires (1) either a functional optimizer + for the ``optimizer_class`` argument or one with a functional + equivalent and (2) registering a DDP communication hook + constructed from one of the functions in ``ddp_zero_hook.py``; + parameters are packed into buckets matching those in + :class:`DistributedDataParallel`, meaning that the + ``parameters_as_bucket_view`` argument is ignored. + If ``False``, :meth:`step` runs disjointly after the backward pass + (per normal). + (default: ``False``) + **defaults: any trailing arguments, which are forwarded to the local + optimizer. + + Example:: + + >>> # xdoctest: +SKIP + >>> import torch.nn as nn + >>> from torch.distributed.optim import ZeroRedundancyOptimizer + >>> from torch.nn.parallel import DistributedDataParallel as DDP + >>> model = nn.Sequential(*[nn.Linear(2000, 2000).to(rank) for _ in range(20)]) + >>> ddp = DDP(model, device_ids=[rank]) + >>> opt = ZeroRedundancyOptimizer( + >>> ddp.parameters(), + >>> optimizer_class=torch.optim.Adam, + >>> lr=0.01 + >>> ) + >>> ddp(inputs).sum().backward() + >>> opt.step() + + .. warning:: + Currently, ``ZeroRedundancyOptimizer`` requires that all of the + passed-in parameters are the same dense type. + + .. warning:: + If you pass ``overlap_with_ddp=True``, be wary of the following: Given + the way that overlapping :class:`DistributedDataParallel` with + :class:`ZeroRedundancyOptimizer` is currently implemented, the first + two or three training iterations do not perform parameter updates in + the optimizer step, depending on if ``static_graph=False`` or + ``static_graph=True``, respectively. This is because it needs + information about the gradient bucketing strategy used by + :class:`DistributedDataParallel`, which is not finalized until the + second forward pass if ``static_graph=False`` or until the third + forward pass if ``static_graph=True``. To adjust for this, one option + is to prepend dummy inputs. + + .. warning:: ZeroRedundancyOptimizer is experimental and subject to change. + """ + + def __init__( + self, + params, + optimizer_class: type[Optimizer], + process_group: Any | None = None, + parameters_as_bucket_view: bool = False, + overlap_with_ddp: bool = False, + **defaults: Any, + ): + r"""Init.""" + # Perform type and assumption checks on the input parameters + params = self._verify_and_init_params(params) + self._verify_same_dense_param_type() + + # NOTE: The parent constructor uses `add_param_group()` which is + # partially overloaded in ZeroRedundancyOptimizer, so we use the + # `initialized` flag to dissociate the behaviour of `add_param_group()` + # between the parent and child. + self.initialized = False + + Optimizer.__init__(self, params, defaults) + Joinable.__init__(self) + # Now, all parameters are held in both `self._all_params` and + # `self.param_groups` + + # Internal data structures (`_cache` indicates lazily evaluated) + self._param_to_rank_cache: dict[torch.Tensor, int] = {} + self._param_to_index_cache: dict[torch.Tensor, int] = {} + self._partition_parameters_cache: list[list[dict]] = [] + self._index_to_param_cache: list[torch.Tensor] = [] + self._device_to_params_per_rank_cache: dict[ + torch.device, list[list[torch.Tensor]] + ] = {} + self._bucket_assignments_per_rank_cache: list[ + dict[int, _DDPBucketAssignment] + ] = [] + self._is_trainable_mask = self._get_is_trainable_mask() + + # Default device for collective communication and buckets + self._default_device = self._all_params[0].device + + self.process_group = ( + process_group if process_group is not None else dist.group.WORLD + ) + self.world_size: int = dist.get_world_size(self.process_group) + self.rank: int = dist.get_rank(self.process_group) + self.global_rank: int = dist.distributed_c10d.get_global_rank( + # pyrefly: ignore [bad-argument-type] + self.process_group, + self.rank, + ) + + self._overlap_with_ddp: bool = overlap_with_ddp + self._optim_defaults = defaults + self._optim_constructor = self._get_optimizer_constructor(optimizer_class) + + # If `overlap_with_ddp=True`, local optimizer initialization is delayed + # to run time after the necessary information has been collected + if not overlap_with_ddp: + self._init_local_optimizer() + else: + self._overlap_info: _OverlapInfo = _OverlapInfo(self.world_size) + if parameters_as_bucket_view: + logger.warning( + "`parameters_as_bucket_view=True` will be ignored since " + "`overlap_with_ddp=True`; instead, a different bucketing " + "strategy will be used" + ) + + # `self._buckets` is used if `parameters_as_bucket_view=True`, in + # which case parameter data is flattened into contiguous bucket tensors + self.parameters_as_bucket_view = parameters_as_bucket_view + self._buckets: list[list[torch.Tensor]] = [] + self._build_param_buckets() + + # Optional consolidated optimizer state, only populated if this rank + # is the target in `consolidate_state_dict()` + self._all_state_dicts: list[dict[str, Any]] = [] + + self.initialized = True + + def _clear_cache(self) -> None: + r"""Clear the cached data structures giving partition information.""" + self._partition_parameters_cache.clear() + self._param_to_rank_cache.clear() + self._index_to_param_cache.clear() + self._param_to_index_cache.clear() + self._device_to_params_per_rank_cache.clear() + self._bucket_assignments_per_rank_cache.clear() + + def add_param_group(self, param_group: dict[str, Any]) -> None: + r""" + Add a parameter group to the :class:`Optimizer` 's ``param_groups``. + + This can be useful when fine tuning a pre-trained network, as frozen + layers can be made trainable and added to the :class:`Optimizer` as + training progresses. + + Arguments: + param_group (dict): specifies the parameters to be optimized and + group-specific optimization options. + + .. warning:: This method handles updating the shards on all partitions + but needs to be called on all ranks. Calling this on a subset of + the ranks will cause the training to hang because communication + primitives are called depending on the managed parameters and + expect all the ranks to participate on the same set of parameters. + """ + if self.initialized and self._overlap_with_ddp: + raise RuntimeError( + "ZeroRedundancyOptimizer with `overlap_with_ddp=True` only " + "supports a single parameter group" + ) + + super().add_param_group(param_group) + # NOTE: The rest of the method assumes that the call to the parent's + # `add_param_group()` appends the new parameter group and preserves + # the previous parameter-group ordering + + if self.initialized: + # Force a re-partitioning of the parameters + self._clear_cache() + param_groups = self._partition_parameters()[self.rank] + # NOTE: All parameters in the old parameter groups should be + # assigned to the same ranks so that the local optimizers do not + # need to be reinitialized + + # Add the parameters assigned to this rank from the new parameter + # group to the local optimizer, if any + if len(param_groups) == len(self.optim.param_groups) + 1: + self.optim.add_param_group(param_groups[-1]) + + # Update the bucketing strategy accordingly + if self.parameters_as_bucket_view: + self._build_param_buckets() + + def consolidate_state_dict(self, to: int = 0) -> None: + r""" + Consolidate a list of ``state_dict`` s (one per rank) on the target rank. + + Arguments: + to (int): the rank that receives the optimizer states (default: 0). + + Raises: + RuntimeError: if ``overlap_with_ddp=True`` and this method is + called before this :class:`ZeroRedundancyOptimizer` instance + has been fully initialized, which happens once + :class:`DistributedDataParallel` gradient buckets have been + rebuilt. + + .. warning:: This needs to be called on all ranks. + """ + self._check_overlap_initialized() + + # Sync the exposed `param_groups` attributes to the local optimizer in + # case they have been updated + self._sync_param_groups(self.param_groups, self.optim.param_groups) + + # Pull the sharded state from all ranks and store them in rank order + empty_messenger = torch.tensor( + [0], dtype=torch.uint8, device=self._default_device + ) + + # NOTE: We wastefully use `broadcast()` (e.g. instead of `gather()`) + # due to compatibility issues with NCCL backend; a possible follow-up + # is to move all sharded state management to RPC RRef + self._all_state_dicts = [] + for rank in range(self.world_size): + global_rank = dist.distributed_c10d.get_global_rank( + # pyrefly: ignore [bad-argument-type] + self.process_group, + rank, + ) + if self.rank == to: + # Consolidate all local `state_dict`s on this rank, storing on + # CPU to save GPU memory + if rank == self.rank: + # Directly append own optimizer state + self._all_state_dicts.append( + _recursive_copy_to_device( + self.optim.state_dict(), + non_blocking=True, + device=torch.device("cpu"), + ) + ) + else: + # Receive the optimizer state from the source rank + local_state_dict = _broadcast_object( + empty_messenger, + src_rank=global_rank, + group=self.process_group, + device=self._default_device, + ) + self._all_state_dicts.append( + _recursive_copy_to_device( + local_state_dict, + non_blocking=True, + device=torch.device("cpu"), + ) + ) + else: + if rank == self.rank: + # Send the optimizer state to the target rank + _ = _broadcast_object( + self.optim.state_dict(), + src_rank=self.global_rank, + group=self.process_group, + device=self._default_device, + ) + elif rank != to: + # Discard the received object; `broadcast()` is used for + # compatibility reasons + _ = _broadcast_object( + empty_messenger, + src_rank=global_rank, + group=self.process_group, + device=self._default_device, + ) + + def _verify_params_per_rank( + self, + params_per_rank: list[list[torch.Tensor]], + ) -> None: + r""" + Verify ``params_per_rank`` for :meth:`_partition_parameters`. + + The verification is done by checking that ``params_per_rank`` has length equal + to the world size and that it does not contain any parameters not passed into the + :class:`ZeroRedundancyOptimizer` constructor. + + The parameters in ``params_per_rank`` being a strict subset of those + passed into the constructor is valid since some parameters may be + frozen. + + Raises: + ValueError: if ``params_per_rank`` does not have length equal to + the world size or if it contains a parameter that was not + passed into the :class:`ZeroRedundancyOptimizer` constructor. + """ + if len(params_per_rank) != self.world_size: + raise ValueError( + "`params_per_rank` must have length equal to the world size" + ) + all_params_set = set(self._all_params) + for params in params_per_rank: + for param in params: + if param not in all_params_set: + raise ValueError( + "Passing a new parameter in `params_per_rank` that " + "was not passed into the ZeroRedundancyOptimizer " + "constructor" + ) + + def _partition_param_group( + self, param_group: dict[str, Any], params_per_rank: list[list[torch.Tensor]] + ) -> None: + r""" + Partition the parameter group ``param_group`` according to ``params_per_rank``. + + The partition will modify the ``self._partition_parameters_cache``. This method should + only be used as a subroutine for :meth:`_partition_parameters`. + + Arguments: + param_group (dict[str, Any]): a parameter group as normally defined + in an optimizer state. + params_per_rank (list[list[torch.Tensor]]): a :class:`list` of + length world size containing :class:`list` s of parameters to + assign to each rank. + """ + for rank, params in enumerate(params_per_rank): + rank_param_group = copy.copy(param_group) + rank_param_group["params"] = params + self._partition_parameters_cache[rank].append(rank_param_group) + + def _partition_parameters( + self, + params_per_rank: list[list[torch.Tensor]] | None = None, + ) -> list[list[dict]]: + r""" + Partitions parameters across distributed data parallel ranks. + + Arguments: + params_per_rank (list[list[torch.Tensor]], optional): a + :class:`list` of length world size containing :class:`list` s + of parameters to assign to each rank; this provides a way to + specify a partition manually. + If ``None``, the parameters are partitioned according to an + internal algorithm. + (default: ``None``) + + Returns: + A :class:`list` where each element of the list contains the + ``param_groups`` for a rank (which itself is a :class:`list` of + :class:`dict`); element 0 corresponds to rank 0, etc.; each rank + stores the ``param_groups`` for all ranks for the collective + communication in :meth:`step`. + + Raises: + ValueError: see :meth:`_validate_params_per_rank`. + RuntimeError: if ``params_per_rank`` is not ``None`` and this + :class:`ZeroRedundancyOptimizer` instance is using more than + one parameter group. + """ + if params_per_rank is None: + # Partition the parameters optimizing for uniformity + if len(self._partition_parameters_cache) == 0: + self._partition_parameters_cache = [[] for _ in range(self.world_size)] + sizes = [0] * self.world_size + for param_group in self.param_groups: + param_group_params_per_rank: list[list] = [ + [] for _ in range(self.world_size) + ] + # Sort the parameters by size (largest first) + params_sorted = sorted( + param_group["params"], key=lambda t: t.numel(), reverse=True + ) + for param in params_sorted: + # Greedily add the parameter to rank with smallest size so far + rank = self._get_min_index(sizes) + param_group_params_per_rank[rank].append(param) + sizes[rank] += param.numel() + # Apply the constructed partition of the parameter group + self._partition_param_group( + param_group, param_group_params_per_rank + ) + + return self._partition_parameters_cache + + # Partition the parameters according to `params_per_rank` + if len(self._partition_parameters_cache) != 0: + raise AssertionError( + "Specifying `params_per_rank` should only be done when the " + "parameters have not been partitioned yet" + ) + if len(self.param_groups) != 1: + raise RuntimeError( + "Specifying `params_per_rank` only supports a single parameter group" + ) + self._verify_params_per_rank(params_per_rank) + self._partition_parameters_cache = [[] for _ in range(self.world_size)] + + # Apply the passed-in partition of the parameter group + param_group = self.param_groups[0] + self._partition_param_group(param_group, params_per_rank) + + return self._partition_parameters_cache + + @property + def _param_to_rank(self) -> dict[torch.Tensor, int]: + r""":class:`dict` mapping parameters to their assigned data parallel rank in the partition.""" + if len(self._param_to_rank_cache) == 0: + for rank, param_groups in enumerate(self._partition_parameters()): + for param_group in param_groups: + for param in param_group["params"]: + self._param_to_rank_cache[param] = rank + return self._param_to_rank_cache + + @property + def _param_to_index(self) -> dict[torch.Tensor, int]: + r""" + :class:`dict` mapping parameters to their indices in the global optimizer state. + + NOTE: This assumes that the global optimizer state's indexing (in + ``state_dict``) follows a linear ordering over the parameter groups. + """ + if len(self._param_to_index_cache) == 0: + self._param_to_index_cache = { + p: i + for i, p in enumerate( + chain.from_iterable(g["params"] for g in self.param_groups) + ) + } + return self._param_to_index_cache + + @property + def _index_to_param(self) -> list[torch.Tensor]: + r"""List mapping parameter indices in the global optimizer scheme to the actual params.""" + if len(self._index_to_param_cache) == 0: + self._index_to_param_cache = list( + chain.from_iterable(g["params"] for g in self.param_groups) + ) + return self._index_to_param_cache + + def _broadcast_params_from_rank(self, rank: int): + r""" + Broadcast the shard of parameters from a given rank to all other ranks asynchronously. + + Arguments: + rank (int): the source rank. + + Returns: + A :class:`list` of async work handles for the ``broadcast()`` s + performed to synchronize the parameters. + """ + if self._overlap_with_ddp: + raise AssertionError( + "`_broadcast_params_from_rank()` should not be used if " + "`overlap_with_ddp=True`; instead, the broadcasting should " + "happen in the DDP communication hook" + ) + handles = [] + if self.parameters_as_bucket_view: + for dev_i_buckets in self._buckets: + bucket = dev_i_buckets[rank] + global_rank = dist.distributed_c10d.get_global_rank( + # pyrefly: ignore [bad-argument-type] + self.process_group, + rank, + ) + handles.append( + dist.broadcast( + tensor=bucket, + src=global_rank, + group=self.process_group, + async_op=True, + ) + ) + else: + param_groups = self._partition_parameters()[rank] + global_rank = dist.distributed_c10d.get_global_rank( + # pyrefly: ignore [bad-argument-type] + self.process_group, + rank, + ) + for param_group in param_groups: + handles.extend( + dist.broadcast( + tensor=param.data, + src=global_rank, + group=self.process_group, + async_op=True, + ) + for param in param_group["params"] + ) + return handles + + def _sync_params(self): + r""" + Sync all parameter shards across the ranks. + + This rank sends its shard of the parameters to all other ranks and + receives a shard from each other rank. This is done using + ``broadcast()``. Parameters are sent bucket-by-bucket if + ``parameters_as_bucket_view=True``and sent parameter-by-parameter + otherwise. + """ + handles = [] + for rank in range(self.world_size): + handles.extend(self._broadcast_params_from_rank(rank)) + _ = [x.wait() for x in handles] + + @property + def _device_to_params_per_rank( + self, + ) -> dict[torch.device, list[list[torch.Tensor]]]: + r""" + Return device parameters assigned per rank. + + :class:`dict` mapping each device to a :class:`list` of the per-rank parameter + lists filtered to only include the parameters stored on that device. + Each per-rank parameter list gives the parameters assigned to that rank + to update. + + This is used for constructing the parameter buckets if + ``parameters_as_bucket_view=True``. + + Let ``dev_i`` denote the ``i``th device for this rank. Then: + ``dev_0`` maps to a list containing: + rank 0's assigned parameters stored on ``dev_0``, + rank 1's assigned parameters stored on ``dev_0``, + ... + ``dev_1`` maps to a list containing: + rank 0's assigned parameters stored on ``dev_1``, + rank 1's assigned parameters stored on ``dev_1``, + ... + ... + """ + if not self.parameters_as_bucket_view: + raise AssertionError( + "`_device_to_params_per_rank` should only be used if " + "`parameters_as_bucket_view=True`" + ) + if len(self._device_to_params_per_rank_cache) == 0: + for rank, param_groups in enumerate(self._partition_parameters()): + for param_group in param_groups: + for param in param_group["params"]: + device = param.device + if device not in self._device_to_params_per_rank_cache: + self._device_to_params_per_rank_cache[device] = [ + [] for _ in range(self.world_size) + ] + self._device_to_params_per_rank_cache[device][rank].append( + param + ) + return self._device_to_params_per_rank_cache + + def _get_min_index( + self, + values: list[int], + disallowed_indices: set[int] | None = None, + ) -> int: + r""" + Return ``values.index(min(values))``, except only uses one pass. + + It also excludes any indices in ``disallowed_indices`` if provided. + + Arguments: + values: (List[int]): :class:`list` of values. + disallowed_indices (Optional[set[int]]): indices that are + disallowed from being the returned min index. + """ + min_index = -1 + min_value = float("inf") + for i, value in enumerate(values): + if disallowed_indices and i in disallowed_indices: + continue + if value < min_value: + min_value = value + min_index = i + if min_index < 0: + raise AssertionError("All indices are disallowed") + return min_index + + def _assign_bucket_subset_to_rank( + self, + bucket_index: int, + bucket_params: list[torch.Tensor], + bucket_offset: int, + assigned_rank: int, + assigned_ranks_per_bucket: list[set[int]], + ) -> None: + r""" + Assign ``bucket_params`` to the rank with the least size assigned so far and collects relevant information. + + The model parameters given by ``bucket_params`` represents a (possibly non-strict) + subset of the parameters corresponding to a :class:`DistributedDataParallel` bucket. + + Arguments: + bucket_index (int): index of the :class:`DistributedDataParallel` + gradient bucket. + bucket_params (List[torch.Tensor]): subset of the parameters + corresponding to the bucket to assign. + bucket_offset (int): offset giving the index of the first element + in ``bucket_params`` in the bucket's full parameter list. + assigned_rank (int): group rank to assign to. + assigned_ranks_per_bucket (list[set[int]]): :class:`set` of group ranks + assigned to each bucket. + """ + overlap_info = self._overlap_info + if len(bucket_params) == 0: + raise ValueError("Empty bucket assignment") + params_per_rank = overlap_info.params_per_rank + offsets = overlap_info.offsets + + self._bucket_assignments_per_rank_cache[assigned_rank][bucket_index] = ( + _DDPBucketAssignment(bucket_index, bucket_params, bucket_offset) + ) + if self.global_rank == assigned_rank: + offsets[bucket_index] = len(params_per_rank[assigned_rank]) + params_per_rank[assigned_rank].extend(bucket_params) + assigned_ranks_per_bucket[bucket_index].add(assigned_rank) + self._overlap_info.num_bucket_assignments += 1 + + @property + def _bucket_assignments_per_rank(self) -> list[dict[int, _DDPBucketAssignment]]: + r""" + Return DDP bucket parameters assigned per rank. + + :class:`list` of length world size consisting of :class:`dict` s + mapping bucket indices to :class:`_DDPBucketAssignment` s for each + rank. + """ + if not self._overlap_with_ddp: + raise AssertionError( + "`_bucket_assignments_per_rank` only be used if `overlap_with_ddp=True`" + ) + if len(self._bucket_assignments_per_rank_cache) > 0: + return self._bucket_assignments_per_rank_cache + + overlap_info = self._overlap_info + if overlap_info.status != _OverlapStatus.INITIALIZED: + raise AssertionError + + self._bucket_assignments_per_rank_cache = [{} for _ in range(self.world_size)] + params_per_bucket = overlap_info.params_per_bucket + + if overlap_info.shard_buckets: + # Define the assignment threshold to approximate uniformity + if overlap_info.total_size is None: + raise AssertionError("`total_size` was not computed") + threshold = overlap_info.total_size / self.world_size # type: ignore[operator] + size_per_rank = [0 for _ in range(self.world_size)] + + num_buckets = len(params_per_bucket) + overlap_info.assigned_ranks_per_bucket = [set() for _ in range(num_buckets)] + assigned_ranks_per_bucket = overlap_info.assigned_ranks_per_bucket + if not overlap_info.shard_buckets: + # Assign each DDP bucket entirely to a single rank + for bucket_index, bucket_params in enumerate(params_per_bucket): + if len(bucket_params) <= 0: + raise AssertionError("Empty bucket") + assigned_rank = self._get_assigned_rank(bucket_index) + self._assign_bucket_subset_to_rank( + bucket_index, + bucket_params, + 0, + assigned_rank, + assigned_ranks_per_bucket, + ) + else: + # Assign each DDP bucket to possibly multiple ranks + # Specifically, sort the DDP buckets by increasing size, and for + # each bucket, iteratively assign the maximal unassigned subset + # with size less than `threshold` to the rank with the least total + # size so far -- each such assignment is represented by a + # `_DDPBucketAssignment` instance and only contains parameters from + # a single DDP bucket + params_per_bucket_enum = sorted( + enumerate(params_per_bucket), key=lambda x: sum(p.numel() for p in x[1]) + ) + for bucket_index, bucket_params in params_per_bucket_enum: + if len(bucket_params) <= 0: + raise AssertionError("Empty bucket") + bucket_offset = 0 + assignment_size = 0 + for param_index, param in enumerate(bucket_params): + param_numel = param.numel() + if ( + # pyrefly: ignore [unbound-name] + assignment_size + param_numel >= threshold + and param_index > bucket_offset + ): + assigned_rank = self._get_min_index( + # pyrefly: ignore [unbound-name] + size_per_rank, + assigned_ranks_per_bucket[bucket_index], + ) + # Include up to but not including the parameter that + # exceeded the threshold + self._assign_bucket_subset_to_rank( + bucket_index, + bucket_params[bucket_offset:param_index], + bucket_offset, + assigned_rank, + assigned_ranks_per_bucket, + ) + # pyrefly: ignore [unbound-name] + size_per_rank[assigned_rank] += assignment_size + bucket_offset = param_index + assignment_size = 0 + assignment_size += param_numel + # Assign the remainder of the bucket so that no assignment + # spans across two buckets + assigned_rank = self._get_min_index( + # pyrefly: ignore [unbound-name] + size_per_rank, + assigned_ranks_per_bucket[bucket_index], + ) + self._assign_bucket_subset_to_rank( + bucket_index, + bucket_params[bucket_offset:], + bucket_offset, + assigned_rank, + assigned_ranks_per_bucket, + ) + # pyrefly: ignore [unbound-name] + size_per_rank[assigned_rank] += assignment_size + + return self._bucket_assignments_per_rank_cache + + def _local_step( + self, + gradients: list[torch.Tensor | None] | None = None, + closure: Callable[[], float] | None = None, + **kwargs: Any, + ) -> float | None: + r""" + Perform a single optimizer step without syncing parameters across ranks. + + Arguments: + gradients (list[Optional[torch.Tensor]], optional): a :class:`list` + of length equal to the number of parameters assigned to this + rank containing gradient tensors or ``None`` as its elements; + a ``None`` in the :class:`list` indicates that the + corresponding parameter should not be updated. + If the argument itself is ``None``, then all parameters are + updated, and the gradients are assumed to be already populated. + (default: ``None``) + closure (Callable): a closure that re-evaluates the model and + returns the loss; optional for most optimizers and should be + ``None`` if ``gradients`` is not ``None``; (default: ``None``) + Returns: + Optional loss depending on the underlying local optimizer. + + .. warning:: + The argument ``gradients`` should only be specified (i.e. not + ``None``) if ``overlap_with_ddp=True``, in which case + :class:`ZeroRedundancyOptimizer` wraps a functional optimizer. + """ + Join.notify_join_context(self) + # Check if the model trainability has changed + is_trainable_mask = self._get_is_trainable_mask() + if is_trainable_mask != self._is_trainable_mask: + if self._overlap_with_ddp: + raise RuntimeError( + "ZeroRedundancyOptimizer with `overlap_with_ddp=True` " + "does not support changing parameter trainability at run " + "time" + ) + logger.warning( + "ZeroRedundancyOptimizer detected that the trainable " + "parameters changed; rebuilding the parameter buckets if " + "enabled" + ) + self._build_param_buckets() + self._is_trainable_mask = is_trainable_mask + + # Sync the exposed `param_groups` attributes to the local optimizer in + # case they have been updated + self._sync_param_groups(self.param_groups, self.optim.param_groups) + + # Run the optimizer step on this shard only + if gradients is None: + loss = ( + self.optim.step(**kwargs) + if closure is None + else self.optim.step(closure=closure, **kwargs) + ) + else: + if not self._overlap_with_ddp: + raise AssertionError( + "Specifying `gradients` should not " + "be used when `overlap_with_ddp=False`" + ) + if closure is not None: + raise AssertionError( + "`closure` is not supported when using a local functional optimizer" + ) + loss = self.optim.step(gradients=gradients) + + # Sync any updated attributes in the local optimizer to the exposed + # `param_groups` + self._sync_param_groups(self.optim.param_groups, self.param_groups) + + return loss + + # pyrefly: ignore [bad-override] + def step( + self, + closure: Callable[[], float] | None = None, + **kwargs: Any, + ) -> float | None: + r""" + Perform a single optimizer step and syncs parameters across all ranks. + + Arguments: + closure (Callable): a closure that re-evaluates the model and + returns the loss; optional for most optimizers. + Returns: + Optional loss depending on the underlying local optimizer. + + .. note:: Any extra parameters are passed to the base optimizer as-is. + """ + if self._overlap_with_ddp: + logger.warning( + "`step()` should not be included in the training loop when " + "`overlap_with_ddp=True`" + ) + return None + + # Perform the local optimizer step + loss = self._local_step(closure=closure, **kwargs) + + # Sync all of the updated parameter shards across the ranks + self._sync_params() + + return loss + + def join_hook(self, **_kwargs: Any) -> JoinHook: + r""" + Return the ZeRO join hook. + + It enables training on uneven inputs by + shadowing the collective communications in the optimizer step. + + Gradients must be properly set before this hook is called. + + Arguments: + kwargs (dict): a :class:`dict` containing any keyword arguments + to modify the behavior of the join hook at run time; all + :class:`Joinable` instances sharing the same join context + manager are forwarded the same value for ``kwargs``. + + This hook does not support any keyword arguments; i.e. ``kwargs`` is + unused. + """ + return _ZeROJoinHook(self) + + @property + def join_device(self) -> torch.device: + r"""Return default device.""" + return self._default_device + + @property + def join_process_group(self) -> Any: + r"""Return process group.""" + return self.process_group + + def load_state_dict(self, state_dict: dict[str, Any]) -> None: + r""" + Load the state pertaining to the given rank from the input ``state_dict``, updating the local optimizer as needed. + + Arguments: + state_dict (dict): optimizer state; should be an object returned + from a call to :meth:`state_dict`. + + Raises: + RuntimeError: if ``overlap_with_ddp=True`` and this method is + called before this :class:`ZeroRedundancyOptimizer` instance + has been fully initialized, which happens once + :class:`DistributedDataParallel` gradient buckets have been + rebuilt. + """ + self._check_overlap_initialized() + + for index, value in state_dict["state"].items(): + param = self._index_to_param[index] + if self._param_to_rank[param] != self.rank: + # Clear any state irrelevant to this rank + state_dict["state"][index] = None + else: + # Load the parameter state to the local optimizer + self.optim.state[param] = _recursive_copy_to_device( + value, non_blocking=True, device=param.device + ) + # Force zero-dimensional tensors (like Adam "step") on CPU + for state_name, state_value in self.optim.state[param].items(): + if torch.is_tensor(state_value) and state_value.dim() == 0: + self.optim.state[param][state_name] = state_value.cpu() + + super().load_state_dict(state_dict) + + # Sync the input state with the exposed and local optimizer states + self._sync_param_groups(state_dict["param_groups"], self.param_groups) + self._sync_param_groups(self.param_groups, self.optim.param_groups) + + def state_dict(self) -> dict[str, Any]: + r""" + Return the last global optimizer state known to this rank. + + .. warning: + If the state has not been consolidated to this rank, this raises a + runtime error, and even if it has, the state may not be up-to-date, + depending on when :meth:`consolidate_state_dict` was last called. + + Raises: + RuntimeError: if ``overlap_with_ddp=True`` and this method is + called before this :class:`ZeroRedundancyOptimizer` instance + has been fully initialized, which happens once + :class:`DistributedDataParallel` gradient buckets have been + rebuilt; or if this method is called without a preceding call + to :meth:`consolidate_state_dict`. + """ + self._check_overlap_initialized() + + if len(self._all_state_dicts) == 0: + raise RuntimeError( + "Optimizer state has not been consolidated on this rank. " + f"Please call `consolidate_state_dict(to={self.rank})` on " + "all ranks beforehand if you meant to save the global state." + ) + + # Get the possibly-stale global optimizer state that uses global + # parameter indexing + state_dict = super().state_dict() + + # Update the global optimizer state with local state information, + # factoring in the translation from local to global indexing + for rank, local_state_dict in enumerate(self._all_state_dicts): + local_param_groups = local_state_dict["param_groups"] + global_param_groups = self._partition_parameters()[rank] + if len(local_param_groups) != len(global_param_groups): + raise AssertionError( + "Mismatch between number of local and global parameter groups" + ) + + for local_param_group, global_param_group in zip( + local_param_groups, global_param_groups + ): + # `local_param_group` stores local indices, while + # `global_param_group` stores the tensors directly + local_param_indices = local_param_group["params"] + global_params = global_param_group["params"] + + if len(local_param_indices) != len(global_params): + raise AssertionError( + "Mismatch between number of local and global " + "parameters in parameter group" + ) + for local_param_index, global_param in zip( + local_param_indices, global_params + ): + # Update the global parameter state, if any + if local_param_index in local_state_dict["state"]: + global_param_index = self._param_to_index[global_param] + state_dict["state"][global_param_index] = local_state_dict[ + "state" + ][local_param_index] + + # Sort the parameters in the state + state_dict["state"] = dict(sorted(state_dict["state"].items())) + return state_dict + + @staticmethod + def _sync_param_groups( + src_param_groups: list[dict[Any, Any]], + dst_param_groups: list[dict[Any, Any]], + ) -> None: + r""" + Sync the attributes from the source parameter groups to the destination parameter groups. + + Example attributes include learning rate or scheduler attributes. The + two parameter groups should have the same length (i.e. same number of + parameter groups). + + Arguments: + src_param_groups (list[dict]): parameter groups giving the + attribute settings to copy. + dst_param_groups (list[dict]): parameter groups giving the + attribute settings to set. + """ + if len(src_param_groups) != len(dst_param_groups): + raise AssertionError( + "Mismatch between number of source and destination parameter groups" + ) + for src_param_group, dst_param_group in zip(src_param_groups, dst_param_groups): + # Sync all attributes except the parameters + for attr in filter(lambda x: x != "params", src_param_group.keys()): + dst_param_group[attr] = src_param_group[attr] + + def _build_param_buckets(self) -> None: + r""" + Build parameter buckets if ``parameters_as_bucket_view=True``. + + For each device that stores this rank's parameters, there is a + bucket (represented as a tensor) containing all of the parameters on + that device that are assigned to a given rank in the parameter update + partition. + + This method is called in the constructor and any time parameter + trainability is changed. + + .. warning:: + The current implementation assumes that all of the parameters in a + bucket are of the same dense type when allocating the bucket's + tensor. + + .. warning:: + If the model parameters are stored across more than one device, + then the storage partitioning must be the same across all + processes in order for parameter synchronization to work. + """ + if not self.parameters_as_bucket_view or self._overlap_with_ddp: + return + + # `self._buckets[i][j]` are the parameters stored on device i and + # assigned to rank j + num_devices = len(self._device_to_params_per_rank) + self._buckets = [[] for _ in range(num_devices)] # type: ignore[assignment] + + for dev_i, (device, params_per_rank) in enumerate( + self._device_to_params_per_rank.items() + ): + for params in params_per_rank: + bucket_size = 0 + dtype = None + trainable_params = [] + for param in params: + if not _is_trainable(param): + # Clone in case the parameter was previously part of + # a bucket to avoid the data from being destroyed + param.data = param.data.detach().clone() + else: + bucket_size += param.numel() + trainable_params.append(param) + dtype = param.dtype # assumes all same dtype + + if bucket_size == 0: + # Create a dummy bucket if there are no parameters + bucket = torch.zeros(1, device=device) + else: + # Construct the bucket (assuming all dense and same dtype) + bucket = torch.empty(bucket_size, dtype=dtype, device=device) + offset = 0 + for param in trainable_params: + offset_next = offset + param.numel() + bucket[offset:offset_next].copy_(param.data.flatten()) + param.data = bucket[offset:offset_next].view_as(param.data) + offset = offset_next + self._buckets[dev_i].append(bucket) # type: ignore[arg-type] + + def _build_ddp_param_buckets(self) -> None: + r""" + Build the DDP bucket with parameters assigned to this rank. + + For each DDP bucket with parameters assigned to this rank, flattens the + data of those parameters into a single tensor and saves the tensor to + the ``tensor`` attribute in the corresponding + :class:`_DDPBucketAssignment` instance stored in + ``self._bucket_assignments_per_rank``. + + :class:`DistributedDataParallel` guarantees that the parameters + corresponding to a gradient bucket have the same device and the same + dtype. + """ + for bucket_assignments in self._bucket_assignments_per_rank: + for bucket_assignment in bucket_assignments.values(): + params = bucket_assignment.parameters + bucket_size = 0 + dtype = None + for param in params: + if not _is_trainable(param): + raise AssertionError( + "Model parameter " + "corresponding to a gradient in a DDP bucket should " + "require a gradient" + ) + bucket_size += param.numel() + dtype = param.dtype # assumes all same dtype + if bucket_size <= 0: + raise AssertionError("Empty bucket") + + # Construct the bucket tensor (assuming all dense and same dtype) + tensor = torch.empty( + bucket_size, dtype=dtype, device=bucket_assignment.device + ) + offset = 0 + for param in params: + offset_next = offset + param.numel() + tensor[offset:offset_next].copy_(param.data.flatten()) + param.data = tensor[offset:offset_next].view_as(param.data) + offset = offset_next + bucket_assignment.tensor = tensor + + def _verify_and_init_params( + self, + params: Any, + ) -> list[torch.Tensor] | list[dict]: + r""" + Verify the type of ``params`` and initializes ``self._all_params`` as a :class:`list` of all parameters. + + The initialization will first make sure that provided ``params`` is valid. + + Arguments: + params (Any): Candidate parameter list or parameter groups to verify. + + Raises: + TypeError: ``params`` has an invalid type. + ValueError: ``params`` is empty. + + Returns: + The persistent form of ``params`` to be passed into the parent + :class:`Optimizer` constructor -- i.e. returns ``params`` as a + :class:`list` to ensure that it can be iterated over again. + """ + if isinstance(params, torch.Tensor): + raise TypeError( + "`params` argument should be an iterable of " + f"Tensors, but got {torch.typename(params)}" + ) + try: + all_params = list(params) + except TypeError as e: + raise TypeError( + "`params` argument should be an iterable of Tensors" + f" or dicts, but got {torch.typename(params)}" + ) from e + if len(all_params) == 0: + raise ValueError("ZeroRedundancyOptimizer got an empty parameter list") + all_tensors = True + all_dicts = True + for param in all_params: + all_tensors &= isinstance(param, torch.Tensor) + all_dicts &= isinstance(param, dict) + if not all_tensors and not all_dicts: + raise TypeError( + "`params` argument should be an iterable of Tensors or dicts" + ) + # Ensure that `self._all_params` contains a list of all parameters + if all_tensors: + self._all_params = all_params + elif all_dicts: + self._all_params = [] + # `all_params` contains parameter groups (not parameters) + for param_group in all_params: + if "params" not in param_group: + raise ValueError( + "Each parameter group passed-in via `params` must " + "have a 'params' key mapping to the parameters in " + "the group" + ) + self._all_params.extend(param_group["params"]) + return all_params + + def _verify_same_dense_param_type(self) -> None: + r""" + Verify that all parameters are of the same dense type. + + The method assumes that ``self._all_params`` has been initialized + and is non-empty. + + Raises: + ValueError: ``params`` contains sparse parameters or parameters + of varying dense types. + + NOTE: This method can be removed once support for sparse parameters + and varying parameter types is added. + """ + typename = torch.typename(self._all_params[0]) + if self._all_params[0].is_sparse: + raise ValueError( + "ZeroRedundancyOptimizer only supports using " + "the same dense type for all parameters but got " + f"{typename}" + ) + for param in self._all_params[1:]: + other_typename = torch.typename(param) + if other_typename != typename: + raise ValueError( + "ZeroRedundancyOptimizer only supports " + "using the same dense type for all " + f"parameters but got both {typename} and " + f"{other_typename}" + ) + + def _get_is_trainable_mask(self) -> list[bool]: + r"""Return a boolean mask indicating if each parameter is trainable (``requires_grad``) or not.""" + return list(map(_is_trainable, self._all_params)) + + def _init_local_optimizer(self) -> None: + r""" + Initialize this rank's local optimizer, responsible for its subset of the parameters. + + The local optimizer is saved in ``self.optim``. + """ + if self._optim_constructor is None: + raise AssertionError("The local optimizer class has not been set") + + param_groups = self._partition_parameters()[self.rank] + # `overlap_with_ddp=True` requires a local functional optimizer + if self._overlap_with_ddp: + # Functional optimizers only support a single parameter group and + # require passing in the parameters as a list + if len(param_groups) != 1: + raise AssertionError( + "Initializing the local functional optimizer " + "with more than one parameter group" + ) + params = param_groups[0]["params"] + # Try to pass `_allow_empty_param_list=True` to avoid erroring + if ( + "_allow_empty_param_list" + in inspect.signature(self._optim_constructor).parameters + ): + self.optim: Any = self._optim_constructor( + params, **self._optim_defaults, _allow_empty_param_list=True + ) + else: + logger.warning( + "%s does not support the argument " + "`_allow_empty_param_list`; ZeroRedundancyOptimizer may " + "error due to an empty parameter list", + self._optim_constructor, + ) + self.optim: Any = self._optim_constructor( + params, **self._optim_defaults + ) # type: ignore[no-redef] + + # Log information about the DDP and ZeRO bucketing + if dist.get_debug_level() != dist.DebugLevel.OFF: + local_numel = sum(p.numel() for p in params) + num_assigned_buckets = len( + self._bucket_assignments_per_rank[self.global_rank] + ) + logger.info( + "rank %s with %s parameters across %s buckets", + self.global_rank, + local_numel, + num_assigned_buckets, + ) + if self.global_rank == 0: + logger.info( + "%s DDP buckets and %s bucket assignments", + len(self._overlap_info.params_per_bucket), + self._overlap_info.num_bucket_assignments, + ) + else: + # NOTE: Passing `param_groups` into the local optimizer constructor + # bypasses the empty parameter list check + self.optim: Optimizer = self._optim_constructor( + param_groups, **self._optim_defaults + ) # type: ignore[no-redef] + + # TODO: Manually add `self.param_groups` if using a functional + # optimizer; remove this if/when the functional optimizers support + # multiple parameter groups + if self._overlap_with_ddp and not hasattr(self.optim, "param_groups"): + if not hasattr(self.optim, "param_group"): + raise AssertionError( + "The functional optimizer should set at least one of " + "the attributes `param_group` or `param_groups`" + ) + self.optim.param_groups = [self.optim.param_group] # type: ignore[attr-defined] + + self._sync_param_groups(self.optim.param_groups, self.param_groups) + + def _init_zero_for_overlap(self) -> None: + r"""Perform a delayed initialization of the local optimizer and the supporting data structures.""" + if not self._overlap_with_ddp: + raise AssertionError( + "`_init_zero_for_overlap()` should only be called when " + "`overlap_with_ddp=True`" + ) + self._overlap_info.status = _OverlapStatus.INITIALIZED + self._clear_cache() + self._partition_parameters(self._overlap_info.params_per_rank) + self._build_ddp_param_buckets() + self._init_local_optimizer() + + def _get_assigned_rank(self, bucket_index: int) -> int: + r""" + Return the single rank assigned to a :class:`DistributedDataParallel` gradient bucket. + + Arguments: + bucket_index (int): index of the :class:`DistributedDataParallel` + bucket for which to get the assigned rank. + """ + if self._overlap_info.shard_buckets: + raise AssertionError( + "The bucket assignment requires global bucket information " + "and will be computed later; there should be no need to " + "use this method" + ) + return bucket_index % self.world_size + + def _check_overlap_initialized(self): + r""" + Check the delayed initialization depending on the value of ``overlap_with_ddp``. + + The delayed initialization has occurred (see + :meth:`_init_zero_for_overlap`) if ``overlap_with_ddp=True``, and + raises a ``RuntimeError`` if not. This should preface methods that + should not be run before that delayed initialization. + + Raises: + RuntimeError: if ``overlap_with_ddp=True`` and + :meth:`_init_zero_for_overlap` has not been called. + """ + if ( + self._overlap_with_ddp + and self._overlap_info.status != _OverlapStatus.INITIALIZED + ): + raise RuntimeError( + "This method should not be called until this " + "ZeroRedundancyOptimizer instance has been fully " + "initialized" + ) + + def _get_optimizer_constructor(self, optimizer_class: Any) -> Any: + r""" + Return the optimizer constructor using validation and transformation depending on ``overlap_with_ddp``. + + Returns: + - ``optimizer_class`` if ``overlap_with_ddp=False`` and + ``optimizer_class`` is not a functional optimizer. + - ``optimizer_class`` if ``overlap_with_ddp=True`` and + ``optimizer_class`` is already a functional optimizer. + - The functional equivalent of ``optimizer_class`` if + ``overlap_with_ddp=True`` and ``optimizer_class`` is not + already a functional optimizer (assuming the equivalent + exists). + + Raises: + ValueError: + + - if ``overlap_with_ddp=True`` but ``optimizer_class`` is + neither a functional optimizer nor translatable to a + functional optimizer. + - if ``overlap_with_ddp=False`` and ``optimizer_class`` is a + functional optimizer. + """ + functional_optims = functional_optim_map.values() + if not self._overlap_with_ddp: + if optimizer_class in functional_optims: + # Using a functional optimizer is only supported when + # `overlap_with_ddp=True` + raise ValueError( + f"Passing in a functional optimizer {optimizer_class} " + "when `overlap_with_ddp=False`" + ) + else: + return optimizer_class + else: + if optimizer_class in functional_optims: + # Already a functional optimizer + return optimizer_class + elif optimizer_class in functional_optim_map: + # Translate the passed-in optimizer class to its functional + # equivalent if `overlap_with_ddp=True` + optim_constructor = functional_optim_map[optimizer_class] + logger.info( + "Using the functional optimizer %s " + "instead of %s since " + "`overlap_with_ddp=True`", + optim_constructor, + optimizer_class, + ) + return optim_constructor + else: + raise ValueError( + "Using `ddp_with_overlap=True` requires using a " + "functional optimizer, but there is no supported functional " + f"optimizer equivalent for {optimizer_class}" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.pyi b/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.pyi new file mode 100644 index 0000000000000000000000000000000000000000..27e3106238f21aa105359a9701a2b656a0f87d4e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/optim/zero_redundancy_optimizer.pyi @@ -0,0 +1,85 @@ +# mypy: allow-untyped-defs +import enum +from collections.abc import Callable +from typing import Any, overload + +import torch +from torch.distributed.algorithms.join import Joinable, JoinHook +from torch.optim import Optimizer + +class _ZeROJoinHook(JoinHook): + zero: Any = ... + def __init__(self, zero: Any) -> None: ... + def main_hook(self) -> None: ... + +class _DDPBucketAssignment: + bucket_index: int + parameters: list[torch.Tensor] + offset: int + device: torch.device + tensor: torch.Tensor | None + +class _OverlapStatus(enum.IntEnum): + UNINITIALIZED = ... + DDP_HAS_REBUILT_BUCKETS = ... + INITIALIZED = ... + +class _OverlapInfo: + status: Any = ... + params_per_bucket: Any = ... + params_per_rank: Any = ... + offsets: Any = ... + broadcast_handles: Any = ... + bucket_index_to_future: Any = ... + bucket_index_to_bucket: Any = ... + bucket_indices_seen: Any = ... + assigned_ranks_per_bucket: list[set[int]] = ... + total_size: int = ... + shard_buckets: bool = ... + def __init__(self) -> None: ... + def wait_for_broadcasts(self) -> None: ... + def clear_per_iter_info(self) -> None: ... + +class ZeroRedundancyOptimizer(Optimizer, Joinable): + functional_optim_map: Any = ... + initialized: bool = ... + process_group: Any = ... + world_size: int = ... + rank: int = ... + global_rank: int = ... + parameters_as_bucket_view: bool = ... + optim: Any = ... + _device_to_device_index: dict[torch.device, int] = ... + _overlap_with_ddp: bool = ... + _overlap_info: _OverlapInfo = ... + _buckets: list[list[torch.Tensor]] = ... + _bucket_assignments_per_rank: list[dict[int, _DDPBucketAssignment]] = ... + def __init__( + self, + params: Any, + optimizer_class: type[Optimizer], + process_group: Any | None = ..., + parameters_as_bucket_view: bool = ..., + overlap_with_ddp: bool = ..., + **defaults: Any, + ) -> None: ... + def add_param_group(self, param_group: dict[str, Any]) -> None: ... + def consolidate_state_dict(self, to: int = ...) -> None: ... + @overload + def step(self, closure: None = None, **kwargs: Any) -> None: ... + @overload + def step(self, closure: Callable[[], float], **kwargs: Any) -> float: ... + def load_state_dict(self, state_dict: dict[str, Any]) -> None: ... + def state_dict(self) -> dict[str, Any]: ... + def _local_step( + self, + gradients: list[torch.Tensor | None] | None = None, + closure: Callable[[], float] | None = None, + **kwargs: Any, + ) -> float | None: ... + def _get_assigned_rank(self, bucket_index: int) -> int: ... + def _init_zero_for_overlap(self) -> None: ... + def join_hook(self, **_kwargs: Any) -> JoinHook: ... + @property + def join_device(self) -> torch.device: ... + def join_process_group(self) -> Any: ... diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_IR.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_IR.py new file mode 100644 index 0000000000000000000000000000000000000000..dc7deea76ed190b9f9db2cd4e656aade711cacfa --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_IR.py @@ -0,0 +1,1296 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import copy +import logging +import operator +from collections import defaultdict +from collections.abc import Callable +from enum import Enum +from inspect import Parameter, Signature, signature +from types import MethodType +from typing import Any + +import torch +import torch.fx as fx +from torch.distributed import ProcessGroup +from torch.export import ExportedProgram +from torch.export.unflatten import ( + _assign_attr, + _AttrKind, + _sink_params, + InterpreterModule, +) +from torch.fx.node import map_aggregate +from torch.fx.passes.split_module import split_module + +from ._backward import _null_coalesce_accumulate, stage_backward +from ._unflatten import _outline_submodules +from ._utils import PipeInfo +from .stage import _PipelineStage + + +logger = logging.getLogger(__name__) + +# TODO: +# 1. investigate gradient sync for shared parameters. how does DDP do it? +# 2. Add parameter movement to split_module + + +PP_SUBMOD_PREFIX = "submod_pp" + + +def get_submod_name(stage_idx: int): + """Returns the name of the submod for a given stage index. + For example, "submod_pp_0", "submod_pp_1", etc. + """ + return "_".join([PP_SUBMOD_PREFIX, str(stage_idx)]) + + +def _find_loss_from_output_and_spec(output_val, spec_val): + if spec_val is False: + return None + if spec_val is True: + if not isinstance(output_val, fx.Node): + raise RuntimeError( + f"Loss spec must specify a dynamic value but got {output_val}" + ) + return output_val + + if isinstance(spec_val, (tuple, list)): + if not isinstance(output_val, (tuple, list)): + raise RuntimeError( + f"Output value {output_val} must match type of loss specification " + f"{spec_val}" + ) + if len(output_val) != len(spec_val): + raise RuntimeError( + f"Output value {output_val} must match length of loss specification " + f"{spec_val}" + ) + for out, spec in zip(output_val, spec_val): + loss_val = _find_loss_from_output_and_spec(out, spec) + if loss_val is not None: + return loss_val + raise RuntimeError(f"Did not find loss value in specification {spec_val}") + + if isinstance(spec_val, dict): + if not isinstance(output_val, dict): + raise RuntimeError( + f"Output value {output_val} must match type of loss specification " + f"{spec_val}" + ) + if set(output_val.keys()) != set(spec_val.keys()): + raise RuntimeError( + f"Output value {output_val} must match keys of loss specification " + f"{spec_val}" + ) + for k in spec_val: + loss_val = _find_loss_from_output_and_spec(output_val[k], spec_val[k]) + if loss_val is not None: + return loss_val + raise RuntimeError(f"Did not find loss value in specification {spec_val}") + + raise RuntimeError(f"Unsupported type {type(spec_val)} in loss specification") + + +def _find_loss_output(mod: torch.nn.Module, g: fx.Graph, output_loss_value_spec): + output_nodes = [n for n in g.nodes if n.op == "output"] + if not len(output_nodes) == 1: + raise AssertionError(f"Expected 1 output node, got {len(output_nodes)}") + output_node = output_nodes[0] + output_val = output_node.args[0] + generated_spec: Any = None + + if isinstance(mod, TrivialLossWrapper): + # TrivialLossWrapper is pre-defined by PiPPy. + # It has loss as the only output so we can safely assume the first output arg is the loss. + if not len(output_node.args) == 1: + raise AssertionError(f"Expected 1 output arg, got {len(output_node.args)}") + loss_node = output_val + generated_spec = TrivialLossWrapper.loss_spec + elif output_loss_value_spec is None: + # Use default spec, i.e. search for "loss" in output values + if isinstance(output_val, dict) and "loss" in output_val: + loss_node = output_val["loss"] + generated_spec = {k: k == "loss" for k in output_val} + else: + loss_node = None + generated_spec = None + else: + loss_node = _find_loss_from_output_and_spec(output_val, output_loss_value_spec) + generated_spec = output_loss_value_spec + + return loss_node, output_node, generated_spec + + +def _insert_stage_symbolic_backward( + g: fx.Graph, + loss_node: fx.Node, + output_node: fx.Node, +): + # Collect metadata about tuple output values. TODO: move this to split_module or FX IR + tuples: dict[fx.Node, tuple] = {} + for node in reversed(g.nodes): + if node.op == "call_function": + # In the forward pass, only emit placeholder, module calls, and + # getitem calls. If we have a target other than getitem in this + # (forward-only) code, there is a bug. + if not node.target == operator.getitem: + raise AssertionError( + "Found non-getitem call in forward pass. Please report a bug to PiPPy" + ) + if not len(node.args) == 2: + raise AssertionError( + "Found malformed getitem call. Please report a bug to PiPPy" + ) + indexed_value, node_idx = tuple(node.args) + + # indexed_value is a collection that we are indexing into. It could + # exist in the tuples map if we've processed another `getitem` + # already. + existing_list_size = ( + len(tuples[indexed_value]) if indexed_value in tuples else -1 + ) + new_list_size = max(node_idx + 1, existing_list_size) + + reconstructed_list = [None for _ in range(new_list_size)] + + # Copy over existing elements if present + if indexed_value in tuples: + for i, val in enumerate(tuples[indexed_value]): + reconstructed_list[i] = val + + # Populate value represented by this node + reconstructed_list[node_idx] = node + + tuples[indexed_value] = tuple(reconstructed_list) + + # Keep track of nodes that dominate the loss node. + # We will only emit backward operations for nodes that can contribute + # to the specified loss value. + live_nodes = {loss_node: None} + val_to_grad: dict[fx.Node, fx.Node | None] = {loss_node: None} + + def assign_or_accumulate_grad(forward_node, grad_value): + if forward_node in val_to_grad and forward_node.op != "placeholder": + grad_value = g.call_function( + _null_coalesce_accumulate, + (val_to_grad[forward_node], grad_value), + ) + val_to_grad[forward_node] = grad_value + + with g.inserting_before(output_node): + for node in reversed(g.nodes): + if node not in live_nodes: + continue + + def add_to_live_nodes(n): + live_nodes.setdefault(n, None) + + fx.node.map_arg(node.args, add_to_live_nodes) + fx.node.map_arg(node.kwargs, add_to_live_nodes) + if node.op == "call_module": + output_grads: tuple[fx.Node | None, ...] | fx.Node | None + if node in tuples: + stage_output = tuples[node] + output_grads = tuple(val_to_grad.get(n) for n in tuples[node]) + outputs_with_grads_idxs = [ + i for i, n in enumerate(tuples[node]) if n in live_nodes + ] + else: + stage_output = (node,) + output_grads = val_to_grad[node] + outputs_with_grads_idxs = [0] + + output_grads = ( + (output_grads,) + if not isinstance(output_grads, tuple) + else output_grads + ) + + grad_call = g.call_function( + stage_backward, + kwargs={ + "stage_output": stage_output, + "output_grads": output_grads, + "input_values": list(node.all_input_nodes), + "outputs_with_grads_idxs": outputs_with_grads_idxs, + }, + ) + # Insert backward stage debug info + kwargs_copy = dict(grad_call.kwargs) + grad_call.kwargs = kwargs_copy + + grad_call_proxy = fx.Proxy(grad_call) + grads = grad_call_proxy.node + + input_nodes = list(node.all_input_nodes) + grads_proxy = fx.Proxy(grads) + for i, input_node in enumerate(input_nodes): + assign_or_accumulate_grad(input_node, grads_proxy[i].node) # type: ignore[index] + + return g + + +def _move_placeholders_to_front(graph: fx.Graph): + """Ensure all placeholder nodes form a contiguous prefix of the graph. + split_module can interleave get_attr nodes with placeholders.""" + last_ph = None + for node in graph.nodes: + if node.op == "placeholder": + last_ph = node + if last_ph is None: + return + nodes_to_move = [] + for node in graph.nodes: + if node is last_ph: + break + if node.op != "placeholder": + nodes_to_move.append(node) + insert_point = last_ph + for node in nodes_to_move: + insert_point.append(node) + insert_point = node + + +class PipeSequential(torch.nn.Sequential): + @staticmethod + def from_sequential(sequential_instance: torch.nn.Sequential): + return PipeSequential(*[copy.copy(m) for m in sequential_instance]) + + def forward(self, input): + for i, module in enumerate(self): + input = module(input) + if i != len(self) - 1: + pipe_split() + return input + + +class LossWrapper(torch.nn.Module): + """ + LossWrapper is a convenient abstract class that allows you to wrap up both + your model as well as its loss function and specify the connectivity between + the inputs, model, loss function, and output value. Example:: + + class MyModelWrapper(LossWrapper): + def forward(self, x, targets): + model_out = self.module(x) + loss_value = self.loss_fn(model_out, targets) + return loss_value + + The above example defines a connectivity where we expect the forward/loss/backward + training procedure to take two arguments (x and targets), pass x into the module + to get the output of the feedforward computation, pass the model output and the + targets value into the loss function, and get and return the loss value, which will + be backpropagated by PiPPy. The above class would then be instantiated like:: + + model = ... # instantiate the model + loss_fn = torch.nn.MSELoss() # for the sake of demonstration + + wrapper = MyModelWrapper(model, loss_fn) + pipe = Pipe.from_tracing(wrapper, ...) + + """ + + def __init__(self, module, loss_fn): + super().__init__() + self.module = module + self.loss_fn = loss_fn + + def forward(self, *args, **kwargs): + raise NotImplementedError( + "This instance of LossWrapper does not have an overridden" + "forward(). Please implement forward() to specify the arguments, " + "connection between the module and loss, and loss output " + "value." + ) + + +class TrivialLossWrapper(LossWrapper): + # pyrefly: ignore [bad-override] + def forward(self, x, targets): + model_out = self.module(x) + return self.loss_fn(model_out, targets) + + loss_spec = True + + +# Pipe model representation +# +# Pipe can be thought of as an `nn.Sequential++`. That is to say: it specifies +# a single topological ordering of pipeline "stages" that, when run in series, +# constitutes all of the operations of the program. However, unlike `nn.Sequential`, +# Pipe allows non-local usages of values, so long as those uses still respect +# topological ordering. In particular: +# +# 1. Non-local activations. This type of usage can appear in, for example, skip +# connections. These values will be directly transmitted from the "def" stage +# to all stages that use them skipping intermediate stages. During autograd, +# gradients will be propagated back through this skip connection reverse +# to how activations propagated in the forward pass. +# 2. Non-local parameter/module invocations. This occurs when a parameter is used +# in a stage downstream of where it is resident. These values can be carried +# forward similarly to (1), but in addition one might want to replicate the +# value on multiple stages. Gradients for these shared parameters will be +# accumulated separately on each stage, but there will be an additional +# gradient accumulation before the optimizer step. + + +# Register `_pipe_split()` as an ATen operator. This is required for Export to +# preserve this marker in the graph. +torch.library.define("pippy::_pipe_split", "() -> ()") + + +@torch.library.impl("pippy::_pipe_split", "BackendSelect") +def _pipe_split(): + return None + + +@torch.library.register_fake("pippy::_pipe_split") # type: ignore[no-redef] +def _pipe_split(): + return None + + +# Add an alias for convenience +aten_pipe_split_alias = torch.ops.pippy._pipe_split.default + +# Ask Export to preserve the `_pipe_split` op. +# See examples in pytorch/torch/fx/node.py +fx.node._side_effectful_functions.add(aten_pipe_split_alias) + + +# User facing API +def pipe_split(): + """ + pipe_split is a special operator that is used to mark the boundary between + stages in a module. It is used to split the module into stages. It is a + no-op if your annotated module is run eagerly. + + Example: + >>> # xdoctest: +SKIP + >>> def forward(self, x): + >>> x = torch.mm(x, self.mm_param) + >>> x = torch.relu(x) + >>> pipe_split() + >>> x = self.lin(x) + >>> return x + + The above example will be split into two stages. + """ + return torch.ops.pippy._pipe_split() + + +class MultiUseParameterConfig(Enum): + TRANSMIT = 1 + REPLICATE = 2 + + +MultiUseParamSpec = MultiUseParameterConfig | dict[str, MultiUseParameterConfig] + + +class DetachExecutor(fx.Interpreter): + """ + Special interpreter to run the split_gm in testing that detaches all inputs to + a module invocation. This is needed so that the values at the boundary are + leaf modules in autograd execution. + """ + + def __init__(self, module, garbage_collect_values=True): + garbage_collect_values = False + super().__init__(module, garbage_collect_values) + self.value_remap = {} + + def run(self, *args, initial_env=None): # type: ignore[override] + self.value_remap = {} + return super().run(*args, initial_env=initial_env) + + def call_module(self, target, args, kwargs): + def detach_tensors(a): + if isinstance(a, torch.Tensor) and a.requires_grad: + if a not in self.value_remap: + new_val = a.detach().requires_grad_(True) + self.value_remap[a] = new_val + return self.value_remap[a] + else: + return a + + """ + def dont_traverse_size(a): + return type(a) is not torch.Size + """ + + args = map_aggregate( + args, + detach_tensors, # dont_traverse_size + ) + kwargs = map_aggregate( + kwargs, + detach_tensors, # dont_traverse_size + ) + + return super().call_module(target, args, kwargs) + + def call_function(self, target, args, kwargs): + # HACK to reroute saved input tensors to point to the detach()ed version + if target is stage_backward: + kwargs = dict(kwargs) + kwargs["input_values"] = [ + self.value_remap.get(v, v) for v in kwargs["input_values"] + ] + return super().call_function(target, args, kwargs) + + +class _NodeReference: + def __init__(self, name): + self.name = name + + name: str + + +class _LinearNodeList: + def __init__(self, node_list): + self.serialize_node_list = [] + for node in node_list: + node_args = fx.node.map_arg(node.args, lambda n: _NodeReference(n.name)) # type: ignore[arg-type,return-value] + node_kwargs = fx.node.map_arg(node.kwargs, lambda n: _NodeReference(n.name)) # type: ignore[arg-type,return-value] + serialize_node = fx.Node( + graph=None, # type: ignore[arg-type] + name=node.name, + op=node.op, + target=node.target, + args=node_args, # type: ignore[arg-type] + kwargs=node_kwargs, # type: ignore[arg-type] + return_type=node.type, + ) + serialize_node.meta = copy.copy(node.meta) + self.serialize_node_list.append(serialize_node) + + def to_graph(self): + graph = fx.Graph() + + ref_str_to_node: dict[str, fx.Node] = {} + + def ref_to_node(arg): + if isinstance(arg, _NodeReference): + return ref_str_to_node[arg.name] + else: + return arg + + for node in self.serialize_node_list: + node_args = map_aggregate(node.args, ref_to_node) + node_kwargs = map_aggregate(node.kwargs, ref_to_node) + deser_node = graph.create_node( + op=node.op, + target=node.target, + args=node_args, # type: ignore[arg-type] + kwargs=node_kwargs, # type: ignore[arg-type] + name=node.name, + type_expr=node.type, + ) + ref_str_to_node[node.name] = deser_node + + return graph + + +def _direct_serialization_deserialize(body, nodes): + """ + Custom `__reduce__` method for serialization. + DO AS I SAY -- NOT AS I DO. This violates the principle that + GraphModules serialize via code export & re-tracing. We allow + for this here because **PIPE STAGES SHOULD NOT BE PERSISTED + TO DISK -- THIS IS ONLY FOR TRANSMISSION VIA RPC**. Persisting + these instances to disk will expose internal implementation + details of `fx.Graph` and related data structures and is + NOT advised. + """ + + class DummyModule(torch.nn.Module): + def __init__(self, body): + super().__init__() + self.__dict__.update(body) + + dummy = DummyModule(body) + + return fx.GraphModule(dummy, nodes.to_graph()) + + +def _direct_serialization_reduce(self): + serialization_dict = dict(self.__dict__) + serialization_dict.pop("_graph") + return ( + _direct_serialization_deserialize, + (serialization_dict, _LinearNodeList(self.graph.nodes)), + ) + + +def _modify_graph_op_device( + gm: torch.fx.GraphModule, + new_device: torch.device, +): + """ + Modify the device argument of all "call_function" nodes in the graph. This + is useful for moving the graph to a different device. In particular for + generator ops, like torch.ones. + """ + modified = False + for node in gm.graph.nodes: + if node.op == "call_function": + if "device" in node.kwargs and node.kwargs["device"] != new_device: + logger.debug( + f"Changing device of Node {node.name} from {node.kwargs['device']} to {new_device}" # noqa: G004 + ) + node.update_kwarg("device", new_device) + modified = True + elif node.op == "call_module": + # Recursively modify "device" in submodules + submod = gm.get_submodule(node.target) + if isinstance(submod, torch.fx.GraphModule): + _modify_graph_op_device(submod, new_device) + elif isinstance(submod, InterpreterModule): + # If unflattening has been performed, we need to access its graph module by `.graph_module` + _modify_graph_op_device(submod.graph_module, new_device) # type: ignore[arg-type] + else: + logger.warning( + f"Skipping device modification for submodule {node.target} because it is a {type(submod)}" # noqa: G004 + ) + + if modified: + gm.recompile() + + +class Pipe(torch.nn.Module): + def __init__( + self, + split_gm: fx.GraphModule, + num_stages: int, + has_loss_and_backward: bool, + loss_spec, + ): + # TODO: is there a way not to hard wire init? + torch.nn.Module.__init__(self) + self.split_gm: fx.GraphModule = split_gm + self.executor: DetachExecutor = DetachExecutor(self.split_gm) + self.num_stages: int = num_stages + self.has_loss_and_backward = has_loss_and_backward + self.loss_spec = loss_spec + + for node in split_gm.graph.nodes: + if not ( + node.op in {"call_module", "placeholder", "output"} + or (node.op, node.target) == ("call_function", operator.getitem) + or (node.op, node.target) == ("call_method", "backward") + or (node.op, node.target) == ("call_function", stage_backward) + or (node.op, node.target) + == ("call_function", _null_coalesce_accumulate) + ): + raise AssertionError(f"Unexpected node: {node}") + + # Detect replicated parameters so we know that we have to do an additional allreduce + # before applying the optimizer + # + # Note that this also handles the case where there were multiple calls to a single + # module from different stages, regardless of whether that module invocation + # was handled by the logic above. + + # Map parameter value to a dictionary that maps the user pipeline module + # to the local qualname within that module + params_to_users: dict[torch.nn.Parameter, dict[str, str]] = {} + + for m_qualname, mod in self.split_gm.named_children(): + for p_qualname, param in mod.named_parameters(): + params_to_users.setdefault(param, {}) + params_to_users[param][m_qualname] = p_qualname + + self.replicated_params: list[dict[str, str]] = [ + use_mapping + for use_mapping in params_to_users.values() + if len(use_mapping) > 1 + ] + + # We must break the aliasing relationship between the replicated parameters for correct + # numerics in reference runs. If we do not do this, the autograd tape in separate stages + # will have a reference to the same tensor value and will erroneously apply gradient + # updates multiple times. Therefore, for each replicated parameter set, we deepcopy the + # values so that we have separate instances. + for param_mapping in self.replicated_params: + for submod_name, param_qualname in param_mapping.items(): + submod = getattr(self.split_gm, submod_name) + atoms = param_qualname.split(".") + for atom in atoms[:-1]: + submod = getattr(submod, atom) + setattr(submod, atoms[-1], copy.deepcopy(getattr(submod, atoms[-1]))) + + def throw(self, *args, **kwargs): + raise RuntimeError( + "To run pipeline locally, invoke the Pipe object directly, not `split_gm`" + ) + + self.split_gm.forward = throw + + # Make submodules use custom direct-serialized GraphModule + i = 0 + while True: + try: + name = get_submod_name(i) + submod = getattr(self.split_gm, name) + submod.__class__.__reduce__ = _direct_serialization_reduce + i += 1 + except AttributeError: + break + + def forward(self, *args, **kwargs): + executor_args = args + if len(kwargs) > 0: + parameters = [] + for node in self.split_gm.graph.nodes: + if node.op == "placeholder": + if node.args and len(node.args) > 0: + parameters.append( + Parameter( + node.target, + Parameter.POSITIONAL_OR_KEYWORD, + default=node.args[0], + ) + ) + else: + parameter_kind = Parameter.POSITIONAL_OR_KEYWORD + param_name = node.target + if node.target.startswith("**"): + parameter_kind = Parameter.VAR_KEYWORD # type: ignore[assignment] + param_name = param_name[2:] + elif node.target.startswith("*"): + parameter_kind = Parameter.VAR_POSITIONAL # type: ignore[assignment] + param_name = param_name[1:] + parameters.append(Parameter(param_name, parameter_kind)) + signature = Signature(parameters) + ba = signature.bind(*args, **kwargs) + ba.apply_defaults() + executor_args = ba.arguments.values() # type: ignore[assignment] + + res = self.executor.run(*executor_args) + + return res + + def get_stage_module(self, stage_idx: int) -> torch.nn.Module: + """ + Return a stage module corresponding to `stage_idx` of the `pipe`. + """ + if stage_idx < 0 or stage_idx >= self.num_stages: + raise ValueError(f"Invalid stage index {stage_idx}!") + + submod_name = get_submod_name(stage_idx) + return getattr(self.split_gm, submod_name) + + @staticmethod + def _number_and_count_forward_stages(gm: fx.GraphModule): + num_stages = 0 + found_idxs: dict[int, None] = {} + for node in gm.graph.nodes: + if node.op == "call_module" and node.target.startswith(PP_SUBMOD_PREFIX): + node.meta["stage_idx"] = int(node.target[len(PP_SUBMOD_PREFIX) + 1 :]) + found_idxs.setdefault(node.meta["stage_idx"]) + num_stages += 1 + + # this assert will fail if a split point is inserted before the first layer, which creates empty first submodule + # Update: the following assert may fail against some torch versions >= + # 2.2.0, as: + # submod_0, submod_1, submod_2, ... + # may be named as + # submod_0, submod_2, submod_4, ... + # TODO: investigate + # assert all(i in found_idxs for i in range(num_stages)) + + return num_stages + + @staticmethod + def _from_traced( + mod: torch.nn.Module, + exported_program: ExportedProgram, + multi_use_param_spec: MultiUseParamSpec | None = None, + output_loss_value_spec=None, + split_policy: Callable[[torch.fx.GraphModule], torch.fx.GraphModule] + | None = None, + ): + """ + Additionally, the ``output_loss_value_spec`` value can be specified to disambiguate + which value in the output of `forward` is the loss value on which PiPPy should apply + backpropagation. For example, if your ``forward`` returns a tuple ``(loss, model_out)``, + you can specify ``output_loss_value_spec=(True, False)``. Or, if your ``forward`` returns + a dict ``{'loss': loss_value, 'model_out': model_out}``, you can specify + ``output_loss_value_spec={'loss': True, 'model_out': False}`` + """ + + traced = exported_program.module(check_guards=False) + + if split_policy is not None: + logger.info("Auto-splitting model") + traced = split_policy(traced) # type: ignore[arg-type] + + logger.debug(traced.print_readable(print_output=False)) # type: ignore[operator] + + # Deduplicate `get_attr` nodes that refer to the same parameter . Downstream code for moving + # parameters relies on the invariant that parameter accesses happen once. This is not necessarily + # the case (especially with custom tracers), so fix that up here. + get_attr_nodes: dict[str, fx.Node] = {} + for node in traced.graph.nodes: # type: ignore[union-attr] + if node.op == "get_attr": + get_attr_nodes.setdefault(node.target, node) + + if get_attr_nodes[node.target] != node: + node.replace_all_uses_with(get_attr_nodes[node.target]) + traced.graph.erase_node(node) # type: ignore[operator, union-attr] + + # avoid looking at next node by keeping track of previous pipe_split + prev_pipe_split_idx = -1 + pipe_split_nodes_to_erase = set() + for i, node in enumerate(traced.graph.nodes): # type: ignore[arg-type, union-attr] + if (node.op, node.target) == ("call_function", pipe_split): + if prev_pipe_split_idx == i - 1: + pipe_split_nodes_to_erase.add(node) + prev_pipe_split_idx = i + + for node in pipe_split_nodes_to_erase: + traced.graph.erase_node(node) # type: ignore[operator, union-attr] + + traced.recompile() # type: ignore[operator] + + part_idx = 0 + + def split_callback(n: fx.Node): + nonlocal part_idx + if (n.op, n.target) == ( + "call_function", + aten_pipe_split_alias, + ): + logger.debug(f"Found pipe_split {part_idx}") # noqa: G004 + part_idx += 1 + return part_idx + + # TODO: what does split do with module invocations? does it move the modules + # into the submodules? + split = split_module(traced, mod, split_callback, partition_affix="pp") # type: ignore[arg-type] + # a (custom) tracer can produce dead code like orphan get_attr nodes + split.graph.eliminate_dead_code() + + # peephole to remove pipe_split + for submodule in split.modules(): + if isinstance(submodule, fx.GraphModule): + for node in submodule.graph.nodes: + if (node.op, node.target) == ( + "call_function", + aten_pipe_split_alias, + ): + submodule.graph.erase_node(node) + submodule.recompile() + + for name, submodule in split.named_children(): + if isinstance(submodule, fx.GraphModule): + _move_placeholders_to_front(submodule.graph) + new_submod = _outline_submodules(submodule.graph) + # Replace old submod + split.register_module(name, new_submod) + + # TODO: backport this into split_module + def delete_user_reference(node, user): + """ + Delete reference of `node` from `user`'s arg list. + Args: + - node: a `get_attr` node at root. + - user: a submodule node that uses `node`. + """ + if not len(user.kwargs) == 0: + raise AssertionError( + f"Expected user.kwargs to be empty, got {len(user.kwargs)}" + ) + use_idxs = [i for i, arg in enumerate(user.args) if arg == node] + if not len(use_idxs) == 1: + raise AssertionError(f"Expected 1 use index, got {len(use_idxs)}") + args_copy = list(user.args) + args_copy.pop(use_idxs[0]) + user.args = tuple(args_copy) + logger.debug( + f"Deleted {node} from user {user}, arg index = {use_idxs[0]}" # noqa: G004 + ) + + # A list of param referrals for deferred deletion. + # To be accumulated in `move_param_to_callee`. + to_delete = [] + + def _recursive_getattr_with_parent(mod, fqn): + # Returns getattr call given a nested FQN, and the last parent + atoms = fqn.split(".") + for atom in atoms[:-1]: + if not hasattr(mod, atom): + return None, None + mod = getattr(mod, atom) + if not hasattr(mod, atoms[-1]): + return mod, None + attr = getattr(mod, atoms[-1]) + return mod, attr + + def move_param_to_callee( + root, + callee_name, + param_fqn, + ): + """ + Move a parameter from the root module to a submodule. + Args: + root: The root module. + callee_name: The name of the submodule to move the parameter to. + param_fqn: The fully qualified name of the parameter to move. + """ + # `atoms` is a list of strings representing the path to the + # parameter in the original model + atoms = param_fqn.split(".") + mod_itr, param_val = _recursive_getattr_with_parent(split, param_fqn) + # Check whether the parameter is a buffer or a parameter + is_buffer = atoms[-1] in mod_itr._buffers + + # Check whether the parameter is a tensor + if not isinstance(param_val, torch.Tensor): + raise AssertionError( + f"Expected '{param_fqn}' to be {torch.Tensor} but got {type(param_val)}." + + ( + f" It might happen if module '{param_fqn}' was passed to some 'leaf function'" + f"(see https://pytorch.org/docs/stable/fx.html#fx.wrap). Please inspect " + f"usages of '{param_fqn}' in the traced graph." + if isinstance(param_val, torch.nn.Module) + else "" + ) + ) + + # Get submodule + callee = root.get_submodule(callee_name) + if hasattr(callee, param_fqn): + raise AssertionError( + f"Module {callee_name} already has a parameter named {param_fqn}" + ) + + # Assign the parameter to the submodule + if is_buffer: + _assign_attr( + param_val, + callee, + param_fqn, + attr_kind=_AttrKind.BUFFER, + persistent=True, # TODO: handle non-persistent buffer + ) + else: + _assign_attr( + param_val, + callee, + param_fqn, + attr_kind=_AttrKind.PARAMETER, + ) + logger.debug(f"Moved parameter {param_fqn} to {callee_name}") # noqa: G004 + + # Next step is to replace placeholder of submodule with a get_attr. + # Those placeholders are created by `split_module` inside each + # submodule. + # Update: this step is now moved to `_sink_params` because + # `_sink_params` can do it recursively (i.e. for modules inside + # submodule) + + to_delete.append((mod_itr, atoms[-1])) + + # Get the list of all parameters in the root module + attr_nodes = list(filter(lambda n: n.op == "get_attr", split.graph.nodes)) + for node in attr_nodes: + # Check whether the parameter is used in only one submodule + if len(node.users) > 1: + logger.info( + f"Parameter {node.target} used in multiple stages: {node.users}." # noqa: G004 + ) + for user in node.users: + if not user.op == "call_module": + raise AssertionError( + f"Expected user.op to be 'call_module', got {user.op}" + ) + # Move parameter into submodule + move_param_to_callee( + split, + user.target, + node.target, + ) + + # [aliasing] store tensor id -> list of FQNs, built from state dict + # Also assign non-persistent buffers + id_to_fqns: dict[int, set[str]] = defaultdict(set) + for fqn, tensor in mod.state_dict(keep_vars=True).items(): + id_to_fqns[id(tensor)].add(fqn) + for fqn, tensor in mod.named_buffers(): + id_to_fqns[id(tensor)].add(fqn) + + # After moving the params to their corresponding hierarchies, we also + # need to move the `get_attr` nodes from the root of the graph to those + # hierarchies. + # [aliasing] use id -> fqn mapping to list out all valid FQNs + inputs_to_state: dict[str, list[str]] = {} + for attr in attr_nodes: + _, tensor = _recursive_getattr_with_parent(mod, attr.target) + fqns = list(id_to_fqns[id(tensor)]) + if fqns: + inputs_to_state[attr.name] = fqns + elif attr.target in exported_program.constants: # lifted constants + inputs_to_state[attr.name] = [attr.target] + + # [aliasing] for each submodule split, assign attributes on FQNs that may be used. + # We determine this based on whether or not the FQN attribute parent exists. + # i.e. if the last submodule exists, assign the attribute. + added_attributes: dict[str, list[str]] = defaultdict(list) + for fqn, tensor in mod.state_dict(keep_vars=True).items(): + for name, submod in split.named_children(): + if isinstance(submod, fx.GraphModule): + parent, child = _recursive_getattr_with_parent(submod, fqn) + if ( + parent and child is None + ): # parent exists, attribute doesn't -> assign + added_attributes[name].append(fqn) + setattr(parent, fqn.split(".")[-1], tensor) + + # Deferral deletion: Remove the original attributes (to params) from the + # root GraphModule + for mod_itr, last_atom in to_delete: + try: + delattr(mod_itr, last_atom) + except AttributeError: + # This is expected if the parameter is used in multiple stages + pass + + # This is done by (1) `_sink_params` at each submodule; + for submod in split.children(): + if isinstance(submod, fx.GraphModule): + _sink_params(submod, inputs_to_state, []) + submod.graph.lint() + submod.recompile() + + # [aliasing] This step is not super necessary, but helps reduce parameter usage/memory. + # After _sink_params() routine has run, clean up unused attributes that we previously added. + # Determine this based on the get_attr nodes - if not used, remove it. + for name, attributes in added_attributes.items(): + submod = getattr(split, name) + unused_attributes = set(attributes) + # track used attributes in the submodule, running DFS on subgraph hierarchy + stack = [("", submod)] # (scope, submodule) + while stack: + scope, _mod = stack.pop() + if isinstance(_mod, (fx.GraphModule, InterpreterModule)): + for node in _mod.graph.nodes: + if node.op == "get_attr": + # get_attr might get access deeper level attribute + fqn = scope + "." + node.target if scope else node.target + unused_attributes.discard(fqn) + for _name, _submod in _mod.named_children(): + stack.append((scope + "." + _name if scope else _name, _submod)) + # delete unused attributes + for attr in unused_attributes: + mod_itr, atoms = submod, attr.split(".") + for atom in atoms[:-1]: + mod_itr = getattr(mod_itr, atom) + delattr(mod_itr, atoms[-1]) + + for node in attr_nodes: + # And (2): remove `get_attr` node from submod's arg list + for user in copy.copy(node.users): + if not user.op == "call_module": + raise AssertionError( + f"Expected user.op to be 'call_module', got {user.op}" + ) + delete_user_reference(node, user) + # And (3): remove the `get_attr` node from the root graph. + split.graph.erase_node(node) + + split.delete_all_unused_submodules() + split.graph.lint() + split.recompile() + + num_stages = Pipe._number_and_count_forward_stages(split) + + has_loss_and_backward = False + generated_loss_spec = output_loss_value_spec + + if output_loss_value_spec is not None: + loss_node, output_node, generated_loss_spec = _find_loss_output( + mod, split.graph, output_loss_value_spec + ) + if loss_node is not None: + _insert_stage_symbolic_backward( + split.graph, + loss_node, + output_node, + ) + split.recompile() + has_loss_and_backward = True + logger.debug("Pipeline is in training mode, backward pass generated") + else: + raise RuntimeError( + f"Did not find any loss value according to {output_loss_value_spec=}" + ) + else: + logger.debug("Pipeline is in inference mode, backward pass not generated") + + logger.debug(f"Full pipe model:\n{split}") # noqa: G004 + + return Pipe( + split, + num_stages, + has_loss_and_backward, + generated_loss_spec, + ) + + def print_readable(self): + """ + Print the pipe in a human-readable format. + This will print both the root pipe and each stage module. + """ + self.split_gm.print_readable() + + @staticmethod + def _trace_with_export( + mod: torch.nn.Module, + example_args: tuple[Any, ...], + example_kwargs: dict[str, Any] | None = None, + ) -> ExportedProgram: + logger.info("Tracing model ...") + try: + ep = torch.export.export(mod, example_args, example_kwargs) + except Exception as e: + raise RuntimeError( + "It seems that we cannot capture your model as a full graph. " + "Typical reasons include graph breaks, data/shape-dependent " + "control flow, or missing meta kernels for custom operators. " + "You can use our manual pipeline interfaces, or try to fix the " + "graph breaks, see https://pytorch.org/docs/stable/export.html" + ) from e + + return ep + + @staticmethod + def from_tracing( + mod: torch.nn.Module, + example_args: tuple[Any, ...], + example_kwargs: dict[str, Any] | None = None, + split_policy: Callable[[fx.GraphModule], fx.GraphModule] | None = None, + ): + # If a param will be used in multiple pipeline stages, we default the strategy to REPLICATE'ing the param across + # stages instead of TRANSMIT'ting it + multi_use_param_spec = MultiUseParameterConfig.REPLICATE + + # Figure out which output is loss from output_chunk_spec + output_loss_value_spec: Any = None + # Deprecated + """ + if output_chunk_spec is not None: + output_loss_value_spec = map_aggregate( + output_chunk_spec, lambda v: isinstance(v, _LossReducer) + ) + """ + + # Trace with export + exported_program = Pipe._trace_with_export( + mod, + example_args, + example_kwargs, + ) + + pipe = Pipe._from_traced( + mod, + exported_program, + multi_use_param_spec, + output_loss_value_spec=output_loss_value_spec, + split_policy=split_policy, + ) + + # Users want the first pipeline stage to accept kwargs if the original + # program does. This is controlled by the `_codegen` field of the graph, + # so we make a copy here. Note: we only want the input spec and not the + # output spec, because the output spec is for the last stage. Maybe a + # TODO? Not sure yet. + split = pipe.split_gm + traced = exported_program.module() + submod0 = next(iter(split.children())) + submod0_sign = signature(submod0.forward) + model_sign = signature(traced.forward) + if len(model_sign.parameters) != len(submod0_sign.parameters): + # We don't change the signature of the first stage if it takes + # different number of args than original model + logger.info( + f"Original model takes {len(model_sign.parameters)} args but the " # noqa: G004 + f"first pipeline stage takes {len(submod0_sign.parameters)}. " + "Please provide args to respective pipeline stages." + ) + else: + # Support kwargs for the first stage + submod0.graph._codegen = copy.deepcopy(traced.graph._codegen) # type: ignore[union-attr] + # `_replace` is actually not "private" or internal. based on this doc: + # To prevent conflicts with field names, the method and attribute names + # start with an underscore + submod0.graph._codegen.pytree_info = ( # type: ignore[union-attr] + submod0.graph._codegen.pytree_info._replace(out_spec=None) # type: ignore[operator, union-attr] + ) + submod0.recompile() + + return pipe + + def __str__(self): + return self.split_gm.__str__() + + def __repr__(self): + return self.split_gm.__repr__() + + def info(self) -> PipeInfo: + """ + Get information about the pipe. + + Returns + ------- + PipeInfo + A dataclass containing information about the pipe. + """ + return PipeInfo( + graph=self.split_gm.graph, + num_stages=self.num_stages, + has_loss_and_backward=self.has_loss_and_backward, + ) + + def build_stage( + self, + stage_index: int, + device: torch.device, + group: ProcessGroup | None = None, + ) -> _PipelineStage: + """ + Create a `PipelineStage` given a stage index and distributed group. + The `PipelineStage` can run with `PipelineSchedule`s. + """ + # Find stage module + stage_module = self.get_stage_module(stage_index) + + # Move ops argument to device + # Today PT2 tracer does not treat `x.device` as a symbolic device; + # instead, the device of tracing time got burned into the generated + # code. Here we provide a workaround for users to manually modify the + # "device" kwarg of operations. Such operation may include: + # `torch.ones`, `torch.zeros`, `torch.rand`, etc. + if isinstance(stage_module, torch.fx.GraphModule): + _modify_graph_op_device(stage_module, device) + else: + logger.warning( + f"Expected a `torch.fx.GraphModule` but got {type(stage_module)}" # noqa: G004 + ) + + # Detach pipe info + # Note: be careful what's included in `pipe_info`. We don't want to keep + # a reference to `Pipe` or `Pipe.split_gm` which stops python from + # recycling them. When python recycles them, other stage modules (which + # are irrelevant to current rank) can be automatically freed. + pipe_info = self.info() + return _PipelineStage(stage_module, stage_index, pipe_info, device, group) + + +class SplitPoint(Enum): + """ + Enum representing the points at which a split can occur in the execution of a submodule. + Attributes: + BEGINNING: Represents adding a split point *before* the execution of a certain submodule in the `forward` function. + END: Represents adding a split point *after* the execution of a certain submodule in the `forward` function. + """ + + BEGINNING = 1 + END = 2 + + +# For backward compatibility, we kept the PipeSplitWrapper class because `class +# SplitPoint` used to be defined in this class. +class PipeSplitWrapper: + # Create a class alias for BC + SplitPoint = SplitPoint + + +def _split_before_forward(self, *args, **kwargs): + pipe_split() + return self._orig_forward(*args, **kwargs) + + +def _split_after_forward(self, *args, **kwargs): + try: + return self._orig_forward(*args, **kwargs) + finally: + pipe_split() + + +def annotate_split_points(mod: torch.nn.Module, spec: dict[str, SplitPoint]): + # TODO: make this implementation out-of-place? + for qualname, split_type in spec.items(): + atoms = qualname.split(".") + predecessor_module = mod + for i, atom in enumerate(atoms[:-1]): + try: + predecessor_module = getattr(predecessor_module, atom) + except AttributeError as e: + raise AttributeError( + f"Specified target {qualname} referenced " + f"nonexistent module {'.'.join(atoms[: i + 1])}" + ) from e + + mod_to_wrap = getattr(predecessor_module, atoms[-1]) + mod_to_wrap._orig_forward = mod_to_wrap.forward + if split_type == SplitPoint.BEGINNING: + mod_to_wrap.forward = MethodType(_split_before_forward, mod_to_wrap) + elif split_type == SplitPoint.END: + mod_to_wrap.forward = MethodType(_split_after_forward, mod_to_wrap) + else: + raise ValueError("Unknown split point type.") + + +def pipeline( + module: torch.nn.Module, + mb_args: tuple[Any, ...], + mb_kwargs: dict[str, Any] | None = None, + split_spec: dict[str, SplitPoint] | None = None, + split_policy: Callable[[fx.GraphModule], fx.GraphModule] | None = None, +) -> Pipe: + """ + Split a module based on a specification. + + See `Pipe` for more details. + + Arguments + --------- + module: + The module to be split. + mb_args: + Example positional inputs, in micro-batch form. + mb_kwargs: + Example keyword inputs, in micro-batch form. (default: `None`) + split_spec: + A dictionary using submodule names as split marker. (default: `None`) + split_policy: + The policy to use for splitting the module. (default: `None`) + + Returns + ------- + A pipeline representation of class `Pipe`. + """ + if split_spec is not None and split_policy is not None: + raise ValueError( + "Cannot specify both `split_spec` and `split_policy`. Please use only one of them." + ) + + if split_spec is not None: + # Annotate split points in the module based on user spec + annotate_split_points(module, split_spec) + return Pipe.from_tracing( + mod=module, + example_args=mb_args, + example_kwargs=mb_kwargs, + ) + else: + # Use split policy + return Pipe.from_tracing( + mod=module, + example_args=mb_args, + example_kwargs=mb_kwargs, + split_policy=split_policy, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..aacaf0b7f5e4ae7f5d221906ebb5b1b6ff93dea9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__init__.py @@ -0,0 +1,30 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from ._IR import Pipe, pipe_split, pipeline, SplitPoint +from .schedules import ( + _ScheduleForwardOnly, + Schedule1F1B, + ScheduleDualPipeV, + ScheduleGPipe, + ScheduleInterleaved1F1B, + ScheduleInterleavedZeroBubble, + ScheduleLoopedBFS, + ScheduleZBVZeroBubble, +) +from .stage import build_stage, PipelineStage + + +__all__ = [ + "Pipe", + "pipe_split", + "SplitPoint", + "pipeline", + "PipelineStage", + "build_stage", + "Schedule1F1B", + "ScheduleGPipe", + "ScheduleInterleaved1F1B", + "ScheduleLoopedBFS", + "ScheduleInterleavedZeroBubble", + "ScheduleZBVZeroBubble", + "ScheduleDualPipeV", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_IR.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_IR.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6f718c6cda5620822d30549407f920ddbad9477b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_IR.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fa9fc217d082b57dbcc0a9c62699c65bfdd06756 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_backward.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_backward.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..069756f80f7bdaa64f0e691d8fd4de7733d179ce Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_backward.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_debug.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_debug.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..db3d248cb384efd3317898b3fd07b0f92532cf2d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_debug.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_schedule_visualizer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_schedule_visualizer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4256ea0e0c7bd11e856ebc382411808fe1fdf38e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_schedule_visualizer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_unflatten.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_unflatten.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4ad117dd3e9cc97e523fecf108563f3afa3851de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_unflatten.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7931f4dfd95e33ee8a047751c12e5ebbfa9d06d0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/microbatch.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/microbatch.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..658b500edf833aa7136caf537e702110b1a641d9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/microbatch.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/schedules.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/schedules.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a170cde91882136699c61ca95bb9faad7b5ff323 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/schedules.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d03a7a211c16479ba3448f3e3e72d4f5c17a5d73508629a6f34d753d43b2b37 +size 153162 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/stage.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/stage.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..efca2ca0f6245ed4dc08b994088fb1a5879e51fd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/__pycache__/stage.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d8a314746291a1207367f9389b3501f28744adebd2fada14150e95bb4f81f95 +size 100519 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_backward.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_backward.py new file mode 100644 index 0000000000000000000000000000000000000000..bf55acbc958ff1197f607c37575e147611f199c4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_backward.py @@ -0,0 +1,492 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import collections +import logging +from collections.abc import Iterator, Sequence +from typing import Any + +import torch +from torch.autograd.graph import GradientEdge, Node +from torch.nn import Parameter + +from ._debug import map_debug_info + + +logger = logging.getLogger(__name__) + + +def _get_grad_fn_or_grad_acc(t: torch.Tensor) -> Node | None: + """ + Get the grad function or grad accumulator for a tensor. + + Accumulate grad nodes are lazily created, so we need to a + dummy view in order to trigger its creation. + """ + if t.requires_grad and t.grad_fn is None: + # if no grad function (leaf tensors) we use view + viewed_t = t.view_as(t) + grad_fn = viewed_t.grad_fn + if grad_fn is not None: + return grad_fn.next_functions[0][0] + else: + raise RuntimeError( + "Attempted to get grad_fn, but got None." + "Is this being created in a no-grad context?" + ) + else: + return t.grad_fn + + +def reverse_closure( + roots: list[Node], target_nodes: set[Node], reverse_edges_dict +) -> tuple[set[Node], set[Node]]: + """ + This function returns the reverse closure of the given roots, + i.e. the set of nodes that can be reached from the roots by following the + reverse edges of the graph. The target_nodes are the nodes that we want to + include in the closure. + """ + # Recurse until we reach a target node + closure: set[Node] = set() + visited_target_nodes = set() + q: collections.deque[Node] = collections.deque() + for node in roots: + if node is not None and node not in closure: + closure.add(node) + q.append(node) + while q: + node = q.popleft() + reverse_edges = reverse_edges_dict[node] + for fn in reverse_edges: + if fn in closure or fn is None: + continue + if fn in target_nodes: + visited_target_nodes.add(fn) + continue + closure.add(fn) + q.append(fn) + return closure, visited_target_nodes + + +def construct_reverse_graph(roots: list[Node]) -> dict[Node, list[Node]]: + q: collections.deque[Node] = collections.deque() + root_seen: set[Node] = set() + reverse_edges_dict: dict[Node, list[Node]] = collections.defaultdict(list) + for node in roots: + if node is not None and node not in root_seen: + q.append(node) + root_seen.add(node) + while q: + node = q.popleft() + for fn, _ in node.next_functions: + if fn is not None: + if len(reverse_edges_dict[fn]) == 0: + q.append(fn) + reverse_edges_dict[fn].append(node) + return reverse_edges_dict + + +def get_param_groups( + inputs: list[Node], params: list[Node], reverse_edges_dict +) -> list[dict[str, Any]]: + """ + Given a list of inputs and a list of parameters, return a list of parameter + groups, where each group contains the parameters and the intermediates that + are connected to the parameters. + + The returned list of parameter groups is a list of dictionaries, where each + dictionary contains the following keys: + - "params": a set of parameters + - "intermediates": a set of intermediates + + The returned list of parameter groups is a list of dictionaries, + """ + # reverse graph that starts with inputs, and goes up to the dOutput or the loss, + # but omits weights and any subgraphs connecting weights to this closure + inputs_closure, _ = reverse_closure(inputs, set(), reverse_edges_dict) + param_groups: dict[Node, dict[str, set]] = dict() # keyed on intermediates + for param in params: + closure, intersected = reverse_closure( + [param], inputs_closure, reverse_edges_dict + ) + param_group: dict[str, set] = { + "params": {param}, + "intermediates": intersected, + } + for input_node in intersected: + existing = param_groups.get(input_node) + if existing is not None: + existing["params"] = existing["params"].union(param_group["params"]) + existing["intermediates"] = existing["intermediates"].union( + param_group["intermediates"] + ) + param_group = existing + else: + param_groups[input_node] = param_group + + # Sanity check: union of all param_groups params should be equal to all params + union_params: set[Node] = set() + seen_ids: set[int] = set() + unique_param_groups = [] + for param_group in param_groups.values(): + if id(param_group) not in seen_ids: + seen_ids.add(id(param_group)) + unique_param_groups.append(param_group) + union_params = union_params.union(param_group["params"]) + + # The assert will only be true if the input tensor requires gradients, + # otherwise the autograd graph will miss the first layer of inputs + # assert union_params == set(params) + return unique_param_groups + + +def _autograd_grad_for_inputs( + outputs: Sequence[torch.Tensor], + inputs: Sequence[Any], + grad_outputs: Sequence[torch.Tensor | None] | None = None, + retain_graph: bool = False, + allow_unused: bool = False, +) -> tuple[torch.Tensor | None, ...]: + """Compute input gradients, returning ``None`` for non-grad inputs.""" + # Some inputs may not be used or may not require gradients, so we filter them out + # before calling autograd.grad and place None for those positions in the result. + grad_indices: list[int] = [] + inputs_requiring_grad: list[torch.Tensor] = [] + for i, inp in enumerate(inputs): + if isinstance(inp, torch.Tensor) and inp.requires_grad: + grad_indices.append(i) + inputs_requiring_grad.append(inp) + + if not inputs_requiring_grad: + return tuple(None for _ in inputs) + + grads = torch.autograd.grad( + outputs=outputs, + inputs=inputs_requiring_grad, + grad_outputs=grad_outputs, + retain_graph=retain_graph, + allow_unused=allow_unused, + ) + + result: list[torch.Tensor | None] = [None] * len(inputs) + for idx, g in zip(grad_indices, grads, strict=True): + result[idx] = g + return tuple(result) + + +def stage_backward_input( + stage_outputs_or_loss: list[torch.Tensor], + output_grads: list[torch.Tensor] | None, + input_values: list[Any], + weights: Iterator[Parameter], +) -> tuple[tuple[torch.Tensor | None, ...], list[dict[str, Any]]]: + """ + Compute the gradients for only the stage inputs with + respect to the stage outputs (if non-last stage) or loss (if last stage) + + After computing input gradients, we save the intermediate nodes in `param_groups` + for later use in stage_backward_weight. We don't need to save any other intermediate nodes + that aren't needed for dW because when we do dW calculation, we start from saved intermediates. + Detaching the stage_outputs_or_loss at the end of this function is important as + it frees up the memory that the autograd graph is anticipating to be used later (but doesn't actually need). + """ + valid_outputs: list[torch.Tensor] = [] + valid_output_grads: list[torch.Tensor | None] = [] + for i, stage_output in enumerate(stage_outputs_or_loss): + if not stage_output.requires_grad and stage_output.grad_fn is None: + continue + valid_outputs.append(stage_output) + valid_output_grads.append( + torch.ones_like(stage_output) if output_grads is None else output_grads[i] + ) + + stage_output_grad_fns: list[Node] = list( + filter(None, map(_get_grad_fn_or_grad_acc, valid_outputs)) + ) + stage_input_grad_fns: list[Node] = list( + filter( + None, + ( + _get_grad_fn_or_grad_acc(inp) + for inp in input_values + if isinstance(inp, torch.Tensor) + ), + ) + ) + weight_grad_fns: list[Node] = list( + filter(None, map(_get_grad_fn_or_grad_acc, weights)) + ) + + reverse_edges_dict = construct_reverse_graph(stage_output_grad_fns) + param_groups = get_param_groups( + stage_input_grad_fns, weight_grad_fns, reverse_edges_dict + ) + + handles = [] + try: + for param_group in param_groups: + intermediates = list(param_group["intermediates"]) + param_group["intermediates"] = intermediates + for i, intermediate in enumerate(intermediates): + + def get_hook(param_group, i): + def hook(grad_inputs): + if param_group.get("grads", None) is None: + param_group["grads"] = [None] * len( + param_group["intermediates"] + ) + param_group["grads"][i] = grad_inputs + + return hook + + handle = intermediate.register_prehook(get_hook(param_group, i)) + handles.append(handle) + + if valid_outputs: + dinputs = _autograd_grad_for_inputs( + valid_outputs, + input_values, + valid_output_grads, + retain_graph=True, + allow_unused=True, + ) + else: + dinputs = tuple(None for _ in input_values) + + for inp, dinput in zip(input_values, dinputs): + if isinstance(inp, torch.Tensor) and dinput is not None: + if inp.grad is None: + inp.grad = dinput + else: + inp.grad += dinput + + # drop output side graph state we no longer need + for t in stage_outputs_or_loss: + t.detach_() + + return dinputs, param_groups + except Exception as e: + exc_msg = f""" + Failed to run stage backward input: + Stage output: {map_debug_info(stage_outputs_or_loss)} + Output gradient: {map_debug_info(output_grads)} + Input: {map_debug_info(input_values)} + """ + raise RuntimeError(exc_msg) from e + finally: + for handle in handles: + handle.remove() + + +def stage_backward_weight( + weights: Iterator[Parameter], param_groups: list[dict[str, Any]], retain_graph=False +) -> tuple[torch.Tensor | None, ...]: + # map weights to param_group_weights + grad_acc_to_weight = {} + weight_grads: list[torch.Tensor | None] = [] + for index, weight in enumerate(weights): + grad_acc = _get_grad_fn_or_grad_acc(weight) + grad_acc_to_weight[grad_acc] = weight, index + weight_grads.append(weight.grad) + + for param_group in param_groups: + valid_edges = [] + valid_grad_outputs: list[torch.Tensor] = [] + + for grads_tuple, intermediate in zip( + param_group["grads"], param_group["intermediates"] + ): + for i, grad in enumerate(grads_tuple): + if grad is not None: + valid_edges.append(GradientEdge(intermediate, i)) + # pyrefly: ignore [bad-argument-type] + valid_grad_outputs.append(grad) + + # clamp saved grads to avoid double counting through descendant intermediates. + handles = [] + try: + if len(param_group["intermediates"]) > 1: + for grads_tuple, intermediate in zip( + param_group["grads"], + param_group["intermediates"], + strict=True, + ): + handles.append( + intermediate.register_prehook( + lambda grad_outputs, sg=grads_tuple: sg + ) + ) + + # break the hook -> param_group -> intermediates cycle + del param_group["intermediates"] + + if valid_edges: + weights_edges = tuple(GradientEdge(w, 0) for w in param_group["params"]) + dweights = torch.autograd.grad( + valid_edges, + weights_edges, + grad_outputs=valid_grad_outputs, + retain_graph=retain_graph, + ) + + del param_group["grads"] + + for grad_acc, dw in zip(param_group["params"], dweights): + weight, index = grad_acc_to_weight[grad_acc] + if weight.grad is None: + weight.grad = dw + else: + weight.grad += dw + finally: + for handle in handles: + handle.remove() + return tuple(weight_grads) + + +def stage_backward( + stage_output, + output_grads, + input_values, + outputs_with_grads_idxs: list[int] | None = None, # deprecated, not used +) -> tuple[torch.Tensor | None, ...]: + """ + This is a helper function to: + 1. compute the gradients for the stage inputs, and + 2. accumulate gradients for the stage module's parameters. + + Given the input value(s) and the corresponding gradient for the output + value(s), compute and accumulate gradients for all parameter values (leaves + in the autograd trace) as well as return a list of the gradients for the + input values + """ + if outputs_with_grads_idxs is not None: + # Deprecated, not used in runtime calls, only exists in compiler + stage_output = [stage_output[i] for i in outputs_with_grads_idxs] + output_grads = [output_grads[i] for i in outputs_with_grads_idxs] + + try: + # stage_output may be a composite datatype like dict. Extract all individual + # tensor values here + stage_output_tensors: list[torch.Tensor] = [] + output_grad_tensors: list[torch.Tensor | None] = [] + + def extract_tensors_with_grads( + output_val, + grad_val, + # Don't delete me- see [Note: ref cycle] + extract_tensors_with_grads, + ): + if isinstance(output_val, torch.Tensor): + if not output_val.requires_grad and output_val.grad_fn is None: + return + if not isinstance(grad_val, (torch.Tensor, type(None))): + raise AssertionError( + f"Expected Tensor or None gradient but got {type(grad_val)}" + ) + stage_output_tensors.append(output_val) + output_grad_tensors.append(grad_val) + elif isinstance(output_val, (tuple, list)): + if grad_val is None: + return + if not isinstance(grad_val, (tuple, list)): + raise AssertionError( + f"grad_value expected to have type {type(output_val)} but got {type(grad_val)}" + ) + if not len(output_val) == len(grad_val): + raise AssertionError( + f"Expected len(output_val) == len(grad_val), got {len(output_val)} != {len(grad_val)}" + ) + for ov, gv in zip(output_val, grad_val): + extract_tensors_with_grads( + ov, + gv, + extract_tensors_with_grads, + ) + elif isinstance(output_val, dict): + if grad_val is None: + return + if not isinstance(grad_val, dict): + raise AssertionError(f"Expected dict, got {type(grad_val)}") + if not set(output_val.keys()) == set(grad_val.keys()): + raise AssertionError( + f"Expected keys {set(output_val.keys())}, got {set(grad_val.keys())}" + ) + for k in output_val: + extract_tensors_with_grads( + output_val[k], grad_val[k], extract_tensors_with_grads + ) + else: + # Output is a non-tensor type; just ignore it + pass + + # Note: ref cycle + # break a ref cycle that would keep tensors alive until GC runs + # 1. extract_tensors_with_grads refers to a cell that holds refs to any vars defined in stage_backward + # and used in extract_tensors_with_grads + # 2. extract_tensors_with_grads referred to both stage_output_tensors, output_grad_tensors, + # and to itself (extract_tensors_with_grads) since it makes a recursive call + # 3. stage_output_tensors was kept alive by the above refcycle, and it holds activation tensors, which is bad + # fix -> explicitly pass in the ref to the fn, so there is no gc cycle anymore + extract_tensors_with_grads( + stage_output, output_grads, extract_tensors_with_grads + ) + + torch.autograd.backward( + stage_output_tensors, + grad_tensors=output_grad_tensors, # type: ignore[arg-type] + ) + + # Extract gradients wrt the input values + grad_inputs: list[torch.Tensor | None] = [] + for val in input_values: + if isinstance(val, torch.Tensor): + grad_inputs.append(val.grad) + # Since gradients that will pass back to previous stages do not require gradient accumulation, + # by decrementing the gradients' reference count at this point, the memory of gradients will be + # returned to the allocator as soon as the next micro batch's get_bwd_send_ops comes and current + # asynchronous send completes. + # This prevents the gradients from persisting in GPU memory for the entire duration of step_microbatches + # until clear_runtime_states() is called. + val.grad = None + else: + grad_inputs.append(None) + + # Alternative impl: `torch.autograd.grad`. + # Note that `torch.autograd.grad` will not accumulate gradients into the + # model's parameters. + """ + inputs_with_grad = [] + for val in input_values: + if isinstance(val, torch.Tensor) and val.requires_grad: + inputs_with_grad.append(val) + + grad_inputs = torch.autograd.grad( + stage_output_tensors, inputs_with_grad, output_grad_tensors, # type: ignore[arg-type] + ) + """ + + except Exception as e: + exc_msg = f""" + Failed to run stage backward: + Stage output: {map_debug_info(stage_output)} + Output gradient: {map_debug_info(output_grads)} + Input: {map_debug_info(input_values)} + """ + raise RuntimeError(exc_msg) from e + + return tuple(grad_inputs) + + +# TODO: handling requires_grad=False dynamically. Can we analyze this during initial +# IR emission? +def _null_coalesce_accumulate(lhs, rhs): + """ + Coalesce two values, even if one of them is null, returning the non-null + value. + """ + if lhs is None: + return rhs + elif rhs is None: + return lhs + else: + return torch.add(lhs, rhs) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_debug.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_debug.py new file mode 100644 index 0000000000000000000000000000000000000000..a3201d2d3adf1d05921e070d14b4e544844df88f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_debug.py @@ -0,0 +1,22 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import torch +from torch.fx.node import Argument + + +def friendly_debug_info(v: object) -> Argument: + """ + Helper function to print out debug info in a friendly way. + """ + if isinstance(v, torch.Tensor): + return f"Tensor({v.shape}, grad={v.requires_grad}, dtype={v.dtype})" + else: + return str(v) + + +def map_debug_info(a: Argument) -> Argument: + """ + Helper function to apply `friendly_debug_info` to items in `a`. + `a` may be a list, tuple, or dict. + """ + return torch.fx.node.map_aggregate(a, friendly_debug_info) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_schedule_visualizer.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_schedule_visualizer.py new file mode 100644 index 0000000000000000000000000000000000000000..e088d9124b924a2c07478cf6fb953cb8fbbf666d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_schedule_visualizer.py @@ -0,0 +1,455 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +""" +This visualizer requires matplotlib to be installed. + +Example usage: + +ops = get_schedule_ops("InterleavedZeroBubble", 4, 8) +visualize_schedule(ops, "test.png") +""" + +import collections +from typing import NamedTuple +from unittest import mock + +from torch.distributed.pipelining.schedules import ( + _Action, + _ComputationType, + _PipelineSchedule, + _PipelineScheduleRuntime, + get_schedule_class, + PipelineScheduleMulti, + PipelineScheduleSingle, +) +from torch.distributed.pipelining.stage import PipelineStage + + +class OpKey(NamedTuple): + stage_index: int + computation_type: _ComputationType + microbatch_index: int + + +def get_schedule_ops( + schedule: str | type[_PipelineSchedule], + pp_degree: int, + num_microbatches: int, + num_stages_per_rank: int | None = None, + add_spacing: bool = False, + with_comms: bool = False, + defer_pp_recv: bool = False, +) -> list[list[_Action | None]]: + """ + Get all actions for a given schedule, pp_degree, and num_microbatches. The actions are returned in a list of lists + where each inner list represents a rank and each element in the inner list represents an action. + + The schedule can be specified as a string which is passed into get_schedule_class() or a _PipelineSchedule instance. + + Args: + defer_pp_recv: If True, RECV ops are deferred to right before the consuming + compute op. If False (default), RECV ops are placed as early as possible to + overlap P2P communication with computation. Only takes effect when with_comms=True. + """ + if add_spacing and with_comms: + raise ValueError("Cannot add spacing and view comms at the same time") + + if isinstance(schedule, str): + schedule_class = get_schedule_class(schedule) + elif issubclass(schedule, _PipelineSchedule): + schedule_class = schedule + else: + raise ValueError(f"Invalid schedule: {schedule}") + + # Create a mock of the PipelineStage class + mock_pipeline_stage = mock.create_autospec(PipelineStage, instance=True) + # Set the return values for group_rank and group_size methods + mock_pipeline_stage.group_rank = 0 + mock_pipeline_stage.group_size = pp_degree + mock_pipeline_stage.submod = None + + # Check num_stages_per_rank is valid + if issubclass(schedule_class, PipelineScheduleSingle): + if num_stages_per_rank is None: + num_stages_per_rank = 1 + if not num_stages_per_rank == 1: + raise AssertionError( + f"Expected num_stages_per_rank to be 1, got {num_stages_per_rank}" + ) + stages = mock_pipeline_stage + stages.num_stages = num_stages_per_rank * pp_degree + elif issubclass(schedule_class, PipelineScheduleMulti): + if num_stages_per_rank is None: + num_stages_per_rank = 2 + if not num_stages_per_rank >= 2: + raise AssertionError( + f"Expected num_stages_per_rank >= 2, got {num_stages_per_rank}" + ) + stages = [mock_pipeline_stage for _ in range(num_stages_per_rank)] + for stage in stages: + stage.num_stages = num_stages_per_rank * pp_degree + + else: + raise ValueError(f"Invalid schedule: {schedule_class}") + + # Instantiate the schedule class + # pyrefly: ignore [bad-argument-type] + schedule_instance = schedule_class(stages, num_microbatches) + if schedule_instance.pipeline_order is None: + raise AssertionError("Expected pipeline_order to not be None") + + # Convert to List[List[_Action]] + all_actions: list[list[_Action | None]] = [] + if with_comms: + runtime = _PipelineScheduleRuntime( + stages, num_microbatches, defer_pp_recv=defer_pp_recv + ) + runtime._prepare_schedule_with_comms(schedule_instance.pipeline_order) + for rank in range(pp_degree): + all_actions.append(list(runtime.pipeline_order_with_comms[rank])) + else: + for rank in range(pp_degree): + all_actions.append(schedule_instance.pipeline_order[rank]) + + # Add spacing + if add_spacing: + # remove all Nones, then respace + # TODO: later we can change this at the schedule creation level to not use Nones + all_actions = [ + [action for action in rank if action is not None] for rank in all_actions + ] + all_actions = add_schedule_op_spacing(all_actions) + + # Return the pipeline order + return all_actions + + +class _ComputationTypeVisual: + def __init__( + self, + color: str, + text: str = "", + width: int = 1, + ): + self.color = color + self.width = width + self.text = text + + +# Update the mapping to use _ComputationTypeVisual instances +action_type_to_color_mapping = { + _ComputationType.FORWARD: _ComputationTypeVisual("blue", "Forward"), + _ComputationType.BACKWARD_INPUT: _ComputationTypeVisual("teal", "Backward Input"), + _ComputationType.BACKWARD_WEIGHT: _ComputationTypeVisual( + "green", "Backward Weight" + ), + _ComputationType.FULL_BACKWARD: _ComputationTypeVisual( + "orange", "Full Backward", 2 + ), + _ComputationType.OVERLAP_F_B: _ComputationTypeVisual("purple", "Overlap F+B", 3), +} + + +def add_schedule_op_spacing( + schedule: list[list[_Action | None]], +) -> list[list[_Action | None]]: + """ + Add spacing to the schedule based on dependencies between ranks. + + Before adding an operation to the list, this function checks if there are + dependencies from other ranks. If there are dependencies (other ranks have + not finished processing the required microbatch), it adds None instead. + + For example, Forward microbatch 0 on rank 1 depends on rank 0 processing + Forward microbatch 0 first. + + Args: + schedule: The original schedule as a list of lists where each inner list + represents a rank and each element represents an action. + + Returns: + A new schedule with proper spacing based on dependencies. + """ + if not schedule: + return schedule + + num_stages = ( + max( + action.stage_index + for rank_actions in schedule + for action in rank_actions + if action is not None + ) + + 1 + ) + + num_ranks = len(schedule) + spaced_schedule: list[list[_Action | None]] = [[] for _ in range(num_ranks)] + rank_ops = [collections.deque(ops) for ops in schedule] + + # Track completion times: (stage_index, action_type, microbatch_index) -> completion_time + scheduled_ops: dict[OpKey, int] = {} + + def is_dependency_ready(dependency_key: OpKey, timestep: int) -> bool: + """Check if a dependency operation has completed by the given timestep.""" + return ( + dependency_key in scheduled_ops + and timestep >= scheduled_ops[dependency_key] + ) + + def get_dependencies(action: _Action) -> list[OpKey]: + """Get the list of dependencies for an action.""" + stage_idx = action.stage_index + comp_type = action.computation_type + mb_idx = action.microbatch_index + + # Ensure mb_idx is not None for dependency tracking + if mb_idx is None: + raise AssertionError(f"Action {action} has None microbatch_index") + + # First stage forward has no dependencies + if stage_idx == 0 and comp_type == _ComputationType.FORWARD: + return [] + + # Last stage backward depends on forward from previous stage + if stage_idx == num_stages - 1 and comp_type in ( + _ComputationType.FULL_BACKWARD, + _ComputationType.BACKWARD_INPUT, + ): + return [OpKey(stage_idx - 1, _ComputationType.FORWARD, mb_idx)] + + # Forward depends on previous stage forward + if comp_type == _ComputationType.FORWARD: + return [OpKey(stage_idx - 1, _ComputationType.FORWARD, mb_idx)] + + # Backward depends on next stage backward + if comp_type in ( + _ComputationType.FULL_BACKWARD, + _ComputationType.BACKWARD_INPUT, + ): + return [ + OpKey(stage_idx + 1, _ComputationType.FULL_BACKWARD, mb_idx), + OpKey(stage_idx + 1, _ComputationType.BACKWARD_INPUT, mb_idx), + ] + + # Weight backward depends on input backward + if comp_type == _ComputationType.BACKWARD_WEIGHT: + return [OpKey(stage_idx, _ComputationType.BACKWARD_INPUT, mb_idx)] + + raise RuntimeError(f"Unknown computation type: {comp_type}") + + def is_action_ready(action: _Action, timestep: int) -> bool: + """Check if an action is ready to be scheduled at the given timestep.""" + # For OR dependencies (like backward), check if any dependency is satisfied + if action.computation_type in ( + _ComputationType.FULL_BACKWARD, + _ComputationType.BACKWARD_INPUT, + _ComputationType.BACKWARD_WEIGHT, + ): + dependencies = get_dependencies(action) + return any(is_dependency_ready(dep, timestep) for dep in dependencies) + # For AND dependencies, all must be satisfied + elif action.computation_type == _ComputationType.FORWARD: + dependencies = get_dependencies(action) + return all(is_dependency_ready(dep, timestep) for dep in dependencies) + elif action.computation_type == _ComputationType.OVERLAP_F_B: + if action.sub_actions is None: + raise AssertionError( + f"OVERLAP_F_B action {action} has None sub_actions" + ) + dep_list: list[bool] = [] + for sub_action in action.sub_actions: + dep_list.append(is_action_ready(sub_action, timestep)) + return all(dep_list) + else: + raise RuntimeError(f"Unknown computation type: {action.computation_type}") + + def schedule_action(action: _Action, rank: int, timestep: int) -> int: + """Schedule an action and return completion time.""" + spaced_schedule[rank].append(action) + comp_type = action.computation_type + comp_time = action_type_to_color_mapping[comp_type].width + completion_time = timestep + comp_time + + if comp_type == _ComputationType.OVERLAP_F_B: + # For overlap actions, schedule each sub-action with cumulative timing + if action.sub_actions is None: + raise AssertionError( + f"OVERLAP_F_B action {action} has None sub_actions" + ) + cumulative_time = 0 + for sub_action in action.sub_actions: + if sub_action.microbatch_index is None: + raise AssertionError( + f"Sub-action {sub_action} has None microbatch_index" + ) + sub_comp_time = action_type_to_color_mapping[ + sub_action.computation_type + ].width + cumulative_time += sub_comp_time + scheduled_ops[ + OpKey( + sub_action.stage_index, + sub_action.computation_type, + sub_action.microbatch_index, + ) + ] = timestep + cumulative_time + else: + if action.microbatch_index is None: + raise AssertionError(f"Action {action} has None microbatch_index") + scheduled_ops[ + OpKey(action.stage_index, comp_type, action.microbatch_index) + ] = completion_time + + return completion_time + + # Main scheduling loop + current_timestep = 0 + timesteps_without_progress = 0 + rank_completion_times = dict.fromkeys(range(num_ranks), 0) + while rank_ops: + print(f"Current timestep: {current_timestep}") + # Process all operations during timestep until we run out of ready operations + for rank, op_queue in enumerate(rank_ops): + if not op_queue: + continue + + op_queue = rank_ops[rank] + action = op_queue[0] + print(f"Rank: {rank}, {action=}") + if action is None: + spaced_schedule[rank].append(None) + op_queue.popleft() + timesteps_without_progress = 0 + elif current_timestep >= rank_completion_times[rank] and is_action_ready( + action, current_timestep + ): + rank_completion_times[rank] = schedule_action( + action, rank, current_timestep + ) + op_queue.popleft() + timesteps_without_progress = 0 + + # Add None for ranks that are waiting + for rank in range(num_ranks): + if current_timestep >= rank_completion_times[rank]: + spaced_schedule[rank].append(None) + + # Remove empty queues and advance timestep + rank_ops = [op_queue for op_queue in rank_ops if op_queue] + current_timestep += 1 + timesteps_without_progress += 1 + + if timesteps_without_progress > max( + visual.width for visual in action_type_to_color_mapping.values() + ): + raise RuntimeError("No progress made in scheduling - possible deadlock") + + return spaced_schedule + + +def visualize_schedule( + schedule: list[list[_Action | None]], + filename: str | None = None, +) -> None: + """ + Visualize the schedule using matplotlib. + The schedule is a list of lists where each inner list represents a rank and each element in the inner list represents an action. + The actions are represented as rectangles with different colors based on their computation type. + The filename is optional and if provided, the plot will be saved to that file. + + Args: + schedule: The schedule to visualize. + filename: The filename to save the plot to. If not provided, the plot will be displayed. + add_schedule_spacing: If True, add spacing to the schedule based on dependencies between ranks. + + """ + + import matplotlib.pyplot as plt + from matplotlib.patches import Rectangle + + plt.rcParams["font.family"] = ( + "DejaVu Sans" # or any other font available on your system + ) + num_ranks = len(schedule) + max_actions = max(len(rank) for rank in schedule) + + # Increase the figure size to provide more space for the legend + fig, ax = plt.subplots(figsize=(max_actions + 2, num_ranks + 2)) + max_draw_position = -1 + # Calculate dynamic font size based on figure size + font_size = min(max_actions, num_ranks) + 4 + used_computation = set() + for rank_idx, actions in enumerate(schedule): + draw_position = 0 # Initialize drawing position for each rank + for action in actions: + if action is not None: + comp_type_color = action_type_to_color_mapping.get( + action.computation_type, _ComputationTypeVisual("black") + ) + used_computation.add(action.computation_type) + color = comp_type_color.color + width = comp_type_color.width + + # Check if action has sub_actions to determine styling + if action.sub_actions is not None: + linewidth = 2 # Thicker border for compound actions + text_weight = "normal" # Bold text for compound actions + else: + linewidth = 1 # Default linewidth for regular actions + text_weight = "normal" # Default text weight + + # Draw the rectangle to represent the action duration + rect = Rectangle( + (draw_position, num_ranks - rank_idx - 1), + width, + 1, + facecolor=color, + edgecolor="black", + linewidth=linewidth, + ) + ax.add_patch(rect) + + # Draw the text centered within the rectangle + ax.text( + draw_position + width / 2, + num_ranks - rank_idx - 1 + 0.5, + str(action), + ha="center", + va="center", + fontsize=font_size, + color="white", + weight=text_weight, + ) + + draw_position += width + else: + draw_position += 1 # Move to the next + max_draw_position = max(max_draw_position, draw_position) + ax.set_xlim(-0.5, max_draw_position + 1) + ax.set_ylim(-0.5, num_ranks + 0.5) # Add extra space at the top + # Set y-ticks to be in the middle of each rank's row + ax.set_yticks([num_ranks - rank_idx - 0.5 for rank_idx in range(num_ranks)]) + ax.set_yticklabels([f"Rank {i}" for i in range(num_ranks)], fontsize=font_size) + ax.set_xticklabels([]) + + # Remove grid lines and ticks + ax.grid(False) + # Add legend with larger font size + legend_elements = [ + Rectangle( + (0, 0), + 1, + 1, + facecolor=action_type_to_color_mapping[comp_type].color, + edgecolor="black", + label=action_type_to_color_mapping[comp_type].text, + ) + for comp_type in used_computation + ] + ax.legend(handles=legend_elements, loc="upper right", fontsize=font_size) + # Save to file if filename is provided, otherwise display the plot + if filename: + plt.savefig(filename, bbox_inches="tight") + else: + plt.show() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_unflatten.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_unflatten.py new file mode 100644 index 0000000000000000000000000000000000000000..0ed592f2f8d832de0703fbfa296225f17698afbf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_unflatten.py @@ -0,0 +1,30 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from collections import defaultdict + +import torch +from torch.export.unflatten import _ModuleFrame, _SubmoduleEntry + + +def _outline_submodules(orig_graph: torch.fx.Graph) -> torch.fx.GraphModule: + # Create an empty GraphModule to hold the outlined modules + new_module = torch.fx.GraphModule(torch.nn.Module(), torch.fx.Graph()) + seen_nodes: dict[str, torch.fx.Node] = {} + seen_modules: dict[int, list[_SubmoduleEntry]] = defaultdict(list) + seen_attrs: dict[str, set[str]] = defaultdict(set) + created_modules: dict[str, torch.nn.Module] = {} + _ModuleFrame( + orig_graph, + tuple(orig_graph.nodes), + seen_nodes, + seen_modules, + seen_attrs, + created_modules, + None, + [("", None, 0)], + "", + {}, + module=new_module, + ).run_outer() + new_module.graph.lint() + new_module.recompile() + return new_module diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..d60ce237b0bea65d54c3fc2695568a3a05a5b9de --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/_utils.py @@ -0,0 +1,947 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates + +from __future__ import annotations + +import logging +import warnings +from dataclasses import dataclass, field +from enum import Enum +from typing import cast, Literal, overload, Protocol, TYPE_CHECKING, TypeAlias + +import torch +from torch import fx +from torch.distributed._mesh_layout import _MeshLayout +from torch.distributed.tensor import DTensor +from torch.utils._pytree import tree_flatten, tree_unflatten + + +if TYPE_CHECKING: + from torch.distributed.device_mesh import DeviceMesh + from torch.distributed.tensor.placement_types import Placement + + +logger = logging.getLogger(__name__) + + +class GetMeshCallback(Protocol): + """Callback to create/retrieve a DeviceMesh from its cache key components.""" + + def __call__( + self, + mesh_dim_names: tuple[str, ...], + mesh_layout: _MeshLayout | None, + ) -> DeviceMesh: ... + + +# Key for mesh cache: (mesh_dim_names, mesh_layout) +# mesh_layout is the _MeshLayout object containing shape and stride (not actual ranks). +# This uniquely identifies a mesh within the same "universe" where all stages share +# the same rank tensor. +MeshCacheKey: TypeAlias = tuple[tuple[str, ...], _MeshLayout | None] + + +class PipeliningMetadataError(RuntimeError): + """Raised on metadata mismatches during pipeline communication.""" + + +@dataclass(frozen=True, slots=True) +class _TensorMeta: + """Tensor metadata for recv buffer allocation and validation. + + For plain tensors, these are the tensor's actual attributes. + For DTensors, these are LOCAL shard attributes; global attributes + are stored in :class:`_DTensorMeta`. + """ + + shape: torch.Size + stride: tuple[int, ...] + dtype: torch.dtype + requires_grad: bool + + @staticmethod + def from_tensor(tensor: torch.Tensor) -> _TensorMeta: + """Create metadata from a plain tensor. + + Args: + tensor: A plain ``torch.Tensor`` (not DTensor). + + Returns: + Metadata capturing shape, stride, dtype, and requires_grad. + + Raises: + TypeError: If ``tensor`` is a DTensor. + """ + if isinstance(tensor, DTensor): + raise PipeliningMetadataError( + "Expected plain tensor, got DTensor. Use _DTensorMeta.from_dtensor instead." + ) + return _TensorMeta( + shape=tensor.shape, + stride=tensor.stride(), + dtype=tensor.dtype, + requires_grad=tensor.requires_grad, + ) + + def to_tensor(self, device: torch.device | str) -> torch.Tensor: + """Reconstruct a tensor on ``device`` from this metadata. + + Args: + device: Target device for the tensor. + + Returns: + An empty strided tensor on ``device``. + """ + t = _make_tensor_from_meta(self, device) + if t.is_floating_point(): + t.requires_grad_(self.requires_grad) + return t + + def get_diff(self, other: _TensorMeta) -> list[str]: + """Return field-by-field differences with ``other``. + + Args: + other: Metadata to compare against. + + Returns: + List of human-readable difference strings (empty if equal). + """ + if self == other: + return [] + + diffs = [] + if self.shape != other.shape: + diffs.append(f"shape mismatch: {self.shape} vs {other.shape}") + if self.stride != other.stride: + diffs.append(f"stride mismatch: {self.stride} vs {other.stride}") + if self.dtype != other.dtype: + diffs.append(f"dtype mismatch: {self.dtype} vs {other.dtype}") + # requires_grad is intentionally excluded: it is a runtime concern + # determined by has_backward and grad context, not a metadata invariant. + return diffs + + +@dataclass(frozen=True, slots=True) +class _DTensorMeta(_TensorMeta): + """DTensor metadata extending :class:`_TensorMeta` with distribution info. + + Inherited fields (shape, stride, etc.) are LOCAL shard attributes. + Additional fields capture global shape and placement information + needed to reconstruct a :class:`DTensor` via ``DTensor.from_local()``. + + The :class:`DeviceMesh` is **not** stored (not serializable for P2P); + it is looked up from :class:`_MeshCache` using + ``(mesh_dim_names, mesh_layout)`` as the key. + """ + + # Global DTensor properties (for reconstruction) + global_shape: torch.Size = field(default_factory=lambda: torch.Size([])) + global_stride: tuple[int, ...] = field(default=()) + + # DTensor distribution properties + placements: tuple[Placement, ...] = field( + default=() + ) # e.g., (Shard(0), Replicate()) + + # Mesh identification - used to look up the correct DeviceMesh from cache + mesh_dim_names: tuple[str, ...] = field(default=()) # e.g., ("tp",) or ("dp", "tp") + mesh_layout: _MeshLayout | None = field( + default=None + ) # _MeshLayout with shape/stride - uniquely identifies mesh within the same universe + + @staticmethod + def from_dtensor(dtensor: DTensor) -> _DTensorMeta: + """Create metadata from a DTensor. + + Args: + dtensor: The DTensor to extract metadata from. + + Returns: + Metadata capturing both local and global attributes. + """ + device_mesh = dtensor.device_mesh + + return _DTensorMeta( + # Local tensor attributes (for recv buffer allocation) + shape=dtensor._local_tensor.shape, + stride=dtensor._local_tensor.stride(), + dtype=dtensor.dtype, + requires_grad=dtensor.requires_grad, + # Global DTensor attributes (for reconstruction) + global_shape=dtensor.shape, + global_stride=dtensor.stride(), + # Distribution info + placements=dtensor._spec.placements, + mesh_dim_names=( + tuple(device_mesh.mesh_dim_names) if device_mesh.mesh_dim_names else () + ), + mesh_layout=device_mesh._layout, + ) + + @property + def mesh_cache_key(self) -> MeshCacheKey: + """Cache key ``(mesh_dim_names, mesh_layout)`` for mesh lookup.""" + return (self.mesh_dim_names, self.mesh_layout) + + def to_dtensor(self, device: torch.device | str, mesh: DeviceMesh) -> DTensor: + """Reconstruct a DTensor on ``device`` with placements. + + Args: + device: Target device for the local tensor. + mesh: The ``DeviceMesh`` to attach. + + Returns: + A DTensor on ``device``. + """ + local_tensor = _make_tensor_from_meta(self, device) + # Set requires_grad after from_local() so that the from_local + # operation itself is not recorded in the autograd graph. + dt = DTensor.from_local( + local_tensor, + device_mesh=mesh, + placements=self.placements, + shape=self.global_shape, + stride=self.global_stride, + run_check=False, + ) + if self.requires_grad and dt.is_floating_point(): + dt = dt.requires_grad_(True) + return cast(DTensor, dt) + + def get_diff(self, other: _TensorMeta) -> list[str]: + """Return field-by-field differences, including DTensor-specific fields. + + Args: + other: Metadata to compare against. + + Returns: + List of human-readable difference strings (empty if equal). + """ + if self == other: + return [] + + # Get base class differences (compares local shape/stride/dtype/requires_grad) + # NOTE: Use explicit class call instead of super() because + # @dataclass(slots=True) on both parent and child can break super(). + diffs = _TensorMeta.get_diff(self, other) + + # Add DTensor-specific comparisons if other is also _DTensorMeta + if isinstance(other, _DTensorMeta): + if self.global_shape != other.global_shape: + diffs.append( + f"global_shape mismatch: {self.global_shape} vs {other.global_shape}" + ) + if self.global_stride != other.global_stride: + diffs.append( + f"global_stride mismatch: {self.global_stride} vs {other.global_stride}" + ) + if self.placements != other.placements: + diffs.append( + f"placements mismatch: {self.placements} vs {other.placements}" + ) + if self.mesh_dim_names != other.mesh_dim_names: + diffs.append( + f"mesh_dim_names mismatch: {self.mesh_dim_names} vs {other.mesh_dim_names}" + ) + if self.mesh_layout != other.mesh_layout: + diffs.append( + f"mesh_layout mismatch: {self.mesh_layout} vs {other.mesh_layout}" + ) + else: + diffs.append("type: _DTensorMeta vs _TensorMeta") + + return diffs + + +# Type alias for union of tensor metadata types +TensorMeta: TypeAlias = _TensorMeta | _DTensorMeta + + +# Not frozen: fields are populated incrementally during forward and +# backward metadata inference or from user provided static metadata +@dataclass(slots=True) +class _StageMeta: + """Consolidated tensor metadata for a pipeline stage's forward and backward passes.""" + + inputs: tuple[TensorMeta, ...] | None = None + outputs: tuple[TensorMeta, ...] | None = None + input_grads: tuple[TensorMeta | None, ...] | None = None + output_grads: tuple[TensorMeta | None, ...] | None = None + + def has_any(self) -> bool: + """Check if any metadata field is populated.""" + return any( + v is not None + for v in [self.inputs, self.outputs, self.input_grads, self.output_grads] + ) + + def has_dtensors(self) -> bool: + """Check if any input/output metadata is DTensor type.""" + for metas in [self.inputs, self.outputs]: + if metas and any(isinstance(m, _DTensorMeta) for m in metas if m): + return True + return False + + def is_complete_for_forward(self) -> bool: + """Check if forward metadata is fully populated.""" + return self.inputs is not None and self.outputs is not None + + +@dataclass(frozen=True, slots=True) +class _StageForwardMeta: + """Forward metadata transmitted from stage *i* to stage *i+1* during inference.""" + + forward_metas: tuple[TensorMeta, ...] # Stage i's outputs → Stage i+1's inputs + + +@dataclass(frozen=True, slots=True) +class _StageBackwardMeta: + """Backward metadata transmitted from stage *i* to stage *i-1* during inference. + + Gradient placements may differ from forward activations + (e.g., ``Replicate`` → ``Partial``). + """ + + backward_metas: tuple[ + TensorMeta | None, ... + ] # Stage i's input_grads → Stage i-1's output_grads + + +def _make_tensor_from_meta( + meta: _TensorMeta, + device: torch.device | str, +) -> torch.Tensor: + """Create a tensor from metadata. + + Args: + meta: Metadata with shape, stride, and dtype. + device: Target device for the tensor. + + Returns: + Empty tensor preserving the exact memory layout. + """ + return torch.empty_strided( + size=meta.shape, + stride=meta.stride, + dtype=meta.dtype, + device=device, + ) + + +def _derive_grad_metas( + tensor_metas: tuple[TensorMeta, ...], +) -> tuple[TensorMeta | None, ...]: + """Derive gradient metadata from tensor metadata. + + Returns metadata with the same shape/stride/dtype but ``requires_grad=False``. + Entries where the source has ``requires_grad=False`` become ``None``. + """ + + def derive_one(m: TensorMeta) -> TensorMeta | None: + if not m.requires_grad: + return None + if isinstance(m, _DTensorMeta): + return None + return _TensorMeta( + shape=m.shape, + stride=m.stride, + dtype=m.dtype, + requires_grad=False, + ) + + return tuple(derive_one(m) for m in tensor_metas) + + +class _MeshCache: + """Cache for :class:`DeviceMesh` objects keyed by ``(mesh_dim_names, mesh_layout)``. + + Assumes all pipeline stages share the same rank tensor (true for + TorchTitan-style frameworks where meshes derive from a common world). + """ + + def __init__(self, get_mesh_cb: GetMeshCallback | None = None) -> None: + self._cache: dict[MeshCacheKey, DeviceMesh] = {} + self._get_mesh_cb = get_mesh_cb + + def get_mesh(self, key: MeshCacheKey) -> DeviceMesh: + """Return a cached mesh, or create one via the callback. + + Args: + key: Cache key ``(mesh_dim_names, mesh_layout)``. + + Returns: + The ``DeviceMesh``. + + Raises: + PipeliningMetadataError: If not cached and no callback provided. + """ + if key in self._cache: + return self._cache[key] + + mesh_dim_names, mesh_layout = key + + if self._get_mesh_cb is None: + raise PipeliningMetadataError( + f"Mesh not found in cache for mesh_dim_names={mesh_dim_names}, " + f"mesh_layout={mesh_layout}, and no get_mesh callback provided. " + f"Provide a get_mesh callback or use DTensors in static mode." + ) + + mesh = self._get_mesh_cb(mesh_dim_names, mesh_layout) + if mesh is None: + raise PipeliningMetadataError( + f"Mesh lookup failed: callback returned None for " + f"mesh_dim_names={mesh_dim_names}, mesh_layout={mesh_layout}. " + f"Ensure all stages use meshes from the same universe." + ) + self._cache[key] = mesh + return mesh + + def put(self, key: MeshCacheKey, mesh: DeviceMesh) -> None: + """Add a mesh to the cache.""" + self._cache[key] = mesh + + def update_from_tensors(self, tensors: tuple[torch.Tensor | None, ...]) -> None: + """Extract and cache meshes from any :class:`DTensor` instances in *tensors*.""" + for tensor in tensors: + if isinstance(tensor, DTensor): + mesh = tensor.device_mesh + dim_names = tuple(mesh.mesh_dim_names) if mesh.mesh_dim_names else () + mesh_layout = mesh._layout + key = (dim_names, mesh_layout) + if key not in self._cache: + self._cache[key] = mesh + + def __contains__(self, key: MeshCacheKey) -> bool: + return key in self._cache + + def __len__(self) -> int: + return len(self._cache) + + +# ============================================================================ +# Inference mode enum +# ============================================================================ + + +class InferenceMode(Enum): + """Pipeline-level metadata inference mode, determined collectively across all PP ranks. + + The mode is set by the schedule (not individual stages) because + ``has_backward`` is only known at schedule creation time and all + stages must agree to avoid P2P hangs. + + .. attribute:: STATIC + + All stages have sufficient metadata; runtime inference is skipped. + + .. attribute:: DYNAMIC + + At least one stage requires runtime metadata inference. + """ + + STATIC = "static" + DYNAMIC = "dynamic" + + @classmethod + def needs_dynamic(cls, meta: _StageMeta, stage_has_backward: bool) -> bool: + """Determine whether dynamic metadata inference is needed for a stage. + + Args: + meta: Stage metadata from user-provided args. + stage_has_backward: Whether a backward pass will be performed. + + Returns: + ``True`` if dynamic inference is needed. + """ + # Case 1: Forward metadata incomplete → needs DYNAMIC + if not meta.is_complete_for_forward(): + return True + + # Case 2: No DTensors → STATIC is fine (bwd metadata derivable from fwd metadata) + if not meta.has_dtensors(): + return False + + # Case 3: No backward needed → STATIC is fine (don't need grad metadata) + if not stage_has_backward: + return False + + # Case 4: DTensors with backward but missing ANY grad metadata → needs DYNAMIC + # Both input_grads AND output_grads are required for static mode with DTensors + if meta.input_grads is None or meta.output_grads is None: + return True + + # Case 5: DTensors with complete grads → STATIC is fine + return False + + +# ============================================================================ +# Utility functions +# ============================================================================ + + +def flatten_args(args, *, detach: bool = False): + """Flatten ``args`` into a list, optionally detaching tensors. + + Args: + args: Nested arguments to flatten. + detach: If ``True``, detach tensors while preserving ``requires_grad``. + + Returns: + ``(new_args, flat_detached_args)`` when ``detach=True``; + ``flat_args`` list otherwise. + """ + flat_args, treespec = tree_flatten(args) + + if detach: + flat_detached = [ + a.detach().requires_grad_(a.requires_grad) + if isinstance(a, torch.Tensor) + else a + for a in flat_args + ] + new_args = tree_unflatten(flat_detached, treespec) + return new_args, flat_detached + + return flat_args + + +# Backward compatibility alias +def flatten_args_detach(args): + """Flatten and detach. Deprecated: use ``flatten_args(args, detach=True)``.""" + return flatten_args(args, detach=True) + + +def generate_stage_to_rank_mapping( + pp_size: int, num_stages: int, style: str = "loop" +) -> dict[int, int]: + """ + Compute the stage id to rank mapping for either a looped or V-style schedule. + + Most commonly num_stages == pp_size * 2, but this function can be used to + compute the mapping for any number of stages per rank. + """ + mapping = {} + if style == "loop": + for stage_index in range(num_stages): + mapping[stage_index] = stage_index % pp_size + elif style == "v": + if num_stages % pp_size != 0: + raise ValueError( + f"num_stages {num_stages} must be evenly divisible by pp_size {pp_size} for V schedules" + ) + + rank_index = 0 + for stage_index in range(num_stages): + mapping[stage_index] = rank_index + # don't change rank if we are on the border (to keep v shape) + if (stage_index + 1) % pp_size == 0: + continue + if (stage_index // pp_size) % 2 == 0: + rank_index += 1 + else: + rank_index -= 1 + else: + raise ValueError(f"Style {style} is not supported.") + return mapping + + +def generate_rank_to_stage_mapping( + pp_size: int, num_stages: int, style: str = "loop" +) -> dict[int, list[int]]: + """ + Compute the rank to stage id mapping for either a looped or V-style schedule. + + This function inverts the stage_to_rank_mapping to get which stages are assigned to each rank. + + Returns a dictionary mapping rank -> list of stage indices assigned to that rank. + """ + stage_to_rank = generate_stage_to_rank_mapping(pp_size, num_stages, style) + + # Invert the mapping: rank -> list of stages + rank_to_stages: dict[int, list[int]] = {} + for stage_id, rank in stage_to_rank.items(): + if rank not in rank_to_stages: + rank_to_stages[rank] = [] + rank_to_stages[rank].append(stage_id) + + # Sort the stage lists for each rank to ensure consistent ordering + for stages in rank_to_stages.values(): + stages.sort() + + return rank_to_stages + + +@dataclass(slots=True) +class PipeInfo: + """ + Captures information for a pipeline (`Pipe` object). + """ + + graph: fx.Graph + num_stages: int + has_loss_and_backward: bool + + +# ============================================================================ +# Metadata extraction helpers +# ============================================================================ + + +def extract_tensor_meta(tensor: torch.Tensor) -> TensorMeta: + """Extract metadata from a tensor. + + Handles both plain Tensor and DTensor correctly: DTensors are + dispatched to ``_DTensorMeta.from_dtensor`` which captures local + shard attributes plus global shape/placement info, while plain + tensors use ``_TensorMeta.from_tensor``. + + Args: + tensor: A plain tensor or DTensor. + + Returns: + ``_TensorMeta`` for plain tensors, ``_DTensorMeta`` for DTensors. + """ + if isinstance(tensor, DTensor): + return _DTensorMeta.from_dtensor(tensor) + else: + return _TensorMeta.from_tensor(tensor) + + +@overload +def extract_tensor_metas( + tensors: tuple[torch.Tensor, ...] | None, + *, + allow_none: Literal[False] = ..., +) -> tuple[TensorMeta, ...] | None: ... + + +@overload +def extract_tensor_metas( + tensors: tuple[torch.Tensor | None, ...] | None, + *, + allow_none: Literal[True], +) -> tuple[TensorMeta | None, ...] | None: ... + + +def extract_tensor_metas( + tensors: tuple[torch.Tensor | None, ...] | tuple[torch.Tensor, ...] | None, + *, + allow_none: bool = False, +) -> tuple[TensorMeta | None, ...] | None: + """Extract metadata from a tuple of tensors. + + Args: + tensors: Tuple of tensors (may include ``None`` when ``allow_none=True``). + allow_none: If ``True``, preserve ``None`` elements (for gradients). + + Returns: + Tuple of ``TensorMeta``, or ``None`` if ``tensors`` is ``None``. + + Raises: + PipeliningMetadataError: If ``None`` found and ``allow_none=False``. + """ + if tensors is None: + return None + + metas_with_none: list[TensorMeta | None] = [] + has_none = False + for t in tensors: + if isinstance(t, torch.Tensor): + metas_with_none.append(extract_tensor_meta(t)) + else: + has_none = True + metas_with_none.append(None) + if not allow_none and has_none: + raise PipeliningMetadataError( + "None values are not allowed in tensor metadata tuples. " + "Use allow_none=True for optional values." + ) + return tuple(metas_with_none) + + +def to_local_if_dtensor(tensor: torch.Tensor, detach: bool = False) -> torch.Tensor: + """Convert a DTensor to its local shard, or return a plain tensor as-is. + + When ``detach=True``, the tensor is detached before conversion — + this applies to both DTensors and plain tensors. + + Args: + tensor: A tensor that may be a DTensor. + detach: If ``True``, detach before ``to_local()`` to avoid + redistribution during backward. + + Returns: + The local tensor component. + """ + maybe_detached_tensor = tensor.detach() if detach else tensor + if isinstance(maybe_detached_tensor, DTensor): + return maybe_detached_tensor.to_local() + return maybe_detached_tensor + + +@overload +def validate_and_normalize_to_tuple( + args: torch.Tensor | tuple[torch.Tensor, ...] | list[torch.Tensor] | None, + allow_none: Literal[False] = ..., +) -> tuple[torch.Tensor, ...] | None: ... + + +@overload +def validate_and_normalize_to_tuple( + args: torch.Tensor + | tuple[torch.Tensor | None, ...] + | list[torch.Tensor | None] + | None, + allow_none: Literal[True] = ..., +) -> tuple[torch.Tensor | None, ...] | None: ... + + +def validate_and_normalize_to_tuple( + args: torch.Tensor + | tuple[torch.Tensor, ...] + | tuple[torch.Tensor | None, ...] + | list[torch.Tensor] + | list[torch.Tensor | None] + | None, + allow_none: bool = False, +) -> tuple[torch.Tensor | None, ...] | tuple[torch.Tensor, ...] | None: + """Normalize ``args`` to a tuple and validate that all elements are tensors. + + Args: + args: A single tensor, tuple/list of tensors, or ``None``. + allow_none: If ``True``, permit ``None`` elements (for gradients). + + Returns: + Tuple of tensors, or ``None`` if ``args`` is ``None``. + + Raises: + PipeliningMetadataError: On non-tensor values + (or ``None`` when ``allow_none=False``). + """ + if args is None: + return None + elif isinstance(args, torch.Tensor): + return (args,) + elif isinstance(args, (tuple, list)): + for i, arg in enumerate(args): + if arg is None: + if not allow_none: + raise PipeliningMetadataError( + f"Stage arg[{i}] is None. " + f"Stage args must be tensors. Use kwargs for optional values." + ) + continue + if not isinstance(arg, torch.Tensor): + raise PipeliningMetadataError( + f"Stage arg[{i}] has type {type(arg).__name__}. " + f"All stage args must be tensors. Use kwargs for non-tensor inputs." + ) + # Normalize list to tuple + return tuple(args) if isinstance(args, list) else args + else: + raise PipeliningMetadataError( + f"Stage args must be a tensor, tuple, or list of tensors, got {type(args).__name__}." + ) + + +# ============================================================================ +# Validation functions +# ============================================================================ + + +def validate_metadata( + desc: str, + expected: TensorMeta, + actual: torch.Tensor | TensorMeta, + *, + raise_on_mismatch: bool = False, + warn_on_mismatch: bool = False, +) -> list[str]: + """ + Compare expected metadata against actual tensor or metadata. + + This is the unified validation/comparison function that uses get_diff() from + metadata classes. Works with both plain tensors and DTensors. + + For plain tensors: compares shape/stride/dtype/requires_grad. + For DTensors: compares all properties including global shape and placements. + + Args: + desc: Description for error/warning messages. + expected: Expected tensor metadata (_TensorMeta or _DTensorMeta). + actual: Actual tensor or metadata to compare against. + raise_on_mismatch: If True, raise PipeliningMetadataError on mismatch. + warn_on_mismatch: If True, issue a warning on mismatch. + + Returns: + List of differences (empty if metadata matches). + + Raises: + PipeliningMetadataError: If raise_on_mismatch=True and differences exist. + """ + # Extract metadata if actual is a tensor + if isinstance(actual, torch.Tensor): + actual_meta = extract_tensor_meta(actual) + else: + actual_meta = actual + + # Type check: ensure both are same type for meaningful comparison + if type(expected) is not type(actual_meta): + type_diff = [ + f"type: expected {type(expected).__name__}, got {type(actual_meta).__name__}" + ] + if raise_on_mismatch: + raise PipeliningMetadataError(f"{desc}: {type_diff[0]}") + if warn_on_mismatch: + warnings.warn( + f"{desc}: Metadata type mismatch. {type_diff[0]}. " + f"Using dynamically inferred metadata instead.", + UserWarning, + stacklevel=2, + ) + return type_diff + + # Use get_diff() from the metadata class + diffs = expected.get_diff(actual_meta) + + if diffs: + if raise_on_mismatch: + raise PipeliningMetadataError(f"{desc}: {'; '.join(diffs)}") + if warn_on_mismatch: + warnings.warn( + f"{desc}: Metadata mismatch. {'; '.join(diffs)}. " + f"Using dynamically inferred metadata instead.", + UserWarning, + stacklevel=2, + ) + + return diffs + + +def validate_tensors_metadata( + desc: str, + expected: tuple[TensorMeta | None, ...], + actual: tuple[torch.Tensor | TensorMeta | None, ...], + *, + raise_on_mismatch: bool = True, + warn_on_mismatch: bool = False, +) -> list[str]: + """Validate metadata for a tuple of tensors element-wise. + + Args: + desc: Description prefix for error/warning messages. + expected: Tuple of expected metadata (may include ``None`` for grads). + actual: Tuple of actual tensors or metadata to compare against. + raise_on_mismatch: If ``True``, raise on the first mismatch. + warn_on_mismatch: If ``True``, issue warnings for mismatches. + + Returns: + Aggregated list of difference strings. + + Raises: + PipeliningMetadataError: If lengths differ or on mismatch. + """ + if len(expected) != len(actual): + msg = f"{desc}: expected {len(expected)} tensors, got {len(actual)}" + if raise_on_mismatch: + raise PipeliningMetadataError(msg) + if warn_on_mismatch: + warnings.warn(msg, UserWarning, stacklevel=2) + return [msg] + + all_diffs: list[str] = [] + for i, (exp, act) in enumerate(zip(expected, actual, strict=True)): + if exp is None and act is None: + continue + if exp is None or act is None: + msg = ( + f"{desc}[{i}]: expected {'None' if exp is None else 'metadata'}, " + f"got {'None' if act is None else 'metadata'}" + ) + if raise_on_mismatch: + raise PipeliningMetadataError(msg) + if warn_on_mismatch: + warnings.warn(msg, UserWarning, stacklevel=2) + all_diffs.append(msg) + continue + diffs = validate_metadata( + f"{desc}[{i}]", + exp, + act, + raise_on_mismatch=raise_on_mismatch, + warn_on_mismatch=warn_on_mismatch, + ) + all_diffs.extend(diffs) + return all_diffs + + +def validate_static_arg_grad_correspondence( + stage_index: int, + args: tuple[torch.Tensor, ...], + grads: tuple[torch.Tensor | None, ...], + is_input: bool, +) -> None: + """ + Validate the args↔grads contract for static mode. + + Enforces four rules for each (arg, grad) pair: + 1. len(args) must equal len(grads). + 2. If arg.requires_grad is False, grad must be None. + 3. If arg.requires_grad is True and grad is None, emit a warning + (this is legal at pipeline boundaries but may indicate a bug). + 4. If arg is a DTensor with requires_grad=True and grad is not None, + grad must also be a DTensor. + + Args: + stage_index: The stage index for error messages. + args: Tuple of forward tensors. + grads: Tuple of gradient tensors (can include None). + is_input: True for input_args/input_grads, False for output_args/output_grads. + + Raises: + PipeliningMetadataError: If any hard rule (1, 2, or 4) is violated. + """ + kind = "input" if is_input else "output" + args_name = f"{kind}_args" + grads_name = f"{kind}_grads" + + # Rule 1: lengths must match + if len(args) != len(grads): + raise PipeliningMetadataError( + f"Stage {stage_index}: {grads_name} length ({len(grads)}) does not match " + f"{args_name} length ({len(args)}). Each forward tensor must have a " + f"corresponding gradient entry (use None for tensors that don't require grad)." + ) + + for i, (arg, grad) in enumerate(zip(args, grads, strict=True)): + # Rule 2: no grad for a non-differentiable arg + if not arg.requires_grad and grad is not None: + raise PipeliningMetadataError( + f"Stage {stage_index}: {args_name}[{i}] has requires_grad=False, " + f"but {grads_name}[{i}] is not None ({type(grad).__name__}). " + f"Non-differentiable tensors must have None as their gradient entry." + ) + + # Rule 3: missing grad for a differentiable arg (warn, don't raise) + if arg.requires_grad and grad is None: + warnings.warn( + f"Stage {stage_index}: {args_name}[{i}] has requires_grad=True, " + f"but {grads_name}[{i}] is None. This is legal at pipeline boundaries " + f"but may indicate a missing gradient.", + UserWarning, + stacklevel=2, + ) + + # Rule 4: DTensor arg must have DTensor grad + if ( + isinstance(arg, DTensor) + and arg.requires_grad + and grad is not None + and not isinstance(grad, DTensor) + ): + raise PipeliningMetadataError( + f"Stage {stage_index}: {args_name}[{i}] is a DTensor with requires_grad=True, " + f"but {grads_name}[{i}] is {type(grad).__name__}, expected DTensor or None. " + f"DTensor gradients may have different placements than forward tensors." + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/microbatch.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/microbatch.py new file mode 100644 index 0000000000000000000000000000000000000000..e4330e4ca0a3acbfcdf00096fdb42c2bb8da3d27 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/microbatch.py @@ -0,0 +1,663 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import logging +import operator +from collections.abc import Sequence +from typing import Any, cast + +import torch +from torch.distributed.tensor import DTensor +from torch.fx.node import map_aggregate +from torch.nn.attention.flex_attention import BlockMask +from torch.utils._pytree import tree_flatten, tree_map, tree_unflatten + + +__all__ = [ + "TensorChunkSpec", + "split_args_kwargs_into_chunks", + "merge_chunks", +] + +logger = logging.getLogger(__name__) + +""" +_debug_mask_minibatches specifies to send masked versions of the mini-batch +through instead of micro-batch slices--this can be used for more stable +numerical testing (see [A Note About Correctness Testing]) +""" +_debug_mask_minibatches = False + + +class _CustomReducer: + """ + Custom reducer class that can be used to specify a custom operation that + reduces losses of multiple microbatches into one value. + + Example: + >>> # xdoctest: +SKIP + >>> sum_reducer = _CustomReducer( + >>> torch.tensor(0.0), + >>> lambda a, b: a + b + >>> ) + """ + + def __init__(self, init_value, reduce_fn): + self.init_value = init_value + self.reduce_fn = reduce_fn + + +class _LossReducer(_CustomReducer): + pass + + +sum_reducer = _LossReducer(torch.tensor(0.0), operator.add) + +# Default chunking dimension is 0. This is used for the case where the user did +# not specify a chunking dimension. +DEFAULT_CHUNK_DIM = 0 + + +class TensorChunkSpec: + """ + Class used to specify chunking of inputs + """ + + def __init__(self, split_dim): + self.split_dim = split_dim + + split_dim: int + + def __repr__(self): + return ( + f"{self.__class__.__module__}.{self.__class__.__name__}({self.split_dim})" + ) + + def __str__(self): + return f"TensorChunkSpec({self.split_dim})" + + @staticmethod + def from_tuple( + chunk_dims: tuple[int, ...], + ): + """ + A helper for creating a tuple of `TensorChunkSpec` from a tuple of chunk + dimensions (int's). + Example: + >>> # xdoctest: +SKIP + >>> # There are three positional arguments to the model, and + >>> # we are chunking them along dimension 0, 0 and 1, respectively + >>> args_chunk_spec = TensorChunkSpec.from_tuple((0, 0, 1)) + """ + args_chunk_spec = map_aggregate( + chunk_dims, + lambda dim: TensorChunkSpec(dim), # type: ignore[arg-type,return-value] + ) + return args_chunk_spec + + @staticmethod + def from_dict( + chunk_dims: dict[str, int], + ): + """ + A helper for creating a dictionary of `TensorChunkSpec` from a + dictionary of chunk dimensions (int's). + Example: + >>> # xdoctest: +SKIP + >>> # Chunk dimension 0 for the "id" argument, 1 for the "mask" argument + >>> kwargs_chunk_spec = TensorChunkSpec.from_dict({"id": 0, "mask": 1}) + """ + kwargs_chunk_spec = map_aggregate( + chunk_dims, + lambda dim: TensorChunkSpec(dim), # type: ignore[arg-type,return-value] + ) + return kwargs_chunk_spec + + +# Class used to specify replication of inputs +class _Replicate: + pass + + +def _split_block_mask( + block_mask: BlockMask, + num_chunks: int, +) -> list[BlockMask]: + """Given a block mask, split the block mask along the batch dimension (dim0). + + Args: + block_mask: Block mask to split + num_chunks: Number of chunks to split the block mask into + + Returns: + chunk_block_masks: List of chunked block masks + """ + + # BlockMask will broadcast if B is 1. + if block_mask.kv_num_blocks.size(0) == 1: + return [block_mask] * num_chunks + + if not block_mask.kv_num_blocks.size(0) >= num_chunks: + raise AssertionError( + "Block mask has fewer batch size than the number of chunks. " + ) + + batch_dim = 0 + kv_num_blocks_chunks = torch.tensor_split( + block_mask.kv_num_blocks, num_chunks, batch_dim + ) + kv_indices_chunks = torch.tensor_split(block_mask.kv_indices, num_chunks, batch_dim) + full_kv_num_blocks_chunks = ( + torch.tensor_split(block_mask.full_kv_num_blocks, num_chunks, batch_dim) + if block_mask.full_kv_num_blocks is not None + else [None] * num_chunks + ) + full_kv_indices_chunks = ( + torch.tensor_split(block_mask.full_kv_indices, num_chunks, batch_dim) + if block_mask.full_kv_indices is not None + else [None] * num_chunks + ) + + chunk_block_masks = [] + batch_offset = 0 + for chunk_idx in range(num_chunks): + + def create_mask_mod(idx): + def batch_offset_mask_mod(b, h, q_idx, kv_idx): + b_offset = torch.full_like(b, idx) + return block_mask.mask_mod(b + b_offset, h, q_idx, kv_idx) + + return batch_offset_mask_mod + + chunk_block_masks.append( + BlockMask.from_kv_blocks( + kv_num_blocks=kv_num_blocks_chunks[chunk_idx], + kv_indices=kv_indices_chunks[chunk_idx], + full_kv_num_blocks=full_kv_num_blocks_chunks[chunk_idx], + full_kv_indices=full_kv_indices_chunks[chunk_idx], + BLOCK_SIZE=block_mask.BLOCK_SIZE, + mask_mod=create_mask_mod(batch_offset), + seq_lengths=block_mask.seq_lengths, + ) + ) + batch_offset += kv_num_blocks_chunks[chunk_idx].size(0) + return chunk_block_masks + + +def _split_tensor( + tensor: torch.Tensor, + spec: TensorChunkSpec, + num_chunks: int, +) -> Sequence[torch.Tensor]: + """Given a tensor, and a chunking spec, split the tensor. + Args: + + tensor: Tensor to split + spec: Chunking spec + num_chunks: Number of chunks to split the tensor into + + Returns: + chunk_tensors: List of chunked tensors + """ + + if not tensor.size(spec.split_dim) >= num_chunks: + raise AssertionError( + f"Tensor size {tensor.size(spec.split_dim)} is smaller than num_chunks" + ) + + _is_dtensor = isinstance(tensor, DTensor) + + if _is_dtensor: + # Split locally and wrap each chunk as a DTensor with the correct + # global shape. We avoid DTensor dispatch (which would all-gather + # Shard(split_dim) to Replicate) and avoid local_map (whose + # from_local call assumes even sharding and computes wrong global + # sizes when local shards differ across ranks). + placements = tensor.placements + mesh = tensor.device_mesh + local_tensor = tensor.to_local() + local_chunks = torch.tensor_split(local_tensor, num_chunks, spec.split_dim) + # Compute the correct global chunk sizes along split_dim. + global_split_size = tensor.shape[spec.split_dim] + quotient, remainder = divmod(global_split_size, num_chunks) + global_stride = tensor.stride() + chunk_tensors_list: list[torch.Tensor] = [] + for i, local_chunk in enumerate(local_chunks): + chunk_shape = list(tensor.shape) + chunk_shape[spec.split_dim] = quotient + (1 if i < remainder else 0) + chunk_tensors_list.append( + DTensor.from_local( + local_chunk, + mesh, + placements, + shape=torch.Size(chunk_shape), + stride=global_stride, + run_check=False, + ) + ) + chunk_tensors: Sequence[torch.Tensor] = chunk_tensors_list # type: ignore[assignment] + else: + chunk_tensors = torch.tensor_split(tensor, num_chunks, spec.split_dim) + + # tensor_split on a leaf tensor produces non-leaf views that won't + # accumulate .grad during torch.autograd.backward(). Call retain_grad() + # on those views so that stage_backward() can read .grad from them. + if tensor.requires_grad and tensor.is_leaf: + for chunk in chunk_tensors: + chunk.retain_grad() + + if not _debug_mask_minibatches: + return chunk_tensors + + def _expand_chunks( + orig: torch.Tensor, *chunks: torch.Tensor + ) -> tuple[torch.Tensor, ...]: + expanded = [] + idx = 0 + for chunk in chunks: + new_val = torch.zeros_like(orig) + upper = idx + chunk.size(spec.split_dim) + slices: list[slice] = [slice(None)] * new_val.ndim + slices[spec.split_dim] = slice(idx, upper) + new_val[slices] = chunk + expanded.append(new_val) + idx += chunk.size(spec.split_dim) + return tuple(expanded) + + if _is_dtensor: + placements = tensor.placements + mesh = tensor.device_mesh + global_shape = tensor.shape + global_stride = tensor.stride() + local_expanded = _expand_chunks( + tensor.to_local(), + *(cast(DTensor, c).to_local() for c in chunk_tensors), + ) + return [ + DTensor.from_local( + t, + mesh, + placements, + shape=global_shape, + stride=global_stride, + run_check=False, + ) + for t in local_expanded + ] + else: + return list(_expand_chunks(tensor, *chunk_tensors)) + + +def _shard_dict_of_args( + args_dict, + args_chunk_spec, + num_chunks, +): + """ + Given a dictionary of args, and a dictionary of chunking specs, shard the + args according to the chunking specs. + + Args: + args_dict: Dictionary of args + args_chunk_spec: Dictionary of chunking specs + num_chunks: Number of chunks to shard the args into + + Returns: + args_split: List of sharded args + """ + + if not args_dict: + return [{} for _ in range(num_chunks)] + + if not len(args_dict) == len(args_chunk_spec): + raise AssertionError( + f"args_dict.keys() = {list(args_dict.keys())} " + f"args_chunk_spec.keys() = {list(args_chunk_spec.keys())}" + ) + if args_chunk_spec is None: + raise AssertionError("args_chunk_spec should have been set by caller") + + values, tree_spec = tree_flatten( + args_dict, is_leaf=lambda x: isinstance(x, BlockMask) + ) + chunk_specs, _ = tree_flatten( + args_chunk_spec, is_leaf=lambda x: isinstance(x, BlockMask) + ) + + # First check and find the actual number of chunks + split_sizes = [] + for v, spec in zip(values, chunk_specs, strict=True): + # The original logic is "spec is _Replicate". This doesn't seem to be + # correct. But we keep it for backward compatibility. + if spec is _Replicate or isinstance(spec, _Replicate): + split_sizes.append(num_chunks) + elif isinstance(v, torch.Tensor): + if not isinstance(spec, TensorChunkSpec): + raise AssertionError(f"Expected TensorChunkSpec, got {type(spec)}") + split_sizes.append(v.size(spec.split_dim)) + elif isinstance(v, BlockMask): + if not isinstance(spec, TensorChunkSpec): + raise AssertionError(f"Expected TensorChunkSpec, got {type(spec)}") + if not spec.split_dim == 0: + raise AssertionError("BlockMask only supports split_dim=0") + # BlockMask will broadcast if B is 1. + if v.kv_num_blocks.size(0) == 1: + split_sizes.append(num_chunks) + else: + split_sizes.append(v.kv_num_blocks.size(0)) + else: + raise ValueError( + f"Unsupported chunk spec: {spec} and value: {v} combination." + ) + result_num_chunks = min(*split_sizes, num_chunks) + + flat_split_results: list[Any] = [[] for _ in range(result_num_chunks)] + for v, spec in zip(values, chunk_specs, strict=True): + v_splits: Sequence[Any] = [] + if spec is _Replicate or isinstance(spec, _Replicate): + v_splits = [v] * result_num_chunks + elif isinstance(v, torch.Tensor): + v_splits = _split_tensor(v, spec, result_num_chunks) + elif isinstance(v, BlockMask): + v_splits = _split_block_mask(v, result_num_chunks) + else: + raise ValueError( + f"Unsupported chunk spec: {spec} and value: {v} combination." + ) + + for _flat_split_result, _v_split in zip( + flat_split_results, v_splits, strict=True + ): + _flat_split_result.append(_v_split) + + return [ + tree_unflatten(_flat_split_result, tree_spec) + for _flat_split_result in flat_split_results + ] + + +def split_args_kwargs_into_chunks( + args: tuple[Any, ...], + kwargs: dict[str, Any] | None, + chunks: int, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, +) -> tuple[list[tuple], list[dict]]: + """ + Given a sequence of args and kwargs, split them into a number of chunks + according to their respective chunking specs. + + Args: + args: Tuple of args + kwargs: Dict of kwargs + chunks: Number of chunks to split the args and kwargs into + args_chunk_spec: chunking specs for args, in same shape as args + kwargs_chunk_spec: chunking specs for kwargs, in same shape as kwargs + + Returns: + args_split: List of sharded args + kwargs_split: List of sharded kwargs + """ + # Given `args` and `kwargs`, we want to yield a set of `chunks` args and kwargs such that + # the constituent Tensor values have been sharded/replicated according to the `args_chunk_spec` + # and `kwargs_chunk_spec` specifications. The steps are as follows: + # + # 1. Use pytree.tree_flatten to flatten each arg and its spec into a 1d array of values. + # To use a running example: suppose our inputs look like + # + # args = ([A, [B, C]], D) args_spec = ([None, [None, TensorChunkSpec]], None) + # (kwargs not shown but it's a similar process) + # + # Then for this step we would end up with + # + # args = ([A, B, C], D) args_spec = ([None, None, TensorChunkSpec], None) + # + # 2. Shard or replicate the arguments subject to the policy in the spec. Suppose chunks = 2 + # + # args = ([[A, A], [B, B], [C_1, C_2]], [D, D]) + # + # 3. Rotate the nesting order such that chunks are the outer dimension + # + # args_chunks = [ + # ([A, B, C_1], D), + # ([A, B, C_2], D), + # ] + # + # 4. Unflatten each chunk according to the spec + # + # args_chunks = [ + # ([A, [B, C_1]], D), + # ([A, [B, C_2]], D), + # ] + + # TODO: _debug_mask_minibatches + # Handle the case where kwargs is None + if kwargs is None: + kwargs = {} + + # If user did not provide args_chunk_spec or kwargs_chunk_spec, we extend + # their format and use default chunking along dim 0 + def default_spec(v): + if isinstance(v, torch.Tensor | BlockMask): + return TensorChunkSpec(DEFAULT_CHUNK_DIM) + else: + return _Replicate() + + if args_chunk_spec is None: + args_chunk_spec = tree_map( + default_spec, args, is_leaf=lambda v: isinstance(v, BlockMask) + ) + + if kwargs_chunk_spec is None: + kwargs_chunk_spec = tree_map( + default_spec, kwargs, is_leaf=lambda v: isinstance(v, BlockMask) + ) + + args_split_dict = _shard_dict_of_args( + dict(enumerate(args)), + dict(enumerate(args_chunk_spec)), + chunks, + ) + real_num_chunks = len(args_split_dict) + + kwargs_split = _shard_dict_of_args( + kwargs, + kwargs_chunk_spec, + real_num_chunks, + ) + + if len(kwargs_split) < real_num_chunks: + # In case kwargs are sharded into less chunks + # e.g. when `args` has no tensor, just values + real_num_chunks = len(kwargs_split) + # Re-shard args + args_split_dict = _shard_dict_of_args( + dict(enumerate(args)), + dict(enumerate(args_chunk_spec)), + real_num_chunks, + ) + + if len(args_split_dict) != len(kwargs_split): + raise RuntimeError( + "args and kwargs are split into different number of chunks: " + f"{len(args_split_dict)}, {len(kwargs_split)}" + ) + + args_split = [ + tuple(chunk_args[i] for i in range(len(chunk_args))) + for chunk_args in args_split_dict + ] + + return args_split, kwargs_split + + +def merge_chunks( + chunks: list[Any], + chunk_spec, +): + """ + Given a list of chunks, merge them into a single value according to + the chunk spec. + + Args: + chunks: list of chunks + chunk_spec: Chunking spec for the chunks + + Returns: + value: Merged value + """ + # This is essentially the inverse of `split_args_kwargs_into_chunks`, so the + # steps are similar to the steps in that function but in reverse. Given the + # input values: + # + # chunks = [ + # ([A, [B, C_1]], D), + # ([A, [B, C_2]], D), + # ] + # args_spec = ([None, [None, TensorChunkSpec]], None) + # + # 1. Flatten the chunks according to the chunk_spec + # + # chunks_flat = [ + # ([A, B, C_1], D), + # ([A, B, C_2], D), + # ] + # + # 2. Rotate the nesting order such that chunks are the inner dimension + # + # value_inner = ([A, B, [C_1, C_2]], D) + # + # 3. Concatenate sharded arguments + # + # value_combined = ([A, B, C], D) + # + # 4. Unflatten the combined args given the spec + # + # value = ([A, [B, C]], D) + + # Preliminary: flatten the chunk spec + if chunk_spec is not None: + spec_flattened, flatten_spec = tree_flatten(chunk_spec) + else: + # If chunk_spec is not provided, we will merge chunks along the default dimension (0), for all output fields + # We obtain the output structure by flattening chunk 0 and generate the chunk_spec + chunk0_flat, flatten_spec = tree_flatten(chunks[0]) + spec_flattened = [TensorChunkSpec(DEFAULT_CHUNK_DIM)] * len(chunk0_flat) + + # Stage 1: flatten chunks + # chunks_flattened : [num chunks, num args] + chunks_flattened = [] + + for chunk in chunks: + chunk_flattened, _ = tree_flatten(chunk) + if len(chunk_flattened) != len(spec_flattened): + raise ValueError(f"Chunk {chunk} did not match chunk spec {chunk_spec}") + + chunks_flattened.append(chunk_flattened) + + # Stage 2 and 3: Rotate nesting order s.t. chunks are inner dimension and + # concatenate sharded operands + # args_flattened : [num args] + args_flattened: list[Any] = [] + for arg_idx, arg in enumerate(spec_flattened): + if isinstance(arg, TensorChunkSpec): + partial_values = [ + chunks_flattened[chunk_idx][arg_idx] + for chunk_idx in range(len(chunks_flattened)) + ] + + if _debug_mask_minibatches: + # Infer size of individual chunks by running `tensor_split` again + overall_shape = partial_values[0].shape + for val in partial_values[1:]: + if not val.shape == overall_shape: + raise AssertionError( + f"Expected shape {overall_shape}, got {val.shape}" + ) + meta_chunks = torch.tensor_split( + torch.empty(*overall_shape, device="meta"), + sections=len(partial_values), + dim=arg.split_dim, + ) + + values_to_cat = [] + chunk_start_idx = 0 + if not len(partial_values) == len(meta_chunks): + raise AssertionError( + f"Expected len(partial_values) == len(meta_chunks), got {len(partial_values)} != {len(meta_chunks)}" + ) + + for partial_value, meta_chunk in zip( + partial_values, meta_chunks, strict=True + ): + chunk_end_idx = chunk_start_idx + meta_chunk.size(arg.split_dim) + + slice_indices = [slice(None, None, None)] * partial_value.ndim + slice_indices[arg.split_dim] = slice(chunk_start_idx, chunk_end_idx) + sliced = partial_value[slice_indices] + values_to_cat.append(sliced) + + chunk_start_idx = chunk_end_idx + + else: + values_to_cat = partial_values + + # Validate DTensor consistency: either all values are DTensors + # or none are. A mix indicates a bug in the pipeline stage. + dtensor_flags = [isinstance(v, DTensor) for v in values_to_cat] + if any(dtensor_flags): + if not all(dtensor_flags): + raise AssertionError( + "merge_chunks: expected all values to be DTensors or " + "none to be DTensors, got a mix" + ) + # All DTensors must have matching placements. + placements = values_to_cat[0].placements + for i, v in enumerate(values_to_cat[1:], 1): + if v.placements != placements: + raise AssertionError( + f"merge_chunks: placement mismatch at chunk {i}: " + f"expected {placements}, got {v.placements}" + ) + mesh = values_to_cat[0].device_mesh + local_cat = torch.cat( + [v.to_local() for v in values_to_cat], + dim=arg.split_dim, + ) + cat_shape = list(values_to_cat[0].shape) + cat_shape[arg.split_dim] = sum( + v.shape[arg.split_dim] for v in values_to_cat + ) + cat_shape = torch.Size(cat_shape) + cat_stride = values_to_cat[0].stride() + args_flattened.append( + DTensor.from_local( + local_cat, + mesh, + placements, + shape=cat_shape, + stride=cat_stride, + run_check=False, + ) + ) + else: + args_flattened.append(torch.cat(values_to_cat, dim=arg.split_dim)) + elif isinstance(arg, _CustomReducer): + reduced_val = arg.init_value + + for chunk_idx in range(len(chunks_flattened)): + reduced_val = arg.reduce_fn( + reduced_val, chunks_flattened[chunk_idx][arg_idx] + ) + + args_flattened.append(reduced_val) + else: + value = chunks_flattened[0][arg_idx] + for chunk_idx in range(1, len(chunks_flattened)): + if not chunks_flattened[chunk_idx][arg_idx] == value: + raise AssertionError( + f"Expected {value}, got {chunks_flattened[chunk_idx][arg_idx]}" + ) + args_flattened.append(value) + + # Stage 4: Unflatten combined args + return tree_unflatten(args_flattened, flatten_spec) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/schedules.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/schedules.py new file mode 100644 index 0000000000000000000000000000000000000000..fca427afafd05bd5707799b67844bb726d8c7e5f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/schedules.py @@ -0,0 +1,4097 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates + +import copy +import csv +import itertools +import logging +import re +from abc import ABC, abstractmethod +from collections import Counter, defaultdict +from collections.abc import Callable +from dataclasses import dataclass +from enum import Enum +from functools import lru_cache +from typing import Any, cast, Literal, NamedTuple, Protocol + +import torch +import torch.distributed as dist +from torch._dynamo import OptimizedModule +from torch.distributed.fsdp import FSDPModule, UnshardHandle +from torch.nn.modules.loss import _Loss +from torch.profiler import record_function + +from ._utils import ( + generate_rank_to_stage_mapping, + generate_stage_to_rank_mapping, + InferenceMode, +) +from .microbatch import ( + _split_tensor, + merge_chunks, + split_args_kwargs_into_chunks, + TensorChunkSpec, +) +from .stage import _PipelineStageBase, _RecvInfo, PipelineStage + + +__all__ = [ + "get_schedule_class", + "PipelineScheduleSingle", + "PipelineScheduleMulti", + "Schedule1F1B", + "ScheduleGPipe", + "ScheduleInterleaved1F1B", + "ScheduleLoopedBFS", + "ScheduleInterleavedZeroBubble", + "ScheduleZBVZeroBubble", + "ScheduleDualPipeV", +] + +logger = logging.getLogger(__name__) + + +class _ComputationType(str, Enum): + # TODO(whc) rename to _ActType? + FORWARD = "F" + BACKWARD_INPUT = "I" + BACKWARD_WEIGHT = "W" + UNSHARD = "UNSHARD" + RESHARD = "RESHARD" + SEND_F = "SEND_F" + RECV_F = "RECV_F" + SEND_B = "SEND_B" + RECV_B = "RECV_B" + FULL_BACKWARD = "B" + OVERLAP_F_B = "OVERLAP_F_B" + REDUCE_GRAD = "REDUCE_GRAD" + + @staticmethod + def from_str(action: str) -> "_ComputationType": + try: + return _ComputationType(action) + except ValueError as exc: + raise RuntimeError(f"Invalid computation type {action}") from exc + + +FORWARD = _ComputationType.FORWARD +BACKWARD_INPUT = _ComputationType.BACKWARD_INPUT +BACKWARD_WEIGHT = _ComputationType.BACKWARD_WEIGHT +UNSHARD = _ComputationType.UNSHARD +RESHARD = _ComputationType.RESHARD +SEND_F = _ComputationType.SEND_F +RECV_F = _ComputationType.RECV_F +SEND_B = _ComputationType.SEND_B +RECV_B = _ComputationType.RECV_B +FULL_BACKWARD = _ComputationType.FULL_BACKWARD +OVERLAP_F_B = _ComputationType.OVERLAP_F_B +REDUCE_GRAD = _ComputationType.REDUCE_GRAD + + +# Targets (e.g. labels) are always split along the batch dim (0). Use +# _split_tensor so DTensor targets preserve their Shard placements instead of +# being silently all-gathered to Replicate by the default dispatch path. +_TARGET_CHUNK_SPEC = TensorChunkSpec(0) + + +# Convenience shorthand for compute actions only since they are used in 'simple schedule format' +F = FORWARD +I = BACKWARD_INPUT +W = BACKWARD_WEIGHT +B = FULL_BACKWARD + +# Helper to parse an action string like 1F0 into a tuple of (stage_index, computation_type, microbatch_index) +_action_regex = re.compile( + r"(\d+)(F|I|B|W|UNSHARD|RESHARD|REDUCE_GRAD|SEND_F|RECV_F|SEND_B|RECV_B)(\d*)" +) + + +class _Action(NamedTuple): + stage_index: int + computation_type: _ComputationType + microbatch_index: int | None = None + sub_actions: tuple["_Action", ...] | None = None + + def __str__(self): + return self.__repr__() + + def __repr__(self): + if self.sub_actions is not None: + # Use recursive repr for sub_actions + sub_action_reprs = [repr(sub_action) for sub_action in self.sub_actions] + return f"({';'.join(sub_action_reprs)}){self.computation_type.value}" + else: + repr_str = str(self.stage_index) + # Use .value to get the short string (e.g., "F", "B") instead of the full enum name + repr_str += self.computation_type.value + if self.microbatch_index is not None: + repr_str += str(self.microbatch_index) + return repr_str + + @property + def is_compute_op(self) -> bool: + return self.computation_type in ( + FORWARD, + FULL_BACKWARD, + BACKWARD_INPUT, + BACKWARD_WEIGHT, + OVERLAP_F_B, + ) + + @staticmethod + def from_str(action_string: str): + """ + Reverse of __repr__ + + String should be formatted as [stage][action type][(microbatch)] + e.g. `2F0`, `1UNSHARD`, `3SEND_F1` + """ + action_string = action_string.strip() + if action_string == "": + return None + + # Check for sub_actions format: [sub_action1;sub_action2;...]ComputationType + if action_string.startswith("(") and ")" in action_string: + # Find the closing bracket to separate sub_actions from computation type + bracket_end = action_string.find(")") + sub_part = action_string[ + 1:bracket_end + ] # Remove '[' and get content before ']' + computation_type_part = action_string[ + bracket_end + 1 : + ] # Get part after ']' + + # Parse sub_actions + sub_actions = [] + if sub_part.strip(): + for sub_str in sub_part.split(";"): + sub_action = _Action.from_str(sub_str.strip()) + if sub_action is not None: + sub_actions.append(sub_action) + + # For sub_actions format, we create an action with just the computation type + # The stage_index and microbatch_index are not meaningful for the container action + return _Action( + stage_index=-1, # Placeholder, not meaningful for sub_actions container + computation_type=_ComputationType.from_str(computation_type_part), + microbatch_index=None, + sub_actions=tuple(sub_actions) if sub_actions else None, + ) + + # Handle regular single action format + if match := _action_regex.match(action_string): + stage_index, computation_type, microbatch_index = match.groups() + return _Action( + int(stage_index), + _ComputationType.from_str(computation_type), + int(microbatch_index) if len(microbatch_index) else None, + ) + elif action_string == "": + return None + raise RuntimeError( + f"Invalid action string: {action_string}, should be formatted as [stage][action type][(microbatch)] e.g. 2F0" + ) + + +@lru_cache +def _get_profiler_function_name(action: _Action) -> str: + return f"PP:{str(action)}" + + +def _format_pipeline_order( + pipeline_order: dict[int, list[_Action | None]], + error_step_number: int | None = None, +) -> str: + """ + Formats the pipeline order in a timestep (row) x rank (column) grid of actions + and returns the formatted string. + + If `error_step_number` is passed in, an additional label will be added to signify which step + that it is erroring on. + """ + + # don't mutate the original + pipeline_order = copy.deepcopy(pipeline_order) + + # Replace None with "" + for rank in pipeline_order: + for i in range(len(pipeline_order[rank])): + if pipeline_order[rank][i] is None: + # TODO make a real 'None action' that prints as empty string and make mypy happy + pipeline_order[rank][i] = "" # type: ignore[call-overload] + + # Calculate the maximum number of steps across all ranks + num_steps = max(len(actions) for actions in pipeline_order.values()) + step_labels = [ + "Step " + str(i).zfill(len(str(num_steps - 1))) for i in range(num_steps) + ] + # Sorting the dictionary by keys and retrieving values in that order + rank_actions = [ + pipeline_order.get(key, [""] * num_steps) for key in sorted(pipeline_order) + ] + # Transpose the list of lists (rows to columns) + # pyrefly: ignore [no-matching-overload] + transposed_actions = list(itertools.zip_longest(*rank_actions, fillvalue="")) + # Generate column labels for ranks + num_ranks = len(pipeline_order) + rank_labels = ["Rank " + str(i) for i in range(num_ranks)] + # Calculate the maximum length of each column, considering labels + max_lengths = [ + max(len(str(item)) if item is not None else 0 for item in col) + for col in zip(step_labels, *transposed_actions) + ] + # Format the header row with rank labels + header_row = " " * (len(step_labels[0]) + 2) + " ".join( + f"{label:<{max_lengths[i]}}" for i, label in enumerate(rank_labels) + ) + # Format each row with its corresponding label + formatted_rows = [ + f"{label}: " + + " ".join(f"{str(item):<{max_lengths[i]}}" for i, item in enumerate(row)) + + ( + " <-- ERROR HERE" + if error_step_number is not None + and int(label.split()[1]) == error_step_number + else "" + ) + for label, row in zip(step_labels, transposed_actions) + ] + # Join the rows into a single string + formatted_table = header_row + "\n" + "\n".join(formatted_rows) + "\n" + return formatted_table + + +class _PipelineSchedule(ABC): + def __init__( + self, + n_microbatches: int, + loss_fn: Callable[..., torch.Tensor] | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + ): + # From arguments + self._n_microbatches = n_microbatches + self._loss_fn = loss_fn + + # See documentation in `PipelineScheduleSingle` / `PipelineScheduleMulti` + self.scale_grads = scale_grads + + # Chunking specification for positional inputs. (default: `None`) + self._args_chunk_spec = args_chunk_spec + # Chunking specification for keyword inputs. (default: `None`) + self._kwargs_chunk_spec = kwargs_chunk_spec + self._output_merge_spec = output_merge_spec + """ + # args_chunk_spec and kwargs_chunk_spec specify how to chunk inputs. + # They are used to convert batch to microbatches in `step(x)`. See + # `TensorChunkSpec` for helper methods for creating them. + """ + + # Derived + self._has_backward = self._loss_fn is not None + + # Holds the losses for each microbatch. + self._internal_losses: list[torch.Tensor] = [] + logger.info("Using %s", self.__class__.__name__) + + def _maybe_compute_loss( + self, stage, output, target_mbs, mb_index, loss_kwargs=None + ): + if stage.is_last and self._loss_fn is not None: + loss = self._compute_loss(output, target_mbs[mb_index], loss_kwargs) # type: ignore[index] + self._internal_losses.append(loss) + + def _maybe_get_loss(self, stage, mb_index): + valid_index = 0 <= mb_index < len(self._internal_losses) + if stage.is_last and self._loss_fn is not None and valid_index: + return self._internal_losses[mb_index] + elif len(self._internal_losses) != 0 and not valid_index: + raise RuntimeError( + f"Loss for microbatch {mb_index} is not available. " + f"Available losses for microbatches: {self._internal_losses}" + ) + else: + return None + + def _update_losses(self, stages, losses): + """ + Update the losses to those in the internal state + """ + # if stages not a list turn into a list + if not isinstance(stages, list): + stages = [stages] + contains_last_stage = any(stage.is_last for stage in stages) + + # Return losses if there is a container passed in + if contains_last_stage and losses is not None: + if len(self._internal_losses) != self._n_microbatches: + raise RuntimeError( + f"Expecting {self._n_microbatches} losses but got {len(self._internal_losses)}" + ) + + # Clean external container first + losses.clear() + # Copy internal losses to external container + losses.extend(self._internal_losses) + + self._internal_losses.clear() + + def _warmup_p2p( + self, + stages: list[_PipelineStageBase], + has_backward: bool, + p2p_done: bool, + ) -> None: + """Run the P2P warm-up protocol for the given stages. + + For ``PipelineStage`` instances this executes the forward/backward vote + protocol (which warms up 2-rank sub-communicators) and sets each + stage's ``_inference_mode``. For other stage types it falls back to + the legacy ``_get_init_p2p_neighbors_ops`` + ``_batch_p2p`` path. + + Args: + stages: The pipeline stages owned by this rank. + has_backward: Whether the schedule includes a backward pass. + p2p_done: ``True`` if P2P neighbours have already been initialised + (avoids redundant init on eval↔train mode switches). + """ + if all(isinstance(stage, PipelineStage) for stage in stages): + # A fake process group cannot exchange real data: a cross-rank + # vote recv reads zeros and selects DYNAMIC, which then fails in + # `_recv_meta` (nothing was actually sent). Since dynamic + # inference can never work across ranks of a fake group, decide + # locally in that case: STATIC if every local stage has complete + # metadata, else error. Same-rank-only pipelines (e.g. a + # single-rank fake world) don't communicate, so the normal vote + # still works and DYNAMIC remains usable there. + pp_stages = cast(list[PipelineStage], stages) + has_cross_rank = any( + (not st.is_first and not st._is_same_rank(st.stage_index - 1)) + or (not st.is_last and not st._is_same_rank(st.stage_index + 1)) + for st in pp_stages + ) + if has_cross_rank and any( + dist.get_backend(st.group) == "fake" for st in pp_stages + ): + for st in pp_stages: + if InferenceMode.needs_dynamic(st._user_meta, has_backward): + raise RuntimeError( + f"Stage {st.stage_index} requires dynamic shape " + "inference, which is not supported with a fake " + "process group. Provide complete static metadata " + "(inputs/outputs, plus input_grads/output_grads " + "for DTensors with backward) to the PipelineStage " + "constructor." + ) + st._inference_mode = InferenceMode.STATIC + logger.debug( + "Fake process group detected; set inference_mode=static " + "for %d stage(s) without voting", + len(stages), + ) + return + acc: torch.Tensor | None = None + for stage in cast(list[PipelineStage], stages): + acc = stage._warmup_forward_vote(has_backward, received_acc=acc) + result: torch.Tensor | None = acc + determined_mode: InferenceMode | None = None + for stage in reversed(cast(list[PipelineStage], stages)): + result = stage._warmup_backward_result(received_result=result) + if result is None: + raise RuntimeError("P2P warm-up voting failed") + determined_mode = ( + InferenceMode.STATIC + if result.item() == 1 + else InferenceMode.DYNAMIC + ) + stage._inference_mode = determined_mode + logger.debug( + "Rank determined inference_mode=%s for %d stage(s)", + determined_mode.value if determined_mode else "None", + len(stages), + ) + elif not p2p_done: + all_ops: list[dist.P2POp] = [] + for stage in stages: + all_ops.extend(stage._get_init_p2p_neighbors_ops()) + _wait_batch_p2p(_batch_p2p(all_ops)) + + # TODO: STATIC mode group communicator warm-up gap + # The vote protocol above warms up 2-rank sub-communicators + # (used by `_batch_p2p` homogeneous fast-path). In DYNAMIC mode, + # `_send_meta`/`_recv_meta` (called during `_prepare_forward_infra` → + # `_forward_metadata_inference`) also warm up the *group* communicator + # (used by `_batch_p2p` mixed-op path). In STATIC mode, metadata + # inference is skipped, so the group communicator is NOT warmed up — + # it will be lazily created on the first mixed `_batch_p2p` call + # (e.g., 1F1B steady-state with both sends and recvs). + # Fix: run `_get_init_p2p_neighbors_ops` + `_batch_p2p` after the + # vote, gated by `not p2p_done`. + + def _initialize_pp_stages( + self, + stages: list[_PipelineStageBase], + args: tuple[Any, ...] | Any, + kwargs: dict[str, Any] | None, + target: Any, + fwd_initialized: bool, + bwd_initialized: bool, + loss_kwargs: dict[str, Any] | None = None, + ) -> tuple[bool, bool]: + """Common stage initialization shared by Single and Multi schedules. + + Handles mode-change detection (eval↔train), P2P warm-up, RNG forking, + forward / backward metadata inference, and FSDP cleanup. + + Returns the updated ``(fwd_initialized, bwd_initialized)`` flags. + """ + # Detect eval↔train mode switch: if has_backward changed since last + # init, re-initialize both fwd (recv buffers need different + # requires_grad) and bwd. p2p_done avoids redundant P2P warm-up. + p2p_done = fwd_initialized + if fwd_initialized and (self._has_backward != bwd_initialized): + fwd_initialized = False + bwd_initialized = False + + needs_fwd = not fwd_initialized + needs_bwd = self._has_backward and not bwd_initialized + + if not needs_fwd and not needs_bwd: + return fwd_initialized, bwd_initialized + + if needs_fwd: + self._warmup_p2p(stages, self._has_backward, p2p_done) + + # Fork RNG so metadata inference doesn't perturb training RNG. + devices = list( + { + torch.device(stage.device) + for stage in stages + if torch.device(stage.device).type != "cpu" + } + ) + device_type = devices[0].type if devices else None + # Assert all device types are the same + if device_type is not None and not all( + device.type == device_type for device in devices + ): + device_types = {device.type for device in devices} + raise AssertionError( + "All stages must have the same device type for RNG forking. " + f"Found device types: {device_types}" + ) + pipeline_stages = [ + stage for stage in stages if isinstance(stage, PipelineStage) + ] + for stage in pipeline_stages: + stage._pre_metadata_inference_backup() + + try: + with torch.random.fork_rng(devices=devices, device_type=device_type): + if needs_fwd: + next_stage_args: Any = None + for stage in stages: + stage_args = args if stage.is_first else next_stage_args + next_stage_args = stage._prepare_forward_infra( + self._n_microbatches, + stage_args, + kwargs, + has_backward=self._has_backward, + ) + fwd_initialized = True + + if needs_bwd: + prev_stage_grad_meta: Any = None + for stage in reversed(stages): + prev_stage_grad_meta = stage._prepare_backward_infra( + self._n_microbatches, + loss_fn=self._loss_fn, + target=target, + received_grad_meta=prev_stage_grad_meta, + loss_kwargs=loss_kwargs, + ) + bwd_initialized = True + finally: + for stage in pipeline_stages: + stage._post_metadata_inference_cleanup() + + return fwd_initialized, bwd_initialized + + @abstractmethod + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Run one iteration of the pipeline schedule with list of microbatches. + Will go through all the microbatches according to the schedule + implementation. + + Args: + microbatches: list of microbatch args. + return_outputs: whether to return the outputs from the last stage. + loss_kwargs: extra keyword arguments forwarded to the loss function. + """ + raise NotImplementedError + + @abstractmethod + def step( + self, + *args, + target=None, + losses: list | None = None, + return_outputs=True, + loss_kwargs: dict[str, Any] | None = None, + arg_mbs: Any = None, + kwarg_mbs: Any = None, + target_mbs: Any = None, + **kwargs, + ): + r"""Run one iteration of the pipeline schedule.""" + raise NotImplementedError + + def eval( + self, + *args, + target=None, + losses: list | None = None, + arg_mbs: Any = None, + kwarg_mbs: Any = None, + target_mbs: Any = None, + **kwargs, + ): + r"""Run one forward-only iteration of the pipeline schedule. + + ``eval`` uses the same input contract as ``step``, but temporarily + disables backward execution. By default, ``args``, ``kwargs``, and + ``target`` are full-batch values that the schedule splits into + microbatches. When the caller has already split inputs into + microbatches, pass them through ``arg_mbs``, ``kwarg_mbs``, and + ``target_mbs`` instead. + + Args: + \*args (Any): Whole-batch positional root inputs when this rank owns + the first pipeline stage. Do not pass positional inputs with + pre-split inputs. + target (Any, optional): Whole-batch target for loss computation. + When passing pre-split inputs, pass targets through + ``target_mbs`` instead. Default: ``None``. + losses (list, optional): Mutable list populated with one loss per + microbatch when this schedule owns the last stage and a + ``loss_fn`` was configured. Default: ``None``. + arg_mbs (list[tuple], optional): Pre-split positional inputs, one + tuple per microbatch. Default: ``None``. + kwarg_mbs (list[dict], optional): Pre-split keyword inputs, one + dict per microbatch. Default: ``None``. + target_mbs (list, optional): Pre-split targets, one entry per + microbatch. Default: ``None``. + \*\*kwargs (Any): Whole-batch keyword root inputs when this rank owns + the first pipeline stage. Do not pass keyword inputs with + pre-split inputs. + + Returns: + Any or None: The merged output from the last stage when this rank + owns the last stage; otherwise ``None``. + + Raises: + TypeError: If a pre-split microbatch container has the wrong type. + ValueError: If whole-batch and pre-split inputs are mixed, or if a + pre-split container does not have one entry per microbatch. + + Examples:: + + >>> # xdoctest: +SKIP("requires a constructed distributed pipeline schedule") + >>> output = schedule.eval(x, mask=mask) + >>> arg_mbs = [(x0,), (x1,)] + >>> kwarg_mbs = [{"mask": mask0}, {"mask": mask1}] + >>> output = schedule.eval( + ... arg_mbs=arg_mbs, + ... kwarg_mbs=kwarg_mbs, + ... ) + """ + # Save the original has_backward state + original_has_backward = self._has_backward + try: + self._has_backward = False + return self.step( + *args, + target=target, + losses=losses, + arg_mbs=arg_mbs, + kwarg_mbs=kwarg_mbs, + target_mbs=target_mbs, + **kwargs, + ) + finally: + # Restore the original state + self._has_backward = original_has_backward + + def _check_inputs( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + ) -> tuple[list, list]: + """ + Pre-process/check inputs + """ + + def check_type_and_len(mbs, name: str): + if not isinstance(mbs, list): + raise TypeError(f"{name} must be a list but got a {type(mbs)}") + if len(mbs) != self._n_microbatches: + raise ValueError( + f"Expecting {self._n_microbatches} {name} but got {len(mbs)}" + ) + + if arg_mbs is not None: + check_type_and_len(arg_mbs, "arg_mbs") + else: + arg_mbs = [()] * self._n_microbatches + + if kwarg_mbs is not None: + check_type_and_len(kwarg_mbs, "kwarg_mbs") + else: + kwarg_mbs = [{}] * self._n_microbatches + + if target_mbs is not None: + check_type_and_len(target_mbs, "target_mbs") + + if losses is not None: + if not isinstance(losses, list): + raise TypeError(f"losses must be a list but got a {type(losses)}") + + return arg_mbs, kwarg_mbs + + def _compute_loss(self, output, target, loss_kwargs=None): + return self._loss_fn(output, target, **(loss_kwargs or {})) # type: ignore[misc] + + def _split_inputs( + self, + args: tuple[Any, ...], + kwargs: dict[str, Any] | None = None, + ): + """ + Splits a full-batch input into chunks (i.e. microbatches) and returns + the chunks + """ + if args or kwargs: + args_split, kwargs_split = split_args_kwargs_into_chunks( + args, + kwargs, + self._n_microbatches, + self._args_chunk_spec, + self._kwargs_chunk_spec, + ) + return args_split, kwargs_split + else: + # Empty inputs (e.g. when called on middle stages) + # Return a list of empty tuples/dicts with matching length as chunks + return [()] * self._n_microbatches, [{}] * self._n_microbatches + + def _get_microbatch_inputs( + self, + args: tuple[Any, ...], + kwargs: dict[str, Any], + target: Any, + arg_mbs: Any, + kwarg_mbs: Any, + target_mbs: Any, + ) -> tuple[list | None, list | None, list | None]: + pre_split = any(mbs is not None for mbs in (arg_mbs, kwarg_mbs, target_mbs)) + if not pre_split: + args_split, kwargs_split = self._split_inputs(args, kwargs) + targets_split = ( + list(_split_tensor(target, _TARGET_CHUNK_SPEC, self._n_microbatches)) + if target is not None + else None + ) + return args_split, kwargs_split, targets_split + + if args: + raise ValueError( + "When using pre-split inputs, pass pre-split positional inputs " + "through arg_mbs=... instead of positional args." + ) + + if kwargs: + names = ", ".join(sorted(kwargs)) + raise ValueError( + f"Unexpected keyword arguments with pre-split inputs: {names}. " + "Pass pre-split keyword inputs through kwarg_mbs=..." + ) + + if target is not None: + raise ValueError( + "When using pre-split inputs, pass pre-split targets through " + "target_mbs=... instead of target=..." + ) + + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs) + + for mb_index, (arg_mb, kwarg_mb) in enumerate( + zip(arg_mbs, kwarg_mbs, strict=True) + ): + if not isinstance(arg_mb, tuple): + raise TypeError( + "arg_mbs must be a list of tuples, but " + f"arg_mbs[{mb_index}] is a {type(arg_mb)}" + ) + + if not isinstance(kwarg_mb, dict): + raise TypeError( + "kwarg_mbs must be a list of dicts, but " + f"kwarg_mbs[{mb_index}] is a {type(kwarg_mb)}" + ) + + return arg_mbs, kwarg_mbs, target_mbs + + def _merge_outputs(self, output_chunks: list[Any]) -> Any: + """ + Merge output chunks back to a batch state. + If output_merge_spec is None, the utility will merge output chunks by dimension 0 (batch dim). + """ + return merge_chunks( + output_chunks, + self._output_merge_spec, + ) + + +def _batch_p2p(p2p_ops: list[dist.P2POp], desc: str | None = None) -> list[dist.Work]: + """ + Wrapper over batch_isend_irecv that avoids coalescing for homogeneous + batches (all-send or all-recv). Coalescing serializes ops on a single + CUDA stream, which causes head-of-line blocking when independent P2P ops + could otherwise overlap. Mixed batches still use batch_isend_irecv for + deadlock avoidance. + """ + if len(p2p_ops) == 0: + return [] + desc_str = f"{desc}, " if desc else "" + logger.debug("batch_p2p %s%s", desc_str, p2p_ops) + + # Per-direction P2P (config.pipeline_per_direction_p2p) tags forward and + # backward ops with different communicators. A fused batch (e.g. 1F1B's + # fwd_sends + bwd_recvs) then spans >1 group; issue each group's ops as their + # own batch so they run on separate comms/streams instead of one FIFO. When + # all ops share a group (the default), this is a no-op fast path. + ops_by_group: dict[str, list[dist.P2POp]] = {} + for p in p2p_ops: + ops_by_group.setdefault(p.group.group_name, []).append(p) + if len(ops_by_group) > 1: + works: list[dist.Work] = [] + # Issue the groups in a deterministic, rank-independent order (by group + # name -- identical on every rank for a given group). The same fused + # batch is built in opposite list order on neighboring ranks (e.g. + # ``fwd_sends + bwd_recvs`` vs ``bwd_sends + fwd_recvs``); sorting keeps + # every rank issuing the two communicators in the same order. + for _, group_ops in sorted(ops_by_group.items()): + works += _batch_p2p(group_ops, desc=desc) + return works + + op_types = {p.op for p in p2p_ops} + if op_types == {dist.isend}: + send_works = [ + p.op(p.tensor, group=p.group, tag=p.tag, group_dst=p.group_peer) + for p in p2p_ops + ] + return [work for work in send_works if work is not None] + if op_types == {dist.irecv}: + recv_works = [ + p.op(p.tensor, group=p.group, tag=p.tag, group_src=p.group_peer) + for p in p2p_ops + ] + return [work for work in recv_works if work is not None] + + return dist.batch_isend_irecv(p2p_ops) + + +def _sorted_batch_p2p( + p2p_ops: list[dist.P2POp], desc: str | None = None +) -> dict[int, list[dist.Work]]: + """ + Sorts the list of P2P ops by the peer rank, and then calls + batch_isend_irecv. Return a dictionary of works by peer rank. This function + helps us avoid hangs in case of skip connections. + """ + # Arrange p2p_ops by peer rank: + # int is the peer rank; + # List is the list of ops towards the peer + ops_by_peer: dict[int, list[dist.P2POp]] = defaultdict(list) + work_by_peer: dict[int, list[dist.Work]] = {} + if len(p2p_ops) == 0: + return work_by_peer + + # Classify the ops by peer rank + for op in p2p_ops: + ops_by_peer[op.peer].append(op) + + # Call batch_isend_irecv per peer, in sorted order of the peers (to avoid hangs) + for peer, ops in sorted(ops_by_peer.items()): + work_by_peer[peer] = _batch_p2p(ops, desc=desc) + + return work_by_peer + + +def _wait_batch_p2p(work: list[dist.Work]): + """ + Waits for a list of dist.Work (typically from _batch_p2p / _sorted_batch_p2p). + """ + for w in work: + w.wait() + + +class PipelineScheduleSingle(_PipelineSchedule): + """ + Base class for single-stage schedules. + Implements the `step` method. + Derived classes should implement `_step_microbatches`. + + Gradients are scaled by num_microbatches depending on the `scale_grads` argument, defaulting to True. This setting + should match the configuration of your loss_fn, which may either average losses (scale_grads=True) + or sum losses (scale_grads=False). + """ + + def __init__( + self, + stage: _PipelineStageBase, + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + ): + # Init parent + super().__init__( + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + ) + # Self attributes + self._stage = stage + self._num_stages = stage.num_stages + self._stage_forward_initialized = False + self._stage_backward_initialized = False + + self.pipeline_order: dict[int, list[_Action | None]] | None = ( + self._get_pipeline_order() + ) + + def _initialize_stage(self, args, kwargs, target=None, loss_kwargs=None): + ( + self._stage_forward_initialized, + self._stage_backward_initialized, + ) = self._initialize_pp_stages( + [self._stage], + args, + kwargs, + target, + self._stage_forward_initialized, + self._stage_backward_initialized, + loss_kwargs=loss_kwargs, + ) + + def step( + self, + *args, + target=None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + arg_mbs: Any = None, + kwarg_mbs: Any = None, + target_mbs: Any = None, + **kwargs, + ): + r"""Run one training iteration of a single-stage pipeline schedule. + + By default, ``args``, ``kwargs``, and ``target`` are full-batch values + that the schedule splits into microbatches. When the caller has already + split inputs into microbatches, pass them through ``arg_mbs``, + ``kwarg_mbs``, and ``target_mbs`` instead. + + Args: + \*args (Any): Whole-batch positional inputs for the first pipeline + stage. Do not pass positional inputs with pre-split inputs. + target (Any, optional): Whole-batch target for loss computation. + When passing pre-split inputs, pass targets through + ``target_mbs`` instead. Default: ``None``. + losses (list, optional): Mutable list populated with one loss per + microbatch when this schedule owns the last stage and a + ``loss_fn`` was configured. Default: ``None``. + return_outputs (bool, optional): Whether to merge and return output + chunks on the last stage. Default: ``True``. + loss_kwargs (dict, optional): Extra keyword arguments forwarded to + the configured ``loss_fn``. Default: ``None``. + arg_mbs (list[tuple], optional): Pre-split positional inputs, one + tuple per microbatch. Default: ``None``. + kwarg_mbs (list[dict], optional): Pre-split keyword inputs, one + dict per microbatch. Default: ``None``. + target_mbs (list, optional): Pre-split targets, one entry per + microbatch. Default: ``None``. + \*\*kwargs (Any): Whole-batch keyword inputs for the first pipeline + stage. Do not pass keyword inputs with pre-split inputs. + + Returns: + Any or None: The merged output from the last stage when this rank + owns the last stage and ``return_outputs=True``; otherwise + ``None``. + + Raises: + RuntimeError: If backward computation is enabled and ``step`` is + called under ``torch.no_grad()``. Use :meth:`eval` for + forward-only execution. + TypeError: If a pre-split microbatch container has the wrong type. + ValueError: If whole-batch and pre-split inputs are mixed, or if a + pre-split container does not have one entry per microbatch. + + Examples:: + + >>> # xdoctest: +SKIP("requires a constructed distributed pipeline schedule") + >>> output = schedule.step(x, mask=mask, target=target) + >>> arg_mbs = [(x0,), (x1,)] + >>> kwarg_mbs = [{"mask": mask0}, {"mask": mask1}] + >>> target_mbs = [target0, target1] + >>> output = schedule.step( + ... arg_mbs=arg_mbs, + ... kwarg_mbs=kwarg_mbs, + ... target_mbs=target_mbs, + ... ) + """ + if self._has_backward and not torch.is_grad_enabled(): + raise RuntimeError( + "step() requires gradients to be enabled for backward computation; " + "it should not be used under torch.no_grad() context. " + "Please call eval() instead." + ) + + # Set the same has_backward flag for stage object + self._stage.has_backward = self._has_backward + + # Clean per iteration + self._stage.clear_runtime_states() + + args_split, kwargs_split, targets_split = self._get_microbatch_inputs( + args, + kwargs, + target, + arg_mbs, + kwarg_mbs, + target_mbs, + ) + + # Run microbatches + self._step_microbatches( + args_split, + kwargs_split, + targets_split, + losses, + return_outputs, + loss_kwargs=loss_kwargs, + ) + + # Return merged results per original format + if self._stage.is_last and return_outputs: + return self._merge_outputs(self._stage.output_chunks) + else: + return None + + def _get_pipeline_order(self) -> dict[int, list[_Action | None]] | None: + """ + Returns the pipeline execution order as a schedule IR. + + The returned IR is a dictionary mapping rank IDs to lists of actions. + Each action is either an _Action object representing computation to perform, + or None representing a deliberate idle step. + + The None values are used to represent pipeline bubbles where a rank + must wait for dependencies from other ranks before proceeding. However + during execution, with the _PipelineScheduleRuntime, these Nones are + skipped since the relevant communication (send/recv) will be scheduled and waited on. + + Returns: + A dictionary mapping rank -> list of actions + """ + return None + + +class _ScheduleForwardOnly(PipelineScheduleSingle): + """ + The forward-only schedule. + Will go through all the microbatches and perform only the forward pass + """ + + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Run one iteration of the pipeline schedule + """ + if target_mbs is not None or losses is not None: + raise RuntimeError( + "Forward-only schedule does not support loss computation" + ) + + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs, losses) + maybe_first_target = target_mbs[0] if target_mbs is not None else None + self._initialize_stage(arg_mbs[0], kwarg_mbs[0], maybe_first_target) + + # Delay send waits + fwd_sends_to_wait: list[list[dist.Work]] = [] + + # Run microbatches + for i in range(self._n_microbatches): + with record_function(f"Forward {i}"): + ops = self._stage.get_fwd_recv_ops(i) + works = _sorted_batch_p2p(ops, desc="fwd_recv") + for work in works.values(): + _wait_batch_p2p(work) + + self._stage.forward_one_chunk(i, arg_mbs[i], kwarg_mbs[i]) # type: ignore[index] + + ops = self._stage.get_fwd_send_ops(i) + works = _sorted_batch_p2p(ops, desc="fwd_send") + fwd_sends_to_wait.extend(works.values()) + + logger.debug("[%s] Forwarded microbatch %s", self._stage.stage_index, i) + + # Wait for all forward sends to finish + # This should not have performance impact because by the time the first + # backward arrives all the forward sends should have been finished. + for work in fwd_sends_to_wait: + _wait_batch_p2p(work) + + +class ScheduleGPipe(PipelineScheduleSingle): + """ + The GPipe schedule. + Will go through all the microbatches in a fill-drain manner. + """ + + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Run one iteration of the pipeline schedule with list of microbatches. + Will go through all the microbatches according to the GPipe schedule. + + Args: + microbatches: list of microbatch args. + return_outputs: whether to return the outputs from the last stage. + """ + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs, losses) + maybe_first_target = target_mbs[0] if target_mbs is not None else None + self._initialize_stage( + arg_mbs[0], kwarg_mbs[0], maybe_first_target, loss_kwargs + ) + + # Delay send waits + fwd_sends_to_wait: list[list[dist.Work]] = [] + + # Run microbatches + for i in range(self._n_microbatches): + with record_function(f"Forward {i}"): + ops = self._stage.get_fwd_recv_ops(i) + works = _sorted_batch_p2p(ops, desc="fwd_recv") + for work in works.values(): + _wait_batch_p2p(work) + + output = self._stage.forward_one_chunk( + i, arg_mbs[i], kwarg_mbs[i], save_forward_output=return_outputs + ) # type: ignore[index] + + ops = self._stage.get_fwd_send_ops(i) + works = _sorted_batch_p2p(ops, desc="fwd_send") + fwd_sends_to_wait.extend(works.values()) + + logger.debug("[%s] Forwarded microbatch %s", self._stage.stage_index, i) + + self._maybe_compute_loss(self._stage, output, target_mbs, i, loss_kwargs) + + # Wait for all forward sends to finish + # This should not have performance impact because by the time the first + # backward arrives all the forward sends should have been finished. + for work in fwd_sends_to_wait: + _wait_batch_p2p(work) + + # Run backward + # Delay send waits + bwd_sends_to_wait: list[list[dist.Work]] = [] + for i in range(self._n_microbatches): + with record_function(f"Backward {i}"): + ops = self._stage.get_bwd_recv_ops(i) + works = _sorted_batch_p2p(ops, desc="bwd_recv") + for work in works.values(): + _wait_batch_p2p(work) + + loss = self._maybe_get_loss(self._stage, i) + self._stage.backward_one_chunk( + i, + loss=loss, + last_backward=i == self._n_microbatches - 1, + ) + + ops = self._stage.get_bwd_send_ops(i) + works = _sorted_batch_p2p(ops, desc="bwd_send") + bwd_sends_to_wait.extend(works.values()) + + logger.debug("[%s] Backwarded microbatch %s", self._stage.stage_index, i) + + # Wait for all backward sends to finish + for work in bwd_sends_to_wait: + _wait_batch_p2p(work) + + # Update losses if there is a container passed in + self._update_losses(self._stage, losses) + + self._stage.perform_reduce_grad(self._n_microbatches if self.scale_grads else 1) + + def _get_pipeline_order(self) -> dict[int, list[_Action | None]] | None: + """ + Returns the pipeline order for GPipe schedule. + + See base method in PipelineScheduleSingle for details on the schedule IR format. + """ + pipeline_order = {} + pp_group_size = self._num_stages + + for rank in range(pp_group_size): + actions: list[_Action | None] = [] + + # 1. Initial delay based on rank position + warmup_delay = rank + actions.extend([None] * warmup_delay) + + # 2. Forward passes for all microbatches + for mb_idx in range(self._n_microbatches): + actions.append(_Action(rank, _ComputationType.FORWARD, mb_idx)) + + # 3. Wait period before backward passes can begin + backward_delay = 3 * (pp_group_size - 1 - rank) + actions.extend([None] * backward_delay) + + # 4. Backward passes for all microbatches + for mb_idx in range(self._n_microbatches): + actions.append(_Action(rank, _ComputationType.FULL_BACKWARD, mb_idx)) + + pipeline_order[rank] = _add_reduce_grad(actions, self._n_microbatches) + + return pipeline_order # type: ignore[return-value] + + +class Schedule1F1B(PipelineScheduleSingle): + """ + The 1F1B schedule. + Will perform one forward and one backward on the microbatches in steady state. + """ + + def __init__( + self, + stage: _PipelineStageBase, + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + ): + super().__init__( + stage=stage, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + ) + if n_microbatches < self._num_stages: + raise ValueError( + f"Number of microbatches ({n_microbatches}) must be greater than \ +or equal to the number of stages ({self._num_stages})." + ) + + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Run one iteration of the pipeline schedule with list of microbatches. + Will go through all the microbatches according to the 1F1B schedule. + + Args: + microbatches: list of microbatch args. + return_outputs: whether to return the outputs from the last stage. + loss_kwargs: extra keyword arguments forwarded to the loss function. + """ + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs, losses) + maybe_first_target = target_mbs[0] if target_mbs is not None else None + self._initialize_stage( + arg_mbs[0], kwarg_mbs[0], maybe_first_target, loss_kwargs + ) + + # Last stage has 1 warmup, second-to-last 2 warmups, ... + # first stage `num_stages` warmups + warmup_chunks = min( + self._n_microbatches, + self._num_stages - self._stage.stage_index, + ) + + # Chunk counters + fwd_mb_index = 0 + bwd_mb_index = 0 + + # Warmup phase + send_work: list[dist.Work] = [] + fwd_sends = [] + for _ in range(warmup_chunks): + # Receive activations + fwd_recvs = self._stage.get_fwd_recv_ops(fwd_mb_index) + _wait_batch_p2p(_batch_p2p(fwd_recvs, desc="fwd_recv")) + + # Compute + output = self._stage.forward_one_chunk( + fwd_mb_index, + arg_mbs[fwd_mb_index], + kwarg_mbs[fwd_mb_index], + save_forward_output=return_outputs, + ) # type: ignore[index] + + # Clear previous chunk's forward sends (hopefully they have well + # finished, otherwise, we are heavily communication bound, in which + # case it doesn't create a lot of benefit to compute next chunk + # eagerly either) + _wait_batch_p2p(send_work) + + # Send activations + fwd_sends = self._stage.get_fwd_send_ops(fwd_mb_index) + if fwd_mb_index != warmup_chunks - 1: + # Safe to fire + send_work = _batch_p2p(fwd_sends, desc="fwd_send") + # otherwise: + # The last forward send is left for fuse with first 1B in 1B1F below + + # Compute loss + self._maybe_compute_loss( + self._stage, output, target_mbs, fwd_mb_index, loss_kwargs + ) + fwd_mb_index += 1 + + # Now we should have send ops left over, to be fused with first 1B of 1B1F phase below. + + # 1B1F phase + while True: # Don't worry, we have a break inside + # We actually do 1B first as the `1B1F` name indicates, so prepare its recv ops + bwd_recvs = self._stage.get_bwd_recv_ops(bwd_mb_index) + + # Now, we need to fire the fwd_sends and bwd_recvs together + _wait_batch_p2p(_batch_p2p(fwd_sends + bwd_recvs, desc="fwd_send_bwd_recv")) + + # Backward one chunk + loss = self._maybe_get_loss(self._stage, bwd_mb_index) + self._stage.backward_one_chunk( + bwd_mb_index, + loss=loss, + last_backward=bwd_mb_index == self._n_microbatches - 1, + ) + + # Get the bwd send ops, but don't fire, to be fused with the 1F below + bwd_sends = self._stage.get_bwd_send_ops(bwd_mb_index) + bwd_mb_index += 1 + + if fwd_mb_index == self._n_microbatches: + # We are done with 1B1F, so break with some left-over bwd_sends + break + + # We prepare 1F of the `1B1F` + fwd_recvs = self._stage.get_fwd_recv_ops(fwd_mb_index) + + # Fuse it with bwd_sends above + _wait_batch_p2p(_batch_p2p(bwd_sends + fwd_recvs, desc="bwd_send_fwd_recv")) + + # Now do the fwd + output = self._stage.forward_one_chunk( + fwd_mb_index, + arg_mbs[fwd_mb_index], + kwarg_mbs[fwd_mb_index], + save_forward_output=return_outputs, + ) # type: ignore[index] + + # Compute loss + self._maybe_compute_loss( + self._stage, output, target_mbs, fwd_mb_index, loss_kwargs + ) + + # Get the fwd send ops, but don't fire, leave it for the next iter (wrap-around) + fwd_sends = self._stage.get_fwd_send_ops(fwd_mb_index) + fwd_mb_index += 1 + + # Remember we still have some bwd_sends left over after the break? Now it is time to fire it + send_work = _batch_p2p(bwd_sends, desc="bwd_send") + + # Cooldown + while bwd_mb_index < self._n_microbatches: + # prepare bwd recv ops + bwd_recvs = self._stage.get_bwd_recv_ops(bwd_mb_index) + _wait_batch_p2p(_batch_p2p(bwd_recvs, desc="bwd_recv")) + + # Backward one chunk + loss = self._maybe_get_loss(self._stage, bwd_mb_index) + self._stage.backward_one_chunk( + bwd_mb_index, + loss=loss, + last_backward=bwd_mb_index == self._n_microbatches - 1, + ) + + # Clear previous chunk's backward sends (hopefully they have well finished) + _wait_batch_p2p(send_work) + + # Get the bwd send ops, fire it + bwd_sends = self._stage.get_bwd_send_ops(bwd_mb_index) + send_work = _batch_p2p(bwd_sends, desc="bwd_send") + bwd_mb_index += 1 + + # Wait for the last backward send to finish + _wait_batch_p2p(send_work) + + # Return losses if there is a container passed in + self._update_losses(self._stage, losses) + + self._stage.perform_reduce_grad(self._n_microbatches if self.scale_grads else 1) + + def _get_pipeline_order(self) -> dict[int, list[_Action | None]] | None: + """ + Returns the pipeline order for 1F1B schedule. + + See base method in PipelineScheduleSingle for details on the schedule IR format. + """ + pipeline_order = {} + pp_group_size = self._num_stages + + for rank in range(pp_group_size): + actions: list[_Action | None] = [] + + # 1. Warmup phase: initial delay based on rank + actions.extend([None] * rank) + + # 2. Initial forward passes before 1F1B phase + num_forward = (pp_group_size - 1) - rank + forward_mb = 0 + for i in range(num_forward): + actions.append(_Action(rank, _ComputationType.FORWARD, i)) + forward_mb = i + + # 3. Wait for backward to be ready + wait_for_1f1b = max(0, 2 * (pp_group_size - 1 - rank)) + actions.extend([None] * wait_for_1f1b) + + # 4. 1F1B steady state phase + backward_mb = 0 + remaining_forward = self._n_microbatches - num_forward + + while remaining_forward > 0: + # One forward + forward_mb += 1 + actions.append(_Action(rank, _ComputationType.FORWARD, forward_mb)) + remaining_forward -= 1 + + # One backward + actions.append( + _Action(rank, _ComputationType.FULL_BACKWARD, backward_mb) + ) + backward_mb += 1 + + # 5. Cooldown phase: remaining backward passes + remaining_backward = self._n_microbatches - backward_mb + + while remaining_backward > 0: + # Add None and backward actions in alternating pattern + # based on distance from the last stage + if (pp_group_size - rank) > 0: + actions.append(None) + # Decrement the wait counter only if we still have backward passes to do + if remaining_backward > 0: + actions.append( + _Action(rank, _ComputationType.FULL_BACKWARD, backward_mb) + ) + backward_mb += 1 + remaining_backward -= 1 + else: + # If we're at the last stage, just add backward actions without None + actions.append( + _Action(rank, _ComputationType.FULL_BACKWARD, backward_mb) + ) + backward_mb += 1 + remaining_backward -= 1 + + pipeline_order[rank] = _add_reduce_grad(actions, self._n_microbatches) + return pipeline_order + + +def _requires_reduce_grad(action_type: _ComputationType) -> bool: + return action_type in (W, B) + + +def _add_reduce_grad( + actions: list[_Action | None], n_microbatches: int +) -> list[_Action | None]: + """ + REDUCE_GRAD refers to joint across minibatches grad reduction. + reduce_grad frees memory and we want to schedule it just after the last "backward"-like stage. + """ + actions_with_reduce_grad: list[_Action | None] = [] + cnt: dict[int, int] = defaultdict(int) + + def _leaf_action(a, to_schedule): + if _requires_reduce_grad(a.computation_type): + stage_index = a.stage_index + cnt[stage_index] += 1 + if cnt[stage_index] == n_microbatches: + to_schedule.append(stage_index) + + for a in actions: + if a is None: + continue + actions_with_reduce_grad.append(a) + schedule_reduce_grad_stage_idxs: list[int] = [] + if a.computation_type == OVERLAP_F_B and a.sub_actions is not None: + for sub_action in a.sub_actions: + _leaf_action(sub_action, schedule_reduce_grad_stage_idxs) + else: + _leaf_action(a, schedule_reduce_grad_stage_idxs) + + for stage_idx in schedule_reduce_grad_stage_idxs: + actions_with_reduce_grad.append(_Action(stage_idx, REDUCE_GRAD, None)) + return actions_with_reduce_grad + + +def _add_unshard_reshard( + compute_actions: list[_Action | None], + max_active_stages: int = 3, +) -> list[_Action]: + """Given a basic schedule involving only compute actions (F,B,W,OVERLAP_F_B), add UNSHARD/RESHARD actions for FSDP. + + UNSHARD refers to fetching the full contents of an FSDP-sharded layer, requiring an all-gather operation. + RESHARD does the opposite, releasing memory (but doing no communication) + + We abandon the "timestep lock" during lowering + + max_active_stages controls how many prefetches we allow. It should be measured in mb and tuneable but in practice + 3 stages is probably the thing we want? + (to account for having one f and one b active, and something else prefetching?) + """ + + def next_stage_indices(count: int, next_actions: list[_Action | None]) -> list[int]: + """Remove duplicates (same stage, different microbatch), find next 'count' stages that will do compute.""" + seen: set[int] = set() + ret: list[int] = [] + + for a in next_actions: + if a is not None: + # Handle OVERLAP_F_B actions by checking their sub_actions + if a.computation_type == OVERLAP_F_B and a.sub_actions is not None: + for sub_action in a.sub_actions: + if sub_action.stage_index not in seen: + seen.add(sub_action.stage_index) + ret.append(sub_action.stage_index) + if len(ret) >= count: + break + else: + # Regular action + if a.stage_index not in seen: + seen.add(a.stage_index) + ret.append(a.stage_index) + if len(ret) == count: + break + return ret + + active_stages: set[int] = set() + fsdp_aware_actions: list[_Action] = [] + + def _unshard(stage_index: int): + active_stages.add(stage_index) + fsdp_aware_actions.append(_Action(stage_index, UNSHARD, None)) + + def _reshard(stage_index: int): + active_stages.remove(stage_index) + fsdp_aware_actions.append(_Action(stage_index, RESHARD, None)) + + for i, action in enumerate(compute_actions): + if action is None: + continue + + # We prefetch the next N stages we'll see, dropping existing stages to make room + next_n = next_stage_indices(max_active_stages, compute_actions[i:]) + # Fetch needs to be ordered correctly, so don't use a set + fetch = list(filter(lambda s: s not in active_stages, next_n)) + # Unclear what the best policy is for eviction, but we can maintain order so we do + evict = list(filter(lambda s: s not in next_n, active_stages)) + + # logger.debug( + # "_add_unshard_reshard Step %d active: %s fetch %s, evict %s", + # i, + # active_stages, + # fetch, + # evict, + # ) + + for stage in evict: + _reshard(stage) + for stage in fetch: + _unshard(stage) + fsdp_aware_actions.append(action) + + # Reshard all remaining active stages after processing all operations + for stage in list(active_stages): + _reshard(stage) + + return fsdp_aware_actions + + +def _merge_bw( + compute_actions: list[_Action | None], +) -> list[_Action]: + """Given a basic schedule involving only compute actions (F,I,W), merge adjacent I and W ops into B ops. + (note: I = BACKWARD_INPUT, W = BACKWARD_WEIGHT, B = FULL_BACKWARD) + + B refers to running the whole backward (not separating grad_input and grad_weight), which can be more efficient + in some cases. + """ + merged_actions = [] + while compute_actions: + action = compute_actions.pop(0) + if action is None: + continue + + # Remove any None actions and find the next non-None action + while len(compute_actions) and compute_actions[0] is None: + compute_actions.pop(0) + + # Get the next action if it exists + next_action = compute_actions[0] if len(compute_actions) > 0 else None + + if ( + action.computation_type == BACKWARD_INPUT + and next_action is not None + and next_action.computation_type == BACKWARD_WEIGHT + and action.stage_index == next_action.stage_index + and action.microbatch_index == next_action.microbatch_index + ): + merged_actions.append( + _Action(action.stage_index, FULL_BACKWARD, action.microbatch_index) + ) + compute_actions.pop(0) + else: + merged_actions.append(action) + return merged_actions + + +def _add_send_recv( + compute_actions: dict[int, list[_Action]], + stage_to_rank: Callable[[int], int], + num_stages: int, +) -> dict[int, list[_Action]]: + """ + Transforms a compute-only schedule into a complete schedule with communication actions. + + For actions with sub-actions (OVERLAP_F_B) we ensure that all the subactions have been + computed and the communication is ready + """ + comm_actions: dict[int, list[_Action]] = {rank: [] for rank in compute_actions} + prev_actions: dict[int, set[_Action]] = {rank: set() for rank in compute_actions} + + def _has_comms(action: _Action) -> bool: + if action.computation_type == F: + return action.stage_index != num_stages - 1 and stage_to_rank( + action.stage_index + 1 + ) != stage_to_rank(action.stage_index) + elif action.computation_type in (BACKWARD_INPUT, FULL_BACKWARD): + return action.stage_index != 0 and stage_to_rank( + action.stage_index - 1 + ) != stage_to_rank(action.stage_index) + return False + + def _get_comms(action: _Action) -> tuple[_Action, _Action]: + if not _has_comms(action): + raise AssertionError(f"{action} is not a valid comm action") + stage_idx = action.stage_index + ctype = action.computation_type + mb_idx = action.microbatch_index + send = _Action(stage_idx, SEND_F if ctype == F else SEND_B, mb_idx) + recv_stage_idx = stage_idx + 1 if ctype == F else stage_idx - 1 + recv = _Action(recv_stage_idx, RECV_F if ctype == F else RECV_B, mb_idx) + return send, recv + + def _ready_to_schedule(action: _Action | None, prev_actions: set[_Action]) -> bool: + """We don't put our own recv ops in the schedule, we let a sender on another rank put our recv ops in place. + This helps ensure a sane (non-hanging) ordering of sends and recvs. + But it also means we might not be able to schedule our next compute action yet. + """ + if action is None: + return True + elif action.computation_type == F and action.stage_index != 0: + if ( + _Action(action.stage_index, RECV_F, action.microbatch_index) + in prev_actions + ): + return True + elif ( + _Action(action.stage_index - 1, F, action.microbatch_index) + in prev_actions + ): + return True + return False + elif ( + action.computation_type in (BACKWARD_INPUT, FULL_BACKWARD) + and action.stage_index != num_stages - 1 + ): + if ( + _Action(action.stage_index, RECV_B, action.microbatch_index) + in prev_actions + ): + return True + elif ( + _Action(action.stage_index + 1, BACKWARD_INPUT, action.microbatch_index) + in prev_actions + ): + return True + elif ( + _Action(action.stage_index + 1, FULL_BACKWARD, action.microbatch_index) + in prev_actions + ): + return True + return False + else: + return True + + while compute_actions: + progress = False + # go in order of ranks even if dict keys aren't ordered + for rank in sorted(compute_actions): + if not (len(compute_actions[rank]) > 0): + raise AssertionError(f"{rank=}, {len(compute_actions[rank])=}") + action = compute_actions[rank][0] + # handle case where parent action (e.g. OVERLAP_F_B) can be comprised of subactions + if action is not None and action.sub_actions is not None: + all_actions = action.sub_actions + else: + all_actions = (action,) + + if not all(_ready_to_schedule(a, prev_actions[rank]) for a in all_actions): + continue + + # The action's dependencies are satisfied, so add to schedule + if action is not None: + comm_actions[rank].append(action) + for a in all_actions: + prev_actions[rank].add(a) + if _has_comms(a): + send, recv = _get_comms(a) + # TODO we can avoid send/recv if the 2 stages are on the same rank. + # should we avoid that in the runtime or here? + comm_actions[rank].append(send) + prev_actions[rank].add(send) + comm_actions[stage_to_rank(recv.stage_index)].append(recv) + prev_actions[stage_to_rank(recv.stage_index)].add(recv) + + compute_actions[rank].pop(0) + if len(compute_actions[rank]) == 0: + del compute_actions[rank] + progress = True + if not progress: + raise AssertionError( + "Malformed compute schedule, can't schedule sends/recvs" + ) + return comm_actions + + +def _defer_recv_ops( + actions: dict[int, list[_Action]], + stage_to_rank: Callable[[int], int], +) -> dict[int, list[_Action]]: + """ + Defers RECV operations to reduce interference with unrelated compute ops, + while maintaining deadlock-safe ordering via rank-parity P2P ordering. + + By default, the schedule places RECV ops as early as possible (ASAP) to overlap + P2P communication with computation. However, on some platforms (e.g., AMD ROCm), + a pending RECV can block unrelated compute ops that also use the communication + fabric (e.g., FSDP allgather inside a forward pass), creating pipeline bubbles. + + This function defers each RECV to as late as possible, subject to: + 1. A RECV must appear before the compute op that consumes its data. + 2. Deadlock avoidance via rank-parity ordering (see pytorch/pytorch#172668): + - When current rank > peer rank: SEND before RECV is safe, so deferred + RECVs from that peer are NOT flushed before SENDs to that peer. + - When current rank < peer rank: RECV before SEND is required, so + deferred RECVs from that peer ARE flushed before SENDs to that peer. + + This breaks circular waits: the lower-ranked side always posts RECV first, + providing the matching target for the higher-ranked side's SEND. + """ + RECV_F = _ComputationType.RECV_F + RECV_B = _ComputationType.RECV_B + SEND_F = _ComputationType.SEND_F + SEND_B = _ComputationType.SEND_B + + def _recv_peer_rank(action: _Action) -> int: + if action.computation_type == RECV_F: + return stage_to_rank(action.stage_index - 1) + else: + return stage_to_rank(action.stage_index + 1) + + def _send_peer_rank(action: _Action) -> int: + if action.computation_type == SEND_F: + return stage_to_rank(action.stage_index + 1) + else: + return stage_to_rank(action.stage_index - 1) + + result: dict[int, list[_Action]] = {} + for rank, action_list in actions.items(): + new_actions: list[_Action] = [] + deferred: dict[tuple[int, _ComputationType, int | None], _Action] = {} + + for action in action_list: + if action.computation_type in (RECV_F, RECV_B): + key = ( + action.stage_index, + action.computation_type, + action.microbatch_index, + ) + deferred[key] = action + continue + + if action.computation_type in (SEND_F, SEND_B): + peer = _send_peer_rank(action) + # Rank-parity rule: only flush RECVs when rank < peer + # (lower rank does RECV before SEND) + if rank < peer: + to_flush = [ + k for k, v in deferred.items() if _recv_peer_rank(v) == peer + ] + for key in to_flush: + new_actions.append(deferred.pop(key)) + + # Constraint 1: before a compute op, flush the RECV it consumes + consumers = ( + action.sub_actions if action.sub_actions is not None else (action,) + ) + for sub in consumers: + if sub.computation_type == FORWARD: + key = (sub.stage_index, RECV_F, sub.microbatch_index) + if key in deferred: + new_actions.append(deferred.pop(key)) + elif sub.computation_type in (FULL_BACKWARD, BACKWARD_INPUT): + key = (sub.stage_index, RECV_B, sub.microbatch_index) + if key in deferred: + new_actions.append(deferred.pop(key)) + + new_actions.append(action) + + if deferred: + raise AssertionError( + f"Malformed input schedule on rank {rank}: leftover RECV ops " + f"with no consumer found: {list(deferred.values())}. " + "Every RECV must be consumed by a downstream compute op." + ) + + result[rank] = new_actions + return result + + +def _validate_schedule( + actions: dict[int, list[_Action | None]], + pp_group_size: int, + num_stages: int, + num_microbatches: int, +) -> dict[int, int]: + if not (len(actions) == pp_group_size): + raise AssertionError( + f"Schedule has incorrect number of ranks - expected {pp_group_size}, actual {len(actions)}" + ) + for rank in range(pp_group_size): + if rank not in actions: + raise AssertionError(f"Schedule is missing actions for rank {rank}") + + # We will count all the actions per stage and ensure they happen in a valid order + # (e.g. F before (B, I) before W for a given microbatch) + stage_actions: dict[int, dict[_ComputationType, set]] = { + stage_id: { + F: set(), + B: set(), + I: set(), + W: set(), + } + for stage_id in range(num_stages) + } + stage_index_to_rank_mapping = {} + + def _process_action(action: _Action, rank: int, step: int): + """Process a single action and update stage_actions and stage_index_to_rank_mapping""" + s_id = action.stage_index + ctype = action.computation_type + mb_id = action.microbatch_index + + if ctype == F: + stage_actions[s_id][F].add(mb_id) + elif ctype == B: + if mb_id not in stage_actions[s_id][F]: + error_msg = ( + f"Rank {rank}, step {step}: Running Full Backward for stage {s_id}, " + f"microbatch {mb_id} without first running Forward" + ) + formatted_schedule = _format_pipeline_order( + actions, error_step_number=step + ) + full_error_msg = ( + f"{error_msg}\n\nFull pipeline schedule:\n{formatted_schedule}" + ) + raise AssertionError(full_error_msg) + stage_actions[s_id][B].add(mb_id) + elif ctype == I: + if mb_id not in stage_actions[s_id][F]: + error_msg = ( + f"Rank {rank}, step {step}: Running Backward Input for stage {s_id}, " + f"microbatch {mb_id} without first running Forward" + ) + formatted_schedule = _format_pipeline_order( + actions, error_step_number=step + ) + full_error_msg = ( + f"{error_msg}\n\nFull pipeline schedule:\n{formatted_schedule}" + ) + raise AssertionError(full_error_msg) + stage_actions[s_id][I].add(mb_id) + elif ctype == W: + if mb_id not in stage_actions[s_id][I]: + error_msg = ( + f"Rank {rank}, step {step}: Running Backward Weight for stage {s_id}, " + f"microbatch {mb_id} without first running Backward Input" + ) + formatted_schedule = _format_pipeline_order( + actions, error_step_number=step + ) + full_error_msg = ( + f"{error_msg}\n\nFull pipeline schedule:\n{formatted_schedule}" + ) + raise AssertionError(full_error_msg) + stage_actions[s_id][W].add(mb_id) + + if s_id not in stage_index_to_rank_mapping: + stage_index_to_rank_mapping[s_id] = rank + else: + existing_rank = stage_index_to_rank_mapping[s_id] + if not (rank == existing_rank): + raise AssertionError( + f"Rank {rank}, step {step}: Stage {s_id} is assigned to both rank {rank} and rank {existing_rank}" + ) + + for rank in actions: + for step, action in enumerate(actions[rank]): + if action is None: + continue + if not isinstance(action, _Action): + raise AssertionError( + f"Rank {rank}, step {step}: Got an invalid action: {action}, expected instance of _Action" + ) + + # Check if action has sub_actions + if action.sub_actions is not None: + # Process each sub_action instead of the main action + for sub_action in action.sub_actions: + _process_action(sub_action, rank, step) + else: + # Process the main action normally + _process_action(action, rank, step) + + for s_id in stage_actions: + f_mb = len(stage_actions[s_id][F]) + b_mb = len(stage_actions[s_id][B]) + i_mb = len(stage_actions[s_id][I]) + w_mb = len(stage_actions[s_id][W]) + + if not (f_mb == num_microbatches): + raise AssertionError( + f"Got {f_mb} {F} microbatches for stage {s_id}, expected {num_microbatches}" + ) + + if not (i_mb == w_mb): + raise AssertionError( + f"Invalid backward microbatches for stage {s_id}: I and W must have equal counts, \ + but got I={i_mb}, W={w_mb}" + ) + + if not (b_mb + (i_mb + w_mb) // 2 == num_microbatches): + raise AssertionError( + f"Invalid backward microbatches for stage {s_id}: expected {num_microbatches} total backwards, \ + but got B={b_mb}, I={i_mb}, W={w_mb}" + ) + return stage_index_to_rank_mapping + + +class PipelineScheduleMulti(_PipelineSchedule): + """ + Base class for multi-stage schedules. + Implements the `step` method. + + Gradients are scaled by num_microbatches depending on the `scale_grads` argument, defaulting to True. This setting + should match the configuration of your loss_fn, which may either average losses (scale_grads=True) + or sum losses (scale_grads=False). + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + use_full_backward: bool | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + ): + # Init parent + super().__init__( + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + ) + # Self attributes + self._stages = stages + self._num_stages = stages[0].num_stages + self.pp_group_size = stages[0].group_size + self.rank = stages[0].group_rank + # Set the pipeline stage states + self.stage_index_to_group_rank = generate_stage_to_rank_mapping( + self.pp_group_size, self._num_stages + ) + for stage in self._stages: + stage.stage_index_to_group_rank = self.stage_index_to_group_rank + + self._stages_forward_initialized = False + self._stages_backward_initialized = False + + # avoid putting a reference to 'self' inside the lambda, it creates a ref cycle + has_loss: bool = self._loss_fn is not None + self._should_compute_loss = lambda stage: stage.is_last and has_loss + + # This will be set during init of derived schedules + self.pipeline_order: dict[int, list[_Action | None]] = {} + + # When using a custom backward function, we may or may not need autograd to be used + # for the backward pass. This flag is used to determine whether or torch.is_grad_enabled() + # check should be performed before the step function. + self._backward_requires_autograd = backward_requires_autograd + + if use_full_backward is not None: + logger.warning( + "Deprecation warning: 'use_full_backward' is no longer supported. " + "Simply stop passing it, and everything should still work fine." + ) + + def _initialize_stages( + self, args: tuple[Any, ...], kwargs, target=None, loss_kwargs=None + ): + reinit_for_mode_switch = self._stages_forward_initialized and ( + self._has_backward != self._stages_backward_initialized + ) + forward_initialized_before = self._stages_forward_initialized + ( + self._stages_forward_initialized, + self._stages_backward_initialized, + ) = self._initialize_pp_stages( + self._stages, + args, + kwargs, + target, + self._stages_forward_initialized, + self._stages_backward_initialized, + loss_kwargs=loss_kwargs, + ) + + if self._stages_forward_initialized and ( + not forward_initialized_before or reinit_for_mode_switch + ): + self._validate_adjacent_stage_communication() + + def _validate_adjacent_stage_communication(self) -> None: + """Validate that stage communication follows adjacent-stage topology only.""" + + def _check_stage_indices( + stage_idx: int, + direction: str, + actual_stage_indices: set[int], + expected_stage_indices: set[int], + ) -> None: + non_adjacent_stage_indices = actual_stage_indices - expected_stage_indices + if non_adjacent_stage_indices: + raise RuntimeError( + "PipelineScheduleMulti only supports adjacent-stage " + f"communication, but stage {stage_idx} has {direction} " + f"stages {sorted(actual_stage_indices)} with " + f"non-adjacent stages {sorted(non_adjacent_stage_indices)} " + f"(allowed adjacent stages: " + f"{sorted(expected_stage_indices)}). This commonly " + "indicates skip connections, which are unsupported in " + "this schedule runtime." + ) + + for stage in self._stages: + stage_idx = stage.stage_index + actual_fwd_recv_sources: set[int] = { + info.source + for info in stage.args_recv_info[0] + if isinstance(info, _RecvInfo) and info.source is not None + } + expected_fwd_recv_sources = set() if stage.is_first else {stage_idx - 1} + _check_stage_indices( + stage_idx, + "forward recv", + actual_fwd_recv_sources, + expected_fwd_recv_sources, + ) + + # act_send_info is keyed by output index (not microbatch index), + # so .values() yields per-output destination lists. + actual_fwd_send_dests: set[int] = { + dst + for dsts in stage.act_send_info.values() + for dst in dsts + if dst is not None + } + expected_fwd_send_dests = set() if stage.is_last else {stage_idx + 1} + _check_stage_indices( + stage_idx, + "forward send", + actual_fwd_send_dests, + expected_fwd_send_dests, + ) + + def _validate_and_set_stage_mapping( + self, actions: dict[int, list[_Action | None]] + ) -> None: + """ + Allocates the stage index to rank mapping which is needed for communication + """ + self.stage_index_to_group_rank = _validate_schedule( + actions, + self.pp_group_size, + self._num_stages, + self._n_microbatches, + ) + for stage in self._stages: + stage.stage_index_to_group_rank = self.stage_index_to_group_rank + + def _dump_csv(self, filename): + """Dump a CSV representation of the schedule into a file with the provided filename.""" + with open(filename, "w", newline="") as csvfile: + writer = csv.writer(csvfile) + for rank in self.pipeline_order: + writer.writerow(self.pipeline_order[rank]) + + def _load_csv( + self, + filename: str, + format: Literal["compute_only", "compute_comms"] = "compute_only", + ): + """Load a CSV representation of the schedule from a file with the provided filename. + This API will most likely get renamed/refactored so is marked as internal for now. + + format must be "compute_only" for PipelineScheduleMulti. + """ + if format != "compute_only": + raise AssertionError(f'format must be "compute_only", got {format}') + with open(filename, newline="") as csvfile: + reader = csv.reader(csvfile) + for rank, row in enumerate(reader): + self.pipeline_order[rank] = [_Action.from_str(s) for s in row] + + # Validates the order of the pipeline actions and infers the stage_to_rank_mapping. + # This will overwrite the default stage_to_rank_mapping created in the constructor + self._validate_and_set_stage_mapping(self.pipeline_order) + + def step( + self, + *args, + target=None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + arg_mbs: Any = None, + kwarg_mbs: Any = None, + target_mbs: Any = None, + **kwargs, + ): + r"""Run one training iteration of a multi-stage pipeline schedule. + + By default, ``args``, ``kwargs``, and ``target`` are full-batch values + that the schedule splits into microbatches. When the caller has already + split inputs into microbatches, pass them through ``arg_mbs``, + ``kwarg_mbs``, and ``target_mbs`` instead. + + Args: + \*args (Any): Whole-batch positional root inputs when this rank owns + the first pipeline stage. Do not pass positional inputs with + pre-split inputs. + target (Any, optional): Whole-batch target for loss computation. + When passing pre-split inputs, pass targets through + ``target_mbs`` instead. Default: ``None``. + losses (list, optional): Mutable list populated with one loss per + microbatch when this schedule owns the last stage and a + ``loss_fn`` was configured. Default: ``None``. + return_outputs (bool, optional): Whether to merge and return output + chunks on the last stage. Default: ``True``. + loss_kwargs (dict, optional): Extra keyword arguments forwarded to + the configured ``loss_fn``. Default: ``None``. + arg_mbs (list[tuple], optional): Pre-split positional inputs, one + tuple per microbatch. Default: ``None``. + kwarg_mbs (list[dict], optional): Pre-split keyword inputs, one + dict per microbatch. Default: ``None``. + target_mbs (list, optional): Pre-split targets, one entry per + microbatch. Default: ``None``. + \*\*kwargs (Any): Whole-batch keyword root inputs when this rank owns + the first pipeline stage. Do not pass keyword inputs with + pre-split inputs. + + Returns: + Any or None: The merged output from the last stage when this rank + owns the last stage and ``return_outputs=True``; otherwise + ``None``. + + Raises: + RuntimeError: If backward computation is enabled and ``step`` is + called under ``torch.no_grad()``. Use :meth:`eval` for + forward-only execution. + TypeError: If a pre-split microbatch container has the wrong type. + ValueError: If whole-batch and pre-split inputs are mixed, or if a + pre-split container does not have one entry per microbatch. + + Examples:: + + >>> # xdoctest: +SKIP("requires a constructed distributed pipeline schedule") + >>> output = schedule.step(x, mask=mask, target=target) + >>> arg_mbs = [(x0,), (x1,)] + >>> kwarg_mbs = [{"mask": mask0}, {"mask": mask1}] + >>> target_mbs = [target0, target1] + >>> output = schedule.step( + ... arg_mbs=arg_mbs, + ... kwarg_mbs=kwarg_mbs, + ... target_mbs=target_mbs, + ... ) + """ + if ( + self._has_backward + and self._backward_requires_autograd + and not torch.is_grad_enabled() + ): + raise RuntimeError( + "step() requires gradients to be enabled for backward computation; " + "it should not be used under torch.no_grad() context. " + "Please call eval() instead." + ) + + # Set the same has_backward flag for stage object + for stage in self._stages: + stage.has_backward = self._has_backward + + # Clean per iteration + for stage in self._stages: + stage.clear_runtime_states() + + args_split, kwargs_split, targets_split = self._get_microbatch_inputs( + args, + kwargs, + target, + arg_mbs, + kwarg_mbs, + target_mbs, + ) + + # Run microbatches + self._step_microbatches( + args_split, + kwargs_split, + targets_split, + losses, + return_outputs, + loss_kwargs=loss_kwargs, + ) + + # Return merged results per original format + for stage in self._stages: + if stage.is_last and return_outputs: + return self._merge_outputs(stage.output_chunks) + # Does not contain the last stage or we do not return output chunks + return None + + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Operate on the microbatches for looped schedules (multiple stages on each rank). + + TODO: Does not use sorted_batch_isend_irecv(). As a result, this schedule does + not support models with skip connections. + """ + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs, losses) + maybe_first_target = target_mbs[0] if target_mbs is not None else None + self._initialize_stages( + arg_mbs[0], kwarg_mbs[0], maybe_first_target, loss_kwargs + ) + + # Based on the plan in Step 1 created in __init__: + # 2. Perform communication based on the pipeline_order + stage_index_to_stage: dict[int, _PipelineStageBase] = { + stage.stage_index: stage for stage in self._stages + } + + # determine prev_rank and next_rank based on which ranks are next to + # the stages in the pipeline_order + all_prev_ranks: set[int] = set() + all_next_ranks: set[int] = set() + for stage_index in stage_index_to_stage: + # TODO: assumption that stages only communicate from distances of +1/-1 (no skip connections) + if stage_index > 0: + all_prev_ranks.add(self.stage_index_to_group_rank[stage_index - 1]) + if stage_index < self._num_stages - 1: + all_next_ranks.add(self.stage_index_to_group_rank[stage_index + 1]) + # count either full_backward or backward_weight together, to determine when to sync DP grads + backward_counter: Counter[int] = Counter() + for time_step, action in enumerate(self.pipeline_order[self.rank]): + try: + ops: list[dist.P2POp] = [] + if action is not None: + computation_type = action.computation_type + mb_index = action.microbatch_index + stage_index = action.stage_index + if mb_index is None: + raise AssertionError( + "All currently supported action types require valid microbatch_index" + ) + if computation_type == _ComputationType.FORWARD: + # perform forward computation + stage = stage_index_to_stage[stage_index] + output = stage.forward_one_chunk( + mb_index, + arg_mbs[mb_index], + kwarg_mbs[mb_index], + save_forward_output=return_outputs, + ) + self._maybe_compute_loss( + stage, output, target_mbs, mb_index, loss_kwargs + ) + ops.extend(stage.get_fwd_send_ops(mb_index)) + elif computation_type == _ComputationType.FULL_BACKWARD: + # perform backward computation + stage = stage_index_to_stage[stage_index] + loss = self._maybe_get_loss(stage, mb_index) + backward_counter[stage_index] += 1 + last_backward = ( + backward_counter[stage_index] == self._n_microbatches + ) + grad_scale_factor = ( + self._n_microbatches if self.scale_grads else 1 + ) + stage.backward_one_chunk( + mb_index, + loss=loss, + full_backward=True, + last_backward=last_backward, + ) + if last_backward: + stage.scale_grads(grad_scale_factor) + + ops.extend(stage.get_bwd_send_ops(mb_index)) + elif computation_type == _ComputationType.BACKWARD_INPUT: + # perform backward computation + stage = stage_index_to_stage[stage_index] + loss = self._maybe_get_loss(stage, mb_index) + stage.backward_one_chunk( + mb_index, + loss=loss, + full_backward=False, + last_backward=False, + ) + ops.extend(stage.get_bwd_send_ops(mb_index)) + elif computation_type == _ComputationType.BACKWARD_WEIGHT: + # perform weight update + stage = stage_index_to_stage[stage_index] + backward_counter[stage_index] += 1 + last_backward = ( + backward_counter[stage_index] == self._n_microbatches + ) + grad_scale_factor = ( + self._n_microbatches if self.scale_grads else 1 + ) + stage.backward_weight_one_chunk( + mb_index, + last_backward=last_backward, + ) + if last_backward: + stage.scale_grads(grad_scale_factor) + else: + raise ValueError(f"Unknown computation type {computation_type}") + + # Look at the neighboring ranks for this current timestep and determine whether + # this current rank needs to do any recv communication + for prev_rank in all_prev_ranks: + prev_rank_ops = self.pipeline_order[prev_rank] + prev_rank_action = None + if time_step < len(prev_rank_ops): + prev_rank_action = prev_rank_ops[time_step] + if prev_rank_action is not None: + computation_type = prev_rank_action.computation_type + mb_index = prev_rank_action.microbatch_index + stage_index = prev_rank_action.stage_index + if mb_index is None: + raise AssertionError( + "All currently supported action types require valid microbatch_index" + ) + # Only handle sends for the forward from a previous rank + if computation_type == _ComputationType.FORWARD: + # If not the last stage, then receive fwd activations + if stage_index + 1 in stage_index_to_stage: + # TODO: We are assuming that stage will always receive from stage-1 + # however that is not necessarily true of get_fwd_recv_ops + stage = stage_index_to_stage[stage_index + 1] + ops.extend(stage.get_fwd_recv_ops(mb_index)) + elif computation_type in ( + FULL_BACKWARD, + BACKWARD_INPUT, + BACKWARD_WEIGHT, + ): + # Previous rank doing backward has no influence for the current rank forward recv + pass + else: + raise ValueError( + f"Unknown computation type {computation_type}" + ) + for next_rank in all_next_ranks: + next_rank_ops = self.pipeline_order[next_rank] + next_rank_action = None + if time_step < len(next_rank_ops): + next_rank_action = next_rank_ops[time_step] + if next_rank_action is not None: + computation_type = next_rank_action.computation_type + mb_index = next_rank_action.microbatch_index + stage_index = next_rank_action.stage_index + if not (mb_index is not None): + raise AssertionError( + "All currently supported action types require valid microbatch_index" + ) + # Only handle receives for the backwards from a next rank + if computation_type in (FORWARD, BACKWARD_WEIGHT): + # Next rank doing forward or weight update has no influence for the current rank backward recv + pass + elif computation_type in (BACKWARD_INPUT, FULL_BACKWARD): + # If not the first stage, then receive bwd gradients + if stage_index - 1 in stage_index_to_stage: + # TODO: We are assuming that stage will always receive from stage+1 + # however that is not necessarily true of get_bwd_recv_ops + stage = stage_index_to_stage[stage_index - 1] + ops.extend(stage.get_bwd_recv_ops(mb_index)) + else: + raise ValueError( + f"Unknown computation type {computation_type}" + ) + + # do the communication + _wait_batch_p2p(_batch_p2p(ops)) + except Exception as e: + logger.error( + "[Rank %s] pipeline schedule %s caught the following exception '%s' \ +at time_step %s when running action %s", + self.rank, + self.__class__.__name__, + e, + time_step, + action, + ) + logger.error( + "%s", + _format_pipeline_order( + self.pipeline_order, error_step_number=time_step + ), + ) + raise e + # Return losses if there is a container passed in + self._update_losses(self._stages, losses) + + +@dataclass +class _PipelineContext: + """Context passed to custom functions during pipeline execution.""" + + schedule_ref: _PipelineSchedule + arg_mbs: list[tuple] | None = None + kwarg_mbs: list[dict] | None = None + target_mbs: list | None = None + losses: list | None = None + + +class _CustomFunctionProtocol(Protocol): + def __call__(self, action: _Action, ctx: _PipelineContext) -> None: ... + + +class _PipelineScheduleRuntime(PipelineScheduleMulti): + """ + Provides a simple runtime that requires a 'schedule IR' including specified communication operations. + + Can be instantiated directly by creating _PipelineScheduleRuntime and calling load_csv, or can be + subclassed and the subclass can be responsible for creating a schedule IR. + """ + + def __init__(self, *args, **kwargs): + self._defer_pp_recv: bool = kwargs.pop("defer_pp_recv", False) + super().__init__(*args, **kwargs) + # Action to custom function mapping + self._comp_type_to_function_map: dict[_ComputationType, Callable] = {} + # count either full_backward or backward_weight together, to determine when to sync DP grads + self.backward_counter: Counter[int] = Counter() + + # recv ops indexed by (stage_idx, mb_idx) need to be waited on before use + self.bwd_recv_ops: dict[tuple[int, int], list[dist.Work]] = {} + self.fwd_recv_ops: dict[tuple[int, int], list[dist.Work]] = {} + + # we track which stages are 'active' when used with FSDP, and wait on unshard ops before computing on stages + self.unshard_ops: dict[int, list[UnshardHandle]] = defaultdict(list) + self.unsharded_stages = set() + + def register_custom_function( + self, + computation_type: _ComputationType, + custom_function: _CustomFunctionProtocol, + ) -> None: + """ + Register a custom function to be executed for a specific computation type. + + Args: + computation_type: The computation type for which to register the custom function + custom_function: The function to execute when this computation type is encountered. + Must have signature: (action: _Action, ctx: _PipelineContext) -> None + """ + # Ensure that the computation type is valid + if computation_type not in ( + FORWARD, + FULL_BACKWARD, + BACKWARD_INPUT, + BACKWARD_WEIGHT, + OVERLAP_F_B, + UNSHARD, + RESHARD, + REDUCE_GRAD, + ): + raise ValueError( + f"Invalid computation type {computation_type}. Only FORWARD, FULL_BACKWARD, \ + BACKWARD_INPUT, BACKWARD_WEIGHT, OVERLAP_F_B, UNSHARD, RESHARD and REDUCE_GRAD are supported." + ) + + # Check if computation_type is already registered + if computation_type in self._comp_type_to_function_map: + logger.warning( + "Computation type %s is already registered. " + "Overwriting the existing custom function.", + computation_type, + ) + + self._comp_type_to_function_map[computation_type] = custom_function + + def _prepare_schedule_with_comms( + self, + actions: dict[int, list[_Action | None]], + format: Literal["compute_only", "compute_comms"] = "compute_only", + ): + """ + Given an in-memory representation for a simple compute-only schedule, lower it to a complex schedule including + communication actions. Stores the schedule in self, and must be called before running step_mo() + """ + # validate the provided actions are valid and overrides the default stage_index_to_group_rank + super()._validate_and_set_stage_mapping(actions) + + self.pipeline_order_with_comms: dict[int, list[_Action]] = {} + if format == "compute_comms": + for rank in actions: + self.pipeline_order_with_comms[rank] = [] + for action in actions[rank]: + if action is None: + raise AssertionError( + f"Expected action to be not None, got {type(action)}" + ) + self.pipeline_order_with_comms[rank].append(action) + # TODO what level of validation should we offer for compute+comms schedule? + elif format == "compute_only": + # Validate that the schedule does not have comms already added to it + for rank, action_list in actions.items(): + for i, action in enumerate(action_list): + if action is not None: + if not action.is_compute_op: + raise ValueError( + f"Expected compute-only schedule but found communication action " + f"'{action}' at rank {rank}, position {i}. " + f"Communication actions (e.g. SEND_F, RECV_F, etc.) " + f"should not be present when format='compute_only'." + ) + + # Perform schedule lowering + for rank in actions: + self.pipeline_order_with_comms[rank] = _add_unshard_reshard( + actions[rank] + ) + self.pipeline_order_with_comms[rank] = _add_reduce_grad( # type: ignore[assignment] + self.pipeline_order_with_comms[rank], # type: ignore[arg-type] + self._n_microbatches, + ) + + self.pipeline_order_with_comms = _add_send_recv( + self.pipeline_order_with_comms, + stage_to_rank=lambda s: self.stage_index_to_group_rank[s], + num_stages=self._num_stages, + ) + + if self._defer_pp_recv: + self.pipeline_order_with_comms = _defer_recv_ops( + self.pipeline_order_with_comms, + stage_to_rank=lambda s: self.stage_index_to_group_rank[s], + ) + else: + raise NotImplementedError(f"{format=} is not implemented") + + def _load_csv( + self, + filename: str, + format: Literal["compute_only", "compute_comms"] = "compute_only", + ): + """Loads a csv in simple format and then lowers it to include communication actions + + format must be either "compute_only" or "compute_comms". If compute_only, the lowering passes + will automatically be run to generate a compute_comms schedule. + """ + if format == "compute_only": + # this will populate self.pipeline_order + super()._load_csv(filename) + # this will populate self.pipeline_order_with_comms + self._prepare_schedule_with_comms(self.pipeline_order) + elif format == "compute_comms": + actions = {} + with open(filename, newline="") as csvfile: + reader = csv.reader(csvfile) + for rank, row in enumerate(reader): + actions[rank] = [_Action.from_str(s) for s in row] + self._prepare_schedule_with_comms(actions, format=format) + else: + raise NotImplementedError(f"{format=} is not implemented") + + def _dump_csv( + self, + filename: str, + format: Literal["compute_only", "compute_comms"] = "compute_comms", + ): + """Dump a CSV representation of the schedule into a file with the provided filename.""" + if format == "compute_only": + if self.pipeline_order is None: + raise AssertionError("Compute only schedule must be available") + with open(filename, "w", newline="") as csvfile: + writer = csv.writer(csvfile) + for rank in self.pipeline_order: + writer.writerow(self.pipeline_order[rank]) + elif format == "compute_comms": + if self.pipeline_order_with_comms is None: + raise AssertionError( + "Must initialize compute_comms schedule before dump_csv" + ) + with open(filename, "w", newline="") as csvfile: + writer = csv.writer(csvfile) + for rank in self.pipeline_order_with_comms: + writer.writerow(self.pipeline_order_with_comms[rank]) + + def _simulate(self): + return _simulate_comms_compute( + self.pipeline_order_with_comms, + lambda s: self.stage_index_to_group_rank[s], + self._num_stages, + ) + + def _assert_unsharded(self, stage: _PipelineStageBase): + """If an unshard is active for `stage_idx`, wait() it and mark `stage_idx` unshared.""" + stage_uses_fsdp = isinstance(stage.submod, FSDPModule) + if stage_uses_fsdp: + stage_idx = stage.stage_index + if stage_idx in self.unshard_ops: + for op in self.unshard_ops[stage_idx]: + op.wait() + del self.unshard_ops[stage_idx] + self.unsharded_stages.add(stage_idx) + if stage_idx not in self.unsharded_stages: + raise AssertionError(f"Attempted to compute on sharded {stage_idx=}") + + def _step_microbatches( + self, + arg_mbs: list | None = None, + kwarg_mbs: list | None = None, + target_mbs: list | None = None, + losses: list | None = None, + return_outputs: bool = True, + loss_kwargs: dict[str, Any] | None = None, + ): + """ + Operate on the microbatches for looped schedules (multiple stages on each rank). + + TODO: Does not use sorted_batch_isend_irecv(). As a result, this schedule does + not support models with skip connections. + """ + arg_mbs, kwarg_mbs = self._check_inputs(arg_mbs, kwarg_mbs, target_mbs, losses) + maybe_first_target = target_mbs[0] if target_mbs is not None else None + self._initialize_stages( + arg_mbs[0], kwarg_mbs[0], maybe_first_target, loss_kwargs + ) + + # Based on the plan in Step 1 created in __init__: + # 2. Perform communication based on the pipeline_order + stage_index_to_stage: dict[int, _PipelineStageBase] = { + stage.stage_index: stage for stage in self._stages + } + + if self.pipeline_order_with_comms is None: + raise AssertionError( + "Must call _prepare_schedule_with_comms() before calling _step_microbatches()" + ) + + # send ops should be waited on before step() exists, mainly for hygiene + send_ops: list[list[dist.Work]] = [] + + def _perform_action(action: _Action) -> None: + comp_type = action.computation_type + mb_index: int = ( + action.microbatch_index if action.microbatch_index is not None else -1 + ) + if not ( + mb_index >= 0 + or comp_type + in ( + UNSHARD, + RESHARD, + REDUCE_GRAD, + ) + ): + raise AssertionError(f"{action=} missing mb_index") + stage_idx = action.stage_index + stage = stage_index_to_stage[stage_idx] + stage_uses_fsdp = isinstance(stage.submod, FSDPModule) + # see [Note: V-schedule special case] + is_next_stage_on_this_rank = stage_idx + 1 in stage_index_to_stage + is_prev_stage_on_this_rank = stage_idx - 1 in stage_index_to_stage + + # TODO(whc) it's not actually safe to use _batch_p2p here in the uncommon case the model has skip-connections, + # since we do not want to batch up ops between more than a pair of ranks. _sorted_batch_p2p would be + # safe to use instead. + # However, I was wondering if I should avoid calling batched operators at all in the case that there is + # only one operator per batch. I could iterate through the 'fwd_send_ops' one by one and run them. + if comp_type == SEND_F: + send_ops.append(_batch_p2p(stage.get_fwd_send_ops(mb_index))) + elif comp_type == SEND_B: + send_ops.append(_batch_p2p(stage.get_bwd_send_ops(mb_index))) + elif comp_type == RECV_F: + if (stage_idx, mb_index) in self.fwd_recv_ops: + raise AssertionError( + f"Recv twice for {stage_idx=} {mb_index=} without executing forward" + ) + self.fwd_recv_ops[(stage_idx, mb_index)] = _batch_p2p( + stage.get_fwd_recv_ops(mb_index) + ) + elif comp_type == RECV_B: + if (stage_idx, mb_index) in self.bwd_recv_ops: + raise AssertionError( + f"Recv twice for {stage_idx=} {mb_index=} without executing backward" + ) + self.bwd_recv_ops[(stage_idx, mb_index)] = _batch_p2p( + stage.get_bwd_recv_ops(mb_index) + ) + elif comp_type == UNSHARD: + if stage_uses_fsdp: + if not ( + stage_idx not in self.unsharded_stages + and stage_idx not in self.unshard_ops + ): + raise AssertionError(f"Unsharding the same {stage_idx=} twice") + for submodule in stage.submod.modules(): + if not isinstance(submodule, FSDPModule): + continue + handle = cast(UnshardHandle, submodule.unshard(async_op=True)) + self.unshard_ops[stage_idx].append(handle) + elif comp_type == RESHARD: + if stage_uses_fsdp: + if stage_idx not in self.unsharded_stages: + raise AssertionError( + f"Resharding {stage_idx=} without unsharding" + ) + if stage_idx in self.unshard_ops: + raise AssertionError( + f"Resharding {stage_idx=} before finishing unshard" + ) + for submodule in stage.submod.modules(): + if not isinstance(submodule, FSDPModule): + continue + submodule.reshard() + self.unsharded_stages.remove(stage_idx) + elif comp_type == FORWARD: + self._assert_unsharded(stage) + + if ( + not stage.is_first + # no recv op expected for V-schedule special case (see [Note: V-schedule special case]) + and not is_prev_stage_on_this_rank + ): + if (stage_idx, mb_index) not in self.fwd_recv_ops: + raise AssertionError( + f"Computing {action=} before receiving input" + ) + _wait_batch_p2p(self.fwd_recv_ops.pop((stage_idx, mb_index))) + + output = stage.forward_one_chunk( + mb_index, + arg_mbs[mb_index], # type: ignore[index] + kwarg_mbs[mb_index], # type: ignore[index] + save_forward_output=return_outputs, + ) + self._maybe_compute_loss( + stage, output, target_mbs, mb_index, loss_kwargs + ) + + # SEND/RECV op are avoided for special case with 2 adjacent stages on same rank + # see [Note: V-schedule special case] + if is_next_stage_on_this_rank: + stage_index_to_stage[stage_idx + 1].set_local_fwd_input( + output, mb_index + ) + + elif comp_type == FULL_BACKWARD: + self._assert_unsharded(stage) + + if ( + not stage.is_last + # no recv op expected for V-schedule special case (see [Note: V-schedule special case]) + and not is_next_stage_on_this_rank + ): + if (stage_idx, mb_index) not in self.bwd_recv_ops: + raise AssertionError( + f"Attempted to run compute {action=} before receiving input" + ) + _wait_batch_p2p(self.bwd_recv_ops.pop((stage_idx, mb_index))) + loss = self._maybe_get_loss(stage, mb_index) + self.backward_counter[stage_idx] += 1 + last_backward = self.backward_counter[stage_idx] == self._n_microbatches + stage.backward_one_chunk( + mb_index, + loss=loss, + full_backward=True, + last_backward=last_backward, + ) + # SEND/RECV op are avoided for special case with 2 adjacent stages on same rank + # see [Note: V-schedule special case] + if is_prev_stage_on_this_rank: + stage_index_to_stage[stage_idx - 1].set_local_bwd_input( + stage.get_local_bwd_output(mb_index), mb_index + ) + elif comp_type == BACKWARD_INPUT: + self._assert_unsharded(stage) + + if not stage.is_last and not is_next_stage_on_this_rank: + if (stage_idx, mb_index) not in self.bwd_recv_ops: + raise AssertionError( + f"Attempted to run compute {action=} before receiving input" + ) + _wait_batch_p2p(self.bwd_recv_ops.pop((stage_idx, mb_index))) + loss = self._maybe_get_loss(stage, mb_index) + stage.backward_one_chunk( + mb_index, + loss=loss, + full_backward=False, + last_backward=False, + ) + # SEND/RECV op are avoided for special case with 2 adjacent stages on same rank + # see [Note: V-schedule special case] + if is_prev_stage_on_this_rank: + stage_index_to_stage[stage_idx - 1].set_local_bwd_input( + stage.get_local_bwd_output(mb_index), mb_index + ) + elif comp_type == BACKWARD_WEIGHT: + self._assert_unsharded(stage) + self.backward_counter[stage_idx] += 1 + last_backward = self.backward_counter[stage_idx] == self._n_microbatches + stage.backward_weight_one_chunk( + mb_index, + last_backward=last_backward, + ) + elif comp_type == REDUCE_GRAD: + grad_scale_factor = self._n_microbatches if self.scale_grads else 1 + stage.perform_reduce_grad(grad_scale_factor) + else: + raise ValueError(f"{action=} is unknown or unsupported") + + # count either full_backward or backward_weight together, to determine when to sync DP grads + self.backward_counter.clear() + for time_step, action in enumerate(self.pipeline_order_with_comms[self.rank]): + logger.debug( + "_PipelineScheduleRuntime running time_step %d, action %s", + time_step, + action, + ) + try: + with record_function(_get_profiler_function_name(action)): + if action.computation_type in self._comp_type_to_function_map: + ctx = _PipelineContext( + self, + arg_mbs, + kwarg_mbs, + target_mbs, + losses, + ) + self._comp_type_to_function_map[action.computation_type]( + action, ctx + ) + elif action.computation_type == OVERLAP_F_B: + if action.sub_actions is None: + raise AssertionError("sub_actions must be set") + for sub_a in action.sub_actions: + _perform_action(sub_a) + else: + _perform_action(action) + except Exception as e: + logger.error( + "_PipelineScheduleRuntime caught exception at step %s when running action %s. Full Schedule:", + time_step, + action, + ) + logger.error( + _format_pipeline_order( + self.pipeline_order_with_comms, # type: ignore[arg-type] + error_step_number=time_step, + ) + ) + raise e + + # Mostly these operations should have finished long ago, but there isn't an obvious time when to wait for them + while send_ops: + _wait_batch_p2p(send_ops.pop()) + + if len(self.unshard_ops) != 0: + raise AssertionError("Unused unshard operations") + + # Return losses if there is a container passed in + self._update_losses(self._stages, losses) + + +class ScheduleLoopedBFS(_PipelineScheduleRuntime): + """ + Breadth-First Pipeline Parallelism. + See https://arxiv.org/abs/2211.05953 for details. + Similar to Interleaved 1F1B, Looped BFS supports multiple stages per rank. + What is different is that when microbatches are ready for multiple local + stages, Loops BFS will prioritizes the earlier stage, running all available + microbatches at once. + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | _Loss | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + defer_pp_recv: bool = False, + ): + super().__init__( + stages=stages, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + backward_requires_autograd=backward_requires_autograd, + defer_pp_recv=defer_pp_recv, + ) + + # 1. Create the pipeline_order (all ranks do this calculation) + # This will be used to keep track of the current state of the entire pipeline + # pipeline_order[rank] = [Action(computation_type, microbatch_index, stage_index), ...] + self.pipeline_order: dict[int, list[_Action | None]] = {} + # ======================================================================== + for rank in range(self.pp_group_size): + rank_ops = self._calculate_single_rank_operations(rank) + self.pipeline_order[rank] = rank_ops + + # Initialize the pipeline order with communication necessary to run with _PipelineScheduleRuntime + self._prepare_schedule_with_comms(self.pipeline_order) + + def _calculate_single_rank_operations(self, rank): + n_local_stages = len(self._stages) + stage_indices = range( + rank, self.pp_group_size * n_local_stages, self.pp_group_size + ) + + # Store the list of operations used for that rank + # Pre-padding, rank starts with no-ops based on the warmup. + rank_ops: list[_Action | None] = [None for _ in range(rank)] + + for stage_index in stage_indices: + rank_ops.extend( + _Action(stage_index, _ComputationType.FORWARD, mb_index) + for mb_index in range(self._n_microbatches) + ) + + # wait for the first backward to trickle up + # which is 2 for every hop away + post_warmup_ops = 2 * (self.pp_group_size - 1 - rank) + rank_ops.extend([None] * post_warmup_ops) + + for stage_index in reversed(stage_indices): + rank_ops.extend( + _Action(stage_index, _ComputationType.FULL_BACKWARD, mb_index) + for mb_index in reversed(range(self._n_microbatches)) + ) + return rank_ops + + +def _get_1f1b_rank_ops( + n_local_stages, + pp_group_size, + warmup_ops, + fwd_bwd_ops, + cooldown_ops, + rank, + forward_stage_index, + backward_stage_index, + num_1f1b_microbatches=0, + enable_zero_bubble=False, +): + # All stages start with handling microbatch 0 + fwd_stage_mb_index: dict[int, int] = defaultdict(int) + bwd_stage_mb_index: dict[int, int] = defaultdict(int) + weight_stage_mb_index: dict[int, int] = defaultdict(int) + + # Store the list of operations used for that rank + # Pre-padding, rank starts with no-ops based on the warmup. + rank_ops: list[_Action | None] = [None for _ in range(rank)] + # These are used to calculate the number of slots to fill with no-ops, to account for the delay in warmup + # when we want to wait for the backward to trickle back up and start 1f1b to align all ranks. + # Formula: + # pre-padding + warmup_ops + post_warmup_ops = earliest time step of first backward + # post_warmup_ops = [earliest time step of first backward] - (warmup_ops + pre-padding) + # earliest time step of first backward = [local_stages * group_size + 2 * (group_size - 1 - rank)] + # warmup_ops = calculated above + post_warmup_ops = ( + n_local_stages * pp_group_size + 2 * (pp_group_size - 1 - rank) + ) - (warmup_ops + rank) + + if enable_zero_bubble: + post_warmup_ops = pp_group_size - rank - 1 + + total_ops = warmup_ops + fwd_bwd_ops + cooldown_ops + + backward_op_ids = [] + weight_op_count = 0 + + FULL_BACKWARD_OR_BACKWARD_INPUT = ( + BACKWARD_INPUT if enable_zero_bubble else FULL_BACKWARD + ) + + for op in range(total_ops): + # Warmup phase + if op < warmup_ops: + fwd_stage_index = forward_stage_index(op) + # This will assign the current microbatch index and update it as well + fwd_stage_mb_index[fwd_stage_index] = ( + mb_index := fwd_stage_mb_index[fwd_stage_index] + ) + 1 + rank_ops.append( + _Action(fwd_stage_index, _ComputationType.FORWARD, mb_index) + ) + if op == warmup_ops - 1: + # This is the last step in the warmup phase, so we need to wait for the backward to trickle back up + rank_ops.extend([None] * post_warmup_ops) + # 1F1B Phase (forward and backward) + elif warmup_ops <= op < warmup_ops + fwd_bwd_ops: + fwd_stage_index = forward_stage_index(op) + fwd_stage_mb_index[fwd_stage_index] = ( + fwd_mb_index := fwd_stage_mb_index[fwd_stage_index] + ) + 1 + rank_ops.append( + _Action(fwd_stage_index, _ComputationType.FORWARD, fwd_mb_index) + ) + bwd_stage_index = backward_stage_index(op) + bwd_stage_mb_index[bwd_stage_index] = ( + bwd_mb_index := bwd_stage_mb_index[bwd_stage_index] + ) + 1 + rank_ops.append( + _Action(bwd_stage_index, FULL_BACKWARD_OR_BACKWARD_INPUT, bwd_mb_index) + ) + backward_op_ids.append(op) + + if enable_zero_bubble and op - warmup_ops >= num_1f1b_microbatches: + weight_stage_index = backward_stage_index( + backward_op_ids[weight_op_count] + ) + weight_stage_mb_index[weight_stage_index] = ( + weight_mb_index := weight_stage_mb_index[weight_stage_index] + ) + 1 + rank_ops.append( + _Action( + weight_stage_index, + _ComputationType.BACKWARD_WEIGHT, + weight_mb_index, + ) + ) + weight_op_count += 1 + # Cooldown phase + else: + # During cooldown phase, we need steps to align with 1f1b happening in other ranks + # TODO: we don't need to always append, after all 1f1b are finished we can stop appending None + if not enable_zero_bubble: + rank_ops.append(None) + + bwd_stage_index = backward_stage_index(op) + bwd_stage_mb_index[bwd_stage_index] = ( + bwd_mb_index := bwd_stage_mb_index[bwd_stage_index] + ) + 1 + rank_ops.append( + _Action(bwd_stage_index, FULL_BACKWARD_OR_BACKWARD_INPUT, bwd_mb_index) + ) + backward_op_ids.append(op) + + if enable_zero_bubble and op - warmup_ops >= num_1f1b_microbatches: + weight_stage_index = backward_stage_index( + backward_op_ids[weight_op_count] + ) + weight_stage_mb_index[weight_stage_index] = ( + weight_mb_index := weight_stage_mb_index[weight_stage_index] + ) + 1 + rank_ops.append( + _Action( + weight_stage_index, + _ComputationType.BACKWARD_WEIGHT, + weight_mb_index, + ) + ) + weight_op_count += 1 + + while enable_zero_bubble and weight_op_count < len(backward_op_ids): + weight_stage_index = backward_stage_index(backward_op_ids[weight_op_count]) + weight_stage_mb_index[weight_stage_index] = ( + weight_mb_index := weight_stage_mb_index[weight_stage_index] + ) + 1 + rank_ops.append( + _Action( + weight_stage_index, _ComputationType.BACKWARD_WEIGHT, weight_mb_index + ) + ) + weight_op_count += 1 + + return rank_ops + + +def _get_warmup_ops( + rank: int, + n_local_stages: int, + microbatches_per_round: int, + pp_group_size: int, + n_microbatches: int, + multiply_factor: int = 2, +) -> int: + """ + Calculate the number of warmup operations for interleaved schedules. + """ + # Warmup operations for last stage + warmups_ops_last_stage = (n_local_stages - 1) * microbatches_per_round + # Increment warmup operations by multiply_factor for each hop away from the last stage + warmup_ops = warmups_ops_last_stage + multiply_factor * ((pp_group_size - 1) - rank) + # We cannot have more warmup operations than there are number of microbatches, so cap it there + return min(warmup_ops, n_microbatches * n_local_stages) + + +class ScheduleInterleaved1F1B(_PipelineScheduleRuntime): + """ + The Interleaved 1F1B schedule. + See https://arxiv.org/pdf/2104.04473 for details. + Will perform one forward and one backward on the microbatches in steady + state and supports multiple stages per rank. When microbatches are ready for + multiple local stages, Interleaved 1F1B prioritizes the earlier microbatch + (also called "depth first"). + + This schedule is mostly similar to the original paper. + It differs by being relaxing the requirement of num_microbatch % pp_size == 0. + Using the flex_pp schedule, we will have num_rounds = max(1, n_microbatches // pp_group_size) and + it works as long as n_microbatches % num_rounds is 0. As a few examples, support + + 1. pp_group_size = 4, n_microbatches = 10. We will have num_rounds = 2 and n_microbatches % 2 is 0. + 2. pp_group_size = 4, n_microbatches = 3. We will have num_rounds = 1 and n_microbatches % 1 is 0. + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + defer_pp_recv: bool = False, + ): + self.pp_group_size = stages[0].group_size + super().__init__( + stages=stages, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + backward_requires_autograd=backward_requires_autograd, + defer_pp_recv=defer_pp_recv, + ) + self.n_local_stages = len(stages) + self.rank = stages[0].group_rank + self.number_of_rounds = max(1, n_microbatches // self.pp_group_size) + self.microbatches_per_round = n_microbatches // self.number_of_rounds + if n_microbatches % self.number_of_rounds != 0: + raise ValueError( + "Interleaved 1F1B requires the number of microbatches to be a " + f"multiple of the number of rounds ({self.number_of_rounds}), " + f"but got {n_microbatches}." + ) + # 1. Create the pipeline_order (all ranks do this calculation) + # This will be used to keep track of the current state of the entire pipeline + # pipeline_order[rank] = [Action(computation_type, microbatch_index, stage_index), ...] + self.pipeline_order: dict[int, list[_Action | None]] = {} + for rank in range(self.pp_group_size): + rank_ops = self._calculate_single_rank_operations(rank) + self.pipeline_order[rank] = rank_ops + + # Initialize the pipeline order with communication necessary to run with _PipelineScheduleRuntime + self._prepare_schedule_with_comms(self.pipeline_order) + + def _calculate_single_rank_operations(self, rank) -> list[_Action | None]: + warmup_ops = _get_warmup_ops( + rank, + self.n_local_stages, + self.microbatches_per_round, + self.pp_group_size, + self._n_microbatches, + multiply_factor=2, + ) + microbatch_ops = self.n_local_stages * self._n_microbatches + # fwd_bwd_ops should encompass the remaining forwards + fwd_bwd_ops = microbatch_ops - warmup_ops + # cooldown_ops should encompass the remaining backwards + cooldown_ops = microbatch_ops - fwd_bwd_ops + # total ops encompass both forward and backward ops + total_ops = warmup_ops + fwd_bwd_ops + cooldown_ops + # warmup_ops + fwd_bwd_ops * 2 + cooldown_ops == microbatch_ops * 2 + logger.debug( + "rank %s, warmup_ops %s, 1f1b %s, cooldown_ops %s total_ops %s", + rank, + warmup_ops, + fwd_bwd_ops, + cooldown_ops, + total_ops, + ) + + # Calculates the stage index based on step and pp_group_size + def forward_stage_index(step): + # Get the local index from 0 to n_local_stages-1 + local_index = (step // self.microbatches_per_round) % self.n_local_stages + return (local_index * self.pp_group_size) + rank + + def backward_stage_index(step): + local_index = ( + self.n_local_stages + - 1 + - ((step - warmup_ops) // self.microbatches_per_round) + % self.n_local_stages + ) + return (local_index * self.pp_group_size) + rank + + return _get_1f1b_rank_ops( + self.n_local_stages, + self.pp_group_size, + warmup_ops, + fwd_bwd_ops, + cooldown_ops, + rank, + forward_stage_index, + backward_stage_index, + ) + + +class ScheduleInterleavedZeroBubble(_PipelineScheduleRuntime): + """ + The Interleaved Zero Bubble schedule. + See https://arxiv.org/pdf/2401.10241 for details. + Will perform one forward and one backward on inputs for the microbatches in steady + state and supports multiple stages per rank. Uses the backward for weights to fill in + the pipeline bubble. + + In particular this is implementing the ZB1P schedule in the paper. + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + defer_pp_recv: bool = False, + ): + # TODO: we don't support input/weight backward split with torch.compile + _check_torch_compile_compatibility(stages, self.__class__.__name__) + self.pp_group_size = stages[0].group_size + super().__init__( + stages=stages, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + backward_requires_autograd=backward_requires_autograd, + defer_pp_recv=defer_pp_recv, + ) + self.n_local_stages = len(stages) + self.rank = stages[0].group_rank + self.number_of_rounds = max(1, n_microbatches // self.pp_group_size) + self.microbatches_per_round = n_microbatches // self.number_of_rounds + if n_microbatches % self.number_of_rounds != 0: + raise ValueError( + "Zero bubble requires the number of microbatches to be a " + f"multiple of the number of rounds ({self.number_of_rounds}), " + f"but got {n_microbatches}." + ) + # 1. Create the pipeline_order (all ranks do this calculation) + # This will be used to keep track of the current state of the entire pipeline + # pipeline_order[rank] = [Action(computation_type, microbatch_index, stage_index), ...] + self.pipeline_order: dict[int, list[_Action | None]] = {} + for rank in range(self.pp_group_size): + rank_ops = self._calculate_single_rank_operations(rank) + self.pipeline_order[rank] = rank_ops + + # This function add bubbles to the generated schedule based on dependencies of actions + # Note that the ZB1P schedule will not require bubbles to be manually added and it is + # only useful when n_microbatches <= microbatches_per_round + self.pipeline_order = self._add_bubbles_to_actions( + self.n_local_stages * self.pp_group_size, + ) + + # Initialize the pipeline order with communication necessary to run with _PipelineScheduleRuntime + self._prepare_schedule_with_comms(self.pipeline_order) + + def _calculate_single_rank_operations(self, rank) -> list[_Action | None]: + warmup_ops = _get_warmup_ops( + rank, + self.n_local_stages, + self.microbatches_per_round, + self.pp_group_size, + self._n_microbatches, + multiply_factor=1, + ) + microbatch_ops = self.n_local_stages * self._n_microbatches + # fwd_bwd_ops should encompass the remaining forwards + fwd_bwd_ops = microbatch_ops - warmup_ops + # cooldown_ops should encompass the remaining backwards + cooldown_ops = microbatch_ops - fwd_bwd_ops + # total ops encompass both forward and backward ops + total_ops = warmup_ops + fwd_bwd_ops + cooldown_ops + # warmup_ops + fwd_bwd_ops * 2 + cooldown_ops == microbatch_ops * 2 + logger.debug( + "rank %s, warmup_ops %s, 1f1b %s, cooldown_ops %s total_ops %s", + rank, + warmup_ops, + fwd_bwd_ops, + cooldown_ops, + total_ops, + ) + + # Calculates the stage index based on step and pp_group_size + + def forward_stage_index(step): + # Get the local index from 0 to n_local_stages-1 + local_index = (step // self.microbatches_per_round) % self.n_local_stages + return (local_index * self.pp_group_size) + rank + + def backward_stage_index(step): + local_index = ( + self.n_local_stages + - 1 + - ((step - warmup_ops) // self.microbatches_per_round) + % self.n_local_stages + ) + return (local_index * self.pp_group_size) + rank + + num_1f1b_microbatches = rank + + return _get_1f1b_rank_ops( + self.n_local_stages, + self.pp_group_size, + warmup_ops, + fwd_bwd_ops, + cooldown_ops, + rank, + forward_stage_index, + backward_stage_index, + num_1f1b_microbatches, + enable_zero_bubble=True, + ) + + def _add_bubbles_to_actions(self, num_stages_global): + actions = self.pipeline_order + + def need_bubble(stage, op, microbatch, num_stages_global, seen_ops): + if op == _ComputationType.FORWARD: + if stage != 0 and (stage - 1, op, microbatch) not in seen_ops: + return True + elif op == _ComputationType.FULL_BACKWARD: + if stage == num_stages_global - 1: + return (stage, _ComputationType.FORWARD, microbatch) not in seen_ops + return (stage + 1, op, microbatch) not in seen_ops + return False + + seen_ops: set[tuple[int, _ComputationType, int]] = set() + result: dict[int, list[_Action | None]] = {} + next_pointer: dict[int, int] = {} + bubbles_added: dict[int, int] = {} + total_bubbles_added = 0 + + for rank in range(self.pp_group_size): + result[rank] = [] + next_pointer[rank] = 0 + bubbles_added[rank] = 0 + + while True: + should_stop = True + + temp_seen_ops: set[tuple[int, _ComputationType, int]] = set() + + for rank in range(self.pp_group_size): + timestamp = next_pointer[rank] + if timestamp >= len(actions[rank]): + continue + + should_stop = False + + if actions[rank][timestamp] is not None: + temp_action = actions[rank][timestamp] + if temp_action is None: + raise AssertionError( + f"Expected temp_action to be not None, got {type(temp_action)}" + ) + stage_index, op, microbatch, _ = temp_action + if not need_bubble( + stage_index, op, microbatch, num_stages_global, seen_ops + ): + result[rank].append(actions[rank][timestamp]) + if microbatch is not None: + temp_seen_ops.add((stage_index, op, microbatch)) + next_pointer[rank] += 1 + else: + result[rank].append(None) + bubbles_added[rank] += 1 + else: + next_pointer[rank] += 1 + result[rank].append(None) + + seen_ops.update(temp_seen_ops) + if should_stop: + break + + if total_bubbles_added > 0: + logger.warning( + "Non zero bubbles added: total_bubbles_added=%s bubbles_added=%s", + total_bubbles_added, + bubbles_added, + ) + return result + + +class ScheduleZBVZeroBubble(_PipelineScheduleRuntime): + """ + The Zero Bubble schedule (ZBV variant). + See https://arxiv.org/pdf/2401.10241 Section 6 for details. + + This schedules requires exactly two stages per rank. + + This schedule will perform one forward and one backward on inputs for the microbatches in steady + state and supports multiple stages per rank. Uses backward with respect to weights to fill in + the pipeline bubble. + + This ZB-V schedule would have the "zero bubble" property only if time forward == time backward input == time backward weights. + In practice, this is not likely true for real models so alternatively + a greedy scheduler could be implemented for unequal/unbalanced time. + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + defer_pp_recv: bool = False, + ): + # TODO: we don't support input/weight backward split with torch.compile + _check_torch_compile_compatibility(stages, self.__class__.__name__) + self.pp_group_size = stages[0].group_size + super().__init__( + stages=stages, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + backward_requires_autograd=backward_requires_autograd, + defer_pp_recv=defer_pp_recv, + ) + self.stage_index_to_group_rank = generate_stage_to_rank_mapping( + self.pp_group_size, self._num_stages, style="v" + ) + for stage in self._stages: + stage.stage_index_to_group_rank = self.stage_index_to_group_rank + + self.n_local_stages = len(stages) + if self.n_local_stages != 2: + raise ValueError( + "ZBV requires exactly 2 stages per rank, but got " + f"{self.n_local_stages}." + ) + + self.rank = stages[0].group_rank + self.num_stages = stages[0].num_stages + + # 1. Create the pipeline_order (all ranks do this calculation) + # This will be used to keep track of the current state of the entire pipeline + # pipeline_order[rank] = [Action(computation_type, microbatch_index, stage_index), ...] + self.pipeline_order: dict[int, list[_Action | None]] = {} + for rank in range(self.pp_group_size): + rank_ops = self._calculate_single_rank_operations(rank) + self.pipeline_order[rank] = rank_ops + + # Initialize the pipeline order with communication necessary to run with _PipelineScheduleRuntime + self._prepare_schedule_with_comms(self.pipeline_order) + + def _calculate_single_rank_operations(self, rank) -> list[_Action | None]: + # max(2 * self.pp_group_size - 1, ...) ensure the number of microbatches is at least + # as large of the number of microbatches needed to fully utilize the pipeline + n_micro = max(2 * self.pp_group_size - 1, self._n_microbatches) + rank_ops: list[_Action | None] = [None for _ in range(rank)] + + # Forward and backward action counts for stage chunk 0 and chunk 1 + f0_cnt, f1_cnt, b0_cnt, b1_cnt = 0, 0, 0, 0 + # warm-up phase + warmup_n1 = 2 * (self.pp_group_size - rank) - 1 + stage_id_chunk0 = rank + stage_id_chunk1 = self.num_stages - 1 - rank + + for _ in range(warmup_n1): + rank_ops.append( + _Action(stage_id_chunk0, computation_type=F, microbatch_index=f0_cnt) + ) + f0_cnt += 1 + warmup_n2 = rank + for _ in range(warmup_n2): + rank_ops.append( + _Action(stage_id_chunk1, computation_type=F, microbatch_index=f1_cnt) + ) + f1_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk0, computation_type=F, microbatch_index=f0_cnt) + ) + f0_cnt += 1 + warmup_n3 = self.pp_group_size - rank + for _ in range(warmup_n3): + rank_ops.append( + _Action(stage_id_chunk1, computation_type=F, microbatch_index=f1_cnt) + ) + f1_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk1, computation_type=I, microbatch_index=b1_cnt) + ) + rank_ops.append( + _Action(stage_id_chunk1, computation_type=W, microbatch_index=b1_cnt) + ) + b1_cnt += 1 + # stable phase + while f1_cnt < f0_cnt or f0_cnt < n_micro: + if f0_cnt < n_micro: + rank_ops.append( + _Action( + stage_id_chunk0, computation_type=F, microbatch_index=f0_cnt + ) + ) + f0_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk0, computation_type=I, microbatch_index=b0_cnt) + ) + rank_ops.append( + _Action(stage_id_chunk0, computation_type=W, microbatch_index=b0_cnt) + ) + b0_cnt += 1 + + rank_ops.append( + _Action(stage_id_chunk1, computation_type=F, microbatch_index=f1_cnt) + ) + f1_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk1, computation_type=I, microbatch_index=b1_cnt) + ) + rank_ops.append( + _Action(stage_id_chunk1, computation_type=W, microbatch_index=b1_cnt) + ) + b1_cnt += 1 + # cool-down phase + w0_cnt, w1_cnt = b0_cnt, b1_cnt + cooldown_n1 = rank + for _ in range(cooldown_n1): + rank_ops.append( + _Action(stage_id_chunk0, computation_type=I, microbatch_index=b0_cnt) + ) + b0_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk1, computation_type=I, microbatch_index=b1_cnt) + ) + b1_cnt += 1 + cooldown_n2 = self.pp_group_size - rank + for _ in range(cooldown_n2): + rank_ops.append( + _Action(stage_id_chunk0, computation_type=I, microbatch_index=b0_cnt) + ) + b0_cnt += 1 + rank_ops.append( + _Action(stage_id_chunk0, computation_type=W, microbatch_index=w0_cnt) + ) + w0_cnt += 1 + while w1_cnt < b1_cnt: + rank_ops.append( + _Action(stage_id_chunk1, computation_type=W, microbatch_index=w1_cnt) + ) + w1_cnt += 1 + while w0_cnt < b0_cnt: + rank_ops.append( + _Action(stage_id_chunk0, computation_type=W, microbatch_index=w0_cnt) + ) + w0_cnt += 1 + + if not (w0_cnt == b0_cnt and b0_cnt == f0_cnt): + raise AssertionError( + f"Expected w0_cnt == b0_cnt == f0_cnt, got w0_cnt={w0_cnt}, b0_cnt={b0_cnt}, f0_cnt={f0_cnt}" + ) + if not (w1_cnt == b1_cnt and b1_cnt == f1_cnt): + raise AssertionError( + f"Expected w1_cnt == b1_cnt == f1_cnt, got w1_cnt={w1_cnt}, b1_cnt={b1_cnt}, f1_cnt={f1_cnt}" + ) + # We use max() in the n_micro computation above, so we may need to + # remove redundant microbatches + rank_ops = [ + ( + action + if action is not None + and action.microbatch_index is not None + and action.microbatch_index < self._n_microbatches + else None + ) + for action in rank_ops + ] + return rank_ops + + +class ScheduleDualPipeV(_PipelineScheduleRuntime): + """ + The DualPipeV schedule. A more efficient schedule variant based on the + DualPipe schedule introduced by DeepSeek in https://arxiv.org/pdf/2412.19437 + + Based on the open sourced code from https://github.com/deepseek-ai/DualPipe + """ + + def __init__( + self, + stages: list[_PipelineStageBase], + n_microbatches: int, + loss_fn: Callable | None = None, + args_chunk_spec: tuple[TensorChunkSpec, ...] | None = None, + kwargs_chunk_spec: dict[str, TensorChunkSpec] | None = None, + output_merge_spec: dict[str, Any] | tuple[Any] | None = None, + scale_grads: bool = True, + backward_requires_autograd: bool = True, + defer_pp_recv: bool = False, + ): + # TODO: we don't support input/weight backward split with torch.compile + _check_torch_compile_compatibility(stages, self.__class__.__name__) + self.pp_group_size = stages[0].group_size + super().__init__( + stages=stages, + n_microbatches=n_microbatches, + loss_fn=loss_fn, + args_chunk_spec=args_chunk_spec, + kwargs_chunk_spec=kwargs_chunk_spec, + output_merge_spec=output_merge_spec, + scale_grads=scale_grads, + backward_requires_autograd=backward_requires_autograd, + defer_pp_recv=defer_pp_recv, + ) + self.stage_index_to_group_rank = generate_stage_to_rank_mapping( + self.pp_group_size, self._num_stages, style="v" + ) + for stage in self._stages: + stage.stage_index_to_group_rank = self.stage_index_to_group_rank + + self.n_local_stages = len(stages) + if self.n_local_stages != 2: + raise ValueError( + "ZBV requires exactly 2 stages per rank, but got " + f"{self.n_local_stages}." + ) + if n_microbatches < self._num_stages: + raise ValueError( + "DualPipeV requires at least as many microbatches as stages, but got " + f"{n_microbatches} microbatches and {self._num_stages} stages." + ) + + self.rank = stages[0].group_rank + self.num_stages = stages[0].num_stages + + # 1. Create the pipeline_order (all ranks do this calculation) + # This will be used to keep track of the current state of the entire pipeline + # pipeline_order[rank] = [Action(computation_type, microbatch_index, stage_index), ...] + self.pipeline_order: dict[int, list[_Action | None]] = {} + for rank in range(self.pp_group_size): + rank_ops = self._calculate_single_rank_operations(rank) + self.pipeline_order[rank] = rank_ops + + # Initialize the pipeline order with communication necessary to run with _PipelineScheduleRuntime + self._prepare_schedule_with_comms(self.pipeline_order) + + def _calculate_single_rank_operations(self, rank) -> list[_Action | None]: + actions: list[_Action | None] = [] + counters: dict[ + tuple[int, _ComputationType], int + ] = {} # (stage_index, computation_type) -> mb_index + weight_queue = [] # Queue of (stage_index, mb_index) for pending weight actions + + num_ranks = self.pp_group_size + num_chunks = self._n_microbatches + + rank_to_stages = generate_rank_to_stage_mapping( + num_ranks, num_ranks * 2, style="v" + ) + stage0_index, stage1_index = rank_to_stages[rank] + + def increment_backward_counts(stage_index: int): + """Helper method to increment BACKWARD_INPUT and BACKWARD_WEIGHT counters when FULL_BACKWARD is used.""" + input_key = (stage_index, BACKWARD_INPUT) + weight_key = (stage_index, BACKWARD_WEIGHT) + counters[input_key] = counters.get(input_key, 0) + 1 + counters[weight_key] = counters.get(weight_key, 0) + 1 + + def add_overlap_f_b( + actions: list, + forward_stage: int, + backward_stage: int, + ): + """Helper method to add an overlapped forward+backward action which tracks microbatch index.""" + # Create new overlapped forward+backward action with sub_actions + forward_key = (forward_stage, FORWARD) + backward_key = (backward_stage, BACKWARD_INPUT) + + forward_mb = counters.get(forward_key, 0) + backward_mb = counters.get(backward_key, 0) + + sub_actions = ( + _Action(forward_stage, FORWARD, forward_mb), + _Action(backward_stage, FULL_BACKWARD, backward_mb), + ) + actions.append(_Action(-1, OVERLAP_F_B, None, sub_actions)) + + # Update counters for sub_actions + counters[forward_key] = forward_mb + 1 + increment_backward_counts(backward_stage) + + def add_action( + actions: list, + stage_index: int, + computation_type: _ComputationType, + ): + # Regular single action, for FULL_BACKWARD we only use the BACKWARD_INPUT counter + key = ( + (stage_index, computation_type) + if computation_type != FULL_BACKWARD + else (stage_index, BACKWARD_INPUT) + ) + mb_index = counters.get(key, 0) + actions.append(_Action(stage_index, computation_type, mb_index)) + + # If FULL_BACKWARD is used, just increment the separate BACKWARD_INPUT and BACKWARD_WEIGHT counters + if computation_type == FULL_BACKWARD: + increment_backward_counts(stage_index) + else: + # If BACKWARD_INPUT is updated, add corresponding weight action to queue + if computation_type == BACKWARD_INPUT: + # Add weight action to queue for later processing + weight_queue.append((stage_index, mb_index)) + counters[key] = mb_index + 1 + + def add_weight_action_if_pending(actions: list): + """Helper method to add a weight action from the queue.""" + if not weight_queue: + return # No pending weight actions, skip + # Pop the oldest weight action from the queue + actual_stage_index, weight_mb_index = weight_queue.pop(0) + actions.append( + _Action( + actual_stage_index, + BACKWARD_WEIGHT, + weight_mb_index, + ) + ) + # Update the counter for the actual stage that was processed + weight_key = (actual_stage_index, BACKWARD_WEIGHT) + counters[weight_key] = counters.get(weight_key, 0) + 1 + + # Step 1: F0 + step_1 = (num_ranks - rank - 1) * 2 + for _ in range(step_1): + add_action(actions, stage0_index, FORWARD) + + # Step 2: F0F1 + step_2 = rank + 1 + for _ in range(step_2): + add_action(actions, stage0_index, FORWARD) + add_action(actions, stage1_index, FORWARD) + + # Step 3: I1W1F1 (Use zero bubble) + step_3 = num_ranks - rank - 1 + for _ in range(step_3): + add_action(actions, stage1_index, BACKWARD_INPUT) + add_weight_action_if_pending(actions) + add_action(actions, stage1_index, FORWARD) + + # Step 4 (Main step): F0B1-F1B0 (combined, overlapped forward+backward) + step_4 = num_chunks - num_ranks * 2 + rank + 1 + for i in range(step_4): + if i == 0 and rank == num_ranks - 1: + # NOTE: We don't overlap these two chunks to further reduce bubble size. + add_action(actions, stage0_index, FORWARD) + add_action(actions, stage1_index, FULL_BACKWARD) + else: + add_overlap_f_b( + actions, + forward_stage=stage0_index, + backward_stage=stage1_index, + ) + add_overlap_f_b( + actions, + forward_stage=stage1_index, + backward_stage=stage0_index, + ) + + # Step 5: B1-F1B0 + step_5 = num_ranks - rank - 1 + for _ in range(step_5): + add_action(actions, stage1_index, FULL_BACKWARD) + add_overlap_f_b( + actions, + forward_stage=stage1_index, + backward_stage=stage0_index, + ) + + # Step 6: B1B0 (The second half of the chunks use zero bubble) + step_6 = rank + 1 + enable_zb = False + for i in range(step_6): + if i == step_6 // 2 and rank % 2 == 1: + enable_zb = True + comp_type = BACKWARD_INPUT if enable_zb else FULL_BACKWARD + add_action(actions, stage1_index, comp_type) + if i == step_6 // 2 and rank % 2 == 0: + enable_zb = True + comp_type = BACKWARD_INPUT if enable_zb else FULL_BACKWARD + add_action(actions, stage0_index, comp_type) + + # Step 7: W0B0 + step_7 = num_ranks - rank - 1 + for _ in range(step_7): + add_weight_action_if_pending(actions) + comp_type = BACKWARD_INPUT if enable_zb else FULL_BACKWARD + add_action(actions, stage0_index, comp_type) + + # Step 8: W0 + step_8 = rank + 1 + for _ in range(step_8): + add_weight_action_if_pending(actions) + + return actions + + +def get_schedule_class(schedule_name: str): + """ + Maps a schedule name (case insensitive) to its corresponding class object. + + Args: + schedule_name (str): The name of the schedule. + """ + schedule_map = { + "1F1B": Schedule1F1B, + "Interleaved1F1B": ScheduleInterleaved1F1B, + "GPipe": ScheduleGPipe, + "LoopedBFS": ScheduleLoopedBFS, + "InterleavedZeroBubble": ScheduleInterleavedZeroBubble, + "PipelineScheduleSingle": PipelineScheduleSingle, + "PipelineScheduleMulti": PipelineScheduleMulti, + "ZBVZeroBubble": ScheduleZBVZeroBubble, + "DualPipeV": ScheduleDualPipeV, + } + lowercase_keys = {k.lower(): k for k in schedule_map} + lowercase_schedule_name = schedule_name.lower() + if lowercase_schedule_name not in lowercase_keys: + raise ValueError( + f"Unknown schedule name '{schedule_name}'. The valid options are {list(schedule_map.keys())}" + ) + return schedule_map[lowercase_keys[lowercase_schedule_name]] + + +def _simulate_comms_compute( + pipeline_order, stage_to_rank: Callable[[int], int], num_stages: int +): + """This function dry-run simulates the actions in the schedule from the perspective of all ranks, and flags + any deadlocks caused by missing or misordered communications. It also simulates any bubbles in time where a rank + can not execute any action due to waiting for unmet dependencies. The total number of simulator steps can be used + as a metric for unit tests involving IR optimization passes as reordering and merging of IR can reduce the number + of simulated steps. + + The simulation is not high-fidelity and does not model overlapping of compute and communication, or cuda streams. + Future work may be to enhance this and model the compute time, comms overlap, and even memory. + """ + pipeline_order = { + rank: [a for a in pipeline_order[rank] if a is not None] + for rank in sorted(pipeline_order) + } + _schedule: dict[int, list[_Action | None]] = { + rank: [] for rank in sorted(pipeline_order) + } + + _prev_ops_rank: dict[int, set[_Action]] = {rank: set() for rank in _schedule} + + def add_to_schedule(rank: int, action: _Action | None): + _schedule[rank].append(action) + if action is not None: + _prev_ops_rank[rank].add(action) + + def _ready_to_schedule(action: _Action | None) -> bool: + if action is None: + return True + + stage_idx = action.stage_index + prev_ops = _prev_ops_rank[stage_to_rank(stage_idx)] + if action.computation_type == F: + if action.stage_index == 0: + return True + elif ( + _Action(action.stage_index, RECV_F, action.microbatch_index) in prev_ops + ): + return True + elif ( + _Action(action.stage_index - 1, F, action.microbatch_index) in prev_ops + ): + return True + return False + elif action.computation_type in (BACKWARD_INPUT, FULL_BACKWARD): + if action.stage_index == num_stages - 1: + return True + if _Action(action.stage_index, RECV_B, action.microbatch_index) in prev_ops: + return True + if ( + _Action(action.stage_index + 1, BACKWARD_INPUT, action.microbatch_index) + in prev_ops + ): + return True + if ( + _Action(action.stage_index + 1, FULL_BACKWARD, action.microbatch_index) + in prev_ops + ): + return True + return False + elif action.computation_type == BACKWARD_WEIGHT: + return True + elif action.computation_type == SEND_F: + expected_f = _Action(action.stage_index, F, action.microbatch_index) + return expected_f in prev_ops + elif action.computation_type == RECV_F: + peer_stage_idx = stage_idx - 1 + expected_send = _Action(peer_stage_idx, SEND_F, action.microbatch_index) + return expected_send in _prev_ops_rank[stage_to_rank(peer_stage_idx)] + elif action.computation_type == SEND_B: + expected_b = _Action( + action.stage_index, BACKWARD_INPUT, action.microbatch_index + ) + expected_bw = _Action( + action.stage_index, FULL_BACKWARD, action.microbatch_index + ) + return expected_b in prev_ops or expected_bw in prev_ops + elif action.computation_type == RECV_B: + peer_stage_idx = stage_idx + 1 + expected_send = _Action(peer_stage_idx, SEND_B, action.microbatch_index) + return expected_send in _prev_ops_rank[stage_to_rank(peer_stage_idx)] + else: + raise ValueError(f"Unsupported action type {action}") + + while pipeline_order: + progress = False + for rank in sorted(pipeline_order): + if len(pipeline_order[rank]) == 0: + continue + + action = pipeline_order[rank][0] + if _ready_to_schedule(action): + if action is not None: + add_to_schedule(rank, action) + pipeline_order[rank].pop(0) + progress = True + else: + add_to_schedule(rank, None) + + for i in sorted(pipeline_order, reverse=True): + if len(pipeline_order[i]) == 0: + del pipeline_order[i] + + # hacky, but do a second pass to replace any 'none' at this timestep with a real action, if it got unblocked + # by one of the later ranks + for rank in sorted(pipeline_order): + if len(pipeline_order[rank]) == 0: + continue + + if _schedule[rank][-1] is not None: + continue + + action = pipeline_order[rank][0] + if _ready_to_schedule(action): + if action is not None: + _schedule[rank][-1] = action + _prev_ops_rank[rank].add(action) + pipeline_order[rank].pop(0) + + for i in sorted(pipeline_order, reverse=True): + if len(pipeline_order[i]) == 0: + del pipeline_order[i] + + if not progress: + print("WIP comms schedule:\n", _format_pipeline_order(_schedule)) + for rank in pipeline_order: + print(f"{rank=} next action= {pipeline_order[rank][0]}") + raise ValueError("Schedule is not progressing") + + return _schedule + + +def _dump_chrometrace(schedule, filename): + """ + This function dumps a schedule IR into a chrometrace format so it can be visualized. + + It is currently very basic and only serves as a graphical alternative to dumping the schedule IR as text. + + As future work we may extend this to include more accurate heuristics for durations, or let users input durations, + add 'flow events' to let the UI show the connection between sends and recvs, and model cuda streams for comm/compute + as separate streams on the chrometrace view. + """ + events = [] + for rank in sorted(schedule): + for timestep, action in enumerate(schedule[rank]): + if action is None: + continue + events.append( + { + "name": str(action), + "cat": ( + "computation" + if action.computation_type in (F, B, W) + else "communication" + ), + "ph": "X", + "pid": rank, + "tid": rank, + "ts": timestep, + "dur": 1, + } + ) + import json + + with open(filename, "w") as f: + json.dump({"traceEvents": events}, f) + + +def _check_torch_compile_compatibility( + stages: list[_PipelineStageBase], schedule_name: str +): + """ + Check if the schedule is compatible with torch.compile. + + Args: + stages: List of pipeline stages to check + schedule_name: Name of the schedule for error message + + Raises: + RuntimeError: If any stage uses torch.compile + """ + for stage in stages: + if not isinstance(stage.submod, torch.nn.Module): + continue + + for module in stage.submod.modules(): + if isinstance(module, OptimizedModule): + raise RuntimeError( + f"The {schedule_name} schedule is not supported with " + "stage modules that have used torch.compile. " + f"Found OptimizedModule in {type(module).__name__}" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/stage.py b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/stage.py new file mode 100644 index 0000000000000000000000000000000000000000..caee33c80d83124e74f55bad0507667546330be1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/pipelining/stage.py @@ -0,0 +1,2454 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import logging +import operator +import warnings +import weakref +from abc import ABC, abstractmethod +from collections.abc import Callable +from typing import Any, cast + +import torch +import torch.distributed as dist +import torch.distributed.config as dist_config +import torch.fx as fx +import torch.nn as nn +from torch._subclasses.fake_tensor import is_fake_tensor +from torch.distributed._composable.replicate_with_fsdp import replicate, ReplicateModule +from torch.distributed.fsdp import FSDPModule, fully_shard +from torch.distributed.pipelining._utils import ( + _derive_grad_metas, + _DTensorMeta, + _make_tensor_from_meta, + _MeshCache, + _StageBackwardMeta, + _StageForwardMeta, + _StageMeta, + _TensorMeta, + extract_tensor_meta, + extract_tensor_metas, + flatten_args, + GetMeshCallback, + InferenceMode, + PipeInfo, + PipeliningMetadataError, + TensorMeta, + to_local_if_dtensor, + validate_and_normalize_to_tuple, + validate_static_arg_grad_correspondence, + validate_tensors_metadata, +) +from torch.distributed.tensor import DTensor +from torch.fx.node import Argument, map_aggregate +from torch.nn.parallel import DistributedDataParallel + +from ._backward import ( + _autograd_grad_for_inputs, + stage_backward, + stage_backward_input, + stage_backward_weight, +) +from ._debug import map_debug_info + + +__all__ = [ + "PipelineStage", + "build_stage", +] + +logger = logging.getLogger(__name__) + + +def _normalize_model_output_as_tuple(output: Any) -> tuple[Any]: + """[Note: pipeline model output type] + + The output of the model passed to pipelining can be any type, controlled by the user. + + However, there are 2 API surfaces that complicate this. + (1) the outputs of intermediate stages are passed via Send/Recv ops to subsequent stages. The implicit assumption + is that each element of the outputs is a tensor. Otherwise, Send/Recv would not be supported. The exception + is the last layer of the model, which can output anything any which won't be communicated via Send/Recv. + (2) the outputs of the last layer of the model are returned to the user, or, passed to the loss function. + The loss function can be written in any way, such that its inputs match the outputs of the model. + + It would be convenient if we could strictly type the output signature of the pipeline stage wrapping the model, + but we do not want to impose an unnecessary constraint on user provided models. + + Currently, we let user provided models return either a Tensor or a tuple of Tensors from each stage. Due to + torch.export tracing, compiled models may also return a list instead of a Tuple, which we will normalize back to a + tuple for consistency. + + TODO: should we be stricter about asserting that stage modules (intermediate and output) all return only Tensor + values? + """ + if type(output) is list: + # HACK: this is a hacky workaround for the fact that export creates + # output in list format + output = tuple(output) + + # Unify output form to tuple for easy correspondence with + # `act_send_info` + output_tuple = output if type(output) is tuple else (output,) + return output_tuple + + +class _RecvInfo: + """Input tensor descriptor for a pipeline stage. + + Handles both received activations from a previous stage + (``is_root_arg=False``) and root-level model inputs provided + by the user (``is_root_arg=True``). + """ + + def __init__( + self, + input_name: str, + source: int | None, + buffer: torch.Tensor | None, + tensor_meta: TensorMeta | None, + *, + is_root_arg: bool = False, + ): + # Name of this input + self.input_name = input_name + # Stage index of the source of this input (None for root args) + self.source = source + # Buffer to receive the input into (None for root args) + self.buffer = buffer + # Tensor metadata for validation and DTensor reconstruction + self.tensor_meta = tensor_meta + # Whether this is a root-level model input (no recv needed) + self.is_root_arg = is_root_arg + + def __repr__(self): + if self.is_root_arg: + return f"_RecvInfo(input={self.input_name}, root_arg=True)" + meta_type = type(self.tensor_meta).__name__ if self.tensor_meta else "None" + buffer_shape = self.buffer.size() if self.buffer is not None else "None" + return f"_RecvInfo(input={self.input_name}, source={self.source}, shape={buffer_shape}, meta={meta_type})" + + +# Cache of per-direction P2P communicators, keyed (weakly) by the PP process +# group they are derived from. Looped/V schedules construct several stage chunks +# per rank that share one PP group; they must share the same forward/backward +# comms (and issue the split_group collective only once) so creation stays +# consistent and cheap across all ranks. Stage chunks are constructed serially +# within a rank, so the first miss performs the split and the rest hit the cache. +# The key is a weakref, so entries are +# dropped automatically once the parent group is destroyed (e.g. via +# destroy_process_group / reinitialization), avoiding stale dead communicators. +_PP_DIRECTION_GROUP_CACHE: "weakref.WeakKeyDictionary[dist.ProcessGroup, tuple[dist.ProcessGroup, dist.ProcessGroup]]" = weakref.WeakKeyDictionary() + + +def _build_p2p_direction_groups( + group: dist.ProcessGroup | None, +) -> tuple[dist.ProcessGroup, dist.ProcessGroup]: + """Create two communicators over the same ranks as ``group``, one per data-flow + direction: ``downstream`` carries traffic flowing ``r -> r+1`` (forward + activations) and ``upstream`` carries ``r -> r-1`` (backward gradients). + + Pipeline P2P normally shares a single communicator for both directions, which + serializes every send/recv in one FIFO. Coalescing makes a single mixed + send+recv batch deadlock-free, but across *separate* batches (pipeline skew, + looped / V schedules, skip connections) the shared FIFO can still form a + dependency cycle and deadlock. Routing the two directions onto separate + communicators / streams removes that cross-batch coupling and restores + full-duplex bandwidth. + + Uses ``split_group``, which is collective over ``group``'s own ranks (not the + whole world), so it composes with PP as a sub-axis of a larger device mesh. + Requires the default process group to be device-bound (e.g. + ``init_process_group(..., device_id=...)``), which ``split_group`` needs for + NCCL; torchcomms binds the device automatically. + """ + parent = group if group is not None else dist.distributed_c10d._get_default_group() + cached = _PP_DIRECTION_GROUP_CACHE.get(parent) + if cached is not None: + return cached + + split_ranks = [list(range(dist.get_world_size(parent)))] + # split_group splits the parent's communicator, so the default process group + # must be device-bound (NCCL) -- torchcomms binds the device automatically. If + # it is not, split_group raises its own device error. + downstream = dist.split_group( + parent_pg=group, split_ranks=split_ranks, group_desc="pp_p2p_downstream" + ) + upstream = dist.split_group( + parent_pg=group, split_ranks=split_ranks, group_desc="pp_p2p_upstream" + ) + # All parent ranks are members of the single split. + if not isinstance(downstream, dist.ProcessGroup): + raise AssertionError(f"expected dist.ProcessGroup, got {type(downstream)}") + if not isinstance(upstream, dist.ProcessGroup): + raise AssertionError(f"expected dist.ProcessGroup, got {type(upstream)}") + logger.info("Pipeline P2P: using per-direction (downstream/upstream) communicators") + _PP_DIRECTION_GROUP_CACHE[parent] = (downstream, upstream) + return downstream, upstream + + +class _PipelineStageBase(ABC): + """Base class for pipeline stages. + + Defines common methods used by ``_PipelineStage`` (tracing frontend) + and ``PipelineStage`` (manual frontend). + """ + + def __init__( + self, + submodule: torch.nn.Module, + stage_index: int, + num_stages: int, + device: torch.device, + group: dist.ProcessGroup | None = None, + dw_builder: Callable[[], Callable[..., None]] | None = None, + ): + """ + Args: + submodule: The module to be executed in this stage. + stage_index: The index of this stage. + num_stages: The total number of stages in this pipeline. + device: The device to run this stage on. + group: Process group for communication. Defaults to the + default process group if ``None``. + dw_builder: Builder function that produces a ``dw_runner`` + for deferred weight updates in F/I/W zero-bubble + schedules. If ``None``, a runner is generated + automatically via autograd graph traversal. + """ + super().__init__() + if stage_index >= num_stages: + raise ValueError( + f"Stage index {stage_index} is out of range of {num_stages}" + ) + + self.submod = submodule + self.stage_index = stage_index + self.num_stages = num_stages + self.device = device + self.group = group + + # Downstream (data flowing r -> r+1: forward activations) and upstream + # (r -> r-1: backward gradients) P2P communicators. Auto-enabled when + # TorchComms is in use (its split path is always available and the + # single-comm FIFO deadlock is most acute there); the config flag + # torch.distributed.config.pipeline_per_direction_p2p (env + # TORCH_DISTRIBUTED_PIPELINE_PER_DIRECTION_P2P) force-enables it on other + # backends. When disabled both alias ``self.group`` so behavior is + # byte-for-byte unchanged. + self.p2p_per_direction = ( + dist_config.pipeline_per_direction_p2p + or dist.distributed_c10d._use_torchcomms_enabled() + ) + self._downstream_group: dist.ProcessGroup | None + self._upstream_group: dist.ProcessGroup | None + if self.p2p_per_direction: + self._downstream_group, self._upstream_group = _build_p2p_direction_groups( + group + ) + else: + self._downstream_group = group + self._upstream_group = group + + self.dw_builder = dw_builder + + # backward state + self.backward_state: dict[int, tuple[Any, ...]] = {} + + # store dw_runner per microbatch_id + self.dw_runner: dict[int, Callable[..., None]] = {} + + # `group_rank` is rank in process group `group`. + self.group_rank = dist.get_rank(self.group) + self.group_size = dist.get_world_size(self.group) + if self.group_size > self.num_stages: + raise RuntimeError( + f"Pipeline group size {self.group_size} cannot be larger than number of stages {self.num_stages}" + ) + + # Run time states + # map microbatch ID to list of forward tensor args + self.fwd_cache: dict[int, tuple[Any, list[torch.Tensor]]] = {} + # map microbatch ID to list of backward grad tensor args + self.bwd_cache: dict[int, tuple[torch.Tensor | None, ...]] = {} + # Caching chunk outputs for final output merge or reduction + self.output_chunks: list[Any] = [] + + # Initialize has_backward to false; this will be set to true if loss + # function is passed to pipeline schedule + self.has_backward = False + # Log prefix + self.log_prefix = f"[Stage {self.stage_index}]" + + # Forward infra + self.args_recv_info: dict[int, tuple[_RecvInfo, ...]] = {} + self.act_send_info: dict[int, list] = {} + + # Backward infra will be created lazily + self.grad_recv_info: dict = {} + self.grad_send_info: list | None = None + + # To be populated later by the Schedule + self.chunks: int | None = None + self.stage_index_to_group_rank: dict[int, int] = { + i: i % self.group_size for i in range(self.num_stages) + } + + # DTensor support: mesh cache for looking up DeviceMesh by (dim_names, layout) + self._mesh_cache = _MeshCache() + + # Per-chunk runtime validation is expensive; only enable under + # TORCH_DISTRIBUTED_DEBUG=DETAIL for debugging shape/dtype mismatches. + self._runtime_validate = dist.get_debug_level() == dist.DebugLevel.DETAIL + + # DTensor support: consolidated stage metadata container + # Contains inputs, outputs, input_grads, output_grads metadata + self._stage_meta = _StageMeta() + + @property + def has_backward(self) -> bool: + """ + Returns true if this stage has a backward pass. + """ + return self._has_backward + + @has_backward.setter + def has_backward(self, has_backward: bool): + self._has_backward = has_backward + + @property + def is_first(self): + """ + Returns true if this stage is the first stage in the pipeline. + """ + return self.stage_index == 0 + + @property + def is_last(self): + """ + Returns true if this stage is the last stage in the pipeline. + """ + return self.stage_index == self.num_stages - 1 + + def _validate_stage_tensors( + self, + desc: str, + expected: tuple[TensorMeta | None, ...] | None, + actual: tuple[torch.Tensor | None, ...], + ) -> None: + """Validate actual tensors against expected metadata. + + Raises: + PipeliningMetadataError: If metadata is missing or mismatched. + """ + if expected is None: + raise PipeliningMetadataError(f"{desc}: no metadata available") + validate_tensors_metadata(desc, expected, actual) + + def _check_chunk_id(self, chunk_id: int): + if self.chunks is None: + raise RuntimeError( + "Attempted to access chunk_id before chunks have been configured." + ) + if chunk_id >= self.chunks: + raise RuntimeError( + f"Chunk id {chunk_id} is out of range [0, {self.chunks})" + ) + + def _create_grad_send_info( + self, + args_recv_info: tuple, + ) -> list[int | None]: + """ + Create a list of stage indices to send gradients to. + """ + grad_send_info: list[int | None] = [] + + def map_recv_to_send(a): + # Note: we send gradients back to previous stage as long as in + # forward it is a received input, regardless of whether it requires + # grad. It is up to the previous stage to discard this gradient. + if a.is_root_arg: + # Root args don't have a source stage to send gradients to + grad_send_info.append(None) + return None + else: + grad_send_info.append(a.source) + return a.source + + map_aggregate(args_recv_info, map_recv_to_send) + + logger.debug("%s Grad send info: %s", self.log_prefix, grad_send_info) + return grad_send_info + + @abstractmethod + def _prepare_forward_infra( + self, + num_microbatches: int, + args: tuple[Any, ...] | _StageForwardMeta | None, + kwargs: dict[str, Any] | None = None, + has_backward: bool = False, + ) -> _StageForwardMeta | None: + raise NotImplementedError + + @abstractmethod + def _prepare_backward_infra( + self, + num_microbatches: int, + loss_fn: Callable[..., torch.Tensor] | None = None, + target: torch.Tensor | None = None, + received_grad_meta: _StageBackwardMeta | None = None, + loss_kwargs: dict[str, Any] | None = None, + ) -> _StageBackwardMeta | None: + raise NotImplementedError + + def _setup_backward_recv_info(self, num_microbatches: int): + # TODO: this is needed for backward_maybe_with_nosync + self.chunks = num_microbatches + + # IMPORTANT: _create_grad_recv_info reads self._stage_meta.output_grads + # to attach DTensor metadata to _RecvInfo objects. The clear below MUST + # happen after all _create_grad_recv_info calls complete. + for mb_index in range(num_microbatches): + self.grad_recv_info[mb_index] = self._create_grad_recv_info( + self.act_send_info + ) + + @abstractmethod + def _create_grad_recv_info( + self, + act_send_info: dict, + ) -> tuple[_RecvInfo, ...]: + raise NotImplementedError + + def _resolve_peer_global_rank(self, stage_idx: int) -> int: + """Map a pipeline stage index to the corresponding global rank for P2P communication.""" + peer_rank = self.stage_index_to_group_rank[stage_idx] + return dist.get_global_rank( + self.group or dist.distributed_c10d._get_default_group(), + peer_rank, + ) + + def _get_recv_ops( + self, + recv_infos: tuple[_RecvInfo, ...], + group: dist.ProcessGroup | None, + ) -> list[dist.P2POp]: + """ + Helper function shared by `get_fwd_recv_ops` and `get_bwd_recv_ops`. + Returns a list of ops that correspond to the recv infos. ``group`` is the + direction-specific communicator (downstream vs upstream); it equals + ``self.group`` unless per-direction P2P is enabled. + """ + ops: list[dist.P2POp] = [] + for info in recv_infos: + if info.is_root_arg: + # Root args don't need recv operations + continue + # Skip entries with None buffer (None gradients) + if info.buffer is None: + if info.tensor_meta is not None: + raise AssertionError("expected info.tensor_meta to be None") + continue + # At this point, source and buffer are guaranteed non-None + if info.source is None: + raise AssertionError("expected info.source to be not None") + peer_global_rank = self._resolve_peer_global_rank(info.source) + ops.append(dist.P2POp(dist.irecv, info.buffer, peer_global_rank, group)) + + return ops + + """[Note: V-schedule special case] + + V-Schedules have a special case where 2 stages with adjacent stage_id + are on the same rank. + + Example: 2 ranks, 4 stages forms a simple V:: + + rank0: stage 0 stage 3 + rank1: stage 1 stage 2 + + Stages 0/1 and 2/3 communicate via send/recv, but stages 1/2 pass + tensors directly via function call, avoiding communication ops. + """ + + def set_local_fwd_input(self, prev_stage_outputs: Any, mb_index: int) -> None: + """Pass outputs from a same-rank stage as forward inputs (V-schedule). + + Detaches tensors and sets ``requires_grad`` so they serve as autograd + leaves. Handles DTensor activations transparently. + """ + recv_infos: tuple[_RecvInfo, ...] = self.args_recv_info[mb_index] + + # See [Note: pipeline model output type] + prev_stage_outputs = _normalize_model_output_as_tuple(prev_stage_outputs) + + for info, tensor in zip(recv_infos, prev_stage_outputs, strict=True): + if not isinstance(tensor, torch.Tensor): + raise AssertionError( + f"expected tensor values as outputs from prev stage, got {type(tensor)}" + ) + if info.is_root_arg: + raise AssertionError( + "set_local_fwd_input should only be called on non-first stage, which should always have non-root RecvInfo" + ) + + # Pass the activation tensor directly (same rank for local execution). + # Detach to create a new autograd leaf for the fresh autograd graph. + info.buffer = to_local_if_dtensor(tensor, detach=True) + + def get_local_bwd_output(self, mb_index): + """ + Returns the input grad tensors for this stage, which correspond to the stage inputs during forward. + """ + if not self.has_backward: + raise AssertionError( + "can't steal_bwd_input if this stage doesn't have backward" + ) + if self.is_first: + raise AssertionError("can't get bwd output if this stage is first") + + self._check_chunk_id(mb_index) + return self.bwd_cache.pop(mb_index) + + def set_local_bwd_input( + self, next_stage_bwd_outputs: tuple[torch.Tensor | None, ...], mb_index: int + ) -> None: + """ + Moves 'grad input' tensors from the next stage to 'grad_output' on this stage, avoiding a copy or send/recv. + Does not detach or set '_requires_grad'. + Handles DTensor gradients for V-schedule local passing. + """ + if not isinstance(next_stage_bwd_outputs, tuple): + raise AssertionError(f"Expected tuple, got {type(next_stage_bwd_outputs)}") + + if not self.has_backward: + raise AssertionError( + "can't set bwd input if this stage doesn't have backward" + ) + if self.is_last: + raise AssertionError("can't set bwd input if this stage is last") + recv_infos = self.grad_recv_info[mb_index] + for info, tensor in zip(recv_infos, next_stage_bwd_outputs, strict=True): + if tensor is None: + if info.buffer is not None: + info.buffer.zero_() + continue + if not isinstance(tensor, torch.Tensor): + raise AssertionError( + f"expected tensor values as outputs from prev stage, got {type(tensor)}" + ) + if info.is_root_arg: + raise AssertionError( + "set_local_bwd_input should only be called with non-root RecvInfo" + ) + + # Extract local tensor for the buffer (handles DTensor or plain tensor) + info.buffer = to_local_if_dtensor(tensor) + + def get_fwd_recv_ops(self, fwd_chunk_id: int) -> list[dist.P2POp]: + """ + Returns a list of ops that are needed to receive the input arguments + for this stage. + """ + recv_infos: tuple[_RecvInfo, ...] = self.args_recv_info[fwd_chunk_id] + + return self._get_recv_ops(recv_infos, self._downstream_group) + + def get_bwd_recv_ops(self, bwd_chunk_id: int) -> list[dist.P2POp]: + """ + Returns a list of ops that are needed to receive the gradients + for this stage. + """ + if not self.has_backward or self.is_last: + return [] + + recv_infos = self.grad_recv_info[bwd_chunk_id] + return self._get_recv_ops(recv_infos, self._upstream_group) + + def get_fwd_send_ops(self, fwd_chunk_id: int) -> list[dist.P2POp]: + """ + Get the activation send ops for current stage's forward. + Handles DTensor outputs by extracting local tensors. + """ + output_tuple, _ = self.fwd_cache[fwd_chunk_id] + + ops: list[dist.P2POp] = [] + + for idx, out in enumerate(output_tuple): + dst_stages = self.act_send_info[idx] + for dst in dst_stages: + if dst is None: + continue + # Extract local tensor if DTensor + send_tensor = to_local_if_dtensor(out, detach=True) + logger.debug( + "%s Sending tensor to Stage %s: %s", + self.log_prefix, + dst, + send_tensor.size(), + ) + peer_global_rank = self._resolve_peer_global_rank(dst) + ops.append( + dist.P2POp( + dist.isend, + send_tensor, + peer_global_rank, + self._downstream_group, + ) + ) + + return ops + + def _get_grad_send_meta(self, input_idx: int) -> TensorMeta | None: + """Return gradient metadata for ``input_idx`` if a recv was allocated.""" + input_grads = self._stage_meta.input_grads + if input_grads is not None and input_idx < len(input_grads): + return input_grads[input_idx] + + inputs = self._stage_meta.inputs + if inputs is not None and input_idx < len(inputs): + meta = inputs[input_idx] + if meta is not None: + return _derive_grad_metas((meta,))[0] + + raise PipeliningMetadataError( + f"Stage {self.stage_index}: backward produced gradient for input " + f"{input_idx}, but no metadata is available for the gradient send." + ) + + def get_bwd_send_ops(self, bwd_chunk_id: int) -> list[dist.P2POp]: + """ + Get the gradient send ops for current stage's backward. + Handles DTensor gradients by extracting local tensors. + """ + if not self.has_backward or self.is_first: + return [] + + self._check_chunk_id(bwd_chunk_id) + # Create bwd send infra lazily + if self.grad_send_info is None: + # Send info for input grads during backward: + # List of destinations corresponding to input grads + # Can be None if an input has no grad + # `grad_send_info` is a mirror of `args_recv_info` + self.grad_send_info = self._create_grad_send_info(self.args_recv_info[0]) + + ops: list[dist.P2POp] = [] + grads_input = self.bwd_cache.pop(bwd_chunk_id) + + for idx, (grad, grad_recv_stage) in enumerate( + zip(grads_input, self.grad_send_info, strict=True) + ): + if grad_recv_stage is None: + if grad is not None: + raise PipeliningMetadataError( + f"[{self.stage_index}] for chunk {bwd_chunk_id} has gradient {grad} " + f"but no stage to send it to" + ) + continue + + grad_meta = self._get_grad_send_meta(idx) + if grad_meta is None: + if grad is not None: + raise PipeliningMetadataError( + f"Stage {self.stage_index} chunk {bwd_chunk_id} input {idx}: " + f"backward produced a {type(grad).__name__} gradient, but " + "backward metadata for this grad slot is None, so the " + "previous stage did not allocate a recv buffer. DTensor " + "grad metadata cannot be derived from forward DTensor " + "metadata because gradient placements may differ from " + "activation placements. Provide static input_grads " + "metadata or run metadata inference with a microbatch " + "that produces this grad." + ) + continue + + if isinstance(grad, torch.Tensor): + # Extract local tensor if DTensor + send_tensor = to_local_if_dtensor(grad) + logger.debug( + "%s Sending gradient to Stage %s: %s", + self.log_prefix, + grad_recv_stage, + send_tensor.size(), + ) + peer_global_rank = self._resolve_peer_global_rank(grad_recv_stage) + ops.append( + dist.P2POp( + dist.isend, send_tensor, peer_global_rank, self._upstream_group + ) + ) + elif grad is None: + zero_grad = _make_tensor_from_meta(grad_meta, self.device).zero_() + send_tensor = to_local_if_dtensor(zero_grad) + logger.debug( + "%s Sending zero gradient to Stage %s: %s", + self.log_prefix, + grad_recv_stage, + send_tensor.size(), + ) + peer_global_rank = self._resolve_peer_global_rank(grad_recv_stage) + ops.append( + dist.P2POp(dist.isend, send_tensor, peer_global_rank, self.group) + ) + return ops + + def clear_runtime_states(self) -> None: + """ + Clear runtime states of the stage. + """ + # map microbatch ID to list of forward tensor args + self.fwd_cache.clear() + # Caching chunk outputs for final output merge or reduction + self.output_chunks.clear() + + # Clear grad of input buffers in between schedule steps. This is because + # `torch.autograd.backward()` will accumulate gradients into leaf + # tensors by default. For gradients to pass back to previous stages, we + # don't want such accumulation. + for recv_tuple in self.args_recv_info.values(): # iterate over all chunks + for a in recv_tuple: # iterate over all input args + if not a.is_root_arg and a.buffer is not None: + # Set to None is the newer and recommended way to clear grads, compared to `zero_()`. + # See https://github.com/pytorch/pytorch/pull/92731 + a.buffer.grad = None + + def _map_tensor_from_recv_info( + self, + recv_infos: tuple[_RecvInfo, ...], + ): + """ + Map tensors from recv infos to a list. + """ + + def get_recv_tensor(info): + if info.is_root_arg: + raise PipeliningMetadataError("Cannot get recv tensor from root arg") + return info.buffer + + return map_aggregate(cast(Argument, recv_infos), get_recv_tensor) + + def _retrieve_recv_activations( + self, + fwd_chunk_id: int, + ): + """ + Retrieve the activations received for the current stage during forward. + Reconstructs DTensors if the inputs were DTensors. + Also validates DTensor metadata against expected values. + """ + recv_infos = self.args_recv_info[fwd_chunk_id] + + activations = [] + for i, info in enumerate(recv_infos): + if not info.is_root_arg: + # Non-root args have valid buffer and tensor_meta + if info.buffer is None or info.tensor_meta is None: + raise PipeliningMetadataError( + f"Non-root arg '{info.input_name}' has None buffer or tensor_meta" + ) + # Effective requires_grad: metadata captures what the model + # produced, but the runtime context (has_backward, grad mode) + # determines whether we actually need gradients. + effective_requires_grad = ( + info.tensor_meta.requires_grad + and self.has_backward + and torch.is_grad_enabled() + ) + if isinstance(info.tensor_meta, _DTensorMeta): + # Buffer must not require grad so from_local stays out + # of the autograd graph (no grad_placements needed). + if info.buffer.requires_grad: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: recv buffer " + f"'{info.input_name}' unexpectedly requires grad " + f"before DTensor reconstruction" + ) + mesh = self._mesh_cache.get_mesh(info.tensor_meta.mesh_cache_key) + activation = DTensor.from_local( + info.buffer, + device_mesh=mesh, + placements=info.tensor_meta.placements, + shape=info.tensor_meta.global_shape, + stride=info.tensor_meta.global_stride, + run_check=False, + ).requires_grad_(effective_requires_grad) + else: + activation = info.buffer.requires_grad_(effective_requires_grad) + # Activation must be a leaf so backward terminates here. + if effective_requires_grad and not activation.is_leaf: + warnings.warn( + f"Stage {self.stage_index}: activation " + f"'{info.input_name}' is not a leaf " + f"(grad_fn={activation.grad_fn}); using " + f"retain_grad() as fallback", + stacklevel=2, + ) + activation.retain_grad() + activations.append(activation) + else: + raise PipeliningMetadataError( + f"_retrieve_recv_activations expected non-root _RecvInfo but got root arg at index {i}" + ) + + return tuple(activations) + + def _retrieve_recv_grads( + self, + bwd_chunk_id: int, + ): + """ + Retrieve the gradients received for the current stage during backward. + + Handles None gradients gracefully (for inputs that don't require grad). + """ + recv_infos = self.grad_recv_info[bwd_chunk_id] + + grads: list[torch.Tensor | None] = [] + for i, info in enumerate(recv_infos): + if not isinstance(info, _RecvInfo): + raise PipeliningMetadataError( + f"Expected _RecvInfo but got {type(info)}" + ) + if not info.is_root_arg: + # Gradients can be None for non-differentiable outputs + if info.buffer is None: + if info.tensor_meta is not None: + raise PipeliningMetadataError( + f"Grad recv '{info.input_name}': buffer is None but tensor_meta is not None" + ) + grads.append(None) + continue + if info.tensor_meta is None: + raise PipeliningMetadataError( + f"Grad recv '{info.input_name}': buffer is not None but tensor_meta is None" + ) + if isinstance(info.tensor_meta, _DTensorMeta): + # Reconstruct DTensor gradient from local tensor + metadata + mesh = self._mesh_cache.get_mesh(info.tensor_meta.mesh_cache_key) + grad = DTensor.from_local( + info.buffer, + device_mesh=mesh, + placements=info.tensor_meta.placements, + shape=info.tensor_meta.global_shape, + stride=info.tensor_meta.global_stride, + run_check=False, + ) + else: + grad = info.buffer + grads.append(grad) + else: + raise PipeliningMetadataError( + f"grad_recv_info should not contain root args, but found one at index {i}" + ) + + return tuple(grads) + + def forward_maybe_with_nosync(self, *args, **kwargs): + # If submod is wrapped with DDP, we use the `no_sync` context manager to + # avoid gradient all-reduce per microbatch + if isinstance(self.submod, DistributedDataParallel): + with self.submod.no_sync(): # type: ignore[operator] + out_val = self.submod(*args, **kwargs) + else: + out_val = self.submod(*args, **kwargs) + return out_val + + def scale_grads(self, grad_scale_factor: int) -> None: + """Scale gradients model gradients by `grad_scale_factor`, which should be specified in coordination with the + loss function used with pipelining. For loss functions which perform 'mean' loss reduction, `grad_scale_factor` + should be set to num_microbatches. For loss functions that use `sum` reduction, `grad_scale_factor` should + be set to 1. + + Should only be called once per pipeline schedule step, after all backwards passes have completed. + """ + + # PP scales only for its own contribution (microbatches), but relies on DP to scale further + # for DP degree. + if grad_scale_factor != 1: + for p in self.submod.parameters(): + if p.grad is not None: + p.grad.div_(grad_scale_factor) + + def backward_maybe_with_nosync( + self, + backward_type, + bwd_kwargs: dict, + last_backward: bool = False, + ) -> tuple[tuple[torch.Tensor | None, ...], list[dict[str, Any]] | None]: + """ + Whether using PP with FSDP, DDP, or replicate there are some runtime differences between the last backward step and the + other steps. Namely, we need to accumulate gradients on previous steps and reduce them on the last step, but + there are additional state-variables and performance considerations depending on the data parallelism used. + This helper should adapt any pipeline parallel schedule to work with common/supported data parallel libraries. + """ + + def perform_backward( + backward_type, + ) -> Callable[ + [], + tuple[tuple[torch.Tensor | None, ...], list[dict[str, Any]] | None], + ]: + if backward_type == "full": + return lambda: ( + stage_backward( + bwd_kwargs["stage_output"], + bwd_kwargs["output_grads"], + bwd_kwargs["input_values"], + ), + None, + ) + elif backward_type == "input": + return lambda: stage_backward_input( + bwd_kwargs["stage_output"], + bwd_kwargs["output_grads"], + bwd_kwargs["input_values"], + self.submod.parameters(), + ) + elif backward_type == "weight": + return lambda: ( + stage_backward_weight( + self.submod.parameters(), bwd_kwargs["param_groups"] + ), + None, + ) + else: + raise RuntimeError(f"Unknown backward type: {backward_type}") + + # If submod is wrapped by DDP + if isinstance(self.submod, DistributedDataParallel): + if last_backward: + # Last chunk, prepare for gradient reduction + # HACK: reaching into DDP implementation details here. Is there a better way? + self.submod.reducer.prepare_for_backward( # type: ignore[union-attr, operator] + list( + torch.nn.parallel.distributed._find_tensors( # type: ignore[attr-defined] + bwd_kwargs["stage_output"] + ) + ) + ) + result = perform_backward(backward_type)() + else: + with self.submod.no_sync(): # type: ignore[operator] + result = perform_backward(backward_type)() + + # If submod is a FSDP or replicate module + elif isinstance(self.submod, FSDPModule): + self.submod.set_is_last_backward(False) + self.submod.set_reshard_after_backward(False) + self.submod.set_requires_gradient_sync(False) + result = perform_backward(backward_type)() + + else: + # Non-DP submodule, regular backward + result = perform_backward(backward_type)() + + grads, param_groups = result + return grads, param_groups + + def forward_one_chunk( + self, + fwd_chunk_id: int, + args: tuple[Any, ...], + kwargs: dict[str, Any] | None = None, + save_forward_output: bool = True, + ): + """ + Perform forward pass on the stage with one microbatch. + `args` and `kwargs` are the inputs from *external* to this stage. + As of Sept 2024: + - `args` applies to the first stage only, other stages receives args + through activation transmission. + - `kwargs` can be passed to all stages via respective `step` calls. + """ + + if self.is_first: + # First stage doesn't need to receive anything + composite_args = args + else: + # Receive activations for this chunk + # Activations only come in args form + composite_args = self._retrieve_recv_activations(fwd_chunk_id) + + composite_kwargs = kwargs or {} + + if self._runtime_validate: + self._validate_stage_tensors( + f"Stage {self.stage_index} forward inputs", + self._stage_meta.inputs, + composite_args, + ) + + # Compute forward + try: + output = self.forward_maybe_with_nosync(*composite_args, **composite_kwargs) + + except Exception as e: + exc_msg = f""" + {self.log_prefix} failed to run forward: + args: {map_debug_info(composite_args)} + kwargs: {map_debug_info(composite_kwargs)} + """ + raise RuntimeError(exc_msg) from e + + # See [Note: pipeline model output type] + output_tuple = _normalize_model_output_as_tuple(output) + + # Prepare for final output merge or reduction + # Output chunks is only used for the last stage since we only merge the output of the last stage + if self.is_last and save_forward_output: + self.output_chunks.append(output) + # Save activations and inputs for backward + flat_args = flatten_args(composite_args) + flat_kwargs = flatten_args(composite_kwargs) + flatten_input_tensors = flat_args + flat_kwargs + self.fwd_cache[fwd_chunk_id] = ( + output_tuple, # stage_output + flatten_input_tensors, # input_values + ) + + logger.debug( + "%s Forwarded chunk %s, outputs: %s", + self.log_prefix, + fwd_chunk_id, + map_debug_info(output), + ) + # Validate outputs before P2P send; skipped for last stage (outputs + # go to loss/user, not via send/recv). + if self._runtime_validate and not self.is_last: + self._validate_stage_tensors( + f"Stage {self.stage_index} forward outputs", + self._stage_meta.outputs, + output_tuple, + ) + + # We return the original user-provided output, not normalized to tuple. + # See [Note: pipeline model output type] + return output + + def backward_one_chunk( + self, + bwd_chunk_id: int, + loss=None, + full_backward: bool = True, + last_backward=False, + ): + """ + Perform backward pass on the module. + This should only be called once per microbatch. + + If full_backward is True (the default), the full backward pass including weight and input gradients will be run, + and it is an error to call `backward_weight_one_chunk` for this bwd_chunk_id. + + If full_backward is False, it is optional that `dw_runner` was provided to the PipelineStage at __init__ time, + and a subsequent call to `backward_weight_one_chunk` is required to invoke dw_runner and complete the backward. + + last_backward is controlled by the schedule and signals synchronization of gradients across DP groups + after the last backward. + """ + # skip backward computation if backward is not enabled + if not self.has_backward: + return + + self._check_chunk_id(bwd_chunk_id) + + ( + stage_output, + input_values, + ) = self.fwd_cache.pop(bwd_chunk_id) + + # Compute backward + if self.is_last: + # Last stage computes gradients from loss and has no gradients from + # next stage + bwd_kwargs = { + "stage_output": loss, + "output_grads": None, + "input_values": input_values, + } + else: + # Otherwise, receive gradients from next stage + grads_output = self._retrieve_recv_grads(bwd_chunk_id) + if self._runtime_validate: + # Validate backward input (output gradients) for DTensor metadata + self._validate_stage_tensors( + f"Stage {self.stage_index} backward input (output_grads)", + self._stage_meta.output_grads, + grads_output, + ) + # If an input to the pipeline requires gradient, + # `torch.autograd.backward` will accumulate the gradient into the + # `.grad` field of such input + bwd_kwargs = { + "stage_output": stage_output, + "output_grads": grads_output, + "input_values": input_values, + } + + grads_input: tuple[torch.Tensor | None, ...] = () + + # Custom backward function + if self.dw_builder: + # TODO: We may want to change our semantics so we are allowed to ignore + # the 'dw_builder' and call full_backward directly when it is a full_backward op. + grads_input, _ = self.backward_maybe_with_nosync( + "full", + bwd_kwargs, + last_backward=last_backward, + ) + if full_backward: + self.dw_builder()() + else: + self.dw_runner[bwd_chunk_id] = self.dw_builder() + else: + if full_backward: + grads_input, _ = self.backward_maybe_with_nosync( + "full", bwd_kwargs, last_backward=last_backward + ) + else: + param_groups: list[dict[str, Any]] | None = None + # Skip the backward for the first stage since we will perform the weight update with + # autograd.backward in backward_weight_one_chunk + if not self.is_first: + if isinstance(bwd_kwargs["stage_output"], torch.Tensor): + bwd_kwargs["stage_output"] = (bwd_kwargs["stage_output"],) + + # perform the partial backwards for the inputs with a custom backward function + # when the "stage_ouput" is a loss, then it is a tensor, otherwise it is a tuple of tensors + grads_input, param_groups = self.backward_maybe_with_nosync( + "input", bwd_kwargs, last_backward=last_backward + ) + + # TODO: we don't need to save this, add to dw_runner? + self.backward_state[bwd_chunk_id] = ( + bwd_kwargs["input_values"], + param_groups, + bwd_kwargs["stage_output"], + bwd_kwargs["output_grads"], + ) + # Save a placeholder for the dw_runner + self.dw_runner[bwd_chunk_id] = lambda: None + # Note: grads_input may contain gradients for both args and kwargs (from fwd_cache), + # Kwargs are local to each stage and don't need gradient transmission. + # Validate backward output (input gradients) for DTensor metadata + if self._stage_meta.inputs is None: + raise AssertionError("expected self._stage_meta.inputs to be not None") + num_fwd_args = len(self._stage_meta.inputs) + if self._runtime_validate and not self.is_first: + self._validate_stage_tensors( + f"Stage {self.stage_index} backward output (input_grads)", + self._stage_meta.input_grads, + grads_input[:num_fwd_args], + ) + self.bwd_cache[bwd_chunk_id] = grads_input[:num_fwd_args] + + if self.is_last and not self.is_first: + # Autograd dependencies: + # rest_of_autograd_graph -> stage_output -> loss + # stage_output is no longer used in the last stage for backward and only needed + # to return to the user in merge_output_chunks, therefore + # this should be detached to release autograd graph context and free memory earlier + for t in stage_output: + if not t._is_view(): # views are not detachable in-place + t.detach_() + + logger.debug("%s Backwarded chunk %s", self.log_prefix, bwd_chunk_id) + + def backward_weight_one_chunk(self, bwd_chunk_id: int, last_backward=False): + # skip backward computation if backward is not enabled + if not self.has_backward: + return + + if bwd_chunk_id not in self.dw_runner: + raise AssertionError( + f"{self.log_prefix} Attempted to run backward_weight_one_chunk for chunk {bwd_chunk_id}" + " without first calling `backward_one_chunk(full_backward=False)`" + ) + + if self.dw_builder is not None: + self.dw_runner.pop(bwd_chunk_id)() + else: + ( + input_values, + param_groups, + stage_output, + output_grads, + ) = self.backward_state.pop(bwd_chunk_id) + + if self.stage_index != 0: + bwd_kwargs = { + "stage_output": stage_output, + "param_groups": param_groups, + } + self.backward_maybe_with_nosync( + "weight", bwd_kwargs, last_backward=last_backward + ) + else: + # TODO: figure out a better way to do this: + # if inputs does not require gradient, + # then the parameter group will not be fully captured during stage_backward_input + # in this case, we need call grad directly on the parameters + # To solve: make input fn do the intersect compute and then finish it off during W + bwd_kwargs = { + "stage_output": stage_output, + "output_grads": output_grads, + "input_values": input_values, + } + self.backward_maybe_with_nosync( + "full", bwd_kwargs, last_backward=last_backward + ) + + def _get_init_p2p_neighbors_ops(self) -> list[dist.P2POp]: + """ + Get the operations to initialize the p2p communicators between previous and next stages. + This is done so by creating a dummy tensor and sending it to the next stage and receiving + from the previous stage. + """ + ops: list[dist.P2POp] = [] + next_stage_peer_rank = self.stage_index_to_group_rank.get(self.stage_index + 1) + prev_stage_peer_rank = self.stage_index_to_group_rank.get(self.stage_index - 1) + + # Separate recv buffers per direction: with per-direction P2P the + # downstream and upstream recvs run concurrently on different + # communicators/streams, so they must not share a buffer (concurrent + # writes = data race). The send buffer is only read, so it can be shared. + downstream_recv_tensor = torch.zeros(1, device=self.device, dtype=torch.float32) + upstream_recv_tensor = torch.zeros(1, device=self.device, dtype=torch.float32) + send_tensor = torch.tensor( + self.stage_index, device=self.device, dtype=torch.float32 + ) + # downstream traffic (r -> r+1: forward activations) -> downstream comm + if not self.is_first: + ops.append( + dist.P2POp( + dist.irecv, + downstream_recv_tensor, + group_peer=prev_stage_peer_rank, + group=self._downstream_group, + ) + ) + if not self.is_last: + ops.append( + dist.P2POp( + dist.isend, + send_tensor, + group_peer=next_stage_peer_rank, + group=self._downstream_group, + ) + ) + + # upstream traffic (r -> r-1: backward gradients) -> upstream comm + if not self.is_first: + ops.append( + dist.P2POp( + dist.isend, + send_tensor, + group_peer=prev_stage_peer_rank, + group=self._upstream_group, + ) + ) + if not self.is_last: + ops.append( + dist.P2POp( + dist.irecv, + upstream_recv_tensor, + group_peer=next_stage_peer_rank, + group=self._upstream_group, + ) + ) + + return ops + + def perform_reduce_grad(self, grad_scale_factor: int): + """ + Called as a part of schedule IR. + REDUCE_GRAD action is scheduled after all microbatches W, B actions. + + Currently contains "post_backward" functionality for FSDP. + We can try to extract post_backward in a separate IR action in future. + """ + # Manually call post backward for FSDP + if isinstance(self.submod, FSDPModule): + fsdp_module = self.submod + fsdp_module.set_is_last_backward(True) + fsdp_module.set_reshard_after_backward(True) + fsdp_module.set_requires_gradient_sync(True) + + if isinstance(fsdp_module, ReplicateModule): + distributed_state = replicate.state(fsdp_module) # type: ignore[arg-type] + else: + distributed_state = fully_shard.state(fsdp_module) # type: ignore[attr-defined] + + for state in distributed_state._state_ctx.all_states: + for fsdp_param_group in state._fsdp_param_groups: + fsdp_param_group.post_backward() + + # it would be much better if pipelining backward invoked .backward so autograd hooks + # worked and modules like DDP/FSDP behaved as expected. Working around this for the time being, + # we need to call this too to ensure FSDP syncs its grad reduction ops back to the default stream. + distributed_state._root_post_backward_final_callback() + # Call gradient scaling at the end of the backward pass + # NOTE: this must happen after FSDP post_backward is FSDP is enabled + if grad_scale_factor != 1: + self.scale_grads(grad_scale_factor) + + +class _PipelineStage(_PipelineStageBase): + def __init__( + self, + stage_module: torch.nn.Module, + stage_index: int, + pipe_info: PipeInfo, + device: torch.device, + group: dist.ProcessGroup | None = None, + ): + """ + Create a pipeline stage given a stage_module to be wrapped by this stage + and a `pipe_info` describing the stage relationship of the pipeline. + + Args: + stage_module (torch.nn.Module): the module to be wrapped by this stage + stage_index (int): the index of this stage in the pipeline + pipe_info (PipeInfo): information about the pipeline, can be retrieved by `pipe.info()` + device (torch.device): the device to be used by this stage + group (Optional[dist.ProcessGroup]): the process group to be used by this stage + """ + _PipelineStageBase.__init__( + self, + stage_module, + stage_index, + pipe_info.num_stages, + device, + group, + ) + self.pipe_info = pipe_info + + # Find stage nodes in graph + submod_nodes = [ + node for node in pipe_info.graph.nodes if node.op == "call_module" + ] + if len(submod_nodes) != self.num_stages: + raise PipeliningMetadataError( + f"Number of submodules in pipe graph {len(submod_nodes)} does not match number of stages {self.num_stages}" + ) + + # Find my stage node in graph + self.node = submod_nodes[self.stage_index] + self.name = self.node.name + logger.info( + "[%s] Creating PipelineStage %s for %s", + self.group_rank, + stage_index, + self.name, + ) + + # Create mapping from stage name to stage index + self.submod_to_stage_index: dict[str, int] = {} + for i, node in enumerate(submod_nodes): + self.submod_to_stage_index.setdefault(node.name, i) + + # Cast submodule to device + self._move_submod_to_device() + + def _move_submod_to_device(self): + # Move submodule to indicated device if possible + # Note: we cannot move meta module to real devices because meta tensors + # do not support to() method. One needs to do an in-place tensor swap in + # that case. + has_meta_param = any( + is_fake_tensor(p) or p.is_meta for p in self.submod.parameters() + ) + if has_meta_param: + logger.debug("%s Found meta parameters!", self.log_prefix) + else: + self.submod.to(self.device) + + def _prepare_forward_infra( + self, + num_microbatches: int, + args: tuple[Any, ...] | _StageForwardMeta | None, + kwargs: dict[str, Any] | None = None, + has_backward: bool = False, + ) -> _StageForwardMeta | None: + """ + Prepare forward infrastructure for traced pipeline. + + Metadata is created directly from graph placeholders with correct + ``requires_grad`` — received activations get ``requires_grad=True`` + when ``has_backward`` is set, fixing the fact that ``torch.export`` + traces under ``no_grad()``. + + ``_stage_meta.inputs`` is derived from recv infos and aligned with + ``forward_one_chunk``'s ``composite_args``: positional root inputs + on the first stage, received activations only on subsequent stages. + """ + # Step 1: Create recv info for each microbatch. + # _create_act_recv_info is self-contained: it creates _TensorMeta + # directly from graph placeholder values with correct requires_grad. + for chunk in range(num_microbatches): + self.args_recv_info[chunk] = self._create_act_recv_info() + + # Step 2: Derive _stage_meta.inputs from recv infos. + # forward_one_chunk builds composite_args as: + # - First stage: args (positional root inputs, excludes kwargs) + # - Non-first stages: received activations only (no root kwargs) + # _stage_meta.inputs must match composite_args for validation. + recv_infos = self.args_recv_info[0] + if self.is_first: + # All placeholders are root args. Only the first len(args) + # correspond to positional inputs (composite_args); the rest + # are kwargs passed separately via composite_kwargs. + # First stage always receives real tensor args, never _StageForwardMeta. + if not isinstance(args, tuple): + raise AssertionError("First stage requires real tensor args") + n_positional = len(args) + self._stage_meta.inputs = tuple( + info.tensor_meta # type: ignore[misc] + for info in recv_infos[:n_positional] + ) + else: + self._stage_meta.inputs = tuple( + info.tensor_meta # type: ignore[misc] + for info in recv_infos + if not info.is_root_arg + ) + + # Step 3: Create send info and output metadata. + self.act_send_info = self._create_act_send_info() + + return None + + def _prepare_backward_infra( + self, + num_microbatches: int, + loss_fn: Callable[..., torch.Tensor] | None = None, + target: torch.Tensor | None = None, + received_grad_meta: _StageBackwardMeta | None = None, + loss_kwargs: dict[str, Any] | None = None, + ) -> _StageBackwardMeta | None: + """ + Prepare backward infrastructure for traced pipeline. + Derives input_grads metadata from inputs (plain tensors only). + + Note: DTensors are NOT supported in the traced frontend. + """ + # Derive input_grads from inputs (for plain tensors, grad shape == input shape) + if self._stage_meta.inputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: inputs metadata required for backward inference." + ) + + self._stage_meta.input_grads = _derive_grad_metas(self._stage_meta.inputs) + + # Setup backward recv info (calls _create_grad_recv_info which sets output_grads). + # Note: grad_send_info is created lazily in get_bwd_send_ops() since + # it mirrors args_recv_info (already populated during forward). + self._setup_backward_recv_info(num_microbatches) + + return None + + def get_stage_index_of_submod( + self, + submod_name: str, + ): + """ + Given a submodule name, return the stage index of the submodule. + """ + if submod_name not in self.submod_to_stage_index: + raise PipeliningMetadataError(f"Stage id of {submod_name} not found") + + return self.submod_to_stage_index[submod_name] + + def _create_act_recv_info( + self, + ): + """ + Create a tuple of `_RecvInfo` for inputs to the stage. + + Self-contained: creates ``_TensorMeta`` directly from graph + placeholder values with correct ``requires_grad``. + ``torch.export`` traces under ``no_grad()`` so traced metadata + always has ``requires_grad=False``; for received activations we + set ``requires_grad=True`` when ``has_backward`` is set. + + Note: DTensors are NOT supported in the traced frontend. + """ + + def create_recv_tensor(placeholder, arg_node): + example_value = placeholder.meta["val"] + + # Reject DTensors in traced frontend + if isinstance(example_value, DTensor): + raise PipeliningMetadataError( + f"{self.log_prefix} DTensor detected in traced pipeline input " + f"'{placeholder.name}'. DTensor metadata propagation is NOT " + f"supported for the traced frontend (_PipelineStage). " + f"Use the manual PipelineStage frontend for full DTensor support." + ) + + if arg_node.op == "placeholder": + # Root-level placeholder: an input argument to the entire + # model. Keep original metadata from the trace. + return _RecvInfo( + input_name=f"root_input_{placeholder.name}", + source=None, + buffer=None, + tensor_meta=_TensorMeta.from_tensor(example_value), + is_root_arg=True, + ) + + # Received activation from a previous stage. + while arg_node.target is operator.getitem: + arg_node = arg_node.args[0] + + if arg_node.op != "call_module": + raise PipeliningMetadataError( + f"Expecting call_module, got {arg_node.op}" + ) + src_stage = self.get_stage_index_of_submod(arg_node.name) + + # Create metadata directly with correct requires_grad. + # Guard for non-float tensors. + tensor_meta = _TensorMeta( + shape=example_value.shape, + stride=example_value.stride(), + dtype=example_value.dtype, + requires_grad=self.has_backward and example_value.is_floating_point(), + ) + + logger.debug( + "%s Creating recv buffer for input '%s' : %s, %s", + self.log_prefix, + placeholder.name, + tensor_meta.shape, + tensor_meta.dtype, + ) + buffer = _make_tensor_from_meta(tensor_meta, self.device) + if self.has_backward and example_value.is_floating_point(): + buffer.requires_grad_(True) + + return _RecvInfo( + arg_node.name, + src_stage, + buffer, + tensor_meta, + ) + + args_recv_info: list[_RecvInfo] = [] + placeholders = filter( # type: ignore[var-annotated] + lambda node: node.op == "placeholder", # type: ignore[arg-type] + self.submod.graph.nodes, # type: ignore[arg-type,union-attr] + ) + # `placeholders` are nodes internal to submod. + # `self.node.args` are dependency nodes in the outer graph. + # The two are 1:1. + for placeholder, arg_node in zip(placeholders, self.node.args, strict=True): + args_recv_info.append(create_recv_tensor(placeholder, arg_node)) + + logger.debug( + "%s Activation recv / args info: %s", self.log_prefix, args_recv_info + ) + return tuple(args_recv_info) + + def find_dst_rank( + self, + user: fx.Node, + ) -> int | None: + """ + Find the destination rank of a `user` node. + If the `user` is not a submod, `None` may be returned. + """ + if user.op == "call_module": + # User is a stage (`call_module`) + return self.get_stage_index_of_submod(user.name) + else: + # - If user.op == "output": + # No need to send back to rank 0 + # - If user.target is stage_backward: + # No need to send assuming submod output is stored locally or + # should be re-calculated in case of activation checkpointing + return None + + def _create_act_send_info(self): + """ + Create a dict of send info for activations and output metadata. + + Output metadata is created directly with correct ``requires_grad`` + (``torch.export`` traces under ``no_grad()``, so traced values + always have ``requires_grad=False``; at runtime, stage outputs + carry ``requires_grad=True`` when training). + """ + # Output index: List of receiver ranks + act_send_info: dict[int, list] = {} + out_idx = 0 + + for user in self.node.users: + if user.target is operator.getitem: + gi_dsts = act_send_info.setdefault(out_idx, []) + for gi_user in user.users: + dst_rank = self.find_dst_rank(gi_user) + if dst_rank is not None: + gi_dsts.append(dst_rank) + out_idx += 1 + else: + dsts = act_send_info.setdefault(out_idx, []) + dst_rank = self.find_dst_rank(user) + if dst_rank is not None: + dsts.append(dst_rank) + + output_node = self._get_output_node() + output_vals: tuple[torch.Tensor] = tuple( + v.meta["val"] for v in flatten_args(output_node.args) + ) + # Reject DTensors and create output metadata directly with + # correct requires_grad. + output_metas: list[_TensorMeta] = [] + for i, val in enumerate(output_vals): + if isinstance(val, DTensor): + raise PipeliningMetadataError( + f"{self.log_prefix} DTensor detected in traced pipeline output index {i}. " + f"DTensor metadata propagation is NOT supported for the traced frontend " + f"(_PipelineStage). Use the manual PipelineStage frontend for full DTensor support." + ) + output_metas.append( + _TensorMeta( + shape=val.shape, + stride=val.stride(), + dtype=val.dtype, + requires_grad=self.has_backward and val.is_floating_point(), + ) + ) + self._stage_meta.outputs = tuple(output_metas) + + logger.debug("%s Send info: %s", self.log_prefix, act_send_info) + return act_send_info + + def _get_output_node(self): + output_nodes = [node for node in self.submod.graph.nodes if node.op == "output"] # type: ignore[union-attr] + if len(output_nodes) != 1: + raise PipeliningMetadataError( + f"Expected 1 output node, got {len(output_nodes)}" + ) + output_node = output_nodes[0] + return output_node + + def _create_grad_recv_info(self, act_send_info: dict) -> tuple[_RecvInfo, ...]: + """ + Create a tuple of `_RecvInfo` for gradients. + Reuses output metadata from _stage_meta.outputs (populated by _create_act_send_info). + """ + if self._stage_meta.outputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: outputs metadata required for grad recv info. " + f"Ensure _create_act_send_info is called first." + ) + + outputs_meta = self._stage_meta.outputs + output_grads_metas: list[TensorMeta | None] = [] + grad_recv_infos: list[_RecvInfo] = [] + + for out_idx, out_meta in enumerate(outputs_meta): + dst_list = act_send_info.get(out_idx, []) + + # Determine the source stage for gradients + grad_src = dst_list[0] if dst_list else self.stage_index + 1 + + # Check if this output needs gradients + if not dst_list or not out_meta.requires_grad: + output_grads_metas.append(None) + grad_recv_infos.append( + _RecvInfo( + input_name=f"recv_grad_for_{self.stage_index}_none_{out_idx}", + source=grad_src, + buffer=None, + tensor_meta=None, + ) + ) + else: + # Derive grad metadata from output metadata (same shape, requires_grad=False) + grad_meta = _TensorMeta( + shape=out_meta.shape, + stride=out_meta.stride, + dtype=out_meta.dtype, + requires_grad=False, + ) + output_grads_metas.append(grad_meta) + + if len(dst_list) != 1: + raise PipeliningMetadataError( + "Backward of skip connections not supported yet" + ) + + logger.debug( + "%s Creating grad recv buffer for output %s : %s, %s", + self.log_prefix, + out_idx, + grad_meta.shape, + grad_meta.dtype, + ) + + grad_recv_infos.append( + _RecvInfo( + input_name=f"recv_grad_for_{self.stage_index}_from_{grad_src}", + source=grad_src, + buffer=_make_tensor_from_meta(grad_meta, self.device), + tensor_meta=grad_meta, + ) + ) + + self._stage_meta.output_grads = tuple(output_grads_metas) + if self._stage_meta.inputs is not None: + self._stage_meta.input_grads = _derive_grad_metas(self._stage_meta.inputs) + logger.debug("%s Grad recv info: %s", self.log_prefix, grad_recv_infos) + return tuple(grad_recv_infos) + + +# A helper function to create a pipeline stage based on traced pipeline information +def build_stage( + stage_module: torch.nn.Module, + stage_index: int, + pipe_info: PipeInfo, + device: torch.device, + group: dist.ProcessGroup | None = None, +) -> _PipelineStage: + """ + Create a pipeline stage given a stage_module to be wrapped by this stage + and pipeline information. + + Args: + stage_module (torch.nn.Module): the module to be wrapped by this stage + stage_index (int): the index of this stage in the pipeline + pipe_info (PipeInfo): information about the pipeline, can be retrieved by `pipe.info()` + device (torch.device): the device to be used by this stage + group (Optional[dist.ProcessGroup]): the process group to be used by this stage + + Returns: + _PipelineStage: a pipeline stage that can run with `PipelineSchedules`. + """ + return _PipelineStage( + stage_module, + stage_index, + pipe_info, + device, + group, + ) + + +class PipelineStage(_PipelineStageBase): + """A pipeline stage for pipeline parallelism with sequential model partitioning. + + Supports both **static** and **dynamic** metadata inference: + + Static mode: + All of ``input_args``, ``output_args`` (and ``input_grads``/``output_grads`` + when DTensors are present) are provided at construction time. + + Dynamic mode: + Metadata is inferred from the first microbatch at runtime; any + statically provided args are used for validation only. + + Args: + submodule: The ``nn.Module`` wrapped by this stage. + stage_index: Zero-based stage ID. + num_stages: Total number of stages in the pipeline. + device: Device this stage runs on. + input_args: Example input tensors (single tensor or tuple). Optional. + output_args: Example output tensors. Optional. + output_grads: Example output gradients (received from next stage). Optional. + input_grads: Example input gradients (sent to previous stage). Optional. + group: Process group for P2P communication. Defaults to the + world process group. + dw_builder: Builder for deferred weight-update runners used by + zero-bubble (F/I/W) schedules. + get_mesh: `GetMeshCallback` used during + dynamic DTensor inference. Ignored in fully static DTensor mode. + """ + + def __init__( + self, + submodule: nn.Module, + stage_index: int, + num_stages: int, + device: torch.device, + input_args: torch.Tensor | tuple[torch.Tensor, ...] | None = None, + output_args: torch.Tensor | tuple[torch.Tensor, ...] | None = None, + output_grads: torch.Tensor | tuple[torch.Tensor | None, ...] | None = None, + input_grads: torch.Tensor | tuple[torch.Tensor | None, ...] | None = None, + group: dist.ProcessGroup | None = None, + dw_builder: Callable[[], Callable[..., None]] | None = None, + get_mesh: GetMeshCallback | None = None, + ): + super().__init__(submodule, stage_index, num_stages, device, group, dw_builder) + + self._mesh_cache = _MeshCache(get_mesh_cb=get_mesh) + self._inference_mode: InferenceMode | None = None + self._fwd_outputs_for_bwd_meta: tuple[torch.Tensor, ...] | None = None + self._fwd_inputs_for_bwd_meta: tuple[torch.Tensor, ...] | None = None + self._fwd_kwargs_tensors_for_bwd_meta: tuple[torch.Tensor, ...] | None = None + self._metadata_inference_buffer_backup: ( + list[tuple[torch.Tensor, torch.Tensor]] | None + ) = None + + # Validate and normalize args to tuples + inputs = validate_and_normalize_to_tuple(input_args) + outputs = validate_and_normalize_to_tuple(output_args) + in_grads = validate_and_normalize_to_tuple(input_grads, allow_none=True) + out_grads = validate_and_normalize_to_tuple(output_grads, allow_none=True) + + self._user_meta = _StageMeta( + inputs=extract_tensor_metas(inputs), + outputs=extract_tensor_metas(outputs), + input_grads=extract_tensor_metas(in_grads, allow_none=True), + output_grads=extract_tensor_metas(out_grads, allow_none=True), + ) + + # Cache meshes from user-provided DTensors + for args in (inputs, outputs, in_grads, out_grads): + if args is not None: + self._mesh_cache.update_from_tensors(args) + + # Validate DTensor↔grad correspondence independently for inputs and outputs + if self._user_meta.has_dtensors(): + if inputs and in_grads: + validate_static_arg_grad_correspondence( + self.stage_index, inputs, in_grads, is_input=True + ) + if outputs and out_grads: + validate_static_arg_grad_correspondence( + self.stage_index, outputs, out_grads, is_input=False + ) + + def _recv_meta(self, src_stage: int) -> Any: + """Receive metadata object from a stage on a different rank via P2P.""" + objects: list[Any] = [None] + dist.recv_object_list( + objects, + src=self._resolve_peer_global_rank(src_stage), + group=self.group, + device=self.device, + use_batch=True, + ) + if len(objects) != 1: + raise PipeliningMetadataError( + f"Expected exactly one object to be received but got: {len(objects)}" + ) + return objects[0] + + def _send_meta(self, meta: Any, dst_stage: int) -> None: + """Send metadata object to a stage on a different rank via P2P.""" + dist.send_object_list( + [meta], + dst=self._resolve_peer_global_rank(dst_stage), + group=self.group, + device=self.device, + use_batch=True, + ) + + def _is_same_rank(self, other_stage: int) -> bool: + """Check if another stage is on the same rank as this stage.""" + return self.stage_index_to_group_rank[other_stage] == self.group_rank + + def _warmup_forward_vote( + self, has_backward: bool, received_acc: torch.Tensor | None = None + ) -> torch.Tensor: + """Forward phase of the warm-up vote protocol (stage 0 → N−1). + + Each stage computes a vote (1 = STATIC, 0 = DYNAMIC) based on + ``InferenceMode.needs_dynamic``, multiplies it with the accumulated + product from the previous stage, and forwards the result to the next + stage. The final product at stage N−1 is 1 iff *every* stage voted + STATIC. + + Args: + has_backward: Whether the schedule includes a backward pass. + received_acc: Accumulated product tensor from the previous + same-rank stage (V-schedule), or ``None`` for the first + stage / cross-rank. + + Returns: + The accumulated product tensor after this stage's vote. + """ + my_vote = 0 if InferenceMode.needs_dynamic(self._user_meta, has_backward) else 1 + + my_vote_t = torch.tensor([my_vote], dtype=torch.int32, device=self.device) + + if self.is_first: + acc = my_vote_t + elif self._is_same_rank(self.stage_index - 1): + if received_acc is None: + raise AssertionError("expected received_acc to be not None") + acc = received_acc * my_vote_t + else: + peer_global = self._resolve_peer_global_rank(self.stage_index - 1) + acc = torch.zeros(1, dtype=torch.int32, device=self.device) + dist.recv(acc, src=peer_global, group=self.group) + acc = acc * my_vote_t + + if not self.is_last and not self._is_same_rank(self.stage_index + 1): + peer_global = self._resolve_peer_global_rank(self.stage_index + 1) + dist.send(acc, dst=peer_global, group=self.group) + + return acc + + def _warmup_backward_result( + self, received_result: torch.Tensor | None = None + ) -> torch.Tensor: + """Backward phase of the warm-up vote protocol (stage N−1 → 0). + + Propagates the final accumulated product (computed in the forward + phase) back through the pipeline so every stage learns the global + inference mode. + + Args: + received_result: Result tensor from the next same-rank stage + (V-schedule), or ``None`` for the last stage / cross-rank. + + Returns: + The global vote result tensor for this stage. + """ + if self.is_last or self._is_same_rank(self.stage_index + 1): + if received_result is None: + raise AssertionError("expected received_result to be not None") + result = received_result + else: + peer_global = self._resolve_peer_global_rank(self.stage_index + 1) + result = torch.zeros(1, dtype=torch.int32, device=self.device) + dist.recv(result, src=peer_global, group=self.group) + + if not self.is_first and not self._is_same_rank(self.stage_index - 1): + peer_global = self._resolve_peer_global_rank(self.stage_index - 1) + dist.send(result, dst=peer_global, group=self.group) + + return result + + def _compute_outputs( + self, + *args: torch.Tensor, + module: torch.nn.Module, + **kwargs: Any, + ) -> torch.Tensor | tuple[torch.Tensor, ...] | list[torch.Tensor] | None: + """Compute outputs of the submodule.""" + return module(*args, **kwargs) + + def _compute_input_grads( + self, + outputs: list[torch.Tensor], + all_fwd_inputs: list[torch.Tensor], + grad_outputs: list[torch.Tensor | None] | None = None, + ) -> tuple[torch.Tensor | None, ...]: + """Compute input gradients via :func:`_autograd_grad_for_inputs`.""" + return _autograd_grad_for_inputs( + outputs, + all_fwd_inputs, + grad_outputs, + allow_unused=True, + ) + + def _to_tensor(self, arg: torch.Tensor | TensorMeta) -> torch.Tensor: + """Convert a tensor or metadata to a real tensor on ``self.device``. + + Real tensors are detached and re-set requires_grad to create a fresh + autograd leaf, isolating metadata inference from the user's graph. + TensorMeta is materialized as an empty tensor (or DTensor via mesh cache). + """ + if isinstance(arg, torch.Tensor): + return arg.detach().requires_grad_(arg.requires_grad) + elif isinstance(arg, TensorMeta): + if isinstance(arg, _DTensorMeta): + mesh = self._mesh_cache.get_mesh(arg.mesh_cache_key) + return arg.to_dtensor(self.device, mesh) + else: + return arg.to_tensor(self.device) + else: + raise PipeliningMetadataError( + f"Unsupported type {type(arg)} for _to_tensor: {arg}" + ) + + def _ones_from_metadata(self, meta: TensorMeta) -> torch.Tensor: + """Create a ones tensor from metadata for backward inference grad_outputs.""" + local_ones = torch.ones( + meta.shape, + dtype=meta.dtype, + device=self.device, + ) + if isinstance(meta, _DTensorMeta): + mesh = self._mesh_cache.get_mesh(meta.mesh_cache_key) + return DTensor.from_local( + local_ones, + device_mesh=mesh, + placements=meta.placements, + shape=meta.global_shape, + stride=meta.global_stride, + run_check=False, + ) + return local_ones + + def _pre_metadata_inference_backup(self) -> None: + """Back up state that dynamic metadata inference must not publish. + + Dynamic metadata inference executes representative real forwards and + backwards before the first runtime microbatch. Those executions are + allowed to observe module buffers, but they must not publish training + state updates such as running counters or routing statistics. + """ + if self._inference_mode != InferenceMode.DYNAMIC: + return + if self._metadata_inference_buffer_backup is not None: + raise RuntimeError( + "Metadata inference buffer backup already exists. " + "Run _post_metadata_inference_cleanup before backing up again." + ) + self._metadata_inference_buffer_backup = [ + (buffer, buffer.detach().clone()) + for _, buffer in self.submod.named_buffers(remove_duplicate=False) + ] + + def _forward_metadata_inference( + self, + args: tuple[torch.Tensor, ...] | _StageForwardMeta | None, + kwargs: dict[str, Any] | None = None, + has_backward: bool = False, + ) -> _StageForwardMeta | None: + """Run forward metadata inference (Stage 0 → N). + + Args: + args: Real tensors (first stage), ``_StageForwardMeta`` + (same-rank), or ``None`` (cross-rank P2P). + kwargs: Keyword arguments forwarded to the submodule. + has_backward: Whether backward inference follows. + + Returns: + ``_StageForwardMeta`` for the next stage, or ``None`` if sent via P2P. + """ + kwargs = kwargs or {} + + # === RECEIVE: Get input metadata and create meta tensors === + if self.is_first: + # First stage: extract metadata from real tensors + if args is None or isinstance(args, _StageForwardMeta): + raise PipeliningMetadataError( + f"Stage {self.stage_index}: First stage requires real tensors, " + f"got {type(args).__name__}." + ) + tensor_args = validate_and_normalize_to_tuple(args) + if tensor_args is None: + raise AssertionError("expected tensor_args to be not None") + self._stage_meta.inputs = extract_tensor_metas(tensor_args) + inference_args = tuple(self._to_tensor(a) for a in tensor_args) + elif self._is_same_rank(self.stage_index - 1): + # Same-rank: _StageForwardMeta passed via argument + if not isinstance(args, _StageForwardMeta): + raise PipeliningMetadataError( + f"Stage {self.stage_index}: Expected _StageForwardMeta from same-rank " + f"previous stage, got {type(args).__name__}." + ) + self._stage_meta.inputs = args.forward_metas + inference_args = tuple(self._to_tensor(m) for m in args.forward_metas) + else: + # Cross-rank: receive _StageForwardMeta via P2P + recv_meta = self._recv_meta(self.stage_index - 1) + if not isinstance(recv_meta, _StageForwardMeta): + raise PipeliningMetadataError( + f"Stage {self.stage_index}: Expected _StageForwardMeta from P2P, " + f"got {type(recv_meta).__name__}." + ) + self._stage_meta.inputs = recv_meta.forward_metas + inference_args = tuple(self._to_tensor(m) for m in recv_meta.forward_metas) + + inference_kwargs = { + k: self._to_tensor(v) if isinstance(v, torch.Tensor) else v + for k, v in kwargs.items() + } + + # Isolate metadata inference from user's grad context. + # has_backward → enable_grad() so backward tracing sees grad_fn; + # no backward → no_grad() for cross-rank consistency. + ctx = torch.enable_grad() if has_backward else torch.no_grad() + with ctx: + outputs = self._compute_outputs( + *inference_args, module=self.submod, **inference_kwargs + ) + + # Normalize outputs to tuple + outputs = validate_and_normalize_to_tuple(outputs) + + self._stage_meta.outputs = extract_tensor_metas(outputs) + + # Store for backward metadata inference (always, even during eval) + fwd_kwargs_tensors = tuple( + v for v in flatten_args(inference_kwargs) if isinstance(v, torch.Tensor) + ) + self._fwd_outputs_for_bwd_meta = outputs + self._fwd_inputs_for_bwd_meta = inference_args + self._fwd_kwargs_tensors_for_bwd_meta = fwd_kwargs_tensors + + # === SEND: Pass output metadata to next stage === + if self._stage_meta.outputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output metadata is required for forward inference." + ) + fwd_meta = _StageForwardMeta(forward_metas=self._stage_meta.outputs) + + if self.is_last or self._is_same_rank(self.stage_index + 1): + # Same-rank or last: return for caller to pass + return fwd_meta + else: + # Cross-rank: send via P2P + self._send_meta(fwd_meta, self.stage_index + 1) + return None + + def _backward_metadata_inference( + self, + loss_fn: Callable[..., torch.Tensor] | None = None, + target: torch.Tensor | None = None, + received_grad_meta: _StageBackwardMeta | None = None, + loss_kwargs: dict[str, Any] | None = None, + ) -> _StageBackwardMeta | None: + """Run backward metadata inference (Stage N → 0). + + Args: + loss_fn: Loss function (required for the last stage). + target: Target tensor (required for the last stage). + received_grad_meta: Grad metadata from next same-rank stage + (V-schedule only). + loss_kwargs: Extra keyword arguments forwarded to the loss + function (e.g. ``global_valid_tokens``). + + Returns: + ``_StageBackwardMeta`` for the previous stage, or ``None`` if sent via P2P. + """ + fwd_outputs = self._fwd_outputs_for_bwd_meta + fwd_inputs = self._fwd_inputs_for_bwd_meta + if fwd_outputs is None or fwd_inputs is None: + raise PipeliningMetadataError( + "Backward metadata inference requires forward metadata inference to run first" + ) + kwargs_tensors = self._fwd_kwargs_tensors_for_bwd_meta or () + all_fwd_inputs = list(fwd_inputs) + list(kwargs_tensors) + # Clear temporary storage early — local refs are sufficient from here + self._fwd_outputs_for_bwd_meta = None + self._fwd_inputs_for_bwd_meta = None + self._fwd_kwargs_tensors_for_bwd_meta = None + # === RECEIVE: Get output grad metadata (except last stage) === + if self.is_last: + if loss_fn is None or target is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: loss_fn and target required for last stage" + ) + inference_target = self._to_tensor(target) + loss = loss_fn( + fwd_outputs[0] if len(fwd_outputs) == 1 else fwd_outputs, + inference_target, + **(loss_kwargs or {}), + ) + self._stage_meta.output_grads = None + all_input_grads = self._compute_input_grads( + [loss], + all_fwd_inputs, + ) + else: + # Non-last stage: receive grad metadata from next stage + if self._is_same_rank(self.stage_index + 1): + # Same-rank: _StageBackwardMeta passed via argument + if not isinstance(received_grad_meta, _StageBackwardMeta): + raise PipeliningMetadataError( + f"Stage {self.stage_index}: Expected _StageBackwardMeta from same-rank " + f"next stage, got {type(received_grad_meta).__name__}." + ) + self._stage_meta.output_grads = received_grad_meta.backward_metas + else: + # Cross-rank: receive _StageBackwardMeta via P2P + recv_meta = self._recv_meta(self.stage_index + 1) + if not isinstance(recv_meta, _StageBackwardMeta): + raise PipeliningMetadataError( + f"Stage {self.stage_index}: Expected _StageBackwardMeta from P2P, " + f"got {type(recv_meta).__name__}." + ) + self._stage_meta.output_grads = recv_meta.backward_metas + + # === COMPUTE: Build grad_outputs and compute input grads === + # Extract output tensors and corresponding grad_outputs from metadata + # Must iterate together to maintain alignment + if self._stage_meta.output_grads is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output_grads metadata is required for backward inference." + ) + stage_output_grad_metas = self._stage_meta.output_grads + + filtered_fwd_outputs: list[torch.Tensor] = [] + filtered_output_grads: list[torch.Tensor | None] = [] + + for idx, (fwd_out, grad_meta) in enumerate( + zip(fwd_outputs, stage_output_grad_metas, strict=True) + ): + # Match _backward.py behavior: skip if output doesn't require grad AND has no grad_fn + if not fwd_out.requires_grad: + if grad_meta is not None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output {idx} requires_grad=False, " + f"but output_grads metadata is provided: {grad_meta}." + ) + continue + filtered_fwd_outputs.append(fwd_out) + # For outputs that require grad, include them even if grad_meta is None + # (runtime passes None grad_outputs to autograd.backward in this case) + filtered_output_grads.append( + self._ones_from_metadata(grad_meta) if grad_meta else None + ) + + if filtered_fwd_outputs: + all_input_grads = self._compute_input_grads( + filtered_fwd_outputs, all_fwd_inputs, filtered_output_grads + ) + # Free intermediate references early + filtered_fwd_outputs.clear() + filtered_output_grads.clear() + all_fwd_inputs.clear() + # Only positional input grads flow to previous stage + else: + all_input_grads = tuple(None for _ in range(len(all_fwd_inputs))) + + input_grads = all_input_grads[: len(fwd_inputs)] + processed_metas: list[TensorMeta | None] = [] + for i, g in enumerate(input_grads): + if isinstance(g, torch.Tensor): + processed_metas.append(extract_tensor_meta(g)) + elif fwd_inputs[i].requires_grad: + processed_metas.append( + _derive_grad_metas((extract_tensor_meta(fwd_inputs[i]),))[0] + ) + else: + processed_metas.append(None) + self._stage_meta.input_grads = tuple(processed_metas) + + # === SEND: Pass input grad metadata to previous stage === + bwd_meta = _StageBackwardMeta(backward_metas=self._stage_meta.input_grads) + + if self.is_first or self._is_same_rank(self.stage_index - 1): + # First rank or Same-rank: return for caller to pass + return bwd_meta + else: + # Cross-rank: send via P2P + self._send_meta(bwd_meta, self.stage_index - 1) + return None + + def _post_metadata_inference_cleanup(self) -> None: + """Restore state and clean up side effects from metadata inference. + + Buffer restore is intentionally delayed until after both forward and + backward metadata inference. Restoring immediately after forward can + bump autograd version counters for buffers saved by the forward graph. + """ + buffer_backup = self._metadata_inference_buffer_backup + if buffer_backup is not None: + with torch.no_grad(): + for buffer, saved_buffer in buffer_backup: + buffer.copy_(saved_buffer) + self._metadata_inference_buffer_backup = None + + # Clear stored inference tensors (frees autograd graph + activations) + self._fwd_outputs_for_bwd_meta = None + self._fwd_inputs_for_bwd_meta = None + self._fwd_kwargs_tensors_for_bwd_meta = None + + # Metadata inference runs real fwd/bwd, which unshards FSDP params and + # accumulates grads. Reshard to free memory and clear stale grads. + for module in self.submod.modules(): + if isinstance(module, FSDPModule): + module.reshard() + for param in module.parameters(): + param.grad = None + + def _prepare_backward_infra( + self, + num_microbatches: int, + loss_fn: Callable[..., torch.Tensor] | None = None, + target: torch.Tensor | None = None, + received_grad_meta: "_StageBackwardMeta | None" = None, + loss_kwargs: dict[str, Any] | None = None, + ) -> "_StageBackwardMeta | None": + """Run backward metadata inference and prepare backward infrastructure. + + Returns: + ``_StageBackwardMeta`` for the previous same-rank stage, or ``None``. + """ + grad_meta_result: _StageBackwardMeta | None = None + if self._inference_mode == InferenceMode.DYNAMIC: + # DYNAMIC mode: run backward metadata inference + # received_grad_meta is used for same-rank V-schedule stages + grad_meta_result = self._backward_metadata_inference( + loss_fn=loss_fn, + target=target, + received_grad_meta=received_grad_meta, + loss_kwargs=loss_kwargs, + ) + # Validate dynamically inferred metadata against user-provided metadata + self._validate_inferred_metadata() + else: + # STATIC mode: metadata comes from user inputs, no validation needed + self._stage_meta.input_grads = self._user_meta.input_grads + self._stage_meta.output_grads = self._user_meta.output_grads + # For STATIC mode with plain tensors, if output_grads is not set but + # we have outputs, derive output_grads from outputs. + # (gradient shape == output shape, but requires_grad=False for gradients) + if self._stage_meta.output_grads is None: + if self._stage_meta.outputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output metadata is required for backward inference." + ) + self._stage_meta.output_grads = _derive_grad_metas( + self._stage_meta.outputs + ) + # Similarly, derive input_grads from inputs if not provided + if self._stage_meta.input_grads is None: + if self._stage_meta.inputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: input metadata is required for backward inference." + ) + self._stage_meta.input_grads = _derive_grad_metas( + self._stage_meta.inputs + ) + + # Note: grad_send_info is created lazily in get_bwd_send_ops() since + # it mirrors args_recv_info (already populated during forward). + self._setup_backward_recv_info(num_microbatches) + return grad_meta_result + + def _validate_inferred_metadata(self) -> None: + """Validate dynamically inferred metadata against user-provided metadata.""" + pairs = [ + (self._user_meta.inputs, self._stage_meta.inputs, "input"), + (self._user_meta.outputs, self._stage_meta.outputs, "output"), + (self._user_meta.input_grads, self._stage_meta.input_grads, "input_grad"), + ( + self._user_meta.output_grads, + self._stage_meta.output_grads, + "output_grad", + ), + ] + for user_val, stage_val, label in pairs: + if user_val and stage_val: + validate_tensors_metadata( + f"Stage {self.stage_index} {label}", + user_val, + stage_val, + warn_on_mismatch=True, + ) + + def _prepare_forward_infra( + self, + num_microbatches: int, + args: tuple[Any, ...] | _StageForwardMeta | None, + kwargs: dict[str, Any] | None = None, + has_backward: bool = False, + ) -> _StageForwardMeta | None: + """Prepare the stage infrastructure for forward pass. + + Returns: + ``_StageForwardMeta`` for next stage (same-rank), or ``None`` if sent via P2P. + """ + if self._inference_mode is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: inference mode not set. " + f"Run warmup vote protocol first." + ) + + fwd_meta_output: _StageForwardMeta | None = None + + if self._inference_mode == InferenceMode.DYNAMIC: + # DYNAMIC mode: run forward metadata inference + # args may be _StageForwardMeta for same-rank V-schedule stages + fwd_meta_output = self._forward_metadata_inference( + args, kwargs, has_backward + ) + # Validate dynamically inferred metadata against user-provided metadata + self._validate_inferred_metadata() + # STATIC mode: metadata comes from user inputs, no validation needed + else: + self._stage_meta.inputs = self._user_meta.inputs + self._stage_meta.outputs = self._user_meta.outputs + + # Setup recv and send info + self._setup_forward_recv_info(num_microbatches, has_backward) + self._setup_forward_send_info() + + return fwd_meta_output + + def _setup_forward_recv_info( + self, num_microbatches: int, has_backward: bool + ) -> None: + """Setup receive info for forward pass.""" + if self._stage_meta.inputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: inputs metadata required for recv info." + ) + for chunk_id in range(num_microbatches): + if self.is_first: + # First stage: all inputs are root arguments (no recv needed) + self.args_recv_info[chunk_id] = tuple( + _RecvInfo( + input_name=f"root_input_{idx}", + source=None, + buffer=None, + tensor_meta=meta, + is_root_arg=True, + ) + for idx, meta in enumerate(self._stage_meta.inputs) + ) + else: + # Non-first stages: receive from previous stage + self.args_recv_info[chunk_id] = tuple( + _RecvInfo( + input_name=f"recv_for_{self.stage_index}_from_{self.stage_index - 1}", + source=self.stage_index - 1, + buffer=_make_tensor_from_meta(meta, self.device), + tensor_meta=meta, + ) + for meta in self._stage_meta.inputs + ) + + def _setup_forward_send_info(self) -> None: + """Setup send info for forward pass.""" + self.act_send_info: dict[int, list] = {} + if self._stage_meta.outputs is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: outputs metadata required for recv info." + ) + for idx in range(len(self._stage_meta.outputs)): + self.act_send_info[idx] = [self.stage_index + 1] if not self.is_last else [] + + def _create_grad_recv_info( + self, + act_send_info: dict, + ) -> tuple[_RecvInfo, ...]: + grad_recv_infos: list[_RecvInfo] = [] + if not self.is_last: + # Ensure output_grads metadata is available + if self._stage_meta.output_grads is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output_grads metadata is required for " + f"creating grad recv info. Ensure backward metadata is populated." + ) + + # Receiving gradients from multiple sources is not supported + # hence we only take the first destination + # Use a helper function to safely extract the metadata + output_grads = self._stage_meta.output_grads + for idx, dst_list in act_send_info.items(): + if dst_list is None: + raise PipeliningMetadataError( + f"Stage {self.stage_index}: output {idx} is not sent to any stage." + ) + src = dst_list[0] + grad_meta = output_grads[idx] + grad_recv_infos.append( + _RecvInfo( + input_name=f"recv_grad_for_{self.stage_index}_from_{src}", + source=src, + buffer=_make_tensor_from_meta(grad_meta, self.device) + if grad_meta + else None, + tensor_meta=grad_meta, + ) + ) + return tuple(grad_recv_infos) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/remote_device.py b/.venv/lib/python3.12/site-packages/torch/distributed/remote_device.py new file mode 100644 index 0000000000000000000000000000000000000000..c03727a33584ab17f3bb766aecc129b6b25f05b6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/remote_device.py @@ -0,0 +1,119 @@ +# mypy: allow-untyped-defs + +import torch + + +class _remote_device: + """ + Represents a device on a remote worker. + + Args: + remote_device (str or torch.device): Represents a device on a remote worker. + The string format should be one of the following: + + 1. "/", where the device field can be parsed as torch.device type. + E.g., "trainer0/cpu", "trainer0", "ps0/cuda:0". + In addition, the device field can be optional and the default value is "cpu". + 2. "rank:/", where is the rank of the + process and device can be parsed as torch.device type. + E.g., "rank:0/cpu", "rank:0", "rank:0/cuda:0" + 3. and are optional and formats like "cpu" + and "cuda:1", just represent local devices. + """ + + def __init__(self, remote_device: str | torch.device): + PARSE_ERROR = ( + f"Could not parse remote_device: {remote_device}. The valid format is " + "'/' or 'rank:/' or ''" + ) + self._worker_name = None + self._rank = None + self._device: str | int | torch.device | None = None + + if isinstance(remote_device, torch.device): + self._device = remote_device + elif isinstance(remote_device, str): + fields = remote_device.split("/") + if len(fields) == 2: + self._worker_name, self._device = fields + elif len(fields) == 1: + # Check if this is a valid device. + if _remote_device._is_valid_local_device(fields[0]): + self._device = fields[0] + else: + self._worker_name = fields[0] + self._device = "cpu" + else: + raise ValueError(PARSE_ERROR) + else: + raise TypeError(f"Invalid type for remote_device: {type(remote_device)}") + + # Do some basic sanity check (no empty string) + if self._worker_name is not None and not self._worker_name: + raise ValueError(PARSE_ERROR) + + # Validate the device. + self._device = torch.device(self._device) + + # Check for rank based format. + if self._worker_name is not None: + fields = self._worker_name.split(":") + if len(fields) == 2: + # rank:/device format, extract rank + if fields[0] == "rank" and fields[1].isdigit(): + self._rank = int(fields[1]) # type: ignore[assignment] + self._worker_name = None + else: + raise ValueError(PARSE_ERROR) + elif len(fields) > 2: + raise ValueError(PARSE_ERROR) + + @staticmethod + def _is_valid_local_device(device): + # Check for torch.device + try: + torch.device(device) + return True + except Exception: + return False + + def worker_name(self) -> str | None: + """Return the name of remote worker representing the remote device and ``None`` if no worker name is available.""" + return self._worker_name + + def rank(self) -> int | None: + """ + Returns the rank of remote worker representing the remote device. + Returns ``None`` if no rank is available. + """ + return self._rank + + def device(self) -> torch.device: + """Return the local device on the remote worker.""" + return self._device # type: ignore[return-value] + + def __repr__(self): + if self._device is not None: + if self._worker_name is not None: + return f"{self._worker_name}/{self._device}" + elif self._rank is not None: + return f"rank:{self._rank}/{self._device}" + else: + return str(self._device) + else: + if self._worker_name is not None: + return f"{self._worker_name}" + elif self._rank is not None: + return f"{self._rank}" + else: + raise RuntimeError("Invalid state!") + + def __eq__(self, other): + return isinstance(other, _remote_device) and ( + self._worker_name == other._worker_name + and self._device == other._device + and self._rank == other._rank + ) + + def __hash__(self): + return hash(self._worker_name) ^ hash(self._device) ^ hash(self._rank) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rendezvous.py b/.venv/lib/python3.12/site-packages/torch/distributed/rendezvous.py new file mode 100644 index 0000000000000000000000000000000000000000..8013307e2b00022d577358f117ebe06b57cb9d92 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rendezvous.py @@ -0,0 +1,308 @@ +# mypy: allow-untyped-defs +try: + from urllib.parse import urlparse, urlunparse +except ImportError as e: + raise ImportError( + "urllib cannot be found, urlparse from python2 is no longer supported." + ) from e + +import numbers +import os +import sys +from collections.abc import Callable, Iterator +from datetime import timedelta + +from torch.distributed import FileStore, Store, TCPStore + +from .constants import default_pg_timeout + + +_rendezvous_handlers: dict[str, Callable[..., Iterator[tuple[Store, int, int]]]] = {} + +__all__ = ["register_rendezvous_handler", "rendezvous"] + + +def register_rendezvous_handler(scheme, handler): + """ + Register a new rendezvous handler. + + Before we can run collective algorithms, participating processes + need to find each other and exchange information to be able to + communicate. We call this process rendezvous. + + The outcome of the rendezvous process is a triplet containing a + shared key/value store, the rank of the process, and the total + number of participating processes. + + If none of the bundled rendezvous methods apply to your execution + environment you can opt to register your own rendezvous handler. + Pick a unique name and use the URL scheme to identify it when + calling the `rendezvous()` function. + + Args: + scheme (str): URL scheme to identify your rendezvous handler. + handler (function): Handler that is invoked when the + `rendezvous()` function is called with a URL that uses + the corresponding scheme. It must be a generator function + that yields the triplet. + """ + global _rendezvous_handlers + if scheme in _rendezvous_handlers: + raise RuntimeError(f"Rendezvous handler for {scheme}:// already registered") + _rendezvous_handlers[scheme] = handler + + +# Query will have format "rank=0&world_size=1" and is +# converted into {"rank": 0, "world_size": 1} +def _query_to_dict(query: str) -> dict[str, str]: + return { + pair[0]: pair[1] + for pair in (pair.split("=") for pair in filter(None, query.split("&"))) + } + + +def _get_use_libuv_from_query_dict(query_dict: dict[str, str]) -> bool: + # libuv is the default backend for TCPStore. To enable the non-libuv backend, + # user can explicitly specify ``use_libuv=0`` in the URL parameter. + if sys.platform == "win32": + # PyTorch is built without libuv support on windows, so default to 0 + return query_dict.get("use_libuv", os.environ.get("USE_LIBUV", "0")) == "1" + return query_dict.get("use_libuv", os.environ.get("USE_LIBUV", "1")) == "1" + + +def _rendezvous_helper(url: str, rank: int, world_size_opt: int | None, **kwargs): + result = urlparse(url) + if world_size_opt is None: + world_size = -1 + if result.scheme == "env": + rank = int(os.environ.get("RANK", rank)) + # If the world_size env variable is not present then it is a dynamic group + world_size = int(os.environ.get("WORLD_SIZE", world_size)) + else: + world_size = world_size_opt + if rank != -1 or world_size != -1 or world_size_opt is None: + query_dict = _query_to_dict(result.query) + if "rank" in query_dict or "world_size" in query_dict: + raise AssertionError( + f"The url: {url} has node-specific arguments(rank, world_size) already." + ) + if rank != -1: + query_dict["rank"] = str(rank) + if world_size != -1 or world_size_opt is None: + query_dict["world_size"] = str(world_size) + result = result._replace( + query=f"{'&'.join([f'{k}={v}' for k, v in query_dict.items()])}" + ) + # pyrefly: ignore [bad-assignment] + url = urlunparse(result) + + if result.scheme not in _rendezvous_handlers: + raise RuntimeError(f"No rendezvous handler for {result.scheme}://") + return _rendezvous_handlers[result.scheme](url, **kwargs) + + +def rendezvous(url: str, rank: int = -1, world_size: int = -1, **kwargs): + if not isinstance(url, (str, bytes)): + raise RuntimeError(f"`url` must be a string. {type(url)}: {url}") + + if not isinstance(rank, numbers.Integral): + raise RuntimeError(f"`rank` must be an integer. {rank}") + + if not isinstance(world_size, numbers.Integral): + raise RuntimeError(f"`world_size` must be an integer. {world_size}") + + return _rendezvous_helper(url, rank, world_size, **kwargs) + + +def _create_store_from_options(backend_options, rank): + store, _, _ = next(_rendezvous_helper(backend_options.init_method, rank, None)) + return store + + +def _rendezvous_error(msg): + return ValueError("Error initializing torch.distributed using " + msg) + + +def _file_rendezvous_handler(url: str, **kwargs): + def _error(msg): + return _rendezvous_error("file:// rendezvous: " + msg) + + result = urlparse(url) + path = result.path + if sys.platform == "win32": + import urllib.request + + full_path = result.netloc + result.path + path = urllib.request.url2pathname(full_path) + if path: + # Normalizing an empty string produces ".", which is not expected. + path = os.path.normpath(path) + + if not path: + raise _error("path missing") + query_dict = _query_to_dict(result.query) + if "rank" not in query_dict: + raise _error("rank parameter missing") + if "world_size" not in query_dict: + raise _error("world size parameter missing") + + rank = int(query_dict["rank"]) + world_size = int(query_dict["world_size"]) + store = FileStore(path, world_size) + yield (store, rank, world_size) + + # If this configuration is invalidated, there is nothing we can do about it + raise RuntimeError("Unable to perform rerendezvous using file:// method") + + +def _torchelastic_use_agent_store() -> bool: + return os.environ.get("TORCHELASTIC_USE_AGENT_STORE", None) == str(True) + + +def _agent_store_serves(hostname, port) -> bool: + """Whether the elastic agent is the one serving a store at ``hostname:port``. + + ``TORCHELASTIC_USE_AGENT_STORE`` only promises a server at the agent's own + ``MASTER_ADDR:MASTER_PORT``, which the agent exports alongside it. Honoring + the flag for any other endpoint makes every rank a client of a store nobody + hosts. ``TCPStore``'s C++ constructor applies the same address check before + it forgives a failed server bind. + """ + return ( + _torchelastic_use_agent_store() + and os.environ.get("MASTER_ADDR") == hostname + and os.environ.get("MASTER_PORT") == str(port) + ) + + +def _create_c10d_store( + hostname, port, rank, world_size, timeout, use_libuv=True +) -> Store: + """ + Smartly creates a c10d Store object on ``rank`` based on whether we need to reuse agent store. + + The TCPStore server is assumed to be hosted + on ``hostname:port``. + + By default, the TCPStore server uses the asynchronous implementation + ``LibUVStoreDaemon`` which utilizes libuv. + + If ``torchelastic_use_agent_store()`` is ``True`` *and* ``hostname:port`` is + the agent's own ``MASTER_ADDR:MASTER_PORT``, then it is assumed that the + agent leader (node rank 0) hosts the TCPStore server there. Hence + ALL ranks will create and return a TCPStore client (e.g. ``start_daemon=False``). + + Otherwise rank 0 will host the TCPStore (with multi-tenancy) and it is + assumed that rank 0's hostname and port are correctly passed via + ``hostname`` and ``port``. All non-zero ranks will create and return a + TCPStore client. + """ + # check if port is uint16_t + if not 0 <= port < 2**16: + raise ValueError(f"port must have value from 0 to 65535 but was {port}.") + + if _agent_store_serves(hostname, port): + # We create a new TCPStore for every retry so no need to add prefix for each attempt. + return TCPStore( + host_name=hostname, + port=port, + world_size=world_size, + is_master=False, + timeout=timeout, + ) + else: + start_daemon = rank == 0 + return TCPStore( + host_name=hostname, + port=port, + world_size=world_size, + is_master=start_daemon, + timeout=timeout, + multi_tenant=True, + use_libuv=use_libuv, + ) + + +def _tcp_rendezvous_handler( + url: str, timeout: timedelta = default_pg_timeout, **kwargs +): + def _error(msg): + return _rendezvous_error("tcp:// rendezvous: " + msg) + + result = urlparse(url) + if result.port is None: + raise _error("port number missing") + query_dict = _query_to_dict(result.query) + if "rank" not in query_dict: + raise _error("rank parameter missing") + if "world_size" not in query_dict: + raise _error("world size parameter missing") + + rank = int(query_dict["rank"]) + world_size = int(query_dict["world_size"]) + use_libuv = _get_use_libuv_from_query_dict(query_dict) + + if result.hostname is None: + raise AssertionError("hostname cannot be None") + + store = _create_c10d_store( + result.hostname, result.port, rank, world_size, timeout, use_libuv + ) + + yield (store, rank, world_size) + + # If this configuration is invalidated, there is nothing we can do about it + raise RuntimeError("Unable to perform re-rendezvous using tcp:// method") + + +def _env_rendezvous_handler( + url: str, timeout: timedelta = default_pg_timeout, **kwargs +): + def _error(msg): + return _rendezvous_error("env:// rendezvous: " + msg) + + def _env_error(var): + return _error(f"environment variable {var} expected, but not set") + + def _get_env_or_raise(env_var: str) -> str: + env_val = os.environ.get(env_var, None) + if not env_val: + raise _env_error(env_var) + else: + return env_val + + result = urlparse(url) + query_dict = _query_to_dict(result.query) + + rank: int + world_size: int + master_port: int + master_addr: str + + if "rank" in query_dict: + rank = int(query_dict["rank"]) + else: + rank = int(_get_env_or_raise("RANK")) + + if "world_size" in query_dict: + world_size = int(query_dict["world_size"]) + else: + world_size = int(_get_env_or_raise("WORLD_SIZE")) + + master_addr = _get_env_or_raise("MASTER_ADDR") + master_port = int(_get_env_or_raise("MASTER_PORT")) + use_libuv = _get_use_libuv_from_query_dict(query_dict) + + store = _create_c10d_store( + master_addr, master_port, rank, world_size, timeout, use_libuv + ) + + yield (store, rank, world_size) + + # If this configuration is invalidated, there is nothing we can do about it + raise RuntimeError("Unable to perform re-rendezvous using env:// method") + + +register_rendezvous_handler("tcp", _tcp_rendezvous_handler) +register_rendezvous_handler("env", _env_rendezvous_handler) +register_rendezvous_handler("file", _file_rendezvous_handler) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..2b075f81b7d5070544b23f80528ecffdbbe04070 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__init__.py @@ -0,0 +1,258 @@ +# mypy: allow-untyped-defs +import logging +import os +import threading +import warnings +from collections.abc import Generator +from datetime import timedelta +from urllib.parse import urlparse + +import torch +import torch.distributed as dist + + +__all__ = ["is_available"] + + +logger = logging.getLogger(__name__) + + +_init_counter = 0 +_init_counter_lock = threading.Lock() + + +def is_available() -> bool: + return hasattr(torch._C, "_rpc_init") + + +if is_available() and not torch._C._rpc_init(): + raise RuntimeError("Failed to initialize torch.distributed.rpc") + + +if is_available(): + _is_tensorpipe_available = hasattr( + torch._C._distributed_rpc, "_TensorPipeRpcBackendOptionsBase" + ) + + import numbers + + import torch.distributed.autograd as dist_autograd + from torch._C._distributed_c10d import Store + from torch._C._distributed_rpc import ( + _cleanup_python_rpc_handler, + _DEFAULT_INIT_METHOD, + _DEFAULT_RPC_TIMEOUT_SEC, + _delete_all_user_and_unforked_owner_rrefs, + _destroy_rref_context, + _disable_jit_rref_pickle, + _disable_server_process_global_profiler, + _enable_jit_rref_pickle, + _enable_server_process_global_profiler, + _get_current_rpc_agent, + _invoke_remote_builtin, + _invoke_remote_python_udf, + _invoke_remote_torchscript, + _invoke_rpc_builtin, + _invoke_rpc_python_udf, + _invoke_rpc_torchscript, + _is_current_rpc_agent_set, + _reset_current_rpc_agent, + _rref_context_get_debug_info, + _set_and_start_rpc_agent, + _set_profiler_node_id, + _set_rpc_timeout, + _UNSET_RPC_TIMEOUT, + enable_gil_profiling, + get_rpc_timeout, + PyRRef, + RemoteProfilerManager, + RpcAgent, + RpcBackendOptions, + WorkerInfo, + ) + + if _is_tensorpipe_available: + from torch._C._distributed_rpc import ( + _DEFAULT_NUM_WORKER_THREADS, + _TensorPipeRpcBackendOptionsBase, + TensorPipeAgent, + ) + + from . import api, backend_registry, functions + from .api import * # noqa: F403 + from .backend_registry import BackendType + from .options import TensorPipeRpcBackendOptions + from .server_process_global_profiler import _server_process_global_profile + + rendezvous_iterator: Generator[tuple[Store, int, int], None, None] + + __all__ += ["init_rpc", "BackendType", "TensorPipeRpcBackendOptions"] + # pyrefly: ignore [unresolvable-dunder-all] + __all__ = __all__ + api.__all__ + backend_registry.__all__ + + def init_rpc( + name, + backend=None, + rank=-1, + world_size=None, + rpc_backend_options=None, + ): + r""" + Initializes RPC primitives such as the local RPC agent + and distributed autograd, which immediately makes the current + process ready to send and receive RPCs. + + Args: + name (str): a globally unique name of this node. (e.g., + ``Trainer3``, ``ParameterServer2``, ``Master``, ``Worker1``) + Name can only contain number, alphabet, underscore, colon, + and/or dash, and must be shorter than 128 characters. + backend (BackendType, optional): The type of RPC backend + implementation. Supported values is + ``BackendType.TENSORPIPE`` (the default). + See :ref:`rpc-backends` for more information. + rank (int): a globally unique id/rank of this node. + world_size (int): The number of workers in the group. + rpc_backend_options (RpcBackendOptions, optional): The options + passed to the RpcAgent constructor. It must be an agent-specific + subclass of :class:`~torch.distributed.rpc.RpcBackendOptions` + and contains agent-specific initialization configurations. By + default, for all agents, it sets the default timeout to 60 + seconds and performs the rendezvous with an underlying process + group initialized using ``init_method = "env://"``, + meaning that environment variables ``MASTER_ADDR`` and + ``MASTER_PORT`` need to be set properly. See + :ref:`rpc-backends` for more information and find which options + are available. + """ + torch._C._log_api_usage_once("torch.distributed.init_rpc") + if backend is not None and not isinstance( + backend, backend_registry.BackendType + ): + raise TypeError("Argument backend must be a member of BackendType") + + if rpc_backend_options is not None and not isinstance( + rpc_backend_options, RpcBackendOptions + ): + raise TypeError( + "Argument rpc_backend_options must be an instance of RpcBackendOptions" + ) + + # Try to detect the backend from the options + if backend is None and rpc_backend_options is not None: + for candidate_backend in BackendType: + if isinstance( + rpc_backend_options, + type( + backend_registry.construct_rpc_backend_options( + candidate_backend + ) + ), + ): + backend = candidate_backend + break + else: + raise TypeError( + f"Could not infer backend for options {rpc_backend_options}" + ) + # Ignore type error because mypy doesn't handle dynamically generated type objects (#4865) + if backend != BackendType.TENSORPIPE: # type: ignore[attr-defined] + logger.warning( + "RPC was initialized with no explicit backend but with options " # type: ignore[attr-defined] + "corresponding to %(backend)s, hence that backend will be used " + "instead of the default BackendType.TENSORPIPE. To silence this " + "warning pass `backend=%(backend)s` explicitly.", + {"backend": backend}, + ) + + if backend is None: + backend = BackendType.TENSORPIPE # type: ignore[attr-defined] + + if rpc_backend_options is None: + # default construct a set of RPC backend options. + rpc_backend_options = backend_registry.construct_rpc_backend_options( + backend + ) + + # Create store, performs rendezvous for static RPC group. + if not world_size: + # If world_size is not set in construction and also not set in environment variables + # The store will be created for the dynamic group setting + store = dist._create_store_from_options(rpc_backend_options, rank) + else: + # This rendezvous state sometimes is destroyed before all processes + # finishing handshaking. To avoid that issue, we make it global to + # keep it alive. + global rendezvous_iterator + rendezvous_iterator = dist.rendezvous( + rpc_backend_options.init_method, rank=rank, world_size=world_size + ) + store, _, _ = next(rendezvous_iterator) + # Use same timeout as RPC. + store.set_timeout(timedelta(seconds=rpc_backend_options.rpc_timeout)) + + # Use a PrefixStore to distinguish multiple invocations. + with _init_counter_lock: + global _init_counter + store = dist.PrefixStore(str(f"rpc_prefix_{_init_counter}"), store) + _init_counter += 1 + + # Initialize autograd before RPC since _init_rpc_backend guarantees all + # processes sync via the store. If we initialize autograd after RPC, + # there could be a race where some nodes might have initialized autograd + # and others might not have. As a result, a node calling + # torch.distributed.autograd.backward() would run into errors since + # other nodes might not have been initialized. + dist_autograd._init(rank) + + _set_profiler_node_id(rank) + # Initialize RPC. + _init_rpc_backend(backend, store, name, rank, world_size, rpc_backend_options) + + def _validate_rpc_args(backend, store, name, rank, world_size, rpc_backend_options): + type_mapping = { + backend: backend_registry.BackendType, + store: dist.Store, + name: str, + rank: numbers.Integral, + # world_size can be None for a dynamic group + world_size: (numbers.Integral, type(None)), + rpc_backend_options: RpcBackendOptions, + } + for arg, arg_type in type_mapping.items(): + if not isinstance(arg, arg_type): # type: ignore[arg-type] + raise RuntimeError( + f"Argument {arg} must be of type {arg_type} but got type {type(arg)}" + ) + + def _init_rpc_backend( + backend=BackendType.TENSORPIPE, # type: ignore[attr-defined] + store=None, + name=None, + rank=-1, + world_size=None, + rpc_backend_options=None, + ): + _validate_rpc_args(backend, store, name, rank, world_size, rpc_backend_options) + + if _is_current_rpc_agent_set(): + raise RuntimeError("RPC is already initialized") + + # Initialize RPC. + rpc_agent = backend_registry.init_backend( + backend, + store=store, + name=name, + rank=rank, + world_size=world_size, + rpc_backend_options=rpc_backend_options, + ) + + api._init_rpc_states(rpc_agent) + + @api._require_initialized + def _get_debug_info(): + info = _rref_context_get_debug_info() + info.update(api._get_current_rpc_agent().get_debug_info()) + info.update(dist_autograd._get_debug_info()) + return info diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7180c8d6f31ca48c82db46e52a925167541c365a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..87d604d67591de1b5d9ca4dee596dde5a75e5258 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..53d363b1725ec73d1aadb8e1490b1904b3fbf9d8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/backend_registry.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/backend_registry.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..048120ddcf9cf2b17bdb5b239430a3bfbd77528a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/backend_registry.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/constants.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/constants.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..937da71520bff0a44f06649cc742a660f73122c3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/constants.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/functions.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/functions.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9e1974c3e5cd729ee8b1907859986a39278865db Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/functions.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/internal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/internal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7cfdb46ea55b14fd16731e9026075833675f9dea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/internal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/options.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/options.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..3e3f46ee7371ef106145596e418eda3cd9f77d59 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/options.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/rref_proxy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/rref_proxy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..44b8cfd18c182a04c81d850f22960fcf64c10bb1 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/rref_proxy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/server_process_global_profiler.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/server_process_global_profiler.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ea671ca3c6ceec4f79fe79d5a06d396b70952e65 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/__pycache__/server_process_global_profiler.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..0abd737becafbae33b0b63799c1eb43c913e1998 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__init__.py @@ -0,0 +1,18 @@ +import torch + + +def is_available() -> bool: + return hasattr(torch._C, "_faulty_agent_init") + + +if is_available() and not torch._C._faulty_agent_init(): + raise RuntimeError("Failed to initialize torch.distributed.rpc._testing") + +if is_available(): + # Registers FAULTY_TENSORPIPE RPC backend. + from torch._C._distributed_rpc_testing import ( + FaultyTensorPipeAgent, + FaultyTensorPipeRpcBackendOptions, + ) + + from . import faulty_agent_backend_registry diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..debdac621fb6046c7c0c54ed0b82655fc7bfa989 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/faulty_agent_backend_registry.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/faulty_agent_backend_registry.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ed9f9952247900418f6bfdc162e8c339a88b0ca6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/__pycache__/faulty_agent_backend_registry.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/faulty_agent_backend_registry.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/faulty_agent_backend_registry.py new file mode 100644 index 0000000000000000000000000000000000000000..d04882e16e79a94f74ddc1350e94f547ef625611 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_testing/faulty_agent_backend_registry.py @@ -0,0 +1,62 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +import torch.distributed as dist +import torch.distributed.rpc as rpc + + +def _faulty_tensorpipe_construct_rpc_backend_options_handler( + rpc_timeout, + init_method, + num_worker_threads, + messages_to_fail, + messages_to_delay, + num_fail_sends, + **kwargs, +): + from . import FaultyTensorPipeRpcBackendOptions + + return FaultyTensorPipeRpcBackendOptions( + num_worker_threads=num_worker_threads, + rpc_timeout=rpc_timeout, + init_method=init_method, + messages_to_fail=messages_to_fail, + messages_to_delay=messages_to_delay, + num_fail_sends=num_fail_sends, + ) + + +def _faulty_tensorpipe_init_backend_handler( + store, name, rank, world_size, rpc_backend_options +): + from torch.distributed.rpc import api + + from . import FaultyTensorPipeAgent, FaultyTensorPipeRpcBackendOptions + + if not isinstance(store, dist.Store): + raise TypeError(f"`store` must be a c10d::Store. {store}") + + if not isinstance(rpc_backend_options, FaultyTensorPipeRpcBackendOptions): + raise TypeError( + f"`rpc_backend_options` must be a `FaultyTensorPipeRpcBackendOptions`. {rpc_backend_options}" + ) + + agent = FaultyTensorPipeAgent( + store, + name, + rank, + world_size, + rpc_backend_options, + {}, # reverse_device_map + [], # devices + ) + api._init_rpc_states(agent) + + return agent + + +rpc.backend_registry.register_backend( + "FAULTY_TENSORPIPE", + _faulty_tensorpipe_construct_rpc_backend_options_handler, + _faulty_tensorpipe_init_backend_handler, +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..a0021ff1e43d8653df457cb99e7ea3637a508851 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/_utils.py @@ -0,0 +1,47 @@ +# mypy: allow-untyped-defs +import logging +from contextlib import contextmanager +from typing import cast + + +logger = logging.getLogger(__name__) + + +@contextmanager +def _group_membership_management(store, name, is_join): + token_key = "RpcGroupManagementToken" + join_or_leave = "join" if is_join else "leave" + my_token = f"Token_for_{name}_{join_or_leave}" + while True: + # Retrieve token from store to signal start of rank join/leave critical section + returned = store.compare_set(token_key, "", my_token).decode() + if returned == my_token: + # Yield to the function this context manager wraps + yield + # Finished, now exit and release token + # Update from store to signal end of rank join/leave critical section + store.set(token_key, "") + # Other will wait for this token to be set before they execute + store.set(my_token, "Done") + break + else: + # Store will wait for the token to be released + try: + store.wait([returned]) + except RuntimeError: + logger.error( + "Group membership token %s timed out waiting for %s to be released.", + my_token, + returned, + ) + raise + + +def _update_group_membership(worker_info, my_devices, reverse_device_map, is_join): + from . import api, TensorPipeAgent + + agent = cast(TensorPipeAgent, api._get_current_rpc_agent()) + ret = agent._update_group_membership( + worker_info, my_devices, reverse_device_map, is_join + ) + return ret diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/api.py new file mode 100644 index 0000000000000000000000000000000000000000..25b97d2129ef40d1f973396d54a39de5e26cf2ed --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/api.py @@ -0,0 +1,978 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs + +import collections +import contextlib +import functools +import inspect +import logging +import threading +from collections.abc import Callable +from typing import Any, Generic, TYPE_CHECKING, TypeVar +from typing_extensions import ParamSpec + +import torch +from torch._C._distributed_rpc import ( + _cleanup_python_rpc_handler, + _delete_all_user_and_unforked_owner_rrefs, + _destroy_rref_context, + _get_current_rpc_agent, + _invoke_remote_builtin, + _invoke_remote_python_udf, + _invoke_remote_torchscript, + _invoke_rpc_builtin, + _invoke_rpc_python_udf, + _invoke_rpc_torchscript, + _is_current_rpc_agent_set, + _reset_current_rpc_agent, + _set_and_start_rpc_agent, + get_rpc_timeout, + PyRRef, + RemoteProfilerManager, + WorkerInfo, +) +from torch.futures import Future + +from ._utils import _group_membership_management, _update_group_membership +from .constants import DEFAULT_SHUTDOWN_TIMEOUT, UNSET_RPC_TIMEOUT +from .internal import ( + _build_rpc_profiling_key, + _internal_rpc_pickler, + PythonUDF, + RPCExecMode, +) + + +_P = ParamSpec("_P") +_R = TypeVar("_R") + + +__all__ = [ + "shutdown", + "get_worker_info", + "remote", + "rpc_sync", + "rpc_async", + "RRef", + "AllGatherStates", + "method_factory", + "new_method", +] + + +logger = logging.getLogger(__name__) + +# NB: Ignoring RRef leaks during shutdown. Without this, applications have to +# make sure there is no references to any RRef in the application code and +# Python GC has done its job to delete those RRefs. This is could result in bad +# debugging experiences especially when for large applications. Therefore, by +# default, we are going to ignore RRef leaks during shutdown. This is usually +# fine as shutdown means applications have done training and no longer care +# about states. +# +# To enable RRef leak checking, set this _ignore_rref_leak to False +_ignore_rref_leak = True +_default_pickler = _internal_rpc_pickler + + +@contextlib.contextmanager +def _use_rpc_pickler(rpc_pickler): + r""" + rpc_pickler: (.internal._InternalRPCPickler) Overrides the default RPC pickler + """ + global _default_pickler + _default_pickler = rpc_pickler + try: + yield + finally: + _default_pickler = _internal_rpc_pickler + + +def _require_initialized(func: Callable[_P, _R]) -> Callable[_P, _R]: + @functools.wraps(func) + def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> _R: + if not _is_current_rpc_agent_set(): + raise RuntimeError( + "RPC has not been initialized. Call " + "torch.distributed.rpc.init_rpc first." + ) + return func(*args, **kwargs) + + return wrapper + + +class AllGatherStates: + def __init__(self): + # Each `gathered_objects` is an empty dict at beginning. + # The leader worker is elected as the first worker in a sorted worker + # name list. Whenever there is a worker entering `_all_gather()`, it + # runs `_gather_to_leader()` on the leader to add its own name and + # data obj to this dict. The leader also adds itself's name to the dict + # on calling `_all_gather()`. + # Once `set(gathered_objects.keys()) == _ALL_WORKER_NAMES`, the leader + # will broadcast the gathered dict to all follower workers and set their + # `gathered_objects` field and the `proceed_signal` field. + self.gathered_objects = {} + # All workers wait on this signal until it receives all gathered + # objects. + self.proceed_signal = threading.Event() + + +# States used by `def _all_gather()`. +# `_ALL_WORKER_NAMES` is initialized on initializing RPC layer. +_ALL_WORKER_NAMES: set[Any] = set() +_all_gather_dict_lock = threading.RLock() +_all_gather_sequence_id: dict[str, int] = {} +_all_gather_sequence_id_to_states: collections.defaultdict = collections.defaultdict( + AllGatherStates +) + + +def _init_rpc_states(agent): + worker_infos = agent.get_worker_infos() + global _ALL_WORKER_NAMES + _ALL_WORKER_NAMES = {worker_info.name for worker_info in worker_infos} + + # NB: backend implementation might have already set the rpc_agent. + if not _is_current_rpc_agent_set(): + _set_and_start_rpc_agent(agent) + + +def _gather_to_leader(sequence_id, worker_name, obj, worker_names=None): + with _all_gather_dict_lock: + if not worker_names: + worker_names = _ALL_WORKER_NAMES + if worker_name not in worker_names: + raise AssertionError(f"{worker_name} is not expected by leader.") + states = _all_gather_sequence_id_to_states[sequence_id] + if worker_name in states.gathered_objects: + raise AssertionError( + f"{worker_name} reported intent sequence id {sequence_id} twice. " + ) + states.gathered_objects[worker_name] = obj + if worker_names == set(states.gathered_objects.keys()): + states.proceed_signal.set() + + +def _broadcast_to_followers(sequence_id, objects_map): + with _all_gather_dict_lock: + states = _all_gather_sequence_id_to_states[sequence_id] + + if states.proceed_signal.is_set(): + raise AssertionError( + f"Termination signal sequence id {sequence_id} got set twice." + ) + states.gathered_objects = objects_map + states.proceed_signal.set() + + +_thread_local_var = threading.local() + + +@contextlib.contextmanager +def _wait_all(): + r""" + A context manager that collects all futures returned by ``rpc_async`` and + waits them on the context manager's exit; relieving the user of needing + to explicitly call wait. + + + Example:: + >>> # xdoctest: +SKIP("distributed") + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> with rpc._wait_all(): + >>> fut_1 = rpc.rpc_async(dst, torch.add, (torch.ones(2, 2), 1)) + >>> fut_2 = rpc.rpc_async(dst, torch.add, (torch.ones(2, 2), 1)) + >>> #fut_1 and fut_2 are waited on + """ + _thread_local_var.future_list = [] + try: + yield + finally: + try: + torch.futures.wait_all(_thread_local_var.future_list) + finally: + del _thread_local_var.future_list + + +@_require_initialized +def _all_gather(obj, worker_names=None, timeout: float = UNSET_RPC_TIMEOUT): + r""" + This is similar to torch.distributed.all_gather(), but is using RPC. It + picks the worker with the smallest name (alphabetic order) as the leader. + Then all followers send their data ``obj`` to the leader. After the leader + has received all, it will broadcast the results back to all followers. This + function blocks until all workers have received the gathered results. + """ + if not worker_names: + if _ALL_WORKER_NAMES is None: + raise AssertionError( + "`_ALL_WORKER_NAMES` is not initialized for `def _all_gather`." + ) + worker_names = _ALL_WORKER_NAMES + leader_name = min(worker_names) + + self_name = _get_current_rpc_agent().get_worker_info().name + + with _all_gather_dict_lock: + concat_names = "".join(sorted(worker_names)) + sequence_num = _all_gather_sequence_id.get(concat_names, 0) + _all_gather_sequence_id[concat_names] = sequence_num + 1 + sequence_id = concat_names + str(sequence_num) + + is_leader = leader_name == self_name + + if timeout == UNSET_RPC_TIMEOUT: + # Timeout is specified by agent for RPC calls + rpc_timeout = get_rpc_timeout() + # No timeout for signal + signal_timeout = None + elif timeout == DEFAULT_SHUTDOWN_TIMEOUT: + # No timeout for RPC + rpc_timeout = timeout + # No timeout for signal + signal_timeout = None + else: + # Signal and RPC timeout use the same timeout + signal_timeout = rpc_timeout = timeout + + # Phase 1: Followers send it's object to the leader + if is_leader: + _gather_to_leader(sequence_id, self_name, obj, worker_names) + else: + rpc_sync( + leader_name, + _gather_to_leader, + args=(sequence_id, self_name, obj, worker_names), + timeout=rpc_timeout, + ) + + with _all_gather_dict_lock: + states = _all_gather_sequence_id_to_states[sequence_id] + + # Timeout is either set by function parameter or None (which is indefinite) + states.proceed_signal.wait(timeout=signal_timeout) + + # Phase 2: Leader broadcast gathered results to all followers + # Leader's signal is the first to be unblocked, after receiving all + # followers' data objects. + if is_leader: + worker_name_to_response_future_dict = {} + for follower_name in worker_names - {leader_name}: + fut = rpc_async( + follower_name, + _broadcast_to_followers, + args=(sequence_id, states.gathered_objects), + timeout=rpc_timeout, + ) + worker_name_to_response_future_dict[follower_name] = fut + + errors = [] + for follower_name, fut in worker_name_to_response_future_dict.items(): + try: + fut.wait() + except RuntimeError as ex: + errors.append((follower_name, ex)) + + if errors: + raise RuntimeError( + f"Followers {[e[0] for e in errors]} timed out in _all_gather " + f"after {rpc_timeout:.2f} seconds. The first exception is {errors[0][1]}" + ) + + # Clean up for the states using the sequence_id + with _all_gather_dict_lock: + states = _all_gather_sequence_id_to_states.pop(sequence_id) + return states.gathered_objects + + +@_require_initialized +def _barrier(worker_names): + r""" + Synchronizes local and remote RPC processes. + + This will block until all local and remote RPC processes specified under worker_names + reach this method to wait for all outstanding work to complete. + + Args: + worker_names (List[str]): The set of workers to synchronize. + + """ + try: + _all_gather(None, set(worker_names)) + except RuntimeError: + logger.exception("Failed to complete barrier") + + +@_require_initialized +def _wait_all_workers(timeout=DEFAULT_SHUTDOWN_TIMEOUT): + r""" + Block until all local and remote RPC processes reach this method and wait + for all outstanding work to complete. Every RPC process must call this + method before exit to perform a graceful shutdown. This should be used to + terminate the RPC framework, and there is no guarantee that the RPC + framework will work after this method returns. + """ + try: + _all_gather(None, timeout=timeout) + except RuntimeError as ex: + logger.exception("Failed to respond to 'Shutdown Proceed' in time") + raise ex + + +@_require_initialized +def shutdown(graceful=True, timeout=DEFAULT_SHUTDOWN_TIMEOUT): + r""" + Perform a shutdown of the RPC agent, and then destroy the RPC agent. This + stops the local agent from accepting outstanding requests, and shuts + down the RPC framework by terminating all RPC threads. If ``graceful=True``, + this will block until all local and remote RPC processes reach this method + and wait for all outstanding work to complete. Otherwise, if + ``graceful=False``, this is a local shutdown, and it does not wait for other + RPC processes to reach this method. + + .. warning:: + For :class:`~torch.futures.Future` objects returned by + :meth:`~torch.distributed.rpc.rpc_async`, ``future.wait()`` should not + be called after ``shutdown()``. + + Args: + graceful (bool): Whether to do a graceful shutdown or not. If True, + this will 1) wait until there is no pending system + messages for ``UserRRefs`` and delete them; 2) block + until all local and remote RPC processes have reached + this method and wait for all outstanding work to + complete. + + Example:: + Make sure that ``MASTER_ADDR`` and ``MASTER_PORT`` are set properly + on both workers. Refer to :meth:`~torch.distributed.init_process_group` + API for more details. For example, + + export MASTER_ADDR=localhost + export MASTER_PORT=5678 + + Then run the following code in two different processes: + + >>> # xdoctest: +SKIP + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> # do some work + >>> result = rpc.rpc_sync("worker1", torch.add, args=(torch.ones(1), 1)) + >>> # ready to shutdown + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> # wait for worker 0 to finish work, and then shutdown. + >>> rpc.shutdown() + """ + if graceful: + try: + agent = _get_current_rpc_agent() + from torch._C._distributed_rpc import TensorPipeAgent + + if not isinstance(agent, TensorPipeAgent) or agent.is_static_group: + _wait_all_workers(timeout) + _delete_all_user_and_unforked_owner_rrefs() + agent.join(shutdown=True, timeout=timeout) + else: + # This is a dynamic group so we need to grab the token for the operation + my_worker_info = agent.get_worker_info() + my_name = my_worker_info.name + with _group_membership_management(agent.store, my_name, False): + all_worker_infos = agent.get_worker_infos() + for worker in all_worker_infos: + if worker.name != my_name: + rpc_sync( + worker.name, + _update_group_membership, + args=(my_worker_info, [], {}, False), + ) + agent.join(shutdown=True, timeout=timeout) + finally: + # In case of errors, continue to complete the local shutdown. + _finalize_shutdown() + else: + _finalize_shutdown() + + +def _finalize_shutdown(): + try: + # This raises a `TORCH_CHECK()` exception on RRef leak detected. + _destroy_rref_context(_ignore_rref_leak) + finally: + _get_current_rpc_agent().shutdown() + # clean up python rpc handler in shutdown(), see comments in + # PythonRpcHandler::cleanup(), call it in python API because the + # cleanup() function has python dependency, it assumes python + # interpreter exists. + # No matter if RRef leak exception is raised, this clean-up code + # must run to avoid destruction segfault in Python 3.5. + # + # future.wait() should not be called after shutdown(). + # pythonRpcHandler is cleaned up in shutdown(), after + # shutdown(), python objects returned from rpc python call can not be + # resolved. + _cleanup_python_rpc_handler() + _reset_current_rpc_agent() + + +@_require_initialized +def get_worker_info(worker_name=None): + r""" + Get :class:`~torch.distributed.rpc.WorkerInfo` of a given worker name. + Use this :class:`~torch.distributed.rpc.WorkerInfo` to avoid passing an + expensive string on every invocation. + + Args: + worker_name (str): the string name of a worker. If ``None``, return the + id of the current worker. (default ``None``) + + Returns: + :class:`~torch.distributed.rpc.WorkerInfo` instance for the given + ``worker_name`` or :class:`~torch.distributed.rpc.WorkerInfo` of the + current worker if ``worker_name`` is ``None``. + """ + if worker_name is not None: + return _get_current_rpc_agent().get_worker_info(worker_name) + else: + return _get_current_rpc_agent().get_worker_info() + + +def _to_worker_info(to): + if isinstance(to, WorkerInfo): + return to + elif isinstance(to, (str, int)): + return get_worker_info(to) + else: + raise ValueError(f"Cannot get WorkerInfo from name {to}") + + +def _rref_typeof_on_owner(rref, blocking: bool = True): + rref_type = type(rref.local_value()) + if blocking: + return rref_type + else: + # Wrap result into a completed Future. This is so that if blocking=`False` + # is specified, we return a future regardless of if this call is on user + # or owner. + future = Future[type]() + future.set_result(rref_type) + return future + + +def _rref_typeof_on_user( + rref, timeout: float = UNSET_RPC_TIMEOUT, blocking: bool = True +): + fut = rpc_async(rref.owner(), _rref_typeof_on_owner, args=(rref,), timeout=timeout) + if blocking: + return fut.wait() + else: + return fut + + +T = TypeVar("T") +# pyrefly: ignore [invalid-annotation] +GenericWithOneTypeVar = Generic[T] + + +if TYPE_CHECKING: + + class RRef(PyRRef[T], Generic[T]): + pass + +else: + try: + # Combine the implementation class and the type class. + class RRef(PyRRef, Generic[T]): + pass + + except TypeError: + # TypeError: metaclass conflict: the metaclass of a derived class + # must be a (non-strict) subclass of the metaclasses of all its bases + # Mypy doesn't understand __class__ (mypy bug #4177) + class RRefMeta(PyRRef.__class__, GenericWithOneTypeVar.__class__): # type: ignore[name-defined, misc, valid-type] + pass + + # Combine the implementation class and the type class. + # Types for classes expecting a certain generic parameter (mypy bug #7791) + class RRef(PyRRef, GenericWithOneTypeVar, metaclass=RRefMeta): # type: ignore[misc, no-redef, valid-type] + pass + + +# Install docstrings from `PyRRef` to `RRef`. +# +# This is for the fact that pybind11 generates the parameter +# `self` as type `rpc.PyRRef`, so a `:inherited-members:` +# under `.. autoclass:: RRef` does not work. +# we have to do the following process to replace `rpc.PyRRef` with `rpc.RRef`. +# +def method_factory(method_name, docstring): + def method(self, *args, **kwargs): + return getattr(super(RRef, self), method_name)(*args, **kwargs) + + if method.__doc__: + method.__doc__ = docstring + return method + + +for method_name, method in inspect.getmembers(PyRRef): + # Ignore magic methods, except "__str__". + if method_name.startswith("_") and method_name != "__str__": + continue + + # Get pybind11 generated docstring. + # It's like, + """ + to_here(self: torch.distributed.rpc.PyRRef, timeout: float=-1.0) -> object + + Blocking call that copies the value of the RRef from the owner + to the local node and returns it. If the current node is the + owner, returns a reference to the local value. + """ + docstring = getattr(method, "__doc__", None) + if docstring is None: + raise AssertionError("RRef user-facing methods should all have docstrings.") + + # Do surgery on pybind11 generated docstrings. + docstring = docstring.replace( + "torch.distributed.rpc.PyRRef", "torch.distributed.rpc.RRef" + ) + + # Attach user-facing RRef method with modified docstring. + new_method = method_factory(method_name, docstring) + setattr(RRef, method_name, new_method) + + +@_require_initialized +def remote(to, func, args=None, kwargs=None, timeout=UNSET_RPC_TIMEOUT): + r""" + Make a remote call to run ``func`` on worker ``to`` and return an + :class:`~torch.distributed.rpc.RRef` to the result value immediately. + Worker ``to`` will be the owner of the returned + :class:`~torch.distributed.rpc.RRef`, and the worker calling ``remote`` is + a user. The owner manages the global reference count of its + :class:`~torch.distributed.rpc.RRef`, and the owner + :class:`~torch.distributed.rpc.RRef` is only destructed when globally there + are no living references to it. + + Args: + to (str or WorkerInfo or int): name/rank/``WorkerInfo`` of the destination worker. + func (Callable): a callable function, such as Python callables, builtin + operators (e.g. :meth:`~torch.add`) and annotated + TorchScript functions. + args (tuple): the argument tuple for the ``func`` invocation. + kwargs (dict): is a dictionary of keyword arguments for the ``func`` + invocation. + + timeout (float, optional): timeout in seconds for this remote call. If the + creation of this + :class:`~torch.distributed.rpc.RRef` on worker + ``to`` is not successfully processed on this + worker within this timeout, then the next time + there is an attempt to use the RRef (such as + ``to_here()``), a timeout will be raised + indicating this failure. A value of 0 indicates + an infinite timeout, i.e. a timeout error will + never be raised. If not provided, the default + value set during initialization or with + ``_set_rpc_timeout`` is used. + + Returns: + A user :class:`~torch.distributed.rpc.RRef` instance to the result + value. Use the blocking API :meth:`torch.distributed.rpc.RRef.to_here` + to retrieve the result value locally. + + .. warning :: + The ``remote`` API does not copy storages of argument tensors until + sending them over the wire, which could be done by a different thread + depending on the RPC backend type. The caller should make sure that the + contents of those tensors stay intact until the returned RRef is + confirmed by the owner, which can be checked using the + :meth:`torch.distributed.rpc.RRef.confirmed_by_owner` API. + + .. warning :: + Errors such as timeouts for the ``remote`` API are handled on a + best-effort basis. This means that when remote calls initiated by + ``remote`` fail, such as with a timeout error, we take a best-effort + approach to error handling. This means that errors are handled and set + on the resulting RRef on an asynchronous basis. If the RRef has not been + used by the application before this handling (such as ``to_here`` or + fork call), then future uses of the ``RRef`` will appropriately raise + errors. However, it is possible that the user application will use the + ``RRef`` before the errors are handled. In this case, errors may not be + raised as they have not yet been handled. + + Example:: + + Make sure that ``MASTER_ADDR`` and ``MASTER_PORT`` are set properly + on both workers. Refer to :meth:`~torch.distributed.init_process_group` + API for more details. For example, + + export MASTER_ADDR=localhost + export MASTER_PORT=5678 + + Then run the following code in two different processes: + + >>> # xdoctest: +SKIP + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> rref1 = rpc.remote("worker1", torch.add, args=(torch.ones(2), 3)) + >>> rref2 = rpc.remote("worker1", torch.add, args=(torch.ones(2), 1)) + >>> x = rref1.to_here() + rref2.to_here() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + Below is an example of running a TorchScript function using RPC. + + >>> # On both workers: + >>> @torch.jit.script + >>> def my_script_add(tensor: torch.Tensor, scalar: int): + >>> return torch.add(tensor, scalar) + + >>> # On worker 0: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> rref = rpc.remote("worker1", my_script_add, args=(torch.ones(2), 3)) + >>> rref.to_here() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + """ + torch._C._log_api_usage_once("torch.distributed.rpc_remote") + qualified_name = torch.jit._builtins._find_builtin(func) + dst_worker_info = _to_worker_info(to) + should_profile = _get_should_profile() + + ctx_manager = _enable_rpc_profiler( + should_profile, qualified_name, func, RPCExecMode.REMOTE, dst_worker_info + ) + + with ctx_manager as rf: + args = args if args else () + kwargs = kwargs if kwargs else {} + + is_async_exec = hasattr(func, "_wrapped_async_rpc_function") + + if is_async_exec: + wrapped = func._wrapped_async_rpc_function + if isinstance(wrapped, torch.jit.ScriptFunction): + func = wrapped + + if qualified_name is not None: + rref = _invoke_remote_builtin( + dst_worker_info, qualified_name, timeout, *args, **kwargs + ) + elif isinstance(func, torch.jit.ScriptFunction): + rref = _invoke_remote_torchscript( + dst_worker_info.name, + torch._jit_internal._qualified_name(func), + timeout, + is_async_exec, + *args, + **kwargs, + ) + else: + (pickled_python_udf, tensors) = _default_pickler.serialize( + PythonUDF(func, args, kwargs) + ) + rref = _invoke_remote_python_udf( + dst_worker_info, pickled_python_udf, tensors, timeout, is_async_exec + ) + # attach profiling information + if should_profile: + if not torch.autograd._profiler_enabled(): + raise AssertionError + if rf is None: + raise AssertionError + fut = rf._call_end_callbacks_on_future(rref._get_future()) + rref._set_profiling_future(fut) + + return rref + + +def _invoke_rpc( + to, func, rpc_type, args=None, kwargs=None, rpc_timeout: float = UNSET_RPC_TIMEOUT +): + if not callable(func): + raise TypeError("function should be callable.") + + qualified_name = torch.jit._builtins._find_builtin(func) + dst_worker_info = _to_worker_info(to) + + should_profile = _get_should_profile() + + ctx_manager = _enable_rpc_profiler( + should_profile, qualified_name, func, rpc_type, dst_worker_info + ) + + with ctx_manager as rf: + args = args if args else () + kwargs = kwargs if kwargs else {} + + is_async_exec = hasattr(func, "_wrapped_async_rpc_function") + + if is_async_exec: + # pyrefly: ignore [missing-attribute] + wrapped = func._wrapped_async_rpc_function + if isinstance(wrapped, torch.jit.ScriptFunction): + func = wrapped + + if qualified_name is not None: + fut = _invoke_rpc_builtin( + dst_worker_info, qualified_name, rpc_timeout, *args, **kwargs + ) + elif isinstance(func, torch.jit.ScriptFunction): + fut = _invoke_rpc_torchscript( + dst_worker_info.name, + torch._jit_internal._qualified_name(func), + args, + kwargs, + rpc_timeout, + is_async_exec, + ) + else: + (pickled_python_udf, tensors) = _default_pickler.serialize( + PythonUDF(func, args, kwargs) + ) + fut = _invoke_rpc_python_udf( + dst_worker_info, pickled_python_udf, tensors, rpc_timeout, is_async_exec + ) + if should_profile: + if not torch.autograd._profiler_enabled(): + raise AssertionError + if rf is None: + raise AssertionError + # Schedule profiling callbacks to run when the future completes. + # This returns a future that is completed when the original future + # completes and the profiling callbacks have been completed as well, + # to guarantee that fut.wait() completes the profiling. This new + # future will contain the same value as the original future. + fut = rf._call_end_callbacks_on_future(fut) + return fut + + +@_require_initialized +def rpc_sync(to, func, args=None, kwargs=None, timeout: float = UNSET_RPC_TIMEOUT): + r""" + Make a blocking RPC call to run function ``func`` on worker ``to``. RPC + messages are sent and received in parallel to execution of Python code. This + method is thread-safe. + + Args: + to (str or WorkerInfo or int): name/rank/``WorkerInfo`` of the destination worker. + func (Callable): a callable function, such as Python callables, builtin + operators (e.g. :meth:`~torch.add`) and annotated + TorchScript functions. + args (tuple): the argument tuple for the ``func`` invocation. + kwargs (dict): is a dictionary of keyword arguments for the ``func`` + invocation. + timeout (float, optional): timeout in seconds to use for this RPC. If + the RPC does not complete in this amount of + time, an exception indicating it has + timed out will be raised. A value of 0 + indicates an infinite timeout, i.e. a timeout + error will never be raised. If not provided, + the default value set during initialization + or with ``_set_rpc_timeout`` is used. + + Returns: + Returns the result of running ``func`` with ``args`` and ``kwargs``. + + Example:: + Make sure that ``MASTER_ADDR`` and ``MASTER_PORT`` are set properly + on both workers. Refer to :meth:`~torch.distributed.init_process_group` + API for more details. For example, + + export MASTER_ADDR=localhost + export MASTER_PORT=5678 + + Then run the following code in two different processes: + + >>> # xdoctest: +SKIP + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> ret = rpc.rpc_sync("worker1", torch.add, args=(torch.ones(2), 3)) + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + Below is an example of running a TorchScript function using RPC. + + >>> # On both workers: + >>> @torch.jit.script + >>> def my_script_add(tensor: torch.Tensor, scalar: int): + >>> return torch.add(tensor, scalar) + + >>> # On worker 0: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> ret = rpc.rpc_sync("worker1", my_script_add, args=(torch.ones(2), 3)) + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + """ + torch._C._log_api_usage_once("torch.distributed.rpc_sync") + fut = _invoke_rpc(to, func, RPCExecMode.SYNC, args, kwargs, timeout) + return fut.wait() + + +@_require_initialized +def rpc_async(to, func, args=None, kwargs=None, timeout=UNSET_RPC_TIMEOUT): + r""" + Make a non-blocking RPC call to run function ``func`` on worker ``to``. RPC + messages are sent and received in parallel to execution of Python code. This + method is thread-safe. This method will immediately return a + :class:`~torch.futures.Future` that can be awaited on. + + Args: + to (str or WorkerInfo or int): name/rank/``WorkerInfo`` of the destination worker. + func (Callable): a callable function, such as Python callables, builtin + operators (e.g. :meth:`~torch.add`) and annotated + TorchScript functions. + args (tuple): the argument tuple for the ``func`` invocation. + kwargs (dict): is a dictionary of keyword arguments for the ``func`` + invocation. + timeout (float, optional): timeout in seconds to use for this RPC. If + the RPC does not complete in this amount of + time, an exception indicating it has + timed out will be raised. A value of 0 + indicates an infinite timeout, i.e. a timeout + error will never be raised. If not provided, + the default value set during initialization + or with ``_set_rpc_timeout`` is used. + + + Returns: + Returns a :class:`~torch.futures.Future` object that can be waited + on. When completed, the return value of ``func`` on ``args`` and + ``kwargs`` can be retrieved from the :class:`~torch.futures.Future` + object. + + .. warning :: + Using GPU tensors as arguments or return values of ``func`` is not + supported since we don't support sending GPU tensors over the wire. You + need to explicitly copy GPU tensors to CPU before using them as + arguments or return values of ``func``. + + .. warning :: + The ``rpc_async`` API does not copy storages of argument tensors until + sending them over the wire, which could be done by a different thread + depending on the RPC backend type. The caller should make sure that the + contents of those tensors stay intact until the returned + :class:`~torch.futures.Future` completes. + + Example:: + Make sure that ``MASTER_ADDR`` and ``MASTER_PORT`` are set properly + on both workers. Refer to :meth:`~torch.distributed.init_process_group` + API for more details. For example, + + export MASTER_ADDR=localhost + export MASTER_PORT=5678 + + Then run the following code in two different processes: + + >>> # xdoctest: +SKIP + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> fut1 = rpc.rpc_async("worker1", torch.add, args=(torch.ones(2), 3)) + >>> fut2 = rpc.rpc_async("worker1", min, args=(1, 2)) + >>> result = fut1.wait() + fut2.wait() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + + Below is an example of running a TorchScript function using RPC. + + >>> # On both workers: + >>> @torch.jit.script + >>> def my_script_add(tensor: torch.Tensor, scalar: int): + >>> return torch.add(tensor, scalar) + + >>> # On worker 0: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> fut = rpc.rpc_async("worker1", my_script_add, args=(torch.ones(2), 3)) + >>> ret = fut.wait() + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> rpc.shutdown() + """ + torch._C._log_api_usage_once("torch.distributed.rpc_async") + fut = _invoke_rpc(to, func, RPCExecMode.ASYNC, args, kwargs, timeout) + if hasattr(_thread_local_var, "future_list"): + _thread_local_var.future_list.append(fut) + return fut + + +def _get_should_profile(): + # Legacy profiler should be enabled. RPC profiling is not supported with + # Kineto profiler. + ActiveProfilerType = torch._C._profiler.ActiveProfilerType + return ( + torch.autograd._profiler_enabled() + and torch._C._autograd._profiler_type() == ActiveProfilerType.LEGACY # type: ignore[attr-defined] + ) + + +def _enable_rpc_profiler( + should_profile, qualified_name, func, rpc_type, dst_worker_info +): + ctx_manager = contextlib.nullcontext() + + if should_profile: + # Create appropriate string representation based on type of func + # (builtin, script, python) + if qualified_name is None: + func_name = ( + torch._jit_internal._qualified_name(func) + if isinstance(func, torch.jit.ScriptFunction) + else func.__qualname__ + ) + else: + func_name = qualified_name + # Build RPC profiling key. + rpc_profiling_key = _build_rpc_profiling_key( + rpc_type, + func_name, + get_worker_info().name, + dst_worker_info.name, + ) + RemoteProfilerManager.set_current_profiling_key(rpc_profiling_key) + # Mypy doesn't support re-def of a variable not in the same block (#1174) + ctx_manager = torch.autograd.profiler.record_function(rpc_profiling_key) # type: ignore[assignment] + + return ctx_manager diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/backend_registry.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/backend_registry.py new file mode 100644 index 0000000000000000000000000000000000000000..56aeb1a6903277a709304ea9c952012cbf77c34f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/backend_registry.py @@ -0,0 +1,437 @@ +# mypy: allow-untyped-defs + + +import collections +import enum +from typing import cast + +import torch +import torch.distributed as dist + +from . import api, constants as rpc_constants +from ._utils import _group_membership_management, _update_group_membership + + +__all__ = [ + "backend_registered", + "register_backend", + "construct_rpc_backend_options", + "init_backend", + "BackendValue", + "BackendType", +] + +BackendValue = collections.namedtuple( + "BackendValue", ["construct_rpc_backend_options_handler", "init_backend_handler"] +) + + +def _backend_type_repr(self): + return "BackendType." + self.name + + +_backend_type_doc = """ + An enum class of available backends. + + PyTorch ships with a builtin ``BackendType.TENSORPIPE`` backend. + Additional ones can be registered using the + :func:`~torch.distributed.rpc.backend_registry.register_backend` function. +""" + +# Create an enum type, `BackendType`, with empty members. +# Can't handle Function Enum API (mypy bug #9079) +BackendType = enum.Enum(value="BackendType", names={}) # type: ignore[misc] +# Unable to assign a function a method (mypy bug #2427) +BackendType.__repr__ = _backend_type_repr # type: ignore[assignment] + +if BackendType.__doc__: + BackendType.__doc__ = _backend_type_doc + + +def backend_registered(backend_name): + """ + Checks if backend_name is registered as an RPC backend. + + Args: + backend_name (str): string to identify the RPC backend. + Returns: + True if the backend has been registered with ``register_backend``, else + False. + """ + return backend_name in BackendType.__members__ + + +def register_backend( + backend_name, construct_rpc_backend_options_handler, init_backend_handler +): + """Registers a new RPC backend. + + Args: + backend_name (str): backend string to identify the handler. + construct_rpc_backend_options_handler (function): + Handler that is invoked when + ``rpc_backend.construct_rpc_backend_options(**dict)`` is called. + init_backend_handler (function): Handler that is invoked when the + `_init_rpc_backend()` function is called with a backend. + This returns the agent. + """ + global BackendType + if backend_registered(backend_name): + raise RuntimeError(f"RPC backend {backend_name}: already registered") + # Create a new enum type, `BackendType`, with extended members. + existing_enum_dict = {member.name: member.value for member in BackendType} + extended_enum_dict = dict( + { + backend_name: BackendValue( + construct_rpc_backend_options_handler=construct_rpc_backend_options_handler, + init_backend_handler=init_backend_handler, + ) + }, + **existing_enum_dict, + ) + # Can't handle Function Enum API (mypy bug #9079) + BackendType = enum.Enum(value="BackendType", names=extended_enum_dict) # type: ignore[misc] + # Unable to assign a function a method (mypy bug #2427) + BackendType.__repr__ = _backend_type_repr # type: ignore[assignment] + if BackendType.__doc__: + BackendType.__doc__ = _backend_type_doc + + return BackendType[backend_name] + + +def construct_rpc_backend_options( + backend, + rpc_timeout=rpc_constants.DEFAULT_RPC_TIMEOUT_SEC, + init_method=rpc_constants.DEFAULT_INIT_METHOD, + **kwargs, +): + return backend.value.construct_rpc_backend_options_handler( + rpc_timeout, init_method, **kwargs + ) + + +def init_backend(backend, *args, **kwargs): + return backend.value.init_backend_handler(*args, **kwargs) + + +def _init_process_group(store, rank, world_size): + # Initialize ProcessGroup. + process_group_timeout = rpc_constants.DEFAULT_PROCESS_GROUP_TIMEOUT + + # We're using a bunch of private APIs here since `new_group` requires the + # default group to be initialized. + group = dist.ProcessGroupGloo(store, rank, world_size, process_group_timeout) + + if group is None: + raise AssertionError("Failed to initialize default ProcessGroup.") + + if (rank != -1) and (rank != group.rank()): + raise RuntimeError(f"rank argument {rank} doesn't match pg rank {group.rank()}") + if (world_size != -1) and (world_size != group.size()): + raise RuntimeError( + f"world_size argument {world_size} doesn't match pg size {group.size()}" + ) + return group + + +def _tensorpipe_construct_rpc_backend_options_handler( + rpc_timeout, + init_method, + num_worker_threads=rpc_constants.DEFAULT_NUM_WORKER_THREADS, + _transports=None, + _channels=None, + **kwargs, +): + from . import TensorPipeRpcBackendOptions + + return TensorPipeRpcBackendOptions( + rpc_timeout=rpc_timeout, + init_method=init_method, + num_worker_threads=num_worker_threads, + _transports=_transports, + _channels=_channels, + ) + + +def _tensorpipe_validate_devices(devices, device_count): + device_type = ( + acc.type if (acc := torch.accelerator.current_accelerator()) else "cpu" + ) + return all( + d.type == "cpu" or (d.type == device_type and 0 <= d.index < device_count) + for d in devices + ) + + +# detect if any worker has invalid device_map configurations, and return +# reverse device maps +def _tensorpipe_exchange_and_check_all_device_maps( + my_name, my_device_count, my_device_maps, my_devices, group +): + gathered: list[ + tuple[str, int, dict[str, dict[torch.device, torch.device]], list[torch.device]] + ] = [("", 0, {}, []) for _ in range(group.size())] + dist.all_gather_object( + gathered, (my_name, my_device_count, my_device_maps, my_devices), group + ) + all_names = [name for name, _, _, _ in gathered] + all_device_counts = {name: count for name, count, _, _ in gathered} + all_device_maps = {name: map_ for name, _, map_, _ in gathered} + all_devices = {name: devices for name, _, _, devices in gathered} + + _validate_device_maps(all_names, all_device_counts, all_device_maps, all_devices) + + # passed all checked, construct reverse mapping and get list of devices handled by this agent + reverse_device_maps = _create_reverse_mapping(my_name, all_names, all_device_maps) + my_devices = _create_device_list(my_devices, my_device_maps, reverse_device_maps) + return reverse_device_maps, my_devices + + +def _validate_device_maps( + all_names, all_device_counts, all_device_maps, all_devices, is_static_group=True +): + for node in all_names: + devices = all_devices[node] + if len(set(devices)) != len(devices): + raise ValueError(f"Node {node} has duplicated devices\ndevices = {devices}") + if not _tensorpipe_validate_devices(devices, all_device_counts[node]): + raise ValueError( + f"Node {node} has devices with invalid indices\n" + f"devices = {devices}\n" + f"device count = {all_device_counts[node]}" + ) + + for source_node in all_names: + # For dynamic group (non-static) do not check the target node name since it may not have joined yet + if is_static_group and not set(all_device_maps[source_node].keys()).issubset( + all_names + ): + raise ValueError( + f"Node {source_node} has invalid target node names in its device maps\n" + f"device maps = {all_device_maps[source_node].keys()}\n" + f"node names = {all_names}" + ) + for target_node, map_ in all_device_maps[source_node].items(): + if len(set(map_.values())) != len(map_): + raise ValueError( + f"Node {source_node} has duplicated target devices " + f"in its device map for {target_node}\n" + f"device map = {map_}" + ) + if all_devices[source_node]: + if not set(map_.keys()).issubset(all_devices[source_node]): + raise ValueError( + f"Node {source_node} has unexpected source devices " + f"in its device map for {target_node}\n" + f"device map = {map_}\n" + f"devices = {all_devices[source_node]}" + ) + elif not _tensorpipe_validate_devices( + map_.keys(), all_device_counts[source_node] + ): + raise ValueError( + f"Node {source_node} has source devices with invalid indices " + f"in its device map for {target_node}\n" + f"device map = {map_}\n" + f"device count = {all_device_counts[source_node]}" + ) + if all_devices.get(target_node, []): + if not set(map_.values()).issubset(all_devices[target_node]): + raise ValueError( + f"Node {source_node} has unexpected target devices " + f"in its device map for {target_node}\n" + f"device map = {map_}\n" + f"devices = {all_devices[target_node]}" + ) + elif target_node in all_device_counts and not _tensorpipe_validate_devices( + map_.values(), all_device_counts[target_node] + ): + raise ValueError( + f"Node {source_node} has target devices with invalid indices " + f"in its device map for {target_node}\n" + f"device map = {map_}\n" + f"device count = {all_device_counts[target_node]}" + ) + + +def _create_device_list(my_devices, my_device_maps, reverse_device_maps): + if not my_devices: + devices_set: set[torch.device] = set() + for map_ in my_device_maps.values(): + devices_set.update(map_.keys()) + for map_ in reverse_device_maps.values(): + devices_set.update(map_.keys()) + devices_set.discard(torch.device("cpu")) + my_devices = list(devices_set) + my_devices = sorted(my_devices, key=lambda d: d.index) + return my_devices + + +def _create_reverse_mapping(my_name, all_names, all_device_maps): + reverse_device_maps: dict[str, dict[torch.device, torch.device]] = {} + for node in all_names: + if my_name in all_device_maps[node]: + reverse_device_maps[node] = { + v: k for k, v in all_device_maps[node][my_name].items() + } + return reverse_device_maps + + +def _get_device_infos(): + from . import TensorPipeAgent + + agent = cast(TensorPipeAgent, api._get_current_rpc_agent()) + opts = agent._get_backend_options() + device_count = torch.accelerator.device_count() + if torch.accelerator.is_available() and opts.devices: + mod = torch.get_device_module(torch.accelerator.current_accelerator()) + mod.init() + return device_count, opts.device_maps, opts.devices + + +def _set_devices_and_reverse_device_map(agent): + from . import TensorPipeAgent + + agent = cast(TensorPipeAgent, agent) + # Group state is retrieved from local agent + # On initialization, tensorpipe agent retrieves information from all existing workers, so group state is valid + my_worker_info = agent.get_worker_info() + my_name = my_worker_info.name + all_worker_infos = agent.get_worker_infos() + # One round to get device_maps of all workers and construct reverse device maps + all_device_counts, all_device_maps, all_devices, all_names = {}, {}, {}, [] + for worker_info in all_worker_infos: + worker_name = worker_info.name + if worker_name != my_name: + # TODO: make async? + device_count, device_map, devices = api.rpc_sync( + worker_name, _get_device_infos + ) + else: + opts = agent._get_backend_options() + device_count, device_map, devices = ( + torch.accelerator.device_count(), + opts.device_maps, + opts.devices, + ) + all_device_counts[worker_name] = device_count + all_device_maps[worker_name] = device_map + all_devices[worker_name] = devices + all_names.append(worker_name) + + _validate_device_maps( + all_names, + all_device_counts, + all_device_maps, + all_devices, + is_static_group=False, + ) + reverse_device_maps = _create_reverse_mapping(my_name, all_names, all_device_maps) + + # Perform RPC call to all workers, including itself, to include newly joined worker information and device maps + for worker_name in all_names: + # Set device list for each worker + all_devices[worker_name] = _create_device_list( + all_devices[worker_name], all_device_maps[worker_name], reverse_device_maps + ) + api.rpc_sync( + worker_name, + _update_group_membership, + args=(my_worker_info, all_devices[worker_name], reverse_device_maps, True), + ) + + +def _tensorpipe_init_backend_handler( + store, name, rank, world_size, rpc_backend_options +): + from . import TensorPipeAgent, TensorPipeRpcBackendOptions + + if not isinstance(store, dist.Store): + raise TypeError(f"`store` must be a c10d::Store. {store}") + + if not isinstance(rpc_backend_options, TensorPipeRpcBackendOptions): + raise TypeError( + f"`rpc_backend_options` must be a `TensorPipeRpcBackendOptions`. {rpc_backend_options}" + ) + + device_count = torch.accelerator.device_count() + + is_static_group = bool(world_size) + # world_size is specified so this is a static group (ranks cannot join and leave) + if is_static_group: + # The agent's join method is required to behave like a barrier and perform + # collective operations, for which it relies on a process group, instead of + # re-implementing this on top of RPCs. + group = _init_process_group(store, rank, world_size) + + reverse_device_maps, devices = _tensorpipe_exchange_and_check_all_device_maps( + name, + device_count, + rpc_backend_options.device_maps, + rpc_backend_options.devices, + group, + ) + + if torch.accelerator.is_available() and devices: + # It's necessary to initialize PyTorch CUDA states here (e.g., + # CUDACachingAllocator). If this is missing, we could hit errors like + # "allocator not initialized", because other processes might send + # CUDA-related RPC request to this process before user code in this + # process initializes its PyTorch CUDA states. + mod = torch.get_device_module(torch.accelerator.current_accelerator()) + mod.init() + + # TODO: add try-except and destroy _agent in all processes if any fails. + agent = TensorPipeAgent( + store, + name, + rank, + world_size, + rpc_backend_options, + reverse_device_maps, + devices, + ) + + api._init_rpc_states(agent) + + # Run one dummy round of RPC to initialize channels/transports. Without + # this, it's easy to hit timeout in rpc.shutdown() if there is no other RPC + # on that process before rpc.shutdown(), as the agent initialization can + # take longer than 5s. + api._all_gather(None, timeout=rpc_backend_options.rpc_timeout) + # Need a barrier here to make sure no peers leave before the rank0 finishes + # _all_gather + group.barrier().wait() + + return agent + # initialization for dynamic rpc (ranks can join and leave) + else: + with _group_membership_management(store, name, True): + # Construct TPAgent with empty reverse_device_map and devices + # these properties will be updated after initialization + agent = TensorPipeAgent( + store, + name, + rank, + world_size, + rpc_backend_options, + {}, + [], + ) + api._init_rpc_states(agent) + + try: + # Notify all workers in group this rank has joined and set devices and reverse_device_map + # This is a synchronous operation that completes once all existing ranks are updated + _set_devices_and_reverse_device_map(agent) + except Exception: + api.shutdown() + raise + return agent + + +register_backend( + "TENSORPIPE", + _tensorpipe_construct_rpc_backend_options_handler, + _tensorpipe_init_backend_handler, +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/constants.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/constants.py new file mode 100644 index 0000000000000000000000000000000000000000..f0eaf92b8aef56dc96700c1ddb42bfb988542650 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/constants.py @@ -0,0 +1,24 @@ +from datetime import timedelta + +from torch._C._distributed_rpc import ( + _DEFAULT_INIT_METHOD, + _DEFAULT_NUM_WORKER_THREADS, + _DEFAULT_RPC_TIMEOUT_SEC, + _UNSET_RPC_TIMEOUT, +) + + +# For any RpcAgent. +DEFAULT_RPC_TIMEOUT_SEC: float = _DEFAULT_RPC_TIMEOUT_SEC +DEFAULT_INIT_METHOD: str = _DEFAULT_INIT_METHOD +DEFAULT_SHUTDOWN_TIMEOUT: float = 0 + +# For TensorPipeAgent. +DEFAULT_NUM_WORKER_THREADS: int = _DEFAULT_NUM_WORKER_THREADS +# Ensure that we don't time out when there are long periods of time without +# any operations against the underlying ProcessGroup. +DEFAULT_PROCESS_GROUP_TIMEOUT: timedelta = timedelta(milliseconds=2**31 - 1) +# Value indicating that timeout is not set for RPC call, and the default should be used. +UNSET_RPC_TIMEOUT: float = _UNSET_RPC_TIMEOUT + +__all__: list[str] = [] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/functions.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/functions.py new file mode 100644 index 0000000000000000000000000000000000000000..e48ea8cc534ab87838965c947bbd0ed76d4d64c7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/functions.py @@ -0,0 +1,169 @@ +# mypy: allow-untyped-defs +import functools + + +def async_execution(fn): + r""" + A decorator for a function indicating that the return value of the function + is guaranteed to be a :class:`~torch.futures.Future` object and this + function can run asynchronously on the RPC callee. More specifically, the + callee extracts the :class:`~torch.futures.Future` returned by the wrapped + function and installs subsequent processing steps as a callback to that + :class:`~torch.futures.Future`. The installed callback will read the value + from the :class:`~torch.futures.Future` when completed and send the + value back as the RPC response. That also means the returned + :class:`~torch.futures.Future` only exists on the callee side and is never + sent through RPC. This decorator is useful when the wrapped function's + (``fn``) execution needs to pause and resume due to, e.g., containing + :meth:`~torch.distributed.rpc.rpc_async` or waiting for other signals. + + .. note:: To enable asynchronous execution, applications must pass the + function object returned by this decorator to RPC APIs. If RPC detected + attributes installed by this decorator, it knows that this function + returns a ``Future`` object and will handle that accordingly. + However, this does not mean this decorator has to be outmost one when + defining a function. For example, when combined with ``@staticmethod`` + or ``@classmethod``, ``@rpc.functions.async_execution`` needs to be the + inner decorator to allow the target function be recognized as a static + or class function. This target function can still execute asynchronously + because, when accessed, the static or class method preserves attributes + installed by ``@rpc.functions.async_execution``. + + + Example:: + The returned :class:`~torch.futures.Future` object can come from + :meth:`~torch.distributed.rpc.rpc_async`, + :meth:`~torch.futures.Future.then`, or :class:`~torch.futures.Future` + constructor. The example below shows directly using the + :class:`~torch.futures.Future` returned by + :meth:`~torch.futures.Future.then`. + + >>> from torch.distributed import rpc + >>> + >>> # omitting setup and shutdown RPC + >>> + >>> # On all workers + >>> @rpc.functions.async_execution + >>> def async_add_chained(to, x, y, z): + >>> # This function runs on "worker1" and returns immediately when + >>> # the callback is installed through the `then(cb)` API. In the + >>> # mean time, the `rpc_async` to "worker2" can run concurrently. + >>> # When the return value of that `rpc_async` arrives at + >>> # "worker1", "worker1" will run the lambda function accordingly + >>> # and set the value for the previously returned `Future`, which + >>> # will then trigger RPC to send the result back to "worker0". + >>> return rpc.rpc_async(to, torch.add, args=(x, y)).then( + >>> lambda fut: fut.wait() + z + >>> ) + >>> + >>> # On worker0 + >>> # xdoctest: +SKIP + >>> ret = rpc.rpc_sync( + >>> "worker1", + >>> async_add_chained, + >>> args=("worker2", torch.ones(2), 1, 1) + >>> ) + >>> print(ret) # prints tensor([3., 3.]) + + When combined with TorchScript decorators, this decorator must be the + outmost one. + + >>> from torch import Tensor + >>> from torch.futures import Future + >>> from torch.distributed import rpc + >>> + >>> # omitting setup and shutdown RPC + >>> + >>> # On all workers + >>> @torch.jit.script + >>> def script_add(x: Tensor, y: Tensor) -> Tensor: + >>> return x + y + >>> + >>> @rpc.functions.async_execution + >>> @torch.jit.script + >>> def async_add(to: str, x: Tensor, y: Tensor) -> Future[Tensor]: + >>> return rpc.rpc_async(to, script_add, (x, y)) + >>> + >>> # On worker0 + >>> ret = rpc.rpc_sync( + >>> "worker1", + >>> async_add, + >>> args=("worker2", torch.ones(2), 1) + >>> ) + >>> print(ret) # prints tensor([2., 2.]) + + When combined with static or class method, this decorator must be the + inner one. + + >>> from torch.distributed import rpc + >>> + >>> # omitting setup and shutdown RPC + >>> + >>> # On all workers + >>> class AsyncExecutionClass: + >>> + >>> @staticmethod + >>> @rpc.functions.async_execution + >>> def static_async_add(to, x, y, z): + >>> return rpc.rpc_async(to, torch.add, args=(x, y)).then( + >>> lambda fut: fut.wait() + z + >>> ) + >>> + >>> @classmethod + >>> @rpc.functions.async_execution + >>> def class_async_add(cls, to, x, y, z): + >>> ret_fut = torch.futures.Future() + >>> rpc.rpc_async(to, torch.add, args=(x, y)).then( + >>> lambda fut: ret_fut.set_result(fut.wait() + z) + >>> ) + >>> return ret_fut + >>> + >>> @rpc.functions.async_execution + >>> def bound_async_add(self, to, x, y, z): + >>> return rpc.rpc_async(to, torch.add, args=(x, y)).then( + >>> lambda fut: fut.wait() + z + >>> ) + >>> + >>> # On worker0 + >>> ret = rpc.rpc_sync( + >>> "worker1", + >>> AsyncExecutionClass.static_async_add, + >>> args=("worker2", torch.ones(2), 1, 2) + >>> ) + >>> print(ret) # prints tensor([4., 4.]) + >>> + >>> ret = rpc.rpc_sync( + >>> "worker1", + >>> AsyncExecutionClass.class_async_add, + >>> args=("worker2", torch.ones(2), 1, 2) + >>> ) + >>> print(ret) # prints tensor([4., 4.]) + + This decorator also works with RRef helpers, i.e., . + :meth:`torch.distributed.rpc.RRef.rpc_sync`, + :meth:`torch.distributed.rpc.RRef.rpc_async`, and + :meth:`torch.distributed.rpc.RRef.remote`. + + >>> from torch.distributed import rpc + >>> + >>> # reuse the AsyncExecutionClass class above + >>> rref = rpc.remote("worker1", AsyncExecutionClass) + >>> ret = rref.rpc_sync().static_async_add("worker2", torch.ones(2), 1, 2) + >>> print(ret) # prints tensor([4., 4.]) + >>> + >>> rref = rpc.remote("worker1", AsyncExecutionClass) + >>> ret = rref.rpc_async().static_async_add("worker2", torch.ones(2), 1, 2).wait() + >>> print(ret) # prints tensor([4., 4.]) + >>> + >>> rref = rpc.remote("worker1", AsyncExecutionClass) + >>> ret = rref.remote().static_async_add("worker2", torch.ones(2), 1, 2).to_here() + >>> print(ret) # prints tensor([4., 4.]) + """ + + @functools.wraps(fn) + def wrapper(*args, **kwargs): + return fn(*args, **kwargs) + + # Can't declare and use attributes of function objects (mypy#2087) + wrapper._wrapped_async_rpc_function = fn # type: ignore[attr-defined] + return wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/internal.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/internal.py new file mode 100644 index 0000000000000000000000000000000000000000..1b5666ec6b0052d15d60acb1bea3214a98af7dda --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/internal.py @@ -0,0 +1,286 @@ +# mypy: allow-untyped-defs +import collections +import copyreg +import io +import pickle +import sys +import threading +import traceback +from enum import Enum + +import torch +import torch.distributed as dist +from torch._C._distributed_rpc import _get_current_rpc_agent + + +__all__ = ["RPCExecMode", "serialize", "deserialize", "PythonUDF", "RemoteException"] + +# Thread local tensor tables to store tensors while pickling torch.Tensor +# objects +_thread_local_tensor_tables = threading.local() +_pickler = pickle.Pickler +_unpickler = pickle.Unpickler + + +class RPCExecMode(Enum): + SYNC = "sync" + ASYNC = "async" + ASYNC_JIT = "async_jit" + REMOTE = "remote" + + +class _InternalRPCPickler: + r""" + This class provides serialize() and deserialize() interfaces to serialize + data to be "binary string + tensor table" format + So for RPC python UDF function and args, non tensor data will be serialized + into regular binary string, tensor data will be put into thread local tensor + tables, this serialization format is consistent with builtin operator and args + using JIT pickler. This format will make tensor handling in C++ much easier, + e.g. attach tensor to distributed autograd graph in C++ + """ + + def __init__(self): + # Ignore type error because dispatch_table is defined in third-party package + self._dispatch_table = copyreg.dispatch_table.copy() # type: ignore[attr-defined] + self._dispatch_table[torch.Tensor] = self._tensor_reducer + # Used for registering customized picklers. + self._class_reducer_dict = {} + + def _register_reducer(self, obj_class, reducer): + # For the same class, only register the reducer once. + if obj_class not in self._class_reducer_dict: + self._class_reducer_dict[obj_class] = reducer + + @classmethod + def _tensor_receiver(cls, tensor_index): + global _thread_local_tensor_tables + return _thread_local_tensor_tables.recv_tables[tensor_index] + + def _tensor_reducer(self, tensor): + global _thread_local_tensor_tables + _thread_local_tensor_tables.send_tables.append(tensor) + tensor_index = len(_thread_local_tensor_tables.send_tables) - 1 + return (_InternalRPCPickler._tensor_receiver, (tensor_index,)) + + @classmethod + def _py_rref_receiver(cls, rref_fork_data): + return dist.rpc.PyRRef._deserialize(rref_fork_data) + + def _py_rref_reducer(self, py_rref): + rref_fork_data = py_rref._serialize() + return (_InternalRPCPickler._py_rref_receiver, (rref_fork_data,)) + + def _rref_reducer(self, rref): + return self._py_rref_reducer(rref) + + @classmethod + def _script_module_receiver(cls, script_module_serialized): + """ + Given a serialized representation of a ScriptModule created with torch.jit.save, + loads and returns the ScriptModule. + """ + f = io.BytesIO(script_module_serialized) + m = torch.jit.load(f) + return m + + def _script_module_reducer(self, script_module): + """ + Serializes a ScriptModule. + """ + f = io.BytesIO() + torch.jit.save(script_module, f) + return (_InternalRPCPickler._script_module_receiver, (f.getvalue(),)) + + def serialize(self, obj): + r""" + Serialize non tensor data into binary string, tensor data into + tensor table + """ + f = io.BytesIO() + p = _pickler(f) + p.dispatch_table = self._dispatch_table + + # rpc api could accept user picklers inheriting from _InternalRPCPickler to serialize rref, + # user picklers could have different initialization function from _InternalRPCPickler, + # but all the user picklers should call serialize() and use _rref_reducer to pickle rref + # in python. also, when _internal_rpc_pickler is imported to rpc/api.py, rpc.RRef is not + # compiled yet, it is not good place to access rpc.RRef inside _InternalRPCPickler constructor, + # so putting rref's dispatch table here + # + # The return value of a `rpc.remote(..)` call is type of `rpc.PyRRef`. + # The deserialized RRef object on an RPC receiver side is type of `rpc.PyRRef`. + # Ignore type error because dispatch_table is defined in third-party package + p.dispatch_table[dist.rpc.PyRRef] = self._py_rref_reducer # type: ignore[index] + # An RRef created locally by RRef Python constructor is type of `rpc.RRef`. + # Ignore type error because dispatch_table is defined in third-party package + p.dispatch_table[dist.rpc.RRef] = self._rref_reducer # type: ignore[index] + + # Add dispatch pickling for ScriptModule or its subclass. + if isinstance(obj, torch.jit.ScriptModule): + # Ignore type error because dispatch_table is defined in third-party package + p.dispatch_table[obj.__class__] = self._script_module_reducer # type: ignore[index] + + # Install customized picklers. + for class_name in self._class_reducer_dict: + p.dispatch_table[class_name] = self._class_reducer_dict[class_name] # type: ignore[index] + + # save _thread_local_tensor_tables.send_tables if it is in nested call + global _thread_local_tensor_tables + if hasattr(_thread_local_tensor_tables, "send_tables"): + old_send_tables = _thread_local_tensor_tables.send_tables + else: + old_send_tables = None + _thread_local_tensor_tables.send_tables = [] + + p.dump(obj) + + # restore _thread_local_tensor_tables.send_tables if return + # from nested call, otherwise clean up the table + tensors = _thread_local_tensor_tables.send_tables + if old_send_tables is not None: + _thread_local_tensor_tables.send_tables = old_send_tables + else: + del _thread_local_tensor_tables.send_tables + + return (f.getvalue(), tensors) + + def deserialize(self, binary_data, tensor_table): + r""" + Deserialize binary string + tensor table to original obj + """ + # save _thread_local_tensor_tables.recv_tables if it is in nested call + global _thread_local_tensor_tables + if hasattr(_thread_local_tensor_tables, "recv_tables"): + old_recv_tables = _thread_local_tensor_tables.recv_tables + else: + old_recv_tables = None + _thread_local_tensor_tables.recv_tables = tensor_table + + try: + unpickler = _unpickler(io.BytesIO(binary_data)) + ret = unpickler.load() + except AttributeError as e: + # Occurs when function is not found on module/class during + # unpickling. + except_str = ( + str(e) + + """ Default RPC pickler does not serialize + function code. Ensure that UDFs are defined on both caller and + callee modules.""" + ) + ret = AttributeError(except_str) + # Ensure the stack trace gets preserved + ret.__cause__ = e + + # restore _thread_local_tensor_tables.recv_tables if return + # from nested call, otherwise clean up the table + if old_recv_tables is not None: + _thread_local_tensor_tables.recv_tables = old_recv_tables + else: + del _thread_local_tensor_tables.recv_tables + + return ret + + +# Create _internal_rpc_pickler only once to initialize _dispatch_table only once +_internal_rpc_pickler = _InternalRPCPickler() + + +def serialize(obj): + return _internal_rpc_pickler.serialize(obj) + + +def deserialize(binary_data, tensor_table): + return _internal_rpc_pickler.deserialize(binary_data, tensor_table) + + +def _run_function(python_udf): + r""" + This function is exclusively called from C++. + See ``torch/csrc/distributed/rpc/python_rpc_handler.cpp``. + + Runs a Python UDF and returns its return value. + Wraps any exception in ``RemoteException`` if the function raises. + """ + try: + if isinstance(python_udf, AttributeError): + raise python_udf + result = python_udf.func(*python_udf.args, **python_udf.kwargs) + except Exception as e: + # except str = exception info + traceback string + except_str = ( + f"On {_get_current_rpc_agent().get_worker_info()}:\n" + f"{repr(e)}\n{traceback.format_exc()}" + ) + print(except_str, file=sys.stderr) + result = RemoteException(except_str, type(e)) + return result + + +def _handle_exception(result): + if isinstance(result, RemoteException): + exception_msg = result.msg.encode("utf-8").decode("unicode_escape") + # We wrap exception re-creation here in case some exception classes + # cannot be constructed directly from a string. + exc = None + try: + exc = result.exception_type(exception_msg) + except BaseException as e: + raise RuntimeError( + f"Failed to create original exception type. Error msg was {str(e)}" + f" Original exception on remote side was {exception_msg}" + ) from e + + if exc is not None: + raise exc + + +def _build_rpc_profiling_key( + exec_type, func_name, current_worker_name, dst_worker_name +): + """ + Builds the key that RPC calls are profiled with using the autograd profiler. + This will be the name of the corresponding Event recorded in the profiler. + + Args: + exec_type (RPCExecMode): Type of RPC/RRef call + func_name (str): Name of function being profiled. + current_worker_name (str): Name of current worker. + dst_worker_name (str): Name of the destination worker. + + Returns: + String representing profiling key + """ + profile_key = ( + f"rpc_{exec_type.value}#{func_name}({current_worker_name} -> {dst_worker_name})" + ) + return profile_key + + +def _start_record_function(exec_type, func_name, current_worker_name, dest_worker_name): + """ + This function should be called from RPC/RRef functions to create a + RecordFunction object for profiling. This function also runs the before + callbacks that start the profiling, though the user is responsible for + running the appropriate callbacks when the function to be profiled finishes. + + Args: + exec_type (RPCExecMode): Type of RPC/RRef call + func_name (str): Name of function being profiled. + current_worker_name (str): Name of current worker. + dest_worker_name (str): Name of the destination worker. + + Returns: + An instance of `torch.autograd._RecordFunction`. + """ + if not torch.autograd._profiler_enabled(): + raise AssertionError("Autograd profiler should be enabled.") + profile_key = f"rpc_{exec_type.value}#{str(func_name)}({current_worker_name} -> {dest_worker_name})" + rf = torch.autograd._RecordFunction() # type: ignore[attr-defined] + torch.autograd._run_before_callbacks(rf, profile_key) # type: ignore[attr-defined] + return rf + + +PythonUDF = collections.namedtuple("PythonUDF", ["func", "args", "kwargs"]) +RemoteException = collections.namedtuple("RemoteException", ["msg", "exception_type"]) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/options.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/options.py new file mode 100644 index 0000000000000000000000000000000000000000..99471974e3e5f4cd07bcd55ede1411e1e9c13ad8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/options.py @@ -0,0 +1,179 @@ +# mypy: allow-untyped-defs +import torch + +from . import _is_tensorpipe_available, constants as rpc_contants + + +DeviceType = int | str | torch.device + +__all__ = ["TensorPipeRpcBackendOptions"] + + +def _to_device(device: DeviceType) -> torch.device: + device = torch.device(device) + if device.type != "cuda": + raise ValueError( + "`set_devices` expect a list of CUDA devices, but got " + f"device type {device.type}." + ) + return device + + +def _to_device_map( + device_map: dict[DeviceType, DeviceType], +) -> dict[torch.device, torch.device]: + full_device_map: dict[torch.device, torch.device] = {} + reverse_map: dict[torch.device, torch.device] = {} + for k, v in device_map.items(): + k, v = torch.device(k), torch.device(v) + if v in reverse_map: + raise ValueError( + "`device_map` only supports 1-to-1 mapping, " + f"trying to map {k} and {reverse_map[v]} to {v}" + ) + full_device_map[k] = v + reverse_map[v] = k + return full_device_map + + +def _to_device_list(devices: list[DeviceType]) -> list[torch.device]: + return list(map(_to_device, devices)) + + +if _is_tensorpipe_available: # type: ignore[has-type] + from torch._C._distributed_rpc import _TensorPipeRpcBackendOptionsBase +else: + _TensorPipeRpcBackendOptionsBase = object # type: ignore[assignment, misc] + + +# pyrefly: ignore [invalid-inheritance] +class TensorPipeRpcBackendOptions(_TensorPipeRpcBackendOptionsBase): + r""" + The backend options for + :class:`~torch.distributed.rpc.TensorPipeAgent`, derived from + :class:`~torch.distributed.rpc.RpcBackendOptions`. + + Args: + num_worker_threads (int, optional): The number of threads in the + thread-pool used by + :class:`~torch.distributed.rpc.TensorPipeAgent` to execute + requests (default: 16). + rpc_timeout (float, optional): The default timeout, in seconds, + for RPC requests (default: 60 seconds). If the RPC has not + completed in this timeframe, an exception indicating so will + be raised. Callers can override this timeout for individual + RPCs in :meth:`~torch.distributed.rpc.rpc_sync` and + :meth:`~torch.distributed.rpc.rpc_async` if necessary. + init_method (str, optional): The URL to initialize the distributed + store used for rendezvous. It takes any value accepted for the + same argument of :meth:`~torch.distributed.init_process_group` + (default: ``env://``). + device_maps (Dict[str, Dict], optional): Device placement mappings from + this worker to the callee. Key is the callee worker name and value + the dictionary (``Dict`` of ``int``, ``str``, or ``torch.device``) + that maps this worker's devices to the callee worker's devices. + (default: ``None``) + devices (List[int, str, or ``torch.device``], optional): all local + CUDA devices used by RPC agent. By Default, it will be initialized + to all local devices from its own ``device_maps`` and corresponding + devices from its peers' ``device_maps``. When processing CUDA RPC + requests, the agent will properly synchronize CUDA streams for + all devices in this ``List``. + """ + + def __init__( + self, + *, + num_worker_threads: int = rpc_contants.DEFAULT_NUM_WORKER_THREADS, + rpc_timeout: float = rpc_contants.DEFAULT_RPC_TIMEOUT_SEC, + init_method: str = rpc_contants.DEFAULT_INIT_METHOD, + device_maps: dict[str, dict[DeviceType, DeviceType]] | None = None, + devices: list[DeviceType] | None = None, + _transports: list | None = None, + _channels: list | None = None, + ): + full_device_maps = ( + {} + if device_maps is None + else {k: _to_device_map(v) for k, v in device_maps.items()} + ) + full_device_list = [] if devices is None else _to_device_list(devices) + super().__init__( + num_worker_threads, + _transports, + _channels, + rpc_timeout, + init_method, + full_device_maps, + full_device_list, + ) + + def set_device_map(self, to: str, device_map: dict[DeviceType, DeviceType]): + r""" + Set device mapping between each RPC caller and callee pair. This + function can be called multiple times to incrementally add + device placement configurations. + + Args: + to (str): Callee name. + device_map (Dict of int, str, or torch.device): Device placement + mappings from this worker to the callee. This map must be + invertible. + + Example: + >>> # xdoctest: +SKIP("distributed") + >>> # both workers + >>> def add(x, y): + >>> print(x) # tensor([1., 1.], device='cuda:1') + >>> return x + y, (x + y).to(2) + >>> + >>> # on worker 0 + >>> options = TensorPipeRpcBackendOptions( + >>> num_worker_threads=8, + >>> device_maps={"worker1": {0: 1}} + >>> # maps worker0's cuda:0 to worker1's cuda:1 + >>> ) + >>> options.set_device_map("worker1", {1: 2}) + >>> # maps worker0's cuda:1 to worker1's cuda:2 + >>> + >>> rpc.init_rpc( + >>> "worker0", + >>> rank=0, + >>> world_size=2, + >>> backend=rpc.BackendType.TENSORPIPE, + >>> rpc_backend_options=options + >>> ) + >>> + >>> x = torch.ones(2) + >>> rets = rpc.rpc_sync("worker1", add, args=(x.to(0), 1)) + >>> # The first argument will be moved to cuda:1 on worker1. When + >>> # sending the return value back, it will follow the invert of + >>> # the device map, and hence will be moved back to cuda:0 and + >>> # cuda:1 on worker0 + >>> print(rets[0]) # tensor([2., 2.], device='cuda:0') + >>> print(rets[1]) # tensor([2., 2.], device='cuda:1') + """ + full_device_map = _to_device_map(device_map) + curr_device_maps = super().device_maps + + if to in curr_device_maps: + for k, v in full_device_map.items(): + if k in curr_device_maps[to] and v != curr_device_maps[to][k]: + raise ValueError( + "`set_device_map` only supports 1-to-1 mapping, trying" + f" to map {k} to {v} and {curr_device_maps[to][k]}" + ) + + super()._set_device_map(to, full_device_map) + + def set_devices(self, devices: list[DeviceType]): + r""" + Set local devices used by the TensorPipe RPC agent. When processing + CUDA RPC requests, the TensorPipe RPC agent will properly synchronize + CUDA streams for all devices in this ``List``. + + Args: + devices (List of int, str, or torch.device): local devices used by + the TensorPipe RPC agent. + """ + self.devices = _to_device_list(devices) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/rref_proxy.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/rref_proxy.py new file mode 100644 index 0000000000000000000000000000000000000000..81b1d43b07dcf4f1feaaf882e07ad457c59dd900 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/rref_proxy.py @@ -0,0 +1,80 @@ +# mypy: allow-untyped-defs +from functools import partial + +import torch +from torch.futures import Future + +from . import functions, rpc_async +from .constants import UNSET_RPC_TIMEOUT + + +def _local_invoke(rref, func_name, args, kwargs): + return getattr(rref.local_value(), func_name)(*args, **kwargs) + + +@functions.async_execution +def _local_invoke_async_execution(rref, func_name, args, kwargs): + return getattr(rref.local_value(), func_name)(*args, **kwargs) + + +def _invoke_rpc(rref, rpc_api, func_name, timeout, *args, **kwargs): + def _rref_type_cont(rref_fut): + rref_type = rref_fut.value() + + _invoke_func = _local_invoke + # Bypass ScriptModules when checking for async function attribute. + bypass_type = issubclass(rref_type, torch.jit.ScriptModule) or issubclass( + rref_type, torch._C.ScriptModule + ) + if not bypass_type: + func = getattr(rref_type, func_name) + if hasattr(func, "_wrapped_async_rpc_function"): + _invoke_func = _local_invoke_async_execution + + return rpc_api( + rref.owner(), + _invoke_func, + args=(rref, func_name, args, kwargs), + timeout=timeout, + ) + + rref_fut = rref._get_type(timeout=timeout, blocking=False) + + if rpc_api is not rpc_async: + rref_fut.wait() + return _rref_type_cont(rref_fut) + else: + # A little explanation on this. + # rpc_async returns a Future pointing to the return value of `func_name`, it returns a `Future[T]` + # Calling _rref_type_cont from the `then` lambda causes Future wrapping. IOW, `then` returns a `Future[Future[T]]` + # To address that, we return a Future that is completed with the result of the async call. + result: Future = Future() + + def _wrap_rref_type_cont(fut): + try: + _rref_type_cont(fut).then(_complete_op) + except BaseException as ex: + result.set_exception(ex) + + def _complete_op(fut): + try: + result.set_result(fut.value()) + except BaseException as ex: + result.set_exception(ex) + + rref_fut.then(_wrap_rref_type_cont) + return result + + +# This class manages proxied RPC API calls for RRefs. It is entirely used from +# C++ (see python_rpc_handler.cpp). +class RRefProxy: + def __init__(self, rref, rpc_api, timeout=UNSET_RPC_TIMEOUT): + self.rref = rref + self.rpc_api = rpc_api + self.rpc_timeout = timeout + + def __getattr__(self, func_name): + return partial( + _invoke_rpc, self.rref, self.rpc_api, func_name, self.rpc_timeout + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/rpc/server_process_global_profiler.py b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/server_process_global_profiler.py new file mode 100644 index 0000000000000000000000000000000000000000..33ebd56ca95bb375c09d5cbc373ebf6477bec4f1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/rpc/server_process_global_profiler.py @@ -0,0 +1,188 @@ +#!/usr/bin/python3 +# mypy: allow-untyped-defs + +import itertools + +import torch + +# pyrefly: ignore [deprecated] +from torch.autograd.profiler_legacy import profile + +from . import ( + _disable_server_process_global_profiler, + _enable_server_process_global_profiler, +) + + +__all__: list[str] = [] + + +class _server_process_global_profile(profile): + """ + It has the same API as ``torch.autograd.profiler.profile`` class, + except that it enables profiling on all threads running RPC server request callbacks. + + Context manager that manages autograd profiler state and holds a summary of results. + Under the hood it just records events of functions being executed in C++ and + exposes those events to Python. You can wrap any code into it and it will + only report runtime of PyTorch functions. + Note: profiler is thread local and is automatically propagated into the async tasks + + Args: + enabled (bool, optional): Setting this to False makes this context manager a no-op. + Default: ``True``. + + use_cuda (bool, optional): Enables timing of CUDA events as well using the cudaEvent API. + Adds approximately 4us of overhead to each tensor operation. + Default: ``False`` + + record_shapes (bool, optional): If shapes recording is set, information + about input dimensions will be collected. This allows one to see which + dimensions have been used under the hood and further group by them + using prof.key_averages(group_by_input_shape=True). Please note that + shape recording might skew your profiling data. It is recommended to + use separate runs with and without shape recording to validate the timing. + Most likely the skew will be negligible for bottom most events (in a case + of nested function calls). But for higher level functions the total + self cpu time might be artificially increased because of the shape + collection. + + profile_memory (bool, optional): Whether to report memory usage, default: ``False`` + + .. warning:: + Enabling memory profiling incurs additional profiler overhead + + .. warning:: + Due to some CUDA multiprocessing limitations (see :ref:`multiprocessing-cuda-note`), + one cannot use the profiler with ``use_cuda = True`` to benchmark + DataLoaders with ``num_workers > 0``. If you wish to benchmark data loading, + please use ``use_cuda = False`` or ``num_workers = 0``. + + Example: + >>> # xdoctest: +SKIP + >>> # On worker 0: + >>> import torch + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker0", rank=0, world_size=2) + >>> x, y = torch.tensor(1), torch.tensor(2) + >>> outer_profile_rref = rpc.remote( + ... dst_worker_name, rpc._server_process_global_profile + ... ) + >>> outer_profile_rref.rpc_sync().__enter__() + >>> rpc.rpc_sync(dst_worker_name, torch.add, (x, y)) + >>> inner_profile_rref = rpc.remote( + ... dst_worker_name, rpc._server_process_global_profile + ... ) + >>> inner_profile_rref.rpc_sync().__enter__() + >>> rpc.rpc_sync(dst_worker_name, torch.sub, (x, y)) + >>> inner_profile_rref.rpc_sync().__exit__(None, None, None) + >>> outer_profile_rref.rpc_sync().__exit__(None, None, None) + >>> print(inner_profile_rref.rpc_sync().key_averages()) + --------- --------------- --------------- --------------- --------------- --------------- --------------- + Name Self CPU total % Self CPU total CPU total % CPU total CPU time avg Number of Calls + --------- --------------- --------------- --------------- --------------- --------------- --------------- + sub 85.06% 76.275us 100.00% 89.667us 89.667us 1 + empty 14.94% 13.392us 14.94% 13.392us 13.392us 1 + --------- --------------- --------------- --------------- --------------- --------------- --------------- + Self CPU time total: 89.667us + >>> print(outer_profile_rref.rpc_sync().key_averages()) + --------- --------------- --------------- --------------- --------------- --------------- --------------- + Name Self CPU total % Self CPU total CPU total % CPU total CPU time avg Number of Calls + --------- --------------- --------------- --------------- --------------- --------------- --------------- + sub 35.65% 76.275us 41.91% 89.667us 89.667us 1 + empty 12.67% 27.101us 12.67% 27.101us 13.551us 2 + add 51.68% 110.550us 58.09% 124.259us 124.259us 1 + --------- --------------- --------------- --------------- --------------- --------------- --------------- + Self CPU time total: 213.926us + >>> rpc.shutdown() + + >>> # On worker 1: + >>> import torch.distributed.rpc as rpc + >>> rpc.init_rpc("worker1", rank=1, world_size=2) + >>> # wait for worker 0 to finish work, and then shutdown. + >>> rpc.shutdown() + """ + + def __init__(self, *args, **kwargs): + super().__init__(*args, **kwargs) + + def __enter__(self): + """ + Turn on server-side process-global profiling. + This enables thread-local profiler on all RPC threads running server-side request callbacks. + """ + if not self.enabled: + return + + if self.entered: # type: ignore[has-type] + raise RuntimeError("autograd profiler traces are not reentrant") + self.entered = True + + profiler_kind = ( + torch.autograd.ProfilerState.CUDA + if self.use_cuda + else torch.autograd.ProfilerState.CPU + ) + profiler_config = torch.autograd.ProfilerConfig( + profiler_kind, + self.record_shapes, + self.profile_memory, + False, + False, + False, + torch.profiler._ExperimentalConfig(), + ) + _enable_server_process_global_profiler(profiler_config) + return self + + def __exit__(self, exc_type, exc_val, exc_tb): + """ + Turn off server-side process-global profiling. + Aggregate all profiling events recorded by RPC threads. + + These attributes are assigned on exiting context. + + Attributes: + function_events (torch.autograd.profiler.EventList). It's a list that has helper + methods, like 1) show record items in a pretty-print table. + 2) do averaging by grouping on keys. 3) and more. + + process_global_function_events (List[torch.autograd.profiler.FunctionEvent]). + It's a list of ``FunctionEvent`` elements. Every element is a profiling result + of an RPC request handling within the profiling range. + """ + if not self.enabled: + return + + process_global_events = _disable_server_process_global_profiler() + + # Every element in this list is a thread profiling result from an RPC request handling. + process_global_function_events = [] + for thread_local_events in process_global_events: + # Parse from ``Event``s to ``FunctionEvent``s. + thread_local_function_events = ( + torch.autograd.profiler_legacy._parse_legacy_records( + thread_local_events + ) + ) + thread_local_function_events.sort( + key=lambda function_event: [ + function_event.time_range.start, + -(function_event.time_range.end), + ] + ) + process_global_function_events.append(thread_local_function_events) + + flattened_function_events = list( + itertools.chain.from_iterable(process_global_function_events) + ) + self.function_events = torch.autograd.profiler_util.EventList( + flattened_function_events, + use_device="cuda" if self.use_cuda else None, + profile_memory=self.profile_memory, + ) + self.function_events._build_tree() + + self.process_global_function_events = process_global_function_events + + return False diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/run.py b/.venv/lib/python3.12/site-packages/torch/distributed/run.py new file mode 100644 index 0000000000000000000000000000000000000000..7f11276373cac492f3fd67515fa16397e2745a3f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/run.py @@ -0,0 +1,1142 @@ +#!/usr/bin/env python3 +# mypy: allow-untyped-defs + +# Copyright (c) Facebook, Inc. and its affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Module ``torch.distributed.run``. + +``torch.distributed.run`` is a module that spawns up multiple distributed +training processes on each of the training nodes. + +``torchrun`` is a python +`console script `_ +to the main module +`torch.distributed.run `_ +declared in the ``entry_points`` configuration in +`setup.py `_. +It is equivalent to invoking ``python -m torch.distributed.run``. + +``torchrun`` can be used for single-node distributed training, in which one or +more processes per node will be spawned. It can be used for either +CPU training or GPU training. If it is used for GPU training, +each distributed process will be operating on a single GPU. This can achieve +well-improved single-node training performance. ``torchrun`` can also be used in +multi-node distributed training, by spawning up multiple processes on each node +for well-improved multi-node distributed training performance as well. +This will especially be beneficial for systems with multiple Infiniband +interfaces that have direct-GPU support, since all of them can be utilized for +aggregated communication bandwidth. + +In both cases of single-node distributed training or multi-node distributed +training, ``torchrun`` will launch the given number of processes per node +(``--nproc-per-node``). If used for GPU training, this number needs to be less +or equal to the number of GPUs on the current system (``nproc_per_node``), +and each process will be operating on a single GPU from *GPU 0 to +GPU (nproc_per_node - 1)*. + +.. versionchanged:: 2.0.0 + + ``torchrun`` will pass the ``--local-rank=`` argument to your script. + From PyTorch 2.0.0 onwards, the dashed ``--local-rank`` is preferred over the + previously used underscored ``--local_rank``. + + For backward compatibility, it may be necessary for users to handle both + cases in their argument parsing code. This means including both ``"--local-rank"`` + and ``"--local_rank"`` in the argument parser. If only ``"--local_rank"`` is + provided, ``torchrun`` will trigger an error: "error: unrecognized arguments: + --local-rank=". For training code that only supports PyTorch 2.0.0+, + including ``"--local-rank"`` should be sufficient. + + :: + + >>> # xdoctest: +SKIP + >>> import argparse + >>> parser = argparse.ArgumentParser() + >>> parser.add_argument("--local-rank", "--local_rank", type=int) + >>> args = parser.parse_args() + +Usage +----- + +Single-node multi-worker +++++++++++++++++++++++++ + +:: + + torchrun + --standalone + --nnodes=1 + --nproc-per-node=$NUM_TRAINERS + YOUR_TRAINING_SCRIPT.py (--arg1 ... train script args...) + +.. note:: ``--nproc-per-node`` may be + ``"gpu"`` (spawn one process per GPU), + ``"cpu"`` (spawn one process per CPU), + ``"xpu"`` (spawn one process per XPU), + ``"auto"`` (equivalent to ``"gpu"`` if CUDA is available, + else equivalent to ``"xpu"`` if XPU is available, + else equivalent to ``"cpu"``), + or an integer specifying the number of processes. + See `torch.distributed.run.determine_local_world_size + `_ + for more details. + +Stacked single-node multi-worker +++++++++++++++++++++++++++++++++ + +To run multiple instances (separate jobs) of single-node, multi-worker on the +same host, we need to make sure that each instance (job) is +setup on different ports to avoid port conflicts (or worse, two jobs being merged +as a single job). To do this you have to run with ``--rdzv-backend=c10d`` +and specify a different port by setting ``--rdzv-endpoint=localhost:$PORT_k``. +For ``--nodes=1``, its often convenient to let ``torchrun`` pick a free random +port automatically instead of manually assigning different ports for each run. + +:: + + torchrun + --rdzv-backend=c10d + --rdzv-endpoint=localhost:0 + --nnodes=1 + --nproc-per-node=$NUM_TRAINERS + YOUR_TRAINING_SCRIPT.py (--arg1 ... train script args...) + + +Fault tolerant (fixed sized number of workers, no elasticity, tolerates 3 failures) ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + +:: + + torchrun + --nnodes=$NUM_NODES + --nproc-per-node=$NUM_TRAINERS + --max-restarts=3 + --rdzv-id=$JOB_ID + --rdzv-backend=c10d + --rdzv-endpoint=$HOST_NODE_ADDR + YOUR_TRAINING_SCRIPT.py (--arg1 ... train script args...) + +``HOST_NODE_ADDR``, in form [:] (e.g. node1.example.com:29400), specifies the node and +the port on which the C10d rendezvous backend should be instantiated and hosted. It can be any +node in your training cluster, but ideally you should pick a node that has a high bandwidth. + +.. note:: + If no port number is specified ``HOST_NODE_ADDR`` defaults to 29400. + +Elastic (``min=1``, ``max=4``, tolerates up to 3 membership changes or failures) +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + +:: + + torchrun + --nnodes=1:4 + --nproc-per-node=$NUM_TRAINERS + --max-restarts=3 + --rdzv-id=$JOB_ID + --rdzv-backend=c10d + --rdzv-endpoint=$HOST_NODE_ADDR + YOUR_TRAINING_SCRIPT.py (--arg1 ... train script args...) + +``HOST_NODE_ADDR``, in form [:] (e.g. node1.example.com:29400), specifies the node and +the port on which the C10d rendezvous backend should be instantiated and hosted. It can be any +node in your training cluster, but ideally you should pick a node that has a high bandwidth. + +.. note:: + If no port number is specified ``HOST_NODE_ADDR`` defaults to 29400. + +Shell completion +---------------- + +``torchrun`` can emit a completion script for ``bash``, ``zsh`` or ``tcsh``. The +script is generated from the argument parser, so it stays in sync with the +options above. This requires the optional `shtab `_ +package (``pip install shtab``); torchrun only imports it when the flag is used. + +:: + + # zsh + torchrun --print-completion zsh > ~/.zsh/completions/_torchrun + + # bash + torchrun --print-completion bash > ~/.local/share/bash-completion/completions/torchrun + +Refer to your shell's documentation for the directory it loads completions from; +the paths above are the common defaults. + +Note on rendezvous backend +-------------------------- + +For multi-node training you need to specify: + +1. ``--rdzv-id``: A unique job id (shared by all nodes participating in the job) +2. ``--rdzv-backend``: An implementation of + :py:class:`torch.distributed.elastic.rendezvous.RendezvousHandler` +3. ``--rdzv-endpoint``: The endpoint where the rendezvous backend is running; usually in form + ``host:port``. + +Currently ``c10d`` (recommended), ``etcd-v2``, and ``etcd`` (legacy) rendezvous backends are +supported out of the box. To use ``etcd-v2`` or ``etcd``, setup an etcd server with the ``v2`` api +enabled (e.g. ``--enable-v2``). + +.. warning:: + ``etcd-v2`` and ``etcd`` rendezvous use etcd API v2. You MUST enable the v2 API on the etcd + server. Our tests use etcd v3.4.3. + +.. warning:: + For etcd-based rendezvous we recommend using ``etcd-v2`` over ``etcd`` which is functionally + equivalent, but uses a revised implementation. ``etcd`` is in maintenance mode and will be + removed in a future version. + +Definitions +----------- + +1. ``Node`` - A physical instance or a container; maps to the unit that the job manager works with. + +2. ``Worker`` - A worker in the context of distributed training. + +3. ``WorkerGroup`` - The set of workers that execute the same function (e.g. trainers). + +4. ``LocalWorkerGroup`` - A subset of the workers in the worker group running on the same node. + +5. ``RANK`` - The rank of the worker within a worker group. + +6. ``WORLD_SIZE`` - The total number of workers in a worker group. + +7. ``LOCAL_RANK`` - The rank of the worker within a local worker group. + +8. ``LOCAL_WORLD_SIZE`` - The size of the local worker group. + +9. ``rdzv_id`` - A user-defined id that uniquely identifies the worker group for a job. This id is + used by each node to join as a member of a particular worker group. + +9. ``rdzv_backend`` - The backend of the rendezvous (e.g. ``c10d``). This is typically a strongly + consistent key-value store. + +10. ``rdzv_endpoint`` - The rendezvous backend endpoint; usually in form ``:``. + +A ``Node`` runs ``LOCAL_WORLD_SIZE`` workers which comprise a ``LocalWorkerGroup``. The union of +all ``LocalWorkerGroups`` in the nodes in the job comprise the ``WorkerGroup``. + +Environment Variables +--------------------- + +The following environment variables are made available to you in your script: + +1. ``LOCAL_RANK`` - The local rank. + +2. ``RANK`` - The global rank. + +3. ``GROUP_RANK`` - The rank of the worker group. A number between 0 and ``max_nnodes``. When + running a single worker group per node, this is the rank of the node. + +4. ``ROLE_RANK`` - The rank of the worker across all the workers that have the same role. The role + of the worker is specified in the ``WorkerSpec``. + +5. ``LOCAL_WORLD_SIZE`` - The local world size (e.g. number of workers running locally); equals to + ``--nproc-per-node`` specified on ``torchrun``. + +6. ``WORLD_SIZE`` - The world size (total number of workers in the job). + +7. ``ROLE_WORLD_SIZE`` - The total number of workers that was launched with the same role specified + in ``WorkerSpec``. + +8. ``MASTER_ADDR`` - The FQDN of the host that is running worker with rank 0; used to initialize + the Torch Distributed backend. + +9. ``MASTER_PORT`` - The port on the ``MASTER_ADDR`` that can be used to host the C10d TCP store. + +10. ``TORCHELASTIC_RESTART_COUNT`` - The number of worker group restarts so far. + +11. ``TORCHELASTIC_MAX_RESTARTS`` - The configured maximum number of restarts. + +12. ``TORCHELASTIC_RUN_ID`` - Equal to the rendezvous ``run_id`` (e.g. unique job id). + +13. ``PYTHON_EXEC`` - System executable override. If provided, the python user script will + use the value of ``PYTHON_EXEC`` as executable. The `sys.executable` is used by default. + +Logging +------- + +By default each worker's ``stdout``/``stderr`` go to the console unchanged, which +interleaves the output of every rank with no way to tell them apart. + +``--redirects`` writes the streams to log files under ``--log-dir`` instead of the +console; ``--tee`` writes them to log files *and* echoes them to the console. Both +take the same format: a single value applies to all workers (``3`` for both +streams, ``1`` for stdout, ``2`` for stderr), or a per-local-rank mapping such as +``0:1,1:2``. For example ``--tee 3`` tees both streams for every worker. + +Tee'd console lines are prefixed with ``[${role_name}${local_rank}]:`` (e.g. +``[default3]: foobar``). Use ``--log-line-prefix-template`` to change that; the +macros ``${role_name}``, ``${local_rank}``, ``${rank}`` and ``${hostname}`` are +substituted per worker. ``${hostname}`` is the name of the node the worker runs +on, which is what identifies the offending host in a multi-node job:: + + torchrun --nnodes 2 --nproc-per-node 8 --tee 3 \ + --log-line-prefix-template "${hostname}:${rank}: " train.py + + r12i0n8:3: python: src/psm2_nccl_net.c:756: Assertion `r->used' failed. + r12i0n8:3: Fatal Python error: Segmentation fault + +The template may also be set with the ``TORCHELASTIC_LOG_LINE_PREFIX_TEMPLATE`` +environment variable; the command line option takes precedence. +``--local-ranks-filter`` restricts which ranks reach the console, without +affecting the log files written by ``--redirects``/``--tee``. + +Deployment +---------- + +1. (Not needed for the C10d backend) Start the rendezvous backend server and get the endpoint (to be + passed as ``--rdzv-endpoint`` to ``torchrun``) + +2. Single-node multi-worker: Start ``torchrun`` on the host to start the agent process which + creates and monitors a local worker group. + +3. Multi-node multi-worker: Start ``torchrun`` with the same arguments on all the nodes + participating in training. + +When using a job/cluster manager, the entry point command to the multi-node job should be ``torchrun``. + +Failure Modes +------------- + +1. Worker failure: For a training job with ``n`` workers, if ``k<=n`` workers fail all workers + are stopped and restarted up to ``max_restarts``. + +2. Agent failure: An agent failure results in a local worker group failure. It is up to the job + manager to fail the entire job (gang semantics) or attempt to replace the node. Both behaviors + are supported by the agent. + +3. Node failure: Same as agent failure. + +Membership Changes +------------------ + +1. Node departure (scale-down): The agent is notified of the departure, all existing workers are + stopped, a new ``WorkerGroup`` is formed, and all workers are started with a new ``RANK`` and + ``WORLD_SIZE``. + +2. Node arrival (scale-up): The new node is admitted to the job, all existing workers are stopped, + a new ``WorkerGroup`` is formed, and all workers are started with a new ``RANK`` and + ``WORLD_SIZE``. + +NUMA Binding +------------ + +On multi-GPU systems with NUMA (Non-Uniform Memory Access) architecture, you can improve +performance by binding worker processes to CPUs near their assigned GPUs. Use the +``--numa-binding`` flag: + +:: + + torchrun --numa-binding=node --nproc-per-node=8 train.py + +See :ref:`numa-api` for more details. + +Important Notices +----------------- + +1. This utility and multi-process distributed (single-node or + multi-node) GPU training currently only achieves the best performance using + the NCCL distributed backend. Thus NCCL backend is the recommended backend to + use for GPU training. + +2. The environment variables necessary to initialize a Torch process group are provided to you by + this module, no need for you to pass ``RANK`` manually. To initialize a process group in your + training script, simply run: + +:: + + >>> # xdoctest: +SKIP("stub") + >>> import torch.distributed as dist + >>> dist.init_process_group(backend="gloo|nccl") + +3. In your training program, you can either use regular distributed functions + or use :func:`torch.nn.parallel.DistributedDataParallel` module. If your + training program uses GPUs for training and you would like to use + :func:`torch.nn.parallel.DistributedDataParallel` module, + here is how to configure it. + +:: + + local_rank = int(os.environ["LOCAL_RANK"]) + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], output_device=local_rank + ) + +Please ensure that ``device_ids`` argument is set to be the only GPU device id +that your code will be operating on. This is generally the local rank of the +process. In other words, the ``device_ids`` needs to be ``[int(os.environ("LOCAL_RANK"))]``, +and ``output_device`` needs to be ``int(os.environ("LOCAL_RANK"))`` in order to use this +utility + + +4. On failures or membership changes ALL surviving workers are killed immediately. Make sure to + checkpoint your progress. The frequency of checkpoints should depend on your job's tolerance + for lost work. + +5. This module only supports homogeneous ``LOCAL_WORLD_SIZE``. That is, it is assumed that all + nodes run the same number of local workers (per role). + +6. ``RANK`` is NOT stable. Between restarts, the local workers on a node can be assigned a + different range of ranks than before. NEVER hard code any assumptions about the stable-ness of + ranks or some correlation between ``RANK`` and ``LOCAL_RANK``. + +7. When using elasticity (``min_size!=max_size``) DO NOT hard code assumptions about + ``WORLD_SIZE`` as the world size can change as nodes are allowed to leave and join. + +8. It is recommended for your script to have the following structure: + +:: + + def main(): + load_checkpoint(checkpoint_path) + initialize() + train() + + + def train(): + for batch in iter(dataset): + train_step(batch) + + if should_checkpoint: + save_checkpoint(checkpoint_path) + +9. (Recommended) On worker errors, this tool will summarize the details of the error + (e.g. time, rank, host, pid, traceback, etc). On each node, the first error (by timestamp) + is heuristically reported as the "Root Cause" error. To get tracebacks as part of this + error summary print out, you must decorate your main entrypoint function in your + training script as shown in the example below. If not decorated, then the summary + will not include the traceback of the exception and will only contain the exitcode. + For details on torchelastic error handling see: https://pytorch.org/docs/stable/elastic/errors.html + +:: + + from torch.distributed.elastic.multiprocessing.errors import record + + + @record + def main(): + # do train + pass + + + if __name__ == "__main__": + main() +""" + +import os +import sys +import uuid +from argparse import Action as _Action, ArgumentParser, REMAINDER +from collections.abc import Callable +from importlib import metadata + +import torch +from torch.distributed.argparse_util import check_env, env +from torch.distributed.elastic.multiprocessing import DefaultLogsSpecs, LogsSpecs, Std +from torch.distributed.elastic.multiprocessing.errors import record +from torch.distributed.elastic.rendezvous.utils import _parse_rendezvous_config +from torch.distributed.elastic.utils import macros +from torch.distributed.elastic.utils.logging import get_logger +from torch.distributed.launcher.api import elastic_launch, LaunchConfig +from torch.numa.binding import ( + AffinityMode as _AffinityMode, # Signify as private with _ + NumaOptions as _NumaOptions, +) +from torch.utils.backend_registration import _get_custom_mod_func + + +logger = get_logger(__name__) + + +class _PrintCompletionAction(_Action): + """Print a shell completion script for ``torchrun`` and exit. + + ``shtab`` derives the script from this parser, so completions stay in sync + with the options defined below. It is an optional dependency: torchrun only + imports it when this flag is used. + """ + + def __call__(self, parser, namespace, values, option_string=None): + try: + import shtab + except ImportError: + parser.exit( + 1, + f"{option_string} requires the 'shtab' package: pip install shtab\n", + ) + else: + # Completions are for the `torchrun` console script. parser.prog is + # `__main__.py` when invoked as `python -m torch.distributed.run`, + # which would bind the completion to the wrong command. + parser.prog = "torchrun" + print(shtab.complete(parser, shell=values)) + parser.exit() + + +def get_args_parser() -> ArgumentParser: + """Parse the command line options.""" + parser = ArgumentParser(description="Torch Distributed Elastic Training Launcher") + + def comma_separated_list(value): + placeholder = "" + value = value.replace(",,", placeholder) + items = value.split(",") + items = [item.replace(placeholder, ",") for item in items] + return items + + # + # Worker/node size related arguments. + # + + parser.add_argument( + "--nnodes", + action=env, + type=str, + default="1:1", + help="Number of nodes, or the range of nodes in form :.", + ) + parser.add_argument( + "--nproc-per-node", + "--nproc_per_node", + action=env, + type=str, + default="1", + help="Number of workers per node; supported values: [auto, cpu, gpu, xpu, int].", + ) + + # + # Rendezvous related arguments + # + + parser.add_argument( + "--rdzv-backend", + "--rdzv_backend", + action=env, + type=str, + default="static", + help="Rendezvous backend.", + ) + parser.add_argument( + "--rdzv-endpoint", + "--rdzv_endpoint", + action=env, + type=str, + default="", + help="Rendezvous backend endpoint; usually in form :.", + ) + parser.add_argument( + "--rdzv-id", + "--rdzv_id", + action=env, + type=str, + default="none", + help="User-defined group id.", + ) + parser.add_argument( + "--rdzv-conf", + "--rdzv_conf", + action=env, + type=str, + default="", + help="Additional rendezvous configuration (=,=,...).", + ) + parser.add_argument( + "--standalone", + action=check_env, + help="Start a local standalone rendezvous backend that is represented by a C10d TCP store " + "on a free port. Useful when launching single-node, multi-worker job. If specified " + "--rdzv-backend, --rdzv-endpoint, --rdzv-id are auto-assigned and any explicitly set values " + "are ignored.", + ) + + # + # User-code launch related arguments. + # + + parser.add_argument( + "--max-restarts", + "--max_restarts", + action=env, + type=int, + default=0, + help="Maximum number of worker group restarts before failing.", + ) + parser.add_argument( + "--monitor-interval", + "--monitor_interval", + action=env, + type=float, + default=0.1, + help="Interval, in seconds, to monitor the state of workers.", + ) + parser.add_argument( + "--start-method", + "--start_method", + action=env, + type=str, + default="spawn", + choices=["spawn", "fork", "forkserver"], + help="Multiprocessing start method to use when creating workers.", + ) + parser.add_argument( + "--event-log-handler", + "--event_log_handler", + action=env, + type=str, + default="null", + help="name of a registered event logging handler (see: https://docs.pytorch.org/docs/stable/elastic/events.html)", + ) + parser.add_argument( + "--role", + action=env, + type=str, + default="default", + help="User-defined role for the workers.", + ) + parser.add_argument( + "-m", + "--module", + action=check_env, + help="Change each process to interpret the launch script as a Python module, executing " + "with the same behavior as 'python -m'.", + ) + parser.add_argument( + "--no-python", + "--no_python", + action=check_env, + help="Skip prepending the training script with 'python' - just execute it directly. Useful " + "when the script is not a Python script.", + ) + + parser.add_argument( + "--run-path", + "--run_path", + action=check_env, + help="Run the training script with runpy.run_path in the same interpreter." + " Script must be provided as an abs path (e.g. /abs/path/script.py)." + " Takes precedence over --no-python.", + ) + parser.add_argument( + "--log-dir", + "--log_dir", + action=env, + type=str, + default=None, + help="Base directory to use for log files (e.g. /var/log/torch/elastic). The same " + "directory is reused for multiple runs (a unique job-level sub-directory is created with " + "rdzv_id as the prefix).", + ) + parser.add_argument( + "-r", + "--redirects", + action=env, + type=str, + default="0", + help="Redirect std streams into a log file in the log directory (e.g. [-r 3] redirects " + "both stdout+stderr for all workers, [-r 0:1,1:2] redirects stdout for local rank 0 and " + "stderr for local rank 1).", + ) + parser.add_argument( + "-t", + "--tee", + action=env, + type=str, + default="0", + help="Tee std streams into a log file and also to console (see --redirects for format).", + ) + + parser.add_argument( + "--log-line-prefix-template", + "--log_line_prefix_template", + action=env, + type=str, + default="", + help="Template for the prefix prepended to each console log line of a tee'd stream " + "(e.g. [--log-line-prefix-template '${hostname}:${rank}: '] renders as 'r12i0n8:3: '). " + "Available macros: ${role_name}, ${local_rank}, ${rank}, ${hostname}. Defaults to " + "'[${role_name}${local_rank}]:'.", + ) + + parser.add_argument( + "--local-ranks-filter", + "--local_ranks_filter", + action=env, + type=str, + default="", + help="Only show logs from specified ranks in console (e.g. [--local_ranks_filter=0,1,2] will " + "only show logs from rank 0, 1 and 2). This will only apply to stdout and stderr, not to" + "log files saved via --redirect or --tee", + ) + + parser.add_argument( + "--duplicate-stdout-filters", + "--duplicate_stdout_filters", + action=env, + type=comma_separated_list, + default=[], + help="Duplicates logs streamed to stdout to another specified file with a list of filters (e.g. " + "[--duplicate_stdout_filters 'apple,orange'] will duplicate log lines matching 'apple' " + "OR 'orange'. An empty filters list won't duplicate any lines. Use double comma to escape a comma) ", + ) + + parser.add_argument( + "--duplicate-stderr-filters", + "--duplicate_stderr_filters", + action=env, + type=comma_separated_list, + default=[], + help="Duplicates logs streamed to stderr to another specified file with a list of filters (e.g. " + "[--duplicate_stdout_filters 'apple,orange'] will duplicate log lines matching 'apple' " + "OR 'orange'. An empty filters list won't duplicate any lines. Use double comma to escape a comma) ", + ) + + # + # Backwards compatible parameters with caffe2.distributed.launch. + # + + parser.add_argument( + "--node-rank", + "--node_rank", + type=int, + action=env, + default=0, + help="Rank of the node for multi-node distributed training. It is only used for static " + "rendezvous (i.e., when ``--rdzv-backend=static``).", + ) + parser.add_argument( + "--master-addr", + "--master_addr", + default="127.0.0.1", + type=str, + action=env, + help="Address of the master node (rank 0) that only used for static rendezvous. It should " + "be either the IP address or the hostname of rank 0. For single node multi-proc training " + "the --master-addr can simply be 127.0.0.1; IPv6 should have the pattern " + "`[0:0:0:0:0:0:0:1]`.", + ) + parser.add_argument( + "--master-port", + "--master_port", + default=None, + type=int, + action=env, + help="Port on the master node (rank 0) to be used for communication during distributed " + "training. It is only used for static rendezvous. Defaults to 29500.", + ) + parser.add_argument( + "--local-addr", + "--local_addr", + default=None, + type=str, + action=env, + help="Address of the local node. If specified, will use the given address for connection. " + "Else, will look up the local node address instead. Else, it will be default to local " + "machine's FQDN.", + ) + + parser.add_argument( + "--logs-specs", + "--logs_specs", + default=None, + type=str, + help="torchrun.logs_specs group entrypoint name, value must be type of LogsSpecs. " + "Can be used to override custom logging behavior.", + ) + + parser.add_argument( + "--numa-binding", + "--numa_binding", + type=str, + choices=[mode.value for mode in _AffinityMode], + default=None, + help="Bind worker processes to CPUs near their assigned GPUs for better performance. " + "See torch/numa/binding.py for available modes and details.", + ) + + parser.add_argument( + "--signals-to-handle", + "--signals_to_handle", + action=env, + type=str, + default="SIGTERM,SIGINT,SIGHUP,SIGQUIT", + help="Comma-separated list of signals to handle and forward to subprocesses. " + "Default: SIGTERM,SIGINT,SIGHUP,SIGQUIT. " + "Common additional signals: SIGUSR1,SIGUSR2 (used in SLURM environments).", + ) + + parser.add_argument( + "--shutdown-timeout", + "--shutdown_timeout", + action=env, + type=int, + default=None, + help="Time in seconds to wait for graceful shutdown of worker processes before " + "sending SIGKILL. If not specified, uses TORCH_ELASTIC_SHUTDOWN_TIMEOUT environment " + "variable or defaults to 30 seconds.", + ) + + parser.add_argument( + "--virtual-local-rank", + "--virtual_local_rank", + action=check_env, + help="Enable virtual local rank mode for workers. When enabled, LOCAL_RANK is set to 0 " + "for all workers and CUDA_VISIBLE_DEVICES is adjusted so each worker accesses its " + "assigned GPU at device index 0.", + ) + + # + # Shell completion. + # + + try: + import shtab + except ImportError: + # shtab is optional; fall back to the shells it supports today so the + # flag still validates its argument when shtab is not installed + choices = ["bash", "zsh", "tcsh"] + else: + choices = shtab.SUPPORTED_SHELLS + + parser.add_argument( + "--print-completion", + "--print_completion", + action=_PrintCompletionAction, + choices=choices, + help="Print a shell completion script for torchrun to stdout and exit " + "(e.g. [--print-completion zsh > ~/.zsh/completions/_torchrun]). " + "Requires the optional 'shtab' package.", + ) + + # + # Positional arguments. + # + + parser.add_argument( + "training_script", + type=str, + help="Full path to the (single GPU) training program/script to be launched in parallel, " + "followed by all the arguments for the training script.", + ) + + # Rest from the training program. + parser.add_argument("training_script_args", nargs=REMAINDER) + + return parser + + +def parse_args(args): + parser = get_args_parser() + return parser.parse_args(args) + + +def parse_min_max_nnodes(nnodes: str): + arr = nnodes.split(":") + + if len(arr) == 1: + min_nodes = max_nodes = int(arr[0]) + elif len(arr) == 2: + min_nodes = int(arr[0]) + max_nodes = int(arr[1]) + else: + raise RuntimeError(f'nnodes={nnodes} is not in "MIN:MAX" format') # noqa: E231 + + return min_nodes, max_nodes + + +def determine_local_world_size(nproc_per_node: str): + try: + logger.info("Using nproc_per_node=%s.", nproc_per_node) + return int(nproc_per_node) + except ValueError as e: + if nproc_per_node == "cpu": + num_proc = torch._utils.cpu_count() + device_type = "cpu" + elif nproc_per_node == "gpu": + if not torch.cuda.is_available(): + raise ValueError("Cuda is not available.") from e + device_type = "gpu" + num_proc = torch.cuda.device_count() + elif nproc_per_node == "xpu": + if not torch.xpu.is_available(): + raise ValueError("Xpu is not available.") from e + device_type = "xpu" + num_proc = torch.xpu.device_count() + elif nproc_per_node == torch._C._get_privateuse1_backend_name(): + if not _get_custom_mod_func("is_available")(): + raise ValueError(f"{nproc_per_node} is not available.") from e + device_type = nproc_per_node + num_proc = _get_custom_mod_func("device_count")() + elif nproc_per_node == "auto": + if torch.accelerator.is_available(): + num_proc = torch.accelerator.device_count() + device_type = torch.accelerator.current_accelerator().type # type: ignore[union-attr] + else: + num_proc = torch._utils.cpu_count() + device_type = "cpu" + else: + raise ValueError( + f"Unsupported nproc_per_node value: {nproc_per_node}" + ) from e + + logger.info( + "Using nproc_per_node=%s, setting nproc_per_node to %s since the instance has %s %s", + nproc_per_node, + num_proc, + num_proc, + device_type, + ) + return num_proc + + +def get_rdzv_endpoint(args): + if args.rdzv_backend == "static" and not args.rdzv_endpoint: + return f"{args.master_addr}:{args.master_port}" # noqa: E231 + return args.rdzv_endpoint + + +def get_use_env(args) -> bool: + """ + Retrieve ``use_env`` from the args. + + ``use_env`` is a legacy argument, if ``use_env`` is False, the + ``--node-rank`` argument will be transferred to all worker processes. + ``use_env`` is only used by the ``torch.distributed.launch`` and will + be deprecated in future releases. + """ + if not hasattr(args, "use_env"): + return True + return args.use_env + + +def _get_logs_specs_class(logs_specs_name: str | None) -> type[LogsSpecs]: + """ + Attempts to load `torchrun.logs_spec` entrypoint with key of `logs_specs_name` param. + Provides plugin mechanism to provide custom implementation of LogsSpecs. + + Returns `DefaultLogsSpecs` when logs_spec_name is None. + Raises ValueError when entrypoint for `logs_spec_name` can't be found in entrypoints. + """ + logs_specs_cls = None + if logs_specs_name is not None: + eps = metadata.entry_points() + group = eps.select(group="torchrun.logs_specs") + if group.select(name=logs_specs_name): + # pyrefly: ignore [bad-index] + logs_specs_cls = group[logs_specs_name].load() + + if logs_specs_cls is None: + raise ValueError( + f"Could not find entrypoint under 'torchrun.logs_specs[{logs_specs_name}]' key" + ) + + logger.info( + "Using logs_spec '%s' mapped to %s", logs_specs_name, logs_specs_cls + ) + else: + logs_specs_cls = DefaultLogsSpecs + + return logs_specs_cls + + +def config_from_args(args) -> tuple[LaunchConfig, Callable | str, list[str]]: + # If ``args`` not passed, defaults to ``sys.argv[:1]`` + min_nodes, max_nodes = parse_min_max_nnodes(args.nnodes) + if not (0 < min_nodes <= max_nodes): + raise AssertionError( + f"min_nodes must be > 0 and <= max_nodes, got min_nodes={min_nodes}, max_nodes={max_nodes}" + ) + if args.max_restarts < 0: + raise AssertionError("max_restarts must be >= 0") + + if ( + hasattr(args, "master_addr") + and args.rdzv_backend != "static" + and not args.rdzv_endpoint + ): + logger.warning( + "master_addr is only used for static rdzv_backend and when rdzv_endpoint " + "is not specified." + ) + + if ( + hasattr(args, "node_rank") + and args.node_rank != 0 + and args.rdzv_backend != "static" + ): + logger.warning( + "node_rank is only used for static rdzv_backend. It will be ignored " + "for rdzv_backend=%s.", + args.rdzv_backend, + ) + + nproc_per_node = determine_local_world_size(args.nproc_per_node) + if "OMP_NUM_THREADS" not in os.environ and nproc_per_node > 1: + omp_num_threads = 1 + logger.warning( + "\n*****************************************\n" + "Setting OMP_NUM_THREADS environment variable for each process to be " + "%s in default, to avoid your system being overloaded, " + "please further tune the variable for optimal performance in " + "your application as needed. \n" + "*****************************************", + omp_num_threads, + ) + # This env variable will be passed down to the subprocesses + os.environ["OMP_NUM_THREADS"] = str(omp_num_threads) + + log_line_prefix_template = args.log_line_prefix_template or os.getenv( + "TORCHELASTIC_LOG_LINE_PREFIX_TEMPLATE" + ) + + rdzv_configs = _parse_rendezvous_config(args.rdzv_conf) + + if args.rdzv_backend == "static": + rdzv_configs["rank"] = args.node_rank + + rdzv_endpoint = get_rdzv_endpoint(args) + + ranks: set[int] | None = None + if args.local_ranks_filter: + try: + ranks = set(map(int, args.local_ranks_filter.split(","))) + if not ranks: + raise AssertionError("ranks set cannot be empty") + except Exception as e: + raise ValueError( + "--local_ranks_filter must be a comma-separated list of integers e.g. --local_ranks_filter=0,1,2" + ) from e + + logs_specs_cls: type[LogsSpecs] = _get_logs_specs_class(args.logs_specs) + + logs_specs = logs_specs_cls( + log_dir=args.log_dir, + redirects=Std.from_str(args.redirects), + tee=Std.from_str(args.tee), + local_ranks_filter=ranks, + ) + numa_options = ( + None + if args.numa_binding is None + else _NumaOptions(affinity_mode=_AffinityMode(args.numa_binding)) + ) + + config = LaunchConfig( + min_nodes=min_nodes, + max_nodes=max_nodes, + nproc_per_node=nproc_per_node, + run_id=args.rdzv_id, + role=args.role, + rdzv_endpoint=rdzv_endpoint, + rdzv_backend=args.rdzv_backend, + rdzv_configs=rdzv_configs, + max_restarts=args.max_restarts, + monitor_interval=args.monitor_interval, + start_method=args.start_method, + log_line_prefix_template=log_line_prefix_template, + local_addr=args.local_addr, + logs_specs=logs_specs, + event_log_handler=args.event_log_handler, + numa_options=numa_options, + signals_to_handle=args.signals_to_handle, + duplicate_stdout_filters=args.duplicate_stdout_filters, + duplicate_stderr_filters=args.duplicate_stderr_filters, + virtual_local_rank=args.virtual_local_rank, + shutdown_timeout=args.shutdown_timeout, + ) + + with_python = not args.no_python + cmd: Callable | str + cmd_args = [] + use_env = get_use_env(args) + if args.run_path: + cmd = run_script_path + cmd_args.append(args.training_script) + else: + if with_python: + cmd = os.getenv("PYTHON_EXEC", sys.executable) + cmd_args.append("-u") + if args.module: + cmd_args.append("-m") + cmd_args.append(args.training_script) + else: + if args.module: + raise ValueError( + "Don't use both the '--no-python' flag" + " and the '--module' flag at the same time." + ) + cmd = args.training_script + if not use_env: + cmd_args.append(f"--local-rank={macros.local_rank}") + cmd_args.extend(args.training_script_args) + + return config, cmd, cmd_args + + +def run_script_path(training_script: str, *training_script_args: str): + """ + Run the provided `training_script` from within this interpreter. + + Usage: `script_as_function("/abs/path/to/script.py", "--arg1", "val1")` + """ + import runpy + import sys + + sys.argv = [training_script] + [*training_script_args] + runpy.run_path(sys.argv[0], run_name="__main__") + + +def run(args): + torch.multiprocessing._set_thread_name("pt_elastic") + + if args.standalone: + args.rdzv_backend = "c10d" + args.rdzv_endpoint = "localhost:0" + args.rdzv_id = str(uuid.uuid4()) + logger.info( + "\n**************************************\n" + "Rendezvous info:\n" + "--rdzv-backend=%s " + "--rdzv-endpoint=%s " + "--rdzv-id=%s\n" + "**************************************\n", + args.rdzv_backend, + args.rdzv_endpoint, + args.rdzv_id, + ) + elif ( + args.rdzv_backend == "static" + and not args.rdzv_endpoint + and args.master_port is None + ): + _, max_nodes = parse_min_max_nnodes(args.nnodes) + if max_nodes == 1: + args.rdzv_backend = "c10d" + args.rdzv_endpoint = "localhost:0" + args.rdzv_id = str(uuid.uuid4()) + + # master_port is only used for the static rendezvous backend, not c10d + if args.master_port is None: + args.master_port = 29500 + + config, cmd, cmd_args = config_from_args(args) + elastic_launch( + config=config, + entrypoint=cmd, + )(*cmd_args) + + +@record +def main(args=None): + args = parse_args(args) + run(args) + + +if __name__ == "__main__": + main() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..c69c336304f96f131b81fab72c782f382e39ee9a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__init__.py @@ -0,0 +1,102 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates + +import torch +import torch.distributed.tensor._ops # force import all built-in dtensor ops +from torch.distributed.device_mesh import DeviceMesh, init_device_mesh +from torch.distributed.tensor._api import ( + distribute_module, + distribute_tensor, + DTensor, + empty, + full, + linspace, + logspace, + ones, + rand, + randn, + zeros, +) +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) +from torch.optim.optimizer import ( + _foreach_supported_types as _optim_foreach_supported_types, +) +from torch.utils._foreach_utils import ( + _foreach_supported_types as _util_foreach_supported_types, +) + + +# All public APIs from dtensor package +__all__ = [ + "DTensor", + "distribute_tensor", + "distribute_module", + "Shard", + "Replicate", + "Partial", + "Placement", + "ones", + "empty", + "full", + "linspace", + "logspace", + "rand", + "randn", + "zeros", +] + +# For weights_only torch.load +from ._dtensor_spec import ( + DTensorSpec as _DTensorSpec, + ShardOrderEntry as _ShardOrderEntry, + TensorMeta as _TensorMeta, +) + + +torch.serialization.add_safe_globals( + [ + DeviceMesh, + _DTensorSpec, + _TensorMeta, + _ShardOrderEntry, + DTensor, + Partial, + Replicate, + Shard, + _StridedShard, + ] +) + + +# Append DTensor to the list of supported types for foreach implementation for optimizer +# and clip_grad_norm_ so that we will try to use foreach over the for-loop implementation on CUDA. +if DTensor not in _optim_foreach_supported_types: + _optim_foreach_supported_types.append(DTensor) + +if DTensor not in _util_foreach_supported_types: + _util_foreach_supported_types.append(DTensor) # type: ignore[arg-type] + + +# Set namespace for exposed private names +DTensor.__module__ = "torch.distributed.tensor" +distribute_tensor.__module__ = "torch.distributed.tensor" +distribute_module.__module__ = "torch.distributed.tensor" +ones.__module__ = "torch.distributed.tensor" +empty.__module__ = "torch.distributed.tensor" +full.__module__ = "torch.distributed.tensor" +linspace.__module__ = "torch.distributed.tensor" +logspace.__module__ = "torch.distributed.tensor" +rand.__module__ = "torch.distributed.tensor" +randn.__module__ = "torch.distributed.tensor" +zeros.__module__ = "torch.distributed.tensor" + +# Register DTensor dispatch for higher order operators +from torch._higher_order_ops.print import _register_dtensor_impl + + +_register_dtensor_impl() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..282c4bb1f82dfd14e316b67994d960b0793d57ea Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f21d0a28e13a0ecde7ba5cad233412abd2fe97c5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_collective_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_collective_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..89bec2c0aa8c13eb39c9ca4b6a43cf32aaaa2bcf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_collective_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_decompositions.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_decompositions.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1f6f84d085026e952a37d71f4b6b59c658b26465 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_decompositions.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dispatch.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dispatch.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fa0766f02183b94cec6d75fc38689cbda8894400 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dispatch.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dtensor_spec.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dtensor_spec.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9240cac4f8dc993678ec69a107f392e3d818d385 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_dtensor_spec.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_nonlinear_redux.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_nonlinear_redux.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2bf3a05a23758811bee06b9b3b3bbf527fe247ad Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_nonlinear_redux.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_op_schema.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_op_schema.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4f762094debfac3dc54c1ac5e9a2464c3aff7a44 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_op_schema.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_random.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_random.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..167bc394d2d1c00d3fa97a8b36da9bb668b095e2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_random.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_redistribute.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_redistribute.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6b1ab022ed763eb7302178656bba94515f1f56fd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_redistribute.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_sharding_prop.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_sharding_prop.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..aab597c564e41d22471e82322c902190a0200947 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_sharding_prop.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_shards_wrapper.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_shards_wrapper.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2d66cba9bf4cd429671d67eb652d01040a73d660 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_shards_wrapper.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_tp_conv.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_tp_conv.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b22d025f6d6f6f442fe15c0f90230dadd371e556 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_tp_conv.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5f7afc54fcc166b402a83769eafea53cf6918eae Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/device_mesh.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/device_mesh.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..497d2b6cd2360c632355b21af3ed85995368f185 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/device_mesh.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/placement_types.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/placement_types.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2e36776049647da13e7266621b75aebb6f735ad6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/__pycache__/placement_types.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_api.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_api.py new file mode 100644 index 0000000000000000000000000000000000000000..bb4c2c5747ea3d6f73ac2e92647dd53876528d42 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_api.py @@ -0,0 +1,1934 @@ +# mypy: allow-untyped-decorators +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import copy +import hashlib +import inspect +import warnings +from collections.abc import Callable, Sequence +from typing import Any +from typing_extensions import deprecated + +import torch +import torch.distributed.tensor._dispatch as op_dispatch +import torch.distributed.tensor._random as random +import torch.nn as nn +from torch._export.wrappers import mark_subclass_constructor_exportable_experimental +from torch._logging import LazyString +from torch.distributed._local_tensor import maybe_run_for_local_tensor +from torch.distributed.device_mesh import ( + _mesh_resources, + _register_distributed_opaque_types, + DeviceMesh, +) +from torch.distributed.tensor._collective_utils import check_tensor_meta, mesh_broadcast +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._redistribute import ( + Redistribute, + redistribute_local_tensor, +) +from torch.distributed.tensor._utils import ( + assert_no_mixed_partial_types, + compute_global_tensor_info, + compute_local_shape_and_global_offset, + ExplicitRedistributionContext, + normalize_to_torch_size, +) +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) + + +__all__ = [ + "DTensor", + "distribute_tensor", + "distribute_module", + "ones", + "empty", + "full", + "linspace", + "logspace", + "rand", + "randn", + "zeros", +] + +aten = torch.ops.aten + + +def _normalize_placements_for_grad( + placements: tuple[Placement, ...], +) -> tuple[Placement, ...]: + """ + Normalize gradient placements by converting Partial to Replicate. + + See the gradient placement guarantees documented in DTensor.from_local's docstring + for why Partial forward placements map to Replicate gradient placements. We do this + for both from_local and to_local backward. + """ + normalized: list[Placement] = [] + for p in placements: + if p.is_partial(): + normalized.append(Replicate()) + else: + normalized.append(p) + return tuple(normalized) + + +# NOTE [Autograd interaction between torch.Tensor] +# +# The autograd functions defined below are being used by the public +# facing APIs (i.e. from_local, to_local) to ensure DTensor to work +# together with torch.Tensor within the autograd engine. This +# allows DTensor to only exist on part of the module hierarchy. +# +# As an example, we have a module that consists of submodules +# A, B, and C, the execution flow would be like: +# input(torch.Tensor) -> Module A -> Module B -> Module C -> output (torch.Tensor) +# +# Suppose I only want to make Module B be a sharded module with +# DTensor params, the following forward/backward should work: +# +# input(torch.Tensor) -> Module A +# -> DTensor input (from_local) -> Sharded Module B -> DTensor output +# -> torch.Tensor output (to_local) -> Module C +# +# So from_local/to_local must be Autograd functions. +# +class _ToTorchTensor(torch.autograd.Function): + @staticmethod + def forward( # type: ignore[override] + ctx, + input: "DTensor", + grad_placements: Sequence[Placement] | None, + ): + ctx.dtensor_spec = input._spec + ctx.grad_placements = grad_placements + ctx.set_materialize_grads(False) + local_tensor = input._local_tensor + ctx.local_tensor_stride = local_tensor.stride() + + # We need to return a fresh Tensor object there as autograd metadata + # will be inplaced into it. So we don't want to pollute the Tensor + # object stored in the _local_tensor of this DTensor. + return local_tensor.view_as(local_tensor) + + @staticmethod + def backward(ctx, grad_output: torch.Tensor | None): # type: ignore[override] + if grad_output is None: + return None, None + + dtensor_spec = ctx.dtensor_spec + mesh = dtensor_spec.mesh + grad_placements = ctx.grad_placements + dtensor_meta = dtensor_spec.tensor_meta + + # user should provide grad_placements as there's no guarantee on input gradient placement + # if grad_placement is None, we provide default placement + if grad_placements is None: + # See DTensor.from_local docstring for gradient placement guarantees + grad_placements = _normalize_placements_for_grad(dtensor_spec.placements) + + from torch._prims_common import check_contiguous_sizes_strides + from torch.fx.experimental.symbolic_shapes import guard_or_false + + def strides_provably_match(stride, other): + return len(stride) == len(other) and all( + guard_or_false(a == b) for a, b in zip(stride, other) + ) + + def strides_provably_differ(stride, other): + return len(stride) != len(other) or any( + guard_or_false(a != b) for a, b in zip(stride, other) + ) + + def strides_are_contiguous_for_shape(shape, stride, other): + if not (len(shape) == len(stride) == len(other)): + return False + return check_contiguous_sizes_strides( + shape, stride, false_if_dde=True + ) and check_contiguous_sizes_strides(shape, other, false_if_dde=True) + + def strides_may_match(shape, stride, other): + return ( + strides_provably_match(stride, other) + or strides_are_contiguous_for_shape(shape, stride, other) + or not strides_provably_differ(stride, other) + ) + + def check_stride_matches(shape, stride, other): + if strides_provably_match( + stride, other + ) or strides_are_contiguous_for_shape(shape, stride, other): + return + for a, b in zip(stride, other): + torch._check( + a == b, + lambda: ( + f"Expected matching DTensor local strides, got {stride} and {other}" + ), + ) + + # Same-placement to_local() backward preserves the forward DTensor + # layout. With symbolic shapes, the local grad stride can be equivalent + # to the saved stride while carrying different symbol provenance. + if ( + grad_placements == dtensor_spec.placements + and strides_may_match( + grad_output.shape, grad_output.stride(), ctx.local_tensor_stride + ) + and strides_may_match( + dtensor_meta.shape, ctx.local_tensor_stride, dtensor_meta.stride + ) + ): + check_stride_matches( + grad_output.shape, grad_output.stride(), ctx.local_tensor_stride + ) + check_stride_matches( + dtensor_meta.shape, ctx.local_tensor_stride, dtensor_meta.stride + ) + # Avoid actual sharing of specs in case they're modified during (e.g.) + # sharding propagation. + grad_spec = copy.copy(dtensor_spec) + else: + _, tensor_stride = compute_global_tensor_info( + grad_output, mesh, dtensor_spec.placements + ) + tensor_stride = tuple(tensor_stride) + grad_spec = DTensorSpec( + mesh, + grad_placements, + tensor_meta=TensorMeta( + shape=dtensor_meta.shape, + stride=tensor_stride, + dtype=dtensor_meta.dtype, + ), + ) + return ( + DTensor.from_local( + grad_output, + grad_spec.device_mesh, + grad_spec.placements, + shape=grad_spec.shape, + stride=grad_spec.stride, + ), + None, + ) + + +class _FromTorchTensor(torch.autograd.Function): + @staticmethod + def forward( # type: ignore[override] + ctx, # pyre-ignore[2]: Parameter must be annotated. + input: torch.Tensor, + device_mesh: DeviceMesh, + placements: tuple[Placement, ...], + run_check: bool, + shape: torch.Size | None = None, + stride: tuple[int, ...] | None = None, + grad_placements: tuple[Placement, ...] | None = None, + ) -> "DTensor": + ctx.forward_input_placements = placements + ctx.forward_input_device_mesh = device_mesh + ctx.grad_placements = grad_placements + ctx.set_materialize_grads(False) + + if shape and stride: + tensor_shape, tensor_stride = shape, stride + elif not shape and not stride: + # if it's not by default run_check, we assume user is certain that each + # rank has the same tensor shape, and we just use that to calculate the + # global shape + global_shape, global_stride = compute_global_tensor_info( + input, device_mesh, placements + ) + tensor_shape, tensor_stride = torch.Size(global_shape), tuple(global_stride) + else: + raise RuntimeError( + f"Found shape:{shape}, stride:{stride}.", + "Please pass both shape and stride at the same time.", + ) + + if not device_mesh._is_current_rank_part_of_mesh(): + # if the global rank is not participating in the device mesh, we + # simply set the local tensor to an empty tensor + input = input.new_empty(0, requires_grad=input.requires_grad) + elif run_check: + # TODO: support uneven sharding when global shape/stride not passed, by + # building the global TensorMeta during check_tensor_meta + check_shape_stride = not shape and not stride + check_tensor_meta(input, check_shape_stride=check_shape_stride) + # TODO: See if we need to make this run_check logic + # have a corresponding backward. + for idx, placement in enumerate(placements): + if placement.is_replicate(): + # broadcast rank 0 tensor to all ranks + # only broadcast if run_check is True + input = input.contiguous() + mesh_broadcast(input, device_mesh, mesh_dim=idx) + + dist_spec = DTensorSpec( + device_mesh, + placements, + tensor_meta=TensorMeta( + tensor_shape, + tensor_stride, + input.dtype, + ), + ) + + # We want a fresh Tensor object that shares memory with the input tensor + # pyrefly: ignore [bad-argument-type] + dist_tensor = DTensor( + # pyrefly: ignore [bad-argument-count] + input.view_as(input), + dist_spec, + # requires_grad of the dist tensor depends on if input + # requires_grad or not + # pyrefly: ignore [unexpected-keyword] + requires_grad=input.requires_grad, + ) + return dist_tensor + + @staticmethod + def backward(ctx, grad_output: "DTensor | None"): # type: ignore[override] + forward_input_placements = ctx.forward_input_placements + forward_input_device_mesh = ctx.forward_input_device_mesh + grad_placements = ctx.grad_placements + + if grad_output is None: + return None, None, None, None, None, None, None + + # reshard to the placement when creating DistributedTensor + # so that the gradient layout matches, and we could return + # local gradients directly + # if forward placement is partial, we always redistribute grad_input to Replicate: + # Partial(fwd) - Replicate(grad_output) - Replicate(grad_input): no redistribution needed + # Partial(fwd) - Partial(grad_output) - Replicate(grad_input): redistribute Partial to Replicate + + # The second case is BC breaking: + # was: Partial(fwd) - Partial(grad_output) - Partial(grad_input) + # now: Partial(fwd) - Partial(grad_output) - Replicate(grad_input) + # See DTensor.from_local docstring for gradient placement guarantees + + # user should provide grad_placements as there's no guarantee on input gradient placement + # if grad_placements is None, we provide default placement + if grad_placements is not None: + normalized_placements = grad_placements + else: + normalized_placements = _normalize_placements_for_grad( + forward_input_placements + ) + + if grad_output.placements != normalized_placements: + current_spec: DTensorSpec = grad_output._spec + target_spec = DTensorSpec( + forward_input_device_mesh, + normalized_placements, + tensor_meta=grad_output._spec.tensor_meta, + ) + ExplicitRedistributionContext.observe_redistribution( + current_spec, + target_spec, + LazyString( + lambda: "Implicit redistribution occurred in DTensor.from_local backward" + ), + ) + local_tensor = grad_output._local_tensor + output = redistribute_local_tensor( + local_tensor, + current_spec, + target_spec, + ) + # TODO: return the redistributed local tensor directly without + # differentiable backward. see if this make sense for all cases. + return output, None, None, None, None, None, None + + # TODO: backward is also differentiable now, add a test + # to test higher level gradients. + return grad_output.to_local(), None, None, None, None, None, None + + +class DTensor(torch.Tensor): + """ + ``DTensor`` (Distributed Tensor) is a subclass of ``torch.Tensor`` that provides single-device like + abstraction to program with multi-device ``torch.Tensor``. It describes the distributed tensor sharding + layout (DTensor Layout) through the :class:`DeviceMesh` and following types of :class:`Placement`: + + * :class:`Shard`: Tensor sharded on the tensor dimension ``dim`` on the devices of the ``DeviceMesh`` dimension + * :class:`Replicate`: Tensor replicated on the devices of the ``DeviceMesh`` dimension + * :class:`Partial`: Tensor is pending reduction on the devices of the ``DeviceMesh`` dimension + + When calling PyTorch operators, ``DTensor`` overrides the PyTorch operators to perform sharded computation and issue + communications whenever necessary. Along with the operator computation, ``DTensor`` will transform or propagate the + placements (DTensor Layout) properly (based on the operator semantic itself) and generate new ``DTensor`` outputs. + + To ensure numerical correctness of the ``DTensor`` sharded computation when calling PyTorch operators, ``DTensor`` + requires every Tensor argument of the operator be DTensor. + + .. note:: Directly using the Tensor subclass constructor here is not the recommended way to create a ``DTensor`` + (i.e. it does not handle autograd correctly hence is not the public API). Please refer to the `create_dtensor`_ + section to see how to create a ``DTensor``. + """ + + _local_tensor: torch.Tensor + _spec: DTensorSpec + __slots__ = ["_local_tensor", "_spec"] + + # _op_dispatcher instance as a class attribute to handle runtime dispatching logic + _op_dispatcher: op_dispatch.OpDispatcher = op_dispatch.OpDispatcher() + + # This implementation is just to convince mypy _spec and _local_tensor are + # initialized; it is immediately overridden below. + def __new__( + cls, + local_tensor: torch.Tensor, + spec: DTensorSpec, + *, + requires_grad: bool, + ) -> "DTensor": + r = torch.Tensor._dtensor__new__( + cls, local_tensor, spec, requires_grad=requires_grad + ) + r._spec = spec + r._local_tensor = local_tensor + return r + + __new__ = torch.Tensor._dtensor__new__ # type: ignore[assignment] # noqa: F811 + + @torch._disable_dynamo + @mark_subclass_constructor_exportable_experimental + def __init__(self, *args, **kwargs): + """ + Construct a DTensor from a local tensor, device mesh, and placement and + other tensor properties (i.e. shape, requires_grad, strides, etc). + .. note:: This is not a public API and it's only supposed to be used by the + operator implementations and internals. If you want to construct a + DTensor from a local tensor, consider using ``DTensor.from_local``, if + you want to construct a DTensor from a "global" tensor (where you + already have tensor initialized and want to shard this tensor), + consider using ``distribute_tensor``. + """ + super().__init__() + + # pyre-fixme[14]: `__repr__` overrides method defined in `DTensor` inconsistently. + # pyre-fixme[3]: Return type must be annotated. + def __repr__(self): # type: ignore[override] + # TODO: consider all_gather the local tensors for better debugging + return f"DTensor(local_tensor={self._local_tensor}, device_mesh={self._spec.mesh}, placements={self._spec.placements})" + + def __tensor_flatten__(self): + """ + protocol to inform how to flatten a DTensor to local tensor + for PT2 tracing + """ + return ["_local_tensor", "device_mesh"], ( + self._spec.placements, + self._spec.tensor_meta, + self._spec.shard_order, + self.requires_grad, + ) + + @staticmethod + def __tensor_unflatten__(inner_tensors, flatten_spec, outer_size, outer_stride): + if flatten_spec is None: + raise AssertionError( + "Expecting spec to be not None from `__tensor_flatten__` return value!" + ) + local_tensor = inner_tensors["_local_tensor"] + mesh = inner_tensors["device_mesh"] + placements, old_tensor_meta, shard_order, requires_grad = flatten_spec + unflatten_tensor_meta = TensorMeta( + shape=outer_size, + stride=outer_stride, + dtype=old_tensor_meta.dtype, + ) + unflatten_spec = DTensorSpec( + mesh, + placements, + tensor_meta=unflatten_tensor_meta, + shard_order=shard_order, + ) + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + local_tensor, + unflatten_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=requires_grad, + ) + + def _stable_hash_for_caching(self) -> str: + """ + Return a stable hash for AOT autograd caching. + [See note: Tensor subclass stable hashing for AOT autograd cache] + """ + # Include local tensor device so different ranks produce distinct keys. + # AOTAutogradCachePickler has no device_id_agnostic normalization (unlike + # FxGraphCachePickler), so ranks sharing a disk cache need separate entries. + device = str(self._local_tensor.device) + cache_data = self._spec._stable_hash() + str(self.requires_grad) + device + return hashlib.blake2b(cache_data.encode(), digest_size=16).hexdigest() + + def __coerce_tangent_metadata__(self): + if not any(isinstance(p, Partial) for p in self.placements): + return self + placements = [ + Replicate() if isinstance(p, Partial) else p for p in self.placements + ] + return self.redistribute(device_mesh=self.device_mesh, placements=placements) + + def __coerce_same_metadata_as_tangent__(self, flatten_spec, expected_type=None): + if expected_type is not None: + return None + + (placements, _, _, _) = flatten_spec + return self.redistribute( + device_mesh=self.device_mesh, + placements=placements, + ) + + @classmethod + def __torch_dispatch__(cls, func, types, args=(), kwargs=None): # type: ignore[override] + # Base DTensor is normally dispatched via a C++ fast path (see #167051) + # and never reaches here. This implementation exists so that DTensor + # subclasses can delegate back via super().__torch_dispatch__(). + # It unwraps subclass instances to base DTensor and re-calls the op, + # which re-enters dispatch and hits the C++ fast path. + def unwrap(t): + if isinstance(t, DTensor) and type(t) is not DTensor: + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + t._local_tensor, + t._spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=t.requires_grad, + ) + return t + + args = torch.utils._pytree.tree_map(unwrap, args) + kwargs = torch.utils._pytree.tree_map(unwrap, kwargs or {}) + return func(*args, **kwargs) + + @staticmethod + def from_local( + local_tensor: torch.Tensor, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, + *, + run_check: bool = False, + shape: torch.Size | None = None, + stride: tuple[int, ...] | None = None, + grad_placements: Sequence[Placement] | None = None, + ) -> "DTensor": + """ + Create a :class:`DTensor` from a local torch.Tensor on each rank + according to the ``device_mesh`` and ``placements`` specified. + + Args: + local_tensor (torch.Tensor): local torch.Tensor on each rank. + device_mesh (:class:`DeviceMesh`, optional): DeviceMesh to place the + tensor, if not specified, must be called under a DeviceMesh + context manager, default: None + placements (List[:class:`Placement`], optional): the placements that + describes how to place the local torch.Tensor on DeviceMesh, must + have the same number of elements as ``device_mesh.ndim``. + + Keyword args: + run_check (bool, optional): at a cost of extra communications, perform + sanity check across ranks to check each local tensor's meta information + to ensure correctness. If have :class:`Replicate` in ``placements``, the + data on first rank of the device mesh dimension will be broadcasted + to other ranks. default: False + shape (torch.Size, optional): A List of int which specifies the size of + DTensor which build on top of `local_tensor`. Note this needs to be + provided if the shape of ``local_tensor`` are different across the ranks. + If not provided, ``shape`` will be computed assuming the given distributed + tensor is evenly sharded across ranks. default: None + stride (tuple, optional): A List of int which specifies the stride of DTensor. + If not provided, ``stride`` will be computed assuming the given distributed + tensor is evenly sharded across ranks. default: None + grad_placements (List[:class:`Placement`], optional): specifies the expected + input gradient placements. The input gradient (a plain tensor) will be + redistributed to this placement before exiting DTensor. If not + specified, follows the default placement guarantees below. default: None + + Returns: + A :class:`DTensor` object + + Raises: + ValueError: If ``placements`` contains mixed :class:`Partial` reduce types + (e.g., both ``Partial("sum")`` and ``Partial("max")``). All Partial + placements must use the same reduce operation. + + .. note:: When ``run_check=False``, it is the user's responsibility to ensure the + local tensor passed in is correct across ranks (i.e. the tensor is sharded for + the ``Shard(dim)`` placement or replicated for the ``Replicate()`` placement). + If not, the behavior of the created DTensor is undefined. + + .. note:: ``from_local`` is differentiable, the `requires_grad` of the created + `DTensor` object will depend on if `local_tensor` requires_grad or not. + + .. note:: During backward, ``from_local`` provides the following gradient placement + guarantees. For each mesh dimension, the gradient placement maps as follows: + + +---------------------+--------------------+ + | Forward Placement | Gradient Placement | + +=====================+====================+ + | ``Shard`` | ``Shard`` | + +---------------------+--------------------+ + | ``Replicate`` | ``Replicate`` | + +---------------------+--------------------+ + | ``Partial`` | ``Replicate`` | + +---------------------+--------------------+ + + When the forward placement is :class:`Partial`, we always redistribute the gradient + to :class:`Replicate` instead of keeping it :class:`Partial`. This may not be the most + efficient option, but it avoids ambiguity and provides clearer gradient semantics to users. + """ + # `local_tensor` argument cannot be DTensor + if isinstance(local_tensor, DTensor): + raise RuntimeError( + f"the local_tensor argument only accepts torch.Tensor but got {type(local_tensor)} value." + ) + + # if same shape/dtype, no need to run_check, if not, must allgather + # the metadatas to check the size/dtype across ranks + # There should be no data communication unless there's replication + # strategy, where we broadcast the replication from the first rank + # in the mesh dimension + device_mesh = device_mesh or _mesh_resources.get_current_mesh() + device_type = device_mesh.device_type + + # convert the local tensor to desired device base on device mesh's device_type + if device_type != local_tensor.device.type and not local_tensor.is_meta: + local_tensor = local_tensor.to(device_type) + + # set default placements to replicated if not specified + if placements is None: + placements = [Replicate() for _ in range(device_mesh.ndim)] + else: + placements = list(placements) + for idx, placement in enumerate(placements): + # normalize shard dim to be positive + if isinstance(placement, Shard | _StridedShard): + if placement.dim < 0: + normalized_dim = placement.dim + local_tensor.ndim + if type(placement) is _StridedShard: + placements[idx] = _StridedShard( + normalized_dim, split_factor=placement.split_factor + ) + elif type(placement) is Shard: + placements[idx] = Shard(normalized_dim) + + # Validate that placements don't contain mixed Partial reduce types + assert_no_mixed_partial_types(placements) + + # `from_local` is differentiable, and the gradient of the dist tensor this function + # created should flow back the gradients to the local_tensor, so we call an autograd + # function to construct the dist tensor instead. + return _FromTorchTensor.apply( # pyre-ignore[16]: autograd func + local_tensor, + device_mesh, + tuple(placements), + run_check, + shape, + stride, + tuple(grad_placements) if grad_placements is not None else None, + ) + + def to_local( + self, *, grad_placements: Sequence[Placement] | None = None + ) -> torch.Tensor: + """ + Get the local tensor of this DTensor on its current rank. For sharding it returns + a local shard of the logical tensor view, for replication it returns the replica on + its current rank. + + Keyword args: + grad_placements (List[:class:`Placement`], optional): the placements describes + the future layout of any gradient layout of the Tensor returned from this + function. + `to_local` converts DTensor to local tensor and the returned local tensor + might not be used as the original DTensor layout later in the code. This + argument is the hint that user can give to autograd in case the gradient + layout of the returned tensor does not match the original DTensor layout. + If not specified, we will assume the gradient layout remains the same + as the original DTensor and use that for gradient computation. + + Returns: + A :class:`torch.Tensor` or ``AsyncCollectiveTensor`` object. it represents the + local tensor on its current rank. When an ``AsyncCollectiveTensor`` object is returned, + it means the local tensor is not ready yet (i.e. communication is not finished). In this + case, user needs to call ``wait`` to wait the local tensor to be ready. + + .. note:: ``to_local`` is differentiable, the ``requires_grad`` of the local tensor returned + will depend on if the `DTensor` requires_grad or not. + """ + if not torch.is_grad_enabled(): + result = self._local_tensor + else: + if grad_placements is not None and not isinstance(grad_placements, tuple): + grad_placements = tuple(grad_placements) + result = _ToTorchTensor.apply( + self, grad_placements + ) # pyre-ignore[16]: autograd func + + # Preserve nn.Parameter-ness: if self is an nn.Parameter (i.e. has the + # _is_param flag, which is how Parameter is represented for custom + # tensor subclasses like DTensor), make the returned local tensor also + # satisfy isinstance(result, nn.Parameter). See gh-166156. + # + # BC: this branch only runs when self has `_is_param` set, i.e. self + # is already a Parameter(DTensor). All raw-DTensor call sites are + # byte-for-byte unchanged, including the historical identity guarantee + # `to_local() is self._local_tensor` in the no-grad path. The only + # observable change is the targeted bug fix: Parameter(DTensor) inputs + # no longer return a plain Tensor. + # + # We probe `_is_param` via getattr rather than isinstance(self, + # nn.Parameter) for two reasons: (1) it avoids importing torch.nn at + # this layer, and (2) it matches the exact mechanism that + # nn.parameter._ParameterMeta.__instancecheck__ uses to recognize + # Parameter-ness on custom tensor subclasses. + if getattr(self, "_is_param", False) and not getattr( + result, "_is_param", False + ): + if result is self._local_tensor: + # Avoid mutating the internal storage of this DTensor by + # returning a fresh view to attach the flag to. + result = result.view_as(result) + result._is_param = True + + return result + + def redistribute( + self, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, + *, + async_op: bool = False, + forward_dtype: torch.dtype | None = None, + backward_dtype: torch.dtype | None = None, + ) -> "DTensor": + """ + ``redistribute`` performs necessary collective operations that redistribute the current + DTensor from its current placements to a new placements, or from its current DeviceMesh + to a new DeviceMesh. i.e. we can turn a Sharded DTensor to a Replicated DTensor by + specifying a Replicate placement for each dimension of the DeviceMesh. + + When redistributing from current to the new placements on one device mesh dimension, we + will perform the following operations including communication collective or local operation: + + 1. ``Shard(dim)`` -> ``Replicate()``: ``all_gather`` + 2. ``Shard(src_dim)`` -> ``Shard(dst_dim)``: ``all_to_all`` + 3. ``Replicate()`` -> ``Shard(dim)``: local chunking (i.e. ``torch.chunk``) + 4. ``Partial()`` -> ``Replicate()``: ``all_reduce`` + 5. ``Partial()`` -> ``Shard(dim)``: ``reduce_scatter`` + 6. ``Shard(dim)`` -> ``Partial("sum")``: local zero-filling. This + explicit-only conversion materializes a global-shape local tensor + on every rank and does not communicate. + + Of conversions to ``Partial``, this public API supports only + ``Shard(dim)`` -> ``Partial("sum")``. Each rank places its local shard + at the corresponding offset in a zero-filled logical-shape tensor, so + summing across ranks reconstructs the original logical tensor. + + ``redistribute`` would correctly figure out the necessary redistribute steps for DTensors + that are created either on 1-D or N-D DeviceMesh. + + Args: + device_mesh (:class:`DeviceMesh`, optional): DeviceMesh to place the + DTensor. If not specified, it would use the current DTensor's DeviceMesh. + default: None + placements (List[:class:`Placement`], optional): the new placements that + describes how to place the DTensor into the DeviceMesh, must + have the same number of elements as ``device_mesh.ndim``. + default: replicate on all mesh dimensions + + Keyword args: + async_op (bool, optional): whether to perform the DTensor redistribute operation + asynchronously or not. Default: False + forward_dtype (torch.dtype, optional): cast the local tensor to + ``forward_dtype`` before running the forward collective. + The result DTensor will be in ``forward_dtype``. If ``None``, + no conversion is performed. Default: None + backward_dtype (torch.dtype, optional): cast the gradient to + ``backward_dtype`` before running the backward collective. + After the collective the gradient is always cast back to the + input DTensor's dtype. If ``None``, the backward collective + runs at the input DTensor's dtype (not ``forward_dtype``). + Default: None + + Returns: + A :class:`DTensor` object + + .. note:: ``redistribute`` is twice-differentiable, which means user do not need to worry about + the backward formula of the redistribute operation, or its compatibility with autograd for + second-order gradients. Higher-order differentiation has not been tested (but may work). + + .. note:: ``redistribute`` currently only supports redistributing DTensor on the same DeviceMesh, + Please file an issue if you need to redistribute DTensor to different DeviceMesh. + """ + # NOTE: This redistribute API currently only supports out + # of place redistribution, i.e. it always create a new + # DTensor object and leave the original one unchanged. + + # if device_mesh is not specified, use the current device_mesh + device_mesh = device_mesh or self.device_mesh + # raise error if new placements not specified + if placements is None: + raise RuntimeError("placements is needed for redistribute!") + + placements = list(placements) + for i, placement in enumerate(placements): + if placement.is_partial() and self.placements[i] != placement: + source = self.placements[i] + if not ( + isinstance(source, Shard) + and type(placement) is Partial + and placement.reduce_op == "sum" + ): + raise RuntimeError( + f"Can not redistribute from {source} to {placement}, " + "only Shard to Partial(sum) redistribution is supported!" + ) + elif isinstance(placement, Shard) and placement.dim < 0: + # normalize shard dim to be positive + placements[i] = Shard(placement.dim + self.ndim) + elif isinstance(placement, _StridedShard) and placement.dim < 0: + placements[i] = _StridedShard( + placement.dim + self.ndim, split_factor=placement.split_factor + ) + placements = tuple(placements) + + input_dtype = self._local_tensor.dtype + forward_dtype = forward_dtype or input_dtype + # pyre-fixme[16]: `Redistribute` has no attribute `apply`. + return Redistribute.apply( + self, + device_mesh, + placements, + async_op, + { + "op_dtype": forward_dtype, + "out_dtype": forward_dtype, + "backward_options": { + # Absent backward_dtype means the backward collective runs + # at the input's storage dtype (matching pre-fix behavior). + "op_dtype": backward_dtype or input_dtype, + "out_dtype": input_dtype, + }, + }, + ) + + def full_tensor( + self, *, grad_placements: Sequence[Placement] | None = None + ) -> torch.Tensor: + """ + Return the full tensor of this DTensor. It will perform necessary collectives + to gather the local tensors from other ranks in its DeviceMesh and concatenate + them together. It's a syntactic sugar of the following code: + + ``dtensor.redistribute(placements=[Replicate()] * mesh.ndim).to_local()`` + + Keyword args: + grad_placements (List[:class:`Placement`], optional): the placements describes + the future layout of any gradient layout of the full Tensor returned from this + function. + `full_tensor` converts DTensor to a full torch.Tensor and the returned torch.tensor + might not be used as the original replicated DTensor layout later in the code. This + argument is the hint that user can give to autograd in case the gradient + layout of the returned tensor does not match the original replicated DTensor layout. + If not specified, we will assume the gradient layout of the full tensor be replicated. + + Returns: + A :class:`torch.Tensor` object that represents the full tensor of this DTensor. + + .. note:: ``full_tensor`` is differentiable. + """ + + redist_res = self.redistribute( + placements=[Replicate()] * self.device_mesh.ndim, async_op=False + ) + return _ToTorchTensor.apply(redist_res, grad_placements) + + @property + def device_mesh(self) -> DeviceMesh: + """ + The :class:`DeviceMesh` attribute that associates with this DTensor object. + + .. note:: ``device_mesh`` is a read-only property, it can not be set. + """ + return self._spec.mesh + + @property + def placements(self) -> tuple[Placement, ...]: + """ + The placements attribute of this DTensor that describes the layout of this + DTensor on the its DeviceMesh. + + .. note:: ``placements`` is a read-only property, it can not be set. + """ + return self._spec.placements + + def _raise_if_contains_partial_placements(self) -> None: + """ + Raise an error if the DTensor contains partial placements. + """ + for placement in self._spec.placements: + if not isinstance(placement, Partial): + continue + + raise ValueError( + "Any checkpointing related operations are not supported for " + "DTensor with partial placements!" + ) + + def __create_write_items__(self, fqn: str, object: Any): + self._raise_if_contains_partial_placements() + from torch.distributed.checkpoint.planner_helpers import ( + _create_write_items_for_dtensor, + ) + + if hasattr(self._local_tensor, "__create_write_items__"): + return self._local_tensor.__create_write_items__(fqn, object) # type: ignore[attr-defined] + elif isinstance(self._local_tensor, torch.Tensor): + return [_create_write_items_for_dtensor(fqn, object)] + else: + raise RuntimeError("Unsupported tensor type!") + + def __create_chunk_list__(self): + """ + Return a list of ChunkStorageMetadata, which is a dataclass that describes the size/offset of the local shard/replica + on current rank. For DTensor, each rank will have a single local shard/replica, so the returned list usually only + has one element. + + This dunder method is primariy used for distributed checkpoint purpose. + + Returns: + A List[:class:`ChunkStorageMetadata`] object that represents the shard size/offset on the current rank. + """ + self._raise_if_contains_partial_placements() + from torch.distributed.checkpoint.planner_helpers import ( + _create_chunk_from_dtensor, + ) + + if hasattr(self._local_tensor, "__create_chunk_list__"): + return self._local_tensor.__create_chunk_list__() # type: ignore[attr-defined] + elif isinstance(self._local_tensor, torch.Tensor): + return [_create_chunk_from_dtensor(self)] + else: + raise RuntimeError("Unsupported tensor type!") + + def __get_tensor_shard__(self, index): + self._raise_if_contains_partial_placements() + if hasattr(self._local_tensor, "__get_tensor_shard__"): + return self._local_tensor.__get_tensor_shard__(index) # type: ignore[attr-defined] + elif isinstance(self._local_tensor, torch.Tensor): + return self.to_local() + else: + raise RuntimeError("Unsupported tensor type!") + + +def distribute_tensor( + tensor: torch.Tensor, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, + *, + src_data_rank: int | None = 0, +) -> DTensor: + """ + Distribute a leaf ``torch.Tensor`` (i.e. nn.Parameter/buffers) to the ``device_mesh`` according + to the ``placements`` specified. The rank of ``device_mesh`` and ``placements`` must be the + same. The ``tensor`` to distribute is the logical or "global" tensor, and the API would use + the ``tensor`` from first rank of the DeviceMesh dimension as the source of truth to preserve + the single-device semantic. If you want to construct a DTensor in the middle of the Autograd + computation, please use :meth:`DTensor.from_local` instead. + + Args: + tensor (torch.Tensor): torch.Tensor to be distributed. Note that if you + want to shard a tensor on a dimension that is not evenly divisible by + the number of devices in that mesh dimension, we use ``torch.chunk`` + semantic to shard the tensor and scatter the shards. The uneven sharding + behavior is experimental and subject to change. + device_mesh (:class:`DeviceMesh`, optional): DeviceMesh to distribute the + tensor, if not specified, must be called under a DeviceMesh context + manager, default: None + placements (List[:class:`Placement`], optional): the placements that + describes how to place the tensor on DeviceMesh, must have the same + number of elements as ``device_mesh.ndim``. If not specified, we will + by default replicate the tensor across the ``device_mesh`` from the + first rank of each dimension of the `device_mesh`. + + Keyword args: + src_data_rank (int, optional): the rank of the source data for the logical/global tensor, it is + used by :meth:`distribute_tensor` to scatter/broadcast the shards/replicas to other ranks. + By default, we use ``group_rank=0`` on each DeviceMesh dimension as the source data to preserve + the single-device semantic. If passing ``None`` explicitly, :meth:`distribute_tensor` simply uses + its local data instead of trying to preserve the single-device semantic via scatter/broadcast. + Default: 0 + + Returns: + A :class:`DTensor` or ``XLAShardedTensor`` object. + + Raises: + ValueError: If ``placements`` contains mixed :class:`Partial` reduce types + (e.g., both ``Partial("sum")`` and ``Partial("max")``). All Partial + placements must use the same reduce operation. + + .. note:: + When initialize the DeviceMesh with the ``xla`` device_type, ``distribute_tensor`` + return `XLAShardedTensor` instead. see `this issue `__ + for more details. The XLA integration is experimental and subject to change. + """ + + torch._C._log_api_usage_once("torch.dtensor.distribute_tensor") + + # get default device mesh if there's nothing specified + device_mesh = device_mesh or _mesh_resources.get_current_mesh() + device_type = device_mesh.device_type + if device_type == "xla": + try: + # call PyTorch/XLA SPMD for `xla` backend type device mesh. + # This returns XLAShardedTensor + from torch_xla.distributed.spmd import ( # type:ignore[import] + xla_distribute_tensor, + ) + + return xla_distribute_tensor(tensor, device_mesh, placements) # type:ignore[return-value] + except ImportError as e: + msg = "To use DTensor API with xla, you must install the torch_xla package!" + raise ImportError(msg) from e + + if not tensor.is_leaf: + raise RuntimeError( + "`distribute_tensor` should be used to distribute leaf tensors! but found non-leaf tensor!" + ) + + # convert tensor to the corresponding device type if it's not in that device type + if device_type != tensor.device.type and not tensor.is_meta: + tensor = tensor.to(device_type) + + # set default placements to replicated if not specified + if placements is None: + placements = [Replicate() for _ in range(device_mesh.ndim)] + + if len(placements) != device_mesh.ndim: + raise ValueError( + f"`placements` must have the same length as `device_mesh.ndim`! " + f"Found placements length: {len(placements)}, and device_mesh.ndim: {device_mesh.ndim}." + ) + + # Validate that placements don't contain mixed Partial reduce types + assert_no_mixed_partial_types(placements) + if isinstance(tensor, DTensor): + # if the tensor is already a DTensor, we need to check: + # 1. if we can further shard this DTensor if the two device mesh belong to + # the same parent mesh and further sharding is possible. + # 2. check if device mesh and placements are the same + if tensor.device_mesh != device_mesh: + raise ValueError( + f"Cannot distribute a DTensor with device mesh {tensor.device_mesh} " + f"to a different device mesh {device_mesh}." + ) + if tensor.placements != tuple(placements): + raise ValueError( + f"Cannot distribute a DTensor with placements {tensor.placements} " + f"to a different placements {placements}. do you want to call " + f"`redistribute` instead?" + ) + return tensor + + local_tensor = tensor.detach() + + # TODO(xilun): address sharding order + # distribute the tensor according to the placements. + placements = list(placements) + for idx, placement in enumerate(placements): + if isinstance(placement, Shard | _StridedShard): + placement_dim = ( + placement.dim + tensor.ndim if placement.dim < 0 else placement.dim + ) + if isinstance(placement, Shard): + local_tensor = Shard._make_shard_tensor( + placement_dim, local_tensor, device_mesh, idx, src_data_rank + ) + placements[idx] = Shard(placement_dim) + else: + local_tensor = _StridedShard._make_shard_tensor( + placement_dim, + local_tensor, + device_mesh, + idx, + src_data_rank, + split_factor=placement.split_factor, + ) + placements[idx] = _StridedShard( + placement_dim, split_factor=placement.split_factor + ) + elif isinstance(placement, Replicate): + local_tensor = Replicate._make_replicate_tensor( + local_tensor, device_mesh, idx, src_data_rank + ) + elif isinstance(placement, Partial): + local_tensor = Replicate._make_replicate_tensor( + local_tensor, device_mesh, idx, src_data_rank + ) + local_tensor = placement._partition_value(local_tensor, device_mesh, idx) + else: + raise RuntimeError( + f"Trying to distribute tensor with unsupported placements {placement} on device mesh dimension {idx}!" + ) + placements = tuple(placements) + + if local_tensor is None: + raise AssertionError("distributing a tensor should not be None") + # detach the local tensor passed to DTensor since after the construction + # of DTensor, autograd would work on top of DTensor instead of local tensor + spec = DTensorSpec( + mesh=device_mesh, + placements=placements, + tensor_meta=TensorMeta( + shape=tensor.size(), + stride=tensor.stride(), + dtype=tensor.dtype, + ), + ) + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + local_tensor.requires_grad_(tensor.requires_grad), + spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=tensor.requires_grad, + ) + + +@deprecated("Please use `distribute_tensor` with `src_data_rank=None` instead.") +def _shard_tensor( + full_tensor: torch.Tensor, + placements: Sequence[Shard], + device_mesh: DeviceMesh | None = None, +) -> "DTensor": + """ + Locally shards a full tensor based on indicated sharding arrangement, and + returns a DTensor containing the local shard. + + .. warning:: This is a private API that is subject to change. It skips the + communication otherwise required by `distribute_tensor`. It is only + applicable to cases where all ranks have the same `full_tensor`. For + example, in distributed inference all ranks load from the same + checkpoint. This API will not check for data equality between ranks, it + is thus user's responsibility to ensure the `full_tensor` is the same + across ranks. + + Args: + full_tensor (torch.Tensor): the full tensor to be sharded. + placements (Sequence[:class:`Shard`]): the placements that + describes how to place the local tensor on DeviceMesh. + device_mesh (:class:`DeviceMesh`, optional): DeviceMesh to place the + DTensor. Must have same dimension as the number of placements. + If not specified, would be retrieve from current context. + + Returns: + A :class:`DTensor` object with the shard as its local tensor. + + Examples: + >>> # xdoctest: +SKIP("need world_size and rank") + >>> device_mesh = dist.init_device_mesh("cuda", (world_size,)) + >>> full_tensor = torch.arange(world_size, device=f"cuda:{rank}") + >>> dtensor = _shard_tensor(full_tensor, [Shard(1)], device_mesh) + """ + return distribute_tensor(full_tensor, device_mesh, placements, src_data_rank=None) + + +def distribute_module( + module: nn.Module, + device_mesh: DeviceMesh | None = None, + partition_fn: Callable[[str, nn.Module, DeviceMesh], None] | None = None, + input_fn: Callable[[nn.Module, Any, DeviceMesh], None] | None = None, + output_fn: Callable[[nn.Module, Any, DeviceMesh], None] | None = None, +) -> nn.Module: + """ + Convert a module's parameters/buffers to :class:`DTensor` s, and optionally + convert its inputs/outputs at runtime, using the three user-provided callbacks. + + The behavior of the three callbacks: + + 1. ``partition_fn`` shards the module's parameters/buffers *before* runtime + execution (i.e. it converts plain ``torch.Tensor`` parameters into + :class:`DTensor` parameters). Any parameter/buffer not converted by + ``partition_fn`` is replicated across ``device_mesh`` afterwards. If + ``partition_fn`` is ``None``, *all* parameters/buffers are replicated. + 2. ``input_fn`` / ``output_fn`` convert the module's inputs/outputs *during* + runtime execution (e.g. convert plain tensor inputs into :class:`DTensor`, + or convert :class:`DTensor` outputs back into plain ``torch.Tensor``). + + Args: + module (:class:`nn.Module`): user module to be partitioned. + device_mesh (:class:`DeviceMesh`): the device mesh to place the module. If + ``None``, the current mesh from the ambient mesh context is used. + partition_fn (Callable): a function with signature + ``partition_fn(name: str, module: nn.Module, device_mesh: DeviceMesh) -> None``. + It is called once for *every* submodule (including ``module`` itself), + where ``name`` is the submodule's fully-qualified name (``""`` for the + top-level module) as returned by :meth:`nn.Module.named_modules`. The + function is expected to mutate ``module`` in place, typically by calling + :func:`distribute_tensor` and :meth:`register_parameter` to replace + selected parameters with :class:`DTensor` parameters. Its return value + is ignored. If not specified, all module parameters are replicated + across ``device_mesh``. + input_fn (Callable): a function with signature + ``input_fn(module: nn.Module, inputs: Inputs, device_mesh: DeviceMesh) -> Inputs``, + where ``Inputs`` is the tuple of positional arguments to ``forward`` (one + entry per positional argument, e.g. ``(x,)`` for ``forward(self, x)``; + keyword arguments are not forwarded to ``input_fn``). Installed on the + top-level ``module`` as a ``forward_pre_hook``, it returns the new + argument tuple -- the same structure as ``inputs``, typically with each + ``torch.Tensor`` entry converted to a :class:`DTensor` -- which replaces + ``inputs`` for the actual ``forward`` call. + output_fn (Callable): a function with signature + ``output_fn(module: nn.Module, outputs: Outputs, device_mesh: DeviceMesh) -> Outputs``, + where ``Outputs`` is the return type of ``module.forward`` (a single + tensor, or any nested structure of tensors such as a tuple/dict), with the + tensors typically being :class:`DTensor` s. Installed on the top-level + ``module`` as a ``forward_hook``, it returns the replacement output of the + same structure -- typically with each :class:`DTensor` converted back to a + plain ``torch.Tensor`` via :meth:`DTensor.to_local` or + :meth:`DTensor.full_tensor`, or redistributed to different placements. + + Returns: + A module whose parameters/buffers are all :class:`DTensor` s (the same + ``module`` object, mutated in place). + + .. warning:: + ``input_fn`` and ``output_fn`` previously took two arguments + ``(inputs, device_mesh)`` / ``(outputs, device_mesh)``. That form is + deprecated; use the three-argument form documented above. + + Example:: + + import torch.nn as nn + from torch.distributed.tensor import ( + DTensor, + Replicate, + Shard, + distribute_module, + distribute_tensor, + init_device_mesh, + ) + + + class MyModule(nn.Module): + def __init__(self) -> None: + super().__init__() + self.fc = nn.Linear(8, 8) + + def forward(self, x): + return self.fc(x) + + + mesh = init_device_mesh("cuda", (4,)) + + + def shard_params(name, mod, mesh): + if isinstance(mod, nn.Linear): + for pname, param in mod.named_parameters(recurse=False): + mod.register_parameter( + pname, nn.Parameter(distribute_tensor(param, mesh, [Shard(0)])) + ) + + + # Replicate the input across the mesh before forward. + def to_replicated(mod, inputs, mesh): + return tuple( + distribute_tensor(x, mesh, [Replicate()]) + if isinstance(x, torch.Tensor) + else x + for x in inputs + ) + + + # Convert the DTensor output back to a plain torch.Tensor after forward. + def to_local(mod, outputs, mesh): + return outputs.full_tensor() if isinstance(outputs, DTensor) else outputs + + + sharded = distribute_module( + MyModule(), + mesh, + partition_fn=shard_params, + input_fn=to_replicated, + output_fn=to_local, + ) + + .. note:: + When initializing the ``DeviceMesh`` with the ``xla`` device_type, + ``distribute_module`` returns an ``nn.Module`` with PyTorch/XLA SPMD + annotated parameters. See + `this issue `__ + for more details. The XLA integration is experimental and subject to change. + + """ + + torch._C._log_api_usage_once("torch.dtensor.distribute_module") + + already_distributed = getattr(module, "_distribute_module_applied", False) + if already_distributed: + raise RuntimeError( + "distribute_module should only be called once on a module, " + "but it has already been called on this module!" + ) + + device_mesh = device_mesh or _mesh_resources.get_current_mesh() + device_type = device_mesh.device_type + if device_type == "xla": + try: + # This function annotates all module parameters for auto-partitioning with + # PyTorch/XLA SPMD or explicitly partition to :class:`XLAShardedTensor` parameters + # according to the `partition_fn` specified. + from torch_xla.distributed.spmd import ( # type:ignore[import] + xla_distribute_module, + ) + + return xla_distribute_module( + module, device_mesh, partition_fn, input_fn, output_fn + ) # type:ignore[return-value] + except ImportError as e: + msg = "To use DTensor API with xla, you must install the torch_xla package!" + raise ImportError(msg) from e + + def replicate_module_params_buffers(m: nn.Module, mesh: DeviceMesh) -> None: + # This function loop over the immediate module parameters and + # buffers, replicate all non DTensor params/buffers to DTensor + # parameters/buffers, if they have not been partitioned in the + # partition_fn, we can't easily use `module._apply` here + # because we don't know what happened inside partition_fn as + # user could do anything, i.e. install hooks, and we want to + # preserve those. + full_replicate = [Replicate()] * mesh.ndim + for key, param in m._parameters.items(): + if param is not None and not isinstance(param, DTensor): + m.register_parameter( + key, + nn.Parameter( + distribute_tensor(param.data, mesh, full_replicate), + requires_grad=param.requires_grad, + ), + ) + for key, buffer in m._buffers.items(): + if buffer is not None and not isinstance(buffer, DTensor): + m._buffers[key] = distribute_tensor(buffer, mesh, full_replicate) + + if partition_fn is None: + # if partition_fn not specified, we by default replicate + # all module params/buffers + for submod in module.modules(): + replicate_module_params_buffers(submod, device_mesh) + else: + # apply partition_fun to submodules + for name, submod in module.named_modules(): + partition_fn(name, submod, device_mesh) + replicate_module_params_buffers(submod, device_mesh) + + # register input_fn as module forward pre hook + if input_fn is not None: + # check the input_fn signature + num_args = len(inspect.signature(input_fn).parameters) + if num_args == 2: + # input_fn only takes in inputs and device mesh + warnings.warn( + "Deprecating input_fn that takes two arguments (inputs, device_mesh), " + "please use input_fn that takes in (module, inputs, device_mesh) instead!", + FutureWarning, + stacklevel=2, + ) + module.register_forward_pre_hook( + lambda _, inputs: input_fn(inputs, device_mesh) # type: ignore[call-arg] + ) + elif num_args == 3: + # input_fn takes in module, inputs, device mesh + module.register_forward_pre_hook( + lambda mod, inputs: input_fn(mod, inputs, device_mesh) + ) + else: + raise ValueError( + f"input_fn should take in 3 arguments, but got {num_args} arguments!" + ) + # register output_fn as module forward hook + if output_fn is not None: + num_args = len(inspect.signature(output_fn).parameters) + if num_args == 2: + # output_fn only takes in outputs and device mesh + warnings.warn( + "Deprecating output_fn that takes two arguments (inputs, device_mesh), " + "please use output_fn that takes in (module, inputs, device_mesh) instead!", + FutureWarning, + stacklevel=2, + ) + module.register_forward_hook( + lambda mod, inputs, outputs: output_fn(outputs, device_mesh) # type: ignore[call-arg] + ) + elif num_args == 3: + module.register_forward_hook( + lambda mod, inputs, outputs: output_fn(mod, outputs, device_mesh) + ) + else: + raise ValueError( + f"output_fn should take in 3 arguments, but got {num_args} arguments!" + ) + + module._distribute_module_applied = True # type: ignore[assignment] + return module + + +# Below are tensor factory function APIs, which are used to create a DTensor directly. We need +# to make separate factory function APIs because tensor subclass could not override the tensor +# factory methods, and we need user to call the factory functions with user intended device_mesh +# and placements to create a proper DTensor. + + +@maybe_run_for_local_tensor +def _linspace_local_tensor( + init_op, start, end, total_steps, local_steps, offset, **kwargs +): + local_start = start + local_end = end + if local_steps > 0 and total_steps > 1: + step = (end - start) / (total_steps - 1) + local_start = start + offset * step + local_end = local_start + step * (local_steps - 1) + elif local_steps > 0: + local_end = start + return init_op(local_start, local_end, steps=local_steps, **kwargs) + + +@maybe_run_for_local_tensor +def _logspace_local_tensor( + init_op, start, end, base, total_steps, local_steps, offset, **kwargs +): + local_start = start + local_end = end + if local_steps > 0 and total_steps > 1: + step = (end - start) / (total_steps - 1) + local_start = start + offset * step + local_end = local_start + step * (local_steps - 1) + elif local_steps > 0: + local_end = start + return init_op(local_start, local_end, steps=local_steps, base=base, **kwargs) + + +def _dtensor_init_helper( # type: ignore[no-untyped-def] + init_op, + size: torch.Size, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, + **kwargs, +) -> DTensor: + # if device_mesh is None, use the one from mesh resources + device_mesh = device_mesh or _mesh_resources.get_current_mesh() + kwargs["device"] = device_mesh.device_type + + # set default placements to replicated if not specified + placements = placements or tuple(Replicate() for _ in range(device_mesh.ndim)) + + # check device_mesh against placements + if device_mesh.ndim != len(placements): + raise AssertionError("mesh dimension does not match the length of placements") + + if kwargs["layout"] != torch.strided: + raise AssertionError("layout value not supported!") + torch_stride = torch._prims_common.make_contiguous_strides_for(size) + + # get local tensor shape + local_shape, global_offset = compute_local_shape_and_global_offset( + size, + device_mesh, + placements, + skip_offset=init_op not in (torch.linspace, torch.logspace), + ) + + # initialize the local tensor + if init_op is torch.full: + fill_value = kwargs.pop("fill_value") + local_tensor = init_op(local_shape, fill_value, **kwargs) + elif init_op is torch.linspace: + start = kwargs.pop("start") + end = kwargs.pop("end") + local_tensor = _linspace_local_tensor( + init_op, + start, + end, + size[0], + local_shape[0], + global_offset[0], + **kwargs, + ) + elif init_op is torch.logspace: + start = kwargs.pop("start") + end = kwargs.pop("end") + base = kwargs.pop("base") + local_tensor = _logspace_local_tensor( + init_op, + start, + end, + base, + size[0], + local_shape[0], + global_offset[0], + **kwargs, + ) + elif init_op is torch.rand or init_op is torch.randn: + # this tensor meta is not used except `shape` + dtype = kwargs.get("dtype", torch.get_default_dtype()) + + tensor_meta = TensorMeta(size, torch_stride, dtype) + spec = DTensorSpec(device_mesh, tuple(placements), tensor_meta=tensor_meta) + + if random.is_rng_supported_mesh(device_mesh) and not random._rng_tracker: + random._rng_tracker = random.OffsetBasedRNGTracker(device_mesh) + + if random._rng_tracker is None: + raise AssertionError + with random._rng_tracker._distribute_region(spec): + local_tensor = init_op(local_shape, **kwargs) + else: + local_tensor = init_op(local_shape, **kwargs) + + spec = DTensorSpec( + device_mesh, + tuple(placements), + tensor_meta=TensorMeta( + size, + torch_stride, + local_tensor.dtype, + ), + ) + + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + local_tensor, + spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=kwargs["requires_grad"], + ) + + +def ones( # type: ignore[no-untyped-def] + *size, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + requires_grad: bool = False, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with the scalar value 1, with the shape defined + by the variable argument ``size``. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: ones(1,2,3..) or ones([1,2,3..]) or ones((1,2,3..)) + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned DTensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.ones, + torch_size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def empty( # type: ignore[no-untyped-def] + *size, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + requires_grad: bool = False, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with uninitialized data. The shape of the :class:`DTensor` + is defined by the variable argument ``size``. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: empty(1,2,3..) or empty([1,2,3..]) or empty((1,2,3..)) + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`).\ + layout (:class:`torch.layout`, optional): the desired layout of returned :class:`DTensor`. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.empty, + torch_size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def full( # type: ignore[no-untyped-def] + size, + fill_value, + *, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + requires_grad: bool = False, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with ``fill_value`` according to ``device_mesh`` and + ``placements``, with the shape defined by the argument ``size``. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: ones(1,2,3..) or ones([1,2,3..]) or ones((1,2,3..)) + fill_value(Scalar): the value to fill the output tensor with. + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned DTensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks. + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.full, + torch_size, + fill_value=fill_value, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def linspace( + start, + end, + steps, + *, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + requires_grad: bool = False, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` of size ``steps`` whose values are evenly spaced from + `start` to `end`. + + Args: + start (float or :class:`DTensor`): the starting value for the set of points. If + :class:`DTensor`, it must be 0-dimensional + end (float or :class:`DTensor`): the ending value for the set of points. If + :class:`DTensor`, it must be 0-dimensional + steps (int): size of the constructed :class:`DTensor` + + Keyword args: + dtype (:class:`torch.dtype`, optional): the data type to perform the computation + in. Default: if ``None``, uses the global default dtype + (see :func:`torch.set_default_dtype`) when both `start` and `end` are real, + and corresponding complex dtype when either is complex. + layout (:class:`torch.layout`, optional): the desired layout of returned + :class:`DTensor`. Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + if placements is not None and any(isinstance(p, _StridedShard) for p in placements): + raise AssertionError("linspace does not support _StridedShard placements") + + if isinstance(start, torch.Tensor): + torch._check( + start.dim() == 0, + lambda: "linspace only supports 0-dimensional start and end tensors", + ) + start = start.item() + + if isinstance(end, torch.Tensor): + torch._check( + end.dim() == 0, + lambda: "linspace only supports 0-dimensional start and end tensors", + ) + end = end.item() + + torch_size = normalize_to_torch_size((steps,)) + + return _dtensor_init_helper( + torch.linspace, + torch_size, + start=start, + end=end, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def logspace( + start, + end, + steps, + base=10.0, + *, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + requires_grad: bool = False, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` of size ``steps`` whose values are evenly spaced from + ``base`` :sup:`start` to ``base`` :sup:`end`, inclusive, on a logarithmic scale + with base ``base``. + + Args: + start (float or :class:`DTensor`): the starting value for the set of points. If + :class:`DTensor`, it must be 0-dimensional + end (float or :class:`DTensor`): the ending value for the set of points. If + :class:`DTensor`, it must be 0-dimensional + steps (int): size of the constructed :class:`DTensor` + + Keyword args: + base (float, optional): base of the logarithm function. Default: ``10.0``. + dtype (:class:`torch.dtype`, optional): the data type to perform the computation + in. Default: if ``None``, uses the global default dtype + (see :func:`torch.set_default_dtype`) when both `start` and `end` are real, + and corresponding complex dtype when either is complex. + layout (:class:`torch.layout`, optional): the desired layout of returned + :class:`DTensor`. Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + if placements is not None and any(isinstance(p, _StridedShard) for p in placements): + raise AssertionError("logspace does not support _StridedShard placements") + + if isinstance(start, torch.Tensor): + torch._check( + start.dim() == 0, + lambda: "logspace only supports 0-dimensional start and end tensors", + ) + start = start.item() + + if isinstance(end, torch.Tensor): + torch._check( + end.dim() == 0, + lambda: "logspace only supports 0-dimensional start and end tensors", + ) + end = end.item() + + torch_size = normalize_to_torch_size((steps,)) + + return _dtensor_init_helper( + torch.logspace, + torch_size, + start=start, + end=end, + base=base, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def rand( # type: ignore[no-untyped-def] + *size, + requires_grad: bool = False, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with random numbers from a uniform distribution + on the interval ``[0, 1)``. The shape of the tensor is defined by the variable + argument ``size``. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: ones(1,2,3..) or ones([1,2,3..]) or ones((1,2,3..)) + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned DTensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks. + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.rand, + torch_size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def randn( # type: ignore[no-untyped-def] + *size, + requires_grad: bool = False, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with random numbers from a normal distribution + with mean 0 and variance 1. The shape of the tensor is defined by the variable + argument ``size``. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: ones(1,2,3..) or ones([1,2,3..]) or ones((1,2,3..)) + + Keyword args: + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned DTensor. + Default: ``torch.strided``. + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks. + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.randn, + torch_size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +def zeros( # type: ignore[no-untyped-def] + *size, + requires_grad: bool = False, + dtype: torch.dtype | None = None, + layout: torch.layout = torch.strided, + device_mesh: DeviceMesh | None = None, + placements: Sequence[Placement] | None = None, +) -> DTensor: + """ + Returns a :class:`DTensor` filled with the scalar value 0. + + Args: + size (int...): a sequence of integers defining the shape of the output :class:`DTensor`. + Can be a variable number of arguments or a collection like a list or tuple. + E.g.: zeros(1,2,3..) or zeros([1,2,3..]) or zeros((1,2,3..)) + Keyword args: + requires_grad (bool, optional): If autograd should record operations on the + returned :class:`DTensor`. Default: ``False``. + dtype (:class:`torch.dtype`, optional): the desired data type of returned :class:`DTensor`. + Default: if ``None``, uses a global default (see :func:`torch.set_default_dtype`). + layout (:class:`torch.layout`, optional): the desired layout of returned :class:`DTensor`. + Default: ``torch.strided``. + device_mesh: :class:`DeviceMesh` type, contains the mesh info of ranks + placements: a sequence of :class:`Placement` type: ``Shard``, ``Replicate`` + + Returns: + A :class:`DTensor` object on each rank + """ + torch_size = normalize_to_torch_size(size) + + return _dtensor_init_helper( + torch.zeros, + torch_size, + dtype=dtype, + layout=layout, + requires_grad=requires_grad, + device_mesh=device_mesh, + placements=placements, + ) + + +# Module-level alias for DTensor.from_local to support FX code generation. +# FX uses __name__ instead of __qualname__ when generating code, which loses +# the class context for static methods. This alias allows generated code like +# `torch.distributed.tensor._api.from_local(...)` to work correctly. +from_local = DTensor.from_local + +_register_distributed_opaque_types() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_collective_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_collective_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..f134b9e6b64d81aa6c2066844a08a73cb31f4110 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_collective_utils.py @@ -0,0 +1,619 @@ +# mypy: allow-untyped-defs +import logging +import math +from dataclasses import dataclass +from functools import lru_cache +from typing import Optional + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.tensor._dtensor_spec as dtensor_spec +from torch._C._distributed_c10d import _resolve_process_group +from torch._logging import warning_once +from torch.distributed._functional_collectives import _are_we_tracing +from torch.distributed._local_tensor import ( + local_tensor_mode, + maybe_run_for_local_tensor, +) +from torch.distributed.device_mesh import _mesh_resources, DeviceMesh +from torch.distributed.distributed_c10d import ( + broadcast, + get_group_rank, + get_rank, + GroupName, + ProcessGroup, + scatter, + Work, +) +from torch.fx.experimental.symbolic_shapes import guard_or_false +from torch.types import IntLikeType + + +logger = logging.getLogger(__name__) + +# Opaque types must be registered before defining schemas that reference them, +# so the schema parser recognizes the type names and uses PyObjectType (which +# wraps Python objects as ConcretePyObjectHolder) instead of AnyType (which +# calls toTypeInferredIValue and fails for Python-only opaque types). +from torch.distributed.device_mesh import _register_distributed_opaque_types + + +_register_distributed_opaque_types() + +_dtensor_lib = torch.library.Library("_dtensor", "FRAGMENT") +_dtensor_lib.define( + "mesh_get_process_group(" + "torch.distributed.device_mesh.DeviceMesh mesh, int dim" + ") -> torch.distributed.distributed_c10d.ProcessGroup" +) + + +@torch.library.impl("_dtensor::mesh_get_process_group", "CompositeExplicitAutograd") +def _mesh_get_process_group_impl(mesh, dim): + return mesh.get_group(dim) + + +@torch.library.register_fake("_dtensor::mesh_get_process_group") +def _mesh_get_process_group_fake(mesh, dim): + from torch._library.fake_class_registry import maybe_unwrap_fake_script_object + + real_mesh = maybe_unwrap_fake_script_object(mesh) + return real_mesh.get_group(dim) + + +@torch.library.register_fake("_dtensor::shard_dim_alltoall") +def _shard_dim_alltoall_meta( + input, gather_dim, shard_dim, group_name: GroupName | ProcessGroup +): + if isinstance(group_name, str): + # pyrefly: ignore[bad-argument-type] # pyrefly bug + group_name = _resolve_process_group(group_name) + group_size = group_name.size() + stacked_list = [torch.empty_like(input) for _ in range(group_size)] + group_rank = get_group_rank(group_name, get_rank()) + + cat_tensor = torch.cat(stacked_list, dim=gather_dim) + # pyrefly: ignore [unsupported-operation] + chunk_size = cat_tensor.size(shard_dim) // group_size + chunk = torch.narrow(cat_tensor, shard_dim, group_rank * chunk_size, chunk_size) + return chunk.contiguous() + + +def _shard_dim_alltoall_backward(ctx, grad_output): + return ( + torch.ops._dtensor.shard_dim_alltoall.default( + grad_output.contiguous(), ctx.shard_dim, ctx.gather_dim, ctx.group_name + ), + None, + None, + None, + ) + + +def _shard_dim_alltoall_setup_context(ctx, inputs, output): + input, gather_dim, shard_dim, group_name = inputs + ctx.gather_dim = gather_dim + ctx.shard_dim = shard_dim + ctx.group_name = group_name + + +torch.library.register_autograd( + "_dtensor::shard_dim_alltoall", + _shard_dim_alltoall_backward, + setup_context=_shard_dim_alltoall_setup_context, +) + + +def shard_dim_alltoall(input, gather_dim, shard_dim, mesh, mesh_dim): + if mesh.device_type == "cpu" and local_tensor_mode() is None: + # Gloo does not support alltoall, so falling back to allgather + chunk + warning_once( + logger, + "CPU process group does not support alltoall yet, falling back with allgather + chunk!", + ) + out = funcol.all_gather_single(input, gather_dim, (mesh, mesh_dim)) + if isinstance(out, funcol.AsyncCollectiveTensor): + # stick to the same behavior for the alltoall case, remove this once we enable alltoall async + out = out.wait() + from torch.distributed.tensor.placement_types import Shard + + out = Shard._custom_chunk(out, mesh.size(mesh_dim), dim=shard_dim)[ + mesh.get_local_rank(mesh_dim) + ] + return out.contiguous() + + group = funcol._resolve_group((mesh, mesh_dim)) + # TODO: enable async op for shard_dim_alltoall + return torch.ops._dtensor.shard_dim_alltoall( + input, gather_dim, shard_dim, funcol._group_or_group_name(group) + ) + + +def mesh_scatter( + output: torch.Tensor, + scatter_list: list[torch.Tensor], + mesh: DeviceMesh, + mesh_dim: int = 0, + async_op: bool = False, + *, + group_src: int = 0, +) -> Work | None: + """ + scatter a list of tensors to a device mesh dimension. We by default + use the first rank of the mesh dimension as the source of truth, i.e + for a 2d mesh [[0, 1], [2, 3]], if we scatter on mesh_dim = 1, we will + scatter the tensor list on rank 0 to rank 0/1, and tensor list on rank + 2 to rank 2/3. + + Args: + output (torch.Tensor): the tensor to receive the scattered list. + scatter_list (List[torch.Tensor]): the tensor list to be scattered. + mesh_dim (int, optional): indicate which mesh dimension we want + to scatter on, we by default choose the first rank on the + mesh dimension as source of truth. + + Keyword args: + group_src (int, optional): the group rank of the source data for the + logical/global tensor, on the specific mesh dimension. By default, we + use ``group_rank=0`` on each DeviceMesh dimension as the source data + to preserve the single-device semantic. If passing ``None`` explicitly, + this method simply uses its local data with no communication. + + Returns: + A :class:`Work` object + """ + # TODO: Ideally we should use the meta tensor way + # (to register a meta kernel for the collective op) + # so that it would avoid the communication. Need to + # remove the check below once that is done. + if output.is_meta: + return None + dim_group = mesh.get_group(mesh_dim) + if not isinstance(dim_group, ProcessGroup): + raise AssertionError + + if group_src == get_rank(dim_group): + fut = scatter( + output, + scatter_list=scatter_list, + group=dim_group, + async_op=async_op, + group_src=group_src, + ) + else: + fut = scatter( + output, + scatter_list=None, + group=dim_group, + async_op=async_op, + group_src=group_src, + ) + + return fut + + +def mesh_broadcast( + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int = 0, + async_op: bool = False, + *, + group_src: int = 0, +) -> Work | None: + """ + broadcast the tensor to a device mesh dimension. We by default + use the first rank of the mesh dimension as the source of truth, i.e + for a 2d mesh [[0, 1], [2, 3]], if we broadcast on mesh_dim = 1, we will + broadcast the tensor on rank 0 to rank 0/1, and tensor on rank 2 + to rank 2/3. + + Args: + tensor (torch.Tensor): tensor to broadcast. + mesh_dim (int, optional): indicate which mesh dimension we want + to scatter on, we by default choose the first rank on the + mesh dimension as source of truth. + + Keyword args: + group_src (int, optional): the group rank of the source data for the + logical/global tensor, on the specific mesh dimension. By default, we + use ``group_rank=0`` on each DeviceMesh dimension as the source data + to preserve the single-device semantic. If passing ``None`` explicitly, + this method simply uses its local data with no communication. + + Returns: + A :class:`Work` object + """ + # TODO: Ideally we should use the meta tensor way + # (to register a meta kernel for the collective op) + # so that it would avoid the communication. Need to + # remove the check below once that is done. + if tensor.is_meta: + return None + dim_group = mesh.get_group(mesh_dim) + if not isinstance(dim_group, ProcessGroup): + raise AssertionError + + return broadcast(tensor, group=dim_group, async_op=async_op, group_src=group_src) + + +@maybe_run_for_local_tensor +def pad_tensor( + tensor: torch.Tensor, pad_dim: int, pad_size: IntLikeType +) -> torch.Tensor: + # During tracing, always emit the pad op even when pad_size=0 so all + # ranks produce identical FX graph structure (SPMD). + # In eager with concrete pad_size=0, guard_or_false returns True and we + # skip the no-op pad. Check _are_we_tracing() first to avoid + # guard_or_false creating a guard that concretizes symbolic pad sizes + # during make_fx tracing. + if isinstance(pad_size, int): + # Fast path: avoids _are_we_tracing() which is costly at compile + # time due to multiple C++ dispatch mode checks. + if pad_size == 0: + return tensor + elif not _are_we_tracing() and guard_or_false(pad_size == 0): + return tensor + if _are_we_tracing(): + from torch.fx.experimental.symbolic_shapes import has_free_unbacked_symbols + + if has_free_unbacked_symbols(tensor): + pad_shape: list[IntLikeType] = list(tensor.shape) + pad_shape[pad_dim] = pad_size + return torch.cat([tensor, tensor.new_zeros(pad_shape)], dim=pad_dim) + + pad = [0, 0] * (tensor.ndim - pad_dim) + pad[-1] = pad_size # pyrefly: ignore[unsupported-operation] + return torch.nn.functional.pad(tensor, pad) + + +@maybe_run_for_local_tensor +def unpad_tensor( + tensor: torch.Tensor, pad_dim: int, pad_size: IntLikeType +) -> torch.Tensor: + # During tracing, always emit the narrow op even when pad_size=0 so all + # ranks produce identical FX graph structure (SPMD). + # In eager with concrete pad_size=0, guard_or_false returns True and we + # skip the no-op narrow. Check _are_we_tracing() first to avoid + # guard_or_false creating a guard that concretizes symbolic pad sizes + # during make_fx tracing. + if isinstance(pad_size, int): + # Fast path: avoids _are_we_tracing() which is costly at compile + # time due to multiple C++ dispatch mode checks. + if pad_size == 0: + return tensor + elif not _are_we_tracing() and guard_or_false(pad_size == 0): + return tensor + return tensor.narrow( + pad_dim, + start=0, + length=tensor.size(pad_dim) - pad_size, + ) + + +def fill_empty_tensor_to_shards( + shards: list[torch.Tensor], shard_dim: int, num_empty_tensors: int +) -> list[torch.Tensor]: + if num_empty_tensors == 0: + return shards + tensor_size = list(shards[0].size()) + tensor_size[shard_dim] = 0 + tensor = shards[0].new_zeros(tensor_size) + shards.extend(tensor for _ in range(num_empty_tensors)) + return shards + + +def check_tensor_meta( + local_tensor, check_shape_stride=False +) -> Optional["dtensor_spec.TensorMeta"]: + local_metadata = { + "dtype": local_tensor.dtype, + "requires_grad": local_tensor.requires_grad, + } + + if check_shape_stride: + local_metadata.update( + {"shape": local_tensor.shape, "stride": local_tensor.stride()} + ) + + gathered_metadata = [None for _ in range(torch.distributed.get_world_size())] + torch.distributed.all_gather_object(gathered_metadata, local_metadata) + + # Check if metadata is consistent across ranks + if not all(meta == local_metadata for meta in gathered_metadata): + raise ValueError( + "Inconsistent tensor metadata (including shape and stride) across ranks." + ) + return None + + +def spec_to_bytes(spec: "dtensor_spec.DTensorSpec") -> int: + if spec.tensor_meta is None: + raise AssertionError("spec should have tensor meta defined!") + return spec.tensor_meta.dtype.itemsize * math.prod(spec.shape) + + +@dataclass +class MeshTopoInfo: + """ + Mesh information for collective cost estimation + """ + + mesh: DeviceMesh + mesh_dim_devices: list[int] + mesh_dim_bandwidth: list[float] + mesh_dim_latency: list[float] + + @staticmethod + @lru_cache(None) + def build_from_mesh(mesh: DeviceMesh) -> "MeshTopoInfo": + # Generate mesh topology info for intra-host/inter-host communication pattern + # Note that we made bunch of assumptions for simplicity: + # 1. we assume the mesh is homogeneous, and it's gpu/nccl model + # 2. we assume gpu arch is Ampere or Hopper + # 3. we assume collectives are all ring base algo for now + num_devices_per_host = _mesh_resources.num_devices_per_host(mesh.device_type) + # the base bw number (intra-node), GB/s + base_bw = 87.7 + mesh_dim_bandwidth = [base_bw] * mesh.ndim + # the latency in terms of us (intra-node, nv-link) + mesh_dim_latency = [0.6] * mesh.ndim + mesh_dim_devices = [1] * mesh.ndim + + total_num_devices = 1 + for mesh_dim in reversed(range(mesh.ndim)): + num_devices = mesh.size(mesh_dim) + mesh_dim_devices[mesh_dim] = num_devices + total_num_devices *= num_devices + if total_num_devices > num_devices_per_host: + # magic number for inter-host communication bandwidth/latency factor + # This number assumes latest GPU arch, i.e. Ampere or Hopper + # TODO: see if we need to tweak this or offer a way for user + # to specify the bandwidths/latency + mesh_dim_bandwidth[mesh_dim] *= 0.22 + # set to ethernet latency for inter-host + mesh_dim_latency[mesh_dim] = 2.7 + + return MeshTopoInfo( + mesh, mesh_dim_devices, mesh_dim_bandwidth, mesh_dim_latency + ) + + +def allgather_cost(bytes_gb: float, mesh_topo: MeshTopoInfo, mesh_dim: int) -> float: + num_devices_on_mesh_dim = mesh_topo.mesh_dim_devices[mesh_dim] + mesh_dim_bandwidth = mesh_topo.mesh_dim_bandwidth[mesh_dim] + num_hops = num_devices_on_mesh_dim - 1 + # base latency + comm latency + latency = 6.6 + num_hops * mesh_topo.mesh_dim_latency[mesh_dim] # us + bw = (bytes_gb * num_hops / num_devices_on_mesh_dim) / mesh_dim_bandwidth # s + return latency + bw * 1e6 # rescale to us + + +def allreduce_cost(bytes_gb: float, mesh_topo: MeshTopoInfo, mesh_dim: int) -> float: + num_devices_on_mesh_dim = mesh_topo.mesh_dim_devices[mesh_dim] + mesh_dim_bandwidth = mesh_topo.mesh_dim_bandwidth[mesh_dim] + # allreduce have almost 2x comm bytes compare to allgather/reduce_scatter + num_hops = 2 * (num_devices_on_mesh_dim - 1) + + latency = 6.6 + num_hops * mesh_topo.mesh_dim_latency[mesh_dim] + bw = (bytes_gb * num_hops / num_devices_on_mesh_dim) / mesh_dim_bandwidth + return latency + bw * 1e6 + + +def reduce_scatter_cost( + bytes_gb: float, + mesh_topo: MeshTopoInfo, + mesh_dim: int, +) -> float: + num_devices_on_mesh_dim = mesh_topo.mesh_dim_devices[mesh_dim] + mesh_dim_bandwidth = mesh_topo.mesh_dim_bandwidth[mesh_dim] + num_hops = num_devices_on_mesh_dim - 1 + # base latency + comm latency + latency = 6.6 + num_hops * mesh_topo.mesh_dim_latency[mesh_dim] + bw = (bytes_gb * num_hops / num_devices_on_mesh_dim) / mesh_dim_bandwidth + return latency + bw * 1e6 + + +def _compute_placement_transition_cost( + current_placement: "dtensor_spec.Placement", + target_placement: "dtensor_spec.Placement", + mesh_topo: MeshTopoInfo, + mesh_dim: int, + comm_bytes_gb: float, +) -> tuple[float, float]: + """ + Compute the cost of transitioning from one placement to another on a single mesh dimension. + + Args: + current_placement: The current placement on the mesh dimension. + target_placement: The target placement on the mesh dimension. + mesh_topo: Mesh topology information for cost estimation. + mesh_dim: The mesh dimension where the transition happens. + comm_bytes_gb: The communication bytes in GB for this step. + + Returns: + A tuple of (cost, updated_comm_bytes_gb): + - cost: The communication cost for this transition. ``float("inf")`` + means that the strategy planner must not select the transition + implicitly; the explicit redistribution API may still support it. + - updated_comm_bytes_gb: The updated communication bytes after this step. + """ + if current_placement == target_placement: + return 0.0, comm_bytes_gb + + num_devices_on_mesh_dim = mesh_topo.mesh_dim_devices[mesh_dim] + + # NOTE: is_shard() does not match _StridedShard; see _is_shard_like(). + # Safe today: redistribute_cost bails with inf when shard_order is None. + if current_placement.is_shard() and target_placement.is_replicate(): + # allgather gives larger comm bytes + comm_bytes_gb *= num_devices_on_mesh_dim + return allgather_cost(comm_bytes_gb, mesh_topo, mesh_dim), comm_bytes_gb + elif current_placement.is_shard() and target_placement.is_shard(): + # should be alltoall comm, since we haven't implement it yet, add 1.0 as penalty + # to favor allgather instead + # TODO: add alltoall_cost + return allgather_cost(comm_bytes_gb, mesh_topo, mesh_dim) + 1.0, comm_bytes_gb + elif current_placement.is_partial() and target_placement.is_replicate(): + return allreduce_cost(comm_bytes_gb, mesh_topo, mesh_dim), comm_bytes_gb + elif current_placement.is_partial() and target_placement.is_shard(): + cost = reduce_scatter_cost(comm_bytes_gb, mesh_topo, mesh_dim) + # after reduce_scatter the comm bytes for further collectives halved. + comm_bytes_gb /= num_devices_on_mesh_dim + return cost, comm_bytes_gb + elif current_placement.is_shard() and target_placement.is_partial(): + # Shard -> Partial("sum") is a valid explicit redistribution, but it + # materializes a logical-shape tensor and copies the local shard into it. + # The cost model accounts only for communication, so assigning zero cost + # would make the strategy planner over-prefer this memory- and copy-heavy + # transition. Exclude it from implicit strategy selection instead. + return float("inf"), comm_bytes_gb + elif current_placement.is_partial() and target_placement.is_partial(): + # we already handled the == case at the top, and we ban converting between partial types. + return float("inf"), comm_bytes_gb + elif current_placement.is_replicate() and target_placement.is_shard(): + comm_bytes_gb /= num_devices_on_mesh_dim + return 0.0, comm_bytes_gb + + return 0.0, comm_bytes_gb + + +def one_step_redistribute_cost( + current_spec: "dtensor_spec.DTensorSpec", + target_spec: "dtensor_spec.DTensorSpec", +) -> float: + """ + Calculate the cost of a single redistribution step between two DTensorSpecs. + + This function computes the communication cost for a one-step redistribution + where the current and target specs differ by exactly one placement on one + mesh dimension. + + Args: + current_spec: The current DTensorSpec. + target_spec: The target DTensorSpec. + + Returns: + The communication cost for this step (float("inf") if invalid). + """ + if current_spec.mesh != target_spec.mesh: + return float("inf") + + if current_spec.placements == target_spec.placements: + return 0.0 + + # Find the mesh dimension that differs + mesh_dim = -1 + current_placement = None + target_placement = None + for i, (cur, tgt) in enumerate( + zip(current_spec.placements, target_spec.placements) + ): + if cur != tgt: + if mesh_dim != -1: + # More than one dimension differs - not a single step + raise ValueError( + "one_step_redistribute_cost expects specs that differ by exactly one placement" + ) + mesh_dim = i + current_placement = cur + target_placement = tgt + + if mesh_dim == -1: + return 0.0 + + if current_placement is None or target_placement is None: + raise AssertionError + + mesh_topo = MeshTopoInfo.build_from_mesh(current_spec.mesh) + comm_bytes_gb = ( + spec_to_bytes(current_spec) / current_spec.num_shards / 1024 / 1024 / 1024 + ) + + cost, _ = _compute_placement_transition_cost( + current_placement, target_placement, mesh_topo, mesh_dim, comm_bytes_gb + ) + return cost + + +def redistribute_cost( + current_spec: "dtensor_spec.DTensorSpec", + target_spec: "dtensor_spec.DTensorSpec", +) -> float: + """ + This function returns the cost of redistribute from current to target DTensorSpec. + + NOTE: + 1. Only consider communication cost here, since computation costs for redistribute + are quite trivial (i.e. we only need to narrow or simple division) + 2. Only consider redistribute cost on same mesh, cross mesh communication cost is + not quite needed for operator strategy estimation/selection. + """ + if current_spec.mesh != target_spec.mesh: + # make infinite cost if meshes are not same + # TODO: see if we want to support this once there's cross mesh communication + return float("inf") + if current_spec.is_replicated(): + # short-cut: comm cost is 0 if current spec is already full replication + return 0.0 + + # TODO(zpcore): test placements with _StridedShard if we replace shard_order + # with _StridedShard. + if ( + current_spec.placements == target_spec.placements + and current_spec.shard_order == target_spec.shard_order + ): + return 0.0 + + # For sub-meshes, ranks not participating in the mesh should not compute + # redistribution costs. Return 0 since they won't actually participate. + if not current_spec.mesh._is_current_rank_part_of_mesh(): + return 0.0 + + mesh_topo = MeshTopoInfo.build_from_mesh(current_spec.mesh) + cost = 0.0 + comm_bytes_gb = ( + spec_to_bytes(current_spec) / current_spec.num_shards / 1024 / 1024 / 1024 + ) + # Transformation that considered for redistribute cost: + # 1. allgather 2. alltoall + # 3. allreduce 4. reduce_scatter + from torch.distributed._functional_collectives import _are_we_tracing + from torch.distributed.tensor._redistribute import ( + _gen_transform_infos, + _gen_transform_infos_non_cached, + ) + + # TODO(zpcore): Support _StridedShard redistribution. Remove the temporary + # fix, which is to prevent StridedShard erroring out. + if current_spec.shard_order is None or target_spec.shard_order is None: + return float("inf") + + # No redistribution needed when placements are already identical. + # This also prevents potential failures in _gen_transform_infos for certain configurations + # (e.g., sub-meshes) where finding a transform path between identical states may error out. + # TODO(zpcore): test placements with _StridedShard if we replace shard_order + # with _StridedShard. + if ( + current_spec.placements == target_spec.placements + and current_spec.shard_order == target_spec.shard_order + ): + return cost + + if _are_we_tracing(): + transform_infos = _gen_transform_infos_non_cached(current_spec, target_spec) + else: + transform_infos = _gen_transform_infos(current_spec, target_spec) + for transform_info in transform_infos: + if current_spec.tensor_meta is None: + raise AssertionError("spec should have tensor meta defined!") + current = transform_info.src_dst_placements[0] + target = transform_info.src_dst_placements[1] + mesh_dim = transform_info.mesh_dim + step_cost, comm_bytes_gb = _compute_placement_transition_cost( + current, target, mesh_topo, mesh_dim, comm_bytes_gb + ) + if step_cost == float("inf"): + return float("inf") + cost += step_cost + + return cost diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_decompositions.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_decompositions.py new file mode 100644 index 0000000000000000000000000000000000000000..e6db7dc6916f6e24f5e59073f2fca947977129bf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_decompositions.py @@ -0,0 +1,335 @@ +# mypy: allow-untyped-defs +""" +Decomposition-based sharding propagation for DTensor. + +When an operator doesn't have a registered sharding strategy, we derive one by +tracing through its decomposition. The decomposed ops (which do have strategies) +determine how placements propagate through the original op. +""" + +from __future__ import annotations + +import itertools +from typing import Any, TYPE_CHECKING + +import torch +from torch._decomp import decomposition_table +from torch.distributed._functional_collectives import _are_we_tracing +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import OpSchema, OpStrategy, RuntimeSchemaInfo +from torch.distributed.tensor._utils import try_find_mesh_from_args +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Placement, + Replicate, + Shard, +) +from torch.fx.experimental.symbolic_shapes import GuardOnDataDependentSymNode +from torch.utils._python_dispatch import TorchDispatchMode + + +def _infer_schema_info_from_op(op: OpOverload) -> RuntimeSchemaInfo: + """Infer RuntimeSchemaInfo from an operator's schema for decomposition ops""" + schema = op._schema + + # Find first non-tensor positional arg index + static_argnum = None + for i, arg in enumerate(schema.arguments): + if arg.kwarg_only: + break + if arg.type.kind() != "TensorType" and static_argnum is None: + static_argnum = i + break + + # Find keyword-only args that aren't tensors + kwarg_only_names = [] + for arg in schema.arguments: + if arg.kwarg_only and arg.type.kind() != "TensorType": + kwarg_only_names.append(arg.name) + + kwargs = {} + if static_argnum is not None: + kwargs["static_argnum"] = static_argnum + if kwarg_only_names: + # pyrefly: ignore [unsupported-operation] + kwargs["static_kwargkey"] = kwarg_only_names + + # pyrefly: ignore [bad-argument-type] + return RuntimeSchemaInfo(**kwargs) + + +from torch.utils._pytree import tree_any, tree_flatten, tree_map, tree_map_only + + +if TYPE_CHECKING: + from collections.abc import Sequence + + from torch._ops import OpOverload + from torch.distributed.tensor._sharding_prop import ShardingPropagator + + +def _extract_input_specs(op_schema: OpSchema) -> tuple[DTensorSpec | object, ...]: + return op_schema.args_schema + tuple(op_schema.kwargs_schema.values()) + + +class PlacementTrackingMode(TorchDispatchMode): + """ + TorchDispatchMode that tracks DTensor placements through op execution. + + Used during decomposition tracing: intercepts each op, propagates sharding + via the ShardingPropagator, and records output placements on the result tensors. + """ + + def __init__(self, sharding_prop: ShardingPropagator, mesh: DeviceMesh): + super().__init__() + self.sharding_prop = sharding_prop + self.mesh = mesh + + def __torch_dispatch__(self, func, types, args=(), kwargs=None): + args_schema, kwargs_schema = tree_map( + lambda x: getattr(x, "_spec", x) if isinstance(x, torch.Tensor) else x, + (args, kwargs or {}), + ) + + if not tree_any( + lambda x: isinstance(x, DTensorSpec), (args_schema, kwargs_schema) + ): + raise NotImplementedError(f"No DTensorSpec found in args/kwargs for {func}") + + # Set schema_info so the LRU cache key includes static args + op_schema = OpSchema(func, args_schema, kwargs_schema) + schema_info = self.sharding_prop.op_to_schema_info.get(func) + if schema_info is None: + schema_info = ( + self.sharding_prop.op_to_schema_info_for_single_dim_strategy.get(func) + ) + if schema_info is not None: + op_schema.schema_info = schema_info + op_schema._recompute_comparison_key() + + if _are_we_tracing(): + output_sharding = self.sharding_prop.propagate_op_sharding_non_cached( + op_schema + ) + else: + output_sharding = self.sharding_prop.propagate_op_sharding(op_schema) + + if ( + output_sharding.needs_redistribute # pyrefly: ignore [missing-attribute] + and ( + redistribute_schema + := output_sharding.redistribute_schema # pyrefly: ignore [missing-attribute] + ) + is not None + ): + # a pure .needs_redistribute check is too broad; we want to ban redistribution, + # but this flag is set for view ops that convert global shape -> local shape args. + # During decomposition tracing on meta tensors at global shape, the shape adjustment + # is irrelevant — only reject true redistribution. + for orig, desired in zip( + op_schema.args_spec, + redistribute_schema.args_spec, # pyrefly: ignore [missing-attribute] + ): + if orig.placements != desired.placements: + raise RuntimeError( + f"Decomposition requires redistribution for {func}" + ) + + out = func(*args, **kwargs) + # pyrefly: ignore [missing-attribute] + self._record_output_specs(out, output_sharding.output_spec) + return out + + def _record_output_specs(self, output: Any, output_spec: DTensorSpec | Any) -> None: + if isinstance(output, torch.Tensor) and output_spec is not None: + output._spec = output_spec # pyrefly: ignore [missing-attribute] + elif isinstance(output, (tuple, list)) and isinstance( + output_spec, (tuple, list) + ): + for t, s in zip(output, output_spec): + self._record_output_specs(t, s) + + +class DecompShardingStrategy: + """ + Generates sharding strategies for ops by tracing through their decompositions. + + For each candidate input placement combination, runs the decomposition on meta + tensors under PlacementTrackingMode to determine the output placement. These + single-dimension strategies are then expanded to the full mesh. + """ + + def __init__(self, sharding_prop: ShardingPropagator): + self.sharding_prop = sharding_prop + # Cache fake meshes per device type to avoid repeated allocation. + # A fake size-1 mesh ensures identical strategy computation across all ranks + # during decomposition tracing, avoiding potential SPMD divergence. + # False negatives are avoided (all sizes % 1 == 0), while false positives + # are caught on expansion to the real, multi-dim device mesh. + self._fake_meshes: dict[str, DeviceMesh] = {} + + def _get_fake_mesh(self, device_type: str) -> DeviceMesh: + fake_mesh = self._fake_meshes.get(device_type) + if fake_mesh is None: + fake_mesh = DeviceMesh(device_type, [0], _init_backend=False, _rank=0) + self._fake_meshes[device_type] = fake_mesh + return fake_mesh + + @staticmethod + def has_decomp(op: OpOverload) -> bool: + # Check if op has a decomposition (explicit or CIA) + return op in decomposition_table or op._can_decompose() + + def ensure_schema_info(self, op: OpOverload) -> None: + """ + Register schema_info for decomposition op on first invocation. + Needed for correct shard prop cache key. + """ + if op not in self.sharding_prop.op_to_schema_info: + schema_info = _infer_schema_info_from_op(op) + self.sharding_prop.op_to_schema_info[op] = schema_info + + def propagate_strategy( + self, + op_schema: OpSchema, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None = None, + ) -> OpStrategy | None: + if not tree_any( + lambda x: isinstance(x, DTensorSpec), + (op_schema.args_schema, op_schema.kwargs_schema), + ): + return None + + candidate_placements = self._get_candidate_placements(op_schema) + mesh = try_find_mesh_from_args( + op_schema.op, + op_schema.args_schema + tuple(op_schema.kwargs_schema.values()), + ) + + fake_mesh = self._get_fake_mesh(mesh.device_type) + single_dim_strategies = [] + output_placements: list[Placement | tuple[Placement, ...]] = [] + for input_placements in candidate_placements: + try: + output = self._propagate_through_decomp( + op_schema, + input_placements, + fake_mesh, + ) + except NotImplementedError: + return None + except GuardOnDataDependentSymNode: + return None + except (RuntimeError, KeyError, IndexError): + # TODO(pianpwk): RuntimeError is raised when redistribution is detected; switch to a custom error type + # Runtime/KeyError/IndexError can also occur in view ops + continue + + output_placements = ( + [output] if not isinstance(output, tuple) else list(output) + ) + single_dim_strategies.append(output_placements + list(input_placements)) + + if not single_dim_strategies: + raise AssertionError( + "Sharding propagation should have produced at least Replicate() strategy" + ) + + n_outputs = len(output_placements) + strategy_schema = self.sharding_prop._wrap_with_op_strategy(op_schema) + # Import here to avoid circular import at module load time + from torch.distributed.tensor._ops.utils import expand_to_full_mesh_op_strategy + + return expand_to_full_mesh_op_strategy( + mesh, + strategy_schema, + single_dim_strategies, + input_index=n_outputs, + output_tensor_meta=output_tensor_meta, + ) + + def _propagate_through_decomp( + self, + op_schema: OpSchema, + placement: tuple[Placement | None], + mesh: DeviceMesh, + ) -> Placement | tuple[Placement, ...]: + op = op_schema.op + if op in decomposition_table: + decomp_fn = decomposition_table[op] + elif op._can_decompose(): + decomp_fn = op.decompose + else: + raise NotImplementedError(f"No decomposition found for {op}") + + placement_iter = iter(placement) + + def to_meta(x): + p = next(placement_iter) + if isinstance(x, DTensorSpec): + # pyrefly: ignore [missing-attribute] + meta = torch.empty(x.shape, dtype=x.tensor_meta.dtype, device="meta") + # pyrefly: ignore [missing-attribute] + meta._spec = DTensorSpec(mesh, (p,), tensor_meta=x.tensor_meta) + return meta + return x + + # Disable LocalTensorMode during decomposition tracing to prevent + # interference with meta tensor operations + from torch.distributed._local_tensor import maybe_disable_local_tensor_mode + + with maybe_disable_local_tensor_mode(): + # Create meta tensors and run decomposition outside LocalTensorMode + args_meta = tree_map(to_meta, op_schema.args_schema) + kwargs_meta = tree_map(to_meta, op_schema.kwargs_schema) + + with PlacementTrackingMode(self.sharding_prop, mesh): + output = decomp_fn(*args_meta, **kwargs_meta) + + def get_placement(t): + if isinstance(t, torch.Tensor): + spec = getattr(t, "_spec", None) + return spec.placements[0] if spec else None + return None + + result = tree_map(get_placement, output) + if isinstance(result, (tuple, list)): + flat = [p for p in result if p is not None] + return flat[0] if len(flat) == 1 else tuple(flat) + return result + + @staticmethod + def _get_candidate_placements( + op_schema: OpSchema, + ) -> list[tuple[Placement | None]]: + tensor_specs = _extract_input_specs(op_schema) + flat_specs, _ = tree_flatten(list(tensor_specs)) + + # Step 1: Collect unique placements across all DTensorSpec inputs + all_placements: set[Placement] = {Replicate()} + tree_map_only( + DTensorSpec, + lambda spec: all_placements.update(spec.placements), + flat_specs, + ) + + # Step 2: For each input, use the placement set, but expand Shard/StridedShard to all tensor dims + candidates: list[list[Placement | None]] = [] + for spec in flat_specs: + if not isinstance(spec, DTensorSpec): + candidates.append([None]) + else: + options = set(all_placements) + for p in all_placements: + if isinstance(p, _StridedShard): + options |= { + _StridedShard(i, split_factor=p.split_factor) + for i in range(spec.ndim) + } + elif isinstance(p, Shard): + options |= {Shard(i) for i in range(spec.ndim)} + candidates.append(list(options)) + + # pyrefly: ignore [bad-argument-type, no-matching-overload] + return list(itertools.product(*candidates)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dispatch.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dispatch.py new file mode 100644 index 0000000000000000000000000000000000000000..d979aef58f7e4d6f2b7fbf325d04d0fb93d73ec3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dispatch.py @@ -0,0 +1,858 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import contextlib +import logging +import warnings +from collections.abc import Sequence +from typing import cast + +import torch +import torch.distributed as dist +import torch.distributed.tensor._api as dtensor +import torch.distributed.tensor._random as random +from torch._library.utils import fill_defaults +from torch._logging import LazyString +from torch._prims.rng_prims import run_dtensor_rng_op +from torch.distributed._functional_collectives import _are_we_tracing +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._nonlinear_redux import argminmax_handler +from torch.distributed.tensor._op_schema import ( + OpInfo, + OpSchema, + OutputSharding, + OutputSpecType, +) +from torch.distributed.tensor._random import is_rng_supported_mesh +from torch.distributed.tensor._redistribute import redistribute_local_tensor +from torch.distributed.tensor._sharding_prop import ShardingPropagator +from torch.distributed.tensor._tp_conv import ( + convolution_backward_handler, + convolution_handler, +) +from torch.distributed.tensor._utils import ( + _format_implicit_redistribution_msg, + ExplicitRedistributionContext, + try_find_mesh_from_args, +) +from torch.distributed.tensor.placement_types import Partial, Placement, Replicate +from torch.utils._debug_mode import get_active_debug_mode +from torch.utils._python_dispatch import return_and_correct_aliasing + + +try: + from torch.utils import _cxx_pytree as pytree +except ImportError: + from torch.utils import _pytree as pytree # type: ignore[no-redef] + +aten = torch.ops.aten +logger = logging.getLogger(__name__) + +# The C++ DTensor dispatch fast path caches whether debug logging is +# enabled. Wrap setLevel so the cached flag is reset automatically. +_orig_setLevel = logger.setLevel + + +def _setLevel_and_reinit(level: int) -> None: + _orig_setLevel(level) + torch._C._reinit_DTensor_dispatch_logger() + + +logger.setLevel = _setLevel_and_reinit # type: ignore[method-assign] + + +def _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_spec: object, +) -> contextlib.AbstractContextManager[None]: + if not _are_we_tracing() or output_spec is None: + return contextlib.nullcontext() + + from torch.fx.experimental.symbolic_shapes import ( + _ignore_fresh_unbacked_symbols_tls_context, + ) + + # DTensor exposes the global output contract through output_spec. Fresh + # local fake symbols created only while redistributing or computing the + # per-rank tensor are implementation details and may not appear in the + # returned DTensor object. + return _ignore_fresh_unbacked_symbols_tls_context() + + +def _as_strided_permutation( + tensor: "dtensor.DTensor", + size: Sequence[int], + stride: Sequence[int], +) -> list[int] | None: + """Return dims such that ``tensor.permute(dims)`` has the requested layout. + + AOTAutograd regenerates an output that aliases an input by calling + as_strided on the input, which a sharded tensor cannot answer in general -- + it has no single global storage to index into. A permutation of the base + dims (transpose, t, permute, Tensor.T) is the exception: it maps onto + aten.permute, which does have a sharding strategy. + + Returns None when the request is not a permutation, or when two dims have + the same (size, stride) and the permutation is therefore ambiguous -- the + layouts would be identical but the placements need not be. + """ + if len(size) != tensor.dim(): + return None + + base = list(zip(tensor.size(), tensor.stride())) + dims = [] + used = [False] * len(base) + for target in zip(size, stride): + candidates = [ + i for i, b in enumerate(base) if not used[i] and tuple(b) == tuple(target) + ] + if len(candidates) != 1: + return None + used[candidates[0]] = True + dims.append(candidates[0]) + return dims + + +def as_strided_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +): + args, kwargs = fill_defaults(op_call._schema, args, kwargs) + if kwargs: + raise AssertionError + tensor, size, stride, storage_offset = args + if storage_offset is None or tensor.storage_offset() == storage_offset: + if tensor.size() == tuple(size) and tensor.stride() == tuple(stride): + return torch.ops.aten.alias.default(tensor) + dims = _as_strided_permutation(tensor, size, stride) + if dims is not None: + return torch.ops.aten.permute.default(tensor, dims) + raise RuntimeError("as_strided not supported with DTensor") + + +def is_same_size_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> bool: + lhs = cast(torch.Tensor, args[0]) + rhs = cast(torch.Tensor, args[1]) + return lhs.shape == rhs.shape + + +def is_pinned_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> bool: + tensor = cast(dtensor.DTensor, args[0]) + return tensor._local_tensor.is_pinned() + + +def found_inf_reduce_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> None: + op_info = dtensor.DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + local_tensor_args = pytree.tree_unflatten( + cast(list[object], op_info.local_args), + op_info.args_tree_spec, # type: ignore[arg-type] + ) + local_tensor_args = cast(tuple[object, ...], local_tensor_args) + op_call(*local_tensor_args, **op_info.local_kwargs) + + grad_dtensor = cast(list[dtensor.DTensor], args[0])[0] + grad_placements = grad_dtensor.placements + mesh = grad_dtensor.device_mesh + + found_inf_placements: list[Placement] = [] + for placement in grad_placements: + if isinstance(placement, Replicate): + found_inf_placements.append(placement) + else: + found_inf_placements.append(Partial("max")) + + target_tensor = cast(torch.Tensor, args[1]) + spec = DTensorSpec( + mesh=mesh, + placements=tuple(found_inf_placements), + tensor_meta=TensorMeta( + shape=target_tensor.size(), + stride=target_tensor.stride(), + dtype=target_tensor.dtype, + ), + ) + # pyrefly: ignore [bad-argument-type] + found_inf_dtensor = dtensor.DTensor( + local_tensor=target_tensor, # pyrefly: ignore [unexpected-keyword] + spec=spec, # pyrefly: ignore [unexpected-keyword] + requires_grad=False, # pyrefly: ignore [unexpected-keyword] + ) + found_inf = found_inf_dtensor.full_tensor() + target_tensor.copy_(found_inf) + + +class OpDispatcher: + """ + Op dispatching class instance to handle args/kwargs pre-processing (un-wrapping), sharding + propagation, redistribute local args, local compute, and post-processing (re-wrapping). It + also handles any op specific logic if necessary. + + NOTE: Given the runtime overhead of Tensor subclass (__torch_dispatch__), the OpDispatcher + is designed to minimize the CPU overhead by using the tricks of proper unflattening, faster + pytree if needed, and leveraging various caching mechanisms implemented in the sharding + propagation and redistribute modules. The CPU overhead is critical to eager mode performance, + one need to carefully measure the CPU overhead when making significant changes to the + OpDispatcher and ShardingPropagator. + """ + + def __init__(self) -> None: + self.sharding_propagator = ShardingPropagator() + # NOTE: must stay in sync with is_random_op in + # torch/csrc/autograd/python_variable.cpp + self._random_ops = { + aten.native_dropout.default, + aten.normal_.default, + aten.rand.default, + aten.rand_like.default, + aten.randn.default, + aten.randn_like.default, + aten.randint_like.default, + aten.randint_like.low_dtype, + aten.randint_like.low_dtype_out, + aten.uniform_.default, + aten.bernoulli.default, + aten.bernoulli_.float, + } + self._custom_op_handlers = { + aten.is_same_size.default: is_same_size_handler, + aten.is_pinned.default: is_pinned_handler, + aten.convolution.default: convolution_handler, + aten.convolution_backward.default: convolution_backward_handler, + aten._amp_foreach_non_finite_check_and_unscale_.default: found_inf_reduce_handler, + aten.as_strided.default: as_strided_handler, + aten.argmin.default: argminmax_handler, + aten.argmax.default: argminmax_handler, + } + + # ******************************************************************************************** + # def dispatch(...) + # + # NOTE: this class no longer contains the top-level dispatch entrypoint! + # See #167051 for details + # + # The entrypoint has been moved to C++, and it handles common cases and then calls back into + # OpDispatcher python to handle corner cases. + # See dispatchDTensorOp() defined in python_variable.cpp and called from python_arg_parser.cpp + # ******************************************************************************************** + + # This flag is used internally to control whether we treat the torch.Tensor(non-DTensor) + # as implicitly replicated or we throw error to user. + # NOTE: It is EXTREMELY UNSAFE to turn this flag on by default so we intentionally leave + # it as False by default. + @property + def _allow_implicit_replication(self) -> bool: + return torch._C._get_dtensor_allow_implicit_replication() + + @_allow_implicit_replication.setter + def _allow_implicit_replication(self, value: bool) -> None: + return torch._C._set_dtensor_allow_implicit_replication(value) + + def _propagate_op_sharding_dispatch_slow_path( + self, + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], + op_info: OpInfo, + # The logic here is a bit messy. There are several reasons why the + # C++ fastpath may have bailed out. If we just cache missed, we will + # come here because we need to actually calculate the real thing. + # There's no need to have a SECOND Python cache lookup; the C++ native + # cache completely subsumes it. But sometimes, we will have failed + # to compute the cache key in C++ entirely. In this case, we DO need + # to do a cache lookup in Python, as the missing cache key in C++ + # means we don't have access to it all. Furthermore, without duping + # this function, we need to do the try_cache test inside of the + # try-except block so that either case hits the inference mode / + # exception rewrapping case. + # + # This should be cleaned up. First, ensuring the C++ codepath can + # always compute a key will be a big help. Second, we should properly + # fastpath inference mode composite implicit autograd so that you + # don't have to throw an exception even in "fastpath". + try_cache: bool, + ) -> object: + # NOTE: schema should always be populated when calling this function, + # as it's only called from C++ after unwrap_to_op_info (create_schema=True). + # See dispatchDTensorOp in python_variable.cpp line 1453-1460. + if op_info.schema is None: + raise AssertionError( + "op_info.schema should not be None in sharding propagation. " + "This function should only be called after unwrap_to_op_info." + ) + try: + # We have basically inlined propagate() here, but WITHOUT the + # output_sharding assignment + if try_cache and not _are_we_tracing(): + result = self.sharding_propagator.propagate_op_sharding(op_info.schema) + else: + result = self.sharding_propagator.propagate_op_sharding_non_cached( + op_info.schema + ) + if logger.handlers and logger.isEnabledFor(logging.DEBUG): + logger.debug( + "sharding_prop MISS (C++ fast path): %s -> %s", + op_info.schema, + # pyrefly: ignore [missing-attribute] + result.output_spec, + ) + return result + except NotImplementedError: + if torch._C._dispatch_has_kernel_for_dispatch_key( + op_call.name(), torch._C.DispatchKey.CompositeImplicitAutograd + ): + # When running under inference mode, CompositeImplicitAutograd ops show up in __torch_dispatch__, + # so we manually decompose them, here + out = op_call.decompose(*args, **kwargs) + if out is NotImplemented: + raise AssertionError from None + return out + else: + raise + except Exception as e: + raise RuntimeError( + f"{e}\n\nSharding propagation failed for {op_info.schema or op_call}" + ) from e + + def _dispatch_get_local_results_slow_path( + self, + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + op_info: OpInfo, + ) -> object: + output_sharding = op_info.output_sharding + if output_sharding is None: + raise AssertionError("output sharding should not be None") + if op_info is None: + raise AssertionError("op_info should never be None") + + # Record output placements for debugging + debug_mode = get_active_debug_mode() + if debug_mode is not None and output_sharding.output_spec is not None: + debug_mode.record_output_placements(output_sharding.output_spec) + + mesh = op_info.compute_mesh + participating = mesh._is_current_rank_part_of_mesh() + local_results = None + if participating: + # computation that happens in the current rank of the mesh, normal case + if output_sharding.needs_redistribute: + # If sharding propagation decision needs redistribute, perform redistribute + # on args first, which could potentially modify args (i.e. allgather certain arg) + if output_sharding.redistribute_schema is None: + raise AssertionError + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + self.redistribute_local_args( + op_info, + output_sharding.redistribute_schema, + output_sharding.use_val_from_redistribute_schema, + ) + + local_tensor_args = ( + pytree.tree_unflatten( + cast(list[object], op_info.local_args), + # pyrefly: ignore [bad-argument-type] + op_info.args_tree_spec, + ) + if op_info.args_tree_spec + else op_info.local_args + ) + + # run local op computation with potentially modified args/kwargs + local_tensor_args = cast(tuple[object, ...], local_tensor_args) + if op_call in self._random_ops: + if not random._rng_tracker and is_rng_supported_mesh(mesh): + # Default to `OffsetBasedRNGTracker` if the parallelism API did not already construct one + # Skip RNG state sync during tracing to avoid lazily initializing real RNG state under fake mode. + run_state_sync = not _are_we_tracing() + if not run_state_sync: + logger.info( + "DTensor RNG tracker is being lazily initialized during tracing. " + "RNG states may not be synchronized across ranks, which can lead " + "to silent incorrectness. Please call `torch.manual_seed()` with " + "the same seed on all ranks before compiling DTensor random ops.", + stacklevel=2, + ) + random._rng_tracker = random.OffsetBasedRNGTracker( + mesh, run_state_sync + ) + + first_arg, first_local_arg = ( + cast(dtensor.DTensor, args[0]), + cast(torch.Tensor, local_tensor_args[0]), + ) + + # If the user provided a generator, we hook it up to our RNG manager, but we also pop it from kwargs + # so the op_call does not directly use it (we want op_call to fall back to the 'default' which is + # our RNG manager) + maybe_user_generator = op_info.local_kwargs.pop("generator", None) + if not ( + maybe_user_generator is None + or isinstance(maybe_user_generator, torch.Generator) + ): + raise AssertionError + + if ( + random._rng_tracker + and not first_local_arg.is_meta + and random._rng_tracker.distribute_region_enabled + ): + if ( + maybe_user_generator is not None + or first_local_arg.device.type != "cuda" + or ( + not _are_we_tracing() + and type(first_local_arg) is not torch.Tensor + ) + ): + with random._rng_tracker._distribute_region( + first_arg._spec, generator=maybe_user_generator + ): + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + local_results = op_call( + *local_tensor_args, **op_info.local_kwargs + ) + else: + # CUDA device without user generator, use HOP for traceability + if not isinstance( + random._rng_tracker, random.OffsetBasedRNGTracker + ): + raise AssertionError + start_offset_incr, end_offset_incr = ( + random._rng_tracker._compute_rng_offsets(first_arg._spec) + ) + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + local_results = run_dtensor_rng_op( + start_offset_incr, + end_offset_incr, + op_call, + *local_tensor_args, + **op_info.local_kwargs, + ) + else: + # No rng_tracker, meta tensor, or distribute_region disabled + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + local_results = op_call( + *local_tensor_args, **op_info.local_kwargs + ) + else: + # normal case, run local sharded op computation + if ( + output_sharding.needs_redistribute + and output_sharding.redistribute_schema is not None + and output_sharding.redistribute_schema.op != op_call + ): + # Op was rewritten (e.g., squeeze.default → squeeze.dims) + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + local_results = output_sharding.redistribute_schema.op( + *local_tensor_args, **op_info.local_kwargs + ) + else: + with _ignore_fresh_unbacked_symbols_for_dtensor_tracing( + output_sharding.output_spec + ): + local_results = op_call( + *local_tensor_args, **op_info.local_kwargs + ) + + else: + # For a non-participating device (happens on rank that does not belong to + # the device mesh), we do: + # 1. if the return type is scalar, set the local result to None. + # 2. if the return type is Tensor or List[Tensor], return empty + # tensor(s) with correct dtype. + spec = output_sharding.output_spec + ret_list = op_call._schema.returns + + if spec is None: + # For a scalar return type, the non-participating device has None + # as its local result + local_results = None + else: + + def default_tensor(spec: DTensorSpec) -> torch.Tensor: + if spec.tensor_meta is not None: + shape = spec.tensor_meta.shape + dtype = spec.tensor_meta.dtype + if len(shape) == 0: + # scalar tensor + return torch.zeros((), dtype=dtype) + else: + # non-scalar tensor + return torch.tensor([], dtype=dtype) + else: + raise RuntimeError(f"{spec} has no tensor metadata.") + + if isinstance(spec, DTensorSpec): + # return a Tensor value + local_results = default_tensor(spec) + elif isinstance(spec, Sequence): + # return a List[Tensor] value + local_results = [ + default_tensor(s) if s is not None else None for s in spec + ] + if not isinstance(local_results, list): + raise AssertionError + if None in local_results: + ret_type = str(ret_list[0].type) + raise NotImplementedError( + f"return type {ret_type} in DTensor op is not supported" + ) + return local_results + + def _dispatch_fast_path_python_tail( + self, + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], + compute_mesh: DeviceMesh, + output_sharding: OutputSharding, + local_results: object, + participating: bool, + is_inplace_op: bool, + is_out_variant_op: bool, + ) -> object: + """ + Tail of main dispatching logic, called from C++ fast path. + """ + + # Record output placements for debugging + debug_mode = get_active_debug_mode() + if debug_mode is not None and output_sharding.output_spec is not None: + debug_mode.record_output_placements(output_sharding.output_spec) + + if output_sharding.output_spec is None: + if op_call == aten.equal.default: + # The output of the equal op is a bool, by converting it into + # a single value tensor, we can use all-reduce with min reduce op + # to simulate logical and. + if not (local_results is None or isinstance(local_results, bool)): + raise AssertionError + r = torch.tensor( + int(local_results) if local_results is not None else 1, + device=compute_mesh.device_type, + ) + dist.all_reduce(r, op=dist.ReduceOp.MIN) + local_results = bool(r.item()) + + if is_inplace_op: + # inplace op should return self instead of re-wrapping + if output_sharding.output_spec is not None: + output_spec = output_sharding.output_spec + if not isinstance(output_spec, DTensorSpec): + raise AssertionError + if not isinstance(args[0], dtensor.DTensor): + raise AssertionError + + # Fast path: placements unchanged (common case: add_, mul_, etc.) + if args[0]._spec.placements == output_spec.placements: + return args[0] + + # Placement reindexed (e.g. squeeze_ removing a non-sharded + # dim: Shard(1) → Shard(0)). No redistribution — the local + # tensor data is unchanged, only dim indices shift. + # strict_view=True in sharding prop prevents the illegal + # case (squeezing a sharded dim) from reaching here. + args[0]._spec = output_spec + return return_and_correct_aliasing(op_call, args, kwargs, args[0]) + else: + return None + elif is_out_variant_op: + # out variant could possibly have multiple out args (i.e. lu_unpack.out) + output_specs = ( + (output_sharding.output_spec,) + if not isinstance(output_sharding.output_spec, tuple) + else output_sharding.output_spec + ) + out_dts = [] + spec_idx = 0 + for argument in op_call._schema.arguments: + if argument.is_out: + out_dt = cast(dtensor.DTensor, kwargs[argument.name]) + out_dt._spec = cast(DTensorSpec, output_specs[spec_idx]) + out_dts.append(out_dt) + spec_idx += 1 + + if len(out_dts) < 1: + raise AssertionError("out variant should have at least one out arg") + return tuple(out_dts) if len(out_dts) > 1 else out_dts[0] + else: + if op_call != aten.equal.default: + raise AssertionError(op_call) + ret = self.wrap(local_results, output_sharding.output_spec) # type: ignore[possibly-undefined] + if participating and op_call._schema._is_view_op(): + return return_and_correct_aliasing(op_call, args, kwargs, ret) + else: + return ret + + @staticmethod + def redistribute_local_args( + op_info: OpInfo, + suggested_input_schema: OpSchema, + use_val_from_redistribute_schema: bool, + ) -> None: + debug_mode = get_active_debug_mode() + + # NOTE: it's very rare that we need to reshard kwargs so we intentionally skip it + if op_info.args_tree_spec is not None: + flatten_args_schema_to_reshard = tuple( + pytree.tree_leaves(suggested_input_schema.args_schema) + ) + else: + flatten_args_schema_to_reshard = suggested_input_schema.args_schema + + new_local_args: list[object] = [] + for i, arg_spec in enumerate(op_info.flat_args_schema): + reshard_arg_spec = flatten_args_schema_to_reshard[i] + if isinstance(arg_spec, DTensorSpec): + local_tensor = cast(torch.Tensor, op_info.local_args[i]) + if arg_spec != reshard_arg_spec: + redistribute_context = ( + debug_mode.record_redistribute_calls( # type: ignore[union-attr] + i, arg_spec, reshard_arg_spec + ) + if debug_mode is not None + else contextlib.nullcontext() + ) + + ExplicitRedistributionContext.observe_redistribution( + arg_spec, + # pyrefly: ignore [bad-argument-type] + reshard_arg_spec, + LazyString( + _format_implicit_redistribution_msg, + op_info.schema or suggested_input_schema.op, + ), + ) + with redistribute_context: + resharded_local_tensor = redistribute_local_tensor( + local_tensor, + arg_spec, + # pyrefly: ignore [bad-argument-type] + reshard_arg_spec, + ) + new_local_args.append(resharded_local_tensor) + else: + new_local_args.append(local_tensor) + else: + if use_val_from_redistribute_schema: + # args can be updated for view related ops, we refer to the + # update in redistribute_schema. + new_local_args.append(reshard_arg_spec) + else: + new_local_args.append(arg_spec) + + # Append extra non-tensor args from rewritten schema (e.g., dims tuple). + if use_val_from_redistribute_schema: + for i in range( + len(op_info.flat_args_schema), len(flatten_args_schema_to_reshard) + ): + new_local_args.append(flatten_args_schema_to_reshard[i]) + + op_info.local_args = tuple(new_local_args) + + def unwrap_to_op_info( + self, + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], + ) -> OpInfo: + return self._unwrap_to_op_info_impl(op_call, args, kwargs, True) + + def _unwrap_to_op_info_impl( + self, + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], + create_schema: bool, + ) -> OpInfo: + # get runtime schema info to determine whether to use pytree to flatten inputs + runtime_schema_info = self.sharding_propagator.op_to_schema_info.get( + op_call, None + ) + if runtime_schema_info is None: + runtime_schema_info = ( + self.sharding_propagator.op_to_schema_info_for_single_dim_strategy.get( + op_call, None + ) + ) + + # Auto-detect needs_pytree if any arg is a list/tuple containing tensors + def _contains_tensor(arg: object) -> bool: + if isinstance(arg, (list, tuple)): + return any(isinstance(item, torch.Tensor) for item in arg) + return False + + needs_pytree = ( + runtime_schema_info is not None and runtime_schema_info.needs_pytree + ) or any(_contains_tensor(arg) for arg in args) + + if needs_pytree: + # flatten args/kwargs when op says necessary or args contain lists/tuples + tree_args, args_spec = pytree.tree_flatten(args) + args_list: Sequence[object] = tree_args + else: + args_list, args_spec = args, None + + args_schema: list[object] = [] + kwargs_schema: dict[str, object] = {} + local_args: list[object] = [] + local_kwargs: dict[str, object] = {} + compute_mesh: DeviceMesh | None = None + + for arg in args_list: + if isinstance(arg, dtensor.DTensor): + local_args.append(arg._local_tensor) + args_schema.append(arg._spec) + if compute_mesh is None: + # record the first compute device mesh from args + compute_mesh = arg.device_mesh + elif isinstance(arg, torch.Tensor): + compute_mesh = compute_mesh or try_find_mesh_from_args( + op_call, args_list + ) + args_schema.append( + self._try_replicate_spec_for_scalar_tensor( + op_call, arg, compute_mesh + ) + ) + local_args.append(arg) + else: + # non DTensor/Tensor args (i.e. int/float/bool), just add to args_schema/local_args + args_schema.append(arg) + local_args.append(arg) + + for k, v in kwargs.items(): + if isinstance(v, dtensor.DTensor): + local_kwargs[k] = v._local_tensor + kwargs_schema[k] = v._spec + if compute_mesh is None: + # record the first compute device mesh from kwargs + compute_mesh = v.device_mesh + elif isinstance(v, torch.Tensor): + compute_mesh = compute_mesh or try_find_mesh_from_args( + op_call, args_list + ) + kwargs_schema[k] = self._try_replicate_spec_for_scalar_tensor( + op_call, + v, + compute_mesh, + ) + local_kwargs[k] = v + else: + # non DTensor/Tensor args (i.e. int/float/bool), just add to args_schema/local_args + kwargs_schema[k] = v + local_kwargs[k] = v + + if compute_mesh is None: + raise AssertionError( + f"found no DeviceMesh from dtensor args for {op_call}!" + ) + op_info = OpInfo( + compute_mesh, + OpSchema( + op_call, + ( + # pyrefly: ignore [bad-argument-type] + pytree.tree_unflatten(args_schema, args_spec) + if args_spec + else tuple(args_schema) + ), + kwargs_schema, + schema_info=runtime_schema_info, + ) + if create_schema + else None, # type: ignore[arg-type] + args_schema, + tuple(local_args), + local_kwargs, + args_spec, + ) + return op_info + + @staticmethod + def wrap(res: object, spec: OutputSpecType) -> object: + if isinstance(res, torch.Tensor): + if spec is not None: + if not isinstance(spec, DTensorSpec): + raise AssertionError( + f"output spec does not match with output! Expected DTensorSpec, got {spec}." + ) + # pyrefly: ignore [bad-argument-type, bad-argument-count, unexpected-keyword] + return dtensor.DTensor(res, spec, requires_grad=res.requires_grad) + else: + # if output does not have a DTensorSpec due to specific ops, it must be a scalar tensor + if res.ndim != 0: + raise AssertionError("output tensor should be scalar!") + return res + elif isinstance(res, (list, tuple)): + if not (spec is not None and isinstance(spec, (list, tuple))): + raise AssertionError( + f"output spec does not match with output! Expected list/tuple, got {spec}." + ) + res_list = [] + for e, s in zip(res, spec): + # pyrefly: ignore [bad-argument-type] + res_list.append(OpDispatcher.wrap(e, s)) + + return tuple(res_list) if isinstance(res, tuple) else res_list + else: + # if the res contains only non tensor values (i.e. int/float/none), we simply return it + # without rewrapping to DTensor. + return res + + def _try_replicate_spec_for_scalar_tensor( + self, + op_call: torch._ops.OpOverload, + tensor_arg: torch.Tensor, + compute_mesh: DeviceMesh, + ) -> DTensorSpec: + # util function to produce a replicate spec for a scalar tensor arg/kwarg + if tensor_arg.numel() == 1 and tensor_arg.ndim == 1: + warnings.warn( + "Found a non-scalar tensor with numel=1 and ndim!=0, " + "we are implicitly creating a replicated DTensor for it. " + "However, please consider changing it to a scalar tensor " + "or explicitly create a DTensor under distributed environment.", + stacklevel=2, + ) + + if tensor_arg.numel() == 1 or self._allow_implicit_replication: + # scalar tensor can be safely treated as replicated + replication_spec = DTensorSpec( + compute_mesh, + (Replicate(),) * compute_mesh.ndim, + tensor_meta=TensorMeta( + shape=tensor_arg.shape, + stride=tensor_arg.stride(), + dtype=tensor_arg.dtype, + ), + ) + else: + raise RuntimeError( + f"{op_call}: got mixed torch.Tensor and DTensor, need to convert all" + " torch.Tensor to DTensor before calling distributed operators!" + " Please see https://docs.pytorch.org/docs/main/distributed.tensor.html#mixed-tensor-and-dtensor-operations" + " for more details." + ) + return replication_spec diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dtensor_spec.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dtensor_spec.py new file mode 100644 index 0000000000000000000000000000000000000000..77de53ead924ae4c53ab3a747915ca40b84c154b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_dtensor_spec.py @@ -0,0 +1,766 @@ +import hashlib +import itertools +import math +from collections import defaultdict +from dataclasses import dataclass +from typing import Any, cast, NamedTuple + +import torch +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _MaskPartial, + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) +from torch.utils._debug_mode import _stringify_shape +from torch.utils._dtype_abbrs import dtype_abbrs + + +# Defined here (not in placement_types.py) because decoding split_factor into +# a shard order is a DTensorSpec concern — placement_types doesn't know about +# shard orders. +class _StridedShardNotDecodableError(ValueError): + """Raised when _StridedShard split_factor cannot be decoded into a shard order.""" + + +class ShardOrderEntry(NamedTuple): + """ + Represents how a single tensor dimension is sharded across mesh dimensions. + + Attributes: + tensor_dim: The tensor dimension being sharded (e.g., 0, 1, 2 for a 3D tensor). + mesh_dims: Tuple of mesh dimensions across which this tensor dimension is sharded, + in execution order. The first mesh dim is applied first, second is applied + second, etc. This tuple is guaranteed to be non-empty. + + Examples: + >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_DISTRIBUTED) + >>> # Tensor dim 1 sharded across mesh dim 2, then mesh dim 0 + >>> ShardOrderEntry(tensor_dim=1, mesh_dims=(2, 0)) + + >>> # Tensor dim 0 sharded only on mesh dim 1 + >>> ShardOrderEntry(tensor_dim=0, mesh_dims=(1,)) + """ + + tensor_dim: int + mesh_dims: tuple[int, ...] # guaranteed to be non-empty + + +# Type alias for the complete shard order specification +# A tuple of ShardOrderEntry, one per sharded tensor dimension +# +# Example: +# shard_order = ( +# ShardOrderEntry(tensor_dim=0, mesh_dims=(1,)), +# ShardOrderEntry(tensor_dim=2, mesh_dims=(0, 3)), +# ) +# This means: +# - Tensor dimension 0 is sharded on mesh dimension 1 +# - Tensor dimension 2 is sharded on mesh dimension 0 first, then mesh dimension 3 +ShardOrder = tuple[ShardOrderEntry, ...] + + +class TensorMeta(NamedTuple): + # simple named tuple to represent tensor metadata + # intentionally to stay simple only for sharding + # propagation purposes. + shape: torch.Size + stride: tuple[int, ...] + dtype: torch.dtype + + +# used internally to propagate the placements +@dataclass +class DTensorSpec: + mesh: DeviceMesh + placements: tuple[Placement, ...] + + # tensor meta will only be set during sharding propagation + tensor_meta: TensorMeta | None = None + + # When a tensor dimension is sharded across multiple mesh axes, + # `shard_order` specifies the sequence in which these shardings are applied. + # This order determines how tensor shards are mapped and distributed across + # devices. + # + # Example: + # For a tensor of shape [8, 16] and a 3D device mesh, if dim 0 is sharded over + # mesh dim 1, and dim 1 is sharded over mesh dim 0 and then mesh dim 2, + # the shard_order would be: + # shard_order = ( + # ShardOrderEntry(tensor_dim=0, mesh_dims=(1,)), + # ShardOrderEntry(tensor_dim=1, mesh_dims=(0, 2)), + # ) + shard_order: ShardOrder = None # type: ignore[assignment] + + # When True, _StridedShard placements encode the shard order, and the + # shard_order field must be left as None (it will be derived on demand). + # Set explicitly to False to treat _StridedShard as a regular Shard, + # e.g., in view propagation. + use_strided_shard_as_shard_order: bool | None = None + + def __post_init__(self) -> None: + if not isinstance(self.placements, tuple): + self.placements = tuple(self.placements) + if self.use_strided_shard_as_shard_order is None: + if any(isinstance(p, _StridedShard) for p in self.placements): + self.use_strided_shard_as_shard_order = True + else: + self.use_strided_shard_as_shard_order = False + if self.use_strided_shard_as_shard_order: + if self.shard_order is not None: + raise ValueError( + "DTensorSpec doesn't allow specify shard_order when " + "use_strided_shard_as_shard_order is True. This may result " + "in conflicting shard order." + ) + else: + if self.shard_order is None: + self.shard_order = self.compute_default_shard_order(self.placements) + + self._hash: int | None = None + + @staticmethod + def _normalize_placements_into_shard_order( + placements: tuple[Placement, ...], + mesh: DeviceMesh, + use_strided_shard_as_shard_order: bool = True, + ) -> tuple[tuple[Placement, ...], ShardOrder]: + # If use_strided_shard_as_shard_order, it means the StridedShard/Shard + # combinations should be interpreted as shard order. + if use_strided_shard_as_shard_order: + # _StridedShard in placements, try check if it can be decoded as shard order + shard_order = DTensorSpec._maybe_convert_StridedShard_to_shard_order( + placements, mesh + ) + if shard_order is None: + raise _StridedShardNotDecodableError( + f"_StridedShard placements {placements} cannot be decoded " + "into a corresponding shard_order" + ) + normalized_placements = tuple( + [ + p if not isinstance(p, _StridedShard) else Shard(p.dim) + for p in placements + ] + ) + return normalized_placements, shard_order + else: + return placements, DTensorSpec.compute_default_shard_order(placements) + + @staticmethod + def compute_default_shard_order( + placements: tuple[Placement, ...], + ) -> ShardOrder: + """ + Compute the default shard order from placements. + + Returns a ShardOrder where each ShardOrderEntry maps a tensor dimension + to the mesh dimensions it's sharded on, in left-to-right order. + + Args: + placements: Tuple of Placement objects representing how a tensor is + distributed across mesh dimensions. + """ + # follow default left-to-right device order if shard_order is not specified + tensor_dim_to_mesh_dims: defaultdict[int, list[int]] = defaultdict(list) + mesh_ndim = len(placements) + for mesh_dim in range(mesh_ndim): + if _is_shard_like(placements[mesh_dim]): + placement = placements[mesh_dim] + shard_dim = placement.dim # pyrefly: ignore [missing-attribute] + if shard_dim < 0: + raise AssertionError( + f"Shard dim {shard_dim} in placements {placements} must be normalized" + ) + tensor_dim_to_mesh_dims[shard_dim].append(mesh_dim) + + # Convert dict into ShardOrderEntry tuples + default_shard_order = tuple( + ShardOrderEntry(tensor_dim=key, mesh_dims=tuple(value)) + for key, value in sorted(tensor_dim_to_mesh_dims.items()) + if value + ) + return default_shard_order + + @staticmethod + def _convert_shard_order_to_StridedShard( + shard_order: ShardOrder, placements: tuple[Placement, ...], mesh: DeviceMesh + ) -> tuple[Placement, ...]: + """ + Convert ShardOrder to placements with _StridedShard. + + This function converts a ShardOrder specification into a tuple of Placement objects, + using _StridedShard when a tensor dimension is sharded across multiple mesh dimensions + in a non-default order. The split_factor of each _StridedShard is determined by the + product of mesh dimension sizes that appear earlier in the shard order but later in + the placement tuple. + + Args: + shard_order: ShardOrder specification indicating which tensor dimensions are + sharded on which mesh dimensions and in what execution order. + placements: Tuple of Placement objects that does not contain _StridedShard. + mesh: DeviceMesh containing the size information for each mesh dimension. + + Returns: + Updated tuple of Placement objects with Shard or _StridedShard placements. + + Algorithm: + For each ShardOrderEntry in shard_order: + - For each mesh dimension in the entry's mesh_dims (in order): + - Calculate split_factor as the product of mesh sizes for all mesh dimensions + that appear: + 1. Earlier in the shard order (lower index in mesh_dims), and + 2. Later in the placement tuple (higher mesh dimension index) + - If split_factor == 1: use normal Shard + - Otherwise: use _StridedShard with the calculated split_factor + + Example: + >>> # xdoctest: +SKIP("Requires DeviceMesh") + >>> # Tensor dimension 0 sharded on mesh dims [2, 0, 1] in that order + >>> # mesh = DeviceMesh([4, 3, 2]) # sizes: mesh[0]=4, mesh[1]=3, mesh[2]=2 + >>> shard_order = (ShardOrderEntry(tensor_dim=0, mesh_dims=(2, 0, 1)),) + >>> placements = (Shard(0), Shard(0), Shard(0)) + >>> # For mesh_dim=2 (index 0 in mesh_dims): no earlier dims, split_factor=1 + >>> # -> placements[2] = Shard(0) + >>> # For mesh_dim=0 (index 1 in mesh_dims): mesh_dim=2 is earlier and has index 2>0 + >>> # -> split_factor = mesh.size(2) = 2 + >>> # -> placements[0] = _StridedShard(0, split_factor=2) + >>> # For mesh_dim=1 (index 2 in mesh_dims): mesh_dim=2 is earlier and has index 2>1 + >>> # -> split_factor = mesh.size(2) = 2 + >>> # -> placements[1] = _StridedShard(0, split_factor=2) + >>> # Result: (_StridedShard(0, sf=2), _StridedShard(0, sf=2), Shard(0)) + """ + placements_list = list(placements) + for entry in shard_order: + tensor_dim = entry.tensor_dim + mesh_dims = entry.mesh_dims + for idx in range(len(mesh_dims)): + # TODO(zpcore): split_factor from `view` and `shard order` + # should be able to be multiplied into one. Need to loosen the + # condition here. + mesh_dim = mesh_dims[idx] + if type(placements[mesh_dim]) is not Shard: + raise ValueError( + f"Only Shard placement can be converted to _StridedShard, " + f"found {placements[mesh_dim]} in {placements=}." + ) + split_factor = math.prod( + mesh.size(i) for i in mesh_dims[:idx] if i > mesh_dim + ) + if split_factor == 1: + # use normal Shard + placements_list[mesh_dim] = Shard(tensor_dim) + else: + placements_list[mesh_dim] = _StridedShard( + tensor_dim, split_factor=split_factor + ) + return tuple(placements_list) + + @staticmethod + def _maybe_convert_StridedShard_to_shard_order( + placements: tuple[Placement, ...], mesh: DeviceMesh + ) -> ShardOrder | None: + """ + Try to convert _StridedShard placements to ShardOrder. + + This is the inverse of `_convert_shard_order_to_StridedShard`. It reconstructs the shard + order by examining the split_factor of each _StridedShard and determining its position + in the execution order. If the _StridedShard configuration cannot be represented as a + valid ShardOrder (i.e., there's no shard order that produces the observed split_factors), + this function returns None. + + Args: + placements: Tuple of Placement objects that may contain _StridedShard. + mesh: DeviceMesh containing the size information for each mesh dimension. + + Returns: + ShardOrder if conversion is possible, None otherwise. For placements without + _StridedShard, returns the default shard order. + + Algorithm: + 1. If no _StridedShard in placements, return default shard order + 2. Create an empty list for each tensor dimension to represent mesh dim ordering + 3. Iterate through placements in reverse order (right to left): + - For each Shard/_StridedShard on a tensor dimension: + - Extract its split_factor (1 for Shard, split_factor for _StridedShard) + - Find the position in mesh_dims_order where accumulated_sf equals split_factor + - accumulated_sf is the product of mesh sizes of mesh dimensions that appear + earlier in mesh_dims_order (lower indices) + - Insert mesh_dim at the found position + 4. If no valid position found for any split_factor, return None (unable to convert) + 5. Construct ShardOrderEntry for each tensor dimension from mesh_dims_order + + Example: + >>> # xdoctest: +SKIP("Requires DeviceMesh") + >>> # mesh = DeviceMesh([4, 3, 2]) # sizes: mesh[0]=4, mesh[1]=3, mesh[2]=2 + >>> # placements = (_StridedShard(0, sf=2), _StridedShard(0, sf=2), Shard(0)) + >>> # Process tensor_dim=0 from right to left: + >>> # - mesh_dim=2: Shard(0) with sf=1 + >>> # Try position 0: accumulated_sf=1, matches! Insert at position 0 + >>> # Current mesh_dims_order order: [2] + >>> # - mesh_dim=1: _StridedShard(0, sf=2) with sf=2 + >>> # Try position 0: accumulated_sf=1, no match + >>> # Try position 1: accumulated_sf=1*mesh.size(2)=2, matches! Insert at position 1 + >>> # Current mesh_dims_order order: [2, 1] + >>> # - mesh_dim=0: _StridedShard(0, sf=2) with sf=2 + >>> # Try position 0: accumulated_sf=1, no match + >>> # Try position 1: accumulated_sf=1*mesh.size(2)=2, matches! Insert at position 1 + >>> # Final mesh_dims_order order: [2, 0, 1] + >>> # Result: ShardOrder((ShardOrderEntry(tensor_dim=0, mesh_dims=(2, 0, 1)),)) + >>> # This means: first shard on mesh_dim=2, then mesh_dim=0, then mesh_dim=1 + + Note: + This function validates that _StridedShard can be represented as a ShardOrder. + Not all _StridedShard configurations are valid - the split_factor must match + the product of mesh sizes in some execution order. + """ + if not any(isinstance(p, _StridedShard) for p in placements): + return DTensorSpec.compute_default_shard_order(placements) + max_tensor_dim = max([i.dim for i in placements if _is_shard_like(i)]) + 1 + shard_order = [] + + tensor_dim_to_mesh_dims_order: list[list[int]] = [ + [] for i in range(max_tensor_dim) + ] + for mesh_dim in reversed(range(len(placements))): + cur_placement = placements[mesh_dim] + if _is_shard_like(cur_placement): + tensor_dim = cur_placement.dim + mesh_dims_order = tensor_dim_to_mesh_dims_order[tensor_dim] + cur_sf = 1 + if isinstance(cur_placement, _StridedShard): + cur_sf = cur_placement.split_factor + accumulated_sf = 1 + find_order = False + for i in range(len(mesh_dims_order) + 1): + if accumulated_sf == cur_sf: + mesh_dims_order.insert(i, mesh_dim) + find_order = True + break + if i < len(mesh_dims_order): + accumulated_sf *= mesh.size(mesh_dims_order[i]) + if not find_order: + # _StridedShard is not convertible to ShardOrder + return None + else: + if not isinstance(cur_placement, Replicate | Partial | _MaskPartial): + raise ValueError( + f"Unsupported placement type {type(cur_placement)} encountered in " + f"{placements}; expected Replicate, Partial, or _MaskPartial." + ) + for tensor_dim in range(max_tensor_dim): + if len(tensor_dim_to_mesh_dims_order[tensor_dim]) > 0: + shard_order.append( + ShardOrderEntry( + tensor_dim=tensor_dim, + mesh_dims=tuple(tensor_dim_to_mesh_dims_order[tensor_dim]), + ) + ) + return tuple(shard_order) + + def _verify_shard_order(self, shard_order: ShardOrder) -> None: + """Verify that the shard_order is valid and matches the placements.""" + total_shard = 0 + if any(isinstance(p, _StridedShard) for p in self.placements): + # _StridedShard shard_order validation not yet supported; + # the Shard-only checks below (line 390, 394) would fail. + return + prev_tensor_dim = -1 + for entry in shard_order: + tensor_dim = entry.tensor_dim + mesh_dims = entry.mesh_dims + if len(mesh_dims) <= 0: + raise AssertionError(f"shard_order {shard_order} has empty mesh dim") + if tensor_dim < 0: + raise AssertionError( + f"shard_order {shard_order} has invalid tensor dim {tensor_dim}" + ) + if tensor_dim <= prev_tensor_dim: + raise AssertionError("tensor dim should be sorted in shard_order") + prev_tensor_dim = tensor_dim + total_shard += len(mesh_dims) + for mesh_dim in mesh_dims: + if not (0 <= mesh_dim < len(self.placements)): + raise AssertionError( + f"shard_order {shard_order} has invalid mesh dim {mesh_dims}" + ) + if self.placements[mesh_dim] != Shard(tensor_dim): + raise AssertionError( + f"placement[{mesh_dim}] doesn't have a matching shard in shard_order" + ) + if total_shard != sum(1 for p in self.placements if isinstance(p, Shard)): + raise AssertionError + + def __setattr__(self, attr: str, value: Any) -> None: + if attr == "shard_order" and value is not None: + self._verify_shard_order(value) + super().__setattr__(attr, value) + # Make sure to recompute the hash in case any of the hashed attributes + # change (though we do not expect `mesh`, `placements` or `shard_order` + # to change) + if hasattr(self, "_hash") and attr in ( + "mesh", + "placements", + "tensor_meta", + "shard_order", + ): + self._hash = None + # This assert was triggered by buggy handling for dict outputs in some + # FX passes, where you accidentally iterate over a dict and try to put + # keys into TensorMeta. See https://github.com/pytorch/pytorch/issues/157919 + if attr == "tensor_meta" and value is not None: + from torch.fx.passes.shape_prop import TensorMetadata + + # TODO: the TensorMetadata arises from + # test/distributed/tensor/experimental/test_tp_transform.py::TensorParallelTest::test_tp_transform_e2e + # but I actually can't reproduce it, maybe it is also a bug! + if not isinstance(value, TensorMeta | TensorMetadata): + raise AssertionError(repr(value)) + + def _hash_key(self) -> tuple[Any, ...]: + """Return the tuple used for hashing. Used by both __hash__ and _stable_hash.""" + if self.tensor_meta is not None: + return ( + self.mesh, + self.placements, + self.shard_order, + self.tensor_meta.shape, + self.tensor_meta.stride, + self.tensor_meta.dtype, + ) + return (self.mesh, self.placements, self.shard_order) + + def _hash_impl(self) -> int: + # hashing and equality check for DTensorSpec are used to cache the sharding + # propagation results. We only need to consider the mesh, placements, shape + # dtype and stride. + # Caveat: we need to keep this in mind and sync hash and eq if we add more + # fields to them. + return hash(self._hash_key()) + + def __hash__(self) -> int: + # We lazily cache the spec to avoid recomputing the hash upon each + # use, where we make sure to update the hash when the `tensor_meta` + # changes by overriding `__setattr__`. This must be lazy so that Dynamo + # does not try to hash non-singleton `SymInt`s for the stride. + if self._hash is None: + self._hash = self._hash_impl() + return self._hash + + def _stable_hash(self) -> str: + """ + Return a stable hash for AOT autograd caching. + [See note: Tensor subclass stable hashing for AOT autograd cache] + """ + from torch._library.fake_class_registry import maybe_unwrap_fake_script_object + + # Get hash key, but replace mesh with its stable hash + key = self._hash_key() + # First element is mesh, replace with its stable hash + mesh = maybe_unwrap_fake_script_object(self.mesh) + stable_key = (mesh._stable_hash(),) + key[1:] + return hashlib.blake2b(repr(stable_key).encode(), digest_size=16).hexdigest() + + def _check_equals(self, other: object, skip_shapes: bool = False) -> bool: + if not ( + isinstance(other, DTensorSpec) + and self.mesh == other.mesh + and self.placements == other.placements + and self.shard_order == other.shard_order + ): + return False + if self.tensor_meta is None or other.tensor_meta is None: + return self.tensor_meta == other.tensor_meta + + if skip_shapes: + return self.tensor_meta.dtype == other.tensor_meta.dtype + return ( + self.tensor_meta.shape == other.tensor_meta.shape # type: ignore[union-attr] + and self.tensor_meta.stride == other.tensor_meta.stride # type: ignore[union-attr] + and self.tensor_meta.dtype == other.tensor_meta.dtype # type: ignore[union-attr] + ) + + def __eq__(self, other: object, /) -> bool: + return self._check_equals(other) + + def __str__(self) -> str: + """ + human readable representation of the DTensorSpec + """ + placement_str = self.format_shard_order_str(self.placements, self.shard_order) + if self.tensor_meta is not None: + tensor_shape = _stringify_shape(self.tensor_meta.shape) + tensor_dtype = dtype_abbrs[self.tensor_meta.dtype] + else: + tensor_shape = "unknown shape" + tensor_dtype = "unknown dtype" + + return f"Spec({tensor_dtype}{tensor_shape}({placement_str}))" + + @staticmethod + def is_default_device_order(shard_order: ShardOrder) -> bool: + """ + Check if the device order is the default left-to-right order. + """ + if shard_order is None: + # Missing `shard_order` attribute, possibly due to _StridedShard in + # placements. + return False + for entry in shard_order: + mesh_dims = entry.mesh_dims + is_increasing = all( + prev < nxt for prev, nxt in itertools.pairwise(mesh_dims) + ) + if not is_increasing: + return False + return True + + @staticmethod + def format_shard_order_str( + placements: tuple[Placement, ...], + shard_order: ShardOrder | None = None, + ) -> str: + """ + Format DTensor sharding information as a human-readable string. + + This method formats the sharding pattern in mesh-centric order, showing the placement + for each mesh dimension sequentially. When a tensor dimension is sharded across multiple + mesh dimensions, the order index indicates the execution sequence of the sharding operations. + + Args: + placements: Tuple of placement objects for each mesh dimension. + shard_order: Optional ShardOrder specifying the sharding order. + + Returns: + String representation of the sharding pattern in mesh-centric format. + + Example: + For a 3D tensor on a 2x2x2x2 mesh (16 devices) with:: + + placements = [Partial(), Shard(1), Shard(1), Replicate()] + shard_order = (ShardOrderEntry(tensor_dim=1, mesh_dims=(2, 1)),) + + Mesh configuration: + - mesh_dim_0: Partial reduction (sum) + - mesh_dim_1: Shard tensor dimension 1 (executed second, order index 1) + - mesh_dim_2: Shard tensor dimension 1 (executed first, order index 0) + - mesh_dim_3: Replicate + + Output: ``"PS(1)[1]S(1)[0]R"`` + + Explanation: + - ``P``: mesh dimension 0 has partial reduction + - ``S(1)[1]``: mesh dimension 1 shards tensor dimension 1 (order index 1 means second) + - ``S(1)[0]``: mesh dimension 2 shards tensor dimension 1 (order index 0 means first) + - ``R``: mesh dimension 3 replicates + + The format follows mesh dimension order (0, 1, 2, 3), and when a tensor dimension + is sharded across multiple mesh dimensions, the bracketed index shows the execution + order: ``[0]`` is executed first, ``[1]`` is executed second, etc. + """ + out_str = "" + # native dtensor-style sharding representation: map from mesh + # dim to tensor dim + for mesh_dim, placement in enumerate(placements): + if _is_shard_like(placement): + if shard_order is not None: + for entry in shard_order: + tensor_dim = entry.tensor_dim + mesh_dims = entry.mesh_dims + + if placement.dim == tensor_dim: + if mesh_dim not in mesh_dims: + raise AssertionError + if len(mesh_dims) > 1: + out_str += f"{placement}[{mesh_dims.index(mesh_dim)}]" + else: + # no need to show device order if the tensor dim is + # only sharded in one mesh dim + out_str += str(placement) + break + else: + out_str += str(placement) + else: + out_str += str(placement) + return out_str + + @property + def shape(self) -> torch.Size: + if self.tensor_meta is None: + raise ValueError("tensor_meta is not set") + return self.tensor_meta.shape + + @property + def stride(self) -> tuple[int, ...]: + if self.tensor_meta is None: + raise ValueError("tensor_meta is not set") + return self.tensor_meta.stride + + @property + def ndim(self) -> int: + if self.tensor_meta is None: + raise ValueError("tensor_meta is not set") + return len(self.tensor_meta.shape) + + @property + def num_shards(self) -> int: + num_shards = 1 + for i, placement in enumerate(self.placements): + if _is_shard_like(placement): + num_shards *= self.mesh.size(i) + return num_shards + + @property + def device_mesh(self) -> DeviceMesh: + # simple aliasing for the mesh field, make some + # checks that mixes DTensor/DTensorSpec easier + return self.mesh + + @property + def dim_map(self) -> list[int]: + """ + dim_map is a property we derive from `placements` of + the distributed tensor. It simply return a list of ints + where dim_map[i] denotes the sharding mapping to the mesh + dimension, and len(dim_map) == dist_tensor.ndim + dim_map[i] = -1: means tensor dim i replicate on mesh + dim_map[i] = j: means tensor dim i shard on mesh dim j + + For example, we have a dist tensor that have the shape of + [18, 20, 30], and device_mesh([0, 1, 2, 3]), placements: + [Shard(1)], the dim_map of this placement would be: + [-1, 0, -1]. This representation is pretty helpful during + sharding propagation where we could know exactly each + tensor dimension is sharded or not. + + Note that if placements contains `_Partial`, we have to + explicitly deal with it, so that when we create a DTensorSpec + with dim_map, we could properly record the pending sums. + """ + # dims mapping of dist tensor sharding + # return size of tensor ndim, -1 represent replicate + # and int >=0 represent shard on that device mesh dim + r = [-1] * self.ndim + for i, placement in enumerate(self.placements): + if _is_shard_like(placement): + shard_dim = placement.dim + if r[shard_dim] > -1: + raise ValueError( + f"Tensor dim {shard_dim} is already sharded on mesh dim {r[shard_dim]}," + " DTensor operator implementation does not support things like hybrid" + " sharding strategies yet (i.e. [Shard(0), Shard(0)])" + ) + r[shard_dim] = i + return r + + @property + def num_shards_map(self) -> list[int]: + """ + dim_map is a property we derive from `placements` of + the distributed tensor. Unlike `dim_map`, `num_shards_map` + denotes how many shards each tensor dim has. Like `dim_map`: + len(num_shards_map) == dist_tensor.ndim + num_shards_map[i] = 1: means tensor dim i is not sharded + num_shards_map[i] = j: means tensor dim i has j shards in total + + For example, we have a dist tensor of shape [18, 20, 30], + a device_mesh ([[0, 1, 2, 3], [4, 5, 6, 7]]), and placements + ([Shard(1), Shard(0)]), the num_shards_map of this distributed tensor + would be: [4, 2, 1]. + """ + r = [1] * self.ndim + for i, placement in enumerate(self.placements): + if _is_shard_like(placement): + r[placement.dim] *= self.mesh.size(i) + + return r + + @property + def sums(self) -> list[int]: + """ + sums is a property we derive from `placements` of the + distributed tensor. It simply return a list of ints where + sums[i] denotes the pending sum (partial) on mesh dim i + """ + return [ + idx + for idx, placement in enumerate(self.placements) + if placement.is_partial() + ] + + @classmethod + def from_dim_map( + cls, + mesh: DeviceMesh, + dim_map: list[int], + sums: list[int], + tensor_meta: TensorMeta | None = None, + ) -> "DTensorSpec": + """ + Construct a DTensorSpec from dim_map list and pending sum. + + Args: + mesh (class:`DeviceMesh`): device mesh to be used in the DTensorSpec + dim_map (List[int]): a list of integer that represents sharding on each + tensor dimension, see `dim_map` property doc for details + sums (List[int]): a list of integer that represents the dist tensor have + pending sum on which device mesh dimension. + tensor meta (TensorMeta): DTensor metadata + + Return: + a class:`DTensorSpec` object + """ + # by default replicate on device mesh dims + placements: list[Placement] = [Replicate() for _ in range(mesh.ndim)] + + # find all mesh dims that need pending reductions + for s in sums: + placements[s] = Partial() + + for i, m in enumerate(dim_map): + if m >= 0: + placement = placements[m] + if placement.is_shard(): # dim_map only produces Shard placements + placement = cast(Shard, placement) + raise RuntimeError( + f"DeviceMesh dimension can't be mapped to two dimension of the same tensor: {i} and {placement.dim}" + ) + elif placement.is_partial(): + raise RuntimeError( + f"DeviceMesh dimension {m} cannot be both shard and partial!" + ) + placements[m] = Shard(i) + + return cls(mesh, tuple(placements), tensor_meta=tensor_meta) + + def is_replicated(self) -> bool: + """ + return True if the current DTensorSpec replicates on all mesh dims (devices) + """ + return all(placement.is_replicate() for placement in self.placements) + + def is_sharded(self) -> bool: + """ + return True if the current DTensorSpec uses Shard() or _StridedShard() placement on any mesh dims (devices) + """ + return any(_is_shard_like(placement) for placement in self.placements) + + def shallow_copy_with_tensor_meta( + self, tensor_meta: TensorMeta | None + ) -> "DTensorSpec": + """ + Shallow copy the DTensorSpec with a new tensor_meta. + """ + if tensor_meta is None: + raise AssertionError("shallow copy with no tensor_meta!") + return DTensorSpec( + self.mesh, + self.placements, + tensor_meta=tensor_meta, + use_strided_shard_as_shard_order=self.use_strided_shard_as_shard_order, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_nonlinear_redux.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_nonlinear_redux.py new file mode 100644 index 0000000000000000000000000000000000000000..44543a23c4e06a2e4ca1921f5e2607fc30f8155b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_nonlinear_redux.py @@ -0,0 +1,305 @@ +import operator +from functools import reduce +from typing import cast + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.tensor._api as dtensor +from torch.distributed.tensor._op_schema import OutputSharding +from torch.distributed.tensor._utils import compute_local_shape_and_global_offset +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) + + +# Mapping from argmin/argmax ops to their corresponding value ops (min/max) +_ARGMINMAX_REDUCTION_OPS = { + torch.ops.aten.argmax.default: torch.max, + torch.ops.aten.argmin.default: torch.min, +} + + +def _get_output_sharding( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> OutputSharding: + """Get the output sharding for the given op.""" + op_info = dtensor.DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + dtensor.DTensor._op_dispatcher.sharding_propagator.propagate(op_info) + output_sharding = op_info.output_sharding + if output_sharding is None: + raise AssertionError("output sharding should not be None") + return output_sharding + + +def _prep_arguments( + op_call_repr: str, + args: tuple[object, ...], + kwargs: dict[str, object] | None, +) -> tuple[ + torch.Tensor, + torch.Size, + "torch.distributed.device_mesh.DeviceMesh", + tuple[Placement, ...], + int | None, + bool, +]: + """ + Prepare arguments for nonlinear reduction ops. + + Returns: + local_tensor: The local tensor to operate on + global_shape: The global shape of the DTensor + device_mesh: The device mesh + placements: The placements tuple + dim: The reduction dimension (can be None) + keepdim: Whether to keep the reduced dimension + """ + input_dtensor = cast(dtensor.DTensor, args[0]) + dim: int | None = None + keepdim: bool = False + + if not isinstance(input_dtensor, dtensor.DTensor): + raise NotImplementedError + if len(args) > 1: + dim = cast(int, args[1]) + if len(args) > 2: + keepdim = cast(bool, args[2]) + if kwargs: + if "dim" in kwargs: + dim = cast(int, kwargs["dim"]) + if "keepdim" in kwargs: + keepdim = cast(bool, kwargs["keepdim"]) + device_mesh = input_dtensor.device_mesh + placements = input_dtensor.placements + + # check for partial placements and handle it as a replicate. + if any(isinstance(p, Partial) for p in placements): + target_placements = [ + Replicate() if isinstance(p, Partial) else p for p in placements + ] + input_dtensor = input_dtensor.redistribute( + device_mesh=device_mesh, placements=target_placements + ) + placements = input_dtensor.placements + local_tensor = input_dtensor.to_local() + global_shape = input_dtensor.shape + + return local_tensor, global_shape, device_mesh, placements, dim, keepdim + + +def _get_expected_shape( + local_tensor: torch.Tensor, dim: int | None, keepdim: bool +) -> torch.Size: + """Compute the expected output shape after reduction.""" + input_shape = list(local_tensor.shape) + if dim is None: + expected_shape = ( + torch.Size([1] * len(input_shape)) if keepdim else torch.Size([]) + ) + elif keepdim: + if input_shape: + input_shape[dim] = 1 + expected_shape = torch.Size(input_shape) + else: + if input_shape: + input_shape.pop(dim) + expected_shape = torch.Size(input_shape) + + return expected_shape + + +def _collect_shard_mesh_dims( + op_call_repr: str, + local_tensor: torch.Tensor, + placements: tuple[Placement, ...], + dim: int | None, +) -> list[int]: + """Collect mesh dimensions that are sharded along the reduction dimension.""" + shard_mesh_dims: list[int] = [] + for mesh_dim, p in enumerate(placements): + if isinstance(p, Shard): + if dim is None or p.dim == (dim if dim >= 0 else local_tensor.ndim + dim): + shard_mesh_dims.append(mesh_dim) + elif isinstance(p, _StridedShard): + raise NotImplementedError(f"{op_call_repr} does not support _StridedShard!") + return shard_mesh_dims + + +def _convert_to_global_idxs( + local_idx: torch.Tensor, + global_shape: torch.Size, + device_mesh: "torch.distributed.device_mesh.DeviceMesh", + placements: tuple[Placement, ...], + dim: int | None, +) -> tuple[int, torch.Tensor]: + """Convert local indices to global indices.""" + local_shape, global_offset = compute_local_shape_and_global_offset( + global_shape, device_mesh, placements + ) + + if dim is None: + # Convert flat local index → flat global index using arithmetic ops + # instead of torch.unravel_index, which doesn't support SymInt shapes. + gathered_idxs = torch.zeros_like(local_idx) + remaining = local_idx + for i in range(len(local_shape)): + local_stride = reduce(operator.mul, local_shape[i + 1 :], 1) + global_stride = reduce(operator.mul, global_shape[i + 1 :], 1) + coord = remaining // local_stride + remaining = remaining % local_stride + gathered_idxs = gathered_idxs + (coord + global_offset[i]) * global_stride + gather_dim = 0 + else: + gather_dim = dim + gathered_idxs = local_idx + global_offset[dim] + return gather_dim, gathered_idxs + + +def _gather_tensors( + gather_dim: int, + gathered_idxs: torch.Tensor, + local_redux: torch.Tensor, + device_mesh: "torch.distributed.device_mesh.DeviceMesh", + shard_mesh_dims: list[int], +) -> tuple[torch.Tensor, torch.Tensor]: + """ + Gather the min or max of the tensors and their corresponding indices. + + Args: + gather_dim: The dim to stack the collected min/max tensors. + gathered_idxs: The local tensor holding the corresponding indices. + local_redux: The local tensor holding the operator's value i.e. min/max. + device_mesh: Device mesh of the DTensor. + shard_mesh_dims: List of mesh dimensions that are sharded. + + Returns: + All gathered tensors (gathered_redux, gathered_idxs) of the reducing operator. + """ + gathered_redux = local_redux + for mesh_dim in shard_mesh_dims: + gathered_redux = funcol.all_gather_single( + gathered_redux, + gather_dim=gather_dim, + group=(device_mesh, mesh_dim), + ) + gathered_idxs = funcol.all_gather_single( + gathered_idxs, + gather_dim=gather_dim, + group=(device_mesh, mesh_dim), + ) + return gathered_redux, gathered_idxs + + +def argminmax_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + """ + Handler for aten.argmin.default and aten.argmax.default ops. + + This is a pure function handler that doesn't require instantiation. + """ + if op_call not in _ARGMINMAX_REDUCTION_OPS: + raise NotImplementedError(f"Unsupported reduction op: {op_call}") + + local_tensor, global_shape, device_mesh, placements, dim, keepdim = _prep_arguments( + str(op_call), args, kwargs + ) + output_sharding = _get_output_sharding(op_call, args, kwargs) + + expected_shape = _get_expected_shape(local_tensor, dim, keepdim) + shard_mesh_dims = _collect_shard_mesh_dims( + str(op_call), local_tensor, placements, dim + ) + + # Compute local reduction + if dim is None: + val_op = _ARGMINMAX_REDUCTION_OPS[op_call] + # unsqueeze scalars to 1-d so they can be allgathered + local_redux = val_op(local_tensor).unsqueeze(0) + local_idx = op_call(local_tensor).unsqueeze(0) + else: + val_op = _ARGMINMAX_REDUCTION_OPS[op_call] + local_redux, local_idx = val_op(local_tensor, dim=dim, keepdim=True) + + if not shard_mesh_dims: + return dtensor.DTensor._op_dispatcher.wrap( + local_idx.reshape(expected_shape), output_sharding.output_spec + ) + + gather_dim, gathered_idxs = _convert_to_global_idxs( + local_idx, global_shape, device_mesh, placements, dim + ) + gathered_redux, gather_idxs = _gather_tensors( + gather_dim, gathered_idxs, local_redux, device_mesh, shard_mesh_dims + ) + # Select the rank with the best value; use dim=0 when dim was None since + # the scalars were unsqueezed to 1-d for gathering + select_dim = 0 if dim is None else dim + rank_winner = op_call(gathered_redux, select_dim, True) + final_idx = torch.gather(gather_idxs, dim=gather_dim, index=rank_winner) + + return dtensor.DTensor._op_dispatcher.wrap( + final_idx.reshape(expected_shape), output_sharding.output_spec + ) + + +def minmax_dim_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + """ + Handler for aten.min.dim and aten.max.dim ops. + + This is a pure function handler that doesn't require instantiation. + """ + local_tensor, global_shape, device_mesh, placements, dim, keepdim = _prep_arguments( + str(op_call), args, kwargs + ) + output_sharding = _get_output_sharding(op_call, args, kwargs) + + expected_shape = _get_expected_shape(local_tensor, dim, keepdim) + shard_mesh_dims = _collect_shard_mesh_dims( + str(op_call), local_tensor, placements, dim + ) + + # Compute local reduction - min/max with dim always requires dim + if dim is None: + raise AssertionError + local_redux, local_idx = op_call(local_tensor, dim=dim, keepdim=True) + + if not shard_mesh_dims: + return dtensor.DTensor._op_dispatcher.wrap( + ( + local_redux.reshape(expected_shape), + local_idx.reshape(expected_shape), + ), + output_sharding.output_spec, + ) + + gather_dim, gathered_idxs = _convert_to_global_idxs( + local_idx, global_shape, device_mesh, placements, dim + ) + + gathered_redux, gather_idxs = _gather_tensors( + gather_dim, gathered_idxs, local_redux, device_mesh, shard_mesh_dims + ) + # The op_call here is min/max with dim which returns (values, indices) + final_redux, rank_winner = op_call(gathered_redux, dim, True) + final_idx = torch.gather(gather_idxs, dim=gather_dim, index=rank_winner) + + return dtensor.DTensor._op_dispatcher.wrap( + ( + final_redux.reshape(expected_shape), + final_idx.reshape(expected_shape), + ), + output_sharding.output_spec, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_op_schema.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_op_schema.py new file mode 100644 index 0000000000000000000000000000000000000000..cffbf47ca576bcd1956e1188fd3fa76582486511 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_op_schema.py @@ -0,0 +1,723 @@ +# mypy: allow-untyped-defs +""" +DTensor operator schema definitions and utilities. + +This module defines the core data structures and utilities for describing and managing +distributed tensor operations in PyTorch's DTensor system. It provides the foundational +schema types used for sharding propagation, operator strategy selection, and distributed +execution planning. + +Key components: +- OpSpec: Describes acceptable sharding placements for operations +- OpStrategy: Represents the possible sharding strategies for an operator +- TupleStrategy: Container for multiple strategies when ops have tuple/list of tensors input +- OpSchema: Describes operator input/output schemas with DTensorSpecs +- OutputSharding: Manages output sharding specifications and redistribution +- RuntimeSchemaInfo: Runtime execution metadata for operators +- OpInfo: Complete runtime operator execution information + +These schema definitions enable the DTensor system to: +1. Propagate tensor sharding information to the operator outputs +2. Greedily select sharding strategies for distributed operations +3. Plan and execute tensor redistributions when needed +4. Cache sharding decisions for performance optimization +""" + +from collections.abc import Sequence +from dataclasses import dataclass +from functools import cached_property +from typing import Any +from typing_extensions import deprecated + +import torch +from torch._C import ( + _DTensor_OpSchema_post_init, + _DTensor_OpSchema_recompute_comparison_key, +) +from torch._ops import OpOverload +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor.placement_types import Placement + + +try: + from torch.utils._cxx_pytree import ( + register_pytree_node, + tree_leaves, + tree_map_only, + TreeSpec, + ) +except ImportError: + from torch.utils._pytree import ( # type: ignore[no-redef, assignment] + register_pytree_node, + tree_leaves, + tree_map_only, + TreeSpec, + ) + + +# Common type aliases +ArgsType = tuple[object, ...] +KwargsType = dict[str, object] + +PlacementList = list[Placement | None] + +# ATen op schemas could have Tensor, Tuple[Tensor] and List[Tensor], so output type should +# be the same set of possibilities. +OutputSpecType = DTensorSpec | Sequence[DTensorSpec | None] | None + + +def _rebuild_tensor_from_dtensor_meta(arg) -> object: + """ + This is used to propagate tensor metadata, must be under fake mode + """ + if arg.tensor_meta is None: + raise AssertionError("DTensorSpec does not contain tensor_meta.") + return torch.empty_strided( + arg.tensor_meta.shape, + arg.tensor_meta.stride, + dtype=arg.tensor_meta.dtype, + device=arg.mesh.device_type, + ) + + +def _pretty_print_spec(spec: object) -> str: + if spec is None: + return "None" + elif isinstance(spec, DTensorSpec): + return "".join([str(p) for p in spec.placements]) + elif isinstance(spec, Sequence): + return "(" + ", ".join([_pretty_print_spec(s) for s in spec]) + ")" + else: + raise RuntimeError(f"Unknown spec type to print: spec={spec}") + + +@dataclass +class OpSpec: + """ + An OpSpec describes an acceptable sharding placements of an operation, with the + specified DTensorSpecs for both the output and the inputs. + + note: when the op return value is a single DTensor object, output_specs is + DTensorSpec; when the return value is a tuple of Optional[DTensor], + output_specs is a tuple of Optional[DTensorSpec]. + + note: we MUST produce an DTensorSpec for every output that is a Tensor. None + entries only occur for non-Tensor outputs (e.g., operators that return Optional[Tensor], + or non-Tensor outputs.) + + invariant: the DeviceMesh on all DTensorSpec must be the same + """ + + # output_specs and input_specs are related: for this op, given these input_specs, + # this is the way the output would look + # Note: output_specs can be None for ops that don't return tensors (e.g., _linalg_check_errors) + output_specs: DTensorSpec | tuple[DTensorSpec | None, ...] | None + input_specs: Sequence[DTensorSpec] | None = None + + """ + redistribute_cost tells how expensive it is to redistribute a given input into the + placement specified in this OpSpec. + + outer list: one entry (list) per (tensor) input in the op's arg schema + inner list: one entry (cost value) per possible sharding spec for that input + + Example: + ------- + another_op() -> tensor_a # another_op produces the output that becomes our first input + my_op(tensor_a) + + Let's assume this OpSpec's input_specs are [Replicate()], + but another_op() supports 2 strategies (OpSpecs) which produce outputs of + Replicate() + Shard(0) + + In this example, redistribute_costs would look like this + [ + # one row representing "my_op's first input" (tensor_a) + [ + # two entries, one for each strategies supported by another_op + 0.0, # cost of redistributing tensor_a from 'Replicate()' + K, # cost of redistributing tensor_a from 'Shard(0)' + ], + """ + redistribute_cost: list[list[float]] | None = None + + @cached_property + def output_spec(self) -> DTensorSpec: + """ + This function requires that the strategy have exactly one DTensorSpec as the + output spec. If the output_specs is a tuple, we throw an exception. + """ + if isinstance(self.output_specs, DTensorSpec): + return self.output_specs + else: + raise ValueError( + f"function output_spec expects a single DTensorSpec but got: {self.output_specs}" + ) + + @cached_property + def mesh(self): + if isinstance(self.output_specs, DTensorSpec): + return self.output_specs.mesh + elif isinstance(self.output_specs, tuple): + for out_spec in self.output_specs: + if isinstance(out_spec, DTensorSpec): + return out_spec.mesh + # all entries are None — fall through to input_specs + elif self.output_specs is not None: + raise ValueError( + f"output_specs expects DTensorSpec, tuple of DTensorSpec, or None but got: {self.output_specs}" + ) + # output_specs is None or tuple-of-all-None — try input_specs + if self.input_specs is not None: + for spec in self.input_specs: + if spec is not None: + return spec.mesh + raise AssertionError( + "Cannot determine mesh: no non-None DTensorSpec in output_specs or input_specs" + ) + + def input_spec(self, index: int = 0) -> DTensorSpec: + if self.input_specs is None: + raise AssertionError("input_specs of OpSpec is None!") + if len(self.input_specs) <= index: + raise AssertionError( + f"Invalid index {index} for input_specs of length " + f"{len(self.input_specs)}: {self.input_specs}" + ) + return self.input_specs[index] + + def __str__(self) -> str: + if self.input_specs is not None: + input_specs_str = f"{_pretty_print_spec(self.input_specs)} -> " + else: + input_specs_str = "" + output_spec_str = _pretty_print_spec(self.output_specs) + return f"{input_specs_str}{output_spec_str}" + + def __hash__(self) -> int: + if self.output_specs is None: + output_hash = hash(None) + elif isinstance(self.output_specs, DTensorSpec): + output_hash = hash(self.output_specs) + else: + output_hash = hash(tuple(self.output_specs)) + input_hash = hash(tuple(self.input_specs)) if self.input_specs else 0 + return hash((output_hash, input_hash)) + + def __eq__(self, other: object) -> bool: + if not isinstance(other, OpSpec): + return False + return ( + self.output_specs == other.output_specs + and self.input_specs == other.input_specs + ) + + +class StrategyType: + """ + Base class type for op strategy, We have two StrategyType: + OpStrategy and TupleStrategy + """ + + +class OpStrategy(StrategyType): + """ + OpStrategy that consists of a list of sharding strategies associated with the op, + where each strategy is an OpSpec that describes the acceptable input/output sharding. + + invariant: the DeviceMesh on all OpSpec must be the same + """ + + def __init__(self, strategies: list[OpSpec]) -> None: + super().__init__() + self.strategies: list[OpSpec] = strategies + + def __str__(self) -> str: + strategy_list_str = ", ".join([str(strategy) for strategy in self.strategies]) + try: + mesh_shape = self.mesh_shape + except AssertionError: + return f"OpStrategy[{strategy_list_str}]" + return f"OpStrategy[{strategy_list_str}] @ mesh: {mesh_shape}" + + def max_num_shards(self) -> int: + """ + Returns the max number of shards across all OpSpecs + """ + return max(strategy.output_spec.num_shards for strategy in self.strategies) + + @property + def mesh(self): + return self.strategies[0].mesh + + @property + def mesh_shape(self): + return self.strategies[0].mesh.shape + + @property + def ndim(self): + return self.strategies[0].output_spec.ndim + + @property + def shape(self): + return self.strategies[0].output_spec.shape + + @property + def tensor_meta(self) -> TensorMeta: + # TODO upstream this assert to DTensorSpec itself and fill any missing TensorMetas + if self.strategies[0].output_spec.tensor_meta is None: + raise AssertionError + return self.strategies[0].output_spec.tensor_meta + + def __hash__(self) -> int: + return hash(tuple(self.strategies)) + + def __eq__(self, other: object) -> bool: + return isinstance(other, OpStrategy) and self.strategies == other.strategies + + +class TupleStrategy(StrategyType): + """ + TupleStrategy is a special case for operators that are fundamentally compound or batched such that some subset + of the inputs and outputs are completely unrelated to some other subset. + + Generally, foreach_* ops are the most common use-case for TupleStrategy, because they accept lists of inputs, + but operate independently on each input or tuple of zipped inputs. + + For example, [out_a, out_b] = torch.foreach_add([a, b], scalar): input a's sharding only affects out_a's sharding, + independent of b and out_b. + + An example of an operator that should NOT use TupleStrategy is torch.split. It produces a List[Tensor] + as its output, but the sharding decision of one output is bound together with the decision + of each other output and the common input. + """ + + def __init__( + self, + children: Sequence[StrategyType], + ) -> None: + super().__init__() + self.children: Sequence[StrategyType] = children + + @property + @deprecated( + "TupleStrategy.childs is deprecated, use TupleStrategy.children instead.", # codespell:ignore childs + category=FutureWarning, + ) + def childs(self) -> Sequence[StrategyType]: # codespell:ignore childs + """ + Alias for children, to maintain backward compatibility. + """ + return self.children + + def child_mesh(self, index: int) -> DeviceMesh: + op_strategy = self.children[index] + if not isinstance(op_strategy, OpStrategy): + raise AssertionError + return op_strategy.mesh + + def __str__(self) -> str: + child_strategies_str = ", ".join( + [f"{str(strat)}" for idx, strat in enumerate(self.children)] + ) + return f"TupleStrategy({child_strategies_str})" + + def __hash__(self) -> int: + return hash(tuple(self.children)) + + def __eq__(self, other: object) -> bool: + return isinstance(other, TupleStrategy) and self.children == other.children + + +try: + register_pytree_node( + TupleStrategy, + lambda node: (node.children, None), + lambda children, _: TupleStrategy(tuple(children)), + ) +except ValueError: + # already registered TupleStrategy, skip + pass + + +@dataclass +class RuntimeSchemaInfo: + """ + RuntimeSchemaInfo stores the operator schema related information for runtime (eager) + execution. This is mainly used for two ways: 1. to generate hash for args to determine + whether to re-run sharding prop or not 2. to determine if we need pytree + """ + + # This static_argnum records static arg "starting index" for ops that have non-tensor + # args/kwargs which would affect sharding propagation results. All args starting from + # this index would be hashed to our sharding cache. + # Note that only a few ops need this information, e.g. view, transpose, var.dim, etc. + static_argnum: int = 100 + # This static_kwargkey records static kwarg names which would affect sharding prop + static_kwargkey: list[str] | None = None + # each op can decide if it wants to use pytree flatten/unflatten during operator + # eager execution, by default we don't need to do flatten/unflatten, only if the + # op indicate it needs to, this is to accelerate eager performance. + needs_pytree: bool = False + + +@dataclass +class OpSchema: + """ + OpSchema is a data class that describes an operator input schemas, it includes + DTensorSpecs/OpStrategies (instead of DTensor) and non-tensor args/kwargs (positional + order preserved). It is mainly used by the DTensor's dispatching logic to perform various + actions (i.e. sharding propagation, caching sharding decisions, redistribute, etc.) + + NOTE: this must be used as a read only data class + TODO: make this a frozen dataclass + + Args: + op: the operator overload we are intercepting + args_schema: contains args except that the DTensor args have been replaced + with its DTensorSpec or OpStrategy + kwargs_schema: contains kwargs except that the DTensor kwargs have been replaced + with its DTensorSpec or OpStrategy + """ + + op: OpOverload + args_schema: ArgsType + kwargs_schema: KwargsType + + schema_info: RuntimeSchemaInfo | None = None + + _comparison_key: tuple[object, ...] | None = None + + @property + def args_spec(self) -> tuple[DTensorSpec, ...]: + """ + args_spec: Tuple[DTensorSpec, ...]: contains a clean list of args spec list + with NO non-DTensor positional arguments (i.e. int/float/tuple, etc) + mainly used by sharding propagation to propagate the output spec + """ + args = ( + tree_leaves(self.args_schema) + if self.schema_info is not None and self.schema_info.needs_pytree + else self.args_schema + ) + return tuple(item for item in args if isinstance(item, DTensorSpec)) + + @property + def args_strategy(self) -> tuple[OpStrategy, ...]: + # filter out non-relevant values from args schema to get a clean OpStrategy list + # separate with args_spec for the ease of type annotation + # TODO: see if we should merge this with args_spec + args = ( + tree_leaves(self.args_schema) + if self.schema_info is not None and self.schema_info.needs_pytree + else self.args_schema + ) + return tuple(item for item in args if isinstance(item, OpStrategy)) + + @property + def kwargs_strategy(self) -> tuple[OpStrategy, ...]: + # returns OpStrategy items from kwargs_schema. + kwargs_vals = ( + tree_leaves(self.kwargs_schema) + if self.schema_info is not None and self.schema_info.needs_pytree + else self.kwargs_schema.values() + ) + return tuple(item for item in kwargs_vals if isinstance(item, OpStrategy)) + + @property + def args_meta(self) -> tuple[TensorMeta | Any, ...]: + # Used for calling single_dim strategy functions, which aren't allowed to see DTensorSpecs/Meshes + # like args_spec, but has OpStrategy replaced with corresponding TensorMeta, + # and TupleStrategy replaced with tuple of TensorMeta + # preserves the original pytree structure + # example: + # args_schema = (OpStrategy1, TupleStrategy([OpStrategy2, OpStrategy3]), OpStrategy4) + # args_meta: (TensorMeta1, (TensorMeta2, TensorMeta3), TensorMeta4) + + def convert_to_meta(item): + if isinstance(item, OpStrategy): + return item.tensor_meta + elif isinstance(item, TupleStrategy): + return tuple(convert_to_meta(child) for child in item.children) + elif isinstance(item, (list, tuple)): + converted = [convert_to_meta(child) for child in item] + return type(item)(converted) + else: + return item + + return tuple(convert_to_meta(arg) for arg in self.args_schema) + + @property + def kwargs_meta(self) -> dict[str, object]: + # like args_meta, but for kwargs + + def convert_to_meta(item): + if isinstance(item, OpStrategy): + return item.tensor_meta + elif isinstance(item, TupleStrategy): + return tuple(convert_to_meta(child) for child in item.children) + elif isinstance(item, (list, tuple)): + converted = [convert_to_meta(child) for child in item] + return type(item)(converted) + else: + return item + + return { + key: convert_to_meta(value) for key, value in self.kwargs_schema.items() + } + + def __repr__(self) -> str: + args_schema = ", ".join([str(arg_schema) for arg_schema in self.args_schema]) + return ( + f"OpSchema(op={self.op}," + f" args_schema=({args_schema})," + f" kwargs_schema={self.kwargs_schema})" + ) + + def __str__(self) -> str: + args_schema: list[str] = [] + device_mesh = None + + for arg in self.args_schema: + if isinstance(arg, DTensorSpec): + args_schema.append(str(arg)) + device_mesh = arg.mesh + elif isinstance(arg, OpStrategy): + if len(arg.strategies) != 1: + raise AssertionError + args_schema.append(_pretty_print_spec(arg.strategies[0].output_specs)) + device_mesh = arg.mesh + elif isinstance(arg, TupleStrategy): + first_op_strategy = arg.children[0] + if not isinstance(first_op_strategy, OpStrategy): + raise AssertionError + device_mesh = first_op_strategy.mesh + args_schema.append(str(arg)) + else: + args_schema.append(str(arg)) + + return f"{self.op}({', '.join(args_schema)}) on {device_mesh})" + + def __post_init__(self) -> None: + _DTensor_OpSchema_post_init(self) + + def arg_type_tensor_or_tensor_list_like(self, arg: object) -> bool: + is_tensor = isinstance(arg, DTensorSpec) + if is_tensor: + return True + + if not isinstance(arg, list): + return False + + return all(isinstance(e, DTensorSpec) or e is None for e in arg) + + def return_type_tuple_tensor_like(self) -> bool: + # all dispatch ops could only return Tuple[Tensor] or have None/ints/floats + # in the tuple, but the first element must be a Tensor, so this check is enough + return_types = self.op._schema.returns + return len(return_types) > 1 and isinstance( + return_types[0].type, torch.TensorType + ) + + def return_type_list_tensor_like(self) -> bool: + # returns True if the return type is a List + return_types = self.op._schema.returns + return len(return_types) == 1 and isinstance( + return_types[0].type, torch.ListType + ) + + def return_type_tensor(self) -> bool: + return_types = self.op._schema.returns + # all dispatch ops only return Tensor or Tuple[Tensor] for tensor like + # return types, so this check is enough for tensor like types + return len(return_types) > 0 and isinstance( + return_types[0].type, torch.TensorType + ) + + def get_mesh_from_args(self, validate: bool = True) -> DeviceMesh: + """ + This util can be used to get a mesh from the OpSchema that contains multiple + DTensors as arguments. When `validate` is True, it will try to validate that all the + arguments have the same mesh to avoid unexpected cross mesh errors. + + NOTE: this util currently does not handle TupleStrategy when `validate=True`, + this is because for TupleStrategy there could be different types of checks, i.e.: + - for stack and cat like op, we need to check within a TupleStrategy is every + input is on the same mesh + - for foreach like ops we need to check "zipped" inputs are on the same mesh + for each index. + """ + mesh = None + # Scan all args to find the first DTensorSpec/OpStrategy (not just the first arg) + for arg in self.args_schema: + if isinstance(arg, (DTensorSpec, OpStrategy)): + mesh = arg.mesh + break + elif isinstance(arg, (list, tuple, TupleStrategy)): + # Scan all elements in the list/tuple, not just the first one, + # to handle cases like List[Optional[Tensor]] where first elem may be None + elems = arg.children if isinstance(arg, TupleStrategy) else arg + for elem in elems: + if isinstance(elem, (DTensorSpec, OpStrategy)): + mesh = elem.mesh + break + if mesh is not None: + break + if mesh is None: + raise ValueError(f"Cannot find device mesh from args for op : {self.op}.") + + if validate: + for arg in self.args_schema[1:]: + if isinstance(arg, (DTensorSpec, OpStrategy)) and arg.mesh != mesh: + raise RuntimeError( + f"DTensor does not support cross-mesh operation on {self.op}! " + f"Got meshes: {mesh} {arg.mesh}. " + f"Please make sure all the arguments have the same DeviceMesh." + ) + + return mesh + + def is_inplace_op(self) -> bool: + # simple analysis of function schema to determine + # if this is an inplace variant, it might not + # be entirely correct, but it's good enough for now. + return self.op._schema.name[-1] == "_" + + def is_out_variant_op(self) -> bool: + # simple analysis of function schema to determine + # if this is an out variant. + return any(argument.is_out for argument in self.op._schema.arguments) + + def is_view_op(self) -> bool: + return self.op._schema._is_view_op() + + def _recompute_comparison_key(self) -> None: + _DTensor_OpSchema_recompute_comparison_key(self) + + def __hash__(self) -> int: + return hash(self._comparison_key) + + def __eq__(self, other: object) -> bool: + # early return checks + if not isinstance(other, OpSchema): + return False + + if self.op != other.op: + return False + + if len(self.args_schema) != len(other.args_schema): + return False + + return self._comparison_key == other._comparison_key + + def gen_fake_args(self) -> ArgsType: + """ + gen_fake_args: generate fake args for the operator, this is mainly used + by sharding propagation rules to generate fake args for the operator + to run the local tensor operator and get the output spec. + """ + return tree_map_only( + DTensorSpec, + _rebuild_tensor_from_dtensor_meta, + self.args_schema, + is_leaf=lambda x: isinstance(x, DTensorSpec), + ) + + def gen_fake_kwargs(self) -> KwargsType: + """ + gen_fake_kwargs: generate fake kwargs for the operator, this is mainly used + by sharding propagation rules to generate fake kwargs for the operator + to run the local tensor operator and get the output spec. + """ + return tree_map_only( + DTensorSpec, + _rebuild_tensor_from_dtensor_meta, + self.kwargs_schema, + is_leaf=lambda x: isinstance(x, DTensorSpec), + ) + + def _inplace_rewrap_schema_suggestion(self, origin_schema: "OpSchema") -> None: + suggestion_args_spec = self.args_spec + new_arg_schema: list[object] = [] + idx_of_args_spec = 0 + if ( + origin_schema.schema_info is not None + and origin_schema.schema_info.needs_pytree + ): + args_schema: Sequence[Any] = tree_leaves(origin_schema.args_schema) + else: + args_schema = origin_schema.args_schema + for arg in args_schema: + if isinstance(arg, DTensorSpec): + new_arg_schema.append(suggestion_args_spec[idx_of_args_spec]) + idx_of_args_spec += 1 + else: + new_arg_schema.append(arg) + self.args_schema = tuple(new_arg_schema) + self.kwargs_schema = origin_schema.kwargs_schema + self._recompute_comparison_key() + + +@dataclass +class OutputSharding: + """ + OutputSharding is a data class that is used by the sharding propagation, + it could set the output_spec upon successful propagation. If needs_redistribute + is set to True, a redistribute_schema would be returned together to indicate + the input arguments needs to be redistributed before the op execution. + + NOTE: the redistribute_schema generated by sharding propagation should be + exactly the same as the operator OpSchema, except the DTensorSpecs + """ + + # specifies the output sharding pattern + output_spec: OutputSpecType + # schema for redistribution if needed + redistribute_schema: OpSchema | None = None + # flag indicating if inputs need redistribution + needs_redistribute: bool = False + # flag to use values from `redistribute_schema` + use_val_from_redistribute_schema: bool = False + + @cached_property + def mesh(self): + if isinstance(self.output_spec, DTensorSpec): + return self.output_spec.mesh + elif isinstance(self.output_spec, tuple): + out_spec = self.output_spec[0] + if isinstance(out_spec, DTensorSpec): + return out_spec.mesh + else: + raise ValueError(f"Unknown output spec type: {type(out_spec)}") + else: + raise ValueError(f"Unknown output spec type: {type(self.output_spec)}") + + +@dataclass +class OpInfo: + """ + All Runtime Op execution info are packed here + """ + + # The first compute device mesh recorded from args + # NOTE: one op could have multiple meshes from its args. We just record the first + # mesh here to check if current rank should participate in computation or not. + compute_mesh: DeviceMesh + + # compete runtime operator infos + # NOTE: schema can be None due to C++ fast path optimization. When the C++ + # dispatch layer (dispatchDTensorOp in python_variable.cpp) finds a cached + # sharding decision, it skips creating the full OpSchema to reduce CPU overhead. + # In this case, OpInfo is created with create_schema=False, setting schema to None. + # The operator information is still available through output_sharding.redistribute_schema + # when redistribution is needed. + schema: OpSchema | None + flat_args_schema: list[object] + local_args: Sequence[object] + local_kwargs: dict[str, object] + args_tree_spec: TreeSpec | None = None + + # the output sharding info + output_sharding: OutputSharding | None = None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..316964412e6255c12e59db25ed092abb023163b6 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__init__.py @@ -0,0 +1,13 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from ._conv_ops import * # noqa: F403 +from ._embedding_ops import * # noqa: F403 +from ._math_ops import * # noqa: F403 +from ._matrix_ops import * # noqa: F403 +from ._pointwise_ops import * # noqa: F403 +from ._random_ops import * # noqa: F403 +from ._tensor_ops import * # noqa: F403 +from ._view_ops import * # noqa: F403 +from .autogen import auto_register_op_variants + + +auto_register_op_variants() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..10248517823faba1db9ad3cac348aebfca3eac93 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_common_rules.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_common_rules.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..446074f06c4eb47b086da8a4e3703a28ac833a67 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_common_rules.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_conv_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_conv_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f0ef2fb19edf9020a95d194d6358b9700674b2cf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_conv_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_einsum_strategy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_einsum_strategy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ea817962c150bbcac7521649ac9f9933ac3f4ee8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_einsum_strategy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_embedding_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_embedding_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7ec9b2bda37cbe5f5fe0f5ab6ab64d01a8e7d888 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_embedding_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_mask_buffer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_mask_buffer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..97ba7e6b8287a81e7a718846b8dbd4dc354a9fd7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_mask_buffer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_math_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_math_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d206762ac6466563fe1962f0b5deeb091c7f3e6c Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_math_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_matrix_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_matrix_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fcbff3a2df4f2db90da35f7b0d9c38f4f9c562ed Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_matrix_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_pointwise_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_pointwise_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..692738f9bde97342603cc5b029a1f55f225e717f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_pointwise_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_random_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_random_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4a2b7fba9ffdca34567cfec627065e83fa218f54 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_random_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_tensor_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_tensor_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..913945649407113319ffc2751557da61dfef06ef Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_tensor_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_view_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_view_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..aaca03210964611591e7ec8c213368c2352849ca Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/_view_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/autogen.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/autogen.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eac5e02013c8ea9f10ade41944988ba6628e29d8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/autogen.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/single_dim_strategy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/single_dim_strategy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1c184872f011cf4963e04b3c2ec692a950674ba0 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/single_dim_strategy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/strategy_validation.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/strategy_validation.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7925f12b244c4b90f6e143dc2d021fe12a496290 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/strategy_validation.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1d33a062da0752aa5634023dc721678216048dd6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_common_rules.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_common_rules.py new file mode 100644 index 0000000000000000000000000000000000000000..7e94f74d93199b09472286a3fc79c2b15bf84004 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_common_rules.py @@ -0,0 +1,288 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import string +from typing import cast + +import torch +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import OpSchema, OutputSharding +from torch.distributed.tensor._ops.utils import prod +from torch.distributed.tensor._utils import compute_local_shape_and_global_offset + + +def _replace_char_in_str(string: str, new_char: str, idx: int) -> str: + return string[:idx] + new_char + string[idx + 1 :] + + +def _gen_reshard_suggestions( + op_schema: OpSchema, + input_dims: list[str], + input_specs: tuple[DTensorSpec, ...], + dim_to_sharding: dict[str, int], + pending_sum: list[int], +) -> OutputSharding: + suggested_arg_specs: list[DTensorSpec] = [] + for input_dim, input_spec in zip(input_dims, input_specs): + dim_map = [dim_to_sharding[dim] for dim in input_dim] + suggested_arg_specs.append( + DTensorSpec.from_dim_map( + mesh=input_spec.mesh, + dim_map=dim_map, + sums=pending_sum, + tensor_meta=input_spec.tensor_meta, + ) + ) + suggested_schema = OpSchema(op_schema.op, tuple(suggested_arg_specs), {}) + suggested_schema._inplace_rewrap_schema_suggestion(op_schema) + return OutputSharding( + None, + redistribute_schema=suggested_schema, + ) + + +def einop_rule( + equation: str, + op_schema: OpSchema, + *, + linearity: bool = False, + enforce_sharding: dict[str, int] | None = None, +) -> OutputSharding: + """ + Propagate the sharding of inputs to output for ops whose data moves according to einsum notation. + + This is mostly borrowed from @zdevito's sharding simulator. Examples: + mk,kn->mn - einsum + ij,ij->ij - addition + ij,j->ij - broadcasted addition + ij->i - reduction + Other ops could use this propagation algorithm when applied, note + that einsum propagation only deal with list of specs (DTensor specs) + as it only works on list of tensors! + + linearity in einop_rule means that the calling op `f` follows this rule: + f(a + b) = f(a) + f(b) + + In this case we can propagate the partial sum, note that linearity in einop + only applies to partial sum, not other operations like min/max (which are + associative but not linear). + """ + # parse einop equation and extract arg specs + inputs, outputs = equation.split("->") + input_dims, output_dims = inputs.split(","), outputs.split(",") + input_specs = op_schema.args_spec + # NOTE: only support single output unless needed in future + output_dim = output_dims[0] + + dim_to_sharding: dict[str, int] = {} + dim_to_size: dict[str, int] = {} + # record pending sum, key is mesh dimension, value is pending sum + # counter across input specs + pending_sums_counter: dict[int, int] = {} + seen_shardings: dict[int, str] = {} + needs_reshard = False + + def merge_sharding(dim: str, a: int, b: int) -> int: + # merge the sharding of inputs if it's able to merge, i.e. we can merge + # replicate and shard to shard, but this will trigger an reshard operation + if a != b: + if a == -1 or b == -1: + # reshard the replicate to match the sharded one + nonlocal needs_reshard + needs_reshard = True + return a if a != -1 else b + else: + # TODO: further merge the sharding properly (i.e. reshard one input to replicate) + raise RuntimeError( + f"{equation}: dim {dim} sharded two different ways: {a} and {b}" + ) + else: + return a + + for input_dim, input_spec in zip(input_dims, input_specs): + # deal with partial sums + input_sums = input_spec.sums + for sum_dim in input_sums: + if sum_dim not in pending_sums_counter: + seen_shardings[sum_dim] = "+" + # update pending sum counter for pending sum mesh + # dimension with the occurrence from each input + pending_sums_counter[sum_dim] = pending_sums_counter.get(sum_dim, 0) + 1 + + for idx, (dim, mesh_dim) in enumerate(zip(input_dim, input_spec.dim_map)): + if enforce_sharding and dim in enforce_sharding: + if enforce_sharding[dim] != mesh_dim: + needs_reshard = True + dim_to_sharding[dim] = enforce_sharding[dim] + dim_to_size[dim] = input_spec.shape[idx] + elif dim not in dim_to_sharding: + dim_to_sharding[dim] = mesh_dim + dim_to_size[dim] = input_spec.shape[idx] + else: + dim_to_sharding[dim] = merge_sharding( + dim, dim_to_sharding[dim], mesh_dim + ) + if dim_to_size[dim] != input_spec.shape[idx]: + raise AssertionError + + # after merging sharding, we check if there're multiple + # sharding on the same mesh dim. + merged_sharding_for_dim = dim_to_sharding[dim] + if merged_sharding_for_dim != -1: + if ( + merged_sharding_for_dim in seen_shardings + and dim != seen_shardings[merged_sharding_for_dim] + ): + needs_reshard = True + seen_shardings[merged_sharding_for_dim] += dim + else: + seen_shardings[merged_sharding_for_dim] = dim + + if pending_sums_counter and not linearity: + # return reshard suggestion with no pending sum, because we already properly + # merge the sharding, this reshard suggestion is legit to use + return _gen_reshard_suggestions( + op_schema, input_dims, input_specs, dim_to_sharding, [] + ) + else: + # It's an op that supports linearity, but not all input arguments are partial + # we fail the sharding propagation with suggestion to make all inputs be + # partial on the corresponding mesh dim (all inputs should be partial for + # the mesh dims in order to execute locally and delay the sum reduction) + for value in pending_sums_counter.values(): + if value != len(input_specs): + needs_reshard = True + + for mesh_dim, dims in seen_shardings.items(): + if len(dims) > 1: + # we found different input dims are being sharded on the same mesh dim + # in order to perform local op computation, we need to reshard inputs + # base on some simple heuristics, now we simply pick the one with least comm + # volume. (i.e. the input with least size) + # TODO: consider a more advanced heuristic to pick the best sharding + costs = [] + for d in dims: + cost = 0 + for input_dim, input_spec in zip(input_dims, input_specs): + if ( + d in input_dim + and input_spec.dim_map[input_dim.index(d)] == mesh_dim + ): + if input_spec.tensor_meta is None: + raise AssertionError + global_shape = input_spec.tensor_meta.shape + local_shape, _ = compute_local_shape_and_global_offset( + global_shape, + input_spec.mesh, + input_spec.placements, + skip_offset=True, + ) + cost += prod(local_shape) * input_spec.mesh.size(mesh_dim) + + costs.append(cost) + d_to_keep_sharding = dims[costs.index(max(costs))] + for d in dims: + # update dim_to_sharding to keep the sharding of the dim with + # highest comm and make the rest of the dims to replicate + if d != d_to_keep_sharding: + dim_to_sharding[d] = -1 + + pending_sums = list(pending_sums_counter.keys()) + if needs_reshard: + return _gen_reshard_suggestions( + op_schema, input_dims, input_specs, dim_to_sharding, pending_sums + ) + + # generate output pending sum if a dim is sharded, and it appears in input + # but not output + for dim, shard_on_mesh in dim_to_sharding.items(): + if dim not in output_dims[0] and shard_on_mesh != -1: + pending_sums.append(shard_on_mesh) + + # if no need to reshard, we directly generate the output sharding + output_dim_map = [] + output_shape = [] + for dim in output_dim: + if dim == "1": + # find output dim that is a singleton dimension, mark sharding and shape + output_dim_map.append(-1) + output_shape.append(1) + else: + output_dim_map.append(dim_to_sharding[dim]) + output_shape.append(dim_to_size[dim]) + + # XXX: since we still need to have intermediate shape calculation, we need + # to pass in the shape here. We should remove this once sharding decomp works + # for ops like addmm + if input_specs[0].tensor_meta is None: + raise AssertionError + tensor_meta = TensorMeta( + torch.Size(output_shape), + input_specs[0].tensor_meta.stride, + input_specs[0].tensor_meta.dtype, + ) + return OutputSharding( + DTensorSpec.from_dim_map( + input_specs[0].mesh, + output_dim_map, + pending_sums, + tensor_meta=tensor_meta, + ) + ) + + +def pointwise_rule(op_schema: OpSchema, linearity: bool = False) -> OutputSharding: + """ + Propagate the sharding for pointwise operations. + + Examples: + ij,ij->ij - addition/mul + ij,j->ij - broadcasted addition + """ + alphabet = string.ascii_lowercase + # find the max_dim first in case we need to broadcasting + input_specs = op_schema.args_spec + max_dim = max(input.ndim for input in input_specs) + dimchars = [] + singleton_counter: list[int] = [0] * max_dim + for input in input_specs: + start_dim = max_dim - input.ndim + p = alphabet[start_dim:max_dim] + # handle the "broadcasting to a common shape case" + # see https://pytorch.org/docs/stable/notes/broadcasting.html + # If any of the dimensions is singleton dimension (i.e. 1). + # we mark the dim char as a special "1" to distinguish with + # the non-singleton dimension, so that sharding propagation + # should just ignore the singleton dimension. + if len(input_specs) > 1: + for i in range(max_dim): + if i < start_dim: + # treat the leading miss dim chars as singleton + singleton_counter[i] += 1 + elif input.shape[i - start_dim] == 1: + # mark singleton dim char as a special "1" in einop rule + singleton_counter[i] += 1 + p = _replace_char_in_str(p, "1", (i - start_dim)) + + dimchars.append(p) + out_dimchars = alphabet[:max_dim] + # check if we replace the all inputs dim char with singleton dimension, + # if we replace all inputs, we also need to replace the output dimension. + for output_dim_idx in range(len(out_dimchars)): + if singleton_counter[output_dim_idx] == len(input_specs): + out_dimchars = _replace_char_in_str(out_dimchars, "1", output_dim_idx) + + fmt = f"{','.join(p for p in dimchars)}->{out_dimchars}" + + enforce_sharding: dict[str, int] = {} + if op_schema.is_inplace_op(): + follow_spec = op_schema.args_spec[0] + enforce_sharding.update(zip(out_dimchars, follow_spec.dim_map)) + elif op_schema.is_out_variant_op(): + follow_spec = cast(DTensorSpec, op_schema.kwargs_schema["out"]) + enforce_sharding.update(zip(out_dimchars, follow_spec.dim_map)) + + return einop_rule( + fmt, + op_schema, + linearity=linearity, + enforce_sharding=enforce_sharding, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_conv_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_conv_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..f8d3ddcfee9a0006defd7c0bb5c63988760d30d8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_conv_ops.py @@ -0,0 +1,321 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +# implement matrix related ops for distributed tensor + +from typing import Any + +import torch +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpSchema, + OutputSharding, + RuntimeSchemaInfo, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import register_prop_rule +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) +from torch.fx.experimental.symbolic_shapes import guard_or_false + + +aten = torch.ops.aten + + +@register_prop_rule(aten.convolution.default) +def convolution_rules(op_schema: OpSchema) -> OutputSharding: + ( + input_spec, + weight_spec, + bias_spec, + stride, + padding, + dilation, + _transposed, + _output_padding, + _groups, + ) = op_schema.args_schema + + if not isinstance(input_spec, DTensorSpec): + raise AssertionError + if not isinstance(weight_spec, DTensorSpec): + raise AssertionError + # bias_spec can be None (optional parameter in aten.convolution schema) + if bias_spec is not None: + if not isinstance(bias_spec, DTensorSpec): + raise AssertionError + if input_spec.tensor_meta is None: + raise AssertionError + if weight_spec.tensor_meta is None: + raise AssertionError + in_shape = input_spec.tensor_meta.shape + weight_shape = weight_spec.tensor_meta.shape + if not isinstance(stride, list): + raise AssertionError(f"stride must be list, got {type(stride)}") + if not isinstance(padding, list): + raise AssertionError(f"padding must be list, got {type(padding)}") + if not isinstance(dilation, list): + raise AssertionError(f"dilation must be list, got {type(dilation)}") + # weight_shape might not be torch.Size in all cases (e.g., SymIntArrayRef during tracing) + # so we don't assert its type, just use it + out_conv_shape = [ + (d + 2 * padding[i] - dilation[i] * (weight_shape[i + 1] - 1) - 1) // stride[i] + + 1 + for (i, d) in enumerate(in_shape[2:]) + ] + output_shape = [in_shape[0], weight_shape[0]] + out_conv_shape + output_stride = [1] + for i in range(1, len(output_shape)): + output_stride.insert(0, output_stride[0] * output_shape[-i]) + output_dim_map = input_spec.dim_map + pending_sums = input_spec.sums + + tensor_meta = TensorMeta( + torch.Size(output_shape), + tuple(output_stride), + input_spec.tensor_meta.dtype, + ) + return OutputSharding( + DTensorSpec.from_dim_map( + input_spec.mesh, + output_dim_map, + pending_sums, + tensor_meta=tensor_meta, + ) + ) + + +@register_prop_rule(aten.convolution_backward.default) +def convolution_backward_rules(op_schema: OpSchema) -> OutputSharding: + input_spec = op_schema.args_schema[0] + ( + grad_output_spec, + input_spec, + weight_spec, + bias_shape_opt, + _stride, + _padding, + _dilation, + _transposed, + _output_padding, + _groups, + _output_mask, + ) = op_schema.args_schema + + if not isinstance(grad_output_spec, DTensorSpec): + raise AssertionError + if not isinstance(input_spec, DTensorSpec): + raise AssertionError + if not isinstance(weight_spec, DTensorSpec): + raise AssertionError + # bias_shape_opt can be None (optional parameter in aten.convolution_backward schema) + if bias_shape_opt is not None: + if not isinstance(bias_shape_opt, list): + raise AssertionError + if input_spec.tensor_meta is None: + raise AssertionError + weight_tensor_meta = weight_spec.tensor_meta + + # Only create bias_tensor_meta if bias_shape_opt is not None + if bias_shape_opt is not None: + bias_tensor_meta = TensorMeta( + torch.Size(bias_shape_opt), + (1,), + input_spec.tensor_meta.dtype, + ) + else: + bias_tensor_meta = None + + grad_input_spec = input_spec + grad_weight_spec = DTensorSpec.from_dim_map( + input_spec.mesh, + [-1, -1, -1, -1], + [0], + tensor_meta=weight_tensor_meta, + ) + + # Only create grad_bias_spec if we have bias_tensor_meta + if bias_tensor_meta is not None: + grad_bias_spec = DTensorSpec.from_dim_map( + input_spec.mesh, + [-1], + [0], + tensor_meta=bias_tensor_meta, + ) + else: + grad_bias_spec = None + + # TODO: actually the output_mask is not respected here, we should + # set the corresponding spec to `None` if the output_mask is not `False` + # for a certain output Tensor. This also applies to the conv handler + # in torch/distributed/tensor/_tp_conv.py + return OutputSharding([grad_input_spec, grad_weight_spec, grad_bias_spec]) + + +# Single-dim strategies for autoparallel optimizer support. +# These coexist with the prop_rules above — strategies take precedence +# in the propagation path, while the prop_rules + custom handlers in +# _tp_conv.py continue to handle runtime dispatch. + + +def _supports_last_dim_sharding( + input_meta: TensorMeta, + weight_meta: TensorMeta, + stride: list[Any], + padding: list[Any], + dilation: list[Any], + transposed: bool, +) -> bool: + ndim = len(input_meta.shape) + return ( + ndim == len(weight_meta.shape) + and ndim in (3, 4, 5) + and not transposed + and guard_or_false(padding[-1] == 0) + and guard_or_false(dilation[-1] == 1) + and guard_or_false(stride[-1] == weight_meta.shape[-1]) + ) + + +def _convolution_full_mesh_strategy_filter( + mesh: DeviceMesh, + op_schema: OpSchema, + input_specs: list[DTensorSpec], + _output_specs: DTensorSpec | tuple[DTensorSpec | None, ...], +) -> bool: + if op_schema.op == aten.convolution.default: + input_idx, stride_idx = 0, 3 + elif op_schema.op == aten.convolution_backward.default: + input_idx, stride_idx = 1, 4 + else: + raise AssertionError(f"Unexpected op {op_schema.op}") + + input_spec = input_specs[input_idx] + last_dim = len(input_spec.shape) - 1 + last_dim_mesh_dims = [ + mesh_dim + for mesh_dim, placement in enumerate(input_spec.placements) + if placement.is_shard(last_dim) + ] + if not last_dim_mesh_dims: + return True + # _tp_conv converts placements to dim_map, which cannot represent one tensor + # dimension sharded over multiple mesh dimensions. + if len(last_dim_mesh_dims) != 1: + return False + + last_dim_mesh_dim = last_dim_mesh_dims[0] + last_dim_placement = input_spec.placements[last_dim_mesh_dim] + if not isinstance(last_dim_placement, Shard) or isinstance( + last_dim_placement, _StridedShard + ): + return False + + stride = op_schema.args_schema[stride_idx] + if not isinstance(stride, list): + raise AssertionError(f"Expected stride list, got {type(stride)}") + return guard_or_false( + input_spec.shape[last_dim] % (stride[-1] * mesh.size(last_dim_mesh_dim)) == 0 + ) + + +@register_single_dim_strategy( + [aten.convolution.default], + schema_info=RuntimeSchemaInfo(2), + full_mesh_strategy_filter=_convolution_full_mesh_strategy_filter, +) +def convolution_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + weight_meta = args_schema[1] + if not isinstance(weight_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(weight_meta)}") + bias_meta = args_schema[2] + stride, padding, dilation = args_schema[3:6] + transposed = args_schema[6] + # [output, input, weight, (bias)] + rule: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(0), # output + _ShardingPlaceholder(0), # input + Replicate(), # weight + ] + if bias_meta is not None: + rule.append(Replicate()) # bias + strategies = [rule] + + if _supports_last_dim_sharding( + input_meta, weight_meta, stride, padding, dilation, transposed + ): + last_dim = len(input_meta.shape) - 1 + last_dim_rule: list[Placement | _ShardingPlaceholder] = [ + Shard(last_dim), # output + Shard(last_dim), # input + Replicate(), # weight + ] + if bias_meta is not None: + last_dim_rule.append(Replicate()) # bias + strategies.append(last_dim_rule) + + return strategies + + +@register_single_dim_strategy( + [aten.convolution_backward.default], + schema_info=RuntimeSchemaInfo(3), + full_mesh_strategy_filter=_convolution_full_mesh_strategy_filter, +) +def convolution_backward_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder | None]]: + input_meta = args_schema[1] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + weight_meta = args_schema[2] + if not isinstance(weight_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(weight_meta)}") + bias_sizes = args_schema[3] + stride, padding, dilation = args_schema[4:7] + transposed = args_schema[7] + has_bias = bias_sizes is not None + # outputs: [grad_input, grad_weight, grad_bias] + # inputs: [grad_output, input, weight] + rule: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(0), # grad_input + Partial("sum"), # grad_weight + Partial("sum") if has_bias else None, # grad_bias + _ShardingPlaceholder(0), # grad_output + _ShardingPlaceholder(0), # input + Replicate(), # weight + ] + strategies = [rule] + + if _supports_last_dim_sharding( + input_meta, weight_meta, stride, padding, dilation, transposed + ): + last_dim = len(input_meta.shape) - 1 + strategies.append( + [ + Shard(last_dim), # grad_input + Partial("sum"), # grad_weight + Partial("sum") if has_bias else None, # grad_bias + Shard(last_dim), # grad_output + Shard(last_dim), # input + Replicate(), # weight + ] + ) + + return strategies diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_einsum_strategy.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_einsum_strategy.py new file mode 100644 index 0000000000000000000000000000000000000000..fc60980f4949dbf0a2a2642c772078d3ac1a1e97 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_einsum_strategy.py @@ -0,0 +1,195 @@ +import itertools +from dataclasses import dataclass + +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec +from torch.distributed.tensor._op_schema import OpSpec, OpStrategy +from torch.distributed.tensor.placement_types import ( + Partial, + Placement, + Replicate, + Shard, +) + + +@dataclass +class EinsumDims: + contracting_dims: list[str] + batch_dims: list[str] + lhs_out_only_dims: list[str] + rhs_out_only_dims: list[str] + + @classmethod + def parse_equation(cls, equation: str) -> tuple[list[str], str]: + # parse einop equation and extract arg specs + """ + Parse the einsum equation str to input dim chars and output dim char + """ + inputs, outputs = equation.split("->") + input_dims, output_dims = inputs.split(","), outputs.split(",") + + # NOTE: only support at most two inputs, and single output + # extend to support more inputs if needed in future + if len(input_dims) > 2: + raise AssertionError("Only support at most two inputs") + if len(output_dims) != 1: + raise AssertionError("Only support single output") + output_dim = output_dims[0] + return input_dims, output_dim + + @classmethod + def parse_dims(cls, input_dims: list[str], output_dim: str) -> "EinsumDims": + """ + Parse the dims and extract the contracting, batch, and free dimensions + for the left and right hand sides. + """ + dim_char_set: set[str] = set() + for input_dim in input_dims: + dim_char_set.update(input_dim) + + # get a deterministic order of all dim chars + all_dim_chars = sorted(dim_char_set) + + # parse input and output dimensions + lhs_out_only_dims, rhs_out_only_dims = [], [] + batch_dims, contracting_dims = [], [] + + for dim_char in all_dim_chars: + if dim_char not in output_dim: + contracting_dims.append(dim_char) + else: + is_batch_dim = True + for input_dim in input_dims: + is_batch_dim = is_batch_dim and dim_char in input_dim + + if is_batch_dim: + batch_dims.append(dim_char) + else: + if len(input_dims) != 2: + raise AssertionError( + "free dimension only supported for two inputs!" + ) + lhs, rhs = input_dims + if dim_char in lhs: + lhs_out_only_dims.append(dim_char) + elif dim_char in rhs: + rhs_out_only_dims.append(dim_char) + else: + raise RuntimeError("Invalid dimension character") + + return cls( + contracting_dims=contracting_dims, + batch_dims=batch_dims, + lhs_out_only_dims=lhs_out_only_dims, + rhs_out_only_dims=rhs_out_only_dims, + ) + + +def gen_einsum_strategies( + equation: str, + mesh: DeviceMesh, + *, + linearity: bool = False, +) -> OpStrategy: + """ + Generate a strategy list for the ops that follow einsum style notation. + + In principle, each mesh dim is independent of other device mesh dim when we + generate strategies. So we generate strategy over each device mesh dim and + do product combination on all mesh dims. We basically follow the below rule + for each device mesh dim: + + 1. Shard on contracting dim: When both inputs shard on contracting dim over + the same device dim. The result will be Partial over that device dim. + + 2. Shard on noncontracting dim: + 2.1: Shard on batch dim: output, both inputs all should shard on batch + dim. + 2.2: Shard on lhs only dim or rhs only dim: both output and lhs or rhs + input should shard on this free dim. + + 3. Linearity (Partial): If enabled, set Partial on output and inputs over + the same device mesh dim. + """ + # parse einop equation and extract dims + input_dims, output_dim = EinsumDims.parse_equation(equation) + edims = EinsumDims.parse_dims(input_dims, output_dim) + all_mesh_dim_strategies = [] + + # generate strategies for each mesh dim and do cartesian product for final strategy. E.g., for a 2D mesh, we can have [P(),R,R] + strategies_over_one_mesh_dim = [] + + # placement list stores placements of [output, input1, input2, ...] + # first we always have replicate all for inputs and output + placement_list: list[Placement] = [Replicate()] * (len(input_dims) + 1) + strategies_over_one_mesh_dim.append(placement_list) + + # split batch dim + for batch_dim in edims.batch_dims: + output_batch_dim = output_dim.index(batch_dim) + placement_list = [Shard(output_batch_dim)] + for input_dim in input_dims: + input_batch_dim = input_dim.index(batch_dim) + placement_list.append(Shard(input_batch_dim)) + + strategies_over_one_mesh_dim.append(placement_list) + + # split contracting dim + # NOTE: This is the only strategy that produces a Partial output, and it + # hardcodes Partial("sum"). No strategy accepts Partial as an input + # placement, so Partial inputs are always redistributed to Shard or + # Replicate. This means Partial("avg") inputs cannot be preserved through + # the op. Use gen_single_dim_einsum_strategies with per-input linearity + # for proper Partial support. + for contracting_dim in edims.contracting_dims: + # Contracting dim can shard on same device axis for both inputs. This + # results in the output being Partial on that device axis. For example: + # bmk_{x},k_{x}n -> bmn{Ux} (becomes partial over device axis x) + placement_list = [Partial()] + for input_dim in input_dims: + input_contracting_dim = input_dim.index(contracting_dim) + placement_list.append(Shard(input_contracting_dim)) + + strategies_over_one_mesh_dim.append(placement_list) + + # split lhs free dim + for lhs_dim in edims.lhs_out_only_dims: + lhs_free_dim_output = output_dim.index(lhs_dim) + lhs_free_dim_input = input_dims[0].index(lhs_dim) + # this means split the lhs input and output + # i.e. S(0), R -> S(0) + lhs_placement_list: list[Placement] = [ + Shard(lhs_free_dim_output), + Shard(lhs_free_dim_input), + Replicate(), + ] + strategies_over_one_mesh_dim.append(lhs_placement_list) + + # split rhs free dim + for rhs_dim in edims.rhs_out_only_dims: + rhs_free_dim_output = output_dim.index(rhs_dim) + rhs_free_dim_input = input_dims[1].index(rhs_dim) + rhs_placement_list: list[Placement] = [ + Shard(rhs_free_dim_output), + Replicate(), + Shard(rhs_free_dim_input), + ] + strategies_over_one_mesh_dim.append(rhs_placement_list) + + # linearity strategy + if linearity: + linearity_placement_list: list[Placement] = [Partial()] + for _ in input_dims: + linearity_placement_list.append(Partial()) + strategies_over_one_mesh_dim.append(linearity_placement_list) + + # generate strategies for entire mesh + all_mesh_dim_strategies = [strategies_over_one_mesh_dim] * mesh.ndim + strategy_combs = itertools.product(*all_mesh_dim_strategies) + all_strategies = [] + for strategy_comb in strategy_combs: + spec_list = [DTensorSpec(mesh, tuple(specs)) for specs in zip(*strategy_comb)] + strat = OpSpec(output_specs=spec_list[0], input_specs=spec_list[1:]) + all_strategies.append(strat) + + return OpStrategy(all_strategies) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_embedding_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_embedding_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..a0051c0b77a3a875e0d9714f9f971c13c96694a0 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_embedding_ops.py @@ -0,0 +1,97 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import torch +from torch._ops import OpOverload +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ArgsType, KwargsType, RuntimeSchemaInfo +from torch.distributed.tensor._ops.single_dim_strategy import ( + register_single_dim_strategy, +) +from torch.distributed.tensor.placement_types import ( + _MaskPartial, + Partial, + Placement, + Replicate, + Shard, +) + + +aten = torch.ops.aten + + +@register_single_dim_strategy(aten.embedding.default) +def embedding_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement]]: + """Single-dim strategy for embedding: rowwise, colwise, and batch-dim sharding. + + Placement order: [output, weight, indices] + """ + weight_meta = args_schema[0] + indices_meta = args_schema[1] + if not isinstance(weight_meta, TensorMeta) or not isinstance( + indices_meta, TensorMeta + ): + raise AssertionError + + # _MaskPartial hashes offset_shape, but torch.Size with SymInt (from + # dynamo tracing) is unhashable. Concretize to int, which adds a + # standard dynamo guard (recompiles if the shape changes at runtime). + weight_shape = torch.Size(int(s) for s in weight_meta.shape) + indices_shape = indices_meta.shape + output_emb_dim = len(indices_shape) + + strategies: list[list[Placement]] = [] + + # colwise: output shard on last dim, weight shard on dim 1, indices replicate + strategies.append([Shard(output_emb_dim), Shard(1), Replicate()]) + + # rowwise: output is MaskPartial, weight shard on dim 0, indices MaskPartial + # NOTE: same object for output & indices so the mask buffer is shared + embedding_partial = _MaskPartial(offset_shape=weight_shape, offset_dim=0) + strategies.append([embedding_partial, Shard(0), embedding_partial]) + + # batch dim sharding: weight replicated, indices shard on any dim, output follows + for i in range(len(indices_shape)): + strategies.append([Shard(i), Replicate(), Shard(i)]) + + return strategies + + +@register_single_dim_strategy( + aten.embedding_dense_backward.default, + schema_info=RuntimeSchemaInfo(static_argnum=2), +) +def embedding_dense_backward_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement]]: + """Single-dim strategy for embedding backward. + + Placement order: [output(weight_grad), grad_out, indices] + """ + grad_out_meta = args_schema[0] + indices_meta = args_schema[1] + if not isinstance(grad_out_meta, TensorMeta) or not isinstance( + indices_meta, TensorMeta + ): + raise AssertionError + + grad_out_ndim = len(grad_out_meta.shape) + indices_shape = indices_meta.shape + + strategies: list[list[Placement]] = [] + + # colwise backward: weight grad shard on dim 1, grad_out shard on last dim, indices replicate + strategies.append([Shard(1), Shard(grad_out_ndim - 1), Replicate()]) + + # batch dim sharding: weight grad partial, grad_out/indices shard on same dim + for i in range(len(indices_shape)): + strategies.append([Partial(), Shard(i), Shard(i)]) + + # grad_out partial, indices replicate, weight grad partial + strategies.append([Partial(), Partial(), Replicate()]) + + return strategies diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_mask_buffer.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_mask_buffer.py new file mode 100644 index 0000000000000000000000000000000000000000..a28074184b205f01f9482e1252bb176d7b2e2713 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_mask_buffer.py @@ -0,0 +1,44 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +from dataclasses import dataclass + +import torch + + +@dataclass +class MaskBuffer: + data: torch.Tensor | None = None + # refcount allows shared usage of the MaskBuffer, as long as all users have the same data + refcount: int = 0 + + def materialize_mask(self, mask): + if self.refcount == 0: + self.data = mask + else: + if self.data is None: + raise AssertionError + if not torch.equal(self.data, mask): + raise RuntimeError( + "MaskBuffer has been materialized with conflicting data" + ) + self.refcount += 1 + + def release_mask(self): + if self.refcount == 0 or self.data is None: + raise RuntimeError("MaskBuffer has not been materialized") + self.refcount -= 1 + if self.refcount == 0: + self.data = None + + def apply_mask(self, tensor): + if self.refcount == 0 or self.data is None: + raise RuntimeError("MaskBuffer has not been materialized") + + # NOTE: MaskPartial is being used by the embedding op and the gather op. + # For gather, the mask has the same dimension as the output tensor, whereas + # the output of the embedding op has an additional dimension compare to the input, + # hence the output masking logic below having two different cases. + if tensor.ndim == self.data.ndim: + tensor[self.data] = 0.0 + else: + tensor[self.data, :] = 0.0 diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_math_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_math_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..d6f7a19147198b840fd63aa15a0e07da21229f00 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_math_ops.py @@ -0,0 +1,2043 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import math +from collections.abc import Sequence +from dataclasses import dataclass +from enum import Enum +from typing import Any, cast + +import torch +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpSchema, + OpSpec, + OpStrategy, + RuntimeSchemaInfo, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import ( + as_list, + generate_redistribute_costs, + is_tensor_evenly_shardable_on_dim, + normalize_dim, + normalize_dims, +) +from torch.distributed.tensor._utils import ( + compute_local_shape_and_global_offset, + normalize_to_torch_size, +) +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) +from torch.fx.experimental.symbolic_shapes import guard_or_false + + +aten = torch.ops.aten +prims = torch.ops.prims + + +class Reduction(Enum): + NONE = 0 + MEAN = 1 + SUM = 2 + + +@dataclass(frozen=True) +class NormReduction: + norm_type: int | float + + +ReductionOpType = NormReduction | str + + +@dataclass(frozen=True) +class _NormPartial(Partial): + """ + This placement is used for partial p-norm (p not in {inf, -inf, 0}). + + For p-norms, the p-norm over n elements computes (sum_i x_i^p)^(1/p). + For example, consider 2 ranks, a (4,) tensor sharded on dim-0, and 2-norm: + Rank 0: [t1, t2] | Rank 1: [t3, t4] + After computing 2-norm per gradient (partial placement): + Rank 0: [sqrt(t1^2 + t2^2)] | Rank 1: [sqrt(t3^2 + t4^2)] + Converting from partial to replicate wants to ultimately get: + Rank 0/1: [sqrt(t1^2 + t2^2 + t3^2 + t4^2)] + This is achieved by: x^p -> allreduce sum -> x^(1/p). + """ + + norm_type: int | float = 2 + + def __init__(self, norm_type: int | float = 2): + super().__init__("sum") + object.__setattr__(self, "norm_type", norm_type) + + def _partition_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + return tensor / math.pow(mesh.size(mesh_dim), 1 / self.norm_type) + + def _reduce_shard_value( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + shard_spec: Placement, + ) -> torch.Tensor: + if not isinstance(shard_spec, Shard): + raise AssertionError(f"Expected Shard, got {type(shard_spec)}") + tensor = self._pre_reduce_transform(tensor) + reduced_tensor = super()._reduce_shard_value(tensor, mesh, mesh_dim, shard_spec) + return self._post_reduce_transform(reduced_tensor) + + def _reduce_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + tensor = self._pre_reduce_transform(tensor) + reduced_tensor = super()._reduce_value(tensor, mesh, mesh_dim) + return self._post_reduce_transform(reduced_tensor) + + def _pre_reduce_transform(self, tensor: torch.Tensor) -> torch.Tensor: + return tensor**self.norm_type + + def _post_reduce_transform(self, tensor: torch.Tensor) -> torch.Tensor: + return tensor ** (1.0 / self.norm_type) + + def __eq__(self, other: object) -> bool: + if not isinstance(other, _NormPartial): + return False + return self.norm_type == other.norm_type + + def __hash__(self) -> int: + return 1 + hash(self.norm_type) + + def __repr__(self) -> str: + return f"_NormPartial({self.norm_type})" + + def __str__(self) -> str: + return f"_NormP({self.norm_type})" + + +def _infer_reduction_dims(dims_arg: object, ndim: int) -> list[int] | None: + if dims_arg is None: + return None + dims = cast(list[int], as_list(dims_arg)) + dims = cast(list[int], normalize_dims(dims, ndim)) + empty_dims = [[0], [-1], []] + if ndim == 0 and dims_arg in empty_dims: + return None + return dims + + +def _infer_reduce_dims_map( + reduction_dims: list[int], input_ndim: int, keep_dim=False +) -> list[int]: + reduction_dims_map = [] + new_dim_count = 0 + for input_dim in range(input_ndim): + if input_dim in reduction_dims and not keep_dim: + # if input dim in reduction dims, mark it as -1 + reduction_dims_map.append(-1) + else: + # otherwise mark it as the new dim + reduction_dims_map.append(new_dim_count) + new_dim_count += 1 + + return reduction_dims_map + + +def _replicate_dims_start_at( + placements: Sequence[Placement], start_dim: int = 0 +) -> tuple[Placement, ...]: + new_placements: list[Placement] = [] + for p in placements: + if p.is_partial() or (_is_shard_like(p) and p.dim >= start_dim): + new_placements.append(Replicate()) # make it replicate + else: + new_placements.append(p) # keep the placement + return tuple(new_placements) + + +# return new_placements which align with placements but skip the skipped_dim +# Precondition: no shard-like placement on skipped_dim (callers must +# replicate it first via replicate_reduction_dims). +def _skip_dim( + placements: tuple[Placement, ...], skipped_dim: int +) -> tuple[Placement, ...]: + new_placements: list[Placement] = [] + for p in placements: + if isinstance(p, _StridedShard) and p.dim >= skipped_dim: + new_placements.append(_StridedShard(p.dim - 1, split_factor=p.split_factor)) + elif isinstance(p, Shard) and p.dim >= skipped_dim: + new_placements.append(Shard(p.dim - 1)) + else: + new_placements.append(p) + return tuple(new_placements) + + +def replicate_reduction_dims( + placements: tuple[Placement, ...], reduction_dims: list[int] +) -> tuple[Placement, ...]: + # replicate the reduction dims if not reduction_linear + new_placements: list[Placement] = [] + + for p in placements: + if p.is_partial(): + new_placements.append(Replicate()) + elif _is_shard_like(p) and p.dim in reduction_dims: + new_placements.append(Replicate()) + else: + new_placements.append(p) + + return tuple(new_placements) + + +def map_placements_after_reduction( + placements: tuple[Placement, ...], + reduction_dims: list[int], + reduction_dims_map: list[int], + reduction_op: ReductionOpType, +) -> tuple[Placement, ...]: + """ + Map each placement based on the output shape after reduction. + """ + new_placements: list[Placement] = [] + for placement in placements: + if isinstance(placement, (Replicate, Partial)): + new_placements.append(placement) + else: + if not _is_shard_like(placement): + raise AssertionError( + f"Expected Shard/_StridedShard, got {type(placement)}" + ) + shard_dim = placement.dim + new_shard_dim = reduction_dims_map[shard_dim] + if new_shard_dim == -1 or shard_dim in reduction_dims: + # if new_shard_dim collapsed or its in the reduction dims + # (i.e. for the case where keepdims=True), we generate partial + new_placements.append(get_placement_from_reduction_op(reduction_op)) + else: + if isinstance(placement, _StridedShard): + new_placements.append( + _StridedShard( + new_shard_dim, split_factor=placement.split_factor + ) + ) + elif isinstance(placement, Shard): + new_placements.append(Shard(new_shard_dim)) + return tuple(new_placements) + + +def get_placement_from_reduction_op(reduction_op: ReductionOpType) -> Placement: + if isinstance(reduction_op, NormReduction): + if reduction_op.norm_type == 0: + # return P(sum) for easier reduction_linear handling. + return Partial("sum") + return _NormPartial(norm_type=reduction_op.norm_type) + return Partial(reduction_op) + + +def common_reduction_strategy( + input_strategy: OpStrategy, + reduce_dims: list[int], + keep_dim: bool = False, + reduction_linear: bool = True, + reduction_op: ReductionOpType = "sum", +) -> OpStrategy: + """ + reduction_linear means that the reduction `f` follows this rule: + f([f(a), f(b)]) = f([a, b]) + + reduction linear should be super set of linearity. + """ + reduction_strategy = OpStrategy([]) + + for op_spec in input_strategy.strategies: + is_reduction_linear = reduction_linear + if reduction_op == "avg": + output_spec = op_spec.output_spec + local_shape = list(output_spec.tensor_meta.shape) # type:ignore[union-attr] + for dim in reduce_dims: + if not is_tensor_evenly_shardable_on_dim(local_shape, output_spec, dim): + is_reduction_linear = False + break + + for p in op_spec.output_spec.placements: + if isinstance(p, Partial) and p.reduce_op != reduction_op: + is_reduction_linear = False + break + + if not is_reduction_linear: + input_placements = replicate_reduction_dims( + op_spec.output_spec.placements, reduce_dims + ) + else: + input_placements = op_spec.output_spec.placements + + input_spec = DTensorSpec( + mesh=input_strategy.mesh, + placements=input_placements, + tensor_meta=op_spec.output_spec.tensor_meta, + ) + + reduce_dims_map = _infer_reduce_dims_map(reduce_dims, input_spec.ndim, keep_dim) + out_placements = map_placements_after_reduction( + input_spec.placements, reduce_dims, reduce_dims_map, reduction_op + ) + reduction_strategy.strategies.append( + OpSpec( + output_specs=DTensorSpec( + mesh=input_strategy.mesh, + placements=out_placements, + ), + input_specs=(input_spec,), + redistribute_cost=[ + generate_redistribute_costs(input_strategy, input_spec) + ], + ) + ) + + return reduction_strategy + + +def _reduction_single_dim_strategy( + args_schema: tuple[Any, ...], + reduction_dims: list[int] | None, + keep_dim: bool, + reduction_linear: bool, + reduction_op: ReductionOpType, + extra_partial_rules: list[list[Placement | _ShardingPlaceholder]] | None = None, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Build single-dim strategies for reduction ops. + + For non-reduction dims: shard input and output on that dim (with dim remapping). + For reduction dims (if reduction_linear): shard input on reduction dim, output is + Partial. Also adds partial propagation rules for linear reductions. + + "avg" reductions do not shard reduced dims because averaging uneven local + averages does not produce the global average. + """ + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + reduce_dims = list(range(ndim)) if reduction_dims is None else reduction_dims + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + + for d in range(ndim): + if d in reduce_dims: + if reduction_linear and reduction_op != "avg": + # Shard on reduction dim -> Partial output. + strategies.append( + [ + get_placement_from_reduction_op(reduction_op), + _ShardingPlaceholder(d), + ] + ) + else: + # Shard on non-reduction dim -> shard output (with dim remapping) + if keep_dim: + out_d = d + else: + out_d = d - sum(1 for rd in reduce_dims if rd < d) + strategies.append([_ShardingPlaceholder(out_d), _ShardingPlaceholder(d)]) + + # Partial propagation: Partial(reduction_op) input -> Partial(reduction_op) output. + # Skip for NormReduction: the old OpStrategy path did not propagate + # _NormPartial through, so keep that behavior. + if reduction_linear and not isinstance(reduction_op, NormReduction): + partial_placement = get_placement_from_reduction_op(reduction_op) + strategies.append([partial_placement, partial_placement]) + + if extra_partial_rules: + strategies.extend(extra_partial_rules) + + return strategies + + +# Category C: Linear reductions +# all/any are listed here because the remaining reduction helpers share this map, +# but they are registered separately with reduction_linear=False. +LINEAR_REDUCTION_OP_MAP = { + aten.all.default: "product", + aten.all.dim: "product", + aten.any.default: "sum", + aten.any.dim: "sum", + aten.any.dims: "sum", + aten.any.out: "sum", + aten.sum.default: "sum", + aten.sum.dim_IntList: "sum", + prims.sum.default: "sum", + # These are only valid when there is no padding + aten.prod.default: "product", + aten.prod.dim_int: "product", + aten.prod.int_out: "product", + prims.prod.default: "product", + aten.max.default: "max", + aten.max.out: "max", + aten.min.default: "min", + aten.min.out: "min", + aten.amax.default: "max", + aten.amax.out: "max", + prims.amax.default: "max", + aten.amin.default: "min", + aten.amin.out: "min", + prims.amin.default: "min", + aten.nansum.default: "sum", +} + +FOREACH_LINEAR_REDUCTION_OP_MAP = { + aten._foreach_max.default: "max", +} + +FOREACH_LINEAR_REDUCTION_OPS = list(FOREACH_LINEAR_REDUCTION_OP_MAP) + +# all/any are NOT linear reductions: validation showed S(reduction_dim)->P(...) +# produces incorrect results. They only support sharding on non-reduction dims. +NON_LINEAR_BOOL_REDUCTION_OPS = [ + aten.all.default, + aten.all.dim, + aten.any.default, + aten.any.dim, + aten.any.dims, + aten.any.out, +] + +MEAN_REDUCTION_OPS = [ + aten.mean.default, + aten.mean.dim, + aten.mean.out, +] + +ARGMAX_ARGMIN_OPS = { + aten.argmax.default: "max", + aten.argmin.default: "min", +} + + +LINEAR_REDUCTION_OPS = [ + op for op in LINEAR_REDUCTION_OP_MAP if op not in NON_LINEAR_BOOL_REDUCTION_OPS +] + + +@register_single_dim_strategy( + FOREACH_LINEAR_REDUCTION_OPS, + schema_info=RuntimeSchemaInfo(1, needs_pytree=True), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +@register_single_dim_strategy( + LINEAR_REDUCTION_OPS, + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def linear_reduction_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dims = None + if len(args_schema) > 1: + dims = _infer_reduction_dims(args_schema[1], ndim) + + keep_dim = len(args_schema) > 2 and bool(args_schema[2]) + reduction_op = ( + FOREACH_LINEAR_REDUCTION_OP_MAP[op] + if op in FOREACH_LINEAR_REDUCTION_OP_MAP + else LINEAR_REDUCTION_OP_MAP[op] + ) + return _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=keep_dim, + reduction_linear=True, + reduction_op=reduction_op, + ) + + +def _is_spec_evenly_sharded_on_dim(spec: DTensorSpec, dim: int) -> bool: + shape = spec.shape + dim = normalize_dim(dim, len(shape)) + + num_shards = 1 + for mesh_dim, placement in enumerate(spec.placements): + if _is_shard_like(placement) and placement.dim == dim: + num_shards *= spec.mesh.size(mesh_dim) + if isinstance(placement, _StridedShard): + num_shards *= placement.split_factor + + return guard_or_false(cast(Any, shape[dim]) % num_shards == 0) + + +def _mean_full_mesh_strategy_filter( + mesh: DeviceMesh, + op_schema: OpSchema, + input_specs: list[DTensorSpec], + output_specs: DTensorSpec | tuple[DTensorSpec | None, ...], +) -> bool: + """Reject mean strategies that would average uneven local means. + + mean uses Partial("avg") for shards on reduction dims. That is only correct + when every rank contributes the same number of elements; otherwise averaging + the per-rank means weights small and large shards equally. The single-dim + rule cannot know the final product of mesh dims sharding a tensor dim, so + this filter checks the materialized full-mesh input spec. + """ + input_spec = input_specs[0] + dims = None + if len(op_schema.args_schema) > 1: + dims = _infer_reduction_dims(op_schema.args_schema[1], input_spec.ndim) + reduce_dims = list(range(input_spec.ndim)) if dims is None else dims + + return all(_is_spec_evenly_sharded_on_dim(input_spec, dim) for dim in reduce_dims) + + +@register_single_dim_strategy( + MEAN_REDUCTION_OPS, + schema_info=RuntimeSchemaInfo(1), + full_mesh_strategy_filter=_mean_full_mesh_strategy_filter, +) +def mean_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dims = None + if len(args_schema) > 1: + dims = _infer_reduction_dims(args_schema[1], ndim) + + reduce_dims = list(range(ndim)) if dims is None else dims + keep_dim = len(args_schema) > 2 and bool(args_schema[2]) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d in reduce_dims: + strategies.append([Partial("avg"), _ShardingPlaceholder(d)]) + else: + if keep_dim: + out_d = d + else: + out_d = d - sum(1 for rd in reduce_dims if rd < d) + strategies.append([_ShardingPlaceholder(out_d), _ShardingPlaceholder(d)]) + + strategies.append([Partial("avg"), Partial("avg")]) + return strategies + + +@register_single_dim_strategy( + NON_LINEAR_BOOL_REDUCTION_OPS, + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def bool_reduction_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """all/any: only shard on non-reduction dims (no partial propagation).""" + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dims = None + if len(args_schema) > 1: + dims = _infer_reduction_dims(args_schema[1], ndim) + + keep_dim = len(args_schema) > 2 and bool(args_schema[2]) + return _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=keep_dim, + reduction_linear=False, + reduction_op="sum", + ) + + +def _shard_non_reduction_dim( + args_schema: tuple[Any, ...], + dim: int, + keep_dim: bool, + n_outputs: int, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Shared logic for ops that shard on non-reduction dims with dim remapping. + + Used by dim-reduction ops returning (values, indices) and argmax/argmin. + + Args: + args_schema: Op args with TensorMeta at position 0. + dim: The reduction dimension. + keep_dim: Whether the reduction dim is kept. + n_outputs: Number of output tensors (1 for argmax, 2 for max.dim). + """ + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + + ndim = len(input_meta.shape) + dim = normalize_dim(dim, ndim) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d == dim: + continue + out_d = d if keep_dim or d < dim else d - 1 + strategies.append( + [_ShardingPlaceholder(out_d)] # pyrefly: ignore[bad-argument-type] + * n_outputs + + [_ShardingPlaceholder(d)] + ) + return strategies + + +# Category B: Dim-reduction ops returning (values, indices) +# max.dim/min.dim return (values, indices). Indices are local to each shard +# and cannot be combined across ranks, so we force Replicate on reduction dims +# (same approach as argmax/argmin). +@register_single_dim_strategy( + [aten.max.dim, aten.min.dim], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def max_min_dim_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + dim = cast(int, args_schema[1]) + keep_dim = len(args_schema) > 2 and bool(args_schema[2]) + return _shard_non_reduction_dim(args_schema, dim, keep_dim, n_outputs=2) + + +def _argmax_argmin_reduction_dims( + args_schema: tuple[Any, ...], kwargs_schema: dict[str, Any], ndim: int +) -> list[int]: + if "dim" in kwargs_schema: + dims_arg = kwargs_schema["dim"] + elif len(args_schema) > 1: + dims_arg = args_schema[1] + else: + dims_arg = None + dims = _infer_reduction_dims(dims_arg, ndim) + return list(range(ndim)) if dims is None else dims + + +def _argmax_argmin_full_mesh_strategy_filter( + _mesh: DeviceMesh, + op_schema: OpSchema, + input_specs: list[DTensorSpec], + _output_specs: DTensorSpec | tuple[DTensorSpec | None, ...], +) -> bool: + """ + Filter strategies to match the custom nonlinear-redux handler. + + argminmax_handler computes from the runtime DTensor placements instead of + applying the normal redistribute_schema. It only materializes Partial inputs + and reduction-dim shards to Replicate, so sharding propagation must not pick + a target input spec that would require retargeting those placements to a + non-reduction shard. + """ + args_schema = op_schema.args_schema + input_strategy = args_schema[0] + if not isinstance(input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(input_strategy)}") + + input_meta = input_strategy.tensor_meta + if input_meta is None: + raise AssertionError("Expected input tensor meta") + + reduce_dims = _argmax_argmin_reduction_dims( + args_schema, op_schema.kwargs_schema, len(input_meta.shape) + ) + + input_spec = input_strategy.strategies[0].output_spec + target_input_spec = input_specs[0] + for input_placement, target_placement in zip( + input_spec.placements, target_input_spec.placements + ): + if input_placement.is_partial() or ( + _is_shard_like(input_placement) and input_placement.dim in reduce_dims + ): + if not target_placement.is_replicate(): + return False + return True + + +@register_single_dim_strategy( + list(ARGMAX_ARGMIN_OPS.keys()), + schema_info=RuntimeSchemaInfo(1, ["dim", "keepdim"]), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, + full_mesh_strategy_filter=_argmax_argmin_full_mesh_strategy_filter, +) +def argmax_argmin_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """argmax/argmin return indices only. Shard on non-reduction dims.""" + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + reduce_dims = _argmax_argmin_reduction_dims(args_schema, kwargs_schema, ndim) + keep_dim = ( + bool(kwargs_schema["keepdim"]) + if "keepdim" in kwargs_schema + else len(args_schema) > 2 and bool(args_schema[2]) + ) + + if len(reduce_dims) == ndim: + return [] + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d in reduce_dims: + continue + out_d = d + if not keep_dim: + out_d = d - sum(1 for rd in reduce_dims if rd < d) + strategies.append([_ShardingPlaceholder(out_d), _ShardingPlaceholder(d)]) + return strategies + + +def _shard_except_dim_strategy( + n_placements: int, + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], + active_dim: int, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Build single-dim strategies that shard on every dim except an active dim. + + Used by sort-like ops (sort, topk, cummax, cummin), scan ops (cumsum, cumprod, + logcumsumexp), and softmax-like ops. All outputs and inputs get the same sharding + placeholder. + + Args: + n_placements: Total number of placements per rule (outputs + inputs). + active_dim: The dim to exclude from sharding (e.g. sort dim, softmax dim). + """ + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + active_dim = normalize_dim(active_dim, ndim) + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d != active_dim: + strategies.append([_ShardingPlaceholder(d)] * n_placements) + return strategies + + +# Category A: Sort-like and scan ops +# Scan: 1 output + 1 input = 2 placements +_SCAN_N_PLACEMENTS = 2 +# Sort-like: 2 outputs (values, indices) + 1 input = 3 placements +_SORT_LIKE_N_PLACEMENTS = 3 +# Softmax forward: 1 output + 1 input = 2 placements +_SOFTMAX_FWD_N_PLACEMENTS = 2 +# Softmax backward: 1 output + 2 inputs = 3 placements +_SOFTMAX_BWD_N_PLACEMENTS = 3 + + +@register_single_dim_strategy( + [aten.cumsum.default, aten.cumprod.default, aten.logcumsumexp.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def scan_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + dim = args_schema[1] + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + strategies = _shard_except_dim_strategy( + _SCAN_N_PLACEMENTS, op, args_schema, kwargs_schema, active_dim=dim + ) + # cumsum is linear for sum/avg: partial sums/avgs propagate through + if op == aten.cumsum.default: + strategies.append([Partial("sum"), Partial("sum")]) + strategies.append([Partial("avg"), Partial("avg")]) + return strategies + + +@register_single_dim_strategy( + [aten.median.default, aten.nanmedian.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def global_median_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """Global median reduces all dims — only Replicate is valid.""" + return [] + + +@register_single_dim_strategy( + [aten.median.dim, aten.nanmedian.dim, aten.mode.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def dim_reduction_with_indices_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + dim = cast(int, args_schema[1]) if len(args_schema) > 1 else -1 + keep_dim = len(args_schema) > 2 and bool(args_schema[2]) + return _shard_non_reduction_dim(args_schema, dim, keep_dim, n_outputs=2) + + +@register_single_dim_strategy( + [aten.kthvalue.default], + schema_info=RuntimeSchemaInfo(2), + allow_uneven_sharding=True, +) +def kthvalue_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + dim = cast(int, args_schema[2]) if len(args_schema) > 2 else -1 + keep_dim = len(args_schema) > 3 and bool(args_schema[3]) + return _shard_non_reduction_dim(args_schema, dim, keep_dim, n_outputs=2) + + +@register_single_dim_strategy( + [aten.cummax.default, aten.cummin.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def cummax_cummin_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + dim = cast(int, args_schema[1]) + return _shard_except_dim_strategy( + _SORT_LIKE_N_PLACEMENTS, op, args_schema, kwargs_schema, active_dim=dim + ) + + +_STD_VAR_OPS = [ + aten.std.correction, + aten.std.correction_out, + aten.var.correction, + aten.var.correction_out, + aten.var_mean.correction, + aten.var_mean.correction_out, + prims.var.default, +] + + +@register_single_dim_strategy( + _STD_VAR_OPS, + schema_info=RuntimeSchemaInfo(1, ["keepdim"]), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def std_var_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dims = None + if len(args_schema) > 1: + dims = _infer_reduction_dims(args_schema[1], ndim) + + keep_dim = cast(bool, kwargs_schema.get("keepdim", False)) + strategies = _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=keep_dim, + reduction_linear=False, + reduction_op="sum", + ) + num_outputs = len(op._schema.returns) + if num_outputs == 1 and not any( + isinstance(arg, TensorMeta) for arg in kwargs_schema.values() + ): + return strategies + + expanded_strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for rule in strategies: + output_placements = [rule[0]] * num_outputs + input_placements = rule[1:] + kwarg_placements: list[Placement | _ShardingPlaceholder] = [] + for arg in kwargs_schema.values(): + if isinstance(arg, TensorMeta): + kwarg_placements.append(output_placements[len(kwarg_placements)]) + expanded_strategies.append( + output_placements + input_placements + kwarg_placements + ) + return expanded_strategies + + +def _get_norm_reduction_op(norm_type: int | float | str) -> ReductionOpType: + """Get the reduction op for vector/foreach norm based on norm_type. + + For inf/-inf norms, returns simple reduction ops ("max", "min"). + For other norms (including 0), returns NormReduction which produces the + appropriate Partial placement via get_placement_from_reduction_op. + """ + if norm_type in (float("inf"), "inf"): + return "max" + elif norm_type in (float("-inf"), "-inf"): + return "min" + else: + if not isinstance(norm_type, (int, float)): + raise AssertionError + return NormReduction(norm_type) + + +@register_single_dim_strategy( + [aten.linalg_vector_norm.default, aten.norm.Scalar], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def vector_norm_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """Vector norm: linear reduction with NormPartial. + + S(reduction_dim)->_NormPartial(ord) rules appear as false positives in + strategy_validation because the validator doesn't enumerate _NormPartial + as an output placement. The rules are correct at runtime. + TODO: remove this comment when _NormPartial is deprecated. + """ + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + norm_type = args_schema[1] if len(args_schema) > 1 else 2 + if not isinstance(norm_type, (int, float, str)): + raise AssertionError(f"Expected int, float, or str, got {type(norm_type)}") + dim = args_schema[2] if len(args_schema) > 2 else None + keepdim = args_schema[3] if len(args_schema) > 3 else False + dims = _infer_reduction_dims(dim, ndim) + + return _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=cast(bool, keepdim), + reduction_linear=True, + reduction_op=_get_norm_reduction_op(norm_type), + ) + + +register_single_dim_strategy( + [aten._foreach_norm.Scalar], + schema_info=RuntimeSchemaInfo(1, needs_pytree=True), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +)(vector_norm_single_dim_strategy) + + +@register_single_dim_strategy( + [aten.linalg__powsum.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def powsum_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """linalg__powsum: computes sum(|x|^ord), always reducible with Partial("sum").""" + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dim = args_schema[2] if len(args_schema) > 2 else None + keepdim = args_schema[3] if len(args_schema) > 3 else False + dims = _infer_reduction_dims(dim, ndim) + + return _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=cast(bool, keepdim), + reduction_linear=True, + reduction_op="sum", + ) + + +register_single_dim_strategy( + [aten._foreach_powsum.Scalar], + schema_info=RuntimeSchemaInfo(1, needs_pytree=True), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +)(powsum_single_dim_strategy) + + +_REPLICATE_ONLY_OPS = [ + aten._linalg_svd.default, + aten.linalg_qr.default, + # TODO: The diagonal ops can have an improved sharding strategy for + # shard placements that does not require redistributing to replicate. + aten.diagonal_copy.default, + aten.diag_embed.default, + aten.diag.default, + aten.diagonal.default, + aten.tril.default, + aten.triu.default, + aten._linalg_eigh.default, +] + + +@register_single_dim_strategy( + _REPLICATE_ONLY_OPS, + schema_info=RuntimeSchemaInfo(1), +) +def linalg_replicate_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + """Replicate-only ops: only the all-Replicate strategy is valid.""" + return [] + + +# Maps each pooling op to its spatial rank (number of spatial dimensions). +# Batched inputs have layout (N, C, *spatial) with ndim = spatial_rank + 2; +# unbatched inputs drop the batch dim giving ndim = spatial_rank + 1. +POOL_SPATIAL_RANK: dict[torch._ops.OpOverload, int] = { + aten.avg_pool1d.default: 1, + aten.avg_pool2d.default: 2, + aten.avg_pool3d.default: 3, + aten.adaptive_avg_pool1d.default: 1, + aten._adaptive_avg_pool2d.default: 2, + aten._adaptive_avg_pool3d.default: 3, + aten.adaptive_max_pool1d.default: 1, + aten.adaptive_max_pool2d.default: 2, + aten.adaptive_max_pool3d.default: 3, + aten.fractional_max_pool2d.default: 2, + aten.fractional_max_pool3d.default: 3, + aten.max_pool1d_with_indices.default: 1, + aten.max_pool2d_with_indices.default: 2, + aten.max_pool3d_with_indices.default: 3, +} + +AVG_POOL_OPS = [ + aten.avg_pool1d.default, + aten.avg_pool2d.default, + aten.avg_pool3d.default, + aten.adaptive_avg_pool1d.default, + aten._adaptive_avg_pool2d.default, + aten._adaptive_avg_pool3d.default, +] + +MAX_POOL_OPS = [ + aten.adaptive_max_pool1d.default, + aten.adaptive_max_pool2d.default, + aten.adaptive_max_pool3d.default, + aten.fractional_max_pool2d.default, + aten.fractional_max_pool3d.default, + aten.max_pool1d_with_indices.default, + aten.max_pool2d_with_indices.default, + aten.max_pool3d_with_indices.default, +] + + +@register_single_dim_strategy( + AVG_POOL_OPS + MAX_POOL_OPS, + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def pooling_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + + ndim = len(input_meta.shape) + num_outputs = sum(1 for r in op._schema.returns if "Tensor" in str(r.type)) + num_inputs = sum(1 for a in args_schema if isinstance(a, TensorMeta)) + n = num_outputs + num_inputs + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + + # S(0) for batch dim — always valid + strategies.append([_ShardingPlaceholder(0)] * n) + + # avg_pool is linear: Partial(sum) and Partial(avg) pass through unchanged. + if op in AVG_POOL_OPS: + strategies.append([Partial("sum")] * n) + strategies.append([Partial("avg")] * n) + + # S(1) is safe when dim 1 is the channel dim (pooling never touches it). + # Batched inputs have layout (N, C, *spatial) with ndim = spatial_rank + 2. + spatial_rank = POOL_SPATIAL_RANK[op] + is_batched = ndim >= spatial_rank + 2 + if is_batched: + strategies.append([_ShardingPlaceholder(1)] * n) + + return strategies + + +# Category F: Softmax-like ops +@register_single_dim_strategy( + [aten._log_softmax.default, aten._softmax.default, aten._safe_softmax.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def softmax_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + softmax_dim = cast(int, args_schema[1]) + return _shard_except_dim_strategy( + _SOFTMAX_FWD_N_PLACEMENTS, + op, + args_schema, + kwargs_schema, + active_dim=softmax_dim, + ) + + +@register_single_dim_strategy( + [ + aten._log_softmax_backward_data.default, + aten._softmax_backward_data.default, + ], + schema_info=RuntimeSchemaInfo(2), + allow_uneven_sharding=True, + allow_unbacked_sharding=False, +) +def softmax_backward_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + softmax_dim = cast(int, args_schema[2]) + return _shard_except_dim_strategy( + _SOFTMAX_BWD_N_PLACEMENTS, + op, + args_schema, + kwargs_schema, + active_dim=softmax_dim, + ) + + +@register_single_dim_strategy( + [aten.nll_loss_forward.default, aten.nll_loss2d_forward.default], + schema_info=RuntimeSchemaInfo(3), +) +def nll_loss_forward_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + if len(args_schema) != 5: + raise AssertionError(f"Expected 5 args, got {len(args_schema)}") + + ( + input_meta, + target_meta, + weight_meta, + reduction, + _, + ) = args_schema + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(target_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(target_meta)}") + if weight_meta is not None and not isinstance(weight_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(weight_meta)}") + reduction = cast(int, reduction) + + input_shape = input_meta.shape + channel_dim = 1 if len(input_shape) >= 2 else 0 + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + # weight, if given, has size input_shape[channel_dim], so replicate it. + weight_placement = [Replicate()] if weight_meta is not None else [] + + for input_dim in range(len(input_shape)): + # Input must be replicated along the channel dim. + if input_dim == channel_dim: + continue + # Target has no channel dim and follows input on the other dims. + target_dim = input_dim if input_dim < channel_dim else input_dim - 1 + input_shard = _ShardingPlaceholder(input_dim) + target_shard = _ShardingPlaceholder(target_dim) + if reduction == Reduction.NONE.value: + # With reduction="none", output follows target and total_weight is + # a replicated scalar. + strategies.append( + [ + target_shard, + Replicate(), + input_shard, + target_shard, + *weight_placement, + ] + ) + elif reduction == Reduction.SUM.value: + # The output reduces across all target dims. total_weight is summed + # across ranks if not replicated. + strategies.append( + [ + Partial("sum"), + Partial("sum"), + input_shard, + target_shard, + *weight_placement, + ] + ) + + return strategies + + +@register_single_dim_strategy( + [aten.nll_loss_backward.default, aten.nll_loss2d_backward.default], + schema_info=RuntimeSchemaInfo(4), +) +def nll_loss_backward_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + if len(args_schema) != 7: + raise AssertionError(f"Expected 7 args, got {len(args_schema)}") + ( + grad_out_meta, + input_meta, + target_meta, + weight_meta, + reduction, + _, + total_weight_meta, + ) = args_schema + if not isinstance(grad_out_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(grad_out_meta)}") + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(target_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(target_meta)}") + if weight_meta is not None and not isinstance(weight_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(weight_meta)}") + if not isinstance(total_weight_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(total_weight_meta)}") + reduction = cast(int, reduction) + + input_shape = input_meta.shape + channel_dim = 1 if len(input_shape) >= 2 else 0 + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + # weight, if given, has size input_shape[channel_dim], so replicate it. + weight_placement = [Replicate()] if weight_meta is not None else [] + + for input_dim in range(len(input_shape)): + # Input must be replicated along the channel dim. + if input_dim == channel_dim: + continue + # Target has no channel dim and follows input on the other dims. + target_dim = input_dim if input_dim < channel_dim else input_dim - 1 + input_shard = _ShardingPlaceholder(input_dim) + target_shard = _ShardingPlaceholder(target_dim) + + if reduction == Reduction.NONE.value: + # grad_out follows target when there is no reduction; total_weight + # is unused by the kernel, so keep it replicated here. + strategies.append( + [ + input_shard, + target_shard, + input_shard, + target_shard, + *weight_placement, + Replicate(), + ] + ) + else: + # grad_out is a replicated scalar for reduced outputs. + total_weight_placements: list[Placement] = [Replicate()] + if reduction != Reduction.MEAN.value: + # total_weight is only used for reduction="mean". For sum it is + # unused, so a partial placement is also valid and avoids + # unnecessary redistribution. + total_weight_placements.append(Partial("sum")) + + for total_weight_placement in total_weight_placements: + strategies.append( + [ + input_shard, + Replicate(), + input_shard, + target_shard, + *weight_placement, + total_weight_placement, + ] + ) + + if reduction != Reduction.MEAN.value: + # When total_weight is unused, also allow a replicated input/output + # strategy that preserves Partial("sum") total_weight. + strategies.append( + [ + Replicate(), + Replicate(), + Replicate(), + Replicate(), + *weight_placement, + Partial("sum"), + ] + ) + + return strategies + + +@register_single_dim_strategy( + [aten.native_layer_norm.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def layer_norm_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + normalized_shape = args_schema[1] + weight_meta = args_schema[2] + bias_meta = args_schema[3] + + axis = len(input_meta.shape) - len(normalize_to_torch_size(normalized_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(axis): + # [out, mean, rstd, input, weight?, bias?] + rule: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(dim), # out + _ShardingPlaceholder(dim), # mean + _ShardingPlaceholder(dim), # rstd + _ShardingPlaceholder(dim), # input + ] + if weight_meta is not None: + rule.append(Replicate()) + if bias_meta is not None: + rule.append(Replicate()) + strategies.append(rule) + return strategies + + +@register_single_dim_strategy( + [aten._fused_rms_norm.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def rms_norm_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + normalized_shape = args_schema[1] + weight_meta = args_schema[2] + + axis = len(input_meta.shape) - len(normalize_to_torch_size(normalized_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(axis): + # [out, rrms, input, weight?] + rule: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(dim), # out + _ShardingPlaceholder(dim), # rrms + _ShardingPlaceholder(dim), # input + ] + if weight_meta is not None: + rule.append(Replicate()) + strategies.append(rule) + return strategies + + +@register_single_dim_strategy( + [aten.native_layer_norm_backward.default], + schema_info=RuntimeSchemaInfo(2), +) +def layer_norm_bwd_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder | None]]: + input_meta = args_schema[1] + normalized_shape = args_schema[2] + # mean = args_schema[3], rstd = args_schema[4] + weight_meta = args_schema[5] + bias_meta = args_schema[6] + output_mask = args_schema[7] + compute_d_input = output_mask[0] + compute_d_weight = weight_meta is not None and output_mask[1] + compute_d_bias = bias_meta is not None and output_mask[2] + + axis = len(input_meta.shape) - len(normalize_to_torch_size(normalized_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder | None]] = [] + for dim in range(axis): + # outputs: [d_input, d_weight, d_bias] — always 3 per schema + # Masked-off outputs use None even if the corresponding input exists. + rule: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(dim) if compute_d_input else None, + Partial("sum") if compute_d_weight else None, + Partial("sum") if compute_d_bias else None, + ] + # inputs: [grad_out, input, mean, rstd, weight?, bias?] + rule.extend( + [ + _ShardingPlaceholder(dim), # grad_out + _ShardingPlaceholder(dim), # input + _ShardingPlaceholder(dim), # mean + _ShardingPlaceholder(dim), # rstd + ] + ) + if weight_meta is not None: + rule.append(Replicate()) + if bias_meta is not None: + rule.append(Replicate()) + strategies.append(rule) + + return strategies + + +@register_single_dim_strategy( + [aten._fused_rms_norm_backward.default], + schema_info=RuntimeSchemaInfo(2), +) +def rms_norm_bwd_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder | None]]: + input_meta = args_schema[1] + normalized_shape = args_schema[2] + # rstd = args_schema[3] + weight_meta = args_schema[4] + output_mask = args_schema[5] + compute_d_input = output_mask[0] + compute_d_weight = weight_meta is not None and output_mask[1] + + axis = len(input_meta.shape) - len(normalize_to_torch_size(normalized_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder | None]] = [] + for dim in range(axis): + # outputs: [d_input, d_weight] — always 2 per schema + # Masked-off outputs use None even if the corresponding input exists. + # inputs: [grad_out, input, rstd, weight?] + rule: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(dim) if compute_d_input else None, + Partial("sum") if compute_d_weight else None, + _ShardingPlaceholder(dim), # grad_out + _ShardingPlaceholder(dim), # input + _ShardingPlaceholder(dim), # rstd + ] + if weight_meta is not None: + rule.append(Replicate()) + strategies.append(rule) + + return strategies + + +@register_single_dim_strategy( + [aten.topk.default], + schema_info=RuntimeSchemaInfo(2), + allow_uneven_sharding=True, +) +def topk_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + topk_dim = cast(int, args_schema[2]) if len(args_schema) > 2 else -1 + return _shard_except_dim_strategy( + _SORT_LIKE_N_PLACEMENTS, op, args_schema, kwargs_schema, active_dim=topk_dim + ) + + +@register_single_dim_strategy( + aten.sort.default, + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def sort_default_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + sort_dim = cast(int, args_schema[1]) if len(args_schema) > 1 else -1 + return _shard_except_dim_strategy( + _SORT_LIKE_N_PLACEMENTS, op, args_schema, kwargs_schema, active_dim=sort_dim + ) + + +@register_single_dim_strategy( + aten.sort.stable, + schema_info=RuntimeSchemaInfo( + 1, + static_kwargkey=["dim", "descending", "stable"], + ), + allow_uneven_sharding=True, +) +def sort_stable_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + sort_dim = cast(int, kwargs_schema.get("dim", -1)) + return _shard_except_dim_strategy( + _SORT_LIKE_N_PLACEMENTS, op, args_schema, kwargs_schema, active_dim=sort_dim + ) + + +@register_single_dim_strategy( + [aten.histc.default], + schema_info=RuntimeSchemaInfo(2), + allow_uneven_sharding=True, +) +def histc_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + # histc supports sharded input -> partial output only when min/max are specified + if len(args_schema) == 4: + for dim in range(ndim): + strategies.append([Partial(), _ShardingPlaceholder(dim)]) + return strategies + + +@register_single_dim_strategy( + [aten.logsumexp.default], + schema_info=RuntimeSchemaInfo( + static_argnum=1, + static_kwargkey=["keepdim"], + ), + allow_uneven_sharding=True, +) +def logsumexp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + dims = _infer_reduction_dims(args_schema[1], ndim) + keep_dim = cast(bool, kwargs_schema.get("keepdim", False)) + return _reduction_single_dim_strategy( + args_schema, + reduction_dims=dims, + keep_dim=keep_dim, + reduction_linear=False, + reduction_op="sum", + ) + + +_LINALG_NUM_PLACEMENTS = { + # 1 in 1 out + aten.cholesky_inverse.default: 2, + aten.linalg_matrix_exp.default: 2, + # 2 in 1 out + aten.cholesky_solve.default: 3, + aten.linalg_householder_product.default: 3, + aten.linalg_solve_triangular.default: 3, + # 3 in 1 out + aten.linalg_ldl_solve.default: 4, + aten.linalg_lu_solve.default: 4, + aten.ormqr.default: 4, + # 1 in 2 out + aten.geqrf.default: 3, + aten.linalg_cholesky_ex.default: 3, + aten.linalg_eig.default: 3, + aten.linalg_inv_ex.default: 3, + # 2 in 2 out + aten.triangular_solve.default: 4, + # 1 in 3 out + aten._linalg_det.default: 4, + aten.linalg_ldl_factor_ex.default: 4, + aten.linalg_lu.default: 4, + aten.linalg_lu_factor_ex.default: 4, + # 2 in 3 out + aten.lu_unpack.default: 5, + # 1 in 4 out + aten._linalg_slogdet.default: 5, + # 2 in 4 out + aten._linalg_solve_ex.default: 6, + # 1 in + aten._linalg_check_errors.default: 1, +} + + +def _linalg_batch_dim_strategies( + ndim: int, n_placements: int +) -> list[list[Placement | _ShardingPlaceholder]]: + """Build single-dim strategies for linalg ops that operate on the last 1-2 dims. + + Returns sharding on each batch dim (all dims except the last 2), with all + outputs and inputs sharded on the same dim. + """ + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(ndim - 2): + strategies.append([_ShardingPlaceholder(dim)] * n_placements) + return strategies + + +def _get_ndim(tensor_meta: Any) -> int: + if not isinstance(tensor_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(tensor_meta)}") + return len(tensor_meta.shape) + + +@register_single_dim_strategy( + [ + aten.cholesky_inverse.default, + aten.linalg_matrix_exp.default, + aten.cholesky_solve.default, + aten.linalg_householder_product.default, + aten.linalg_solve_triangular.default, + aten.linalg_ldl_solve.default, + aten.linalg_lu_solve.default, + aten.ormqr.default, + aten.geqrf.default, + aten.linalg_cholesky_ex.default, + aten.linalg_eig.default, + aten.linalg_inv_ex.default, + aten.triangular_solve.default, + aten._linalg_det.default, + aten.linalg_ldl_factor_ex.default, + aten.linalg_lu.default, + aten.linalg_lu_factor_ex.default, + aten.lu_unpack.default, + aten._linalg_slogdet.default, + aten._linalg_solve_ex.default, + aten._linalg_check_errors.default, + ], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def linalg_batch_dim_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + ndim = _get_ndim(args_schema[0]) + if op not in _LINALG_NUM_PLACEMENTS: + raise AssertionError(f"Expected op in _LINALG_NUM_PLACEMENTS, got {op}") + + n_placements = _LINALG_NUM_PLACEMENTS[op] + strategies = _linalg_batch_dim_strategies(ndim, n_placements=n_placements) + + if op == aten.linalg_solve_triangular.default: + # solve_triangular(A, B) -> result: linear in B + strategies.append([Partial(), Replicate(), Partial()]) + strategies.append([Partial("avg"), Replicate(), Partial("avg")]) + # A replicated, B sharded on batch dims (B may have more batch dims than A) + ndim_b = _get_ndim(args_schema[1]) + for dim in range(ndim_b - 2): + strategies.append( + [_ShardingPlaceholder(dim), Replicate(), _ShardingPlaceholder(dim)] + ) + elif op == aten.cholesky_solve.default: + # cholesky_solve(B, A) -> result (B is arg0) + strategies.append([Partial(), Partial(), Replicate()]) + elif op == aten.linalg_lu_solve.default: + # linalg_lu_solve(LU, pivots, B) -> result + strategies.append([Partial(), Replicate(), Replicate(), Partial()]) + elif op == aten.linalg_ldl_solve.default: + # linalg_ldl_solve(LD, pivots, B) -> result + strategies.append([Partial(), Replicate(), Replicate(), Partial()]) + elif op == aten.ormqr.default: + # ormqr(a, tau, C) -> result (linear in C) + strategies.append([Partial(), Replicate(), Replicate(), Partial()]) + elif op == aten._linalg_solve_ex.default: + # _linalg_solve_ex(A, B) -> (result, LU, pivots, info) + strategies.append( + [Partial(), Replicate(), Replicate(), Replicate(), Replicate(), Partial()] + ) + + return strategies + + +# linalg_pinv has optional tensor kwargs atol, rtol (scalar tensors when present). +# Schema: (Tensor self, *, Tensor? atol=None, Tensor? rtol=None, bool hermitian) -> Tensor +# When atol/rtol are None, num_inputs=1; when present, they add to num_inputs. +@register_single_dim_strategy( + [aten.linalg_pinv.atol_rtol_tensor], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def linalg_pinv_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + ndim = _get_ndim(args_schema[0]) + # Count optional tensor kwargs that are actually present + extra_tensors = sum( + isinstance(kwargs_schema.get(k), TensorMeta) for k in ("atol", "rtol") + ) + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(ndim - 2): + s: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(dim), + _ShardingPlaceholder(dim), + ] + # atol, rtol are scalar tensors — always Replicate + s.extend([Replicate()] * extra_tensors) + strategies.append(s) + return strategies + + +# linalg_cross is pointwise on every dim except the cross-product dim (which +# must be size 3). Shard on any other dim. +@register_single_dim_strategy( + [aten.linalg_cross.default], + schema_info=RuntimeSchemaInfo(1), + allow_uneven_sharding=True, +) +def linalg_cross_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + ndim = _get_ndim(args_schema[0]) + cross_dim = kwargs_schema.get("dim", -1) % ndim + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(ndim): + if dim == cross_dim: + continue + strategies.append([_ShardingPlaceholder(dim)] * 3) + return strategies + + +# --------------------------------------------------------------------------- +# Interpolation / upsample / pooling ops +# +# These ops operate on spatial dims and are safely shardable on batch (dim 0) +# and channel (dim 1). grid_sampler is batch-only because the grid tensor has +# no channel dimension. +# --------------------------------------------------------------------------- + + +@register_single_dim_strategy( + [ + # Forward ops + aten.upsample_nearest1d.default, + aten.upsample_nearest2d.default, + aten.upsample_nearest3d.default, + aten._upsample_nearest_exact1d.default, + aten._upsample_nearest_exact2d.default, + aten._upsample_nearest_exact3d.default, + aten._upsample_bilinear2d_aa.default, + aten._upsample_bicubic2d_aa.default, + aten._upsample_lanczos2d_aa.default, + aten.upsample_bicubic2d.default, + aten.upsample_bilinear2d.default, + aten.upsample_linear1d.default, + aten.upsample_trilinear3d.default, + # Backward ops + aten.upsample_nearest1d_backward.default, + aten.upsample_nearest2d_backward.default, + aten.upsample_nearest3d_backward.default, + aten._upsample_nearest_exact1d_backward.default, + aten._upsample_nearest_exact2d_backward.default, + aten._upsample_nearest_exact3d_backward.default, + aten._upsample_bilinear2d_aa_backward.default, + aten._upsample_bicubic2d_aa_backward.default, + aten._upsample_lanczos2d_aa_backward.default, + aten.upsample_bicubic2d_backward.default, + aten.upsample_bilinear2d_backward.default, + aten.upsample_linear1d_backward.default, + aten.upsample_trilinear3d_backward.default, + ], + schema_info=RuntimeSchemaInfo(1), +) +def interp_upsample_1out_1in_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # 1 output + 1 input = 2 placements; shard on batch (0) and channel (1) + # Upsample is a linear transformation so Partial(sum/avg) is valid. + return [ + [_ShardingPlaceholder(0)] * 2, + [_ShardingPlaceholder(1)] * 2, + [Partial("sum"), Partial("sum")], + [Partial("avg"), Partial("avg")], + ] + + +@register_single_dim_strategy( + [ + aten.max_unpool2d.default, + aten.max_unpool3d.default, + aten._adaptive_avg_pool2d_backward.default, + ], + schema_info=RuntimeSchemaInfo(1), +) +def interp_pool_1out_2in_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # 1 output + 2 inputs = 3 placements; shard on batch (0) and channel (1) + return [ + [_ShardingPlaceholder(0)] * 3, + [_ShardingPlaceholder(1)] * 3, + ] + + +@register_single_dim_strategy( + [aten.max_pool2d_with_indices_backward.default], + schema_info=RuntimeSchemaInfo(1), +) +def pool_backward_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # max_pool2d_with_indices_backward(grad_output, self, ..., indices) -> grad_input + # 1 output + 3 tensor inputs = 4 placements + # Order: [output, grad_output, self, indices] + input_meta = cast(TensorMeta, args_schema[0]) + strategies: list[list[Placement | _ShardingPlaceholder]] = [ + [_ShardingPlaceholder(0)] * 4, + ] + if len(input_meta.shape) >= 4: # batched: (N, C, H, W) + strategies.append([_ShardingPlaceholder(1)] * 4) + # The backward is linear in grad_output, so P(sum/avg) pass through. + # indices must be replicated (integer positions, not reducible). + # self is only used for shape, so replicate it too. + r = Replicate() + for reduce_op in ("sum", "avg"): + p = Partial(reduce_op) + strategies.append([p, p, r, r]) + return strategies + + +@register_single_dim_strategy( + [aten.grid_sampler_2d.default, aten.grid_sampler_3d.default], + schema_info=RuntimeSchemaInfo(1), +) +def grid_sampler_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # grid_sampler_{2,3}d(input[N,C,...], grid[N,...,{2,3}]) -> output[N,C,...] + # grid has no channel dim, so only batch sharding applies to both inputs. + # Linear in input: P(sum/avg) on input with replicated grid is valid. + return [ + [_ShardingPlaceholder(0)] * 3, + [Partial("sum"), Partial("sum"), Replicate()], + [Partial("avg"), Partial("avg"), Replicate()], + ] + + +@register_single_dim_strategy( + [aten.grid_sampler_2d_backward.default, aten.grid_sampler_3d_backward.default], + schema_info=RuntimeSchemaInfo(1), +) +def grid_sampler_backward_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder | None]]: + # grid_sampler_{2,3}d_backward: 2 outputs (grad_input, grad_grid) + 3 inputs = 5 placements, batch-only + # The native implementation only honors output_mask[0]; grad_grid is always computed. + output_mask = args_schema[6] + return [ + [ + _ShardingPlaceholder(0) if output_mask[0] else None, + _ShardingPlaceholder(0), + _ShardingPlaceholder(0), # grad_output + _ShardingPlaceholder(0), # input + _ShardingPlaceholder(0), # grid + ] + ] + + +def _adjust_group_norm_scalars( + input_specs: list[DTensorSpec], schema: OpSchema +) -> OpSchema: + """Adjust N, C, HxW scalar args in native_group_norm to local values. + + native_group_norm(input, weight?, bias?, N, C, HxW, group, eps) + The scalar args are derived from the global input shape by the Python frontend. + When the input is sharded, we recompute them from the local input shape. + """ + input_spec = input_specs[0] + if input_spec.tensor_meta is None: + raise AssertionError("input_spec must have tensor_meta") + local_shape, _ = compute_local_shape_and_global_offset( + input_spec.tensor_meta.shape, + input_spec.mesh, + input_spec.placements, + skip_offset=True, + ) + # N = local_shape[0], C = local_shape[1], HxW = product of remaining dims + n_local = local_shape[0] + c_local = local_shape[1] + hxw_local = 1 + for d in local_shape[2:]: + hxw_local *= d + args = list(schema.args_schema) + # Find scalar arg positions: tensor slots (DTensorSpec or None for optionals) + # precede N, C, HxW. Count both to find the offset. + num_tensor_args = sum(isinstance(a, (DTensorSpec, type(None))) for a in args) + args[num_tensor_args] = n_local + args[num_tensor_args + 1] = c_local + args[num_tensor_args + 2] = hxw_local + return OpSchema(schema.op, tuple(args), schema.kwargs_schema) + + +# --------------------------------------------------------------------------- +# Normalization ops +# +# Batch norm reduces over batch (dim 0) + spatial dims (2+), keeping only +# channel (dim 1). Channel-dim sharding is valid for both forward and +# backward: each shard processes independent channels. +# +# Group norm reduces over (C/groups, spatial) within each group per sample. +# Batch dim (0) is safe to shard — each sample is independent. +# --------------------------------------------------------------------------- + +BATCH_NORM_3OUT_OPS = [ + aten.native_batch_norm.default, + aten._native_batch_norm_legit.default, + aten._native_batch_norm_legit.no_stats, + aten._native_batch_norm_legit_no_training.default, +] + +BATCH_NORM_4OUT_OPS = [ + aten._batch_norm_with_update.default, +] + + +@register_single_dim_strategy( + BATCH_NORM_3OUT_OPS + BATCH_NORM_4OUT_OPS, + schema_info=RuntimeSchemaInfo(1), +) +def batch_norm_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # Batch norm normalizes per-channel (reduces over batch + spatial dims), + # so channel-dim sharding is valid: each shard processes independent channels. + # Unlike group_norm, batch_norm infers shapes from tensors (no scalar N/C/HxW). + num_outputs = 4 if op in BATCH_NORM_4OUT_OPS else 3 + num_tensor_inputs = sum(isinstance(a, TensorMeta) for a in args_schema) + # output [N,C,*] shards on dim 1; save_mean, save_invstd [C] shard on dim 0 + rule: list[Placement | _ShardingPlaceholder] = [_ShardingPlaceholder(1)] + rule.extend([_ShardingPlaceholder(0)] * 2) # save_mean, save_invstd + if num_outputs == 4: + rule.append(Replicate()) # reserve: opaque cuDNN workspace + # input [N,C,*] shards on dim 1; weight, bias, running_mean, running_var [C] on dim 0 + rule.append(_ShardingPlaceholder(1)) # input + rule.extend([_ShardingPlaceholder(0)] * (num_tensor_inputs - 1)) + return [rule] + + +@register_single_dim_strategy( + [aten.native_batch_norm_backward.default], + schema_info=RuntimeSchemaInfo(1), +) +def batch_norm_backward_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder | None]]: + # Backward reduces over batch + spatial dims, orthogonal to the channel dim, + # so channel sharding commutes with the op (same principle as forward). + # Tensors are [N,C,*] -> Shard(1) or [C] -> Shard(0). + output_mask = args_schema[9] + num_tensor_inputs = sum(isinstance(a, TensorMeta) for a in args_schema) + rule: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(1) if output_mask[0] else None, # grad_input [N,C,*] + _ShardingPlaceholder(0) if output_mask[1] else None, # grad_weight [C] + _ShardingPlaceholder(0) if output_mask[2] else None, # grad_bias [C] + _ShardingPlaceholder(1), # grad_out [N,C,*] + _ShardingPlaceholder(1), # input [N,C,*] + ] + rule.extend([_ShardingPlaceholder(0)] * (num_tensor_inputs - 2)) + return [rule] + + +@register_single_dim_strategy( + [aten.native_group_norm.default], + schema_info=RuntimeSchemaInfo(1), +) +def group_norm_strategy( + op: torch._ops.OpOverload, + args_schema: tuple[Any, ...], + kwargs_schema: dict[str, Any], +) -> list[list[Placement | _ShardingPlaceholder]]: + # native_group_norm(input, weight?, bias?, N, C, HxW, group, eps) -> (out, mean, rstd) + # Batch dim (0) is independent. The scalar N/C/HxW args are adjusted to local + # values by _adjust_group_norm_scalars in the sharding propagation layer. + num_tensor_inputs = sum(isinstance(a, TensorMeta) for a in args_schema) + # 3 outputs + input all shard on batch dim + placements: list[Placement | _ShardingPlaceholder] = [_ShardingPlaceholder(0)] * 4 + # weight and bias (if present) must be Replicate + placements.extend([Replicate()] * (num_tensor_inputs - 1)) + return [placements] + + +# Register scalar shape adjuster for group_norm so the sharding propagator +# rewrites the N/C/HxW args to local values when the input is sharded. +from torch.distributed.tensor._api import DTensor + + +DTensor._op_dispatcher.sharding_propagator.op_to_scalar_shape_adjuster[ + aten.native_group_norm.default +] = _adjust_group_norm_scalars diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_matrix_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_matrix_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..26ebff73457ca6c90feda51ea844552b3812e19f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_matrix_ops.py @@ -0,0 +1,1368 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +# implement matrix related ops for distributed tensor + + +import copy + +import torch +from torch._ops import OpOverload +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + ArgsType, + KwargsType, + OpSchema, + OpStrategy, + PlacementList, + RuntimeSchemaInfo, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import prod +from torch.distributed.tensor._utils import ( + compute_local_shape_and_global_offset, + compute_local_stride, +) +from torch.distributed.tensor.placement_types import ( + Partial, + Placement, + Replicate, + Shard, +) +from torch.fx.experimental.symbolic_shapes import guard_or_false + + +aten = torch.ops.aten + + +@register_single_dim_strategy(aten.t.default, allow_unbacked_sharding=False) +def transpose_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + if ndim <= 1: + for dim in range(ndim): + strategies.append([_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)]) + else: + strategies.append([_ShardingPlaceholder(1), _ShardingPlaceholder(0)]) + strategies.append([_ShardingPlaceholder(0), _ShardingPlaceholder(1)]) + + for reduce_op in ("sum", "avg", "max", "min"): + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +def _scaled_mm_scale_placement( + data_placement: Placement | _ShardingPlaceholder, + scale_shape: torch.Size, + contracting_dim: int, +) -> Placement | _ShardingPlaceholder | None: + """ + Derive scale placement from data operand placement for _scaled_mm. + + Handles three cases: + + 1. Tensor-wise scale (single element): always Replicate. + 2. 2D (or higher) scale, e.g. row-wise [M,1]: copy data placement directly. + 3. 1D blockwise scale, e.g. MX format [M*K/block_size]: map + non-contracting shard to Shard(0)/_ShardingPlaceholder(0), and reject + contracting-dim shards (returns None). + """ + if prod(scale_shape) == 1: + return Replicate() + + if len(scale_shape) != 1: + return data_placement + + # 1D blockwise scale: Shard(>=1) is invalid on a 1D tensor, so we need + # to map the data operand's placement to a valid 1D placement. + if isinstance(data_placement, _ShardingPlaceholder): + if data_placement.dim == contracting_dim: + return None + return _ShardingPlaceholder(0) + # NOTE: isinstance(_, Shard) does not match _StridedShard; see _is_shard_like(). + elif isinstance(data_placement, Shard): + if data_placement.dim == contracting_dim: + return None + return Shard(0) + elif isinstance(data_placement, (Replicate, Partial)): + return Replicate() + return data_placement + + +from ._einsum_strategy import EinsumDims + + +def gen_single_dim_einsum_strategies( + equation: str, + *, + bias_shape: torch.Size | None = None, +) -> list[list[Placement | _ShardingPlaceholder]]: + """ + Generate a strategy list for the ops that follow einsum style notation. + + In principle, each mesh dim is independent of other device mesh dim when we + generate strategies. So we generate strategy over each device mesh dim and + do product combination on all mesh dims. We basically follow the below rule + for each device mesh dim: + + 1. Shard on contracting dim: When both inputs shard on contracting dim over + the same device dim. The result will be Partial over that device dim. + + 2. Shard on noncontracting dim: + 2.1: Shard on batch dim: output, both inputs all should shard on batch + dim. + 2.2: Shard on lhs only dim or rhs only dim: both output and lhs or rhs + input should shard on this free dim. + + 3. Per-input linearity (Partial): Since matmul is linear in each input + independently, one input can remain Partial while others are Replicate, + producing a Partial output. + + 4. Batch-dimension linearity (all-Partial): When all dims are batch dims + (no contracting or free dims), the operation is element-wise and linear + in all inputs simultaneously, so all inputs can be Partial. + + 5. Bias input (optional): If bias_shape is provided, a bias placement + is inserted after the output placement. The bias placement is derived from + the output placement, accounting for broadcast semantics (based on ndim + difference between output and bias). This is used for addmm-like ops + (addmm, baddbmm) where bias + mat1 @ mat2. + """ + # parse einop equation and extract dims + input_dims, output_dim = EinsumDims.parse_equation(equation) + edims = EinsumDims.parse_dims(input_dims, output_dim) + + # Compute broadcast dims map for bias if provided + # Maps output dims to bias dims, -1 for broadcast dims (dims that don't exist in bias + # or have size 1) + broadcast_dims_map: list[int] | None = None + if bias_shape is not None: + output_ndim = len(output_dim) + bias_ndim = len(bias_shape) + pad_size = output_ndim - bias_ndim + broadcast_dims_map = [] + for i in range(output_ndim): + if i < pad_size: + # Padded dimension (not in bias) + broadcast_dims_map.append(-1) + else: + bias_dim_idx = i - pad_size + if bias_shape[bias_dim_idx] == 1: + # Size-1 dimension (broadcasts) + broadcast_dims_map.append(-1) + else: + broadcast_dims_map.append(bias_dim_idx) + + def _derive_bias_placement( + output_placement: Placement | _ShardingPlaceholder, + ) -> Placement | _ShardingPlaceholder: + """Derive bias placement from output placement, accounting for broadcast.""" + if broadcast_dims_map is None: + return copy.copy(output_placement) + if isinstance(output_placement, _ShardingPlaceholder): + output_dim_idx = output_placement.dim + bias_dim = broadcast_dims_map[output_dim_idx] + if bias_dim == -1: + # Dim doesn't exist in bias (broadcast), replicate + return Replicate() + else: + return _ShardingPlaceholder(bias_dim) + else: + # Clone Partial, Replicate, or other placements + return copy.copy(output_placement) + + def _maybe_add_bias( + placement_list: list[Placement | _ShardingPlaceholder], + ) -> list[Placement | _ShardingPlaceholder]: + """Insert bias placement after output if bias_shape is provided.""" + if bias_shape is None: + return placement_list + output_placement = placement_list[0] + bias_placement = _derive_bias_placement(output_placement) + return [placement_list[0], bias_placement] + placement_list[1:] + + # generate strategies for each mesh dim and do cartesian product for final strategy. E.g., for a 2D mesh, we can have [P(),R,R] + strategies_over_one_mesh_dim: list[list[Placement | _ShardingPlaceholder]] = [] + placement_list: list[Placement | _ShardingPlaceholder] + # split batch dim + for batch_dim in edims.batch_dims: + output_batch_dim = output_dim.index(batch_dim) + placement_list = [_ShardingPlaceholder(output_batch_dim)] + for input_dim in input_dims: + input_batch_dim = input_dim.index(batch_dim) + placement_list.append(_ShardingPlaceholder(input_batch_dim)) + + strategies_over_one_mesh_dim.append(_maybe_add_bias(placement_list)) + + # split contracting dim + for contracting_dim in edims.contracting_dims: + # Contracting dim can shard on same device axis for both inputs. This + # results in the output being Partial on that device axis. For example: + # bmk_{x},k_{x}n -> bmn{Ux} (becomes partial over device axis x) + placement_list = [Partial()] + for input_dim in input_dims: + input_contracting_dim = input_dim.index(contracting_dim) + placement_list.append(_ShardingPlaceholder(input_contracting_dim)) + + strategies_over_one_mesh_dim.append(_maybe_add_bias(placement_list)) + + # split lhs free dim + for lhs_dim in edims.lhs_out_only_dims: + lhs_free_dim_output = output_dim.index(lhs_dim) + lhs_free_dim_input = input_dims[0].index(lhs_dim) + # this means split the lhs input and output + # i.e. S(0), R -> S(0) + lhs_placement_list: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(lhs_free_dim_output), + _ShardingPlaceholder(lhs_free_dim_input), + Replicate(), + ] + strategies_over_one_mesh_dim.append(_maybe_add_bias(lhs_placement_list)) + + # split rhs free dim + for rhs_dim in edims.rhs_out_only_dims: + rhs_free_dim_output = output_dim.index(rhs_dim) + rhs_free_dim_input = input_dims[1].index(rhs_dim) + rhs_placement_list: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(rhs_free_dim_output), + Replicate(), + _ShardingPlaceholder(rhs_free_dim_input), + ] + strategies_over_one_mesh_dim.append(_maybe_add_bias(rhs_placement_list)) + + # Per-input linearity: matmul is linear in each input independently. + # One input Partial, the other Replicate → output Partial. + for reduce_op in Partial.LINEAR_REDUCE_OPS: + output_placement = Partial(reduce_op) + strategies_over_one_mesh_dim.append( + _maybe_add_bias([output_placement, Partial(reduce_op), Replicate()]) + ) + strategies_over_one_mesh_dim.append( + _maybe_add_bias([output_placement, Replicate(), Partial(reduce_op)]) + ) + + # Batch-dimension linearity: when the einsum has no contracting dims and + # no free dims (all dims are batch dims), the operation is element-wise + # and linear in all inputs simultaneously. Add all-Partial strategies. + if ( + not edims.contracting_dims + and not edims.lhs_out_only_dims + and not edims.rhs_out_only_dims + ): + for reduce_op in Partial.LINEAR_REDUCE_OPS: + linearity_placements: list[Placement | _ShardingPlaceholder] = [ + Partial(reduce_op) + ] + [Partial(reduce_op) for _ in input_dims] + strategies_over_one_mesh_dim.append(_maybe_add_bias(linearity_placements)) + + return strategies_over_one_mesh_dim + + +@register_single_dim_strategy(aten.dot.default, allow_unbacked_sharding=True) +def dot_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + return gen_single_dim_einsum_strategies("i,i->") + + +@register_single_dim_strategy(aten.mm.default, allow_unbacked_sharding=True) +def mm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + return gen_single_dim_einsum_strategies("mk,kn->mn") + + +@register_single_dim_strategy(aten.addmm.default, allow_unbacked_sharding=True) +def addmm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + bias_meta = args_schema[0] + if not isinstance(bias_meta, TensorMeta): + raise AssertionError + return gen_single_dim_einsum_strategies("mk,kn->mn", bias_shape=bias_meta.shape) + + +@register_single_dim_strategy(aten.bmm.default, allow_unbacked_sharding=True) +def bmm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + return gen_single_dim_einsum_strategies("bmk,bkn->bmn") + + +@register_single_dim_strategy(aten.baddbmm.default, allow_unbacked_sharding=True) +def baddbmm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + bias_meta = args_schema[0] + if not isinstance(bias_meta, TensorMeta): + raise AssertionError + return gen_single_dim_einsum_strategies("bmk,bkn->bmn", bias_shape=bias_meta.shape) + + +@register_single_dim_strategy(aten._scaled_mm.default, allow_unbacked_sharding=True) +def scaled_mm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + scale_self_meta = args_schema[2] + scale_mat2_meta = args_schema[3] + if not isinstance(scale_self_meta, TensorMeta): + raise AssertionError + if not isinstance(scale_mat2_meta, TensorMeta): + raise AssertionError + if args_schema[4] is not None: + raise AssertionError("_scaled_mm on DTensors doesn't support bias") + if args_schema[5] is not None: + raise AssertionError("_scaled_mm on DTensors doesn't support scale_result") + + # "mk,kn->mn": self_contracting_dim=1, mat2_contracting_dim=0 + base_strategies = gen_single_dim_einsum_strategies("mk,kn->mn") + result = [] + for strat in base_strategies: + # strat is [output, self, mat2]; derive scale placements + scale_self_p = _scaled_mm_scale_placement( + strat[1], scale_self_meta.shape, contracting_dim=1 + ) + scale_mat2_p = _scaled_mm_scale_placement( + strat[2], scale_mat2_meta.shape, contracting_dim=0 + ) + if scale_self_p is None or scale_mat2_p is None: + continue + result.append(strat + [scale_self_p, scale_mat2_p]) + return result + + +def _scaled_dot_product_flash_attention_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + return_debug_mask = len(op_schema.args_schema) >= 6 and op_schema.args_schema[5] + q_input_strategy = op_schema.args_schema[0] + if not isinstance(q_input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(q_input_strategy)}") + # assuming q/k/v have the same shape + + single_mesh_dim_strategies = [] + + # placement list stores placements of [outputs, inputs] + # in the sdpa case, we have 3 valid tensor outputs and 3 tensor inputs + # first we can always accept full replication for both inputs and outputs + all_replicate: PlacementList = [ + Replicate(), + Replicate(), + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + Replicate(), + Replicate(), + Replicate(), + Replicate(), + ] + single_mesh_dim_strategies.append(all_replicate) + + # second we can accept the sharding pattern of tensor parallelism, which + # shard on the num of head dim + qkv_sharding = Shard(1) # num head dim + output_sharding = Shard(1) # num head dim + logsumexp_sharding = Shard(1) # num head dim + if return_debug_mask: + debug_attn_mask_sharding: Placement = Shard(1) # num head dim + else: + # empty debug mask, replicated + debug_attn_mask_sharding = Replicate() + + num_heads_dim_sharding: PlacementList = [ + output_sharding, + logsumexp_sharding, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + debug_attn_mask_sharding, + qkv_sharding, + qkv_sharding, + qkv_sharding, + ] + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # Shard on the batch dimension + debug_attn_mask_sharding = Shard(0) if return_debug_mask else Replicate() + single_mesh_dim_strategies.append( + [ + Shard(0), # output + Shard(0), # logsumexp + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + debug_attn_mask_sharding, # debugattn + Shard(0), # q + Shard(0), # k + Shard(0), # v + ] + ) + return single_mesh_dim_strategies + + +@register_single_dim_strategy( + aten._scaled_dot_product_flash_attention.default, schema_info=RuntimeSchemaInfo(5) +) +def scaled_dot_product_flash_attention_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + q_meta = args_schema[0] + if not isinstance(q_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(q_meta)}") + + return_debug_mask = len(args_schema) >= 6 and args_schema[5] + debug_attn_mask_head: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(1) if return_debug_mask else Replicate() + ) + debug_attn_mask_batch: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(0) if return_debug_mask else Replicate() + ) + + return [ + [ + _ShardingPlaceholder(1), # output + _ShardingPlaceholder(1), # logsumexp + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + debug_attn_mask_head, + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + ], + [ + _ShardingPlaceholder(0), # output + _ShardingPlaceholder(0), # logsumexp + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + debug_attn_mask_batch, + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + ], + ] + + +def _scaled_dot_product_flash_attention_backward_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + q_input_strategy = op_schema.args_schema[1] + if not isinstance(q_input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(q_input_strategy)}") + # assuming q/k/v have the same shape + + tensor_input_indices = [ + i + for i, arg_spec in enumerate(op_schema.args_schema) + if isinstance(arg_spec, OpStrategy) + ] + num_tensor_inputs = len(tensor_input_indices) + + single_mesh_dim_strategies = [] + + # placement list stores placements of [outputs, inputs] + # in the sdpa backward case, we have 3 tensor outputs and 6 to 10 tensor inputs + # first we can always accept full replication for both inputs and outputs + all_replicate: PlacementList = [Replicate()] * (3 + num_tensor_inputs) + + single_mesh_dim_strategies.append(all_replicate) + + # second we can accept the sharding pattern of tensor parallelism, which + # shard on the num of head dim + grad_output_sharding = Shard(1) # num head dim + qkv_sharding = Shard(1) # num head dim + output_sharding = Shard(1) # num head dim + logsumexp_sharding = Shard(1) # num head dim + grad_qkv_sharding = Shard(1) # num head dim + + num_heads_dim_sharding: PlacementList = [ + grad_qkv_sharding, + grad_qkv_sharding, + grad_qkv_sharding, + grad_output_sharding, + qkv_sharding, + qkv_sharding, + qkv_sharding, + output_sharding, + logsumexp_sharding, + ] + # accept replicate on the rest tensor inputs, potentially + # cum_seq_q, cum_seq_k, philox_seed, philox_offset + # at indices 6, 7, 12, 13, respectively + num_heads_dim_sharding.extend([Replicate()] * (num_tensor_inputs - 6)) + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # Batch sharding + batch_dim_sharding: PlacementList = [ + Shard(0), # grad_q + Shard(0), # grad_k + Shard(0), # grad_v + Shard(0), # grad_output + Shard(0), # q + Shard(0), # k + Shard(0), # v + Shard(0), # output + Shard(0), # logsumexp + ] + # accept replicate on the rest tensor inputs, potentially + # cum_seq_q, cum_seq_k, philox_seed, philox_offset + # at indices 6, 7, 12, 13, respectively + batch_dim_sharding.extend([Replicate()] * (num_tensor_inputs - 6)) + single_mesh_dim_strategies.append(batch_dim_sharding) + + return single_mesh_dim_strategies + + +@register_single_dim_strategy(aten._scaled_dot_product_flash_attention_backward.default) +def scaled_dot_product_flash_attention_backward_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + num_tensor_inputs = sum(isinstance(arg, TensorMeta) for arg in args_schema) + if num_tensor_inputs < 6: + raise AssertionError( + f"Expected at least 6 tensor inputs, got {num_tensor_inputs}" + ) + + num_heads_dim_sharding: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(1), # grad_q + _ShardingPlaceholder(1), # grad_k + _ShardingPlaceholder(1), # grad_v + _ShardingPlaceholder(1), # grad_out + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + _ShardingPlaceholder(1), # output + _ShardingPlaceholder(1), # logsumexp + ] + num_heads_dim_sharding.extend([Replicate()] * (num_tensor_inputs - 6)) + + batch_dim_sharding: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(0), # grad_q + _ShardingPlaceholder(0), # grad_k + _ShardingPlaceholder(0), # grad_v + _ShardingPlaceholder(0), # grad_out + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + _ShardingPlaceholder(0), # output + _ShardingPlaceholder(0), # logsumexp + ] + batch_dim_sharding.extend([Replicate()] * (num_tensor_inputs - 6)) + + return [num_heads_dim_sharding, batch_dim_sharding] + + +@register_single_dim_strategy( + aten.constant_pad_nd.default, schema_info=RuntimeSchemaInfo(1) +) +def constant_pad_nd_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # Allow sharding on non-padded dimensions; ban sharding on dims + # that have non-zero padding (where the pad value must be inserted). + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + pad = args_schema[1] + if not isinstance(pad, (list, tuple)): + raise AssertionError(f"Expected list or tuple, got {type(pad)}") + + # pad is [dim_{n-1}_left, dim_{n-1}_right, dim_{n-2}_left, ...] from + # the last dim backwards. Determine which dims have non-zero padding. + padded_dims = set() + for i in range(len(pad) // 2): + if not ( + guard_or_false(pad[i * 2] == 0) and guard_or_false(pad[i * 2 + 1] == 0) + ): + padded_dims.add(ndim - 1 - i) + + # Shard on any non-padded dim: output and input share the same placement. + # All-Replicate is added automatically by the framework. + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(ndim): + if dim not in padded_dims: + strategies.append([_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)]) + + # Partial rules: at padded positions every rank writes the same constant v, + # so reduce(v, v, ..., v) = v for avg/max/min (idempotent). P(sum) only + # works when v=0 since sum(v, ..., v) = N*v != v otherwise. + # When all pad amounts are zero the op is a no-op, so all reduce ops hold. + value = args_schema[2] if len(args_schema) > 2 else 0 + no_padding = all(guard_or_false(pad[i] == 0) for i in range(len(pad))) + if no_padding or guard_or_false(value == 0): + reduce_ops = ("sum", "avg", "max", "min") + else: + reduce_ops = ("avg", "max", "min") + for reduce_op in reduce_ops: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + + return strategies + + +def _scaled_dot_product_efficient_attention_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + q_input_strategy = op_schema.args_schema[0] + if not isinstance(q_input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(q_input_strategy)}") + # assuming q/k/v have the same shape + + has_attn_bias = op_schema.args_schema[3] is not None + compute_log_sumexp = op_schema.args_schema[4] + + single_mesh_dim_strategies: list[PlacementList] = [] + + # placement list stores placements of [outputs, inputs] + # in the sdpa case, we have 2 valid tensor outputs and 3 or 4 tensor inputs + # first we can always accept full replication for both inputs and outputs + all_replicate: PlacementList = [ + Replicate(), + Replicate(), + None, + None, + Replicate(), + Replicate(), + Replicate(), + ] + if has_attn_bias: + all_replicate.append(Replicate()) # attn bias + + single_mesh_dim_strategies.append(all_replicate) + + # second we can accept the sharding pattern of tensor parallelism, which + # shard on the heads dimension + qkv_sharding = Shard(1) + output_sharding = Shard(1) + if compute_log_sumexp: + logsumexp_sharding: Placement = Shard(1) + else: + # empty logsumexp, replicated + logsumexp_sharding = Replicate() + + num_heads_dim_sharding = [ + output_sharding, + logsumexp_sharding, + None, + None, + qkv_sharding, + qkv_sharding, + qkv_sharding, + ] + if has_attn_bias: + num_heads_dim_sharding.append(Shard(1)) + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # batch sharding + if compute_log_sumexp: + logsumexp_sharding_dp: Placement = Shard(0) + else: + # empty logsumexp, replicated + logsumexp_sharding_dp = Replicate() + batch_sharding = [ + Shard(0), # output + logsumexp_sharding_dp, # logsumexp + None, # philox_seed + None, # philox_offset + Shard(0), # q + Shard(0), # k + Shard(0), # v + ] + if has_attn_bias: + batch_sharding.append(Shard(0)) + + single_mesh_dim_strategies.append(batch_sharding) + + return single_mesh_dim_strategies + + +@register_single_dim_strategy( + aten._scaled_dot_product_efficient_attention.default, + schema_info=RuntimeSchemaInfo(4), +) +def scaled_dot_product_efficient_attention_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + q_meta = args_schema[0] + if not isinstance(q_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(q_meta)}") + + has_attn_bias = args_schema[3] is not None + compute_log_sumexp = args_schema[4] + + logsumexp_head: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(1) if compute_log_sumexp else Replicate() + ) + logsumexp_batch: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(0) if compute_log_sumexp else Replicate() + ) + + num_heads_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(1), # output + logsumexp_head, + None, # philox_seed + None, # philox_offset + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + ] + if has_attn_bias: + num_heads_dim_sharding.append(_ShardingPlaceholder(1)) + + batch_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(0), # output + logsumexp_batch, + None, # philox_seed + None, # philox_offset + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + ] + if has_attn_bias: + batch_dim_sharding.append(_ShardingPlaceholder(0)) + + return [num_heads_dim_sharding, batch_dim_sharding] + + +def _scaled_dot_product_efficient_attention_backward_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + q_input_strategy = op_schema.args_schema[1] + if not isinstance(q_input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(q_input_strategy)}") + # assuming q/k/v have the same shape + has_attn_bias = op_schema.args_schema[4] is not None + + single_mesh_dim_strategies = [] + + # placement list stores placements of [outputs, inputs] + # in the sdpa backward case, we have 4 tensor outputs and 8 or 9 tensor inputs + # NOTE: Output sharding of grad_bias on heads dim if attn_bias is present; + # otherwise grad_bias will be empty and its DTensorSpec will be removed. + # first we can always accept full replication for both inputs and outputs + all_replicate: PlacementList = [Replicate()] * (12 + has_attn_bias) + + if not has_attn_bias: + all_replicate[3] = None # grad bias is None if attn_bias is not present + + single_mesh_dim_strategies.append(all_replicate) + + # second we can accept the sharding pattern of tensor parallelism, which + # shard on the heads dimension + grad_output_sharding = Shard(1) + qkv_sharding = Shard(1) + output_sharding = Shard(1) + logsumexp_sharding = Shard(1) + grad_qkv_sharding = Shard(1) + grad_bias_sharding = Shard(1) if has_attn_bias else None + + num_heads_dim_sharding: PlacementList = [ + grad_qkv_sharding, + grad_qkv_sharding, + grad_qkv_sharding, + grad_bias_sharding, + grad_output_sharding, + qkv_sharding, + qkv_sharding, + qkv_sharding, + # the place for optional input attn_bias, + output_sharding, + logsumexp_sharding, + ] + # input sharding of attn_bias on heads dim if present + if has_attn_bias: + num_heads_dim_sharding.insert(8, Shard(1)) + # accept replicate on the rest scalar tensor inputs + # namely philox_seed and philox_offset + num_heads_dim_sharding.extend([Replicate(), Replicate()]) + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # Shards on batch dim + batch_dim_sharding: PlacementList = [ + Shard(0), # grad_q + Shard(0), # grad_k + Shard(0), # grad_v + Shard(0) if has_attn_bias else None, # grad_bias + Shard(0), # grad_output + Shard(0), # q + Shard(0), # k + Shard(0), # v + Shard(0), # output + Shard(0), # logsumexp + ] + # accept replicate on the rest tensor inputs, potentially + # cum_seq_q, cum_seq_k, philox_seed, philox_offset + # at indices 6, 7, 12, 13, respectively + if has_attn_bias: + batch_dim_sharding.insert(8, Shard(0)) + batch_dim_sharding.extend([Replicate(), Replicate()]) + single_mesh_dim_strategies.append(batch_dim_sharding) + + return single_mesh_dim_strategies + + +@register_single_dim_strategy( + aten._scaled_dot_product_efficient_attention_backward.default +) +def scaled_dot_product_efficient_attention_backward_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + has_attn_bias = args_schema[4] is not None + + num_heads_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(1), # grad_q + _ShardingPlaceholder(1), # grad_k + _ShardingPlaceholder(1), # grad_v + _ShardingPlaceholder(1) if has_attn_bias else None, # grad_bias + _ShardingPlaceholder(1), # grad_out + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + ] + if has_attn_bias: + num_heads_dim_sharding.append(_ShardingPlaceholder(1)) + num_heads_dim_sharding.extend( + [ + _ShardingPlaceholder(1), # output + _ShardingPlaceholder(1), # logsumexp + Replicate(), # philox_seed + Replicate(), # philox_offset + ] + ) + + batch_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(0), # grad_q + _ShardingPlaceholder(0), # grad_k + _ShardingPlaceholder(0), # grad_v + _ShardingPlaceholder(0) if has_attn_bias else None, # grad_bias + _ShardingPlaceholder(0), # grad_out + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + ] + if has_attn_bias: + batch_dim_sharding.append(_ShardingPlaceholder(0)) + batch_dim_sharding.extend( + [ + _ShardingPlaceholder(0), # output + _ShardingPlaceholder(0), # logsumexp + Replicate(), # philox_seed + Replicate(), # philox_offset + ] + ) + + return [num_heads_dim_sharding, batch_dim_sharding] + + +def _scaled_dot_product_cudnn_attention_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + ( + query_strategy, # query + _, # key + _, # value + attn_bias_strategy, + compute_log_sumexp, # compute_log_sumexp + *rest_args, # optional args: dropout_p, is_causal, return_debug_mask, scale + ) = op_schema.args_schema + return_debug_mask = len(op_schema.args_schema) >= 8 and rest_args[2] + has_attn_bias = attn_bias_strategy is not None + debug_attn_mask_sharding: Placement | None = ( + Replicate() if return_debug_mask else None + ) + + if not isinstance(query_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(query_strategy)}") + # assuming q/k/v have the same shape + + single_mesh_dim_strategies = [] + + # placement list stores placements of [outputs, inputs] + # in the sdpa case, we have 2 valid tensor outputs and 3 tensor inputs + # first we can always accept full replication for both inputs and outputs + all_replicate: PlacementList = [ + Replicate(), # output + Replicate(), # logsumexp + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + # NOTE: debug_attn_mask is not supported by pytorch and is always an empty tensor + # https://github.com/pytorch/pytorch/blob/60205b0eb2602317856312a66d955c88334ade0b/aten/src/ATen/native/transformers/cuda/attention.cu#L839-L840 + debug_attn_mask_sharding, # debug_attn_mask + Replicate(), # q + Replicate(), # k + Replicate(), # v + ] + if has_attn_bias: + all_replicate.append(Replicate()) # attn bias + + single_mesh_dim_strategies.append(all_replicate) + + # second we can accept the sharding pattern of tensor parallelism, which + # shard on the num of head dim + tp_sharding = Shard(1) # num head dim + qkv_sharding = tp_sharding + output_sharding = tp_sharding + logsumexp_sharding = tp_sharding if compute_log_sumexp else Replicate() + debug_attn_mask_sharding = tp_sharding if return_debug_mask else None + + num_heads_dim_sharding: PlacementList = [ + output_sharding, + logsumexp_sharding, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + debug_attn_mask_sharding, + qkv_sharding, + qkv_sharding, + qkv_sharding, + ] + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # batch parallelism + logsumexp_sharding = Shard(0) if compute_log_sumexp else Replicate() + debug_attn_mask_sharding = Shard(0) if return_debug_mask else None + batch_dim_sharding: PlacementList = [ + Shard(0), # output + logsumexp_sharding, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + debug_attn_mask_sharding, + Shard(0), # q + Shard(0), # k + Shard(0), # v + ] + single_mesh_dim_strategies.append(batch_dim_sharding) + + return single_mesh_dim_strategies + + +@register_single_dim_strategy( + aten._scaled_dot_product_cudnn_attention.default, + schema_info=RuntimeSchemaInfo(4), +) +def scaled_dot_product_cudnn_attention_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + query_meta = args_schema[0] + if not isinstance(query_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(query_meta)}") + + has_attn_bias = args_schema[3] is not None + compute_log_sumexp = args_schema[4] + return_debug_mask = len(args_schema) >= 8 and args_schema[7] + + logsumexp_head: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(1) if compute_log_sumexp else Replicate() + ) + logsumexp_batch: Placement | _ShardingPlaceholder = ( + _ShardingPlaceholder(0) if compute_log_sumexp else Replicate() + ) + debug_attn_mask_head: Placement | _ShardingPlaceholder | None = ( + _ShardingPlaceholder(1) if return_debug_mask else None + ) + debug_attn_mask_batch: Placement | _ShardingPlaceholder | None = ( + _ShardingPlaceholder(0) if return_debug_mask else None + ) + + num_heads_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(1), # output + logsumexp_head, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + debug_attn_mask_head, + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + ] + if has_attn_bias: + num_heads_dim_sharding.append(_ShardingPlaceholder(1)) + + batch_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(0), # output + logsumexp_batch, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + debug_attn_mask_batch, + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + ] + if has_attn_bias: + batch_dim_sharding.append(_ShardingPlaceholder(0)) + + return [num_heads_dim_sharding, batch_dim_sharding] + + +def _scaled_dot_product_cudnn_attention_backward_base_strategies( + op_schema: OpSchema, +) -> list[PlacementList]: + """Helper that returns list of base placement strategies (without CP).""" + if len(op_schema.args_schema) < 15: + raise AssertionError( + f"Expected at least 15 args_schema, got {len(op_schema.args_schema)}" + ) + has_attn_bias = op_schema.args_schema[8] is not None + has_scale = len(op_schema.args_schema) >= 16 and False + + query_strategy = op_schema.args_schema[1] + if not isinstance(query_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(query_strategy)}") + # assuming q/k/v have the same shape + + single_mesh_dim_strategies = [] + + # placement list stores placements of [outputs, inputs] + # cudnn outputs: (Tensor dq, Tensor dk, Tensor dv) + # cudnn inputs: ( + # Tensor grad_out, + # Tensor query, + # Tensor key, + # Tensor value, + # Tensor out, + # Tensor logsumexp, + # Tensor philox_seed, + # Tensor philox_offset, + # Tensor attn_bias, + # Tensor cum_seq_q, + # Tensor cum_seq_k, + # SymInt max_q, + # SymInt max_k, + # float dropout_p, + # bool is_causal, + # int? scale, + # ) + + # case 1: we can always accept full replication for both inputs and outputs + all_replicate_out: PlacementList = [ + Replicate(), # dq + Replicate(), # dk + Replicate(), # dv + ] + all_replicate_inp: PlacementList = [Replicate()] * 6 + all_replicate_inp += [ + Replicate() + ] * 2 # philox_seed, philox_offset is casted to Replicate() in DTensor + all_replicate_inp += [Replicate() if has_attn_bias else None] + all_replicate_inp += [None] * 6 + if has_scale: + all_replicate_inp.append(None) + + all_replicate: PlacementList = all_replicate_out + all_replicate_inp + single_mesh_dim_strategies.append(all_replicate) + + # case 2: we can accept the sharding pattern of tensor parallelism, which + # shards on the num of head dim + qkv_sharding = Shard(1) # num head dim + output_sharding = Shard(1) # num head dim + logsumexp_sharding = Shard(1) # num head dim + + num_heads_dim_sharding_out: PlacementList = [qkv_sharding] * 3 + num_heads_dim_sharding_inp: PlacementList = [qkv_sharding] * 4 + num_heads_dim_sharding_inp += [output_sharding] + num_heads_dim_sharding_inp += [logsumexp_sharding] + num_heads_dim_sharding_inp += [ + Replicate() + ] * 2 # philox_seed, philox_offset is casted to Replicate() in DTensor + num_heads_dim_sharding_inp += [Shard(1) if has_attn_bias else None] + num_heads_dim_sharding_inp += [None] * 6 + if has_scale: + num_heads_dim_sharding_inp.append(None) + + num_heads_dim_sharding = num_heads_dim_sharding_out + num_heads_dim_sharding_inp + single_mesh_dim_strategies.append(num_heads_dim_sharding) + + # case 3: we can accept the sharding pattern of batch parallelism, which + # shards on the batch dimension + qkv_sharding = Shard(0) + output_sharding = Shard(0) + logsumexp_sharding = Shard(0) + + batch_dim_sharding_out: PlacementList = [qkv_sharding] * 3 + batch_dim_sharding_inp: PlacementList = [qkv_sharding] * 4 + batch_dim_sharding_inp += [output_sharding] + batch_dim_sharding_inp += [logsumexp_sharding] + batch_dim_sharding_inp += [ + Replicate() + ] * 2 # philox_seed, philox_offset is casted to Replicate() in DTensor + batch_dim_sharding_inp += [Shard(0) if has_attn_bias else None] + batch_dim_sharding_inp += [None] * 6 + if has_scale: + batch_dim_sharding_inp.append(None) + + batch_dim_sharding = batch_dim_sharding_out + batch_dim_sharding_inp + single_mesh_dim_strategies.append(batch_dim_sharding) + + return single_mesh_dim_strategies + + +@register_single_dim_strategy(aten._scaled_dot_product_cudnn_attention_backward.default) +def scaled_dot_product_cudnn_attention_backward_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + if len(args_schema) < 15: + raise AssertionError(f"Expected at least 15 args, got {len(args_schema)}") + + for arg_index in range(6): + arg = args_schema[arg_index] + if not isinstance(arg, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(arg)}") + + philox_placements: list[Placement] = [] + for arg_index in (6, 7): + arg = args_schema[arg_index] + if isinstance(arg, TensorMeta): + philox_placements.append(Replicate()) + elif not isinstance(arg, torch.Tensor): + raise AssertionError(f"Expected TensorMeta or Tensor, got {type(arg)}") + + has_attn_bias = args_schema[8] is not None + if has_attn_bias and not isinstance(args_schema[8], (TensorMeta, torch.Tensor)): + raise AssertionError( + f"Expected TensorMeta or Tensor, got {type(args_schema[8])}" + ) + + cum_seq_placements: list[None] = [] + for arg_index in (9, 10): + arg = args_schema[arg_index] + if isinstance(arg, TensorMeta): + cum_seq_placements.append(None) + elif arg is None or isinstance(arg, torch.Tensor): + pass + else: + raise AssertionError(f"Expected TensorMeta or Tensor, got {type(arg)}") + + num_heads_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(1), # grad_q + _ShardingPlaceholder(1), # grad_k + _ShardingPlaceholder(1), # grad_v + _ShardingPlaceholder(1), # grad_out + _ShardingPlaceholder(1), # q + _ShardingPlaceholder(1), # k + _ShardingPlaceholder(1), # v + _ShardingPlaceholder(1), # output + _ShardingPlaceholder(1), # logsumexp + ] + num_heads_dim_sharding.extend(philox_placements) + if has_attn_bias and isinstance(args_schema[8], TensorMeta): + num_heads_dim_sharding.append(_ShardingPlaceholder(1)) + num_heads_dim_sharding.extend(cum_seq_placements) + + batch_dim_sharding: list[Placement | _ShardingPlaceholder | None] = [ + _ShardingPlaceholder(0), # grad_q + _ShardingPlaceholder(0), # grad_k + _ShardingPlaceholder(0), # grad_v + _ShardingPlaceholder(0), # grad_out + _ShardingPlaceholder(0), # q + _ShardingPlaceholder(0), # k + _ShardingPlaceholder(0), # v + _ShardingPlaceholder(0), # output + _ShardingPlaceholder(0), # logsumexp + ] + batch_dim_sharding.extend(philox_placements) + if has_attn_bias and isinstance(args_schema[8], TensorMeta): + batch_dim_sharding.append(_ShardingPlaceholder(0)) + batch_dim_sharding.extend(cum_seq_placements) + + return [num_heads_dim_sharding, batch_dim_sharding] + + +def _valid_grouped_mm_strides( + mesh: DeviceMesh, + op_schema: OpSchema, + input_specs: list[DTensorSpec], + output_specs: DTensorSpec | tuple[DTensorSpec | None, ...], +) -> bool: + def local_meta(spec: DTensorSpec) -> TensorMeta: + if not isinstance(spec.tensor_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(spec.tensor_meta)}") + meta: TensorMeta = spec.tensor_meta + local_shape, _ = compute_local_shape_and_global_offset( + meta.shape, mesh, spec.placements, skip_offset=True + ) + local_stride = compute_local_stride(meta.stride, local_shape) + return TensorMeta(torch.Size(local_shape), local_stride, meta.dtype) + + def check_valid_strides(meta: TensorMeta) -> bool: + # copied from `_meta_grouped_mm_common` in meta_registrations.py + end_dim = len(meta.shape) - 1 + alignment = 16 // meta.dtype.itemsize + if meta.stride[end_dim - 1] == 1 and meta.stride[end_dim] >= max( + 1, meta.shape[end_dim - 1] + ): + return meta.stride[end_dim] % alignment == 0 + elif meta.stride[end_dim] == 1 and meta.stride[end_dim - 1] >= max( + 1, meta.shape[end_dim] + ): + return meta.stride[end_dim - 1] % alignment == 0 + else: + return False + + mat1_meta = local_meta(input_specs[0]) + mat2_meta = local_meta(input_specs[1]) + return check_valid_strides(mat1_meta) and check_valid_strides(mat2_meta) + + +@register_single_dim_strategy( + aten._grouped_mm.default, + full_mesh_strategy_filter=_valid_grouped_mm_strides, +) +def grouped_mm_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder | None]]: + mat1_meta = args_schema[0] + if not isinstance(mat1_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(mat1_meta)}") + mat2_meta = args_schema[1] + if not isinstance(mat2_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(mat2_meta)}") + + if len(args_schema) > 3 and args_schema[3] is not None: + raise AssertionError("grouped_mm doesn't support bias yet") + + tensor_inputs_tail: list[Placement | None] = [] + if len(args_schema) > 2 and args_schema[2] is not None: + if not isinstance(args_schema[2], TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(args_schema[2])}") + tensor_inputs_tail.append(Replicate()) + + strategies: list[list[Placement | _ShardingPlaceholder | None]] = [ + [Partial(), Partial(), Replicate(), *tensor_inputs_tail], + [Partial(), Replicate(), Partial(), *tensor_inputs_tail], + ] + + mat1_ndim = len(mat1_meta.shape) + mat2_ndim = len(mat2_meta.shape) + + if mat1_ndim == 2 and mat2_ndim == 3: + strategies.extend( + [ + [ + _ShardingPlaceholder(1), + Replicate(), + _ShardingPlaceholder(2), + *tensor_inputs_tail, + ], + [ + Partial(), + _ShardingPlaceholder(1), + _ShardingPlaceholder(1), + *tensor_inputs_tail, + ], + ] + ) + + if mat1_ndim == 3 and mat2_ndim == 2: + strategies.extend( + [ + [ + Partial(), + _ShardingPlaceholder(2), + _ShardingPlaceholder(0), + *tensor_inputs_tail, + ], + [ + _ShardingPlaceholder(0), + _ShardingPlaceholder(1), + Replicate(), + *tensor_inputs_tail, + ], + ] + ) + + if mat1_ndim == 2 and mat2_ndim == 2: + strategies.extend( + [ + [ + _ShardingPlaceholder(2), + Replicate(), + _ShardingPlaceholder(1), + *tensor_inputs_tail, + ], + [ + _ShardingPlaceholder(1), + _ShardingPlaceholder(0), + Replicate(), + *tensor_inputs_tail, + ], + ] + ) + + if mat1_ndim == 3 and mat2_ndim == 3: + strategies.extend( + [ + [ + _ShardingPlaceholder(2), + Replicate(), + _ShardingPlaceholder(2), + *tensor_inputs_tail, + ], + [ + _ShardingPlaceholder(1), + _ShardingPlaceholder(1), + Replicate(), + *tensor_inputs_tail, + ], + [ + Partial(), + _ShardingPlaceholder(2), + _ShardingPlaceholder(1), + *tensor_inputs_tail, + ], + [ + _ShardingPlaceholder(0), + _ShardingPlaceholder(0), + _ShardingPlaceholder(0), + *tensor_inputs_tail, + ], + ] + ) + + return strategies diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_pointwise_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_pointwise_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..dfc253eff8e860860a6fe76cb21f259907508389 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_pointwise_ops.py @@ -0,0 +1,592 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from collections.abc import Callable + +import torch +from torch._ops import OpOverload +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ArgsType, KwargsType, RuntimeSchemaInfo +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import infer_broadcast_dims_map +from torch.distributed.tensor.placement_types import Partial, Placement, Replicate + + +aten = torch.ops.aten +prims = torch.ops.prims + +# Linear pointwise ops, split by linearity type. +unary_linear_ops = [aten.to.dtype] + + +def _common_pointwise_single_dim_strategy( + partial_extra_rules: list[list[Placement | _ShardingPlaceholder]] | None = None, +) -> Callable[ + [OpOverload, ArgsType, KwargsType], list[list[Placement | _ShardingPlaceholder]] +]: + """Factory for single-dim strategies that add partial placement rules. + + Returns strategies shaped [output, *args] only. Tensor kwarg placements + (e.g. ``out``, ``lr``) are appended by the wrapper in + ``_register_single_dim_pointwise``. + """ + + def strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, + ) -> list[list[Placement | _ShardingPlaceholder]]: + tensor_arg_metas: list[TensorMeta] = [ + arg for arg in args_schema if isinstance(arg, TensorMeta) + ] + common_shape = torch.broadcast_shapes( + *[arg.shape for arg in args_schema if isinstance(arg, TensorMeta)] + ) + # For multi-output ops (e.g. frexp), all outputs share the same + # pointwise sharding, so replicate the output placement. + num_outputs = sum(1 for r in op._schema.returns if "Tensor" in str(r.type)) + placements: list[list[Placement | _ShardingPlaceholder]] = [] + for i in range(len(common_shape)): + shard_placements: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(i) + ] * num_outputs + for arg in tensor_arg_metas: + common_dim_to_arg_dim = infer_broadcast_dims_map( + common_shape, arg.shape + ) + # If the output shard dim maps to an input dim, shard that + # input dim; otherwise it was broadcast, so replicate. + if common_dim_to_arg_dim[i] >= 0: + shard_placements.append( + _ShardingPlaceholder(common_dim_to_arg_dim[i]) + ) + else: + shard_placements.append(Replicate()) + placements.append(shard_placements) + if partial_extra_rules: + n_tensors = len(tensor_arg_metas) + expected_len = num_outputs + n_tensors + for rule in partial_extra_rules: + # Filter rather than assert: some ops (e.g. mul.Tensor) mix + # unary rules (len 2, for scalar promotion) and binary rules + # (len 3, for tensor-tensor), so mismatched lengths are expected. + # see _MUL_RULES to see how _UNARY_LINEAR_RULES handles the + # scalar promotion case + if len(rule) == expected_len: + placements.append(rule) + return placements + + return strategy + + +def _is_list_op(op: OpOverload) -> bool: + """Returns True if op is a foreach, amp_foreach, or fused op.""" + name = op.name() + return name.startswith(("aten::_foreach_", "aten::_amp_foreach_", "aten::_fused_")) + + +# The state_steps arg of fused adam / adamw is a Replicate scalar tensor, which will be put on +# the compute_mesh of an op across all parameter groups, even when not all parameter groups +# are on the same device mesh. This idx will help avoid hitting exceptions or unnecessary +# redistribute during sharding propagation. +_FUSED_OP_SCALAR_IDX = 5 + +# Ops registered with extra Partial rules; populated by _register_single_dim_pointwise +# when partial_extra_rules is not None, to avoid double-registration from tag discovery. +_specially_registered_ops: set[OpOverload] = set() + + +def _register_single_dim_pointwise( + op: OpOverload, + partial_extra_rules: list[list[Placement]] | None = None, + static_argnum: int = 0, +) -> None: + if partial_extra_rules is not None: + _specially_registered_ops.add(op) + inner_fn = _common_pointwise_single_dim_strategy( + partial_extra_rules=partial_extra_rules # pyrefly: ignore[bad-argument-type] + ) + + # Wrap to append tensor kwarg placements in schema declaration order. + # out = output placement (s[0]); everything else (e.g. lr) = Replicate. + # TODO: move kwargs handling upstream if this works + def strategy_fn( + op: OpOverload, + args: ArgsType, + kwargs: KwargsType, + _fn: Callable = inner_fn, + ) -> list[list[Placement | _ShardingPlaceholder]]: + strategies = _fn(op, args, kwargs) + kw_names = [k for k, v in kwargs.items() if isinstance(v, TensorMeta)] + if not kw_names: + return strategies + return [ + s + [s[0] if name == "out" else Replicate() for name in kw_names] + for s in strategies + ] + + if _is_list_op(op): + schema_info = RuntimeSchemaInfo(needs_pytree=True) + else: + schema_info = RuntimeSchemaInfo(static_argnum, static_kwargkey=["out"]) + # Fused ops (e.g. _fused_adam_) have state_steps on a potentially different + # mesh; see the note in expand_to_full_mesh_op_strategy for details. + different_mesh_args: list[int] | None = None + if op.name().startswith("aten::_fused_"): + different_mesh_args = [_FUSED_OP_SCALAR_IDX] + register_single_dim_strategy( + op, + schema_info=schema_info, + allow_uneven_sharding=True, + allow_unbacked_sharding=True, + different_mesh_args=different_mesh_args, + )(strategy_fn) + + +_UNARY_LINEAR_RULES: list[list[Placement]] = [ + [Partial("sum"), Partial("sum")], + [Partial("avg"), Partial("avg")], +] + +binary_additive_ops = [ + aten.add.Tensor, + aten.add_.Tensor, + aten.add.out, + aten.sub.Tensor, + aten.sub_.Tensor, + aten.sub.out, + # foreach variants + aten._foreach_add.List, + aten._foreach_add_.List, + aten._foreach_sub.List, + aten._foreach_sub_.List, +] + +_BINARY_ADDITIVE_RULES: list[list[Placement]] = [ + [Partial("sum"), Partial("sum"), Partial("sum")], + [Partial("avg"), Partial("avg"), Partial("avg")], + # P(x), R -> P(x): adding/subtracting a replicated value preserves partial types + # avg, max, min. sum would result in R being added n times, n = num_ranks + # (the replicated value is constant across ranks, so reduce order is unaffected) + [Partial("avg"), Partial("avg"), Replicate()], + [Partial("max"), Partial("max"), Replicate()], + [Partial("min"), Partial("min"), Replicate()], + # R, P(avg) -> P(avg): avg is linear so this holds for any alpha + # (R, P(max/min) excluded: negative alpha would flip the ordering) + [Partial("avg"), Replicate(), Partial("avg")], +] + +for op in binary_additive_ops: + _register_single_dim_pointwise(op, _BINARY_ADDITIVE_RULES) + +# mul: partials propagate through either arg. div: only through numerator. +binary_mul_ops = [ + aten.mul.Tensor, + aten.mul_.Tensor, + aten.mul.out, + # foreach variants + aten._foreach_mul.List, + aten._foreach_mul_.List, + aten._foreach_mul.Tensor, + aten._foreach_mul_.Tensor, +] +binary_div_ops = [ + aten.div.Tensor, + aten.div_.Tensor, + aten.div.out, + # foreach variants + aten._foreach_div.List, + aten._foreach_div_.List, + aten._foreach_div.Tensor, + aten._foreach_div_.Tensor, +] + +# _UNARY_LINEAR_RULES handles the scalar promotion case: Python's __mul__/__truediv__ +# promote scalars to 0-dim tensors, so aten.mul.Scalar dispatches as aten.mul.Tensor +# with n_tensors=1, matching the length-2 unary rules. +_MUL_RULES: list[list[Placement]] = [ + [Partial("sum"), Partial("sum"), Replicate()], + [Partial("avg"), Partial("avg"), Replicate()], + [Partial("sum"), Replicate(), Partial("sum")], + [Partial("avg"), Replicate(), Partial("avg")], +] + +_DIV_RULES: list[list[Placement]] = [ + [Partial("sum"), Partial("sum"), Replicate()], + [Partial("avg"), Partial("avg"), Replicate()], +] + +for op in binary_mul_ops: + _register_single_dim_pointwise(op, _UNARY_LINEAR_RULES + _MUL_RULES) + +for op in binary_div_ops: + _register_single_dim_pointwise(op, _UNARY_LINEAR_RULES + _DIV_RULES) + +scalar_linear_ops = [ + aten.div.Scalar, + aten.div_.Scalar, + aten.mul.Scalar, + aten.mul_.Scalar, + # foreach variants + aten._foreach_div.Scalar, + aten._foreach_div_.Scalar, + aten._foreach_mul.Scalar, + aten._foreach_mul_.Scalar, + aten._foreach_div.ScalarList, + aten._foreach_div_.ScalarList, + aten._foreach_mul.ScalarList, + aten._foreach_mul_.ScalarList, +] + +for op in scalar_linear_ops: + _register_single_dim_pointwise(op, _UNARY_LINEAR_RULES, static_argnum=1) + + +# Non-decreasing unary ops: f(max(a,b)) = max(f(a),f(b)). +# Only ops that are non-decreasing on their ENTIRE domain belong here. +# Ops with restricted domains (e.g. log on (0,∞), asin on [-1,1]) do NOT qualify +# because P(max) offsets can push inputs outside the valid domain. +non_decreasing_unary_ops = [ + aten.asinh.default, + aten.asinh_.default, + aten.asinh.out, + aten.atan.default, + aten.atan_.default, + aten.atan.out, + aten.ceil.default, + aten.ceil_.default, + aten.ceil.out, + aten.deg2rad.default, + aten.deg2rad_.default, + aten.deg2rad.out, + aten.erf.default, + aten.erf_.default, + aten.erf.out, + aten.exp.default, + aten.exp_.default, + aten.exp.out, + aten.exp2.default, + aten.exp2_.default, + aten.exp2.out, + aten.expm1.default, + aten.expm1_.default, + aten.expm1.out, + aten.floor.default, + aten.floor_.default, + aten.floor.out, + aten.rad2deg.default, + aten.rad2deg_.default, + aten.rad2deg.out, + aten.relu.default, + aten.relu_.default, + aten.round.decimals, + aten.round.default, + aten.round_.decimals, + aten.round_.default, + aten.round.decimals_out, + aten.round.out, + aten.sgn.default, + aten.sgn_.default, + aten.sgn.out, + aten.sigmoid.default, + aten.sigmoid_.default, + aten.sigmoid.out, + aten.sign.default, + aten.sign_.default, + aten.sign.out, + aten.sinh.default, + aten.sinh_.default, + aten.sinh.out, + aten.tanh.default, + aten.tanh_.default, + aten.tanh.out, + aten.trunc.default, + aten.trunc_.default, + aten.trunc.out, + # nan_to_num is non-decreasing on its entire domain (including nan/inf): + # it maps -inf→min, nan→0, inf→max, and is identity elsewhere. + aten.nan_to_num.default, + aten.nan_to_num_.default, + aten.nan_to_num.out, + # hardshrink: x if |x|>lambd else 0. Non-decreasing on entire domain. + aten.hardshrink.default, + # I1(x) is monotonically non-decreasing for all real x. + aten.special_modified_bessel_i1.default, + # threshold(x, t, v): x if x > t else v. Non-decreasing for v <= t (the + # common case, including the default v=0, t=0). + aten.threshold.default, + # foreach variants + aten._foreach_exp.default, + aten._foreach_exp_.default, + aten._foreach_clamp_max_.Scalar, + aten._foreach_clamp_min_.Scalar, +] + +_NON_DECREASING_RULES: list[list[Placement]] = [ + [Partial("max"), Partial("max")], + [Partial("min"), Partial("min")], +] + +for op in non_decreasing_unary_ops: + _register_single_dim_pointwise(op, _NON_DECREASING_RULES) + +# Non-increasing unary ops: f(max(a,b)) = min(f(a),f(b)). +# Note: acos excluded due to domain constraints [-1,1] causing validation failures +non_increasing_unary_ops: list[OpOverload] = [ + aten.erfc.default, + aten.erfc_.default, + aten.erfc.out, + aten.special_erfcx.default, + aten.special_erfcx.out, +] + +_NON_INCREASING_RULES: list[list[Placement]] = [ + [Partial("min"), Partial("max")], + [Partial("max"), Partial("min")], +] + +for op in non_increasing_unary_ops: + _register_single_dim_pointwise(op, _NON_INCREASING_RULES) + +# Bessel K functions are strictly decreasing for x > 0 but undefined at x <= 0. +# Only P(min)->P(max) is safe: P(min) offsets add positive values to the +# non-holding rank, keeping all inputs positive. P(max) offsets subtract, +# which can push inputs to x <= 0 producing NaN. +_POSITIVE_DOMAIN_NON_INCREASING_RULES: list[list[Placement]] = [ + [Partial("max"), Partial("min")], +] + +for op in [ + aten.special_modified_bessel_k0.default, + aten.special_modified_bessel_k1.default, + aten.special_scaled_modified_bessel_k0.default, + aten.special_scaled_modified_bessel_k1.default, +]: + _register_single_dim_pointwise(op, _POSITIVE_DOMAIN_NON_INCREASING_RULES) + +# neg is linear: -(A1 + A2) = -A1 + -A2 +neg_ops = [ + aten.neg.default, + aten.neg_.default, + aten.neg.out, + # foreach variants + aten._foreach_neg.default, + aten._foreach_neg_.default, +] + +_NEG_RULES: list[list[Placement]] = _UNARY_LINEAR_RULES + _NON_INCREASING_RULES + +for op in neg_ops: + _register_single_dim_pointwise(op, _NEG_RULES) + +# xlog1py(x, y) = x * log1p(y). Linear in x with y replicated: +# (a+b)*log1p(y) = a*log1p(y) + b*log1p(y). +_XLOG1PY_RULES: list[list[Placement]] = [ + [Partial("sum"), Partial("sum"), Replicate()], + [Partial("avg"), Partial("avg"), Replicate()], +] + +for op in [aten.special_xlog1py.default, aten.special_xlog1py.other_scalar]: + _register_single_dim_pointwise(op, _XLOG1PY_RULES) + + +# All-partial-preserving unary ops: P(x)->P(x) for all x. +# TODO: positive should be removed once CIA (Copy Is All) optimizes it away. +all_partial_preserving_unary_ops = [ + aten.to.dtype, + aten.positive.default, +] + +_ALL_PARTIAL_PRESERVING_RULES: list[list[Placement]] = [ + [Partial(r), Partial(r)] for r in ("sum", "avg", "max", "min") +] + +for op in all_partial_preserving_unary_ops: + _register_single_dim_pointwise(op, _ALL_PARTIAL_PRESERVING_RULES) + +all_partial_preserving_binary_ops = [ + aten.copy_.default, + prims.copy_to.default, +] + +_ALL_PARTIAL_BINARY_PRESERVING_RULES: list[list[Placement]] = [ + [Partial(r), Partial(r), Partial(r)] for r in ("sum", "avg", "max", "min") +] + +for op in all_partial_preserving_binary_ops: + _register_single_dim_pointwise(op, _ALL_PARTIAL_BINARY_PRESERVING_RULES) + +# Monotonic increasing in both args but don't preserve any specific partial type. +monotonic_binary_ops = [ + aten.logaddexp.default, + aten.logaddexp.out, + aten.logaddexp2.default, + aten.logaddexp2.out, +] + +_MONOTONE_BINARY_BASE_RULES: list[list[Placement]] = [ + [Partial("max"), Partial("max"), Replicate()], + [Partial("max"), Replicate(), Partial("max")], + [Partial("min"), Partial("min"), Replicate()], + [Partial("min"), Replicate(), Partial("min")], +] + +for op in monotonic_binary_ops: + _register_single_dim_pointwise(op, _MONOTONE_BINARY_BASE_RULES) + +# Binary ops monotonically increasing in both arguments. +# max-preserving: P(max)+P(max)->P(max) because max(max(a),max(b)) = max(a,b) +monotonic_max_preserving_binary_ops = [ + aten.clamp_min.Tensor, + aten.fmax.default, + aten.fmax.out, + aten.maximum.default, + aten.maximum.out, + prims.fmax.default, + # foreach variants + aten._foreach_maximum_.List, +] + +_MONOTONE_MAX_PRESERVING_BINARY_BASE_RULES: list[list[Placement]] = [ + *_MONOTONE_BINARY_BASE_RULES, + [Partial("max"), Partial("max"), Partial("max")], +] + +for op in monotonic_max_preserving_binary_ops: + _register_single_dim_pointwise(op, _MONOTONE_MAX_PRESERVING_BINARY_BASE_RULES) + +# min-preserving: P(min)+P(min)->P(min) because min(min(a),min(b)) = min(a,b) +monotonic_min_preserving_binary_ops = [ + aten.clamp_max.Tensor, + aten.fmin.default, + aten.fmin.out, + aten.minimum.default, + aten.minimum.out, + prims.fmin.default, +] + +_MONOTONE_MIN_PRESERVING_BINARY_BASE_RULES: list[list[Placement]] = [ + *_MONOTONE_BINARY_BASE_RULES, + [Partial("min"), Partial("min"), Partial("min")], +] + +for op in monotonic_min_preserving_binary_ops: + _register_single_dim_pointwise(op, _MONOTONE_MIN_PRESERVING_BINARY_BASE_RULES) + + +# Ops that are pointwise for DTensor purposes but lack torch.Tag.pointwise. +# TODO(pianpwk): add torch.Tag.pointwise to these ops in native_functions.yaml +# so this list can be removed. +_extra_pointwise_ops: list[OpOverload] = [ + aten.__irshift__.Scalar, + aten.__irshift__.Tensor, + aten._conj.default, + aten.abs_.default, + aten.copysign_.Scalar, + aten.copysign_.Tensor, + aten.exponential_.default, + aten.float_power.Scalar, + aten.float_power.Scalar_out, + aten.float_power.Tensor_Scalar, + aten.float_power.Tensor_Scalar_out, + aten.float_power.Tensor_Tensor, + aten.float_power.Tensor_Tensor_out, + aten.masked_fill_.Scalar, + aten.native_dropout_backward.out, + aten.polygamma_.default, + aten.rrelu_with_noise.default, + aten.where.self_out, + aten.xlogy_.Scalar_Other, + prims.bessel_i0e.default, + prims.bessel_i1.default, + prims.bessel_i1e.default, + prims.bessel_j0.default, + prims.bessel_j1.default, + prims.div.default, + prims.erfcx.default, + prims.frexp.default, + prims.gcd.default, + prims.ndtri.default, + prims.ne.default, + prims.spherical_bessel_j0.default, + prims.zeta.default, + # foreach variants + aten._foreach_abs.default, + aten._foreach_abs_.default, + aten._foreach_addcdiv_.Scalar, + aten._foreach_addcdiv_.ScalarList, + aten._foreach_addcdiv_.Tensor, + aten._foreach_addcmul.Scalar, + aten._foreach_addcmul_.Scalar, + aten._foreach_addcmul_.ScalarList, + aten._foreach_addcmul_.Tensor, + aten._foreach_lerp_.Scalar, + aten._foreach_pow.List, + aten._foreach_pow.ScalarList, + aten._foreach_reciprocal_.default, + aten._foreach_sub.Scalar, + aten._foreach_sub_.Scalar, + aten._foreach_sub.ScalarList, + aten._foreach_sub_.ScalarList, + aten._foreach_sqrt.default, + aten._foreach_sqrt_.default, + aten._foreach_zero_.default, + aten._foreach_cos.default, + aten._foreach_cos_.default, + aten._foreach_log.default, + aten._foreach_log_.default, + aten._amp_foreach_non_finite_check_and_unscale_.default, + # foreach linearity variants + aten._foreach_add.Scalar, + aten._foreach_add_.Scalar, + aten._foreach_add_.ScalarList, + # fused optimizer ops + aten._fused_adam_.default, + aten._fused_adam.default, + aten._fused_adam.tensor_lr, + aten._fused_adam_.tensor_lr, + aten._fused_adamw_.default, + aten._fused_adamw.default, + aten._fused_adamw.tensor_lr, + aten._fused_adamw_.tensor_lr, +] + + +def _get_pointwise_ops_from_tag() -> list[OpOverload]: + """ + Auto-discover pointwise ops via torch.Tag.pointwise, from ops.aten, ops.prims. + """ + ops = [] + for ns in [torch.ops.aten, torch.ops.prims]: + for attr_name in dir(ns): + attr = getattr(ns, attr_name) + if isinstance(attr, torch._ops.OpOverloadPacket): + for overload_name in attr.overloads(): + op = getattr(attr, overload_name) + if torch.Tag.pointwise in op.tags: + ops.append(op) + return ops + + +pointwise_ops = [ + op + for op in _get_pointwise_ops_from_tag() + _extra_pointwise_ops + if op not in _specially_registered_ops +] + + +for op in pointwise_ops: + _register_single_dim_pointwise(op) + + +def register_inductor_prims() -> None: + """Register DTensor sharding strategies for inductor prims ops. + + Called lazily because inductor prims are created via make_prim() in + torch._inductor.inductor_prims, which is imported after this module. + """ + # TODO: handle other inductor prims ops that may need DTensor sharding + # strategies (e.g. mul_rn, div_rn). Those are more complicated and not + # necessarily pointwise. + _register_single_dim_pointwise(prims.fma.default) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_random_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_random_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..af023f12544314e18a277ca5f96ce3234918461b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_random_ops.py @@ -0,0 +1,81 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import torch +from torch._ops import OpOverload +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ArgsType, KwargsType +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor.placement_types import Placement + + +aten = torch.ops.aten + + +def _random_inplace_single_dim_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Single-dim strategy for in-place random ops (single tensor input, output follows input). + + No Partial inputs: random sampling on partial tensors is undefined. + """ + self_meta = args_schema[0] + if not isinstance(self_meta, TensorMeta): + raise AssertionError + num_outputs = sum(1 for r in op._schema.returns if "Tensor" in str(r.type)) + placements: list[list[Placement | _ShardingPlaceholder]] = [] + for i in range(len(self_meta.shape)): + rule: list[Placement | _ShardingPlaceholder] = [_ShardingPlaceholder(i)] * ( + num_outputs + 1 + ) + placements.append(rule) + return placements + + +# In-place random sampling ops: output follows input sharding exactly. +_inplace_random_ops = [ + aten.normal_.default, + aten.uniform_.default, + aten.native_dropout.default, + aten.bernoulli_.float, + aten.bernoulli.default, + aten.log_normal_.default, + aten.exponential_.default, + aten.geometric_.default, +] + +for _op in _inplace_random_ops: + register_single_dim_strategy(_op, allow_uneven_sharding=True)( + _random_inplace_single_dim_strategy + ) + + +@register_single_dim_strategy(aten.multinomial.default) +def multinomial_single_dim_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Single-dim strategy for multinomial. + + multinomial(self, num_samples, ...) -> Tensor + Input: [*, n_categories], Output: [*, num_samples] (dtype=long) + + Only batch dims (all except the last) can be sharded — the last dim + (categories) is consumed by the sampling and maps to a different + semantic dim (num_samples) in the output. + """ + self_meta = args_schema[0] + if not isinstance(self_meta, TensorMeta): + raise AssertionError + placements: list[list[Placement | _ShardingPlaceholder]] = [] + for i in range(len(self_meta.shape) - 1): + rule: list[Placement | _ShardingPlaceholder] = [ + _ShardingPlaceholder(i), + _ShardingPlaceholder(i), + ] + placements.append(rule) + return placements diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_tensor_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_tensor_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..c612e12b94cedc8b71bc1c151229d2a93b583d95 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_tensor_ops.py @@ -0,0 +1,1362 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +from collections.abc import Callable, Sequence, Sized +from typing import cast + +import torch +from torch._ops import OpOverload +from torch._prims_common import IntLike +from torch.distributed.tensor._dtensor_spec import DTensorSpec +from torch.distributed.tensor._op_schema import ( + ArgsType, + KwargsType, + OpSchema, + OpSpec, + OpStrategy, + RuntimeSchemaInfo, + StrategyType, + TensorMeta, + TupleStrategy, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import ( + generate_redistribute_costs, + is_tensor_dim_sharded, + normalize_dim, + register_op_strategy, + shift_shard_dims_after_insert, + shift_shard_dims_after_remove, +) +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _MaskPartial, + Partial, + Placement, + Replicate, + Shard, +) +from torch.fx.experimental.symbolic_shapes import statically_known_true + + +aten = torch.ops.aten +prims = torch.ops.prims + +_PARTIAL_PASS_THROUGH_REDUCE_OPS = ("sum", "avg", "min", "max") + + +@register_single_dim_strategy( + [ + aten.clone.default, + aten.contiguous.default, + aten.detach.default, + aten.detach_.default, + aten.alias.default, + aten.fill_.Scalar, + aten.view.dtype, + aten.zero_.default, + prims.view_of.default, + ], + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def propagate_single_input_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Propagate placements for ops with one tensor input. + + This is the single-dim form of the old one-to-one OpStrategy: a sharded + input dim produces the same sharded output dim. Supported Partial + placements are also forwarded with the same pending reduction. + """ + input_meta = cast(TensorMeta, args_schema[0]) + strategies: list[list[Placement | _ShardingPlaceholder]] = [ + [_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)] + for dim in range(len(input_meta.shape)) + ] + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +def _partial_needs_reduce_for_dtype_cast( + reduce_op: str, + src_dtype: torch.dtype, + target_dtype: torch.dtype | None, +) -> bool: + """Return True when reduce_op does not commute with the dtype cast.""" + if target_dtype is None or src_dtype == target_dtype: + return False + if target_dtype == torch.bool: + return True + if reduce_op in ("max", "min"): + return False + return src_dtype.is_floating_point and not target_dtype.is_floating_point + + +@register_single_dim_strategy( + aten._to_copy.default, + schema_info=RuntimeSchemaInfo(static_kwargkey=["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def _to_copy_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = cast(TensorMeta, args_schema[0]) + src_dtype = input_meta.dtype + target_dtype = cast(torch.dtype | None, kwargs_schema.get("dtype", None)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + strategies.append([_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)]) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + if not _partial_needs_reduce_for_dtype_cast(reduce_op, src_dtype, target_dtype): + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +@register_single_dim_strategy( + [ + aten.equal.default, + aten.is_same_size.default, + ], + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def equal_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Require both tensor operands to use the same non-partial layout. + + equal/is_same_size compare corresponding local shards, so mismatched + shardings would compare different global elements. Partial inputs must use + the implicit Replicate fallback because comparing unreduced partial values + would be invalid. If either operand is scalar, no sharded rows are + produced and the implicit Replicate rule is used. + """ + self_meta = cast(TensorMeta, args_schema[0]) + other_meta = cast(TensorMeta, args_schema[1]) + return [ + [_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)] + for dim in range(min(len(self_meta.shape), len(other_meta.shape))) + ] + + +register_single_dim_strategy( + aten.empty_like.default, + schema_info=RuntimeSchemaInfo(1, ["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +)(propagate_single_input_single_dim_strategy) + + +@register_single_dim_strategy( + [ + aten.ones_like.default, + aten.rand_like.default, + aten.randn_like.default, + aten.zeros_like.default, + ], + schema_info=RuntimeSchemaInfo(1, ["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +@register_single_dim_strategy( + [aten.full_like.default], + schema_info=RuntimeSchemaInfo(2, ["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +@register_single_dim_strategy( + [ + aten.randint_like.default, + aten.randint_like.low_dtype, + aten.randint_like.low_dtype_out, + ], + schema_info=RuntimeSchemaInfo(3, ["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def create_like_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # These factory-like ops create tensors with the input shape but with + # content independent of the input. Sharding can follow the input, but a + # Partial input maps to a Replicate output so generated values are not + # reduced as if they were accumulated partials. + input_meta = cast(TensorMeta, args_schema[0]) + kw_names = [k for k, v in kwargs_schema.items() if isinstance(v, TensorMeta)] + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + placement = _ShardingPlaceholder(dim) + strategies.append( + [placement, placement] + + [placement if name == "out" else Replicate() for name in kw_names] + ) + for reduce_op in Partial.ALL_REDUCE_OPS: + strategies.append( + [Replicate(), Partial(reduce_op)] + [Replicate() for _ in kw_names] + ) + return strategies + + +@register_single_dim_strategy( + [ + aten.new_empty.default, + aten.new_full.default, + aten.new_ones.default, + aten.new_zeros.default, + aten.new_empty_strided.default, + ], + schema_info=RuntimeSchemaInfo(1, ["dtype"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def new_factory_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # Replicated output is always legal and is inserted by the single-dim + # infra. If input and output have the same shape, sharded inputs can also + # propagate to the new tensor. + input_meta = cast(TensorMeta, args_schema[0]) + output_shape = cast(Sequence[object], args_schema[1]) + same_shape = tuple(input_meta.shape) == tuple(output_shape) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + placement = _ShardingPlaceholder(dim) + if same_shape: + strategies.append([placement, placement]) + else: + strategies.append([Replicate(), placement]) + + # Uninitialized factories (new_empty*) can propagate Partial when the shape + # matches because the storage is expected to be overwritten later, e.g. + # autograd's new_empty_strided + copy_ path. Initialized factories keep a + # Replicate output for Partial inputs to avoid reducing generated constants. + if op in (aten.new_empty.default, aten.new_empty_strided.default): + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + if same_shape: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + else: + strategies.append([Replicate(), Partial(reduce_op)]) + else: + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Replicate(), Partial(reduce_op)]) + + return strategies + + +@register_single_dim_strategy(aten.bucketize.Tensor) +def bucketize_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Bucketize returns indices into a sorted boundary tensor. + + Three families of strategies: + 1. Shard the input (and output) on any dim, keep boundaries replicated. + 2. Shard boundaries on dim 0, replicate input, output is Partial("sum"). + Each rank counts how many of its local boundary values each input + element exceeds; summing across ranks gives the correct global index. + 3. Partial("max") or Partial("min") input with replicated boundaries. + Bucketize is monotonically non-decreasing in its input, so reducing + local bucket indices with max (or min) across ranks gives the same + result as bucketizing the reduced input values. + """ + input_meta, _boundaries_meta = args_schema + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + strategies.append( + [_ShardingPlaceholder(dim), _ShardingPlaceholder(dim), Replicate()] + ) + strategies.append([Partial("sum"), Replicate(), _ShardingPlaceholder(0)]) + for reduce_op in ("max", "min"): + strategies.append([Partial(reduce_op), Partial(reduce_op), Replicate()]) + return strategies + + +@register_single_dim_strategy( + aten.select.int, + schema_info=RuntimeSchemaInfo(1), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def select_int_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Forward all shardings except the selected dimension. + + select removes selected_dim. A shard before that dim keeps its output dim; + a shard after it shifts left by one. If selected_dim is sharded, the + strategy search must pick a replicated/resharded fallback. + """ + input_meta = cast(TensorMeta, args_schema[0]) + selected_dim = normalize_dim(cast(int, args_schema[1]), len(input_meta.shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + if dim == selected_dim: + continue + output_dim = dim if dim < selected_dim else dim - 1 + strategies.append([_ShardingPlaceholder(output_dim), _ShardingPlaceholder(dim)]) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +@register_single_dim_strategy( + aten.select_backward.default, + schema_info=RuntimeSchemaInfo(1), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def select_backward_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Inverse of select: insert the selected dimension into the output layout.""" + grad_meta = cast(TensorMeta, args_schema[0]) + input_sizes = cast(Sequence[object], args_schema[1]) + dim = normalize_dim(cast(int, args_schema[2]), len(input_sizes)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for grad_dim in range(len(grad_meta.shape)): + output_dim = grad_dim if grad_dim < dim else grad_dim + 1 + strategies.append( + [_ShardingPlaceholder(output_dim), _ShardingPlaceholder(grad_dim)] + ) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +@register_single_dim_strategy( + aten.slice.Tensor, + schema_info=RuntimeSchemaInfo(1), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def slice_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Forward shardings except on the sliced dimension. + + A nontrivial slice changes offsets along slice_dim, so a local shard on + that dimension cannot be forwarded without extra global index handling. A + statically full slice is just a view and can keep that sharding. + """ + defaults = (None, 0, None, None, 1) + input_meta, dim, start, end, step = args_schema + defaults[len(args_schema) :] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + input_shape = input_meta.shape + + if start is None: + start = 0 + if not isinstance(start, IntLike): + raise AssertionError(f"Expected IntLike, got {type(start)}") + if end is not None and not isinstance(end, IntLike): + raise AssertionError(f"Expected IntLike, got {type(end)}") + if not isinstance(step, IntLike): + raise AssertionError(f"Expected IntLike, got {type(step)}") + if end is None or statically_known_true(end > input_shape[dim]): + end = input_shape[dim] + + slice_dim = normalize_dim(dim, len(input_shape)) + start = start + input_shape[dim] if statically_known_true(start < 0) else start + end = end + input_shape[dim] if statically_known_true(end < 0) else end + + statically_redundant_slice = ( + statically_known_true(start == 0) + and statically_known_true(end == input_shape[dim]) + and statically_known_true(step == 1) + ) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_shape)): + if dim != slice_dim or statically_redundant_slice: + strategies.append([_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)]) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +@register_single_dim_strategy( + aten.slice_backward.default, + schema_info=RuntimeSchemaInfo(1), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def slice_backward_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Forward grad shardings except along the original sliced dimension.""" + grad_output_meta = args_schema[0] + input_sizes = args_schema[1] + dim = args_schema[2] + if not isinstance(grad_output_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(grad_output_meta)}") + if not isinstance(input_sizes, Sequence): + raise AssertionError(f"Expected Sequence, got {type(input_sizes)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + slice_dim = normalize_dim(dim, len(input_sizes)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(grad_output_meta.shape)): + if dim != slice_dim: + strategies.append([_ShardingPlaceholder(dim), _ShardingPlaceholder(dim)]) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op)]) + return strategies + + +@register_single_dim_strategy( + aten.slice_scatter.default, + schema_info=RuntimeSchemaInfo(2), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def slice_scatter_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Let input, src, and output follow each other off the scatter dimension. + + slice_scatter replaces a slice of self with src. Non-slice dimensions have + matching extents and can be sharded locally; the slice dimension stays + replicated via fallback because local slice coordinates would otherwise + depend on global offsets. + """ + defaults = (None, None, 0) + input_meta, src_meta, dim = args_schema[:3] + defaults[len(args_schema[:3]) :] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(src_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(src_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + slice_dim = normalize_dim(dim, len(input_meta.shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for dim in range(len(input_meta.shape)): + if dim != slice_dim: + strategies.append( + [ + _ShardingPlaceholder(dim), + _ShardingPlaceholder(dim), + _ShardingPlaceholder(dim), + ] + ) + # Replacement is pointwise by position, so matching Partial placements pass + # through for the supported reductions. + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append( + [ + Partial(reduce_op), + Partial(reduce_op), + Partial(reduce_op), + ] + ) + return strategies + + +@register_single_dim_strategy( + [aten.select_scatter.default], + schema_info=RuntimeSchemaInfo(1), +) +def select_scatter_single_dim_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + dim = normalize_dim(cast(int, args_schema[2]), ndim) + # [output, self, src] — src has the select dim removed + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d == dim: + continue + strategies.append( + [ + _ShardingPlaceholder(d), + _ShardingPlaceholder(d), + _ShardingPlaceholder(d if d < dim else d - 1), + ] + ) + return strategies + + +@register_single_dim_strategy( + [aten.diagonal_scatter.default], + schema_info=RuntimeSchemaInfo(1), +) +def diagonal_scatter_single_dim_strategy( + op: OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + # schema: (self, src, offset=0, dim1=0, dim2=1) + dim1 = cast(int, args_schema[3]) if len(args_schema) > 3 else 0 + dim2 = cast(int, args_schema[4]) if len(args_schema) > 4 else 1 + dim1 = normalize_dim(dim1, ndim) + dim2 = normalize_dim(dim2, ndim) + min_d, max_d = min(dim1, dim2), max(dim1, dim2) + # [output, self, src] — src has dim1/dim2 removed and diagonal appended + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d in (dim1, dim2): + continue + removed = (1 if d > min_d else 0) + (1 if d > max_d else 0) + strategies.append( + [ + _ShardingPlaceholder(d), + _ShardingPlaceholder(d), + _ShardingPlaceholder(d - removed), + ] + ) + return strategies + + +@register_single_dim_strategy(aten._local_scalar_dense.default) +def local_scalar_dense_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Only allow replication on the input; the output is a local scalar.""" + return [] + + +@register_single_dim_strategy( + [ + aten.scatter_.value, + aten.scatter.value, + aten.scatter_.src, + aten.scatter.src, + ], + schema_info=RuntimeSchemaInfo(1), +) +def scatter_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Sharding strategy for scatter/scatter_ value and src overloads. + + scatter writes along the dim argument, so that tensor axis must stay + replicated. Any other axis can be sharded only if every tensor operand has + the same size on it; then each rank scatters into its local slice with no + communication. + + Rows are [output, input, index] for scalar value overloads and + [output, input, index, src] when src is a tensor. The all-Replicate + fallback is inserted by the single-dim strategy infra. + """ + input_meta, dim, index_meta = args_schema[:3] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(index_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(index_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + + input_shape = input_meta.shape + index_shape = index_meta.shape + scatter_dim = normalize_dim(dim, len(input_shape)) + + # The scalar value overloads carry no src tensor. scatter.src can also + # receive a Python number, so only TensorMeta creates a src placement slot. + src_meta = args_schema[3] if len(args_schema) > 3 else None + has_tensor_src = isinstance(src_meta, TensorMeta) + src_shape = src_meta.shape if has_tensor_src else None + num_specs = 4 if has_tensor_src else 3 + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + if len(input_shape) == len(index_shape): + for d in range(len(input_shape)): + # Shard a non-scatter axis only when input/index/src shards align. + if d == scatter_dim or input_shape[d] != index_shape[d]: + continue + if src_shape is not None and ( + len(src_shape) != len(index_shape) or src_shape[d] != index_shape[d] + ): + continue + placement = _ShardingPlaceholder(d) + strategies.append([placement] * num_specs) + return strategies + + +@register_single_dim_strategy( + aten.scatter_add.default, schema_info=RuntimeSchemaInfo(1) +) +def scatter_add_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Shard scatter_add only on non-scatter axes with aligned sizes. + + Placement rows are [output, input, index, src]. The scatter axis stays + replicated because indices address the full extent of that dimension. + """ + input_meta, dim, index_meta = args_schema[:3] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(index_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(index_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + input_shape = input_meta.shape + index_shape = index_meta.shape + scatter_dim = normalize_dim(dim, len(input_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + if len(input_shape) == len(index_shape): + for d in range(len(input_shape)): + if d != scatter_dim and input_shape[d] == index_shape[d]: + placement = _ShardingPlaceholder(d) + strategies.append([placement, placement, placement, placement]) + return strategies + + +@register_single_dim_strategy(aten.gather.default, schema_info=RuntimeSchemaInfo(1)) +def gather_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Sharding strategy for gather with rows [output, input, index].""" + input_meta, dim, index_meta = args_schema[:3] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + if not isinstance(index_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(index_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + input_shape = input_meta.shape + index_shape = index_meta.shape + gather_dim = normalize_dim(dim, len(input_shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + # Input sharded on gather_dim: only valid when index has size 1 on that + # dim. Index/output use a mask partial so reducing combines the local + # gather results into the requested global value. + if ( + len(input_shape) > 0 + and gather_dim < len(index_shape) + and index_shape[gather_dim] == 1 + ): + mask_partial = _MaskPartial(offset_shape=input_shape, offset_dim=gather_dim) + strategies.append([mask_partial, Shard(gather_dim), mask_partial]) + + # Index sharded on gather_dim: input must be replicated because each rank + # may request values from the full input; output follows index. + if len(input_shape) > 0 and gather_dim < len(index_shape): + strategies.append([Shard(gather_dim), Replicate(), Shard(gather_dim)]) + + # Any non-gather dimension can be sharded across all tensors. + if len(input_shape) == len(index_shape): + for d in range(len(input_shape)): + if d != gather_dim: + placement = _ShardingPlaceholder(d) + strategies.append([placement, placement, placement]) + + return strategies + + +def _derive_follow_placements_from_tuple_strategy( + op: torch._ops.OpOverload, + tuple_strategy: TupleStrategy, +) -> Sequence[Placement]: + """ + derive the placements to follow from the tuple strategy, mainly used by + aten.stack, where each operand have the same shape, and correspondingly + expecting the same sharding + """ + + def merge_placement( + cur_placement: Placement, new_placement: Placement + ) -> Placement: + # semantic if we already have a follow placement, we + # check each placement for the current arg placement + # to see if we want to merge/adjust the placement to follow + # the priority: Partial -> Shard -> Replicate + # _StridedShard.__eq__ compares both dim and split_factor, + # so two _StridedShard with different split_factor won't match here. + if cur_placement == new_placement: + return cur_placement + + if cur_placement.is_partial(): + if _is_shard_like(new_placement): + # follow new placement + return new_placement + elif new_placement.is_partial(): + # different partial types, we can't merge and have to replicate all here + return Replicate() + else: + # follow partial + return cur_placement + elif _is_shard_like(cur_placement): + if _is_shard_like(new_placement): + # cur/new placement are different sharding (i.e. different shard dim) + # currently fallback to replicate all args + return Replicate() + else: + # for partial/replicate, follow the current shard placement + return cur_placement + else: + # current replicate, just follow new placement + return new_placement + + follow_placements: list[Placement] | None = None + mesh = tuple_strategy.child_mesh(0) + for arg_strategy in tuple_strategy.children: + if not isinstance(arg_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(arg_strategy)}") + if arg_strategy.mesh != mesh: + raise ValueError( + f"All operands in {op} must have the same mesh, " + f"but got {arg_strategy.mesh} and {mesh}." + ) + + for placement_strategy in arg_strategy.strategies: + arg_placements = placement_strategy.output_spec.placements + if follow_placements is None: + follow_placements = list(arg_placements) + continue + if follow_placements is None: + raise AssertionError( + "follow_placements should not be None at this point" + ) + for mesh_idx in range(mesh.ndim): + # merge placements with the priority + follow_placements[mesh_idx] = merge_placement( + follow_placements[mesh_idx], arg_placements[mesh_idx] + ) + if follow_placements is None: + raise AssertionError("follow placements should not be None!") + return follow_placements + + +@register_op_strategy(aten.stack.default, RuntimeSchemaInfo(1, needs_pytree=True)) +def stack_strategy(op_schema: OpSchema) -> StrategyType: + # We keep stack as an OpStrategy because .stack is used w/ NormPartial, and its + # impossible to include every NormPartial rules as there are infinite possibilities + # with ord. Future changes to infra could be made, but we prefer to deprecate + # NormPartial at some point. + + args_schema = op_schema.args_schema + input_tuple_strategy = args_schema[0] + if not isinstance(input_tuple_strategy, TupleStrategy): + raise AssertionError(f"Expected TupleStrategy, got {input_tuple_strategy}") + input_strategies: list[OpStrategy] = [] + for child in input_tuple_strategy.children: + if not isinstance(child, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {child}") + input_strategies.append(child) + first_input_strategy = input_strategies[0] + common_input_ndim = first_input_strategy.ndim + dim = cast(int, args_schema[1]) if len(args_schema) > 1 else 0 + # normalize the dim to be within the output ndim (input ndim + 1), + # since stack inserts a new dimension + dim = normalize_dim(dim, common_input_ndim + 1) + + mesh = first_input_strategy.mesh + + follow_placements = _derive_follow_placements_from_tuple_strategy( + op_schema.op, input_tuple_strategy + ) + + # create op strategy base on the follow placements + op_strategy = OpStrategy([]) + + input_specs = tuple( + DTensorSpec(mesh, tuple(follow_placements)) + for _ in range(len(input_tuple_strategy.children)) + ) + + # stack op would "insert" new dim, so all sharded dim >= the inserted dim need to + # be normalized with the new Shard placement + follow_placements = shift_shard_dims_after_insert(follow_placements, dim) + output_spec = DTensorSpec(mesh, tuple(follow_placements)) + redistribute_cost = [ + generate_redistribute_costs(input_strategies[i], input_specs[i]) + for i in range(len(input_specs)) + ] + op_strategy.strategies.append( + OpSpec( + output_specs=output_spec, + input_specs=input_specs, + redistribute_cost=redistribute_cost, + ) + ) + return op_strategy + + +@register_single_dim_strategy( + aten.cat.default, + RuntimeSchemaInfo(1, needs_pytree=True), + allow_unbacked_sharding=False, +) +def cat_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Forward cat input placements except along the concatenation dimension. + + Concatenation changes only cat_dim; sharding there needs redistribution, so + this rule offers non-cat dims and leaves the replicate fallback to the + single-dim infra. Legacy 1-D empty tensors have no real cat layout, so they + stay Replicate while non-empty tensors follow the output placement. + """ + input_list = args_schema[0] + # Unfortunate naming, but yes it's a TensorList input, and we represent it + # as a tuple of TensorMeta. + if not isinstance(input_list, (tuple, list)): + raise AssertionError(type(input_list)) + if not all(isinstance(tm, TensorMeta) for tm in input_list): + raise AssertionError + + if isinstance(input_list, list): + input_list = tuple(input_list) + + def is_legacy_cat_empty(meta: TensorMeta) -> bool: + return len(meta.shape) == 1 and statically_known_true(meta.shape[0] == 0) + + num_inputs = len(input_list) + if num_inputs == 0: + raise AssertionError("Expected cat inputs to be non-empty list of tensors") + legacy_empty_inputs = tuple(is_legacy_cat_empty(meta) for meta in input_list) + real_ndim_set = { + len(meta.shape) + for meta, legacy_empty in zip(input_list, legacy_empty_inputs, strict=True) + if not legacy_empty + } + if len(real_ndim_set) > 1: + raise AssertionError("Expected all non-empty cat inputs to be the same ndim") + common_ndim = next(iter(real_ndim_set), 1) + cat_dim = cast(int, args_schema[1]) if len(args_schema) > 1 else 0 + cat_dim = normalize_dim(cat_dim, common_ndim) + single_dim_strategies = [] + for i in range(common_ndim): + if i != cat_dim: + single_dim_strategies.append( + [_ShardingPlaceholder(i)] + + [ + Replicate() if legacy_empty else _ShardingPlaceholder(i) + for legacy_empty in legacy_empty_inputs + ] + ) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + single_dim_strategies.append( + [Partial(reduce_op)] # pyrefly: ignore [bad-argument-type] + + [ + Replicate() if legacy_empty else Partial(reduce_op) + for legacy_empty in legacy_empty_inputs + ] + ) + # pyrefly: ignore [bad-return] + return single_dim_strategies + + +@register_single_dim_strategy( + aten.index_select.default, schema_info=RuntimeSchemaInfo(1) +) +def index_select_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + values_meta, dim, index_meta = args_schema + if not isinstance(values_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(values_meta)}") + if not isinstance(dim, int): + raise AssertionError(f"Expected int, got {type(dim)}") + dim = normalize_dim(dim, len(values_meta.shape)) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + + # Shard values on any non-indexed dim (output has same ndim) + for d in range(len(values_meta.shape)): + if d == dim: + continue + strategies.append( + [_ShardingPlaceholder(d), _ShardingPlaceholder(d), Replicate()] + ) + + # Shard index → output sharded on the indexed dim + strategies.append([_ShardingPlaceholder(dim), Replicate(), _ShardingPlaceholder(0)]) + + # Partial passthrough from values + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op), Partial(reduce_op), Replicate()]) + + return strategies + + +@register_single_dim_strategy( + aten.index.Tensor, schema_info=RuntimeSchemaInfo(needs_pytree=True) +) +def index_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + values_meta, multi_indices_meta = args_schema + if not isinstance(values_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(values_meta)}") + if not isinstance(multi_indices_meta, (list, tuple)): + raise AssertionError(f"Expected list or tuple, got {type(multi_indices_meta)}") + + indexed_dims = [i for i, idx in enumerate(multi_indices_meta) if idx is not None] + non_indexed_dims = [ + i for i in range(len(values_meta.shape)) if i not in set(indexed_dims) + ] + + index_metas = [idx for idx in multi_indices_meta if idx is not None] + if not all(isinstance(m, TensorMeta) for m in index_metas): + raise AssertionError("Expected all index metas to be TensorMeta") + broadcast_ndim = max(len(m.shape) for m in index_metas) + num_indices = len(indexed_dims) + + # Determine where index output dims are inserted in the result + all_consecutive = all( + indexed_dims[i + 1] - indexed_dims[i] == 1 for i in range(len(indexed_dims) - 1) + ) + insert_dim = indexed_dims[0] if all_consecutive else 0 + + def values_dim_to_output_dim(d: int) -> int: + if d < insert_dim: + return d + return d + broadcast_ndim - sum(1 for idx_dim in indexed_dims if d > idx_dim) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + + # Shard values on a non-indexed dim, all indices replicated + for d in non_indexed_dims: + out_dim = values_dim_to_output_dim(d) + rule: list[Placement | _ShardingPlaceholder] = [_ShardingPlaceholder(out_dim)] + rule.append(_ShardingPlaceholder(d)) + rule.extend([Replicate()] * num_indices) + strategies.append(rule) + + # Shard indices on the same broadcast dim. Each index tensor may + # have a different ndim, so we map broadcast dim → tensor dim via + # left-padding. Tensors with size 1 on that dim are replicated + # (broadcast semantics). + for bd in range(broadcast_ndim): + per_tensor: list[tuple[int, int]] = [] # (tensor_dim, size) + for m in index_metas: + offset = broadcast_ndim - len(m.shape) + if bd < offset: + per_tensor.append((-1, 1)) # implicit broadcast + else: + td = bd - offset + per_tensor.append((td, m.shape[td])) + if all(s == 1 for _, s in per_tensor): + continue # all broadcast-only, skip + out_dim = bd + insert_dim + rule: list[Placement | _ShardingPlaceholder] = [_ShardingPlaceholder(out_dim)] + rule.append(Replicate()) + for td, s in per_tensor: + if s > 1: + rule.append(_ShardingPlaceholder(td)) + else: + rule.append(Replicate()) + strategies.append(rule) + + # Partial passthrough from values + for reduce_op in Partial.LINEAR_REDUCE_OPS: + rule: list[Placement | _ShardingPlaceholder] = [ + Partial(reduce_op), + Partial(reduce_op), + ] + rule.extend([Replicate()] * num_indices) + strategies.append(rule) + + return strategies + + +@register_single_dim_strategy( + [aten.index_put.default, aten.index_put_.default, aten._index_put_impl_.default], + schema_info=RuntimeSchemaInfo(needs_pytree=True), +) +def index_put_single_dim_strategy( + op: OpOverload, args: ArgsType, kwargs: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Single-dim sharding strategy for index_put(self, indices, values). + + Strategy format: [output, input, *indices, value] + + How index_put works: + + indices is a tuple of index tensors and Nones: + - an index tensor at entry i means self is indexed on dim i. + - a None at entry i means all elements along dim i are selected (like :). + - any trailing dims (if self.ndim > len(indices)) are also not indexed + (i.e. implicit trailing Nones). + + All non-None index tensors are broadcast together to produce a + broadcasted indexing shape. Each position in this broadcasted shape + serves as an indexing coordinate into self. Each coordinate selects a + tensor element, or a slice (if non-indexed dims exist). + + values is a tensor broadcastable to the indexing output shape. + When indexed dims are consecutive starting at dim k, this shape is + (*self[:k], *broadcast_shape, *self[k+n_indexed:]). When indexed + dims are non-consecutive, it is (*broadcast_shape, *non_indexed_dims). + + Sharding rules (possibly conservative and incomplete): + - Index tensors: always Replicate (every rank needs all coordinates). + - Self cannot be sharded on indexed dims (local position != global position). + - Self and values CAN be sharded on non-indexed dims. + The exception is broadcasted value dimensions (size 1) - we require Replicate, but can shard self. + - Additionally, we allow the full Partial rule on non-indexing tensors. + + """ + self_meta = cast(TensorMeta, args[0]) + indices_meta = cast(tuple[TensorMeta | None, ...], args[1]) + values_meta = cast(TensorMeta, args[2]) + + # Determine indexed vs non-indexed dims of self. + indexed_dims = {i for i, idx in enumerate(indices_meta) if idx is not None} + non_indexed_dims = [d for d in range(len(self_meta.shape)) if d not in indexed_dims] + n_indexed = len(indexed_dims) + values_ndim = len(values_meta.shape) + + # Explicitly compute the broadcast shape of the index tensors. + index_shapes = [idx.shape for idx in indices_meta if idx is not None] + broadcast_ndim = len(torch.broadcast_shapes(*index_shapes)) if index_shapes else 0 + + # Strategy format: [output, input, *indices, value] + # The infra flattens the indices list and drops None entries, so only + # non-None index tensors get a placement slot (all Replicate). + # + # Values dim mapping depends on whether indexed dims are contiguous: + # Contiguous (e.g., (None, idx0, idx1)): broadcast replaces indexed block in-place. + # values shape = (*non_indexed_before, *broadcast_shape, *non_indexed_after) + # Non-contiguous (e.g., (idx0, None, idx1)): broadcast goes to front. + # values shape = (*broadcast_shape, *non_indexed_dim_sizes) + indexed_dims_sorted = sorted(indexed_dims) + contiguous_indexed = len(indexed_dims_sorted) <= 1 or ( + indexed_dims_sorted[-1] - indexed_dims_sorted[0] + 1 == len(indexed_dims_sorted) + ) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for i, self_dim in enumerate(non_indexed_dims): + if contiguous_indexed and indexed_dims_sorted: + # Broadcast replaces the indexed block in-place. + first_indexed = indexed_dims_sorted[0] + if self_dim < first_indexed: + values_dim = self_dim + else: + values_dim = self_dim - n_indexed + broadcast_ndim + else: + # Broadcast goes to front (non-contiguous or no indexed dims). + values_dim = broadcast_ndim + i + + # values_dim is the position in the result tensor, but values may + # have fewer dims (right-aligned broadcasting). Convert to the + # actual values tensor dimension. + result_ndim = broadcast_ndim + len(non_indexed_dims) + values_tensor_dim = values_dim - (result_ndim - values_ndim) + + if values_tensor_dim < 0: + values_placement: Placement | _ShardingPlaceholder = Replicate() + elif values_meta.shape[values_tensor_dim] == 1: + values_placement = Replicate() + else: + values_placement = _ShardingPlaceholder(values_tensor_dim) + + strategies.append( + [ + _ShardingPlaceholder(self_dim), + _ShardingPlaceholder(self_dim), + *([Replicate()] * n_indexed), + values_placement, + ] + ) + + # full-partial rule on non-indexing tensors + strategies.append( + [ + Partial(), + Partial(), + *([Replicate()] * n_indexed), + Partial(), + ] + ) + return strategies + + +def _index_dim_strategy( + args_schema: ArgsType, + shard_row: Callable[[int], list[Placement | _ShardingPlaceholder]], + partial_rules: list[list[Placement | _ShardingPlaceholder]] | None = None, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Common strategy for index ops that shard on all dims except the indexed dim. + + Args: + shard_row: given a dim d, returns the strategy row for sharding on that dim. + partial_rules: additional Partial passthrough strategies. + """ + self_meta = cast(TensorMeta, args_schema[0]) + ndim = len(self_meta.shape) + dim = normalize_dim(cast(int, args_schema[1]), ndim) + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d != dim: + strategies.append(shard_row(d)) + if partial_rules: + strategies.extend(partial_rules) + return strategies + + +@register_single_dim_strategy( + [aten.index_fill.int_Scalar, aten.index_fill_.int_Scalar], + schema_info=RuntimeSchemaInfo(1), +) +def index_fill_scalar_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # index_fill(self, dim, index, value) — fills self[..., index, ...] with scalar value. + # Partial rules: each rank fills with the same scalar v, then reduces. + # Only idempotent reduces work: avg(v,v,...,v)=v, max(v,v,...,v)=v, min(v,v,...,v)=v. + # sum and product fail: sum(v,v,...,v)=nv, product(v,v,...,v)=v^n. + return _index_dim_strategy( + args_schema, + lambda d: [ + _ShardingPlaceholder(d), # result + _ShardingPlaceholder(d), # self + Replicate(), # value (scalar, same on all ranks) + ], + [[Partial(op), Partial(op), Replicate()] for op in ("avg", "max", "min")], + ) + + +@register_single_dim_strategy( + [aten.index_fill.int_Tensor, aten.index_fill_.int_Tensor], + schema_info=RuntimeSchemaInfo(1), +) +def index_fill_tensor_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # index_fill(self, dim, index, value) — fills self[..., index, ...] with 0-d tensor value. + # Partial rules: each rank fills with its partial value v_i, then reduces. + # All reduce ops work because reduce(v_0, ..., v_{n-1}) = V (the global value) + # regardless of op, since fill is a pure replacement (no mixing with self). + return _index_dim_strategy( + args_schema, + lambda d: [ + _ShardingPlaceholder(d), # result + _ShardingPlaceholder(d), # self + Replicate(), # index + Replicate(), # value + ], + [ + [Partial(op), Partial(op), Replicate(), Partial(op)] + for op in _PARTIAL_PASS_THROUGH_REDUCE_OPS + ], + ) + + +@register_single_dim_strategy( + [aten.index_reduce.default, aten.index_reduce_.default], + schema_info=RuntimeSchemaInfo(1), +) +def index_reduce_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + # index_reduce(self, dim, index, source, reduce) — reduces source into self at index positions. + # No partial rules: reduce ops are "mean"/"amax"/"amin"/"prod", which don't match + # any Partial reduce op names ("avg"/"max"/"min"/"product"/"sum"). + return _index_dim_strategy( + args_schema, + lambda d: [ + _ShardingPlaceholder(d), # result + _ShardingPlaceholder(d), # self + Replicate(), # index + _ShardingPlaceholder(d), # source + ], + ) + + +@register_single_dim_strategy( + [ + aten.split.Tensor, + aten.split_with_sizes.default, + aten.split_with_sizes_copy.default, + ], + RuntimeSchemaInfo(1), + allow_unbacked_sharding=False, +) +def split_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Forward all shardings except the split dimension. + + Splitting a sharded split_dim would require each output chunk to know + global chunk boundaries, so the single-dim rule excludes that dim and + relies on the redistribute-to-replicate fallback. + """ + input_meta = args_schema[0] + split_size_or_sections = args_schema[1] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + input_ndim = len(input_meta.shape) + split_dim = cast(int, args_schema[2]) if len(args_schema) > 2 else 0 + dim = normalize_dim(split_dim, input_ndim) + + def size_split(N, i) -> list: + # Last chunk will be smaller if the tensor size N + # along the given dimension dim is not divisible by i. + if not i > 0: + raise AssertionError(f"Split size must be positive, got {i}") + return [i] * (N // i) + ([N % i] if N % i != 0 else []) + + output_size_list = ( + size_split(input_meta.shape[dim], split_size_or_sections) + if isinstance(split_size_or_sections, IntLike) + else split_size_or_sections + ) + if not isinstance(output_size_list, Sized): + raise AssertionError(f"Expected Sized, got {type(output_size_list)}") + num_outputs = len(output_size_list) + + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(input_ndim): + if d != dim: + placement = _ShardingPlaceholder(d) + # pyrefly: ignore [bad-argument-type] + strategies.append([placement] * num_outputs + [placement]) + for reduce_op in _PARTIAL_PASS_THROUGH_REDUCE_OPS: + strategies.append([Partial(reduce_op)] * (num_outputs + 1)) + return strategies + + +# TODO: fix remaining failures in xfail("unbind") in test_dtensor_ops.py +# and remove this xfail item +@register_op_strategy(aten.unbind.int, schema_info=RuntimeSchemaInfo(1)) +def gen_unbind_strategy(op_schema: OpSchema) -> StrategyType: + """Forward all shardings except the unbind dimension.""" + input_strategy = op_schema.args_schema[0] + if not isinstance(input_strategy, OpStrategy): + raise AssertionError(f"Expected OpStrategy, got {type(input_strategy)}") + input_ndim = input_strategy.ndim + input_shape = input_strategy.shape + unbind_dim = ( + cast(int, op_schema.args_schema[1]) if len(op_schema.args_schema) > 1 else 0 + ) + unbind_dim = normalize_dim(unbind_dim, input_ndim) + + mesh = input_strategy.mesh + unbind_strategy = OpStrategy([]) + for arg_strategy in input_strategy.strategies: + arg_spec = arg_strategy.output_spec + if is_tensor_dim_sharded(arg_spec, dim=unbind_dim): + raise RuntimeError( + f"Attempted to unbind along the sharded dimension {unbind_dim}. ", + "It cannot be performed without redistribution, which is disallowed " + "by the current operator.", + ) + # Only add the strategy if the unbind dim is not sharded. + output_placements = shift_shard_dims_after_remove( + arg_spec.placements, unbind_dim + ) + output_specs = tuple( + DTensorSpec(mesh, tuple(output_placements)) + for _ in range(input_shape[unbind_dim]) + ) + unbind_strategy.strategies.append( + OpSpec( + output_specs=output_specs, + input_specs=(arg_spec,), + redistribute_cost=[[0.0] * len(input_strategy.strategies)], + ) + ) + return unbind_strategy + + +@register_single_dim_strategy( + aten.eye.m_out, + schema_info=RuntimeSchemaInfo(static_kwargkey=["out"]), + allow_unbacked_sharding=True, + allow_uneven_sharding=True, +) +def eye_out_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + """Strategy for torch.eye(..., out=...). + + Sharded eye needs global index offsets to place the diagonal correctly. + The local out= kernel does not have that context, so only the implicit + all-Replicate strategy is valid. + """ + out_meta = kwargs_schema["out"] + if not isinstance(out_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta for out, got {type(out_meta)}") + + return [] + + +def _pass_through_partials( + num_inputs: int = 1, +) -> list[list[Placement | _ShardingPlaceholder]]: + """Pass-through strategies for all supported reduce ops.""" + return [[Partial(op)] * (1 + num_inputs) for op in ("sum", "avg", "max", "min")] + + +def _shard_inactive_dims( + ndim: int, active_dims: set[int], num_inputs: int = 1 +) -> list[list[Placement | _ShardingPlaceholder]]: + """Single-dim strategies: shard on dims the op doesn't touch.""" + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim): + if d not in active_dims: + strategies.append([_ShardingPlaceholder(d)] * (1 + num_inputs)) + return strategies + + +@register_single_dim_strategy(aten.roll.default, schema_info=RuntimeSchemaInfo(1)) +def roll_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + raw_dims = cast(list[int], args_schema[2]) if len(args_schema) > 2 else [] + # When dims is empty, roll flattens the tensor — all dims are active + if not raw_dims: + raw_dims = list(range(ndim)) + active_dims = {normalize_dim(d, ndim) for d in raw_dims} + return _shard_inactive_dims(ndim, active_dims) + _pass_through_partials() + + +@register_single_dim_strategy(aten.flip.default, schema_info=RuntimeSchemaInfo(1)) +def flip_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + raw_dims = cast(list[int], args_schema[1]) + active_dims = {normalize_dim(d, ndim) for d in raw_dims} + return _shard_inactive_dims(ndim, active_dims) + _pass_through_partials() + + +@register_single_dim_strategy( + [aten._fft_c2c.default, aten._fft_r2c.default, aten._fft_c2r.default], + schema_info=RuntimeSchemaInfo(1), +) +def fft_single_dim_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType +) -> list[list[Placement | _ShardingPlaceholder]]: + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + raw_dims = cast(list[int], args_schema[1]) + active_dims = {normalize_dim(d, ndim) for d in raw_dims} + return _shard_inactive_dims(ndim, active_dims) + _pass_through_partials() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_view_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_view_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..a64ab67f4585b2af503e304e5754a80672974088 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/_view_ops.py @@ -0,0 +1,1518 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import math +from collections.abc import Callable, Iterable, Sequence +from dataclasses import dataclass +from typing import Any, cast, NamedTuple + +import torch +from torch import Tensor +from torch._prims_common import DimsType +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpSchema, + OpSpec, + OpStrategy, + RuntimeSchemaInfo, + StrategyType, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor._ops.utils import ( + generate_redistribute_costs, + normalize_dim, + normalize_dims, + prod, + register_op_strategy, +) +from torch.distributed.tensor.placement_types import ( + _defer_symbolic_equality_assert, + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) + + +aten = torch.ops.aten + +Shape = tuple[int, ...] + + +def _explicit_unbacked_hint(value: object) -> int | None: + from torch.fx.experimental.symbolic_shapes import ( + free_unbacked_symbols, + optimization_hint, + ) + + if not isinstance(value, torch.SymInt): + return None + shape_env = getattr(value.node, "shape_env", None) + if shape_env is None: + return None + unbacked_symbols = free_unbacked_symbols(value.node.expr) + if not unbacked_symbols or not unbacked_symbols.issubset( + shape_env.var_to_hint_override.keys() + ): + return None + return int(optimization_hint(value, fallback=None)) + + +def _split_factor_hint(value: object) -> int | None: + if isinstance(value, int): + return int(value) + return _explicit_unbacked_hint(value) + + +def _split_factor_matches(expected: object, actual: object) -> bool: + from torch.fx.experimental.symbolic_shapes import guard_or_false + + if expected is None: + return False + if guard_or_false(expected == actual): + return True + expected_hint = _split_factor_hint(expected) + actual_hint = _split_factor_hint(actual) + if expected_hint is None or actual_hint is None or expected_hint != actual_hint: + return False + if isinstance(expected, torch.SymInt): + _defer_symbolic_equality_assert( + expected, actual_hint, reason="_StridedShard split_factor match" + ) + elif isinstance(actual, torch.SymInt): + _defer_symbolic_equality_assert( + actual, expected_hint, reason="_StridedShard split_factor match" + ) + return True + + +class ClaimedDim(NamedTuple): + """An (input_dim, output_dim) pair claimed by a mesh dim's _StridedShard rewrite.""" + + input_dim: int + output_dim: int + + +@dataclass +class DimSpec: + """Specifies how an output dimension maps to an input dimension.""" + + def inputs(self) -> Iterable["DimSpec"]: + return () + + +# Rules that map each dimension of the output to dimensions of the input tensor +DimMap = tuple[DimSpec, ...] + + +@dataclass +class Singleton(DimSpec): + """Output dimension is a singleton.""" + + +@dataclass(eq=False) +class InputDim(DimSpec): + """Output dimension maps directly to an input dimension.""" + + input_dim: int + + def __eq__(self, other: object) -> bool: + """Raises TypeError for non-DimSpec comparisons to catch accidental + ``shard.dim == input_dim`` bugs where ``.input_dim`` was intended.""" + if isinstance(other, InputDim): + return self.input_dim == other.input_dim + if not isinstance(other, DimSpec): + raise TypeError( + f"Cannot compare InputDim with {type(other).__name__}. " + f"Did you mean to use .input_dim?" + ) + return NotImplemented + + def __hash__(self) -> int: + return hash((InputDim, self.input_dim)) + + +@dataclass +class Broadcast(DimSpec): + """Output is the broadcast of a singleton input dimension.""" + + dim: DimSpec + dim_size: int + + @classmethod + def new(cls, dim: DimSpec, dim_size: int) -> DimSpec: + return Broadcast(dim, dim_size) + + def inputs(self) -> Iterable[DimSpec]: + return (self.dim,) + + +@dataclass +class NewDim(DimSpec): + """This is a new dimension created by the op.""" + + size: int + + @classmethod + def new(cls, size: int) -> DimSpec: + from torch.fx.experimental.symbolic_shapes import guard_or_false + + return Singleton() if guard_or_false(size == 1) else NewDim(size) + + +@dataclass +class Repeat(DimSpec): + """Output dimension is the input dimension repeated n-times.""" + + input_dim: DimSpec + times: int + + @classmethod + def new(cls, dim: DimSpec, times: int) -> DimSpec: + from torch.fx.experimental.symbolic_shapes import guard_or_false + + if guard_or_false(times == 1): + return dim + elif isinstance(dim, Singleton): + # repeating a singleton is the same as broadcasting it + return Broadcast(dim, times) + else: + return Repeat(dim, times) + + def inputs(self) -> Iterable[DimSpec]: + return (self.input_dim,) + + +@dataclass +class Flatten(DimSpec): + """Flatten a set of input dimensions, ensuring right-most adjacent elements remain adjacent in the output.""" + + input_dims: Sequence[DimSpec] + + @classmethod + def new(cls, dims: Sequence[DimSpec]) -> DimSpec: + if len(dims) == 0: + # flattening a scalar leads to a singleton + return Singleton() + elif len(dims) == 1: + # flattening a single dimension is no-op + return dims[0] + else: + return Flatten(dims) + + def inputs(self) -> Iterable[DimSpec]: + return self.input_dims + + +@dataclass +class Split(DimSpec): + """ + This dimension is a member of a decomposition of the input dim. + + Note that input_dim itself could be a Flattened set of input dims. + """ + + input_dim: DimSpec + group_shape: Shape + split_id: int + + @classmethod + def new(cls, dim: DimSpec, group_shape: tuple[int, ...], idx: int) -> DimSpec: + from torch.fx.experimental.symbolic_shapes import guard_or_false, guard_or_true + + if not len(group_shape) > 0: + raise AssertionError( + f"Expected group_shape length > 0, got {len(group_shape)}" + ) + if len(group_shape) == 1: + # not really a group, just return the input dim back + if not idx == 0: + raise AssertionError(f"Expected idx == 0, got {idx}") + return dim + elif guard_or_false(group_shape[idx] == 1): + return Singleton() + else: + # remove singletons from group + # group_mapping = [(new_index, (shape, old_index)) ...] + group_mapping = list( + enumerate( + (s, i) for i, s in enumerate(group_shape) if guard_or_true(s != 1) + ) + ) + new_group_shape = tuple(m[1][0] for m in group_mapping) + new_idx = next(filter(lambda x: x[1][1] == idx, group_mapping))[0] + return Split(dim, new_group_shape, new_idx) + + def inputs(self) -> Iterable[DimSpec]: + return (self.input_dim,) + + +def dim_pad_left(ndim: int, min_dims: int) -> DimMap: + return (Singleton(),) * max(0, min_dims - ndim) + tuple( + InputDim(i) for i in range(ndim) + ) + + +def dim_atleast_3d(ndim: int) -> DimMap: + if ndim == 0: + return (Singleton(), Singleton(), Singleton()) + elif ndim == 1: + return (Singleton(), InputDim(0), Singleton()) + elif ndim == 2: + return (InputDim(0), InputDim(1), Singleton()) + else: + return tuple(InputDim(i) for i in range(ndim)) + + +def expand(input_shape: Shape, shape: Shape) -> DimMap: + """Implement broadcast on multiple dimensions.""" + from torch.fx.experimental.symbolic_shapes import guard_or_false + + if not len(shape) >= len(input_shape): + raise AssertionError( + f"Expected len(shape) >= len(input_shape), got {len(shape)} < {len(input_shape)}" + ) + + # 1. create padded input dimensions + padded_input = dim_pad_left(len(input_shape), len(shape)) + # 2. check that input shapes are compatible + mapping = [] + for p, desired_s in zip(padded_input, shape): + if isinstance(p, Singleton): + actual_s = 1 + if not desired_s >= 0: + raise AssertionError(f"Expected desired_s >= 0, got {desired_s}") + else: + if not isinstance(p, InputDim): + raise AssertionError(f"DimSpec not supported in expand: {p}") + actual_s = input_shape[p.input_dim] + if not ( + guard_or_false(actual_s == 1) + or guard_or_false(desired_s == -1) + or guard_or_false(desired_s == actual_s) + ): + raise AssertionError( + f"Expected actual_s == 1 or desired_s == -1 or " + f"desired_s == actual_s, got actual_s={actual_s}, desired_s={desired_s}" + ) + mapping.append( + p + if ( + guard_or_false(desired_s == 1) + or guard_or_false(desired_s == -1) + or guard_or_false(desired_s == actual_s) + ) + else Broadcast.new(p, desired_s) + ) + return tuple(mapping) + + +def normalize_sizes(sizes: Shape | tuple[Shape]) -> Shape: + if isinstance(sizes[0], (int, torch.SymInt)): + return cast(Shape, sizes) + elif len(sizes) == 1: + return sizes[0] + else: + raise RuntimeError("Size must be int... or tuple") + + +def dim_flatten(ndim: int, start_dim=0, end_dim=-1) -> DimMap: + if ndim == 0: + return (Singleton(),) + elif ndim == 1: + return (InputDim(0),) + else: + # only flattening dims from start_dim to end_dim (inclusive) + # other dims are passed through + if end_dim < 0: + end_dim += ndim + results: list[DimSpec] = [InputDim(i) for i in range(start_dim)] + results.append( + Flatten.new(tuple(InputDim(i) for i in range(start_dim, end_dim + 1))) + ) + results.extend([InputDim(i) for i in range(end_dim + 1, ndim)]) + return tuple(results) + + +def dim_movedim( + ndim: int, + input: DimsType, + destination: DimsType, +) -> DimMap: + input = normalize_dims(input, ndim) + destination = normalize_dims(destination, ndim) + + if not len(input) == len(destination): + raise AssertionError( + f"Expected len(input) == len(destination), got {len(input)} != {len(destination)}" + ) + input_set = set(input) + if not len(input_set) == len(input): + raise AssertionError("Found repeated input dims") + if not len(set(destination)) == len(destination): + raise AssertionError("Found repeated output dims") + if not max(input) < ndim: + raise AssertionError(f"Expected max(input) < ndim, got {max(input)} >= {ndim}") + if not max(destination) < ndim: + raise AssertionError( + f"Expected max(destination) < ndim, got {max(destination)} >= {ndim}" + ) + + dest = [-1] * ndim + for i, d in zip(input, destination): + dest[d] = i + + unused_inputs_iter = iter(i for i in range(ndim) if i not in input_set) + for i in range(ndim): + if dest[i] == -1: + dest[i] = next(unused_inputs_iter) + + return tuple(InputDim(i) for i in dest) + + +def dim_repeat(ndim: int, sizes: Shape) -> DimMap: + sizes = normalize_sizes(sizes) + if not len(sizes) >= ndim: + raise AssertionError( + f"Number of dimensions of repeat dims {sizes} can not be smaller than number of dimensions of tensor {ndim}." + ) + pad = len(sizes) - ndim + return tuple(Repeat.new(Singleton(), s) for s in sizes[:pad]) + tuple( + Repeat.new(InputDim(i), s) for i, s in enumerate(sizes[pad:]) + ) + + +def infer_size(total_size: int, sizes: Shape) -> Shape: + """ + One dimension input to view may be "-1". + + Infer the size of this dimension given the total_size. + """ + from torch.fx.experimental.symbolic_shapes import guard_or_false + + infers = [i for i, s in enumerate(sizes) if guard_or_false(s == -1)] + size = prod(sizes) + if not len(infers) <= 1: + raise AssertionError("can only infer one size") + if infers: + size = -size + missing_size = total_size // size + torch._check( + total_size % size == 0, + lambda: f"size inferred for -1 is not integral {sizes} should have {total_size} elements.", + ) + return tuple(s if not guard_or_false(s == -1) else missing_size for s in sizes) + torch._check( + size == total_size, + lambda: f"sizes do not match {total_size} vs {size}", + ) + return sizes + + +def view_groups(from_size: Shape, to_size: Shape) -> DimMap: + """ + Decompose a reshape operation into forwarding, flattening, or splitting dimensions for each output dimension. + + A view or reshape operation can be decomposed into a set of 3 types of smaller operations: + 1) Forward a dimension from input to output + 2) Flatten a set of dimensions into a single dimension + 3) Split one dimension into multiple dimensions + + view_groups identifies these operations and returns, for each output dimension, what + operation was performed in the input dimension. For example: + + view_groups([2, 3, 4], [2, 12]) -> ( + InputDim(0), + Flatten((InputDim(1), InputDim(2))) + ) + + - output dimension 0 maps to input dimension 0 + - output dimension 1 maps to a flattened input dimensions 1 and 2 + + + view_groups([2, 3], [3, 2]) -> ( + Split(Flatten((InputDim(0), InputDim(1))), (3, 2), 0), + Split(Flatten((InputDim(0), InputDim(1))), (3, 2), 1), + ) + + - in the above, input is flattened into a single dimension and then split + into two separate dimensions with different sizes from the input. + """ + from torch.fx.experimental.symbolic_shapes import ( + free_symbols, + guard_or_false, + guard_or_true, + ) + + from_nelem = prod(from_size) + to_size = infer_size(from_nelem, normalize_sizes(to_size)) + + torch._check( + from_nelem == prod(to_size), + lambda: "Total view shape does not add up", + ) + + from_idx = 0 + to_idx = 0 + from_len = len(from_size) + to_len = len(to_size) + + result_pp: list[DimSpec] = [] + + while from_idx < from_len or to_idx < to_len: + from_group_dim, to_group_shape = [], [] + + if from_idx >= from_len: + f = 1 + else: + f = from_size[from_idx] + from_group_dim.append(from_idx) + from_idx += 1 + + if to_idx >= to_len: + t = 1 + else: + t = to_size[to_idx] + to_group_shape.append(t) + to_idx += 1 + + # if any of the groups is singleton, great, we need to backtrack though + if guard_or_false(f == 1) and guard_or_true(t != 1): + # produces ([1], []) + to_idx -= 1 + to_group_shape = [] + elif guard_or_true(f != 1) and guard_or_false(t == 1): + # produces ([], [1]) + from_idx -= 1 + from_group_dim = [] + else: + # produces ([1], [1]), ([2], [2]), ([2,3], [6]) + while not guard_or_false(f == t): + if from_idx < from_len and ( + to_idx >= to_len + or guard_or_false(t % f == 0) + or guard_or_false(t > f) + ): # for easier symbolic comparisons, e.g. u0*u1 > u0 + nf = from_size[from_idx] + from_group_dim.append(from_idx) + from_idx += 1 + f *= nf + elif to_idx < to_len and ( + from_idx >= from_len + or guard_or_false(f % t == 0) + or guard_or_false(f > t) + or (free_symbols(f) and not free_symbols(t)) + ): + nt = to_size[to_idx] + to_group_shape.append(nt) + to_idx += 1 + t *= nt + elif from_idx < from_len and free_symbols(t) and not free_symbols(f): + nf = from_size[from_idx] + from_group_dim.append(from_idx) + from_idx += 1 + f *= nf + elif to_idx < to_len and from_idx < from_len: + # With unbacked dimensions, neither divisibility nor + # ordering may be provable without a data-dependent guard. + # Fall back to one conservative reshape group over the + # remaining contiguous dims; sharding propagation can then + # replicate if the exact split is not statically known. + while from_idx < from_len: + nf = from_size[from_idx] + from_group_dim.append(from_idx) + from_idx += 1 + f *= nf + while to_idx < to_len: + nt = to_size[to_idx] + to_group_shape.append(nt) + to_idx += 1 + t *= nt + break + else: + break + + if len(to_group_shape) > 0: + flattened = Flatten.new( + tuple( + InputDim(fi) + for fi in from_group_dim + if guard_or_true(from_size[fi] >= 1) + ) + ) + result_pp += [ + Split.new(flattened, tuple(to_group_shape), i) + for i in range(len(to_group_shape)) + ] + + return tuple(result_pp) + + +def dim_tile(ndim: int, dims: tuple[int, ...]) -> DimMap: + if len(dims) < ndim: + dims = (1,) * (ndim - len(dims)) + dims + return dim_repeat(ndim, dims) + + +def dim_transpose(ndim: int, dim1: int, dim2: int) -> DimMap: + dim1 = normalize_dim(dim1, ndim) + dim2 = normalize_dim(dim2, ndim) + if not dim1 < ndim: + raise AssertionError(f"Expected dim1 < ndim, got {dim1} >= {ndim}") + if not dim2 < ndim: + raise AssertionError(f"Expected dim2 < ndim, got {dim2} >= {ndim}") + dimmap = [InputDim(i) for i in range(ndim)] + swapdim = dimmap[dim1] + dimmap[dim1] = dimmap[dim2] + dimmap[dim2] = swapdim + return tuple(dimmap) + + +def dim_squeeze(shape: Shape, dim: DimsType | None = None) -> DimMap: + # Operates on local shape; sharding_prop rewrites squeeze ops to squeeze.dims + # with only globally-singleton dims before this is called. + from torch.fx.experimental.symbolic_shapes import guard_or_true + + ndim = len(shape) + if dim is None: + target_dims = set(range(ndim)) + elif isinstance(dim, int): + target_dims = {normalize_dim(dim, ndim)} + else: + target_dims = set(normalize_dims(dim, ndim)) + return tuple( + InputDim(i) + for i, s in enumerate(shape) + if guard_or_true(s > 1) or i not in target_dims + ) + + +def dim_unsqueeze(ndim: int, dim: int) -> DimMap: + dims = tuple(InputDim(i) for i in range(ndim)) + if dim < 0: + dim += ndim + 1 + return dims[:dim] + (Singleton(),) + dims[dim:] + + +def dim_view_as_real(shape: Shape) -> DimMap: + ndim = len(shape) + results: list[DimSpec] = [InputDim(i) for i in range(ndim - 1)] + # each complex number is split into two real numbers, + # resulting in one more dimension of size 2 + results.append(Split(InputDim(ndim - 1), (shape[-1], 2), 0)) + results.append(Split(InputDim(ndim - 1), (shape[-1], 2), 1)) + return tuple(results) + + +def dim_reduction(ndim: int, dim_or_dims: DimsType | None, keepdim: bool) -> DimMap: + """ + General fallback for reduction ops where Partial() does not apply. + + This will cause incoming tensor to be replicated on the reducing dimensions. + """ + if dim_or_dims is None: + dim_or_dims = tuple(range(ndim)) + if isinstance(dim_or_dims, int): + dim_or_dims = (dim_or_dims,) + dim_or_dims = tuple(d if d >= 0 else d + ndim for d in dim_or_dims) + return tuple( + InputDim(i) if i not in dim_or_dims else Singleton() + for i in range(ndim) + if i not in dim_or_dims or keepdim + ) + + +dim_maps: dict[Callable[..., torch.Tensor], Callable[..., DimMap]] = { + torch.atleast_1d: lambda x: dim_pad_left(x.ndim, 1), + torch.atleast_2d: lambda x: dim_pad_left(x.ndim, 2), + torch.atleast_3d: lambda x: dim_atleast_3d(x.ndim), + torch.broadcast_to: lambda input, shape: expand(input.shape, shape), + Tensor.expand: lambda self, *sizes: expand(self.shape, normalize_sizes(sizes)), + torch.flatten: lambda tensor: dim_flatten(tensor.ndim), + torch.movedim: lambda input, source, destination: dim_movedim( + input.ndim, source, destination + ), + torch.permute: lambda input, dims: tuple( + InputDim(i) for i in normalize_dims(dims, input.ndim) + ), + torch.ravel: lambda tensor: dim_flatten(tensor.ndim), + Tensor.repeat: lambda self, *sizes: dim_repeat(self.ndim, sizes), + torch.reshape: lambda input, shape: view_groups(input.shape, shape), + torch.squeeze: lambda input, dim=None: dim_squeeze(input.shape, dim), + torch.tile: lambda input, dims: dim_tile(input.ndim, dims), + torch.transpose: lambda input, dim0, dim1: dim_transpose(input.ndim, dim0, dim1), + torch.unsqueeze: lambda input, dim: dim_unsqueeze(input.ndim, dim), + Tensor.view: lambda input, *shape: view_groups(input.shape, shape), + torch.view_as_complex: lambda input: dim_flatten(input.ndim, input.ndim - 2), + torch.view_as_real: lambda input: dim_view_as_real(input.shape), +} + + +def propagate_shape_and_sharding( + input_src_placements: Sequence[Placement], + global_input_shape: Shape, + rule: DimMap, + mesh_sizes: Shape, + strict_view: bool = False, +) -> tuple[Sequence[Placement], Sequence[Placement]]: + """ + Determine input target sharding and output sharding based on + given global tensor shape and input source sharding. + + Sharding propagation follows mapped dimensions: + - An output dimension that maps directly to an input dimension is sharded equally + - An output dimension that is a flattened set of input dimensions can be sharded: + the first sharded dim stays as Shard, non-first sharded dims become _StridedShard + - An output dimension that is a split of the input dimension can only be sharded + if the leftmost split size is divisible by the mesh dimension + """ + propagator = _ViewShardingPropagator( + input_src_placements, global_input_shape, rule, mesh_sizes, strict_view + ) + input_tgt_placements, input_to_output_tensor_dims = propagator.analyze() + output_placements = propagator.rewrite_output_placements( + input_tgt_placements, input_to_output_tensor_dims + ) + return input_tgt_placements, output_placements + + +class _ViewShardingPropagator: + """Two-phase sharding propagator for view ops. + + Phase 1 — ``analyze()``: + Walks the DimMap rule and returns: + - ``input_tgt_placements``: input placements with unshardable dims + demoted to Replicate. + - ``input_to_output_tensor_dims``: maps each input tensor dim to its + output dim(s). Cardinality encodes the op type: 1→1 for InputDim, + N→1 for Flatten, 1→N for Split/unflatten. + + Phase 2 — ``rewrite_output_placements()``: + Consumes both Phase 1 outputs. Iterates mesh dims 0..n-1, maintaining: + - ``strided_shard_claimed_dims``: (input_dim, output_dim) pairs already assigned + to a mesh dim by _StridedShard rewriting. + - ``local_tensor_shapes``: global shape progressively divided by each + mesh dim's shard size. + For each surviving Shard/_StridedShard, looks up the output dim(s) and + produces the final output placement. + """ + + def __init__( + self, + input_src_placements: Sequence[Placement], + global_input_shape: Shape, + rule: DimMap, + mesh_sizes: Shape, + strict_view: bool, + ) -> None: + self.input_src_placements = input_src_placements + self.global_input_shape = global_input_shape + self.rule = rule + self.mesh_sizes = mesh_sizes + self.strict_view = strict_view + self.mesh_ndim = len(mesh_sizes) + + # shard_allowed[input_dim][mesh_dim]: whether input_dim can stay + # sharded on mesh_dim. Populated by _analyze_dim and its helpers. + self.shard_allowed: dict[int, list[bool]] = {} + # Mesh dims whose _StridedShard has already been matched to an output dim. + # Populated by _analyze_split. + self.matched_strided_mesh_dims: set[int] = set() + self.strict_replicate_fallback: set[tuple[int, int]] = set() + + # ------------------------------------------------------------------ + # Public API: analyze → rewrite_output_placements + # ------------------------------------------------------------------ + + def analyze( + self, + ) -> tuple[Sequence[Placement], dict[int, list[int]]]: + """Phase 1: walk the DimMap rule, return (input_tgt_placements, input_to_output_tensor_dims).""" + input_dims_in_rule = self._input_dims_in_rule(self.rule) + + # Default: shardable if the dim appears in the rule. Refined by _analyze_*. + for dim in range(len(self.global_input_shape)): + self.shard_allowed[dim] = [dim in input_dims_in_rule] * self.mesh_ndim + + # Walk the rule to refine shard_allowed and build input_to_output_tensor_dims. + # + # Flatten example: view([2, 3, 4], [6, 4]) + # rule = (Flatten(InputDim(0), InputDim(1)), InputDim(2)) + # output_dim=0 (Flatten): hits the isinstance(cmd, Flatten) branch. + # Maps input dims 0 and 1 to output dim 0. Result: {0: [0], 1: [0]} + # output_dim=1 (InputDim(2)): hits the len(in_dims) > 0 branch. + # Maps input dim 2 to output dim 1. Result: {0: [0], 1: [0], 2: [1]} + # + # Split example: view([6], [2, 3]) + # rule = (Split(InputDim(0), (2,3), 0), Split(InputDim(0), (2,3), 1)) + # output_dim=0 (split_id=0): hits the len(in_dims) > 0 branch. + # Maps input dim 0 to output dim 0. Result: {0: [0]} + # output_dim=1 (split_id=1): hits the isinstance(cmd, Split) branch + # because _analyze_split returns [] for split_id>0. Chases root + # InputDim(0) and appends output dim 1. Result: {0: [0, 1]} + input_to_output_tensor_dims: dict[int, list[int]] = {} + for output_dim, cmd in enumerate(self.rule): + in_dims = self._analyze_dim(cmd) + if isinstance(cmd, Flatten): + for in_dim in in_dims: + if in_dim.input_dim in input_to_output_tensor_dims: + raise AssertionError( + f"Input dim {in_dim.input_dim} already mapped to output dims " + f"{input_to_output_tensor_dims[in_dim.input_dim]}" + ) + input_to_output_tensor_dims[in_dim.input_dim] = [output_dim] + elif len(in_dims) > 0: + # InputDim (identity) or Split(split_id=0). + in_dim = in_dims[0] + if in_dim.input_dim not in input_to_output_tensor_dims: + input_to_output_tensor_dims[in_dim.input_dim] = [output_dim] + else: + input_to_output_tensor_dims[in_dim.input_dim].append(output_dim) + elif isinstance(cmd, Split): + # Split(split_id>0): _analyze_split returned [], so chase the + # root input dim and append this output dim to its existing entry. + # + # Flatten+Split example: view([2, 3], [3, 2]) + # rule = (Split(Flatten(InputDim(0), InputDim(1)), (3,2), 0), + # Split(Flatten(InputDim(0), InputDim(1)), (3,2), 1)) + # output_dim=0 (split_id=0): same as Split example above. + # Result: {0: [0]} + # output_dim=1 (split_id=1): same as Split example, but + # the chase unwraps the inner Flatten to find InputDim(0). + # Result: {0: [0, 1]} + root_spec = cmd.input_dim + while isinstance(root_spec, (Flatten, Split)): + if isinstance(root_spec, Flatten): + # _analyze_flatten always returns input_dims[0] as + # the first element (either as the only shardable dim + # in non-strict mode, or as the fallback when nothing + # is sharded), so split_id=0 uses it as the key in + # input_to_output_tensor_dims. Use [0] here to match. + root_spec = root_spec.input_dims[0] + else: + root_spec = root_spec.input_dim + root = root_spec if isinstance(root_spec, InputDim) else None + if root is not None and root.input_dim in input_to_output_tensor_dims: + input_to_output_tensor_dims[root.input_dim].append(output_dim) + + input_tgt_placements: list[Placement] = [] + for mesh_dim, p in enumerate(self.input_src_placements): + if ( + isinstance(p, Shard | _StridedShard) + and not self.shard_allowed[p.dim][mesh_dim] + ): + if ( + self.strict_view + and (p.dim, mesh_dim) not in self.strict_replicate_fallback + ): + raise RuntimeError( + f"This operation would remove or reshape sharded " + f"dimension {p.dim}, which requires redistribution. " + f"Please redistribute the input first." + ) + input_tgt_placements.append(Replicate()) + else: + input_tgt_placements.append(p) + return input_tgt_placements, input_to_output_tensor_dims + + def rewrite_output_placements( + self, + input_tgt_placements: Sequence[Placement], + input_to_output_tensor_dims: dict[int, list[int]], + ) -> list[Placement]: + """Phase 2: consume analyze() outputs, return final output placements.""" + # (input_dim, output_dim) pairs claimed by earlier mesh dims + # (via _rewrite_strided_shard), to avoid double-assignment. + strided_shard_claimed_dims: set[ClaimedDim] = set() + # Starts as global_input_shape; each mesh dim divides its sharded dim. + local_tensor_shapes: list[int] = list(self.global_input_shape) + + output_placements: list[Placement] = [] + # Process mesh dims in order; _rewrite_*_shard relies on this for + # truncating division safety in local_tensor_shapes. + for mesh_dim, p in enumerate(input_tgt_placements): + if isinstance(p, Shard): + placement, local_tensor_shapes = self._rewrite_plain_shard( + p, + mesh_dim, + input_tgt_placements, + strided_shard_claimed_dims, + local_tensor_shapes, + input_to_output_tensor_dims, + ) + output_placements.append(placement) + elif isinstance(p, _StridedShard): + placement, local_tensor_shapes = self._rewrite_strided_shard( + p, + mesh_dim, + input_tgt_placements, + strided_shard_claimed_dims, + local_tensor_shapes, + input_to_output_tensor_dims, + ) + output_placements.append(placement) + else: + output_placements.append(p) + return output_placements + + # ------------------------------------------------------------------ + # Analysis phase helpers + # ------------------------------------------------------------------ + + @staticmethod + def _input_dims_in_rule(rule: DimMap) -> set[int]: + """Walk the DimMap rule tree and return all input dim indices that appear in it.""" + seen: set[int] = set() + + def _walk(cmd: DimSpec) -> None: + if isinstance(cmd, InputDim): + seen.add(cmd.input_dim) + for inp in cmd.inputs(): + _walk(inp) + + for cmd in rule: + _walk(cmd) + return seen + + def _find_plain_shard( + self, input_dim: InputDim + ) -> tuple[int | None, Shard | _StridedShard | None]: + """Find the mesh dim with a plain Shard on ``input_dim``. + + Only matches Shard, not _StridedShard. Used by both _analyze_flatten + and _analyze_split. _find_shard_for_split is the counterpart that + also matches _StridedShard with split_factor validation. + """ + for mesh_dim, placement in enumerate(self.input_src_placements): + if isinstance(placement, Shard) and placement.dim == input_dim.input_dim: + return mesh_dim, placement + return None, None + + def _find_shard_for_split( + self, + current_dim: int, + cmd: Split, + placements: Sequence[Placement], + ) -> tuple[int | None, Shard | _StridedShard | None]: + """Find the mesh dim and placement for an input dim in Split ops. + + Matches both Shard and _StridedShard: + - Shard: plain unflatten, e.g. [6] Shard(0) → [2, 3]. + - _StridedShard: unflatten after a prior flatten that produced + _StridedShard, e.g. [2,3,4] Shard(1) → flatten → [6,4] + _StridedShard(0,sf=2) → unflatten → [2,3,4]. Validates that + the split_factor matches the expected value for this split_id. + """ + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Shard | _StridedShard): + continue + if placement.dim != current_dim: + continue + if mesh_dim in self.matched_strided_mesh_dims: + continue + + if isinstance(placement, _StridedShard): + expected_sf = self._expected_split_factor( + cmd, current_dim, mesh_dim, placements + ) + if _split_factor_matches(expected_sf, placement.split_factor): + return mesh_dim, placement + else: + return mesh_dim, placement + return None, None + + def _analyze_flatten(self, cmd: Flatten) -> list[InputDim]: + """Fill self.shard_allowed for Flatten; return sharded input dims.""" + from torch.fx.experimental.symbolic_shapes import guard_or_true + + sharded_dims: list[InputDim] = [] + num_input_dims = len(cmd.input_dims) + for i, dim in enumerate(cmd.input_dims): + if not isinstance(dim, InputDim): + raise AssertionError(f"Expected InputDim, got {type(dim)}") + shard_mesh_dim, shard_placement = self._find_plain_shard(dim) + if shard_mesh_dim is None or shard_placement is None: + continue # default from analyze() already covers this + tensor_dim_size = self.global_input_shape[shard_placement.dim] + mesh_dim_size = self.mesh_sizes[shard_mesh_dim] + can_shard_dim = True + if self.strict_view: + is_last_input_dim = i == num_input_dims - 1 + if not is_last_input_dim and guard_or_true( + tensor_dim_size % mesh_dim_size != 0 + ): + raise RuntimeError( + f"Cannot flatten unevenly sharded tensor: " + f"dimension {dim.input_dim} (size {tensor_dim_size}) " + f"is not evenly divisible by mesh dimension " + f"{shard_mesh_dim} (size {mesh_dim_size}). " + f"Please redistribute the tensor before this operation." + ) + sharded_dims.append(dim) + else: + # TODO: non-strict (reshape) should allow can_shard_dim = True + # for non-first flatten dims, since strict_view already does. + # Currently forces redistribution because the rewrite phase + # wasn't originally implemented for this case. + if i == 0: + sharded_dims.append(dim) + if guard_or_true(tensor_dim_size % mesh_dim_size != 0): + can_shard_dim = False + else: + can_shard_dim = False + self.shard_allowed[dim.input_dim] = [can_shard_dim] * self.mesh_ndim + + if len(sharded_dims) > 0: + return sharded_dims + # No sharded dims: e.g. Flatten([InputDim(0), InputDim(1)]) where + # neither dim is sharded. Return the first input dim so that + # input_to_output_tensor_dims is populated for identity rewrites. + if not isinstance(cmd.input_dims[0], InputDim): + raise AssertionError(f"Expected InputDim, got {type(cmd.input_dims[0])}") + return [cmd.input_dims[0]] + + def _analyze_split(self, cmd: Split) -> list[InputDim]: + """Fill self.shard_allowed for Split; return shardable input dims.""" + from torch.fx.experimental.symbolic_shapes import guard_or_false, guard_or_true + + in_dims = self._analyze_dim(cmd.input_dim) + if len(in_dims) == 0: + return [] + in_dim = in_dims[0] + if isinstance(cmd.input_dim, Flatten): + # A plain Shard on a non-first flattened input dim would need an + # intermediate _StridedShard before this Split. The combined + # Flatten+Split rewrite does not preserve that placement exactly, + # so redistribute that mesh dim instead of raising later. + for flat_dim in cmd.input_dim.input_dims[1:]: + if not isinstance(flat_dim, InputDim): + raise AssertionError(f"Expected InputDim, got {type(flat_dim)}") + for mesh_dim, placement in enumerate(self.input_src_placements): + if ( + isinstance(placement, Shard) + and placement.dim == flat_dim.input_dim + ): + self.shard_allowed[flat_dim.input_dim][mesh_dim] = False + self.strict_replicate_fallback.add( + (flat_dim.input_dim, mesh_dim) + ) + out_size = cmd.group_shape[cmd.split_id] + shard_mesh_dim, input_src_placement = self._find_shard_for_split( + in_dim.input_dim, cmd, self.input_src_placements + ) + # split_id == 0 sets the base shard_allowed for this input dim. + # Later split_ids (processed in subsequent rule iterations) refine + # individual mesh_dim entries via the _StridedShard branch below. + if cmd.split_id == 0: + self.shard_allowed[in_dim.input_dim] = [ + guard_or_false(out_size % mesh_dim_size == 0) + for mesh_dim_size in self.mesh_sizes + ] + plain_mesh_dim, _ = self._find_plain_shard(in_dim) + # Non-strict silently redistributes via shard_allowed=False above; + # strict raises so the user knows to redistribute before view(). + if self.strict_view and plain_mesh_dim is not None: + if not self.shard_allowed[in_dim.input_dim][plain_mesh_dim]: + raise RuntimeError( + f"Cannot unflatten unevenly sharded tensor: " + f"output dimension {cmd.split_id} (size {out_size}) " + f"is not evenly divisible by mesh dimension " + f"{plain_mesh_dim} (size {self.mesh_sizes[plain_mesh_dim]}). " + f"Please redistribute the tensor before this operation." + ) + if shard_mesh_dim is not None and isinstance( + input_src_placement, _StridedShard + ): + # The last split dim doesn't require even divisibility because + # its local size is inferred: local_last = local_flat / product + # of earlier dims, and DTensor handles uneven local sizes. + # Non-last dims must be evenly divisible because they appear as + # fixed sizes in the local reshape — uneven division would make + # the stride pattern inconsistent across devices. + # E.g. [12] → [3, 4], _StridedShard targeting dim 1 (last), + # mesh=3: 4%3≠0, but local shapes [3,2],[3,1],[3,1] are valid. + is_last_split_dim = cmd.split_id == len(cmd.group_shape) - 1 + if ( + self.strict_view + and not is_last_split_dim + and guard_or_true(out_size % self.mesh_sizes[shard_mesh_dim] != 0) + ): + raise RuntimeError( + f"Cannot unflatten unevenly sharded tensor: " + f"output dimension {cmd.split_id} (size {out_size}) " + f"is not evenly divisible by mesh dimension {shard_mesh_dim} " + f"(size {self.mesh_sizes[shard_mesh_dim]}). " + f"Please redistribute the tensor before this operation." + ) + # Prevents _find_shard_for_split from matching this mesh dim + # again for a later split_id of the same Split group. + self.matched_strided_mesh_dims.add(shard_mesh_dim) + if in_dim.input_dim in self.shard_allowed: + self.shard_allowed[in_dim.input_dim][shard_mesh_dim] = ( + guard_or_false(out_size % self.mesh_sizes[shard_mesh_dim] == 0) + or is_last_split_dim + ) + # Only split_id==0 returns the input dim for input_to_output_tensor_dims. + # Later split_ids refine shard_allowed above but return [] — their + # output dims are linked via the root-input-dim chase in analyze(). + return [in_dim] if cmd.split_id == 0 else [] + + def _analyze_dim(self, cmd: DimSpec) -> list[InputDim]: + """Dispatch one DimSpec: update self.shard_allowed, return input dim(s) to shard on.""" + if isinstance(cmd, InputDim): + return [cmd] + elif isinstance(cmd, Flatten): + return self._analyze_flatten(cmd) + elif isinstance(cmd, Split): + return self._analyze_split(cmd) + elif isinstance(cmd, Repeat): + in_dims = self._analyze_dim(cmd.input_dim) + for d in in_dims: + self.shard_allowed[d.input_dim] = [False] * self.mesh_ndim + return [] + else: + return [] + + # ------------------------------------------------------------------ + # Rewrite phase helpers + # ------------------------------------------------------------------ + + @staticmethod + def _is_last_shard_in_flatten_range( + mesh_dim: int, + placements: Sequence[Placement], + flatten_start: int, + flatten_end: int, + ) -> bool: + """Check if no later mesh dim shards a dim within the flatten range at or above this one. + + Uneven sharding on dim d breaks stride computation for all earlier dims + that flatten together with d. Only dims within [flatten_start, flatten_end) + matter; shards on dims outside the flatten range are independent. + + Requires: placements[mesh_dim] must be Shard or _StridedShard. + """ + p = placements[mesh_dim] + if not isinstance(p, (Shard, _StridedShard)): + raise AssertionError( + f"Expected Shard or _StridedShard at mesh_dim {mesh_dim}, got {type(p)}" + ) + tensor_dim = p.dim + return not any( + isinstance(other_p, (Shard, _StridedShard)) + and flatten_start <= other_p.dim < flatten_end + and other_p.dim >= tensor_dim + for other_p in placements[mesh_dim + 1 :] + ) + + def _expected_split_factor( + self, + cmd: Split, + sharded_dim: int, + mesh_dim: int, + placements: Sequence[Placement], + ) -> int | torch.SymInt | None: + """Compute the residual split factor for ``cmd`` after earlier mesh dims. + + Starts from ``math.prod(cmd.group_shape[:cmd.split_id])`` and divides + out each earlier mesh dim that shards the same input dim. Returns + ``None`` if any earlier mesh size doesn't divide evenly. + """ + sf = math.prod(cmd.group_shape[: cmd.split_id]) + for m in range(mesh_dim): + other_p = placements[m] + if ( + isinstance(other_p, (_StridedShard, Shard)) + and other_p.dim == sharded_dim + ): + if sf % self.mesh_sizes[m] != 0: + return None + sf //= self.mesh_sizes[m] + return sf + + def _find_keep_ss_dim( + self, + tgt_shard_dims: list[int], + p: _StridedShard, + mesh_dim: int, + ) -> int | None: + """Find an output dim where SS stays as SS. + + Returns the first output dim whose Split can accommodate the combined + sharding (mesh_size * split_factor), or ``None`` if no dim fits. + """ + total_shard = cast(Any, self.mesh_sizes[mesh_dim]) * p.split_factor + input_dim_size = cast(Any, self.global_input_shape[p.dim]) + if input_dim_size % total_shard != 0: + return None + shard_size = input_dim_size // total_shard + for candidate_dim in tgt_shard_dims: + cmd = self.rule[candidate_dim] + if isinstance(cmd, Split): + inner_size = math.prod(cmd.group_shape[cmd.split_id + 1 :]) + # When a Split wraps a Flatten, the per-shard chunk covers + # the sharded dim plus trailing dims flattened together. + trailing_size = 1 + if isinstance(cmd.input_dim, Flatten): + found = False + for flat_dim in cmd.input_dim.input_dims: + if not isinstance(flat_dim, InputDim): + raise AssertionError( + f"Expected InputDim, got {type(flat_dim)}" + ) + if flat_dim.input_dim == p.dim: + found = True + elif found: + trailing_size *= self.global_input_shape[flat_dim.input_dim] + flattened_shard_size = shard_size * trailing_size + if ( + flattened_shard_size >= inner_size + and flattened_shard_size % inner_size == 0 + ): + return candidate_dim + return None + + def _rewrite_plain_shard( + self, + p: Shard, + mesh_dim: int, + placements: Sequence[Placement], + strided_shard_claimed_dims: set[ClaimedDim], + local_tensor_shapes: list[int], + input_to_output_tensor_dims: dict[int, list[int]], + ) -> tuple[Placement, list[int]]: + """Given a plain Shard(dim=X) input placement on a specific mesh dim, + determine what output placement it maps to after the view op. + + For identity and unflatten, produces Shard on the output dim. For + flatten, Shard on the first flattened dim stays Shard, while Shard on + a non-first dim produces _StridedShard (consumed later by + _rewrite_strided_shard). + + Returns the output placement and a new local_tensor_shapes with this + mesh dim's division applied. + """ + # Output dims that input dim p.dim maps to, filtering out any + # already claimed by _StridedShard rewriting on earlier mesh dims. + tgt_shard_dims = [ + d + for d in input_to_output_tensor_dims[p.dim] + if ClaimedDim(p.dim, d) not in strided_shard_claimed_dims + ] + if len(tgt_shard_dims) == 0: + raise AssertionError( + f"No output dim available for Shard(dim={p.dim}) on mesh dim " + f"{mesh_dim}. All output dims already claimed by earlier mesh dims." + ) + if len(tgt_shard_dims) == 1: + tgt_shard_dim = tgt_shard_dims[0] + else: + # Unflatten: one input dim maps to multiple output dims + # (e.g. (24,) → (2, 3, 4) gives 3 splits). Plain Shard + # always targets the split_id=0 output dim. + tgt_shard_dim = next( + ( + d + for d in tgt_shard_dims + if isinstance(self.rule[d], Split) + and cast(Split, self.rule[d]).split_id == 0 + ), + None, + ) + if tgt_shard_dim is None: + raise AssertionError( + f"No Split(split_id=0) found among unclaimed output dims " + f"{tgt_shard_dims} for Shard(dim={p.dim}) on mesh dim {mesh_dim}." + ) + cmd = self.rule[tgt_shard_dim] + if isinstance(cmd, Split) and isinstance(cmd.input_dim, Flatten): + first_dim = cmd.input_dim.input_dims[0] + if isinstance(first_dim, InputDim) and p.dim != first_dim.input_dim: + raise AssertionError( + f"Shard(dim={p.dim}) should have been demoted before " + f"Split(Flatten(...), {cmd.group_shape}) rewrite." + ) + if isinstance(cmd, (Split, InputDim)): + # Split/InputDim: 1:1 dim mapping, sharding transfers directly. + # Flatten needs stride computation below (multiple dims merge). + new_shapes = list(local_tensor_shapes) + new_shapes[p.dim] //= self.mesh_sizes[mesh_dim] + return Shard(tgt_shard_dim), new_shapes + if not isinstance(cmd, Flatten): + raise AssertionError(f"Expected Flatten, got {type(cmd)}") + first_dim = cmd.input_dims[0] + last_dim = cmd.input_dims[-1] + if not isinstance(first_dim, InputDim): + raise AssertionError(f"Expected InputDim, got {type(first_dim)}") + if not isinstance(last_dim, InputDim): + raise AssertionError(f"Expected InputDim, got {type(last_dim)}") + input_start_idx = first_dim.input_dim + if p.dim == input_start_idx: + output_placement: Placement = Shard(tgt_shard_dim) + else: + split_factor = math.prod(local_tensor_shapes[input_start_idx : p.dim]) + if isinstance(split_factor, torch.SymInt): + if _explicit_unbacked_hint(split_factor) is None: + raise RuntimeError( + "Cannot propagate strict view sharding through a symbolic " + "_StridedShard split_factor without an explicit unbacked " + f"hint override: split_factor={split_factor}, " + f"input_shape={self.global_input_shape}, rule={self.rule}." + ) + output_placement = _StridedShard(tgt_shard_dim, split_factor=split_factor) + # Uneven sharding on a non-last flatten dim breaks _StridedShard: + # split_factor (number of groups) must be the same on all devices, + # but uneven division of a non-last dim makes group count vary. + # E.g. [3,4]→[12] Shard(0) mesh=2: device 0 has 2 groups of 4, + # device 1 has 1 group of 4 — no consistent split_factor. + # The last dim is exempt: only group *size* varies, not count. + flatten_end = last_dim.input_dim + 1 + if local_tensor_shapes[p.dim] % self.mesh_sizes[ + mesh_dim + ] != 0 and not self._is_last_shard_in_flatten_range( + mesh_dim, placements, input_start_idx, flatten_end + ): + raise RuntimeError( + f"Cannot shard unevenly distributed tensor: " + f"dimension {p.dim} (size {local_tensor_shapes[p.dim]}) " + f"is not evenly divisible by mesh dimension " + f"{mesh_dim} (size {self.mesh_sizes[mesh_dim]}). " + f"Please redistribute the tensor before this operation." + ) + new_shapes = list(local_tensor_shapes) + new_shapes[p.dim] //= self.mesh_sizes[mesh_dim] + return output_placement, new_shapes + + def _rewrite_strided_shard( + self, + p: _StridedShard, + mesh_dim: int, + placements: Sequence[Placement], + strided_shard_claimed_dims: set[ClaimedDim], + local_tensor_shapes: list[int], + input_to_output_tensor_dims: dict[int, list[int]], + ) -> tuple[Placement, list[int]]: + """Rewrite _StridedShard placement to target the correct output dim. + + _StridedShard inputs arise from a prior flatten on a non-first dim + (produced by _rewrite_plain_shard above). The interesting case is + unflatten (Split rule): the split_factor may resolve to contiguous + sharding (producing Shard) or stay as _StridedShard. For + identity/flatten rules, falls through to the fallback and keeps the + placement as-is. + + Returns the output placement and a new local_tensor_shapes with this + mesh dim's division applied. + """ + tgt_shard_dims = [ + d + for d in input_to_output_tensor_dims[p.dim] + if ClaimedDim(p.dim, d) not in strided_shard_claimed_dims + ] + # Phase 1: resolve SS → Shard. If an output dim's Split has a + # group_shape prefix matching the split_factor, the strided pattern + # is fully captured by the Split, so SS simplifies to Shard. + # E.g. unflatten (6, 4) → (2, 3, 4) with SS(0, sf=2) on mesh (3): + # sf=2 means 2 groups of contiguous data in dim 0. Split into + # (2, 3, 4) gives group_shape=(2, 3); prod(group_shape[:1])=2==sf, + # so the strided pattern lands exactly on output dim 1 → Shard(1). + for candidate_dim in tgt_shard_dims: + cmd = self.rule[candidate_dim] + if isinstance(cmd, Split): + expected_sf = self._expected_split_factor( + cmd, p.dim, mesh_dim, placements + ) + if not _split_factor_matches(expected_sf, p.split_factor): + continue + strided_shard_claimed_dims.add(ClaimedDim(p.dim, candidate_dim)) + new_shapes = list(local_tensor_shapes) + new_shapes[p.dim] //= self.mesh_sizes[mesh_dim] + return Shard(candidate_dim), new_shapes + + # Phase 2: keep SS as SS. Phase 1 is tried first because we prefer + # resolving to the simpler Shard when possible. + tgt_shard_dim = self._find_keep_ss_dim(tgt_shard_dims, p, mesh_dim) + + if tgt_shard_dim is None: + if self.strict_view and any( + isinstance(self.rule[d], Split) for d in tgt_shard_dims + ): + raise RuntimeError( + f"Cannot unflatten tensor with _StridedShard placement: " + f"split_factor={p.split_factor} does not match any output " + f"dimension. This typically means the _StridedShard placement " + f"was constructed with a split_factor that is incompatible " + f"with the unflatten shape. Please redistribute the tensor " + f"before this operation." + ) + if len(tgt_shard_dims) == 0: + raise AssertionError( + f"No unclaimed output dims for _StridedShard(dim={p.dim}) " + f"on mesh dim {mesh_dim}." + ) + # Fallback for identity/flatten: tgt_shard_dims has exactly one + # element, so [0] is correct. For Split rules this is unreachable + # in practice — the analysis phase rejects mismatched split_factors + # via shard_allowed, forcing redistribution before we get here. + tgt_shard_dim = tgt_shard_dims[0] + new_shapes = list(local_tensor_shapes) + new_shapes[p.dim] //= self.mesh_sizes[mesh_dim] + return _StridedShard(tgt_shard_dim, split_factor=p.split_factor), new_shapes + + +def register_op_strategy_map( + aten_op_overload: torch._ops.OpOverload, + local_op_name: Callable[..., torch.Tensor], + schema_info: RuntimeSchemaInfo | None = None, + strict_view: bool = False, +) -> None: + """ + Helper that registers strategies for view-like operators that follow a pattern: + (1) define the way input dims are split/combined to form output dims (dim_maps) + (2) register a strategy for the op schema that uses the dim_map as a sharding prop rule + + strict_view: if True, we will error out if the view-operation would require resharding the input. + Currently, this should be set to 'true' for any "view" ops. + We could diverge behavior for "reshape" ops which could perform a redistribute implicitly. + """ + dim_map: Callable[..., DimMap] = dim_maps[local_op_name] + + @register_op_strategy(aten_op_overload, schema_info=schema_info) + def reshape_strategy(op_schema: OpSchema) -> StrategyType: + rules = dim_map(*op_schema.args_schema, **op_schema.kwargs_schema) + input_strategy = cast(OpStrategy, op_schema.args_schema[0]) + mesh = op_schema.get_mesh_from_args(validate=False) + + global_in_shape = input_strategy.shape + if global_in_shape is None: + raise AssertionError("Shape required.") + + output_strategy = OpStrategy([]) + for input_placement_strategy in input_strategy.strategies: + input_src_spec = input_placement_strategy.output_spec + + input_tgt_placements, output_placements = propagate_shape_and_sharding( + input_src_spec.placements, + tuple(global_in_shape), + rules, + mesh.shape, + strict_view, + ) + + # TODO: optimize this. we shouldn't simply blindly replicate + # unshardable dims ... + # FIXME: this can be wrong for situations where we have + # [Shard(0), Shard(0)] + input_tgt_spec = DTensorSpec( + placements=tuple(input_tgt_placements), + mesh=mesh, + tensor_meta=input_src_spec.tensor_meta, + use_strided_shard_as_shard_order=False, + ) + redistribute_costs: list[list[float]] = [ + generate_redistribute_costs(input_strategy, input_tgt_spec) + ] + + output_spec = DTensorSpec( + mesh=mesh, + placements=tuple(output_placements), + use_strided_shard_as_shard_order=False, + ) + output_strategy.strategies.append( + OpSpec( + output_specs=output_spec, + input_specs=(input_tgt_spec,), + redistribute_cost=redistribute_costs, + ) + ) + + return output_strategy + + +register_op_strategy_map(aten.squeeze.default, torch.squeeze, strict_view=True) +register_op_strategy_map(aten.squeeze_.default, torch.squeeze, strict_view=True) +register_op_strategy_map( + aten.squeeze_.dim, + torch.squeeze, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.squeeze.dim, + torch.squeeze, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.squeeze.dims, + torch.squeeze, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.squeeze_.dims, + torch.squeeze, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.view.default, + Tensor.view, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.view_copy.default, + Tensor.view, + schema_info=RuntimeSchemaInfo(1), +) +register_op_strategy_map( + aten.reshape.default, torch.reshape, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten._unsafe_view.default, + Tensor.view, + schema_info=RuntimeSchemaInfo(1), + strict_view=True, +) +register_op_strategy_map( + aten.unsqueeze.default, torch.unsqueeze, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten.expand.default, Tensor.expand, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten.expand_copy.default, Tensor.expand, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten.permute.default, torch.permute, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten.repeat.default, Tensor.repeat, schema_info=RuntimeSchemaInfo(1) +) +register_op_strategy_map( + aten.transpose.int, torch.transpose, schema_info=RuntimeSchemaInfo(1) +) + + +@register_single_dim_strategy(aten.view_as_complex.default) +def view_as_complex_single_dim_strategy(op, args_schema, kwargs_schema): + # view_as_complex: float [..., 2] -> complex [...] + # Dims 0..ndim-2 map 1:1; last dim (real/imag pair) is consumed. + # P(max)/P(min) invalid: complex numbers have no total ordering. + input_meta = args_schema[0] + if not isinstance(input_meta, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(input_meta)}") + ndim = len(input_meta.shape) + strategies: list[list[Placement | _ShardingPlaceholder]] = [] + for d in range(ndim - 1): + strategies.append([_ShardingPlaceholder(d), _ShardingPlaceholder(d)]) + strategies.append([Partial("sum"), Partial("sum")]) + strategies.append([Partial("avg"), Partial("avg")]) + return strategies + + +register_op_strategy_map(aten.view_as_real.default, torch.view_as_real) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/autogen.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/autogen.py new file mode 100644 index 0000000000000000000000000000000000000000..266f3d2b52c1468179ec5395109114d3de4adcce --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/autogen.py @@ -0,0 +1,575 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +"""Schema-discovered registration of single-dim strategy variants. + +Manual single-dim registrations target the semantic base op. +``auto_register_op_variants`` mechanically covers the related ATen overloads +(in-place, out, functional, and foreach) by reusing the base op's strategy. +""" + +from collections.abc import Sequence +from typing import Any + +from torch._ops import OpOverload +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ArgsType, KwargsType, RuntimeSchemaInfo +from torch.distributed.tensor._ops.single_dim_strategy import ( + _get_overload_packet, + _get_packet_overload, + _normalize_schema_type, + _op_namespace_and_base_name, + _resolve_foreach_elementwise_overload, + _schema_args_match, + _ShardingPlaceholder, + _SingleDimStrategyFunc, + _SingleDimStrategyInfo, +) +from torch.distributed.tensor.placement_types import Placement + + +def _tensor_schema_count(schema_values: Sequence[Any]) -> int: + """Count Tensor-valued entries in an ATen schema sequence. + + Variant discovery uses this to reject overloads with incompatible arity. + """ + return sum(1 for value in schema_values if "Tensor" in str(value.type)) + + +def _schema_tensor_output_count(op: OpOverload) -> int: + """Count Tensor-valued returns for an ATen overload.""" + return _tensor_schema_count(op._schema.returns) + + +def _is_write_arg(arg: Any) -> bool: + """Return whether a schema argument is mutated by the op. + + Schema alias info is the reliable way to identify explicit out args and + mutable inputs whose placements must be preserved. + """ + return arg.alias_info is not None and arg.alias_info.is_write + + +def _is_foreach_like_op_name(op_name: str) -> bool: + """Return whether the op name uses foreach/fused naming conventions. + + These ops are already aggregate/list variants, so they should not be used + as scalar bases for another round of automatic variant registration. + """ + return op_name.startswith( + ("aten::_foreach_", "aten::_amp_foreach_", "aten::_fused_") + ) + + +def _schema_args_are_same( + base_args: Sequence[Any], candidate_args: Sequence[Any] +) -> bool: + """Return whether two schema arg lists are identical for strategy reuse. + + Out variants require this stricter check so only the written output args are + allowed to differ from the base op. + """ + if len(base_args) != len(candidate_args): + return False + return all( + base_arg.name == candidate_arg.name + and base_arg.kwarg_only == candidate_arg.kwarg_only + and _normalize_schema_type(base_arg) == _normalize_schema_type(candidate_arg) + for base_arg, candidate_arg in zip(base_args, candidate_args) + ) + + +def _is_explicit_out_arg(arg: Any) -> bool: + """Return whether a schema argument is an explicit output argument.""" + return bool(getattr(arg, "is_out", False)) + + +def _schema_non_alias_tensor_output_indices(op: OpOverload) -> list[int]: + """Return tensor return indices that are not aliases of mutable inputs.""" + indices: list[int] = [] + for idx, ret in enumerate(op._schema.returns): + if "Tensor" in str(ret.type) and ret.alias_info is None: + indices.append(idx) + return indices + + +def _schema_written_tensor_arg_count(op: OpOverload) -> int: + """Count Tensor-valued arguments written by an op.""" + return sum( + 1 + for arg in op._schema.arguments + if _is_write_arg(arg) and "Tensor" in str(arg.type) + ) + + +def _functional_variant_tensor_output_count(base_op: OpOverload) -> int: + """Return expected tensor outputs for a functionalization variant. + + Mutable ops can have real tensor returns and/or alias returns for mutated + inputs. Functionalization returns the real tensor returns plus updated + copies of written tensor inputs. + """ + return len(_schema_non_alias_tensor_output_indices(base_op)) + ( + _schema_written_tensor_arg_count(base_op) + ) + + +def _iter_packet_overloads(packet: Any | None) -> list[OpOverload]: + """Return overloads from an overload packet, skipping lookup misses.""" + if packet is None: + return [] + overloads: list[OpOverload] = [] + for overload_name in packet.overloads(): + overload = _get_packet_overload(packet, overload_name) + if overload is not None: + overloads.append(overload) + return overloads + + +def _count_tensor_meta_values(value: object) -> int: + """Count TensorMeta leaves in a value that may contain nested tuples/lists. + + Strategy rule indices are based on tensor leaves, not raw schema argument + count. + """ + if isinstance(value, TensorMeta): + return 1 + if isinstance(value, (list, tuple)): + return sum(_count_tensor_meta_values(item) for item in value) + return 0 + + +def _clone_schema_info( + schema_info: RuntimeSchemaInfo | None, *, needs_pytree: bool | None = None +) -> RuntimeSchemaInfo | None: + """Copy schema cache metadata, optionally overriding pytree flattening. + + Auto-registered variants should inherit the base op cache behavior unless + their argument structure changes, as foreach does with Tensor lists. + """ + if schema_info is None: + if needs_pytree is None: + return None + return RuntimeSchemaInfo(needs_pytree=needs_pytree) + + return RuntimeSchemaInfo( + static_argnum=schema_info.static_argnum, + static_kwargkey=( + list(schema_info.static_kwargkey) + if schema_info.static_kwargkey is not None + else None + ), + needs_pytree=schema_info.needs_pytree if needs_pytree is None else needs_pytree, + ) + + +def _clone_strategy_info( + info: _SingleDimStrategyInfo, func: _SingleDimStrategyFunc +) -> _SingleDimStrategyInfo: + """Copy strategy options while replacing the underlying strategy function. + + Variant wrappers must preserve base-op flags such as uneven sharding support. + """ + return _SingleDimStrategyInfo( + func=func, + allow_unbacked_sharding=info.allow_unbacked_sharding, + allow_uneven_sharding=info.allow_uneven_sharding, + different_mesh_args=( + list(info.different_mesh_args) + if info.different_mesh_args is not None + else None + ), + ) + + +def _canonical_variant_base_name(op: OpOverload) -> tuple[str, str]: + """Return the namespace and base name shared by related ATen variants.""" + namespace, base_name = _op_namespace_and_base_name(op) + if base_name.endswith("_functional"): + return namespace, base_name.removesuffix("_functional") + if op._schema.is_mutable and base_name.endswith("_"): + return namespace, base_name.removesuffix("_") + return namespace, base_name + + +def _strip_output_args_for_base_call( + out_op: OpOverload, + output_arg_names: set[str], + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> tuple[ArgsType, KwargsType]: + """Drop explicit out args before invoking the functional base strategy. + + Base strategy functions do not expect the extra writable tensors from the + out overload schema. + """ + base_args: list[object] = [] + positional_idx = 0 + for arg in out_op._schema.arguments: + if arg.kwarg_only: + continue + if positional_idx >= len(args_schema): + break + value = args_schema[positional_idx] + positional_idx += 1 + if arg.name not in output_arg_names: + base_args.append(value) + + base_kwargs = { + name: value + for name, value in kwargs_schema.items() + if name not in output_arg_names + } + return tuple(base_args), base_kwargs + + +def _find_inplace_variant_overloads(base_op: OpOverload) -> list[OpOverload]: + """Find an inplace overload that has the same tensor schema as base_op. + + Inplace variants can reuse the base strategy only when inputs and outputs + have the same tensor structure. + """ + namespace, base_name = _canonical_variant_base_name(base_op) + if base_op._schema.is_mutable: + return [] + if _is_foreach_like_op_name(base_op.name()): + return [] + if any(_is_explicit_out_arg(arg) for arg in base_op._schema.arguments): + return [] + + packet = _get_overload_packet(namespace, f"{base_name}_") + if packet is None: + return [] + + variants: list[OpOverload] = [] + for candidate in _iter_packet_overloads(packet): + if not candidate._schema.is_mutable: + continue + if _schema_tensor_output_count(candidate) != _schema_tensor_output_count( + base_op + ): + continue + if _schema_args_match(base_op._schema.arguments, candidate._schema.arguments): + variants.append(candidate) + return variants + + +def _find_out_variant_overloads( + base_op: OpOverload, +) -> list[tuple[OpOverload, tuple[str, ...]]]: + """Find an out overload and return its written output arg names. + + Out variants need extra strategy inputs for the provided output tensors. + """ + namespace, base_name = _canonical_variant_base_name(base_op) + if _is_foreach_like_op_name(base_op.name()): + return [] + if base_name.endswith("_functional"): + return [] + if any(_is_explicit_out_arg(arg) for arg in base_op._schema.arguments): + return [] + + base_num_outputs = _schema_tensor_output_count(base_op) + if base_num_outputs == 0: + return [] + + packet = _get_overload_packet(namespace, base_name) + if packet is None: + return [] + + variants: list[tuple[OpOverload, tuple[str, ...]]] = [] + for candidate in _iter_packet_overloads(packet): + if candidate is base_op: + continue + if not candidate._schema.is_mutable: + continue + if _schema_tensor_output_count(candidate) != base_num_outputs: + continue + + output_args = [ + arg for arg in candidate._schema.arguments if _is_explicit_out_arg(arg) + ] + if len(output_args) != base_num_outputs: + continue + if not all("Tensor" in str(arg.type) for arg in output_args): + continue + + non_output_args = [ + arg for arg in candidate._schema.arguments if not _is_explicit_out_arg(arg) + ] + if _schema_args_are_same(base_op._schema.arguments, non_output_args): + variants.append((candidate, tuple(arg.name for arg in output_args))) + + return variants + + +def _find_functional_variant_overloads(base_op: OpOverload) -> list[OpOverload]: + """Find the functional variant generated from a mutable base overload. + + Functional variants return updated copies of mutable inputs, so their extra + outputs must follow the mutated input placements. + """ + if not any(_is_write_arg(arg) for arg in base_op._schema.arguments): + return [] + namespace, base_name = _canonical_variant_base_name(base_op) + if _is_foreach_like_op_name(base_op.name()): + return [] + if any(_is_explicit_out_arg(arg) for arg in base_op._schema.arguments): + return [] + + packet = _get_overload_packet(namespace, f"{base_name}_functional") + if packet is None: + return [] + + variants: list[OpOverload] = [] + expected_outputs = _functional_variant_tensor_output_count(base_op) + for candidate in _iter_packet_overloads(packet): + if candidate._schema.is_mutable: + continue + if not _schema_args_match( + base_op._schema.arguments, candidate._schema.arguments + ): + continue + if _schema_tensor_output_count(candidate) != expected_outputs: + continue + variants.append(candidate) + return variants + + +def _find_foreach_variants(base_op: OpOverload) -> list[OpOverload]: + """Find elementwise foreach overloads that can reuse base_op's strategy. + + Foreach variants are registered in one pass because newly registered + variants are not revisited by auto_register_op_variants. + """ + namespace, base_name = _op_namespace_and_base_name(base_op) + if namespace != "aten": + return [] + if base_op._schema.is_mutable or _is_foreach_like_op_name(base_op.name()): + return [] + if _schema_tensor_output_count(base_op) == 0: + return [] + + variants: list[OpOverload] = [] + for foreach_name in (f"_foreach_{base_name}", f"_foreach_{base_name}_"): + packet = _get_overload_packet(namespace, foreach_name) + if packet is None: + continue + for overload_name in packet.overloads(): + if "out" in overload_name: + continue + candidate = _get_packet_overload(packet, overload_name) + if candidate is None: + continue + if _resolve_foreach_elementwise_overload(candidate) == base_op: + variants.append(candidate) + return variants + + +def _make_same_schema_variant_strategy_fn( + base_fn: _SingleDimStrategyFunc, base_op: OpOverload +) -> _SingleDimStrategyFunc: + """Wrap base_fn so a same-schema variant uses base-op strategy logic. + + Passing base_op keeps op-sensitive strategy functions on their original + lookup paths. + """ + + def strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType + ) -> list[list[Placement | _ShardingPlaceholder]]: + return base_fn(base_op, args_schema, kwargs_schema) + + return strategy + + +def _make_out_variant_strategy_fn( + base_fn: _SingleDimStrategyFunc, + base_op: OpOverload, + out_op: OpOverload, + output_arg_names: tuple[str, ...], +) -> _SingleDimStrategyFunc: + """Wrap base_fn so out tensors appear as extra strategy inputs. + + DTensor dispatch cannot redistribute out kwargs, so generated rules must + make each output placement agree with its corresponding out tensor. + """ + output_arg_name_set = set(output_arg_names) + output_arg_to_index = { + name: output_idx for output_idx, name in enumerate(output_arg_names) + } + base_num_outputs = _schema_tensor_output_count(base_op) + + def _output_arg_placements( + rule: list[Placement | _ShardingPlaceholder], + arg_name: str, + value: object, + ) -> list[Placement | _ShardingPlaceholder]: + """Return placements to append for one explicit out argument.""" + tensor_count = _count_tensor_meta_values(value) + if tensor_count == 0: + return [] + if tensor_count > 1 and len(output_arg_names) == 1: + return [rule[i] for i in range(base_num_outputs)] + return [rule[output_arg_to_index[arg_name]]] + + def strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType + ) -> list[list[Placement | _ShardingPlaceholder]]: + base_args, base_kwargs = _strip_output_args_for_base_call( + out_op, output_arg_name_set, args_schema, kwargs_schema + ) + base_rules = base_fn(base_op, base_args, base_kwargs) + rules: list[list[Placement | _ShardingPlaceholder]] = [] + + for rule in base_rules: + output_arg_placements: list[Placement | _ShardingPlaceholder] = [] + positional_idx = 0 + for arg in out_op._schema.arguments: + if arg.kwarg_only: + continue + if positional_idx >= len(args_schema): + break + value = args_schema[positional_idx] + positional_idx += 1 + if arg.name in output_arg_to_index: + output_arg_placements.extend( + _output_arg_placements(rule, arg.name, value) + ) + + for name, value in kwargs_schema.items(): + if name in output_arg_to_index: + output_arg_placements.extend( + _output_arg_placements(rule, name, value) + ) + + rules.append([*rule, *output_arg_placements]) + + return rules + + return strategy + + +def _make_functional_variant_strategy_fn( + base_fn: _SingleDimStrategyFunc, base_op: OpOverload +) -> _SingleDimStrategyFunc: + """Wrap mutable base_fn for functional variants with extra mutated outputs. + + The added functional outputs represent updated mutable inputs and therefore + inherit those input placements. + """ + mutable_arg_names = { + arg.name for arg in base_op._schema.arguments if _is_write_arg(arg) + } + base_num_outputs = _schema_tensor_output_count(base_op) + non_alias_output_indices = _schema_non_alias_tensor_output_indices(base_op) + + def _mutable_input_rule_indices( + args_schema: ArgsType, kwargs_schema: KwargsType + ) -> list[int]: + """Map mutable schema args to their positions in a base strategy rule.""" + mutable_indices: list[int] = [] + tensor_input_idx = 0 + positional_idx = 0 + for arg in base_op._schema.arguments: + value = None + if arg.kwarg_only: + value = kwargs_schema.get(arg.name) + elif positional_idx < len(args_schema): + value = args_schema[positional_idx] + positional_idx += 1 + else: + value = kwargs_schema.get(arg.name) + + tensor_count = _count_tensor_meta_values(value) + if tensor_count == 0: + continue + if arg.name in mutable_arg_names: + mutable_indices.append(base_num_outputs + tensor_input_idx) + tensor_input_idx += tensor_count + return mutable_indices + + def strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType + ) -> list[list[Placement | _ShardingPlaceholder]]: + mutable_input_indices = _mutable_input_rule_indices(args_schema, kwargs_schema) + base_rules = base_fn(base_op, args_schema, kwargs_schema) + rules: list[list[Placement | _ShardingPlaceholder]] = [] + for rule in base_rules: + output_placements = [ + rule[output_idx] for output_idx in non_alias_output_indices + ] + output_placements.extend( + rule[input_idx] for input_idx in mutable_input_indices + ) + rules.append([*output_placements, *rule[base_num_outputs:]]) + return rules + + return strategy + + +def auto_register_op_variants() -> None: + """Register schema-discovered variants for existing single-dim strategies. + + This keeps manual registrations focused on the semantic base op while + covering mechanically related ATen overloads (in-place, out, functional, + and foreach). + """ + from torch.distributed.tensor._api import DTensor + + propagator = DTensor._op_dispatcher.sharding_propagator + registry = propagator.op_single_dim_strategy_funcs + already_registered = set(registry) + + for base_op, info in list(registry.items()): + base_schema_info = propagator.op_to_schema_info_for_single_dim_strategy.get( + base_op + ) + + for inplace_op in _find_inplace_variant_overloads(base_op): + if inplace_op in already_registered: + continue + propagator.register_single_dim_op_strategy( + inplace_op, + _clone_strategy_info( + info, _make_same_schema_variant_strategy_fn(info.func, base_op) + ), + _clone_schema_info(base_schema_info), + ) + already_registered.add(inplace_op) + + for out_variant in _find_out_variant_overloads(base_op): + out_op, output_arg_names = out_variant + if out_op in already_registered: + continue + propagator.register_single_dim_op_strategy( + out_op, + _clone_strategy_info( + info, + _make_out_variant_strategy_fn( + info.func, base_op, out_op, output_arg_names + ), + ), + _clone_schema_info(base_schema_info), + ) + already_registered.add(out_op) + + for functional_op in _find_functional_variant_overloads(base_op): + if functional_op in already_registered: + continue + propagator.register_single_dim_op_strategy( + functional_op, + _clone_strategy_info( + info, _make_functional_variant_strategy_fn(info.func, base_op) + ), + _clone_schema_info(base_schema_info), + ) + already_registered.add(functional_op) + + for foreach_op in _find_foreach_variants(base_op): + if foreach_op in already_registered: + continue + propagator.register_single_dim_op_strategy( + foreach_op, + info, + RuntimeSchemaInfo(needs_pytree=True), + ) + already_registered.add(foreach_op) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/single_dim_strategy.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/single_dim_strategy.py new file mode 100644 index 0000000000000000000000000000000000000000..d5810c169e5c84acad26b2f8f6b39e4fb15925fb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/single_dim_strategy.py @@ -0,0 +1,1235 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import functools +import heapq +import logging +import math +from collections import defaultdict +from collections.abc import Callable, Sequence +from dataclasses import dataclass, field +from itertools import count +from typing import Any, cast, TypeAlias, TypeVar +from typing_extensions import TypeIs + +import torch +from torch._ops import OpOverload +from torch.distributed.tensor._collective_utils import ( + _compute_placement_transition_cost, + MeshTopoInfo, +) +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + ArgsType, + KwargsType, + OpSchema, + OpSpec, + OpStrategy, + RuntimeSchemaInfo, + StrategyType, + TupleStrategy, +) +from torch.distributed.tensor.device_mesh import DeviceMesh +from torch.distributed.tensor.placement_types import ( + _StridedShard, + Placement, + Replicate, + Shard, +) +from torch.utils._pytree import tree_map_only + + +logger = logging.getLogger(__name__) + + +def _is_sharding(p: Placement) -> TypeIs[Shard | _StridedShard]: + return isinstance(p, (Shard, _StridedShard)) + + +class _ShardingPlaceholder: + """ + A placeholder for a sharding placement that has a specified tensor dim, but the other + metadata (e.g. split factor if it's a StridedShard) will be filled in later. + """ + + dim: int + + def __init__(self, dim: int): + self.dim = dim + + def __repr__(self) -> str: + return f"_ShardingPlaceholder(dim={self.dim})" + + +_StrategyTypeT = TypeVar("_StrategyTypeT", bound=StrategyType) +_PlacementT = TypeVar("_PlacementT", bound=Placement) +_ShardingPlaceholderT = TypeVar("_ShardingPlaceholderT", bound=_ShardingPlaceholder) +_SingleDimStrategyFunc: TypeAlias = Callable[ + [OpOverload, ArgsType, KwargsType], list[list[_PlacementT | _ShardingPlaceholderT]] +] +_ExpandedSingleDimStrategyFunc: TypeAlias = Callable[ + [OpOverload, ArgsType, KwargsType], _StrategyTypeT +] +_FullMeshStrategyFilter: TypeAlias = Callable[ + [ + DeviceMesh, + OpSchema, + list[DTensorSpec], + DTensorSpec | tuple[DTensorSpec | None, ...], + ], + bool, +] + + +@dataclass +class _SingleDimStrategyInfo: + func: _SingleDimStrategyFunc + allow_unbacked_sharding: bool | None = field(default=None) + allow_uneven_sharding: bool = field(default=False) + full_mesh_strategy_filter: _FullMeshStrategyFilter | None = field(default=None) + # Positions (in args_schema) of args that may live on a different mesh + # than the op's compute mesh. These args must be Replicate. + # See Note [Multi-mesh args] in expand_to_full_mesh_op_strategy. + different_mesh_args: list[int] | None = field(default=None) + + # Delegate to func so this can be used interchangeably with a raw + # _SingleDimStrategyFunc (e.g. in tests that call strategy functions directly). + def __call__(self, *args, **kwargs): + return self.func(*args, **kwargs) + + +def _insert_single_dim_replication_strategy( + single_dim_strategies_with_placeholders: list[ + list[Placement | _ShardingPlaceholder | None] + ], + num_outputs: int, + num_input_tensors: int, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None = None, +) -> list[list[Placement | _ShardingPlaceholder | None]]: + """ + Inserts the [Replicate(), Replicate(), ...] strategy after asserting that such strategy does not yet exist. + For ops with masked-off outputs (e.g. backward ops with output_mask), output positions + where output_tensor_meta is None are set to None in the all-Replicate rule. + Output positions that are explicitly None in every strategy are also preserved + as None in the inserted rule. + """ + for strategy in single_dim_strategies_with_placeholders: + if all(isinstance(p, Replicate) or p is None for p in strategy): + return single_dim_strategies_with_placeholders + total_len = num_outputs + num_input_tensors + replicate_rule: list[Placement | _ShardingPlaceholder | None] = [ + Replicate() + ] * total_len + # Set None for masked-off output positions based on output_tensor_meta + if isinstance(output_tensor_meta, Sequence) and not isinstance( + output_tensor_meta, TensorMeta + ): + for i, meta in enumerate(output_tensor_meta): + if meta is None and i < num_outputs: + replicate_rule[i] = None + for i in range(num_outputs): + if single_dim_strategies_with_placeholders and all( + strategy[i] is None for strategy in single_dim_strategies_with_placeholders + ): + replicate_rule[i] = None + single_dim_strategies_with_placeholders.insert(0, replicate_rule) + return single_dim_strategies_with_placeholders + + +def _fill_single_dim_strategy_placeholders( + unique_input_placements: set[Placement], + single_dim_strategies_with_placeholders: list[ + list[Placement | _ShardingPlaceholder | None] + ], +) -> list[list[Placement | None]]: + """ + Replace any _ShardingPlaceholder with the specific Sharding types used by the inputs in op_schema. + Supports implicit replication. + + Example: + single_dim_strategies_with_placeholders = [[Partial(), _ShardingPlaceholder(1), _ShardingPlaceholder(0)]] + input0: Shard(0) + input1: StridedShard(1, split_factor=2) + returns: [ + [Partial(), Shard(1), Shard(0)], + [Partial(), StridedShard(1, split_factor=2), StridedShard(0, split_factor=2)], + [Replicate(), Replicate(), Replicate()] + ] + """ + shard_builders: dict[str, Callable[[int], Placement]] = {} + for placement in unique_input_placements: + if isinstance(placement, _StridedShard): + key = f"StridedShard(sf={placement.split_factor})" + if key not in shard_builders: + shard_builders[key] = functools.partial( + _StridedShard, split_factor=placement.split_factor + ) + elif isinstance(placement, Shard): + key = "Shard()" + if key not in shard_builders: + shard_builders[key] = lambda tensor_dim: Shard(tensor_dim) + + # if any of the placements is a placeholder, we need to expand the strategy + # to all possible combinations of placements + expanded_strategies_over_one_mesh_dim: list[list[Placement | None]] = [] + for s in single_dim_strategies_with_placeholders: + if any(isinstance(p, _ShardingPlaceholder) for p in s): + for shard_builder in shard_builders.values(): + expanded_strategy: list[Placement | None] = [] + for maybe_placeholder in s: + if isinstance(maybe_placeholder, _ShardingPlaceholder): + # we combine the tensor dim to shard from the placeholder + # with other metadata (e.g. split_factor) from the sharding class + expanded_strategy.append(shard_builder(maybe_placeholder.dim)) + elif maybe_placeholder is None: + expanded_strategy.append(None) + else: + if not isinstance(maybe_placeholder, Placement): + raise AssertionError + expanded_strategy.append(maybe_placeholder) + expanded_strategies_over_one_mesh_dim.append(expanded_strategy) + else: + if not all(isinstance(p, Placement) or p is None for p in s): + raise AssertionError + expanded_strategies_over_one_mesh_dim.append( + cast(list[Placement | None], (s)) + ) + + return expanded_strategies_over_one_mesh_dim + + +def _get_unique_placements(op_schema: OpSchema) -> set[Placement]: + unique_placements = set() + + def _update_placements(obj: Any): + if isinstance(obj, DTensorSpec): + unique_placements.update(obj.placements) + elif isinstance(obj, OpStrategy): + if len(obj.strategies) != 1: + raise AssertionError + unique_placements.update(obj.strategies[0].output_spec.placements) + elif isinstance(obj, TupleStrategy): + for child in obj.children: + _update_placements(child) + elif isinstance(obj, (list, tuple)): + for child in obj: + _update_placements(child) + + for obj in op_schema.args_schema: + _update_placements(obj) + # Also include placements from kwargs (e.g., "out" tensor) + for obj in op_schema.kwargs_schema.values(): + _update_placements(obj) + + return unique_placements + + +def _get_num_tensor_inputs(op_schema: OpSchema) -> int: + num_inputs = 0 + + def _count(obj: Any) -> int: + if isinstance(obj, OpStrategy): + return 1 + elif isinstance(obj, TupleStrategy): + return sum(1 for child in obj.children if child is not None) + elif isinstance(obj, (list, tuple)): + return sum(_count(child) for child in obj) + return 0 + + for obj in op_schema.args_schema: + num_inputs += _count(obj) + # Also count tensor kwargs (e.g., "out" for out-variant ops) + for obj in op_schema.kwargs_schema.values(): + num_inputs += _count(obj) + return num_inputs + + +def _build_output_specs( + mesh: DeviceMesh, + per_mesh_dim_placements: list[tuple[Placement | None, ...]], + num_outputs: int, + output_metas: tuple[TensorMeta | None, ...], +) -> DTensorSpec | tuple[DTensorSpec | None, ...]: + """Build output spec(s) by transposing per-mesh-dim placements to per-output. + + per_mesh_dim_placements is indexed [mesh_dim][output_idx]. output_metas must + have exactly num_outputs elements. Outputs where output_metas[i] is None + (masked-off outputs) produce None specs. + """ + if num_outputs <= 0: + raise AssertionError(f"Expected num_outputs > 0, got {num_outputs}") + if len(output_metas) != num_outputs: + raise AssertionError( + f"Expected {num_outputs} output_metas, got {len(output_metas)}" + ) + + def _spec_for_output(out_idx: int) -> DTensorSpec | None: + maybe_placements = tuple(out[out_idx] for out in per_mesh_dim_placements) + # All mesh dims agreeing on None means the whole output has no + # DTensorSpec, e.g. local scalar SDPA philox outputs. + if all(placement is None for placement in maybe_placements): + return None + if any(placement is None for placement in maybe_placements): + raise AssertionError( + f"Output {out_idx} mixes None with placements across mesh dims: " + f"{maybe_placements}" + ) + if output_metas[out_idx] is None: + return None + placements = tuple(cast(Placement, placement) for placement in maybe_placements) + return DTensorSpec(mesh, placements, tensor_meta=output_metas[out_idx]) + + if num_outputs > 1: + return tuple(_spec_for_output(i) for i in range(num_outputs)) + else: + spec = _spec_for_output(0) + if spec is None: + raise AssertionError("Single-output op cannot have None output meta") + return spec + + +class _PreparedSingleDimStrategy: + """A single-dim strategy materialized for a specific op. + + Expands a strategy function's placeholder-based rules into concrete + placement rules by filling in the actual shard/partial placements from + the op_schema. The result is a lookup table (strategy_lookup) that maps + input placements to output placements for one mesh dimension. + + Provides try_propagate() for matching a multi-dim placement tuple against + the per-dim rules, and exposes allowed_sharding_per_input / + allowed_partial_per_input for graph search neighbor generation. + """ + + strategy_lookup: dict[tuple[Placement | None, ...], tuple[Placement | None, ...]] + expanded_strategies: list[list[Placement | None]] + num_outputs: int + num_inputs: int + output_metas: tuple[TensorMeta | None, ...] + allowed_sharding_per_input: dict[int, set[Shard | _StridedShard]] + allowed_partial_per_input: dict[int, set[Placement]] + allow_unbacked_sharding: bool | None + + # many, but not all ops are able to support unevenly sharded tensors + # there are existing BC expectations even if we wanted to ban for + # simplicity, see why justification for why pointwise_ops always work + # with uneven sharding at + # https://github.com/pytorch/pytorch/pull/174874#issuecomment-3995152777 + allow_uneven_sharding: bool + + def __init__( + self, + strategy_fn: _SingleDimStrategyInfo + | Callable[ + [OpOverload, ArgsType, KwargsType], + list[list[Placement | _ShardingPlaceholder]], + ], + op_schema: OpSchema, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, + num_inputs: int | None = None, + ) -> None: + # Note: circular import + from torch.distributed.tensor.placement_types import Partial + + if isinstance(strategy_fn, _SingleDimStrategyInfo): + self.allow_unbacked_sharding = strategy_fn.allow_unbacked_sharding + self.allow_uneven_sharding = strategy_fn.allow_uneven_sharding + self.full_mesh_strategy_filter = strategy_fn.full_mesh_strategy_filter + different_mesh_args = strategy_fn.different_mesh_args + func = strategy_fn.func + else: + self.allow_unbacked_sharding = None + self.allow_uneven_sharding = False + self.full_mesh_strategy_filter = None + different_mesh_args = None + func = strategy_fn + + # Determine element_mesh from the first OpStrategy arg. For foreach + # per-element schemas the element's inputs may live on a smaller + # sub-mesh than the global compute_mesh. + self.element_mesh: DeviceMesh | None = None + for arg in op_schema.args_schema: + if isinstance(arg, OpStrategy): + self.element_mesh = arg.strategies[0].output_spec.mesh + break + + # Validate that all inputs are on the same mesh (except + # different_mesh_args which are explicitly allowed to differ). + if self.element_mesh is not None: + allowed = set(different_mesh_args or []) + for i, arg in enumerate(op_schema.args_schema): + if isinstance(arg, OpStrategy) and i not in allowed: + arg_mesh = arg.strategies[0].output_spec.mesh + if arg_mesh != self.element_mesh: + raise ValueError( + f"Cannot run {op_schema.op} on inputs with different " + f"meshes: got {self.element_mesh} and {arg_mesh}" + ) + + # Remap different_mesh_args from args_schema positions to + # OpStrategy-only positions. Non-OpStrategy args (e.g. empty lists) + # are filtered out by expand_to_full_mesh_op_strategy, shifting later + # indices. + self.remapped_different_mesh_args: list[int] | None = None + if different_mesh_args is not None: + schema_to_strategy: dict[int, int] = {} + strategy_pos = 0 + for schema_pos, arg in enumerate(op_schema.args_schema): + if isinstance(arg, OpStrategy): + schema_to_strategy[schema_pos] = strategy_pos + strategy_pos += 1 + self.remapped_different_mesh_args = [ + schema_to_strategy[i] + for i in different_mesh_args + if i in schema_to_strategy + ] + + if num_inputs is None: + num_inputs = _get_num_tensor_inputs(op_schema) + self.num_inputs = num_inputs + + # Strategy functions may return None in output positions for masked-off + # outputs (e.g. backward ops with output_mask). Widen the type here. + strategies_with_placeholders = cast( + list[list[Placement | _ShardingPlaceholder | None]], + func(op_schema.op, op_schema.args_meta, op_schema.kwargs_meta), + ) + + # Validate strategy length against the op outputs. Tuple outputs are + # aligned with output_tensor_meta so mixed Tensor/non-Tensor returns + # keep their None slots for wrapping the real result tuple. + if len(strategies_with_placeholders) > 0: + if isinstance(output_tensor_meta, Sequence) and not isinstance( + output_tensor_meta, TensorMeta + ): + schema_num_outputs = len(output_tensor_meta) + elif any( + isinstance(r.type, torch.ListType) for r in op_schema.op._schema.returns + ): + schema_num_outputs = len(strategies_with_placeholders[0]) - num_inputs + else: + schema_num_outputs = sum( + 1 for r in op_schema.op._schema.returns if "Tensor" in str(r.type) + ) + expected_len = schema_num_outputs + num_inputs + actual_len = len(strategies_with_placeholders[0]) + if actual_len != expected_len: + raise AssertionError( + f"Strategy length {actual_len} != expected {expected_len} " + f"(schema_outputs={schema_num_outputs} + inputs={num_inputs}) " + f"for {op_schema.op}. Strategies must include placements " + f"for all outputs, args, and tensor kwargs." + ) + + # Compute num_outputs from strategy structure or output_tensor_meta + if len(strategies_with_placeholders) > 0: + num_outputs = len(strategies_with_placeholders[0]) - num_inputs + elif output_tensor_meta is None: + num_outputs = 0 + elif isinstance(output_tensor_meta, TensorMeta): + num_outputs = 1 + else: + num_outputs = len(output_tensor_meta) + self.num_outputs = num_outputs + + strategies_with_placeholders = _insert_single_dim_replication_strategy( + strategies_with_placeholders, + num_outputs, + num_inputs, + output_tensor_meta, + ) + + unique_input_placements = _get_unique_placements(op_schema) + self.expanded_strategies = _fill_single_dim_strategy_placeholders( + unique_input_placements, strategies_with_placeholders + ) + + # Build strategy lookup: map input placements -> output placements + self.strategy_lookup = {} + for strategy in self.expanded_strategies: + input_key = tuple(strategy[num_outputs:]) + if input_key not in self.strategy_lookup: + self.strategy_lookup[input_key] = tuple(strategy[:num_outputs]) + + # Precompute allowed placements per input from the expanded rules + self.allowed_sharding_per_input: dict[int, set[Shard | _StridedShard]] = ( + defaultdict(set) + ) + self.allowed_partial_per_input: dict[int, set[Placement]] = defaultdict(set) + for strategy in self.expanded_strategies: + for input_idx in range(num_inputs): + p = strategy[num_outputs + input_idx] + if p is None: + continue + if _is_sharding(p): + self.allowed_sharding_per_input[input_idx].add(p) + elif isinstance(p, Partial): + self.allowed_partial_per_input[input_idx].add(p) + + # Resolve output tensor_meta per output index + if output_tensor_meta is None: + self.output_metas = (None,) * max(num_outputs, 0) + elif isinstance(output_tensor_meta, TensorMeta): + self.output_metas = (output_tensor_meta,) + else: + self.output_metas = tuple(output_tensor_meta) + + def try_propagate( + self, + mesh: DeviceMesh, + input_placements: tuple[tuple[Placement, ...], ...], + input_specs: list[DTensorSpec], + ) -> OpStrategy | None: + """Try to match input placements against single-dim strategy rules on every mesh dim. + + Checks whether the given input placements independently match a rule in + strategy_lookup on each mesh dimension, and that all inputs are shardable + with those placements. If so, returns an OpStrategy with the matched output + placements and zero redistribute costs. + """ + from torch.distributed.tensor._ops.utils import is_tensor_shardable + + selected_output_placements: list[tuple[Placement | None, ...]] = [] + for mesh_dim in range(mesh.ndim): + input_placements_for_dim = tuple( + placements[mesh_dim] for placements in input_placements + ) + output_for_dim = self.strategy_lookup.get(input_placements_for_dim) + if output_for_dim is not None: + selected_output_placements.append(output_for_dim) + else: + return None + + arg_specs = [ + DTensorSpec(mesh, placements, tensor_meta=input_spec.tensor_meta) + for placements, input_spec in zip(input_placements, input_specs) + ] + if not all( + is_tensor_shardable( + spec.tensor_meta.shape, + spec, + allow_unbacked_sharding=self.allow_unbacked_sharding, + ) + or (self.allow_uneven_sharding and input_spec.placements == spec.placements) + for spec, input_spec in zip(arg_specs, input_specs) + if spec.tensor_meta is not None + ): + return None + + output_spec = ( + _build_output_specs( + mesh, + selected_output_placements, + self.num_outputs, + self.output_metas, + ) + if self.num_outputs > 0 + else None + ) + return OpStrategy( + [ + OpSpec( + output_specs=output_spec, + input_specs=arg_specs, + redistribute_cost=[[0.0] for _ in input_specs], + ) + ] + ) + + +def _has_default_value(arg: Any) -> bool: + """Return whether a schema argument can be omitted at runtime. + + Foreach schemas sometimes omit trailing scalar args that exist on the scalar + op only because they have defaults. + """ + has_default_value = getattr(arg, "has_default_value", None) + if callable(has_default_value): + return bool(has_default_value()) + return getattr(arg, "default_value", None) is not None + + +def _normalize_schema_type(arg: Any, *, list_to_element: bool = False) -> str: + """Normalize schema type strings for direct and foreach element matching. + + In foreach matching, a `List[Tensor]` argument represents one per-element + `Tensor` argument to the scalar strategy function. + """ + type_str = str(arg.type) + if list_to_element and type_str.startswith("List[") and type_str.endswith("]"): + return type_str.removeprefix("List[").removesuffix("]") + return type_str + + +def _schema_args_match( + base_args: Sequence[Any], + candidate_args: Sequence[Any], + *, + candidate_lists_are_elements: bool = False, + allow_base_trailing_defaults: bool = False, + allow_kwonly_scalar_defaults: bool = False, +) -> bool: + """Return whether candidate args can safely reuse the base op strategy. + + This is intentionally schema-based so auto-registration does not rely only + on overload names. + """ + if len(candidate_args) > len(base_args): + return False + + for base_arg, candidate_arg in zip(base_args, candidate_args): + if _normalize_schema_type(base_arg) != _normalize_schema_type( + candidate_arg, list_to_element=candidate_lists_are_elements + ): + return False + if base_arg.kwarg_only != candidate_arg.kwarg_only: + base_arg_type = _normalize_schema_type(base_arg) + if not ( + allow_kwonly_scalar_defaults + and base_arg.kwarg_only + and not candidate_arg.kwarg_only + and base_arg_type == "number" + and _has_default_value(base_arg) + ): + return False + + if len(candidate_args) == len(base_args): + return True + + if not allow_base_trailing_defaults: + return False + + return all(_has_default_value(arg) for arg in base_args[len(candidate_args) :]) + + +def _op_namespace_and_base_name(op: OpOverload) -> tuple[str, str]: + """Split an overload into namespace and overload-agnostic base op name.""" + namespace, op_name = op.name().split("::", 1) + return namespace, op_name.split(".", 1)[0] + + +def _get_overload_packet(namespace: str, op_name: str) -> Any | None: + """Look up torch.ops.. without raising on misses.""" + namespace_packet = getattr(torch.ops, namespace, None) + if namespace_packet is None: + return None + try: + return getattr(namespace_packet, op_name) + except AttributeError: + return None + + +def _get_packet_overload(packet: Any, overload_name: str) -> OpOverload | None: + """Look up one overload from an overload packet without raising on misses.""" + if overload_name not in packet.overloads(): + return None + return cast(OpOverload, getattr(packet, overload_name)) + + +def _resolve_foreach_elementwise_overload(foreach_op: OpOverload) -> OpOverload | None: + """Map an elementwise foreach overload to the scalar op for one element. + + Single-dim strategy functions reason about one Tensor at a time; foreach + expansion calls the scalar strategy once for each list element. + """ + namespace, foreach_base_name = _op_namespace_and_base_name(foreach_op) + if namespace != "aten": + return None + if foreach_base_name.startswith("_foreach_"): + base_op_name = foreach_base_name.removeprefix("_foreach_") + elif foreach_base_name.startswith("_amp_foreach_"): + base_op_name = foreach_base_name.removeprefix("_amp_foreach_") + else: + return None + + base_op_name = base_op_name.removesuffix("_") + base_packet = _get_overload_packet(namespace, base_op_name) + if base_packet is None: + return None + + foreach_args = foreach_op._schema.arguments + matches: list[tuple[int, OpOverload]] = [] + for overload_name in base_packet.overloads(): + if "out" in overload_name: + continue + candidate = _get_packet_overload(base_packet, overload_name) + if candidate is None or candidate._schema.is_mutable: + continue + if _schema_args_match( + candidate._schema.arguments, + foreach_args, + candidate_lists_are_elements=True, + allow_base_trailing_defaults=True, + allow_kwonly_scalar_defaults=True, + ): + # Prefer exact schema matches over matches that rely on defaulted + # trailing args, e.g. add.Tensor over add.Scalar when possible. + score = len(candidate._schema.arguments) - len(foreach_args) + matches.append((score, candidate)) + + if not matches: + return None + return min(matches, key=lambda item: item[0])[1] + + +def _expand_single_dim_strategy_to_mesh( + mesh: DeviceMesh, + op_schema: OpSchema, + strategy_info: _SingleDimStrategyInfo, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, +) -> _ExpandedSingleDimStrategyFunc: + """ + Expands the single_mesh_dim impl across all mesh dims, and expands ShardingPlaceholder into all + sharding types used by inputs. + + This supports functional correctness but will generate all possible combinations, which is prohibitively expensive + for larger numbers of mesh dimensions. + + The expanded_strategy function accesses both the args_schema/kwargs_schema, which contains TensorMeta in place of + tensor arguments, but also the op_schema which contains OpStrategy in place of Tensor args. + + Args: + output_tensor_meta: tensor metadata for the output(s), precomputed during sharding prop + """ + # Note: circular import, failed to untangle with #168221, reverted + from torch.distributed.tensor._ops.utils import expand_to_full_mesh_op_strategy + + def _create_expanded_strategy_impl( + op_schema: OpSchema, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, + ) -> Callable[[OpOverload, ArgsType, KwargsType], StrategyType]: + def expanded_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType + ) -> StrategyType: + prepared_strategy = _PreparedSingleDimStrategy( + strategy_info, op_schema, output_tensor_meta + ) + + # Detect inplace ops by checking if the base op name ends with '_' + op_name = op.name() + base_name = op_name.split("::")[1].split(".")[0] + is_inplace = base_name.endswith("_") + + element_mesh = prepared_strategy.element_mesh or mesh + full_mesh_strategy_filter = None + if prepared_strategy.full_mesh_strategy_filter is not None: + + def full_mesh_strategy_filter( + input_specs: list[DTensorSpec], + output_specs: DTensorSpec | tuple[DTensorSpec | None, ...], + ) -> bool: + if prepared_strategy.full_mesh_strategy_filter is None: + raise AssertionError + return prepared_strategy.full_mesh_strategy_filter( + element_mesh, op_schema, input_specs, output_specs + ) + + return expand_to_full_mesh_op_strategy( + element_mesh, + op_schema, + prepared_strategy.expanded_strategies, + output_tensor_meta=output_tensor_meta, + inplace_op=is_inplace, + input_index=prepared_strategy.num_outputs, + allow_unbacked_sharding=prepared_strategy.allow_unbacked_sharding, + allow_uneven_sharding=prepared_strategy.allow_uneven_sharding, + full_mesh_strategy_filter=full_mesh_strategy_filter, + different_mesh_args=prepared_strategy.remapped_different_mesh_args, + ) + + return expanded_strategy + + # Create a cached version of the impl + _cached_create_expanded_strategy = functools.lru_cache( + _create_expanded_strategy_impl + ) + + def _create_expanded_strategy( + op_schema: OpSchema, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, + ) -> Callable[[OpOverload, ArgsType, KwargsType], StrategyType]: + # Try to use cache, but fall back to uncached version if hashing fails + # (e.g., when TensorMeta contains SymInts from dynamic shapes) + try: + return _cached_create_expanded_strategy(op_schema, output_tensor_meta) + except TypeError: + # Unhashable types (SymInts), skip caching + return _create_expanded_strategy_impl(op_schema, output_tensor_meta) + + def _translate_list_op_schema( + op_schema: OpSchema, + output_tensor_meta: Sequence[TensorMeta] | None, + index: int, + ) -> tuple[OpSchema, TensorMeta | None]: + """Translate foreach/fused op to per-element version of schema.""" + # select per-element inputs, outputs + target_args, target_kwargs = tree_map_only( + TupleStrategy, + lambda x: x.children[index], + (op_schema.args_schema, op_schema.kwargs_schema), + is_leaf=lambda x: isinstance(x, TupleStrategy), + ) + # For inplace ops, output_tensor_meta is None + target_output_meta = ( + output_tensor_meta[index] if output_tensor_meta is not None else None + ) + + target_op = _resolve_foreach_elementwise_overload(op_schema.op) + if target_op is None: + # Fused ops or unknown foreach-like ops: keep original op, no translation. + target_op = op_schema.op + + op_schema = OpSchema( + target_op, # type: ignore[arg-type] + args_schema=tuple(target_args), + kwargs_schema=op_schema.kwargs_schema, + ) + return op_schema, target_output_meta + + def expanded_foreach_strategy( + op: OpOverload, args_schema: ArgsType, kwargs_schema: KwargsType + ) -> StrategyType: + tensorlist_len: int | None = None + for i, obj in enumerate(op_schema.args_schema): + if isinstance(obj, TupleStrategy): + if tensorlist_len is None: + tensorlist_len = len(obj.children) + elif len(obj.children) != tensorlist_len: + raise AssertionError( + f"Expected {tensorlist_len} children in index {i}, but found {len(obj.children)}." + ) + + if tensorlist_len is None: + raise AssertionError("Must have at least one tuple input to a foreach op") + + child_strategies: list[StrategyType] = [] + for tensorlist_i in range(tensorlist_len): + per_index_schema, per_index_output_meta = _translate_list_op_schema( + op_schema, + output_tensor_meta, # type: ignore[arg-type] + tensorlist_i, + ) + per_index_strategy = _create_expanded_strategy( + per_index_schema, per_index_output_meta + ) + child_strategies.append( + per_index_strategy( + op, per_index_schema.args_meta, per_index_schema.kwargs_meta + ) + ) + + return TupleStrategy(children=child_strategies) + + # TODO maybe this could be helped by adding a new 'tag' to the OpOverload? + # Only use the foreach path if the op has TupleStrategy inputs (i.e., actual + # list-of-tensor args). The name prefix alone is insufficient because ops like + # _fused_rms_norm share the "_fused_" prefix but are not foreach/fused-optimizer ops. + op_name = op_schema.op.name() + has_tuple_strategy = any( + isinstance(arg, TupleStrategy) for arg in op_schema.args_schema + ) + if has_tuple_strategy and op_name.startswith( + ("aten::_foreach_", "aten::_amp_foreach_", "aten::_fused_") + ): + return expanded_foreach_strategy + + return _create_expanded_strategy(op_schema, output_tensor_meta) + + +def register_single_dim_strategy( + op: torch._ops.OpOverload | list[torch._ops.OpOverload], + schema_info: RuntimeSchemaInfo | None = None, + allow_unbacked_sharding: bool | None = None, + allow_uneven_sharding: bool = False, + full_mesh_strategy_filter: _FullMeshStrategyFilter | None = None, + different_mesh_args: list[int] | None = None, +) -> Callable[[_SingleDimStrategyFunc], _SingleDimStrategyFunc]: + """ + Registers a single_dim_strategy function for the given op. + + A single_dim_strategy enumerates all the non-trivial sharding specifications for the operator over a single mesh + dim. Since it generates the full set of valid shardings regardless of the given input placements, tensor inputs + are represented via TensorMetas instead of OpStrategies like in op_strategy functions. TensorMeta inputs, along + with all other op inputs (int, float, bool, etc) inform which sharding placements are valid. Single-dim strategies + are fed into infra that expands them over multiple mesh dims and fills sharding placeholders with concrete sharding + types. + + Single-dim-strategies should not list the trivial "all Replicate" rule, which is assumed for all ops. + + Sharding placeholders should be used to represent generic sharding rules in single_dim_strategies. For example, + most operators that support sharded inputs and outputs do not care how the shards are laid out globally, as long as + they are consistent. It is just as valid to run a matmul on (Shard(0), Replicate -> Shard(0)) as on + (StridedShard(0), Replicate -> StridedShard(0)). For this reason, we use a 'ShardingPlaceholder' to represent all + generic sharding types. Placeholders will be filled by concrete sharding types seen in runtime inputs, not all + types known to DTensor. + + """ + # Note: circular import, failed to untangle with #168221, reverted + from torch.distributed.tensor._api import DTensor + from torch.distributed.tensor._ops.utils import _get_registration_wrapper + + # For every ATen op that accepts any args in this list, + # the arg itself can impact the strides (and potentially the sharding strategy) + # of the output tensor. + # thus, we will detect ATen schemas with any of these args and ensure + # that they get specialized here. + arg_names_that_require_specializing_cache_strategy = [ + "memory_format", + ] + registration_wrapper = _get_registration_wrapper( + DTensor._op_dispatcher.sharding_propagator.register_single_dim_op_strategy, + op, + schema_info, + arg_names_that_require_specializing_cache_strategy, + ) + + # Wrap impl in _SingleDimStrategyInfo here rather than adding a generic + # transform hook to _get_registration_wrapper, so that single-dim-strategy + # concerns stay in this module and the shared registration util stays simple. + def wrapper(impl): + info = _SingleDimStrategyInfo( + func=impl, + allow_unbacked_sharding=allow_unbacked_sharding, + allow_uneven_sharding=allow_uneven_sharding, + full_mesh_strategy_filter=full_mesh_strategy_filter, + different_mesh_args=different_mesh_args, + ) + registration_wrapper(info) + return impl + + return wrapper + + +@dataclass(order=True) +class _PQEntry: + """Priority queue entry for the Dijkstra search in _dijkstra_expand_single_dim_strategy_to_mesh. + + Ordered by (cost, counter) for heap comparison. The counter breaks ties + in FIFO order so that entries with equal cost are explored in insertion + order rather than by arbitrary tuple comparison on placements. + """ + + cost: float + counter: int + # Per-input placement tuples representing the current search state. + placements: tuple[tuple[Placement, ...], ...] = field(compare=False) + # History of (input_idx, mesh_dim, old_placement, new_placement) transitions + # from the initial state to this state, used for debugging. + transitions: list[tuple[int, int, Placement, Placement]] = field(compare=False) + # Accumulated redistribute cost per input (sum of incremental step costs). + per_input_costs: tuple[float, ...] = field(compare=False) + # Current communication bytes (in GB) per input, updated as placements change. + per_input_comm_bytes_gb: tuple[float, ...] = field(compare=False) + + +def _get_neighbor_placements( + allowed_sharding: set[Shard | _StridedShard], + allowed_partial: set[Placement], + current: Placement, + input_placements: tuple[Placement, ...], + mesh_dim: int, +) -> list[Placement]: + """Return valid one-shot placement transitions for one input on one mesh dim. + + DTensor placements are applied left-to-right, so a tensor dim sharded on + multiple mesh dims has a specific nesting order. A one-shot collective on + mesh_dim M can only produce the correct data layout if no mesh dim to the + RIGHT of M already shards the same tensor dim. For example, going from + (R, S(0)) to (S(0), S(0)) via local chunk on mesh dim 0 produces a + strided-shard layout, not the correct left-to-right (S(0), S(0)) layout. + + Transition rules: + - Replicate -> Shard(d): free local chunk, valid if d not sharded to the right + - Replicate -> Partial: local view, always valid + - Shard(d) -> Replicate: allgather, valid if d not sharded to the right + - Shard(d1) -> Shard(d2): all-to-all, valid if neither d1 nor d2 sharded right + - Partial -> Replicate: allreduce, always valid + - Partial -> Shard(d): reduce-scatter, valid if d not sharded to the right + + Shard -> Partial("sum") is intentionally absent. Although explicit + redistribution supports it, the conversion materializes a global-shape + local tensor. Because the strategy cost model accounts only for communication, + we exclude this transition to avoid selecting it without accounting for its + memory and copy costs. + """ + # Note: circular import + from torch.distributed.tensor.placement_types import Partial + + # Tensor dims sharded by mesh dims to the right of this one. + right_shard_dims: set[int] = set() + for i in range(mesh_dim + 1, len(input_placements)): + p = input_placements[i] + if _is_sharding(p): + right_shard_dims.add(p.dim) + + neighbors: list[Placement] = [] + + if isinstance(current, Replicate): + neighbors.extend(s for s in allowed_sharding if s.dim not in right_shard_dims) + neighbors.extend(allowed_partial) + + elif _is_sharding(current): + cur_dim_ok = current.dim not in right_shard_dims + if cur_dim_ok: + neighbors.append(Replicate()) + for s in allowed_sharding: + if s != current and cur_dim_ok and s.dim not in right_shard_dims: + neighbors.append(s) + + elif isinstance(current, Partial): + neighbors.append(Replicate()) + neighbors.extend(s for s in allowed_sharding if s.dim not in right_shard_dims) + + return neighbors + + +def _dijkstra_expand_single_dim_strategy_to_mesh( + mesh: DeviceMesh, + op_schema: OpSchema, + single_dim_strategy: _SingleDimStrategyInfo + | Callable[ + [OpOverload, ArgsType, KwargsType], list[list[Placement | _ShardingPlaceholder]] + ], + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None = None, + _collect_all_matches: set[tuple[tuple[Placement, ...], ...]] | None = None, +) -> OpStrategy | None: + """ + Find the lowest cost sharding for the given op_schema. + + Uses a Dijkstra-like priority-queue search over input placement states. Each + state is a tuple of per-input placement tuples, and neighbors are generated + by changing one placement on one mesh dim for one input. The search + terminates when a state matches a single-dim strategy on every mesh dim. + + This avoids the O(S^N) exhaustive expansion of _expand_single_dim_strategy_to_mesh + (S = single-dim strategies, N = mesh dims). Benchmarks with mm on fake + process groups show: + + 1D(4): S^N=8, avg 0.2ms + 2D(2,2): S^N=64, avg 2.3ms + 3D(2,2,2): S^N=512, avg 41ms, worst 392ms + + The step count is small (avg 0.6-2.0 pops) but per-step cost is dominated + by cost computation. Each transition computes an incremental cost via + _compute_placement_transition_cost for the single changed placement, matching + the per-step costs used by graph-based transform info planning. + + Returns None if any input has _StridedShard placement, signaling the caller + to fall back to full expansion. + + Args: + _collect_all_matches: Testing-only. When non-None, exhaustively explores the + full transition graph, adding every shardable match to the set. Still + returns the optimal (first) match. + """ + # Extract input DTensorSpecs from OpStrategy-wrapped args. + # Fall back for TupleStrategy (e.g. index tensors in index_put) since the PQ + # search doesn't model variable-length tuple inputs. + input_specs: list[DTensorSpec] = [] + for arg in op_schema.args_schema: + if isinstance(arg, OpStrategy): + if len(arg.strategies) != 1: + raise AssertionError + input_specs.append(arg.strategies[0].output_spec) + elif isinstance(arg, TupleStrategy): + return None + + # Fall back if any kwargs are tensor inputs — the PQ search only tracks + # positional tensor args and would miss redistribute costs for kwargs. + for kwarg in op_schema.kwargs_schema.values(): + if isinstance(kwarg, (OpStrategy, TupleStrategy)): + return None + + if len(input_specs) == 0: + raise AssertionError("broken input") + num_inputs = len(input_specs) + + # Fall back to full expansion if any input has _StridedShard or symbolic shapes + # (symbolic shapes produce SymFloat costs that can't be compared in the PQ) + for spec in input_specs: + if any(isinstance(p, _StridedShard) for p in spec.placements): + return None + if spec.tensor_meta is not None and any( + isinstance(s, torch.SymInt) for s in spec.tensor_meta.shape + ): + return None + + prepared_strategy = _PreparedSingleDimStrategy( + single_dim_strategy, op_schema, output_tensor_meta, num_inputs=num_inputs + ) + + initial_placements = tuple(spec.placements for spec in input_specs) + first_result: OpStrategy | None = None + + # Fast path: if initial placements already match a strategy, skip search + fast_result = prepared_strategy.try_propagate(mesh, initial_placements, input_specs) + if fast_result is not None: + fast_result._pq_transitions = [] # type: ignore[attr-defined] + if _collect_all_matches is not None: + _collect_all_matches.add(initial_placements) + first_result = fast_result + else: + return fast_result + + # Pre-compute mesh topology and per-input comm bytes for cost computation. + # comm_bytes_gb reflects the local shard size given current placements; + # it's tracked per PQ entry and updated as placements change. + mesh_topo = MeshTopoInfo.build_from_mesh(mesh) + initial_comm_bytes_gb: list[float] = [] + for spec in input_specs: + if spec.tensor_meta is None: + raise AssertionError + total_bytes = spec.tensor_meta.dtype.itemsize * math.prod( + spec.tensor_meta.shape + ) + # TODO: is_shard() misses _StridedShard, use spec.num_shards instead. + # Not fixing yet: the overestimate biases Dijkstra toward redistributing + # away from _StridedShard, which is the safer default until _StridedShard + # is fully validated. + num_shards = 1 + for i, p in enumerate(spec.placements): + if p.is_shard(): + num_shards *= mesh_topo.mesh_dim_devices[i] + initial_comm_bytes_gb.append(total_bytes / num_shards / (1024**3)) + + pq: list[_PQEntry] = [] + visited: set[tuple[tuple[Placement, ...], ...]] = set() + next_counter = count() + + initial_per_input_costs = (0.0,) * num_inputs + initial_per_input_comm_bytes = tuple(initial_comm_bytes_gb) + heapq.heappush( + pq, + _PQEntry( + 0.0, + next(next_counter), + initial_placements, + [], + initial_per_input_costs, + initial_per_input_comm_bytes, + ), + ) + + def _push_neighbor( + input_idx: int, + mesh_dim: int, + new_placement: Placement, + source: _PQEntry, + ) -> None: + new_input_placements = [list(ps) for ps in source.placements] + old_placement = new_input_placements[input_idx][mesh_dim] + new_input_placements[input_idx][mesh_dim] = new_placement + candidate_placements = tuple(tuple(ps) for ps in new_input_placements) + if candidate_placements in visited: + return + # Check that the NET transition (original -> proposed) is feasible. + # Individual hops may each be valid (e.g. S->R then R->P) while the + # net redistribution (S->P) is unsupported by the runtime planner. + original_p = initial_placements[input_idx][mesh_dim] + net_cost, _ = _compute_placement_transition_cost( + original_p, + new_placement, + mesh_topo, + mesh_dim, + initial_comm_bytes_gb[input_idx], + ) + if net_cost == float("inf"): + return + step_cost, new_comm_bytes = _compute_placement_transition_cost( + old_placement, + new_placement, + mesh_topo, + mesh_dim, + source.per_input_comm_bytes_gb[input_idx], + ) + if step_cost == float("inf"): + return + changed_cost = source.per_input_costs[input_idx] + step_cost + new_per_input_costs = ( + source.per_input_costs[:input_idx] + + (changed_cost,) + + source.per_input_costs[input_idx + 1 :] + ) + new_per_input_comm_bytes = ( + source.per_input_comm_bytes_gb[:input_idx] + + (new_comm_bytes,) + + source.per_input_comm_bytes_gb[input_idx + 1 :] + ) + new_cost = sum(new_per_input_costs) + new_transitions = source.transitions + [ + (input_idx, mesh_dim, old_placement, new_placement) + ] + heapq.heappush( + pq, + _PQEntry( + new_cost, + next(next_counter), + candidate_placements, + new_transitions, + new_per_input_costs, + new_per_input_comm_bytes, + ), + ) + + while pq: + candidate = heapq.heappop(pq) + + if candidate.placements in visited: + continue + visited.add(candidate.placements) + + match_result = prepared_strategy.try_propagate( + mesh, candidate.placements, input_specs + ) + if match_result is not None: + # Use pre-computed per-input costs from the PQ search instead of + # recomputing via generate_redistribute_costs -> _gen_transform_infos. + match_spec = match_result.strategies[0] + if match_spec.input_specs is None: + raise AssertionError + op_spec = OpSpec( + output_specs=match_spec.output_specs, + input_specs=list(match_spec.input_specs), + redistribute_cost=[[cost] for cost in candidate.per_input_costs], + ) + + exhaustive = len(prepared_strategy.expanded_strategies) ** mesh.ndim + logger.debug( + "returning cost=%f %s, visited=%d, exhaustive=%d, transitions=%s", + candidate.cost, + op_spec, + len(visited), + exhaustive, + candidate.transitions, + ) + result = OpStrategy([op_spec]) + result._pq_transitions = candidate.transitions # type: ignore[attr-defined] + if _collect_all_matches is not None: + _collect_all_matches.add(candidate.placements) + if first_result is None: + first_result = result + else: + return result + + # Generate neighbor states + for mesh_dim in range(mesh.ndim): + for input_idx in range(len(candidate.placements)): + current_p = candidate.placements[input_idx][mesh_dim] + for neighbor_p in _get_neighbor_placements( + prepared_strategy.allowed_sharding_per_input[input_idx], + prepared_strategy.allowed_partial_per_input[input_idx], + current_p, + candidate.placements[input_idx], + mesh_dim, + ): + _push_neighbor(input_idx, mesh_dim, neighbor_p, candidate) + + if _collect_all_matches is not None and first_result is not None: + return first_result + + logger.warning( + "Dijkstra search exhausted without finding a valid strategy for " + "%s on %s (explored %d combinations); falling back to full expansion", + op_schema, + mesh, + len(visited), + ) + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/strategy_validation.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/strategy_validation.py new file mode 100644 index 0000000000000000000000000000000000000000..3d94de2d1455d7f82a13818bbeeba25a823c9198 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/strategy_validation.py @@ -0,0 +1,2304 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +""" +Strategy validation for DTensor sharding rules. + +This module provides utilities to validate DTensor's sharding strategies by: +1. Running operators on full tensors to get ground truth +2. Simulating sharding with various placement combinations +3. Comparing redistributed outputs against ground truth +4. Reporting incorrect rules (DTensor claims valid but wrong) and + missing rules (ground truth valid but DTensor has no rule) + +Run as a module to compare DTensor rules against ground truth: + python -m torch.distributed.tensor._ops.strategy_validation --op add + python -m torch.distributed.tensor._ops.strategy_validation --op relu,mul,add + python -m torch.distributed.tensor._ops.strategy_validation --op "nn.functional.*" + python -m torch.distributed.tensor._ops.strategy_validation --all-registered + python -m torch.distributed.tensor._ops.strategy_validation --op div --incorrect-only + python -m torch.distributed.tensor._ops.strategy_validation --op add --show-repro +""" + +import argparse +import fnmatch +import itertools +import re +import sys +import time +from collections import defaultdict +from collections.abc import Callable +from dataclasses import dataclass, field +from typing import Any + +import torch +import torch.distributed as dist +import torch.testing._internal.common_methods_invocations as common_ops +from torch._ops import OpOverload +from torch.distributed._local_tensor import LocalTensor, LocalTensorMode +from torch.distributed.device_mesh import DeviceMesh, init_device_mesh +from torch.distributed.tensor import distribute_tensor, DTensor, Replicate +from torch.distributed.tensor._decompositions import DecompShardingStrategy +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ( + DTensorSpec, + OpSchema, + OpSpec, + OpStrategy, +) +from torch.distributed.tensor._ops.single_dim_strategy import _ShardingPlaceholder +from torch.distributed.tensor.placement_types import Partial, Placement, Shard +from torch.testing._internal.common_methods_invocations import op_db, SampleInput +from torch.testing._internal.opinfo import core as opinfo_core +from torch.utils import _pytree as pytree + + +# A combo key is (input_placement_strs, output_placement_strs) +# For single-output ops: (("S(0)",), ("P(min)",)) +# For multi-output ops: (("S(0)",), ("P(min)", "P(min)")) +ComboKey = tuple[tuple[str, ...], tuple[str, ...]] + +# Partial reduce ops to enumerate +PARTIAL_REDUCE_OPS = ["sum", "avg", "min", "max"] + +SKIP_OPS: dict[str, str] = { + "bernoulli": "non-deterministic (random sampling)", + "empty_like": "uninitialized memory", + "exponential": "non-deterministic (random sampling)", + "new_empty": "uninitialized memory", + "new_empty_strided": "uninitialized memory", + "nn.functional.dropout": "non-deterministic (random masking)", + "normal": "non-deterministic (random sampling)", + "rand_like": "non-deterministic (random sampling)", + "randint_like": "non-deterministic (random sampling)", + "randn_like": "non-deterministic (random sampling)", + "uniform": "non-deterministic (random sampling)", +} + + +PlacementCombination = tuple[tuple[Placement, ...], tuple[Placement, ...]] + + +@dataclass +class Discrepancy: + """Represents a discrepancy between ground truth and DTensor's rules.""" + + input_placements: tuple[str, ...] + output_placements: tuple[str, ...] + sample_idx: int + input_shapes: tuple[tuple[int, ...], ...] + discrepancy_type: str # "false_positive" or "false_negative" + error_msg: str = "" + scalar_args: tuple[Any, ...] = () + scalar_kwargs: dict[str, Any] = field(default_factory=dict) + aten_op: OpOverload | None = None + variant: str = "" + sample: SampleInput | None = None + + +@dataclass +class ComparisonStats: + """Statistics for comparing ground truth vs DTensor rules.""" + + true_positives: int = 0 + true_negatives: int = 0 + false_positives: list[Discrepancy] = field( + default_factory=list + ) # DTensor has rule, ground truth says invalid + false_negatives: list[Discrepancy] = field( + default_factory=list + ) # Ground truth valid, DTensor has no rule + total_samples: int = 0 + total_combinations: int = 0 + skip_reasons: dict[str, int] = field(default_factory=dict) + no_dtensor_support: bool = False + # Per aten op variant breakdown (e.g. "aten.min.dim" -> 5) + true_positives_by_op: dict[str, int] = field(default_factory=dict) + + +@dataclass +class _FalsePositiveMitigations: + """Bundle of sample variants used to detect false positive validations. + + Contains negated and non-rounded variants of a sample. These are used to + re-test combinations that pass the primary validation, catching cases where + sign patterns or rounding modes mask real differences. + """ + + negated_sample: SampleInput | None = None + negated_tensors: list[tuple[str, torch.Tensor]] | None = None + negated_ground_truth: torch.Tensor | list[torch.Tensor] | None = None + non_rounded_sample: SampleInput | None = None + non_rounded_ground_truth: torch.Tensor | list[torch.Tensor] | None = None + non_rounded_negated_sample: SampleInput | None = None + non_rounded_negated_tensors: list[tuple[str, torch.Tensor]] | None = None + non_rounded_negated_ground_truth: torch.Tensor | list[torch.Tensor] | None = None + + +def parse_placement(s: str) -> Placement | None: + """ + Parse a placement string back to a placement object. + Placement strings are: R, S(dim), P(reduce_op) + """ + s = s.strip() + if s == "R": + return Replicate() + elif s.startswith("S("): + m = re.match(r"S\((\d+)\)", s) + if m: + return Shard(int(m.group(1))) + elif s.startswith("P("): + m = re.match(r"P\((\w+)\)", s) + if m: + return Partial(m.group(1)) + return None + + +def is_fully_replicated(placements: tuple[Placement, ...]) -> bool: + """Check if all placements are Replicate.""" + return all(isinstance(p, Replicate) for p in placements) + + +def is_trivial_shard(p: Placement, tensor_shape: tuple[int, ...]) -> bool: + """Check if placement is a Shard on a size-1 dimension.""" + # NOTE: isinstance(_, Shard) does not match _StridedShard; see _is_shard_like(). + return ( + isinstance(p, Shard) and p.dim < len(tensor_shape) and tensor_shape[p.dim] == 1 + ) + + +def normalize_placement(p: Placement, tensor_shape: tuple[int, ...]) -> Placement: + """ + Normalize a placement for a given tensor shape. + + Converts Shard on a size-1 dimension to Replicate for deduplication. + Shard(0) on a [1, 4] tensor puts all data on rank 0 and an empty [0, 4] + on rank 1. Rank 1's empty computation is vacuous (contributes nothing + after redistribution), so the validation outcome is determined entirely + by rank 0, which has the full data — same as Replicate. We gain no + signal from testing S(0) on a size-1 dim beyond what R already provides, + so we normalize to R to avoid spurious "missing rule" noise when ground + truth and DTensor use different forms for size-1 dims. + """ + if is_trivial_shard(p, tensor_shape): + return Replicate() + return p + + +def normalize_placement_str(p_str: str, shape: tuple[int, ...]) -> str: + """Normalize a placement string, converting trivial shards to Replicate.""" + p = parse_placement(p_str) + if p is None: + return p_str + normalized = normalize_placement(p, shape) + if isinstance(normalized, Replicate): + return "R" + return p_str + + +def normalize_combo_key( + combo_key: ComboKey, + input_shapes: tuple[tuple[int, ...], ...], + output_shapes: tuple[tuple[int, ...], ...], +) -> ComboKey: + """ + Normalize a combo_key by converting trivial shards to Replicate. + + This deduplicates equivalent placement combinations, e.g.: + - P(max) -> S(0) on output [1,1,1] becomes P(max) -> R + - S(0), R -> R on input [1,4] becomes R, R -> R + + Args: + combo_key: (input_placement_strs, output_placement_strs) tuple + input_shapes: Shapes of input tensors + output_shapes: Shapes of output tensors + + Returns: + Normalized combo_key with trivial shards converted to Replicate + """ + input_placement_strs, output_placement_strs = combo_key + + normalized_inputs = tuple( + normalize_placement_str(p_str, shape) + for p_str, shape in zip(input_placement_strs, input_shapes) + ) + + normalized_outputs = tuple( + normalize_placement_str(p_str, shape) + for p_str, shape in zip(output_placement_strs, output_shapes) + ) + + return (normalized_inputs, normalized_outputs) + + +def get_1d_input_placements_for_tensor( + t: torch.Tensor, include_partial: bool = False +) -> list[Placement]: + """ + Get all possible 1-D mesh placements for an INPUT tensor. + + Args: + t: The tensor to get placements for + include_partial: If True, include Partial placements for inputs. + """ + placements: list[Placement] = [Replicate()] + for dim in range(t.ndim): + placements.append(Shard(dim)) + if include_partial and t.dtype != torch.bool: + for reduce_op in PARTIAL_REDUCE_OPS: + placements.append(Partial(reduce_op)) + return placements + + +def get_1d_output_placements_for_tensor(t: torch.Tensor) -> list[Placement]: + """ + Get all possible 1-D mesh placements for an OUTPUT tensor. + """ + placements: list[Placement] = [Replicate()] + for dim in range(t.ndim): + placements.append(Shard(dim)) + + if t.dtype != torch.bool: + for reduce_op in PARTIAL_REDUCE_OPS: + placements.append(Partial(reduce_op)) + return placements + + +def extract_tensors_from_sample( + sample_input: SampleInput, +) -> list[tuple[str, torch.Tensor]]: + """ + Extract all tensor arguments from a SampleInput. + Returns a list of (name, tensor) pairs. + + Uses pytree traversal to match the same order as _replace_with_local + in validate_combination, which uses pytree.tree_map on the same structures. + """ + tensors: list[tuple[str, torch.Tensor]] = [] + idx = 0 + + def _collect(x): + nonlocal idx + if isinstance(x, torch.Tensor): + tensors.append((f"tensor_{idx}", x)) + idx += 1 + return x + + pytree.tree_map(_collect, sample_input.input) + pytree.tree_map(_collect, sample_input.args) + pytree.tree_map(_collect, sample_input.kwargs) + + return tensors + + +def _checkerboard_mask( + tensor: torch.Tensor, tensor_idx: int = 0, mask_shift: int = 0 +) -> torch.Tensor: + """Checkerboard mask that alternates in every dimension. + + Unlike flat-index % 2, which can be uniform along even-stride dimensions + (causing all elements in a reduction group to get the same offset), the + checkerboard uses sum-of-coordinates mod 2 so adjacent elements differ + along every axis. + + Returns a flat bool tensor of shape (numel,). + """ + if tensor.ndim == 0: + return torch.tensor([(tensor_idx + mask_shift) % 2 == 0], device=tensor.device) + coords = [torch.arange(s, device=tensor.device) for s in tensor.shape] + grids = torch.meshgrid(*coords, indexing="ij") + coord_sum = grids[0].clone() + for g in grids[1:]: + coord_sum += g + return ((coord_sum + tensor_idx + mask_shift) % 2 == 0).flatten() + + +def _create_partial_input( + tensor: torch.Tensor, + placement: Partial, + world_size: int, + tensor_idx: int = 0, + mask_shift: int = 0, +) -> LocalTensor: + """ + Create a LocalTensor with values that reduce to the original tensor. + + Uses asymmetric splits to avoid coincidental matches when combining + different Partial types. + + For each placement combination, it creates local tensors that would + reduce to the original (e.g., for P(sum), splits values across ranks so + they sum back), runs the op on those local tensors, wraps the output as + a DTensor, redistributes to Replicate, and compares against ground + truth. + + The main challenge is avoiding false positives where a rule appears + valid on a specific input but is actually incorrect. Several techniques + are used: + + Asymmetric splits for P(sum)/P(avg): instead of splitting evenly + (tensor/2 per rank), uses a 60/40 ratio (varied by tensor index) so + that ops which are not truly linear don't accidentally produce + matching outputs. + + Sign-varying offsets for P(sum)/P(avg): adds an offset that + alternates sign across elements, so local tensors have mixed positive + and negative values. Without this, proportional splits preserve the + sign pattern of the original tensor, causing non-linear ops like abs + to falsely validate P(sum)->P(sum). + + Distinct magnitudes for P(min) vs P(max): P(min) offsets non-holding + ranks by +(range*2+1) while P(max) offsets by -(range*2+1), where + range is the tensor's value range. Using adaptive offsets that exceed + the value range ensures that index-returning ops (argmin/argmax) + produce different results on different ranks, correctly rejecting + P(min)/P(max) inputs for those ops. Using different signs for min vs + max prevents accidental cancellation when both appear in the same + combination. + + Alternating rank ownership for P(min)/P(max): a multi-dimensional + checkerboard mask (sum of coordinates mod 2) controls which rank holds + the true value vs the offset value. Unlike a flat-index mask which can + have uniform parity along an even-stride dimension, the checkerboard + guarantees alternation along EVERY dimension. The mask_shift parameter + allows re-validation with the complementary mask to catch ops where + the result coincidentally matches. + + """ + reduce_op = placement.reduce_op + local_tensors: dict[int, torch.Tensor] = {} + + if reduce_op in ("sum", "avg"): + base_ratio = 0.6 + 0.1 * (tensor_idx % 3) + + # See docstring above: "Sign-varying offsets" + flat = tensor.flatten() + offset_mag = flat.abs() + 1.0 + signs = torch.ones_like(flat) + # Use checkerboard mask so offset sign alternates in every dimension, + # not just along flat index (which can be uniform along even-stride dims). + signs[_checkerboard_mask(tensor, tensor_idx, mask_shift)] = -1.0 + offset = (offset_mag * signs).reshape(tensor.shape) + + scale = world_size if reduce_op == "avg" else 1 + for r in range(world_size): + if r == 0: + local_tensors[r] = tensor.clone() * base_ratio * scale + offset + else: + local_tensors[r] = tensor.clone() * ( + (1 - base_ratio) / (world_size - 1) + ) * scale - offset / (world_size - 1) + + elif reduce_op == "min": + # See docstring above: "Distinct Magnitudes" and "Alternating Rank Ownership" + flat = tensor.flatten() + # Offset must exceed the tensor's value range so that the mask pattern + # determines argmin/argmax, not the original values. + value_range = (flat.max() - flat.min()).item() + min_offset = value_range * 2 + 1 + mask = _checkerboard_mask(tensor, tensor_idx, mask_shift) + for r in range(world_size): + if r == 0: + r_offset = torch.where( + mask, torch.zeros_like(flat), torch.full_like(flat, min_offset) + ) + else: + r_offset = torch.where( + mask, torch.full_like(flat, min_offset), torch.zeros_like(flat) + ) + local_tensors[r] = (flat + r_offset).reshape(tensor.shape) + + elif reduce_op == "max": + # See docstring above: "Distinct Magnitudes" and "Alternating Rank Ownership" + flat = tensor.flatten() + value_range = (flat.max() - flat.min()).item() + max_offset = -(value_range * 2 + 1) + mask = _checkerboard_mask(tensor, tensor_idx, mask_shift) + for r in range(world_size): + if r == 0: + r_offset = torch.where( + mask, torch.zeros_like(flat), torch.full_like(flat, max_offset) + ) + else: + r_offset = torch.where( + mask, torch.full_like(flat, max_offset), torch.zeros_like(flat) + ) + local_tensors[r] = (flat + r_offset).reshape(tensor.shape) + + else: + for r in range(world_size): + local_tensors[r] = tensor.clone() + + # pyrefly: ignore [bad-argument-type, bad-argument-count] + return LocalTensor(local_tensors) + + +def _shard_tensors( + tensors: list[tuple[str, torch.Tensor]], + input_placements: tuple[Placement, ...], + world_size: int, + mesh: DeviceMesh, + mask_shift: int = 0, +) -> list[LocalTensor | torch.Tensor]: + """Create sharded LocalTensors from tensors according to placements.""" + local_tensors: list[LocalTensor | torch.Tensor] = [] + for tensor_idx, ((name, tensor), placement) in enumerate( + zip(tensors, input_placements) + ): + if isinstance(placement, Partial): + local_tensor = _create_partial_input( + tensor, placement, world_size, tensor_idx, mask_shift + ) + elif isinstance(placement, Replicate): + _tmp = {r: tensor.clone() for r in range(world_size)} + # pyrefly: ignore [bad-argument-type, bad-argument-count] + local_tensor = LocalTensor(_tmp) + elif isinstance(placement, Shard): + shard_dim = placement.dim + chunks = tensor.tensor_split(world_size, dim=shard_dim) + _tmp = { + r: chunks[r].clone(memory_format=torch.contiguous_format) + for r in range(world_size) + } + # pyrefly: ignore [bad-argument-type, bad-argument-count] + local_tensor = LocalTensor(_tmp) + else: + dt = distribute_tensor(tensor.clone(), mesh, (placement,)) + local_tensor = dt.to_local() + local_tensors.append(local_tensor) + return local_tensors + + +def _compare_outputs( + local_output: Any, + ground_truth: torch.Tensor | list[torch.Tensor], + output_placements: tuple[Placement, ...], + mesh: DeviceMesh, + world_size: int, +) -> tuple[bool, str]: + """Compare op output (wrapped as DTensor) against ground truth.""" + if isinstance(local_output, (list, tuple)): + local_outputs = list(local_output) + else: + local_outputs = [local_output] + + if isinstance(ground_truth, list): + ground_truths = ground_truth + else: + ground_truths = [ground_truth] + + if len(local_outputs) != len(ground_truths): + return ( + False, + f"Output count mismatch: got {len(local_outputs)}, " + f"expected {len(ground_truths)}", + ) + + if len(local_outputs) != len(output_placements): + return ( + False, + f"Output count mismatch with placements: " + f"got {len(local_outputs)}, expected {len(output_placements)}", + ) + + for i, (local_out, gt, out_plc) in enumerate( + zip(local_outputs, ground_truths, output_placements) + ): + if not isinstance(local_out, torch.Tensor): + return False, f"Local output[{i}] is not a tensor: {type(local_out)}" + + if not isinstance(local_out, LocalTensor): + return False, f"LocalTensor inputs produced non-LocalTensor output[{i}]" + + output_dt = DTensor.from_local( + local_out, + mesh, + (out_plc,), + shape=gt.shape, + stride=gt.stride(), + ) + + if isinstance(out_plc, Replicate): + local_values = [local_out._local_tensors[r] for r in range(world_size)] + all_same = all( + torch.allclose(local_values[0], lv, atol=1e-5, rtol=1e-5) + for lv in local_values[1:] + ) + if not all_same: + return ( + False, + f"Replicate output[{i}] but local values differ across ranks", + ) + + full_output = output_dt.redistribute(mesh, (Replicate(),)).to_local() + + if isinstance(full_output, LocalTensor): + full_output = full_output._local_tensors[0] + + if gt.shape != full_output.shape: + return ( + False, + f"Shape mismatch[{i}]: expected {gt.shape}, got {full_output.shape}", + ) + + if not torch.allclose(gt, full_output, atol=1e-5, rtol=1e-5, equal_nan=True): + max_diff = (gt - full_output).abs().max().item() + return False, f"Value mismatch[{i}]: max_diff={max_diff:.6f}" + + return True, "" + + +def validate_combination( + op: Callable[..., Any], + sample_input: SampleInput, + tensors: list[tuple[str, torch.Tensor]], + combination: PlacementCombination, + ground_truth: torch.Tensor | list[torch.Tensor], + world_size: int = 2, + mesh: DeviceMesh | None = None, + mask_shift: int = 0, +) -> tuple[bool | None, str]: + """ + Validate a single placement combination against ground truth. + + Returns (True, "") if valid, (False, error_msg) if invalid, or + (None, reason) if the combination cannot be tested (e.g. uneven shards). + + The validation logic: + 1. Shard inputs according to input placements to get local tensors + 2. Run the raw op on local tensors (bypassing DTensor dispatch) + 3. Wrap the local output in a DTensor with the claimed output placement + 4. Redistribute to Replicate and compare with ground truth + + Args: + op: The operator function + sample_input: The SampleInput with original arguments + tensors: List of (name, tensor) pairs extracted from sample + combination: The placement combination to validate + ground_truth: Expected output tensor(s). For multi-output ops, a list + of tensors where each element is validated independently against + the same output placement. + world_size: Number of simulated ranks + mesh: Optional pre-created device mesh (for performance) + + Returns: + (is_valid, error_message) + """ + try: + if mesh is None: + device = tensors[0][1].device.type if tensors else "cpu" + mesh = init_device_mesh(device, (world_size,)) + + # Uneven shards produce SymInt in LocalTensor's wrapper shape, + # which breaks C++ overload resolution before __torch_dispatch__ + # can intercept. Return None to signal "untestable". + for (name, tensor), placement in zip(tensors, combination[0]): + if isinstance(placement, Shard): + if tensor.size(placement.dim) % world_size != 0: + return None, "uneven shard" + + local_tensors = _shard_tensors( + tensors, combination[0], world_size, mesh, mask_shift + ) + + local_idx = 0 + + def _replace_with_local(a): + nonlocal local_idx + if isinstance(a, torch.Tensor): + local = local_tensors[local_idx] + local_idx += 1 + return local + return a + + if isinstance(sample_input.input, torch.Tensor): + local_input = _replace_with_local(sample_input.input) + else: + local_input = pytree.tree_map(_replace_with_local, sample_input.input) + + local_args = pytree.tree_map(_replace_with_local, sample_input.args) + local_kwargs = pytree.tree_map(_replace_with_local, sample_input.kwargs) + + local_output = op(local_input, *local_args, **local_kwargs) + + return _compare_outputs( + local_output, ground_truth, combination[1], mesh, world_size + ) + + except Exception as e: + # TODO: This is too broad. Consider: (1) explicit checks for shard dim + # validity and shape compatibility before calling tensor_split/from_local, + # (2) scoped try/except around op() and redistribute() that raise specific + # exceptions (e.g., UnsupportedRedistribute, OpError), and (3) only + # catching those here, letting real bugs propagate. + return False, f"Exception: {type(e).__name__}: {e}" + + +def extract_tensors_from_args( + args: tuple[Any, ...], + kwargs: dict[str, Any], +) -> list[tuple[str, torch.Tensor]]: + """Extract tensor arguments from captured aten args/kwargs. + + Unlike extract_tensors_from_sample which walks SampleInput pytrees, + this walks the flat aten-level args and kwargs directly. + """ + tensors: list[tuple[str, torch.Tensor]] = [] + idx = 0 + + def _collect(x: Any) -> Any: + nonlocal idx + if isinstance(x, torch.Tensor): + tensors.append((f"tensor_{idx}", x)) + idx += 1 + return x + + pytree.tree_map(_collect, args) + pytree.tree_map(_collect, kwargs) + return tensors + + +def validate_aten_combination( + aten_op: OpOverload, + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], + ground_truth: torch.Tensor | list[torch.Tensor], + combination: PlacementCombination, + world_size: int, + mesh: DeviceMesh, + mask_shift: int = 0, +) -> tuple[bool | None, str]: + """Validate a placement combination using aten-level captured args. + + Works directly with aten op args/kwargs instead of SampleInput pytrees. + Replaces tensors in the flat args/kwargs with sharded LocalTensors, + calls the aten op, and compares output. + + Returns (True, ""), (False, error_msg), or (None, reason) if untestable. + """ + try: + tensors = extract_tensors_from_args(captured_args, captured_kwargs) + if not tensors: + return False, "No tensor args in captured aten call" + + for (name, tensor), placement in zip(tensors, combination[0]): + if isinstance(placement, Shard): + if tensor.size(placement.dim) % world_size != 0: + return None, "uneven shard" + + local_tensors = _shard_tensors( + tensors, combination[0], world_size, mesh, mask_shift + ) + + local_idx = 0 + + def _replace_with_local(a: Any) -> Any: + nonlocal local_idx + if isinstance(a, torch.Tensor): + local = local_tensors[local_idx] + local_idx += 1 + return local + return a + + local_args = pytree.tree_map(_replace_with_local, captured_args) + local_kwargs = pytree.tree_map(_replace_with_local, captured_kwargs) + + local_output = aten_op(*local_args, **local_kwargs) + + return _compare_outputs( + local_output, ground_truth, combination[1], mesh, world_size + ) + + except Exception as e: + return False, f"Exception: {type(e).__name__}: {e}" + + +def has_pmin_pmax( + input_placements: tuple[Placement, ...], + output_placements: tuple[Placement, ...], +) -> bool: + """Check if any placement is Partial(min) or Partial(max).""" + for p in (*input_placements, *output_placements): + if isinstance(p, Partial) and p.reduce_op in ("min", "max"): + return True + return False + + +def has_any_partial( + input_placements: tuple[Placement, ...], + output_placements: tuple[Placement, ...], +) -> bool: + """Check if any placement is Partial (any reduce op).""" + for p in (*input_placements, *output_placements): + if isinstance(p, Partial): + return True + return False + + +def negate_all_tensors( + tensors: list[tuple[str, torch.Tensor]], +) -> list[tuple[str, torch.Tensor]]: + """Return a new list with all tensors negated.""" + return [(name, -t) for name, t in tensors] + + +def create_fully_negated_sample(sample: SampleInput) -> SampleInput: + """Create a sample with ALL tensors negated (for P(min)/P(max) sign testing).""" + + def negate_tensor(x): + if isinstance(x, torch.Tensor): + return -x + return x + + new_input = pytree.tree_map(negate_tensor, sample.input) + new_args = pytree.tree_map(negate_tensor, sample.args) + new_kwargs = pytree.tree_map(negate_tensor, sample.kwargs) + + return SampleInput(new_input, args=new_args, kwargs=new_kwargs) + + +def _run_op_on_sample(op: Callable[..., Any], sample: SampleInput) -> Any: + """Run an operator on a SampleInput, handling both tensor and tuple inputs.""" + if isinstance(sample.input, torch.Tensor): + return op(sample.input, *sample.args, **sample.kwargs) + return op(*sample.input, *sample.args, **sample.kwargs) + + +def _extract_rules_from_op_strategy( + op_strategy: Any, + input_shapes: tuple[tuple[int, ...], ...], + output_shapes: tuple[tuple[int, ...], ...], +) -> set[ComboKey]: + """Extract normalized sharding rules from an OpStrategy. + + Called during rule comparison to collect DTensor's claimed-valid placement + combinations. These are compared against ground truth (brute-force + validation) to find false positives (DTensor claims valid but wrong) and + false negatives (valid but DTensor has no rule). + """ + rules: set[ComboKey] = set() + if not isinstance(op_strategy, OpStrategy): + return rules + for spec in op_strategy.strategies: + if spec.input_specs is None: + continue + if isinstance(spec.output_specs, tuple): + output_plcs: list[Placement] = [] + has_none = False + for out_spec in spec.output_specs: + if out_spec is None: + # None means the output placement is undefined for this + # strategy (e.g. indices under P(max) reduction). Skip it. + has_none = True + break + output_plcs.append(out_spec.placements[0]) + if has_none: + continue + else: + # Single DTensorSpec — the propagator duplicates it for all + # outputs of multi-output ops, so we do the same here. + output_plcs = [spec.output_spec.placements[0]] * len(output_shapes) + input_plcs = tuple(s.placements[0] for s in spec.input_specs) + rule_key: ComboKey = ( + tuple(str(p) for p in input_plcs), + tuple(str(p) for p in output_plcs), + ) + normalized_rule = normalize_combo_key(rule_key, input_shapes, output_shapes) + if not is_fully_replicated( + tuple(parse_placement(p) or Replicate() for p in normalized_rule[0]) + ): + rules.add(normalized_rule) + return rules + + +class _CaptureAtenOp(torch.utils._python_dispatch.TorchDispatchMode): + """Dispatch mode that captures aten ops called, their args, and return values.""" + + def __init__(self, target_op_name: str = ""): + self.target_op_name = target_op_name.lower() + self.all_ops: list[tuple[OpOverload, tuple[Any, ...], dict[str, Any], Any]] = [] + self.best_match: OpOverload | None = None + self.best_match_args: tuple[Any, ...] | None = None + self.best_match_kwargs: dict[str, Any] | None = None + self.best_match_result: Any = None + + def __torch_dispatch__(self, func, types, args=(), kwargs=None): + if kwargs is None: + kwargs = {} + result = func(*args, **kwargs) + if func.namespace == "aten": + self.all_ops.append((func, args, kwargs, result)) + op_name = func.name().split("::")[1].split(".")[0].lower() + if self.target_op_name and self.target_op_name in op_name: + if self.best_match is None: + self.best_match = func + self.best_match_args = args + self.best_match_kwargs = kwargs + self.best_match_result = result + return result + + +def get_aten_op_for_sample( + op: Callable[..., Any], sample: SampleInput, op_name: str = "" +) -> _CaptureAtenOp: + """ + Capture aten ops dispatched for a given sample. + + Returns the _CaptureAtenOp object containing all captured ops with their + args, kwargs, and return values. Use best_match for the primary op or + all_ops for exhaustive iteration. + """ + with _CaptureAtenOp(op_name) as capture: + try: + if isinstance(sample.input, torch.Tensor): + op(sample.input, *sample.args, **sample.kwargs) + else: + op(*sample.input, *sample.args, **sample.kwargs) + except Exception: + pass + + # Populate best_match from first op if target match wasn't found + if capture.best_match is None and capture.all_ops: + first_op, first_args, first_kwargs, first_result = capture.all_ops[0] + capture.best_match = first_op + capture.best_match_args = first_args + capture.best_match_kwargs = first_kwargs + capture.best_match_result = first_result + + return capture + + +def query_single_dim_strategy( + op_overload: OpOverload, + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], +) -> list[list[Placement]] | None: + """ + Query DTensor's single-dim strategy for given input tensors. + Returns list of [output_placement, *input_placements] rules. + """ + propagator = DTensor._op_dispatcher.sharding_propagator + + if op_overload not in propagator.op_single_dim_strategy_funcs: + return None + + strategy_func = propagator.op_single_dim_strategy_funcs[op_overload] + + args_meta = tuple( + TensorMeta(shape=a.shape, stride=a.stride(), dtype=a.dtype) + if isinstance(a, torch.Tensor) + else a + for a in captured_args + ) + kwargs_meta = { + k: TensorMeta(shape=v.shape, stride=v.stride(), dtype=v.dtype) + if isinstance(v, torch.Tensor) + else v + for k, v in captured_kwargs.items() + } + + try: + result = strategy_func(op_overload, args_meta, kwargs_meta) + + expanded_result: list[list[Placement]] = [] + for combo in result: + expanded_combo: list[Placement] = [] + for p in combo: + if isinstance(p, _ShardingPlaceholder): + expanded_combo.append(Shard(p.dim)) + else: + expanded_combo.append(p) + expanded_result.append(expanded_combo) + + return expanded_result + except Exception: + return None + + +def get_opinfo_by_name(name: str) -> list[opinfo_core.OpInfo]: + """Find OpInfo entries by exact operator name.""" + matches = [op for op in op_db if op.name == name] + if matches: + return matches + + # Suggest alternatives + candidates = _find_opinfo_candidates(name) + if candidates: + suggestions = ", ".join(f'"{c}"' for c in candidates) + raise ValueError(f'No OpInfo found for "{name}", did you mean: {suggestions}?') + raise ValueError( + f'No OpInfo found for "{name}". OpInfo is required as it provides ' + f"sample inputs for the operator." + ) + + +def _find_opinfo_candidates(name: str) -> list[str]: + """Find OpInfo names that plausibly match a short/incorrect name.""" + candidates: list[str] = [] + seen: set[str] = set() + # Match on aten_name (e.g., "relu" -> OpInfo with aten_name="relu") + for op in op_db: + if op.aten_name == name and op.name not in seen: + candidates.append(op.name) + seen.add(op.name) + # Suffix match: "relu" matches "nn.functional.relu" + suffix = "." + name + for op in op_db: + if op.name.endswith(suffix) and op.name not in seen: + candidates.append(op.name) + seen.add(op.name) + return candidates + + +def resolve_op_names(patterns: list[str]) -> list[str]: + """Resolve user-provided op patterns to exact OpInfo names. + + Supports exact names, comma separation, and glob patterns (e.g., + "nn.functional.*"). Short names like "relu" are resolved unambiguously + or an error is raised with suggestions. + """ + all_opinfo_names = sorted({op.name for op in op_db}) + resolved: list[str] = [] + seen: set[str] = set() + + for pattern in patterns: + # Glob pattern + if "*" in pattern or "?" in pattern: + matches = fnmatch.filter(all_opinfo_names, pattern) + if not matches: + raise ValueError(f'No OpInfo names match pattern "{pattern}".') + for m in matches: + if m not in seen: + resolved.append(m) + seen.add(m) + continue + + # Exact match + if pattern in {op.name for op in op_db}: + if pattern not in seen: + resolved.append(pattern) + seen.add(pattern) + continue + + # Try to resolve shorthand + candidates = _find_opinfo_candidates(pattern) + if len(candidates) == 1: + name = candidates[0] + if name not in seen: + resolved.append(name) + seen.add(name) + elif len(candidates) > 1: + suggestions = ", ".join(f'"{c}"' for c in candidates) + raise ValueError( + f'"{pattern}" is ambiguous, matching: {suggestions}. ' + f"Use the fully qualified name." + ) + else: + raise ValueError(f'No OpInfo found for "{pattern}".') + + return resolved + + +def _is_tensor_output(result: Any) -> bool: + """Check if a result is a tensor or list/tuple of tensors.""" + if isinstance(result, torch.Tensor): + return True + if isinstance(result, (list, tuple)): + has_tensor = any(isinstance(t, torch.Tensor) for t in result) + all_tensor = all(isinstance(t, torch.Tensor) for t in result) + if has_tensor and not all_tensor: + raise NotImplementedError( + f"Mixed tensor/non-tensor tuple outputs are not supported by the " + f"validator. Got types: {[type(t).__name__ for t in result]}" + ) + return all_tensor + return False + + +def _to_ground_truth(result: Any) -> torch.Tensor | list[torch.Tensor]: + """Convert an op result to the ground truth format (tensor or list of tensors).""" + if isinstance(result, torch.Tensor): + return result + return list(result) + + +def _prepare_false_positive_mitigations( + op: Callable[..., Any], + sample: SampleInput, + tensors: list[tuple[str, torch.Tensor]], +) -> _FalsePositiveMitigations: + """Create negated and non-rounded sample variants for false positive detection.""" + m = _FalsePositiveMitigations() + + try: + m.negated_sample = create_fully_negated_sample(sample) + m.negated_tensors = negate_all_tensors(tensors) + result = _run_op_on_sample(op, m.negated_sample) + if _is_tensor_output(result): + m.negated_ground_truth = _to_ground_truth(result) + else: + m.negated_sample = None + except Exception: + m.negated_sample = None + m.negated_tensors = None + + if "rounding_mode" not in sample.kwargs: + return m + + try: + non_rounded_kwargs = { + k: v for k, v in sample.kwargs.items() if k != "rounding_mode" + } + m.non_rounded_sample = SampleInput( + sample.input, args=sample.args, kwargs=non_rounded_kwargs + ) + result = _run_op_on_sample(op, m.non_rounded_sample) + if not _is_tensor_output(result): + m.non_rounded_sample = None + else: + m.non_rounded_ground_truth = _to_ground_truth(result) + m.non_rounded_negated_sample = create_fully_negated_sample( + m.non_rounded_sample + ) + m.non_rounded_negated_tensors = negate_all_tensors(tensors) + nr_neg_result = _run_op_on_sample(op, m.non_rounded_negated_sample) + if _is_tensor_output(nr_neg_result): + m.non_rounded_negated_ground_truth = _to_ground_truth(nr_neg_result) + else: + m.non_rounded_negated_sample = None + except Exception: + m.non_rounded_sample = None + m.non_rounded_ground_truth = None + m.non_rounded_negated_sample = None + + return m + + +def _query_dtensor_rules( + aten_op: OpOverload | None, + tensors: list[tuple[str, torch.Tensor]], + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], + input_shapes: tuple[tuple[int, ...], ...], + output_shapes: tuple[tuple[int, ...], ...], + world_size: int, + verbose: bool, +) -> set[ComboKey]: + """Query DTensor's claimed sharding rules via single-dim, op_strategy, or decomp paths. + + TODO: This reimplements strategy resolution logic from ShardingPropagator. + Refactor ShardingPropagator to expose a public API for querying sharding + rules given an op and tensor metadata, so this function can be replaced + with a single call. + """ + if not aten_op: + return set() + + num_tensors = len(tensors) + non_tensor_kwargs = { + k: v for k, v in captured_kwargs.items() if not isinstance(v, torch.Tensor) + } + n_outputs = len(output_shapes) + propagator = DTensor._op_dispatcher.sharding_propagator + rules: set[ComboKey] = set() + + if aten_op in propagator.op_single_dim_strategy_funcs: + strategy_result = query_single_dim_strategy( + aten_op, captured_args, captured_kwargs + ) + if strategy_result: + for combo in strategy_result: + if len(combo) >= n_outputs + num_tensors: + output_plcs = combo[:n_outputs] + input_plcs = tuple(combo[n_outputs : n_outputs + num_tensors]) + rule_key: ComboKey = ( + tuple(str(p) for p in input_plcs), + tuple(str(p) for p in output_plcs), + ) + normalized_rule = normalize_combo_key( + rule_key, input_shapes, output_shapes + ) + if not is_fully_replicated( + tuple( + parse_placement(p) or Replicate() + for p in normalized_rule[0] + ) + ): + rules.add(normalized_rule) + + elif aten_op in propagator.op_strategy_funcs: + try: + mesh = init_device_mesh("cpu", (world_size,)) + # Build OpStrategy objects for each tensor, keyed by identity + tensor_to_strategy: dict[int, OpStrategy] = {} + for _, t in tensors: + input_placements = get_1d_input_placements_for_tensor( + t, include_partial=True + ) + specs = [] + for p in input_placements: + spec = DTensorSpec( + mesh=mesh, + placements=(p,), + tensor_meta=TensorMeta( + shape=t.shape, stride=t.stride(), dtype=t.dtype + ), + ) + specs.append(OpSpec(output_specs=spec, input_specs=tuple())) + tensor_to_strategy[id(t)] = OpStrategy(specs) + # Interleave strategies and non-tensor args at original positions + args_schema = [ + tensor_to_strategy[id(a)] if isinstance(a, torch.Tensor) else a + for a in captured_args + ] + op_schema = OpSchema(aten_op, tuple(args_schema), non_tensor_kwargs) + strategy_func = propagator.op_strategy_funcs[aten_op] + output_strategy = strategy_func(op_schema) + rules |= _extract_rules_from_op_strategy( + output_strategy, input_shapes, output_shapes + ) + except Exception as e: + if verbose: + print(f" Error querying op_strategy: {e}") + + else: + # Decomp-based strategy: only discovers rules reachable from a single + # seed (Shard(0) on the first input). Rules requiring other input + # placements (e.g., Shard(1), Partial, or sharding on non-first inputs) + # will not be found, so this under-reports DTensor's capabilities. + if DecompShardingStrategy.has_decomp(aten_op): + try: + mesh = init_device_mesh("cpu", (world_size,)) + # Interleave DTensorSpec and non-tensor args at original positions + tensor_idx = 0 + args_schema: list[Any] = [] + for a in captured_args: + if isinstance(a, torch.Tensor): + # First tensor gets Shard(0) to seed candidate + # placement generation in _get_candidate_placements + plc = Shard(0) if tensor_idx == 0 else Replicate() + spec = DTensorSpec( + mesh=mesh, + placements=(plc,), + tensor_meta=TensorMeta( + shape=a.shape, stride=a.stride(), dtype=a.dtype + ), + ) + args_schema.append(spec) + tensor_idx += 1 + else: + args_schema.append(a) + op_schema = OpSchema(aten_op, tuple(args_schema), non_tensor_kwargs) + propagator.decomp_strategy.ensure_schema_info(aten_op) + output_strategy = propagator.decomp_strategy.propagate_strategy( + op_schema, + ) + if output_strategy is not None: + rules |= _extract_rules_from_op_strategy( + output_strategy, input_shapes, output_shapes + ) + except Exception as e: + if verbose: + print(f" Error querying decomp strategy: {e}") + + return rules + + +def _validate_with_mitigations( + op: Callable[..., Any], + sample: SampleInput, + tensors: list[tuple[str, torch.Tensor]], + input_placements: tuple[Placement, ...], + output_placements: tuple[Placement, ...], + ground_truth: torch.Tensor | list[torch.Tensor], + world_size: int, + mesh: DeviceMesh, + mitigations: _FalsePositiveMitigations, +) -> bool | None: + """Validate a combination, including false positive mitigation re-checks. + + Returns True (valid), False (invalid), or None (untestable). + """ + combo: PlacementCombination = (input_placements, output_placements) + is_valid, _ = validate_combination( + op, sample, tensors, combo, ground_truth, world_size, mesh + ) + if is_valid is None: + return None + + # Flipped-mask mitigation: the checkerboard mask that controls offset + # signs (for P(sum)/P(avg)) or rank ownership (for P(min)/P(max)) is + # deterministic per tensor_idx. Re-validate with the complementary mask + # to catch index-returning ops (argmin/argmax) where the result + # coincidentally matches because the dominant value happens to land on + # a position where both mask orientations preserve argmin/argmax. + if is_valid and has_any_partial(input_placements, output_placements): + is_valid, _ = validate_combination( + op, + sample, + tensors, + combo, + ground_truth, + world_size, + mesh, + mask_shift=1, + ) + + if ( + is_valid + and mitigations.negated_sample + and has_pmin_pmax(input_placements, output_placements) + ): + if mitigations.negated_tensors is None: + raise AssertionError("negated_tensors is None") + if mitigations.negated_ground_truth is None: + raise AssertionError("negated_ground_truth is None") + is_valid, _ = validate_combination( + op, + mitigations.negated_sample, + mitigations.negated_tensors, + combo, + mitigations.negated_ground_truth, + world_size, + mesh, + ) + + if ( + is_valid + and mitigations.non_rounded_sample + and has_any_partial(input_placements, output_placements) + ): + if mitigations.non_rounded_ground_truth is None: + raise AssertionError("non_rounded_ground_truth is None") + is_valid, _ = validate_combination( + op, + mitigations.non_rounded_sample, + tensors, + combo, + mitigations.non_rounded_ground_truth, + world_size, + mesh, + ) + + if ( + is_valid + and mitigations.non_rounded_negated_sample + and has_pmin_pmax(input_placements, output_placements) + ): + if mitigations.non_rounded_negated_tensors is None: + raise AssertionError("non_rounded_negated_tensors is None") + if mitigations.non_rounded_negated_ground_truth is None: + raise AssertionError("non_rounded_negated_ground_truth is None") + is_valid, _ = validate_combination( + op, + mitigations.non_rounded_negated_sample, + mitigations.non_rounded_negated_tensors, + combo, + mitigations.non_rounded_negated_ground_truth, + world_size, + mesh, + ) + + return is_valid + + +@dataclass +class _AtenFalsePositiveMitigations: + """Bundle of negated variants for aten-level false positive detection. + + Unlike _FalsePositiveMitigations, this works with captured aten args/kwargs + directly, without SampleInput. The rounding_mode mitigation is skipped + since it's an OpInfo-level concept that doesn't appear in captured aten kwargs. + """ + + negated_args: tuple[Any, ...] | None = None + negated_kwargs: dict[str, Any] | None = None + negated_ground_truth: torch.Tensor | list[torch.Tensor] | None = None + + +def _negate_tensors_in_tree(tree: Any) -> Any: + """Negate all tensors in a pytree structure.""" + + def _negate(x: Any) -> Any: + if isinstance(x, torch.Tensor): + return -x + return x + + return pytree.tree_map(_negate, tree) + + +def _prepare_aten_mitigations( + aten_op: OpOverload, + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], +) -> _AtenFalsePositiveMitigations: + """Create negated variants for aten-level false positive detection.""" + m = _AtenFalsePositiveMitigations() + try: + m.negated_args = _negate_tensors_in_tree(captured_args) + m.negated_kwargs = _negate_tensors_in_tree(captured_kwargs) + result = aten_op(*m.negated_args, **m.negated_kwargs) + if _is_tensor_output(result): + m.negated_ground_truth = _to_ground_truth(result) + else: + m.negated_args = None + m.negated_kwargs = None + except Exception: + m.negated_args = None + m.negated_kwargs = None + return m + + +def _validate_aten_with_mitigations( + aten_op: OpOverload, + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], + input_placements: tuple[Placement, ...], + output_placements: tuple[Placement, ...], + ground_truth: torch.Tensor | list[torch.Tensor], + world_size: int, + mesh: DeviceMesh, + mitigations: _AtenFalsePositiveMitigations, +) -> bool | None: + """Validate an aten-level combination with false positive mitigations. + + Returns True (valid), False (invalid), or None (untestable). + """ + combo: PlacementCombination = (input_placements, output_placements) + is_valid, _ = validate_aten_combination( + aten_op, + captured_args, + captured_kwargs, + ground_truth, + combo, + world_size, + mesh, + ) + if is_valid is None: + return None + + if is_valid and has_any_partial(input_placements, output_placements): + is_valid, _ = validate_aten_combination( + aten_op, + captured_args, + captured_kwargs, + ground_truth, + combo, + world_size, + mesh, + mask_shift=1, + ) + + if ( + is_valid + and mitigations.negated_args is not None + and has_pmin_pmax(input_placements, output_placements) + ): + if mitigations.negated_kwargs is None: + raise AssertionError("negated_kwargs must not be None") + if mitigations.negated_ground_truth is None: + raise AssertionError("negated_ground_truth must not be None") + is_valid, _ = validate_aten_combination( + aten_op, + mitigations.negated_args, + mitigations.negated_kwargs, + mitigations.negated_ground_truth, + combo, + world_size, + mesh, + ) + + return is_valid + + +def _assert_keys_normalized( + keys: set[ComboKey], + input_shapes: tuple[tuple[int, ...], ...], + output_shapes: tuple[tuple[int, ...], ...], +) -> None: + """Assert all combo keys have trivial shards already normalized to Replicate.""" + for key in keys: + if key != normalize_combo_key(key, input_shapes, output_shapes): + raise AssertionError( + f"Key {key} contains un-normalized trivial shards; " + f"call normalize_combo_key before _compare_rules" + ) + + +def _compare_rules( + ground_truth_valid: set[ComboKey], + dtensor_rules: set[ComboKey], + input_shapes: tuple[tuple[int, ...], ...], + output_shapes: tuple[tuple[int, ...], ...], + sample_idx: int, + scalar_args: tuple[Any, ...], + scalar_kwargs: dict[str, Any], + aten_op: OpOverload | None, + variant: str, + stats: ComparisonStats, + sample: SampleInput | None = None, + untestable: set[ComboKey] | None = None, +) -> None: + """Compare ground truth valid rules against DTensor claimed rules, updating stats.""" + if not dtensor_rules: + return + if untestable is None: + untestable = set() + + _assert_keys_normalized(ground_truth_valid, input_shapes, output_shapes) + _assert_keys_normalized(dtensor_rules, input_shapes, output_shapes) + + op_str = str(aten_op) + for combo_key in ground_truth_valid: + if combo_key in dtensor_rules: + stats.true_positives += 1 + stats.true_positives_by_op[op_str] = ( + stats.true_positives_by_op.get(op_str, 0) + 1 + ) + elif combo_key not in untestable: + stats.false_negatives.append( + Discrepancy( + input_placements=combo_key[0], + output_placements=combo_key[1], + sample_idx=sample_idx, + input_shapes=input_shapes, + discrepancy_type="false_negative", + scalar_args=scalar_args, + scalar_kwargs=scalar_kwargs, + aten_op=aten_op, + variant=variant, + sample=sample, + ) + ) + + for combo_key in dtensor_rules: + if combo_key not in ground_truth_valid and combo_key not in untestable: + stats.false_positives.append( + Discrepancy( + input_placements=combo_key[0], + output_placements=combo_key[1], + sample_idx=sample_idx, + input_shapes=input_shapes, + discrepancy_type="false_positive", + scalar_args=scalar_args, + scalar_kwargs=scalar_kwargs, + aten_op=aten_op, + variant=variant, + sample=sample, + ) + ) + + +def _format_op_name(opinfo_name: str) -> str: + """Format an OpInfo name for display, adding aten. prefix for simple names.""" + if "." not in opinfo_name: + return f"aten.{opinfo_name}" + return opinfo_name + + +def _format_sample_repro(sample: SampleInput, aten_op: OpOverload | None = None) -> str: + """Format a SampleInput's values for repro output, using schema arg names.""" + # Get argument names from the aten op schema if available. + # Schema args are ordered: the first is the input, then positional args, + # then keyword-only args (which appear in sample.kwargs). + arg_names: list[str] = [] + if aten_op is not None: + try: + arg_names = [a.name for a in aten_op._schema.arguments] + except Exception: + pass + + parts = [] + # input is the first positional arg + name = arg_names[0] if arg_names else "input" + parts.append(f"{name}={sample.input!r}") + # remaining positional args + for i, arg in enumerate(sample.args): + name = arg_names[1 + i] if 1 + i < len(arg_names) else f"args[{i}]" + parts.append(f"{name}={arg!r}") + # kwargs — use their actual key names (already named) + for k, v in sample.kwargs.items(): + parts.append(f"{k}={v!r}") + return ", ".join(parts) + + +def _print_discrepancy_section( + title: str, discrepancies: list[Discrepancy], show_repro: int = 0 +) -> None: + """Print grouped discrepancies for a section (incorrect or missing).""" + if not discrepancies: + return + print(f"\n{title}") + by_op: dict[str, dict[ComboKey, list[Discrepancy]]] = defaultdict( + lambda: defaultdict(list) + ) + for d in discrepancies: + op_str = str(d.aten_op) + key = (d.input_placements, d.output_placements) + by_op[op_str][key].append(d) + + for op_str in sorted(by_op.keys()): + print(f"\n [{op_str}]") + for (inp, out), discs in sorted(by_op[op_str].items(), key=str): + inp_str = ", ".join(inp) + out_str = out[0] if len(out) == 1 else "(" + ", ".join(out) + ")" + print(f" {inp_str} -> {out_str}") + if show_repro: + limit = len(discs) if show_repro < 0 else show_repro + for d in discs[:limit]: + if d.sample is not None: + print( + f" Repro: {_format_sample_repro(d.sample, d.aten_op)}" + ) + if len(discs) > limit: + print(f" ... and {len(discs) - limit} more") + + +def _print_comparison_summary( + stats: ComparisonStats, + show_repro: int = 0, +) -> None: + """Print discrepancy details for an operator.""" + # Per aten op variant breakdown + fp_by_op: dict[str, set[ComboKey]] = defaultdict(set) + for d in stats.false_positives: + op_str = str(d.aten_op) + fp_by_op[op_str].add((d.input_placements, d.output_placements)) + fn_by_op: dict[str, set[ComboKey]] = defaultdict(set) + for d in stats.false_negatives: + op_str = str(d.aten_op) + fn_by_op[op_str].add((d.input_placements, d.output_placements)) + + all_ops = sorted(set(stats.true_positives_by_op) | set(fp_by_op) | set(fn_by_op)) + if len(all_ops) > 1: + for op_str in all_ops: + tp = stats.true_positives_by_op.get(op_str, 0) + fp = len(fp_by_op.get(op_str, set())) + fn = len(fn_by_op.get(op_str, set())) + print(f" {op_str}: {tp} correct, {fp} incorrect, {fn} missing") + + _print_discrepancy_section( + "Incorrect (has rule but ground truth invalid)", + stats.false_positives, + show_repro, + ) + _print_discrepancy_section( + "Possibly missing (valid in ground truth but no DTensor rule)", + stats.false_negatives, + show_repro, + ) + + +def _has_dtensor_support(aten_op: OpOverload) -> bool: + """Check if an aten op has any DTensor sharding strategy registered.""" + propagator = DTensor._op_dispatcher.sharding_propagator + if aten_op in propagator.op_single_dim_strategy_funcs: + return True + if aten_op in propagator.op_strategy_funcs: + return True + return DecompShardingStrategy.has_decomp(aten_op) + + +def _discover_aten_op( + opinfos: list[opinfo_core.OpInfo], + device: str, + dtype: torch.dtype, +) -> OpOverload | None: + """Discover the aten op dispatched by the first valid sample.""" + for opinfo in opinfos: + try: + samples = list(opinfo.sample_inputs(device, dtype)) + except Exception: + continue + for sample in samples: + tensors = extract_tensors_from_sample(sample) + if not tensors or any(0 in t.shape for _, t in tensors): + continue + capture = get_aten_op_for_sample(opinfo.op, sample, opinfo.name) + aten_op = capture.best_match + if aten_op is not None: + return aten_op + return None + + +def _check_ground_truth( + result: Any, +) -> torch.Tensor | list[torch.Tensor] | None: + """Validate an op result is suitable as ground truth. + + Returns the ground truth tensor(s) or None if the result should be skipped. + """ + if isinstance(result, (list, tuple)): + if not all(isinstance(t, torch.Tensor) for t in result): + return None + gt = list(result) + elif isinstance(result, torch.Tensor): + gt = result + else: + return None + + first_gt = gt[0] if isinstance(gt, list) else gt + if first_gt.numel() == 0: + return None + if (first_gt == 0).all(): + return None + if first_gt.isnan().all(): + return None + return gt + + +def _validate_aten_op_for_sample( + aten_op: OpOverload, + captured_args: tuple[Any, ...], + captured_kwargs: dict[str, Any], + ground_truth: torch.Tensor | list[torch.Tensor], + world_size: int, + incorrect_only: bool, + verbose: bool, + sample_idx: int, + variant: str, + stats: ComparisonStats, + sample: SampleInput | None = None, +) -> tuple[int, int]: + """Validate a single aten op with captured args against ground truth. + + Shared logic used by both default (1:1) and allow_composite modes in + compare_operator. Returns (samples_counted, combinations_counted). + """ + tensors = extract_tensors_from_args(captured_args, captured_kwargs) + if not tensors: + return 0, 0 + if any(0 in t.shape for _, t in tensors): + return 0, 0 + + input_shapes = tuple(t.shape for _, t in tensors) + gt_list = ground_truth if isinstance(ground_truth, list) else [ground_truth] + output_shapes = tuple(tuple(gt.shape) for gt in gt_list) + n_outputs = len(gt_list) + first_gt = gt_list[0] + + scalar_args = tuple(a for a in captured_args if not isinstance(a, torch.Tensor)) + scalar_kwargs = { + k: v for k, v in captured_kwargs.items() if not isinstance(v, torch.Tensor) + } + + mitigations = _prepare_aten_mitigations(aten_op, captured_args, captured_kwargs) + + input_placement_options = [ + get_1d_input_placements_for_tensor(t, include_partial=True) for _, t in tensors + ] + output_placement_options = get_1d_output_placements_for_tensor(first_gt) + + dtensor_rules = _query_dtensor_rules( + aten_op, + tensors, + captured_args, + captured_kwargs, + input_shapes, + output_shapes, + world_size, + verbose, + ) + + ground_truth_valid: set[ComboKey] = set() + total_combinations = 0 + + tensor_device = tensors[0][1].device.type if tensors else "cpu" + with LocalTensorMode(frozenset(range(world_size))): + mesh = init_device_mesh(tensor_device, (world_size,)) + + if incorrect_only: + combinations_to_test = [] + for combo_key in dtensor_rules: + input_plc_strs, output_plc_strs = combo_key + input_plcs_list: list[Placement] = [] + all_valid = True + for s in input_plc_strs: + p = parse_placement(s) + if p is None: + all_valid = False + break + input_plcs_list.append(p) + output_plcs_list: list[Placement] = [] + for s in output_plc_strs: + p = parse_placement(s) + if p is None: + all_valid = False + break + output_plcs_list.append(p) + if not all_valid: + continue + combinations_to_test.append( + ( + tuple(input_plcs_list), + tuple(output_plcs_list), + combo_key, + ) + ) + else: + combinations_to_test = [] + for input_placements in itertools.product(*input_placement_options): + if is_fully_replicated(input_placements): + continue + for output_placement in output_placement_options: + output_placements = tuple( + output_placement for _ in range(n_outputs) + ) + combo_key = ( + tuple(str(p) for p in input_placements), + tuple(str(p) for p in output_placements), + ) + combinations_to_test.append( + (input_placements, output_placements, combo_key) + ) + + untestable: set[ComboKey] = set() + + for ( + input_placements, + output_placements, + combo_key, + ) in combinations_to_test: + total_combinations += 1 + is_valid = _validate_aten_with_mitigations( + aten_op, + captured_args, + captured_kwargs, + input_placements, + output_placements, + ground_truth, + world_size, + mesh, + mitigations, + ) + + if is_valid is None: + normalized_key = normalize_combo_key( + combo_key, input_shapes, output_shapes + ) + untestable.add(normalized_key) + elif is_valid: + normalized_key = normalize_combo_key( + combo_key, input_shapes, output_shapes + ) + if not is_fully_replicated( + tuple(parse_placement(p) or Replicate() for p in normalized_key[0]) + ): + ground_truth_valid.add(normalized_key) + + _compare_rules( + ground_truth_valid, + dtensor_rules, + input_shapes, + output_shapes, + sample_idx, + scalar_args, + scalar_kwargs, + aten_op, + variant, + stats, + sample, + untestable, + ) + + if verbose: + print(f" Sample {sample_idx} [{aten_op}]: shapes={input_shapes}") + print(f" Ground truth valid: {len(ground_truth_valid)}") + print(f" DTensor rules: {len(dtensor_rules)}") + + return 1, total_combinations + + +def compare_operator( + op_name: str, + device: str = "cpu", + dtype: torch.dtype = torch.float32, + world_size: int = 2, + max_samples: int | None = None, + verbose: bool = False, + incorrect_only: bool = False, + allow_composite: bool = False, +) -> ComparisonStats: + """ + Compare DTensor's sharding rules against ground truth for an operator. + + Args: + op_name: Name of the operator to test + device: Device to run on + dtype: Data type for tensors + world_size: Simulated world size + max_samples: Maximum number of samples to test per OpInfo + verbose: Print detailed output + incorrect_only: If True, only test DTensor's claimed rules for correctness. + Skips search for missing rules (much faster). + allow_composite: If True, validate each supported aten op individually for + samples that decompose into multiple aten calls. Default (False) + skips samples where the OpInfo doesn't map 1:1 to a single aten op. + """ + if op_name in SKIP_OPS: + return ComparisonStats() + + opinfos = get_opinfo_by_name(op_name) + + stats = ComparisonStats() + + if not allow_composite: + aten_op = _discover_aten_op(opinfos, device, dtype) + if aten_op is None or not _has_dtensor_support(aten_op): + if verbose: + print(f" ATEN_OP_MAP: {op_name} -> {aten_op} [no_support]") + stats.no_dtensor_support = True + return stats + if verbose: + print(f" ATEN_OP_MAP: {op_name} -> {aten_op} [supported]") + + total_samples = 0 + total_combinations = 0 + skip_reasons: dict[str, int] = defaultdict(int) + + for opinfo in opinfos: + variant = opinfo.variant_test_name + if variant and verbose: + print(f"\n OpInfo variant: {variant}") + + op = opinfo.op + + try: + samples = list(opinfo.sample_inputs(device, dtype)) + except Exception as e: + if verbose: + print(f" Error generating samples: {e}") + continue + + if max_samples: + samples = samples[:max_samples] + + for sample_idx, sample in enumerate(samples): + # Check that SampleInput has tensor inputs and no zero-sized tensors + sample_tensors = extract_tensors_from_sample(sample) + if len(sample_tensors) == 0: + skip_reasons["no tensor inputs"] += 1 + continue + if any(0 in t.shape for _, t in sample_tensors): + skip_reasons["zero-sized tensor"] += 1 + continue + + # Capture all aten ops dispatched for this sample + capture = get_aten_op_for_sample(op, sample, opinfo.name) + if capture.best_match is None: + skip_reasons["no aten op captured"] += 1 + continue + + # Count supported aten ops in the capture + supported_ops = [ + (func, args, kwargs, result) + for func, args, kwargs, result in capture.all_ops + if _has_dtensor_support(func) + ] + num_supported = len(supported_ops) + + if allow_composite: + # Validate each supported aten op individually + if num_supported == 0: + skip_reasons["no supported aten ops"] += 1 + continue + + for func, args, kwargs, result in supported_ops: + gt = _check_ground_truth(result) + if gt is None: + skip_reasons["non-tensor/degenerate aten output"] += 1 + continue + n_samples, n_combos = _validate_aten_op_for_sample( + func, + args, + kwargs, + gt, + world_size, + incorrect_only, + verbose, + sample_idx, + variant, + stats, + sample, + ) + total_samples += n_samples + total_combinations += n_combos + else: + # Default: only validate samples with a single supported aten op + if num_supported != 1: + skip_reasons["non-1:1 aten mapping"] += 1 + continue + + func, args, kwargs, result = supported_ops[0] + gt = _check_ground_truth(result) + if gt is None: + skip_reasons["non-tensor/degenerate aten output"] += 1 + continue + + n_samples, n_combos = _validate_aten_op_for_sample( + func, + args, + kwargs, + gt, + world_size, + incorrect_only, + verbose, + sample_idx, + variant, + stats, + sample, + ) + total_samples += n_samples + total_combinations += n_combos + + stats.total_samples = total_samples + stats.total_combinations = total_combinations + stats.skip_reasons = dict(skip_reasons) + + # In allow_composite mode, check DTensor support after processing + if allow_composite and total_samples == 0 and not skip_reasons: + stats.no_dtensor_support = True + + return stats + + +def get_registered_op_names() -> list[str]: + """Get all op names that have DTensor sharding rules and also have OpInfo. + + Returns OpInfo names (which may differ from aten base names, e.g., + "nn.functional.relu" instead of "relu"). + """ + + propagator = DTensor._op_dispatcher.sharding_propagator + + # Get all registered aten ops + all_registered = set(propagator.op_single_dim_strategy_funcs.keys()) | set( + propagator.op_strategy_funcs.keys() + ) + + # Extract base names (aten.mul.Tensor -> mul) + base_names = set() + for op in all_registered: + parts = str(op).split(".") + if len(parts) >= 2: + base_names.add(parts[1]) + + # Build mappings from OpInfo: both by name and by aten_name + opinfo_by_name = {} + opinfo_by_aten_name: dict[str, list[str]] = {} + for op in op_db: + opinfo_by_name[op.name] = True + opinfo_by_aten_name.setdefault(op.aten_name, []).append(op.name) + + result = set() + for base_name in base_names: + if base_name in opinfo_by_name: + # Direct match (e.g., "add" -> OpInfo named "add") + result.add(base_name) + elif base_name in opinfo_by_aten_name: + # Match via aten_name (e.g., "relu" -> OpInfo "nn.functional.relu" + # which has aten_name="relu") + result.update(opinfo_by_aten_name[base_name]) + + return sorted(result) + + +def report_registrations(verbose: bool = False) -> None: + """Report the number (and optionally full list) of ops by registration method. + + Directly registered categories (mutually exclusive): + - rule: ops registered via register_prop_rule + - op_strategy: ops registered via register_op_strategy + - single_dim_strategy: ops registered via register_single_dim_strategy + + Ops not in any of the above may still be supported at runtime via + DecompShardingStrategy, which traces through the op's decomposition and + propagates placements through the decomposed sub-ops. Whether this + actually works depends on every sub-op having a registered strategy. + We report the decomposition_table entries as a separate (untested) count. + """ + from torch._decomp import decomposition_table + + propagator = DTensor._op_dispatcher.sharding_propagator + + rule_ops = sorted(propagator.op_to_rules.keys(), key=str) + strategy_ops = sorted(propagator.op_strategy_funcs.keys(), key=str) + single_dim_ops = sorted(propagator.op_single_dim_strategy_funcs.keys(), key=str) + + directly_registered = ( + set(propagator.op_to_rules.keys()) + | set(propagator.op_strategy_funcs.keys()) + | set(propagator.op_single_dim_strategy_funcs.keys()) + ) + + # Ops from the explicit decomposition table that aren't directly registered. + # These *may* work via DecompShardingStrategy if all their sub-ops are + # supported, but we can't verify that without tracing each one. + decomp_only_ops = sorted( + (op for op in decomposition_table if op not in directly_registered), + key=str, + ) + + print("=" * 70) + print("DTensor operator registration report") + print("=" * 70) + + print("\nDirectly registered:") + print(f" rule (register_prop_rule): {len(rule_ops):>4}") + print(f" op_strategy (register_op_strategy): {len(strategy_ops):>4}") + print(f" single_dim_strategy: {len(single_dim_ops):>4}") + print(f" total: {len(directly_registered):>4}") + + print(f"\nDecomposition table (not directly registered): {len(decomp_only_ops)}") + print( + " These ops have entries in torch._decomp.decomposition_table but no\n" + " direct DTensor strategy. They may work at runtime via\n" + " DecompShardingStrategy if all decomposed sub-ops are supported.\n" + " Additional ops beyond this count may also be reachable via CIA\n" + " (CompositeImplicitAutograd) decompositions." + ) + + if verbose: + + def _print_ops(label: str, ops: list) -> None: + print(f"\n{label} ({len(ops)}):") + for op in ops: + print(f" {op}") + + _print_ops("rule", rule_ops) + _print_ops("op_strategy", strategy_ops) + _print_ops("single_dim_strategy", single_dim_ops) + _print_ops("decomp table (not directly registered)", decomp_only_ops) + + +if __name__ == "__main__": + # Override common size variables to ensure even sharding across world_size=2. + # These are process-global mutations, but this is a CLI entry point so the + # process exits after running. + opinfo_core.L = 24 # pyrefly: ignore[bad-assignment] + opinfo_core.M = 12 # pyrefly: ignore[bad-assignment] + opinfo_core.S = 4 # pyrefly: ignore[bad-assignment] + opinfo_core.XS = 2 # pyrefly: ignore[bad-assignment] + + common_ops.L = 24 # pyrefly: ignore[bad-assignment] + common_ops.M = 12 # pyrefly: ignore[bad-assignment] + common_ops.S = 4 # pyrefly: ignore[bad-assignment] + common_ops.XS = 2 # pyrefly: ignore[bad-assignment] + + parser = argparse.ArgumentParser( + description="Compare DTensor rules against ground truth" + ) + parser.add_argument( + "--op", + default=None, + help="Operator name(s) to compare (comma-separated, supports glob " + 'patterns, e.g., "relu,add" or "nn.functional.*")', + ) + parser.add_argument( + "--all-registered", + action="store_true", + help="Test all ops with DTensor sharding rules registered", + ) + parser.add_argument( + "--incorrect-only", + action="store_true", + help="Only test DTensor's claimed rules (faster, skips missing detection)", + ) + parser.add_argument( + "--allow-composite", + action="store_true", + help="Validate each supported aten op individually for decomposed ops " + "(default skips non-1:1 aten mappings)", + ) + parser.add_argument("--device", default="cuda", help="Device to use") + parser.add_argument("--dtype", default="float32", help="Dtype to use") + parser.add_argument( + "--world-size", type=int, default=2, help="Simulated world size" + ) + parser.add_argument( + "--max-samples", type=int, default=None, help="Max samples to test" + ) + parser.add_argument( + "--show-repro", + nargs="?", + const=1, + type=int, + default=0, + metavar="N", + help="Show N sample repros per rule (default 1 if flag given, -1 for all)", + ) + parser.add_argument( + "--report", + action="store_true", + help="Show registration statistics (op counts per registration method)", + ) + parser.add_argument( + "--report-full", + action="store_true", + help="Show full op lists with registration statistics (implies --report)", + ) + args = parser.parse_args() + + dtype_map = { + "float32": torch.float32, + "float64": torch.float64, + "float16": torch.float16, + "bfloat16": torch.bfloat16, + } + dtype = dtype_map.get(args.dtype, torch.float32) + + dist.init_process_group("fake", rank=0, world_size=args.world_size) + + if args.report or args.report_full: + report_registrations(verbose=args.report_full) + if not args.all_registered and not args.op: + dist.destroy_process_group() + sys.exit(0) + print() + + try: + if args.all_registered: + op_names = get_registered_op_names() + elif args.op: + patterns = [p.strip() for p in args.op.split(",")] + op_names = resolve_op_names(patterns) + else: + op_names = ["add"] + + # Preamble + display_names = [_format_op_name(n) for n in op_names] + print(f"Testing ops: {', '.join(display_names)}") + if args.allow_composite: + print( + "Mode: allow-composite (validates each aten op in decomposed samples)" + ) + if args.incorrect_only: + print("Mode: incorrect-only (fast)") + print(f"Device: {args.device}, Dtype: {dtype}, World size: {args.world_size}") + + op_results: list[tuple[str, ComparisonStats, float]] = [] + + for i, op_name in enumerate(op_names): + display = _format_op_name(op_name) + + if op_name in SKIP_OPS: + print( + f"\n[{i + 1}/{len(op_names)}] {display}" + f" — skipped: {SKIP_OPS[op_name]}" + ) + continue + + try: + op_start = time.time() + stats = compare_operator( + op_name, + args.device, + dtype, + args.world_size, + args.max_samples, + verbose=True, + incorrect_only=args.incorrect_only, + allow_composite=args.allow_composite, + ) + elapsed = time.time() - op_start + + if stats.no_dtensor_support: + print( + f"\n[{i + 1}/{len(op_names)}] {display}" + f" — skipped: no DTensor support" + ) + continue + + if stats.total_samples == 0 and stats.skip_reasons: + reasons = ", ".join( + f"{r} ({n})" for r, n in stats.skip_reasons.items() + ) + print(f"\n[{i + 1}/{len(op_names)}] {display} — skipped: {reasons}") + continue + + op_results.append((op_name, stats, elapsed)) + + skipped = sum(stats.skip_reasons.values()) + skipped_str = f" ({skipped} skipped)" if skipped else "" + print( + f"\n[{i + 1}/{len(op_names)}] {display}" + f" — Samples: {stats.total_samples}{skipped_str}," + f" Combinations: {stats.total_combinations}" + ) + print("-" * 70) + _print_comparison_summary(stats, show_repro=args.show_repro) + except Exception as e: + print(f"\n[{i + 1}/{len(op_names)}] {display} — Error: {e}") + + if len(op_results) >= 1: + # Summary table + print("\n" + "=" * 70) + print("Summary") + print("=" * 70) + + formatted_results = [ + (_format_op_name(name), stats, elapsed) + for name, stats, elapsed in op_results + ] + + col_widths = { + "op": max(len(name) for name, _, _ in formatted_results), + "tp": 7, + "fp": 9, + "fn": 7, + "time": 6, + } + col_widths["op"] = max(col_widths["op"], 2) # min width + + header = ( + f"{'Op':<{col_widths['op']}} " + f"{'Correct':>{col_widths['tp']}} " + f"{'Incorrect':>{col_widths['fp']}} " + f"{'Missing':>{col_widths['fn']}} " + f"{'Time':>{col_widths['time']}}" + ) + print(header) + print("-" * len(header)) + + total_tp = 0 + total_fp = 0 + total_fn = 0 + total_time = 0.0 + + for name, stats, elapsed in formatted_results: + fp_rules = { + (d.input_placements, d.output_placements) + for d in stats.false_positives + } + fn_rules = { + (d.input_placements, d.output_placements) + for d in stats.false_negatives + } + n_fp = len(fp_rules) + n_fn = len(fn_rules) + total_tp += stats.true_positives + total_fp += n_fp + total_fn += n_fn + total_time += elapsed + + fp_str = str(n_fp) if n_fp else "0" + fn_str = str(n_fn) if n_fn else "0" + + print( + f"{name:<{col_widths['op']}} " + f"{stats.true_positives:>{col_widths['tp']}} " + f"{fp_str:>{col_widths['fp']}} " + f"{fn_str:>{col_widths['fn']}} " + f"{elapsed:>{col_widths['time']}.1f}s" + ) + + print("-" * len(header)) + fp_str = str(total_fp) if total_fp else "0" + fn_str = str(total_fn) if total_fn else "0" + print( + f"{'Total':<{col_widths['op']}} " + f"{total_tp:>{col_widths['tp']}} " + f"{fp_str:>{col_widths['fp']}} " + f"{fn_str:>{col_widths['fn']}} " + f"{total_time:>{col_widths['time']}.1f}s" + ) + finally: + dist.destroy_process_group() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..056dcff0fc8828455cc0d6a9a1e7f787a96d6854 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_ops/utils.py @@ -0,0 +1,726 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import functools +import itertools +import operator +from collections.abc import Callable, Iterable, Sequence +from typing import Any, cast, TypeAlias, TypeVar + +import torch +from torch._prims_common import DimsSequenceType, DimsType +from torch.distributed.tensor._api import DTensor +from torch.distributed.tensor._collective_utils import redistribute_cost +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpSchema, + OpSpec, + OpStrategy, + OutputSharding, + PlacementList, + RuntimeSchemaInfo, + StrategyType, +) +from torch.distributed.tensor.device_mesh import DeviceMesh +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) + + +def _get_registration_wrapper( + registration_fn, + op: torch._ops.OpOverload | list[torch._ops.OpOverload], + schema_info: RuntimeSchemaInfo | None, + arg_names_that_require_specializing_cache_strategy: list[str] | None, +): + def wrapper(impl): + overloads = op if isinstance(op, list) else [op] + for overload in overloads: + curr_schema_info = None + if ( + schema_info is None + and arg_names_that_require_specializing_cache_strategy is not None + ): + specialized_args = [ + a.name + for a in overload._schema.arguments + if a.name in arg_names_that_require_specializing_cache_strategy + ] + if any(specialized_args): + curr_schema_info = RuntimeSchemaInfo( + static_kwargkey=specialized_args + ) + else: + curr_schema_info = schema_info + registration_fn(overload, impl, curr_schema_info) + return impl + + return wrapper + + +# convenient wrapper to register sharding propagation rules +def register_prop_rule( + op: torch._ops.OpOverload | list[torch._ops.OpOverload], + schema_info: RuntimeSchemaInfo | None = None, +) -> Callable[ + [Callable[[OpSchema], OutputSharding]], Callable[[OpSchema], OutputSharding] +]: + return _get_registration_wrapper( + DTensor._op_dispatcher.sharding_propagator.register_sharding_prop_rule, + op, + schema_info, + arg_names_that_require_specializing_cache_strategy=None, + ) + + +# Note: +# using TypeVar here allows the registration decorator to preserve the specific type info of the wrapped strategy, +# while hardcoding the typing on the wrapper (e.g. Callable[[OpSchema], StrategyType]) would mean mypy would treat +# the return value of the wrapped strategy as always being a `StrategyType` even if it were a derived class like +# MyStrategyType(StrategyType). +_OpSchemaT = TypeVar("_OpSchemaT", bound=OpSchema) +_StrategyTypeT = TypeVar("_StrategyTypeT", bound=StrategyType) +_ShardingStrategyFunc: TypeAlias = Callable[[_OpSchemaT], _StrategyTypeT] + + +def register_op_strategy( + op: torch._ops.OpOverload | list[torch._ops.OpOverload], + schema_info: RuntimeSchemaInfo | None = None, +) -> Callable[[_ShardingStrategyFunc], _ShardingStrategyFunc]: + # For every ATen op that accepts any args in this list, + # the arg itself can impact the strides (and potentially the sharding strategy) + # of the output tensor. + # thus, we will detect ATen schemas with any of these args and ensure + # that they get specialized here. + arg_names_that_require_specializing_cache_strategy = [ + "memory_format", + ] + return _get_registration_wrapper( + DTensor._op_dispatcher.sharding_propagator.register_op_strategy, + op, + schema_info, + arg_names_that_require_specializing_cache_strategy, + ) + + +def replicate_op_strategy(op_schema: OpSchema) -> StrategyType: + """ + Fallback strategy all use Replication() + """ + args_strategy = op_schema.args_strategy + kwargs_strategy = op_schema.kwargs_strategy + inputs_strategy = args_strategy + kwargs_strategy + + output_type = [str(ret.type) for ret in op_schema.op._schema.returns] + output_len = output_type.count("Tensor") + # TODO(zpcore): Confirm if view op can be handle properly or not. Prevent + # handling view ops until confirmed. + if op_schema.op.is_view: + raise RuntimeError( + "fallback strategy is unable to handle view ops until confirmed" + ) + if "List[Tensor]" in output_type: + raise RuntimeError( + "fallback strategy is unable to handle ops with List[Tensor] output " + "because size of the list may depend on the op's input value" + ) + + mesh = inputs_strategy[0].mesh + + dim_sharding: PlacementList = [Replicate()] * (output_len + len(inputs_strategy)) + single_dim_placement = [dim_sharding] + return expand_to_full_mesh_op_strategy( + mesh, op_schema, single_dim_placement, input_index=output_len + ) + + +def as_list( + x: list[object] | object, + # pyre-fixme[11]: Annotation `immutable_list` is not defined as a type. +) -> list[object] | torch.fx.immutable_collections.immutable_list: # type: ignore[valid-type] + # During tracing, `aten.sum.dim_IntList` uses `immutable_list` for its args, + # which is an object but treated as a list by the tracer. Therefore, keep + # `immutable_list` intact here as well. + if type(x) is list or isinstance(x, torch.fx.immutable_collections.immutable_list): + return x + else: + return [x] + + +def normalize_dim(dim: int, ndim: int) -> int: + return dim if dim >= 0 else dim + ndim + + +def normalize_dims(dims: DimsType, ndim: int) -> DimsSequenceType: + """Normalize a dim or a sequence of dims, so that they are all positive.""" + if isinstance(dims, int): + dims = (normalize_dim(dims, ndim),) + elif isinstance(dims, list): + dims = [normalize_dim(dim, ndim) for dim in dims] + elif isinstance(dims, tuple): + dims = tuple(normalize_dim(dim, ndim) for dim in dims) + return dims + + +def prod(xs: Iterable[int]) -> int: + return functools.reduce(operator.mul, xs, 1) + + +def is_tensor_shardable( + shape: Sequence[int], + spec: DTensorSpec, + allow_unbacked_sharding: bool | None = None, +) -> bool: + """ + Check if the shape is shardable according to the spec. + + This function handles both `Shard` and `_StridedShard` placements: + - For `Shard`: checks if the tensor dimension size >= number of shards + - For `_StridedShard`: additionally checks if the dimension is shardable after + splitting with the placement's `split_factor` + + allow_unbacked_sharding: determines the fallback value if unbacked shapes are involved, + and the queried shape properties are not statically known. + + e.g. when asking if u0 is shardable on num_shards, and u0 has generic bounds [0, inf], + the behavior of allow_unbacked_sharding is: + + None: will data-dependent error + True: assumes shardability; we return True, allowing zero-size shards at runtime when u0 < num_shards. + False: returns False, and lower-bounding u0, e.g. torch._check(u0 >= num_shards), is needed to enable sharding. + """ + from torch.fx.experimental.symbolic_shapes import ( + free_unbacked_symbols, + guard_or_false, + guard_or_true, + optimization_hint, + ) + + if allow_unbacked_sharding not in [None, True, False]: + raise AssertionError + guard_fn = { + None: bool, + True: guard_or_false, + False: guard_or_true, + }[allow_unbacked_sharding] + + def hint_proves_at_least(size, lower_bound: int) -> bool: + if guard_or_false(size >= lower_bound): + return True + if not ( + isinstance(size, torch.SymInt) + and (shape_env := getattr(size.node, "shape_env", None)) is not None + and (unbacked_symbols := free_unbacked_symbols(size.node.expr)) + and unbacked_symbols.issubset(shape_env.var_to_hint_override.keys()) + and optimization_hint(size, fallback=None) >= lower_bound + ): + return False + + # Shardability is a correctness property, not just an optimization. + # A user-provided unbacked hint can select the shardable branch, but + # the runtime extent must still be checked before the strategy is used. + torch._check(size >= lower_bound) + return True + + # number of shards in each tensor dimension + num_shards = [1] * len(shape) + for i, placement in enumerate(spec.placements): + if _is_shard_like(placement): + shard_dim = placement.dim + if shard_dim >= len(shape): + return False + num_shards[shard_dim] *= spec.mesh.size(i) + if isinstance(placement, _StridedShard): + # make sure tensor dim `shard_dim` is shardable after splitting + # with split_factor + lower_bound = cast(Any, num_shards[shard_dim]) * placement.split_factor + if not hint_proves_at_least(shape[shard_dim], lower_bound) and guard_fn( + shape[shard_dim] < lower_bound + ): + return False + else: + lower_bound = num_shards[shard_dim] + if not hint_proves_at_least(shape[shard_dim], lower_bound) and guard_fn( + shape[shard_dim] < lower_bound + ): + return False + + return True + + +def is_tensor_evenly_shardable(shape: Sequence[int], spec: DTensorSpec) -> bool: + """Check if the shape is evenly shardable according to the spec.""" + # number of shards in each tensor dimension + num_shards = [1] * len(shape) + for i, placement in enumerate(spec.placements): + if _is_shard_like(placement): + shard_dim = placement.dim + if shard_dim >= len(shape): + return False + num_shards[shard_dim] *= spec.mesh.size(i) + if isinstance(placement, _StridedShard): + if ( + cast(Any, shape[shard_dim]) + % (placement.split_factor * num_shards[shard_dim]) + != 0 + ): + return False + else: + if shape[shard_dim] % num_shards[shard_dim] != 0: + return False + return True + + +def is_tensor_evenly_shardable_on_dim( + shape: Sequence[int], spec: DTensorSpec, dim: int +) -> bool: + """Check if the shape is evenly shardable according to the spec on dim.""" + dim = normalize_dim(dim, len(shape)) + + num_shards = 1 + for i, placement in enumerate(spec.placements): + if _is_shard_like(placement) and placement.dim == dim: + num_shards *= spec.mesh.size(i) + if isinstance(placement, _StridedShard): + # _StridedShard._split_tensor first chunks into split_factor + # groups, then into num_shards within each group, so the dim + # must be divisible by the product of both. This is stricter + # than the final num_shards check and implies it. Note: + # num_shards already includes spec.mesh.size(i) from this + # iteration, so the check covers the full shard count. + if cast(Any, shape[dim]) % (placement.split_factor * num_shards) != 0: + return False + + return shape[dim] % num_shards == 0 + + +def is_tensor_dim_sharded(spec: DTensorSpec, dim: int) -> bool: + """Return True if tensor dim is sharded.""" + return any(_is_shard_like(p) and p.dim == dim for p in spec.placements) + + +def is_tensor_partial(spec: DTensorSpec) -> bool: + """Return True if tensor is partial on the mesh.""" + return any(p.is_partial() for p in spec.placements) + + +def infer_broadcast_dims_map( + common_shape: torch.Size, input_shape: torch.Size +) -> list[int]: + # infer the broadcast dims map, where it maps from the common shape dim to the input shape dim + # this is aligned with the broadcast semantics + # e.g. if common_shape = [1, 2, 3, 4] and input_shape = [2, 3, 4], + # broadcast_dims_map will be [-1, 0, 1, 2] + # meaning that dim 0 in the output has no mapping to the input, and dim 1 in the output maps to dim 0 in the input + from torch.fx.experimental.symbolic_shapes import guard_or_false + + common_ndim = len(common_shape) + input_ndim = len(input_shape) + broadcast_dims_map = [-1] * common_ndim + for idx in range(-1, -1 - input_ndim, -1): + if guard_or_false(input_shape[idx] == common_shape[idx]): + broadcast_dims_map[common_ndim + idx] = input_ndim + idx + return broadcast_dims_map + + +def map_placements_after_broadcast( + placements: tuple[Placement, ...], + shape: torch.Size, + broadcast_dims_map: list[int], + partial_to_replicate: bool = False, +) -> tuple[Placement, ...]: + """Map each placement based on the output shape after broadcast.""" + new_placements: list[Placement] = [] + for placement in placements: + if isinstance(placement, Partial): + if partial_to_replicate: + # map the partial placement to replicate + new_placements.append(Replicate()) + else: + new_placements.append(placement) + elif isinstance(placement, Replicate): + new_placements.append(placement) + else: + if not _is_shard_like(placement): + raise AssertionError + shard_dim = normalize_dim(placement.dim, len(shape)) + new_shard_dim = broadcast_dims_map[shard_dim] + if new_shard_dim != -1: + # there's a map from the common shape shard dim to + # the input shape shard dim before broadcasting, + # use that instead + if isinstance(placement, _StridedShard): + new_placements.append( + _StridedShard( + new_shard_dim, split_factor=placement.split_factor + ) + ) + else: + new_placements.append(Shard(new_shard_dim)) + else: + # there's no map between common shape shard dim and + # the input shape shard dim before broadcasting, + # in this case it means implicit broadcasting happen + # in this dim, so we can just mark it as replicate + # and implicit broadcast will broadcast automatically + # to the sharded shape + new_placements.append(Replicate()) + + return tuple(new_placements) + + +def generate_redistribute_costs( + src_strategy: OpStrategy, dst_spec: DTensorSpec +) -> list[float]: + """Generates one row in the 'redistribute_costs' matrix in an OpSpec + The length of the returned list will match the number of strategies in 'src_strategy'. + + Each value in the row is the cost of redistributing from a particular src_strategy to dst_spec. + """ + redistribute_costs: list[float] = [ + redistribute_cost(strat.output_spec, dst_spec) + for strat in src_strategy.strategies + ] + + return redistribute_costs + + +def expand_to_full_mesh_op_strategy( + mesh: DeviceMesh, + op_schema: OpSchema, + single_mesh_dim_strategies: list[PlacementList], + *, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None = None, + input_index: int = 1, + inplace_op: bool = False, + allow_unbacked_sharding: bool | None = None, + allow_uneven_sharding: bool = False, + full_mesh_strategy_filter: Callable[ + [list[DTensorSpec], DTensorSpec | tuple[DTensorSpec | None, ...]], bool + ] + | None = None, + different_mesh_args: list[int] | None = None, +) -> OpStrategy: + """ + Convenience function to allow writing a sharding strategy considering only a single mesh dimension, + and have it expanded combinatorially to all mesh dimensions. + + Args: + mesh (DeviceMesh): the device mesh to expand the strategy to + op_schema (OpSchema): the op schema + single_mesh_dim_strategies (list[PlacementList]): the sharding strategies to expand. The outer list is over + different strategies. The inner PlacementList is over the outputs and inputs of the op. If input_index is 1, + a PlacementList looks like [output_placement, input_placement1, input_placement2, ...]. + output_tensor_meta: tensor metadata for the output(s), used to populate DTensorSpec.tensor_meta field + input_index: the number of outputs of the op, defaults to 1 + inplace_op: whether the op is inplace or not, defaults to False + full_mesh_strategy_filter: callback to filter invalid full-mesh strategy + candidates, defaults to None. + + Example: Let's say `my_op(tensor_x, tensor_y) - > output_tensor` can support sharding or replicating tensor_x, + but always requires tensor_y to be replicated. We can specify these valid combinations ignoring mesh dims. + Then, we can rely on `expand_to_full_mesh_op_strategy` to create every possible combination of these shardings + over multiple mesh dimensions, filtering out any combinations that are invalid based on the actual mesh dim size. + + single_mesh_dim_strategies = [ + # first strategy: return output sharded on first dim, shard tensor_x on its first dim, replicate tensor_y + [Shard(0), Shard(0), Replicate()] + # second strategy: replicate output, and both inputs + [Replicate(), Replicate(), Replicate()] + ] + """ + # Expand the single_mesh_dim_strategies to full mesh dim strategies. + all_mesh_dim_strategies = [single_mesh_dim_strategies] * mesh.ndim + + strategy_combs = itertools.product(*all_mesh_dim_strategies) + + args_strategy = op_schema.args_strategy + kwargs_strategy = op_schema.kwargs_strategy + input_args_strategy = args_strategy + kwargs_strategy + + # Propagate use_strided_shard_as_shard_order from inputs so that + # strategy specs with _StridedShard get the correct flag (and thus + # correct shard_order) at construction time, avoiding shard_order + # mismatches in redistribute_cost computation. + _input_use_strided: bool | None = None + for input_strat in input_args_strategy: + input_spec = input_strat.strategies[0].output_spec + if any(isinstance(p, _StridedShard) for p in input_spec.placements): + _input_use_strided = input_spec.use_strided_shard_as_shard_order + break + + all_strategies = [] + # Track input placements if we skip strategies due to inplace placement mismatch + blocking_inplace_input_placements: tuple[Placement, ...] | None = None + for strategy_comb in strategy_combs: + spec_list: list[DTensorSpec | None] = [] + # Track how many non-None output specs we've seen (for output_tensor_meta indexing). + # This is needed because output_tensor_meta may contain only non-None entries, + # so we can't use position directly when there are None entries in the output. + output_spec_count = 0 + # Track input args separately since not all tensor inputs have OpStrategy + # (e.g., philox_seed/offset in SDPA are scalar tensors without OpStrategy) + input_strategy_counter = 0 + for position, specs in enumerate(zip(*strategy_comb, strict=True)): + if any(spec is None for spec in specs): + if any(spec is not None for spec in specs): + raise AssertionError( + f"{op_schema.op}: strategy position {position} mixes None " + f"with placements across mesh dims: {specs}" + ) + spec_list.append(None) + else: + # Populate tensor_meta field for both output and input specs, + # including for tuple output cases + tensor_meta = None + # Use position to determine output vs input territory + # (position includes None entries, unlike the old spec_index) + if position < input_index: + # This is an output position + if output_tensor_meta is not None: + if isinstance(output_tensor_meta, TensorMeta): + tensor_meta = output_tensor_meta + elif isinstance(output_tensor_meta, (tuple, list)): + if output_spec_count < len(output_tensor_meta): + tensor_meta = output_tensor_meta[output_spec_count] + output_spec_count += 1 + else: + # This is an input position + # Only get tensor_meta if we have a corresponding input_args_strategy entry + if input_strategy_counter < len(input_args_strategy): + tensor_meta = input_args_strategy[ + input_strategy_counter + ].tensor_meta + input_strategy_counter += 1 + + # pyrefly: ignore [bad-argument-type] + use_strided = ( + _input_use_strided + if _input_use_strided is not None + and any(isinstance(p, _StridedShard) for p in specs) + else None + ) + spec_list.append( + DTensorSpec( + mesh, + specs, + tensor_meta=tensor_meta, + use_strided_shard_as_shard_order=use_strided, + ) + ) + + # Skip strategy combinations that would create mixed partial types + # (except sum+avg which commute with each other). + # We check (type, reduce_op) pairs rather than just reduce_op because + # Partial subclasses like _MaskPartial have different reduction semantics + # even when they share the same reduce_op string. + has_mixed_partial = False + for spec in spec_list: + if spec is not None: + partial_kinds = { + (type(p), p.reduce_op) + for p in spec.placements + if isinstance(p, Partial) + } + if len(partial_kinds) > 1: + reduce_ops = {ro for _, ro in partial_kinds} + types = {t for t, _ in partial_kinds} + if not (len(types) == 1 and reduce_ops == {"sum", "avg"}): + has_mixed_partial = True + break + if has_mixed_partial: + continue + + input_specs: list[DTensorSpec] = [ + s for s in spec_list[input_index:] if isinstance(s, DTensorSpec) + ] + + if len(input_specs) != len(input_args_strategy): + raise AssertionError( + f"input_specs({len(input_specs)}) != strategies({len(input_args_strategy)}: " + f"{len(args_strategy)} args + {len(kwargs_strategy)} kwargs)" + ) + + # Note [Multi-mesh args] + # + # Some ops accept args whose DTensor lives on a different DeviceMesh + # than the op's primary compute mesh. We call these "multi-mesh + # args". They arise in fused optimizer ops (e.g. _fused_adam_) + # where *state_steps* is a per-rank scalar counter allocated on a + # smaller sub-mesh (e.g. 1-D DP) while params and grads live on a + # larger mesh (e.g. 2-D DP × TP). + # + # Why must these args be Replicate? + # Sharding implies a specific partitioning of a tensor's data + # across the ranks of a mesh. If a tensor doesn't even *exist* + # on the compute mesh, there is no meaningful way to interpret a + # Shard placement for it. Replicate, on the other hand, is + # mesh-agnostic: every rank already holds the full data, so the + # op can simply read the value regardless of which mesh owns it. + # + # What we do here: + # We preserve the original mesh and Replicate placement for these + # args so the propagator does not try to redistribute them onto + # the compute mesh (which would fail or produce wrong results). + # + # This is distinct from the *element_mesh* handling in + # single_dim_strategy.py, which deals with foreach ops where + # different *elements* in a tensor list may live on different + # sub-meshes (e.g. param group A on 2-D mesh, param group B on + # 1-D mesh). + # TODO: refactor fused_ops handling so that there are no longer + # args on different meshes + if different_mesh_args is not None: + for idx in different_mesh_args: + if idx < len(input_args_strategy): + cross_mesh_input = input_args_strategy[idx] + original_spec = cross_mesh_input.strategies[0].output_spec + if original_spec.mesh != mesh: + if not all(p == Replicate() for p in original_spec.placements): + raise RuntimeError( + f"Cross-mesh input at index {idx} must be Replicate, " + f"but got {original_spec.placements}" + ) + input_specs[idx] = DTensorSpec( + mesh=original_spec.mesh, + placements=original_spec.placements, + tensor_meta=original_spec.tensor_meta, + ) + self_spec = input_args_strategy[0].strategies[0].output_spec + + redistribute_input = self_spec.placements != input_specs[0].placements + mismatching_input_output = ( + spec_list[0] is not None and spec_list[0].placements != self_spec.placements + ) + if inplace_op and (redistribute_input or mismatching_input_output): + # For inplace ops, both the proposed input[0] and the output must + # match self's runtime placement: input[0] because self can't be + # redistributed, output because the result IS self. + if blocking_inplace_input_placements is None: + blocking_inplace_input_placements = self_spec.placements + continue + + # For out= variant ops, output placement must match each out kwarg's + # placement. Kwargs are intentionally not redistributed in dispatch. + if op_schema.is_out_variant_op(): + out_arg_idx = 0 + mismatched_out_arg = False + for argument in op_schema.op._schema.arguments: + if not argument.is_out: + continue + out_arg = op_schema.kwargs_schema.get(argument.name) + if isinstance(out_arg, OpStrategy): + output_spec = spec_list[out_arg_idx] + out_arg_spec = out_arg.strategies[0].output_spec + if ( + output_spec is None + or output_spec.placements != out_arg_spec.placements + ): + mismatched_out_arg = True + break + out_arg_idx += 1 + if mismatched_out_arg: + continue + + output_specs: tuple[DTensorSpec | None, ...] | DTensorSpec | None + has_container_output_meta = isinstance( + output_tensor_meta, Sequence + ) and not isinstance(output_tensor_meta, TensorMeta) + if op_schema.return_type_list_tensor_like() and has_container_output_meta: + # Tensor[] outputs must stay container-shaped even when there is + # only one output spec, so they are distinguishable from Tensor. + output_specs = tuple(spec_list[:input_index]) + elif input_index == 0: + # No outputs (e.g., _linalg_check_errors) + output_specs = None + elif input_index > 1: + output_specs = tuple(spec_list[:input_index]) + else: + if spec_list[0] is not None: + output_specs = spec_list[0] + else: + raise RuntimeError("output spec is None") + + # check all inputs are shardable + if not all( + is_tensor_shardable( + inp.shape, s, allow_unbacked_sharding=allow_unbacked_sharding + ) + or ( + allow_uneven_sharding + and inp.strategies[0].output_spec.placements == s.placements + ) + for inp, s in zip(input_args_strategy, input_specs) + ): + continue + + # perform additional op-specific filtering + # Skip callback for no-output ops (output_specs is None) + if full_mesh_strategy_filter is not None and output_specs is not None: + if not full_mesh_strategy_filter(input_specs, output_specs): + continue + + redistribute_cost = [ + generate_redistribute_costs(input_strategy, input_spec) + for input_strategy, input_spec in zip(input_args_strategy, input_specs) + ] + + strategy = OpSpec( + output_specs=output_specs, + input_specs=input_specs, + redistribute_cost=redistribute_cost, + ) + all_strategies.append(strategy) + + # If all strategies were filtered out due to inplace placement mismatch, + # raise a clear error message instead of returning an empty OpStrategy + # (which would later cause a cryptic "min() arg is an empty sequence" error) + if not all_strategies and blocking_inplace_input_placements is not None: + raise RuntimeError( + f"{op_schema.op}: in-place operations that require placement changes " + f"are not supported. The input has placement {blocking_inplace_input_placements}, " + f"but no valid strategy preserves this placement. " + f"Please use the out-of-place version of this operation instead." + ) + if not all_strategies: + raise RuntimeError( + f"{op_schema.op}: no valid sharding strategy for the input placements." + ) + + return OpStrategy(all_strategies) + + +def shift_shard_dims_after_insert( + placements: Sequence[Placement], insert_dim: int = 0 +) -> Sequence[Placement]: + normalized_placements: list[Placement] = [] + for placement in placements: + if isinstance(placement, _StridedShard) and placement.dim >= insert_dim: + normalized_placements.append( + _StridedShard(placement.dim + 1, split_factor=placement.split_factor) + ) + elif isinstance(placement, Shard) and placement.dim >= insert_dim: + normalized_placements.append(Shard(placement.dim + 1)) + else: + normalized_placements.append(placement) + return normalized_placements + + +def shift_shard_dims_after_remove( + placements: Sequence[Placement], remove_dim: int = 0 +) -> Sequence[Placement]: + normalized_placements: list[Placement] = [] + for placement in placements: + if isinstance(placement, _StridedShard) and placement.dim > remove_dim: + normalized_placements.append( + _StridedShard(placement.dim - 1, split_factor=placement.split_factor) + ) + elif isinstance(placement, Shard) and placement.dim > remove_dim: + normalized_placements.append(Shard(placement.dim - 1)) + else: + normalized_placements.append(placement) + return normalized_placements diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_random.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_random.py new file mode 100644 index 0000000000000000000000000000000000000000..e0166362de2e51921a7793e92f947a1c26d863da --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_random.py @@ -0,0 +1,480 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import contextlib +import warnings +from collections.abc import Sequence +from logging import getLogger +from typing import Optional + +import torch +from torch.distributed._local_tensor import maybe_run_for_local_tensor +from torch.distributed.device_mesh import _get_device_handle, DeviceMesh +from torch.distributed.tensor._dtensor_spec import DTensorSpec +from torch.distributed.tensor.placement_types import _StridedShard, Shard +from torch.types import IntLikeType + + +logger = getLogger(__name__) + +__all__ = [ + "is_rng_supported_mesh", + "manual_seed", + "OffsetBasedRNGTracker", +] + +_rng_tracker: Optional["_RNGStateTracker"] = None + + +def is_rng_supported_mesh(device_mesh: DeviceMesh) -> bool: + """Checks if the current device of ``device_mesh`` supports DTensor's random APIs. + Currently DTensor Random APIs only supports cuda/cuda-like devices. We suggest + users call this API to test the availability before using our random APIs. + + Args: + device_mesh (:class:`DeviceMesh`): The device mesh on which we check if the + random ops APIs are supported. + + Returns: + A bool value. True if ``device_mesh`` supports DTensor Random APIs; False otherwise. + + .. warning:: + Currently we only support correct RNG on cuda/cuda-like devices. + """ + device_handle = _get_device_handle(device_mesh.device_type) + if device_handle and hasattr(device_handle, "set_rng_state"): + return True + else: + # TODO: Logs way too much + warnings.warn( + f"DTensor random operators may not have complete support on {device_mesh.device_type} device mesh", + stacklevel=2, + ) + return False + + +def manual_seed(seed: int, device_mesh: DeviceMesh) -> None: + """Sets the seed for generating random numbers for the calling rank. + + Args: + seed (int): The desired seed. + device_mesh (:class:`DeviceMesh`): The device mesh to set the seed. It is + required that the ``device_mesh`` include the calling rank. This is + to ensure that the SPMD region maintains a synchronous RNG state, which + means no ranks should be initialized with values other than ``seed``. + + Returns: + None + + .. warning:: + :func:`manual_seed` does not check the ``seed`` value correctness. Users must + ensure on their own that the value passed in is the desired ``seed`` for ranks + within ``device_mesh``. + If ``device_mesh`` is a sub-mesh and the calling rank is not a part of it, + ``manual_seed`` will throw an error. + Current implementation only supports a GPU device mesh. + """ + if not is_rng_supported_mesh(device_mesh): + warnings.warn( + "DTensor manual_seed() may not have complete support " + f"on {device_mesh.device_type} device mesh", + stacklevel=2, + ) + return + + # TODO: deprecate this API, but also need to ensure we disable broadcast for PP case, and that's currently + # bundled together with this API. See torchtitan/distributed/utils.py:set_determinism + # warnings.warn( + # "DTensor manual_seed() is deprecated, since DTensor no longer maintains a separate copy of generator state. " + # "Use `torch.manual_seed` instead" + # ) + # Note: we still need to ensure setting `run_state_sync=False` to support the pp case + + # instantiate a RNG tracker if haven't. By default DTensor uses an + # OffsetBasedRNGTracker to perform random operators. + global _rng_tracker + if not _rng_tracker: + _rng_tracker = OffsetBasedRNGTracker(device_mesh, run_state_sync=False) + + if device_mesh.get_coordinate() is None: + raise RuntimeError( + "manual_seed requires the current rank to be a part of the device mesh " + "otherwise DTensor RNG state on the rank will not be initialized and " + "the behavior of DTensor random ops is undefined." + ) + + # DTensor no longer maintains a copy of rng state. manual seed on dtensor is the same thing + # as manual seed on torch. + # + # torch.manual_seed will handle LocalTensor mode correctly by + # iterating through all ranks if seed is a LocalIntNode. + torch.manual_seed(seed) + + +class _PhiloxState: + """ + Convenience accessor for interpreting the packed bits of (seed: uint64, offset: uint64) in the philox state, + which for some reason is actually exposed as a size-16 uint8 tensor. + + The state is always moved to .cpu since it is necessary for it to be on CPU before applying it back to a generator. + """ + + def __init__(self, state: torch.Tensor): + self._state = state.to("cpu") + + @property + def state(self): + return self._state + + @property + def offset(self) -> torch.Tensor: + return self._state[8:].view(dtype=torch.int64) + + @offset.setter + def offset(self, offset: torch.Tensor) -> None: + if offset.numel() != 1: + raise AssertionError + self._state[8:] = offset.view(torch.uint8) + + @property + def seed(self) -> torch.Tensor: + return self._state[:8].view(dtype=torch.uint64) + + @seed.setter + def seed(self, seed: torch.Tensor) -> None: + if seed.numel() != 1: + raise AssertionError + self._state[:8] = seed.view(torch.uint8) + + +class _RNGStateTracker: + """ + _RNGStateTracker stores Random Number Generator (RNG) state (a ByteTensor object) + in a dict, mapping from a corresponding tag to each state tensor. It also provides + a set of convenient utility methods to help access/modify the state tensors. The most + important interface is _distribute_region which will be used when DTensor executes + a random op (an operator that calls RNG). + """ + + def __init__(self, device: torch.device): + self._device = device + self._device_handle = _get_device_handle(self._device.type) + if not (self._device_handle and self._device_handle.is_available()): + raise RuntimeError( + f"{self.__class__.__name__} instantiation requires the presence of " + f"{device.type} device but couldn't find." + ) + self._use_distribute_region = True + + @property + def distribute_region_enabled(self) -> bool: + return self._use_distribute_region + + @distribute_region_enabled.setter + def distribute_region_enabled(self, value) -> None: + self._use_distribute_region = value + + def _distribute_region( + self, spec: DTensorSpec, generator: torch.Generator | None = None + ): + pass + + def _manual_seed(self, parallel_seed: int) -> None: + pass + + +class OffsetBasedRNGTracker(_RNGStateTracker): + """ + This subclass of ``_RNGStateTracker`` defines the default policy of how RNG states + should be shared and synchronized among all ranks to respect the semantics of DTensor + random operators. + + note: _RNGStateTracker only supports cuda/cuda-like device. + """ + + def __init__( + self, + device_mesh: DeviceMesh, + run_state_sync: bool = True, + ): + super().__init__(_resolve_device(device_mesh=device_mesh)) + if self._device_handle is None: + raise AssertionError + # DTensor RNG tracker so far only supports CUDA/CUDA-like devices + if self._device.type == "cpu": + raise RuntimeError( + f"{self.__class__.__name__} instantiation requires the presence of " + f"CUDA/CUDA-like/XPU device. Got {self._device.type} instead." + ) + + if run_state_sync: + rng_state = self._get_device_state() + # synchronize RNG state using rank 0's current one + torch.distributed.broadcast(rng_state, 0) + my_rng_state = self._get_device_state() + if not all(my_rng_state == rng_state): + logger.warning( + "DTensor is synchronizing RNG states of every rank with the state from rank 0. " + "This behavior is deprecated. " + "Please call `torch.manual_seed()` on every rank that participates in SPMD DTensor Operations with " + "the same seed. If using Pipeline Parallelism, each pipeline state would use a different seed, " + "but all ranks belonging to one pipeline stage would use the same seed." + ) + self._set_device_state(rng_state) + + def _get_device_state(self) -> torch.Tensor: + if self._device.type == "hpu": + self._device_handle.set_rng_ctx("philox") + rng_state = self._device_handle.get_rng_state().to(self._device) + if self._device.type == "hpu": + self._device_handle.unset_rng_ctx("philox") + return rng_state + + def _set_device_state(self, state: torch.Tensor): + # It seems that the underlying generator wants a cpu tensor but the dtensor code expects `_get_device_state` + # to convert to a 'device' tensor, probably because we may use it with our backend comms for sync/debug + # for now, we just convert back to cpu here to make sure it always works. + if self._device.type == "hpu": + self._device_handle.set_rng_ctx("philox") + self._device_handle.set_rng_state(state.to("cpu")) + if self._device.type == "hpu": + self._device_handle.unset_rng_ctx("philox") + + @contextlib.contextmanager + def _distribute_region( + self, spec: DTensorSpec, generator: torch.Generator | None = None + ): + from torch.distributed._local_tensor import maybe_enable_local_tracker + + if local_tracker_context := maybe_enable_local_tracker( + self._device.type, self.distribute_region_enabled, spec, generator + ): + with local_tracker_context: + yield + return + + # regular (non-LocalTensor) mode + if generator is not None: + # This is a little hacky, but for any user-passed generator, we store its state under a unique key, + # not because we need to keep a copy of it but because its the easiest way to make it work with the + # existing set/get APIs. We also ensure we remove it from rng_states after each _distribute_region. + state = _PhiloxState(generator.get_state()) + else: + state = _PhiloxState(self._get_device_state()) + + if self.distribute_region_enabled: + if self._device.type == "hpu": + self._device_handle.set_rng_ctx("philox") + old_offset = state.offset.clone() + self._set_pre_op_offset(state, spec) + with torch.random.fork_rng( + devices=[self._device], device_type=self._device.type + ): + if self._device_handle is None: + raise AssertionError + self._device_handle.set_rng_state(state.state) + try: + yield # execute the region code + finally: + # update offset to synchronize among ranks + self._set_post_op_offset(state, spec, old_offset) + if self._device.type == "hpu": + self._device_handle.unset_rng_ctx("philox") + else: + yield + + if generator is not None: + # ensure we (a) propagate the state advancement back to the user's RNG so its visible and impacts any future + # usage of that RNG (dtensor or non-dtensor), (b) drop it from our own cache so that if the user updates + # the seed value in their rng and uses it with DTensor again, we always use the latest value + generator.set_state(state.state) + else: + self._set_device_state(state.state) + + def _set_pre_op_offset(self, state: _PhiloxState, spec: DTensorSpec) -> None: + """Set the starting RNG offset for current device's local shard before actual + op execution. The pre_op_offset value should start from the current RNG offset + and increment by the size of local shard until it reaches the size of the whole + DTensor. For different ranks that hold the same DTensor shard, their pre_op_offset + will be the same. + + Args: + state (:class:`Tensor`): The generator state to modify + spec (:class:`DTensorSpec`): the spec of the DTensor object on which + we prepare the offset for running random ops. + + Returns: + None + + .. warning:: + Note that, current implementation does not consider DTensor's continguity. + + Example: + take a DTensor of shape [8, 16] as an example. Assume that the DTensor + is placed on a device mesh with placements ([Shard(1), Replicate(), Shard(0)]), + and the mesh is: + [[[0, 1], [2, 3]], [[4, 5], [6, 7]]] + ``spec.mesh.get_coordinate()`` provides the coordinate of the current rank + in the mesh. For example, the coordinate of rank 5 is (1, 0, 1). + + Another concept to introduce besides rank coordinate is shard coordinate. + Each rank holds a local shard of the DTensor. In the example, the DTensor + is partitioned into 4 [4, 8] shards. The first shard has 2 replicas and + rank 0 (coord (0, 0, 0)) and rank 2 (coord (0, 1, 0)) have 1 replica each. + That being said, the local shard on rank 0 and rank 2 correspond to the same + shard of the DTensor. To denote each DTensor shard, we use a shard coordinate + (in the example, it will be a tuple (i, j) where shard (i, j) has the slice + DTensor[4 * i : 4 * (i + 1), 8 * j : 8 * (j + 1)], 0 <= i < 2, 0 <= j < 2). + + Once we have rank coordinate and shard coordinate, we can calculate on each rank + what shard of the DTensor the rank holds, with the help of dim_map. The dim_map + of the above DTensor is [2, 0] so the shard coordinate of a rank with rank coord + (x, y, z) is simply (z, x) by taking(rank_coord[dim_map[0]],rank_coord[dim_map[1]]). + Following this calculation, + rank 0 and rank 2 holds the shard of coord (0, 0); + rank 1 and rank 3 holds the shard of coord (0, 1); + rank 4 and rank 6 holds the shard of coord (1, 0); + rank 5 and rank 7 holds the shard of coord (1, 1); + + The last value to calculate before obtaining the starting offset is the shard linear index. + The starting offset for each rank will be its shard_linear_index * local_tensor_numel. + """ + start_offset_incr, _ = self._compute_rng_offsets(spec) + state.offset = state.offset + start_offset_incr + + def _set_post_op_offset( + self, state: _PhiloxState, spec: DTensorSpec, old_offset: torch.Tensor + ) -> None: + """Sets the RNG to a synchronized state after running the local random op. Every + rank should set its RNG offset to `old_offset + DTensor.numel()` where old_offset is + the offset before calling `set_pre_op_offset` i.e. the offset before running DTensor + random ops. + + Args: + state (:class:`Tensor`): The generator state to modify. + spec (:class:`DTensorSpec`): the spec of the DTensor object on which + we post-process the offset for running random ops. + + Returns: + None + """ + _, end_offset_incr = self._compute_rng_offsets(spec) + state.offset = old_offset + end_offset_incr + + def _compute_rng_offsets(self, spec: DTensorSpec) -> tuple[int, int]: + """Compute the RNG offset increments for a distributed random op. + + These values are derived from mesh topology, placements, and tensor shape, + and are static for a given compiled graph. They can be burned into the graph + as integer constants rather than keeping the DTensorSpec around at runtime. + + Returns: + (start_offset_incr, end_offset_incr) — both aligned to multiples of 4. + """ + from torch.distributed.tensor._ops.utils import prod + + mesh = spec.mesh + mesh_coordinate = [mesh._sym_get_coordinate(i) for i in range(mesh.ndim)] + + shard_idx_by_dim, total_num_shards_by_dim = _calc_shard_info( + mesh_coordinate, spec + ) + shard_linear_idx = self._calc_shard_linear_idx( + shard_idx_by_dim, total_num_shards_by_dim + ) + local_size = prod(_calc_first_shard_size(spec)) + # pytorch: offset must be multiple of 4 + # source: aten/src/ATen/cuda/CUDAGeneratorImpl.cpp + start_offset_incr = (shard_linear_idx * local_size + 3) // 4 * 4 + end_offset_incr = (prod(spec.shape) + 3) // 4 * 4 + + return start_offset_incr, end_offset_incr # type: ignore[bad-return] + + def _calc_shard_linear_idx( + self, shard_coord: Sequence[IntLikeType], shard_size: Sequence[IntLikeType] + ) -> IntLikeType: + return _calc_shard_linear_idx(shard_coord, shard_size) + + +def _calc_first_shard_size(spec: DTensorSpec) -> list[int]: + local_size_on_rank_0 = list(spec.shape) + for idx, placement in enumerate(spec.placements): + if isinstance(placement, Shard | _StridedShard): + mesh_dim_size = spec.mesh.size(idx) + shard_dim = placement.dim + local_size_on_rank_0[shard_dim], _ = placement._local_shard_size_and_offset( + spec.shape[shard_dim], + mesh_dim_size, + 0, + ) + return local_size_on_rank_0 + + +def _calc_shard_info( + mesh_coordinate: Sequence[IntLikeType], spec: DTensorSpec +) -> tuple[list[IntLikeType], list[IntLikeType]]: + mesh = spec.mesh + # note: dim_map does not allow double sharding which is the FSDP(fully_shard)+TP + # case. Replace the custom logic with dim_map once we support it. + dim_map: list[int | list[int]] = [-1] * spec.ndim + for i, placement in enumerate(spec.placements): + if isinstance(placement, Shard | _StridedShard): + shard_dim = placement.dim + if dim_map[shard_dim] == -1: + dim_map[shard_dim] = [i] + else: + mesh_dim_list = dim_map[shard_dim] + if not isinstance(mesh_dim_list, list): + raise AssertionError + mesh_dim_list.append(i) + + # Compute shard coordinate: + # The coordinate on each tensor dim is a tuple (idx, range) + # If a DTensor is partitioned on its dim i into n shards, and the current rank + # holds the j-th, then its shard coordinate will be (idx=j, range=n) on dim i + mesh_size = mesh.shape + shard_idx_by_dim = [] + total_num_shards_by_dim: list[ + IntLikeType + ] = [] # total number of shards on each tensor dim + for mesh_dim in dim_map: + shard_idx: IntLikeType = 0 + total_num_shards: IntLikeType = 1 + # the tensor dim is sharded on more than 1 mesh dim + if isinstance(mesh_dim, list): + rank_coord = [mesh_coordinate[d] for d in mesh_dim] + num_shards = [mesh_size[d] for d in mesh_dim] + # compute the shard idx and total number of shards + for idx, size in zip(rank_coord, num_shards): + shard_idx = shard_idx * size + idx + total_num_shards *= size + + shard_idx_by_dim.append(shard_idx) + total_num_shards_by_dim.append(total_num_shards) + return shard_idx_by_dim, total_num_shards_by_dim + + +def _calc_shard_linear_idx( + shard_coord: Sequence[IntLikeType], shard_size: Sequence[IntLikeType] +) -> IntLikeType: + # compute shard linear index + shard_linear_idx: IntLikeType = 0 + shard_coord_stride: IntLikeType = 1 + for idx, size in zip(reversed(shard_coord), reversed(shard_size)): + shard_linear_idx += idx * shard_coord_stride + shard_coord_stride *= size + + return shard_linear_idx + + +def _resolve_device(device_mesh: DeviceMesh) -> torch.device: + device_type = device_mesh.device_type + device_handle = _get_device_handle(device_type) + if device_handle is None: + raise AssertionError + device_idx = device_mesh.get_rank() % device_handle.device_count() + + @maybe_run_for_local_tensor + def get_device(device_idx): + return torch.device(f"{device_type}:{device_idx:d}") + + return get_device(device_idx) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_redistribute.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_redistribute.py new file mode 100644 index 0000000000000000000000000000000000000000..1069c61a232e08a52392f5441ebb02afb4e9375a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_redistribute.py @@ -0,0 +1,2112 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import contextlib +import dataclasses +import itertools +import logging +import math +import weakref +from collections import defaultdict +from collections.abc import Sequence +from functools import cache +from typing import Any, cast, TypedDict + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.tensor._api as dtensor +from torch.distributed._functional_collectives import _are_we_tracing +from torch.distributed._mesh_layout import _MeshLayout +from torch.distributed.tensor._collective_utils import one_step_redistribute_cost +from torch.distributed.tensor._dtensor_spec import ( + _StridedShardNotDecodableError, + DTensorSpec, + ShardOrder, + ShardOrderEntry, + TensorMeta, +) +from torch.distributed.tensor._utils import assert_no_mixed_partial_types +from torch.distributed.tensor.device_mesh import DeviceMesh +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _StridedShard, + Partial, + Placement, + Replicate, + Shard, +) +from torch.types import FloatLikeType, IntLikeType +from torch.utils._debug_mode import get_active_debug_mode +from torch.utils._typing_utils import not_none + + +logger = logging.getLogger(__name__) + +# Global configuration flag to control the redistribution planning strategy. +# When True, forces the graph-based algorithm using Dijkstra's shortest path. +# When False, prefers the greedy algorithm for faster planning. Uses the graph-based algorithm +# only when necessary to support strided-shard redistribution +_FORCE_MIN_COST_REDISTRIBUTION_PLAN: bool | None = None + +# Global kill switch to disable the transform optimization pass in +# _optimize_transform_infos. When True, the optimization that merges +# consecutive same-type collectives into flattened operations is skipped, +# and the unmodified transform_infos list is returned as-is. +_DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION: bool = False + + +def _redistribute_cost_sort_key(cost: FloatLikeType) -> float: + if isinstance(cost, float): + return cost + if isinstance(cost, torch.SymFloat): + expr = cost.node.expr + else: + return float(cost) + + free_symbols = expr.free_symbols + if not free_symbols: + return float(expr) + + shape_env = not_none(cost.node.shape_env) + hint_overrides = shape_env.var_to_hint_override + if free_symbols.issubset(hint_overrides): + return float(expr.xreplace(hint_overrides)) + + return float(shape_env.optimization_hint(expr, fallback=None)) + + +@contextlib.contextmanager +def use_min_cost_redistribution_plan(enabled: bool = True): + """ + Context manager to control the redistribution planning strategy for DTensor operations. + + This context manager allows you to choose between two algorithms for computing the + sequence of collective operations needed to redistribute a DTensor from one placement + to another: + + - **Graph-based**: Uses Dijkstra's algorithm to find the minimum-cost path + through all possible placement transformations. This approach considers the global + cost of all collective operations and finds the optimal sequence. Best for complex + redistribution patterns where reducing communication cost and memory overhead is critical. + + - **Greedy**: Uses a heuristic approach that makes locally optimal choices + at each step. This is faster to compute but may not produce the globally optimal + transformation sequence. Best for simple redistribution patterns or when planning + speed is more important than optimal communication. + + **Default Behavior (without this context manager):** + + When this context manager is NOT used, the algorithm selection follows this priority: + + 1. **Non-default shard orders** + → Always use graph-based algorithm (required for correctness) + + 2. **Explicit `use_graph_based_transform` parameter** to `_gen_transform_infos_non_cached` + → Use the specified algorithm (True = graph-based, False = greedy) + + 3. **No explicit parameter** (default case) + → Use greedy algorithm for faster planning + + **Behavior with this context manager:** + + This context manager overrides the default selection by setting the global flag + `_FORCE_MIN_COST_REDISTRIBUTION_PLAN`, which takes precedence over the explicit + `use_graph_based_transform` parameter (but not over non-default shard order requirements). + + **Cache Considerations:** + + The redistribution planner caches transform info for performance via the `@cache` + decorator on `_gen_transform_infos`. If you need to change the algorithm selection + for the same input specs, clear the cache using `_gen_transform_infos.cache_clear()` + to ensure the new setting takes effect and doesn't reuse cached results from a + previous run. + + Args: + enabled (bool): If True, forces the use of the graph-based algorithm. + If False, forces the use of the greedy algorithm. + Default: True + """ + global _FORCE_MIN_COST_REDISTRIBUTION_PLAN + + old_value = _FORCE_MIN_COST_REDISTRIBUTION_PLAN + _FORCE_MIN_COST_REDISTRIBUTION_PLAN = enabled + try: + yield + finally: + _FORCE_MIN_COST_REDISTRIBUTION_PLAN = old_value + + +@contextlib.contextmanager +def disable_redistribute_transform_optimization(disabled: bool = True): + """ + Context manager to disable the transform optimization pass that merges + consecutive same-type collectives into single flattened operations. + + When the optimization is disabled, ``_optimize_transform_infos`` becomes a + no-op and returns the original list of ``_TransformInfo`` objects unchanged. + This is useful for debugging or isolating issues related to the flattened + collective merging logic. + + The flag can also be set directly:: + + torch.distributed.tensor._redistribute._DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION = True + + Args: + disabled (bool): If True (default), disables the optimization. + If False, explicitly enables it (the normal default). + """ + global _DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION + + old_value = _DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION + _DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION = disabled + try: + yield + finally: + _DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION = old_value + + +@dataclasses.dataclass(frozen=True, slots=True) +class _TransformInfo: + mesh_dim: int + src_dst_placements: tuple[Placement, Placement] + # logical_shape on this mesh dimension + logical_shape: Sequence[IntLikeType] + + def __post_init__(self): + if self.mesh_dim < 0: + raise AssertionError + if self.src_dst_placements[0] == self.src_dst_placements[1]: + raise AssertionError( + "TransformInfo should only be created if it is an op with some effect, not a no-op" + ) + + def _comm_type_key(self) -> str | None: + """ + Return a key for grouping transforms by communication type. + + Returns None for local ops (no communication needed), or a string + that identifies the collective type for potential grouping/merging. + """ + src, dst = self.src_dst_placements + if src.is_partial() and dst.is_replicate(): + return "all_reduce" + elif src.is_partial() and _is_shard_like(dst): + return "reduce_scatter" + elif _is_shard_like(src) and dst.is_replicate(): + return "all_gather" + elif _is_shard_like(src) and _is_shard_like(dst): + return "all_to_all" + else: + # Local ops (Replicate->Shard, Replicate->Partial, noop, etc.) + return None + + +@dataclasses.dataclass(frozen=True, slots=True) +class _FlattenedTransformInfo(_TransformInfo): + """ + Represents a flattened transform that combines multiple mesh dimensions + into a single collective operation using a flattened DeviceMesh. + + Note: inherits the fields from _TransformInfo. Gets an __init__ with parent fields, followed by child fields, + and runs parent validation (post_init) + """ + + # The flattened DeviceMesh to use for the collective operation + mesh: DeviceMesh + # The mesh dimensions from the original mesh that are being flattened (for debugging) + original_mesh_dims: tuple[int, ...] + # Scale factor for merged sum/avg partials (product of avg mesh dim sizes) + # When merging sum/avg partials, we use sum for the collective and divide by this afterward + avg_scale: int | None = None + + def __post_init__(self) -> None: + _TransformInfo.__post_init__(self) + if self.avg_scale is not None: + if not self.avg_scale > 1: + raise AssertionError( + f"avg_scale must be > 1 if set, got {self.avg_scale}" + ) + + +def _update_shard_order_and_placements( + transform_info: _TransformInfo, + current_placements: list[Placement], + shard_order_dict: dict[int, list[int]], +) -> None: + """ + Update current_placements and shard_order_dict in-place to reflect the + effect of a single transform step. + """ + src_placement, dst_placement = transform_info.src_dst_placements + + if isinstance(transform_info, _FlattenedTransformInfo): + mesh_dims = transform_info.original_mesh_dims + else: + mesh_dims = (transform_info.mesh_dim,) + + if isinstance(src_placement, Shard | _StridedShard): + src_dim = src_placement.dim # type: ignore[attr-defined] + removed_dim = set() + for _ in mesh_dims: + if len(shard_order_dict[src_dim]) == 0: + raise ValueError( + "Invalid shard_order update. No entries left to pop for src_dim " + f"{src_dim}. transform_info={transform_info}, " + f"current_placements={current_placements}, " + f"shard_order={shard_order_dict}" + ) + removed_dim.add(shard_order_dict[src_dim].pop()) + + if not set(mesh_dims) == removed_dim: + raise ValueError( + "Mismatch between expected and removed mesh dims during shard_order " + "update. Expected to remove " + f"{set(mesh_dims)}, but removed {removed_dim}. " + f"transform_info={transform_info}, " + f"current_placements={current_placements}, " + f"shard_order={shard_order_dict}" + ) + if isinstance(dst_placement, Shard | _StridedShard): + dst_dim = dst_placement.dim # type: ignore[attr-defined] + if dst_dim not in shard_order_dict: + shard_order_dict[dst_dim] = [] + for mesh_dim in mesh_dims: + shard_order_dict[dst_dim].append(mesh_dim) + + for mesh_dim in mesh_dims: + current_placements[mesh_dim] = dst_placement + + +def _get_flattened_mesh_by_layout_impl( + mesh: DeviceMesh, mesh_dims: tuple[int, ...] +) -> DeviceMesh | None: + """ + Query for an explicitly created flattened mesh using layout comparison. + + Searches root_mesh._flatten_mapping for a mesh whose layout matches + the expected flattened layout for the given dims. Pure Python layout math. + """ + root_mesh = mesh._get_root_mesh() + mesh_dim_names = mesh.mesh_dim_names + + if mesh_dim_names is None: + return None + + # Convert mesh dim indices to dim names + dim_names = tuple(mesh_dim_names[i] for i in mesh_dims) + + # Compute expected layout WITHOUT creating a submesh (avoids tracing issues) + # _get_slice_mesh_layout does pure layout math, no tensor operations + sliced_layout = mesh._get_slice_mesh_layout(dim_names) + expected_layout = _MeshLayout([sliced_layout.collapse()]) + + # Search existing flattened meshes by comparing layouts + for flattened_mesh in root_mesh._flatten_mapping.values(): + if flattened_mesh._layout == expected_layout: + return flattened_mesh + + return None + + +def _get_flattened_mesh_by_layout( + mesh: DeviceMesh, mesh_dims: tuple[int, ...] +) -> DeviceMesh | None: + """ + Query for an explicitly created flattened mesh using layout comparison. + + When tracing with compile_on_one_rank, delegates to a custom op so the + flattened mesh appears as a call_function node derived from mesh (a graph + input) rather than as a get_attr constant holding an unpicklable + ProcessGroup. + """ + import torch.compiler.config + + if _are_we_tracing() and torch.compiler.config.compile_on_one_rank: + # Pre-check: the custom op can't return None (torch.library doesn't + # support Optional opaque return types), so guard here first. + if _get_flattened_mesh_by_layout_impl(mesh, mesh_dims) is None: + return None + from torch.distributed._ops import device_mesh as _ # noqa: F401 + + return torch.ops.device_mesh._get_flattened_submesh(mesh, list(mesh_dims)) + + return _get_flattened_mesh_by_layout_impl(mesh, mesh_dims) + + +# Track (mesh_hash, mesh_dims, reason) we've already warned about to avoid repeated warnings +_warned_flatten_issues: set[tuple[int, tuple[int, ...], str]] = set() + + +def _warn_flatten_optimization_not_possible( + device_mesh: DeviceMesh, + mesh_dims: tuple[int, ...], + src_placements: tuple[Placement, ...], + dst_placements: tuple[Placement, ...], + num_ops: int, + comm_type: str, + reason: str, +) -> None: + """ + Warn once per (mesh, dims, reason) about inability to flatten operations. + + Args: + device_mesh: The device mesh being used + mesh_dims: Tuple of mesh dimensions that could not be flattened + src_placements: Source placements for the redistribution + dst_placements: Target placements for the redistribution + num_ops: Number of sequential operations that will be performed + comm_type: Type of collective operation (e.g., "reduce_scatter") + reason: Either "no_flattened_mesh" or "uneven_tensor_shape" + """ + cache_key = (hash(device_mesh), mesh_dims, reason) + if cache_key in _warned_flatten_issues: + return + _warned_flatten_issues.add(cache_key) + + mesh_dim_names = device_mesh.mesh_dim_names + if mesh_dim_names is not None: + dim_names = [mesh_dim_names[d] for d in mesh_dims] + dims_str = ", ".join(f'"{name}"' for name in dim_names) + else: + dims_str = f"dims {', '.join(str(d) for d in mesh_dims)} of {device_mesh}" + + common_warning = ( + "While redistributing from %s to %s, %d sequential %s " + "operations will be performed. This is suboptimal: " + "multiple collective operations have higher latency " + "(separate kernel launches and synchronization points) " + "and may give inconsistent results between ranks due to different reduction orders. %s" + ) + + if reason == "no_flattened_mesh": + reason_msg = f"To optimize, flatten mesh dimensions [{dims_str}] so DTensor can use a single operation instead." + elif reason == "uneven_tensor_shape": + reason_msg = ( + " Unfortunately, because the tensor dimension is not evenly divisible by the product of " + "the mesh dim sizes that would need to be flattened for the optimization to work, it can not be optimized.", + ) + elif reason == "non_ascending_mesh_dims": + reason_msg = ( + f"it is not possible to merge non-ascending order {comm_type} operations." + ) + else: + raise AssertionError(f"Unexpected reason: {reason}") + + logger.warning( + common_warning, src_placements, dst_placements, num_ops, comm_type, reason_msg + ) + + +def _optimize_transform_infos( + transform_infos: list[_TransformInfo], + device_mesh: DeviceMesh, + src_placements: tuple[Placement, ...], + dst_placements: tuple[Placement, ...], +) -> list[_TransformInfo | _FlattenedTransformInfo]: + """ + Optimize transform infos by merging consecutive same-type collectives into + a single flattened operation when a matching flattened DeviceMesh exists. + + Merging requirements: + - Operations must be consecutive in the transform list (no reordering). + Notably, redistributing from P, P, P -> R, S, R is not optimized here and cannot be optimized due to + optimization needing to fuse non-contiguous reductions, leaving this pattern vulnerable to numerics issues and + suboptimal perf + - Operations must have the same comm type (e.g., all allgather or all reduce_scatter) + - Operations must have identical src_dst_placements (e.g., can't merge + Partial->Shard(0) with Partial->Shard(1)) + - A flattened mesh covering the relevant dimensions must exist + - For reduce_scatter, tensor dim must be evenly divisible by flattened mesh size + + For nested sharding, the merged operation uses the logical_shape from the + outermost mesh dimension (smallest mesh_dim index) which represents the + global tensor shape needed for correct padding/unpadding. + + TODO: + - all_to_all operations are excluded from merging, but it may be possible to merge them in some cases. + + """ + if len(transform_infos) < 2: + return transform_infos + + if _DISABLE_REDISTRIBUTE_TRANSFORM_OPTIMIZATION: + return transform_infos + + # Comm types that are safe to merge (all_to_all excluded for now) + MERGEABLE_COMM_TYPES = frozenset({"all_gather", "all_reduce", "reduce_scatter"}) + + def is_mergeable(key: str | None) -> bool: + """Check if a comm type key represents a mergeable operation.""" + return key in MERGEABLE_COMM_TYPES + + def are_placements_mergeable( + p1: tuple[Placement, Placement], p2: tuple[Placement, Placement] + ) -> bool: + """ + Check if two src_dst_placements can be merged. + + Allows merging of Partial("sum") and Partial("avg") since they can be + combined: perform sum reduction, then scale by avg mesh dims afterward. + """ + if p1 == p2: + return True + + src1, dst1 = p1 + src2, dst2 = p2 + + # Destinations must match exactly + if dst1 != dst2: + return False + + # Both sources must be partial + if not (src1.is_partial() and src2.is_partial()): + return False + + # Only sum and avg can be merged (both use sum reduction, avg just scales) + partial1 = cast(Partial, src1) + partial2 = cast(Partial, src2) + mergeable_reduce_ops = {"sum", "avg"} + return ( + partial1.reduce_op in mergeable_reduce_ops + and partial2.reduce_op in mergeable_reduce_ops + ) + + def try_create_flattened( + infos: list[_TransformInfo], + ) -> tuple[_FlattenedTransformInfo | None, str | None]: + """ + Try to create a flattened transform from 2+ same-type transforms. + + Returns (result, failure_reason) where: + - result is the FlattenedTransformInfo if successful, None otherwise + - failure_reason is None if successful, or one of: + - "too_few_transforms": Less than 2 transforms provided + - "no_flattened_mesh": No flattened mesh exists for the required dimensions + - "uneven_tensor_shape": For reduce_scatter, tensor dim not evenly divisible + """ + if len(infos) < 2: + return None, "too_few_transforms" + + # All transforms must have mergeable src_dst_placements + # (e.g., can't merge Partial->Shard(0) with Partial->Shard(1)) + first_placements = infos[0].src_dst_placements + comm_type = infos[0]._comm_type_key() + if not all( + are_placements_mergeable(info.src_dst_placements, first_placements) + for info in infos + ): + raise AssertionError( + "All transforms must have mergeable src_dst_placements" + ) + mesh_dims = tuple(info.mesh_dim for info in infos) + sorted_mesh_dims = tuple(sorted(mesh_dims)) + + # For reduce_scatter and all_gather, order matters for correctness. + # Flattened meshes only exist for ascending dim order. + if comm_type == "reduce_scatter": + # For reduce_scatter: the transform order determines the operation sequence. + # If transforms are in order (1, 0) but flattened mesh is (0, 1), we can't flatten. + if mesh_dims != sorted_mesh_dims: + return None, "non_ascending_mesh_dims" + elif comm_type == "all_gather": + # For all_gather: transforms come from planner in innermost-to-outermost order + # (descending mesh dims for ascending shard order). If transforms are not in + # descending order, the shard order isn't ascending and we can't flatten. + if mesh_dims != sorted_mesh_dims[::-1]: + return None, "non_ascending_mesh_dims" + # Use sorted dims for mesh lookup (required by DeviceMesh API) + flattened_mesh = _get_flattened_mesh_by_layout(device_mesh, sorted_mesh_dims) + if flattened_mesh is None: + return None, "no_flattened_mesh" + + # For nested sharding, each transform has a different logical_shape. + # We need the outermost transform's logical_shape, which represents the + # tensor shape before any of the transforms in this group are applied. + # The outermost transform has the largest logical_shape on the affected + # tensor dimension (least divided by prior shards). + src, dst = first_placements + if comm_type == "all_gather": + # S->R (all_gather): affected dim is the source shard dim + affected_dim = cast(Shard, src).dim + outermost_info = max(infos, key=lambda x: x.logical_shape[affected_dim]) + elif comm_type == "reduce_scatter": + affected_dim = cast(Shard, dst).dim + outermost_info = max(infos, key=lambda x: x.logical_shape[affected_dim]) + tensor_dim_size = outermost_info.logical_shape[affected_dim] + effective_shard_mesh_size = math.prod( + device_mesh.size(info.mesh_dim) for info in infos + ) + # For reduce_scatter (Partial -> Shard), we cannot flatten if the tensor + # dimension is not evenly divisible by the flattened mesh size. + # The effective size is the product of mesh sizes for dims being transformed + # (not all dims with matching placement - intervening shards are already + # accounted for in logical_shape). + if tensor_dim_size % effective_shard_mesh_size != 0: + return None, "uneven_tensor_shape" + elif comm_type == "all_reduce": + # no shape change, any info works + outermost_info = infos[0] + else: + raise NotImplementedError( + f"Unsupported comm type for try_create_flattened: {comm_type}" + ) + + # For mixed sum/avg partials: use sum for the collective, compute avg scale + avg_scale = None + merged_src = src + if src.is_partial(): + scale = math.prod( + device_mesh.size(info.mesh_dim) + for info in infos + if cast(Partial, info.src_dst_placements[0]).reduce_op == "avg" + ) + if scale > 1: + avg_scale = scale + merged_src = Partial("sum") + + merged_placements = (merged_src, dst) + + return ( + _FlattenedTransformInfo( + mesh_dim=0, + src_dst_placements=merged_placements, + logical_shape=outermost_info.logical_shape, + mesh=flattened_mesh, + original_mesh_dims=sorted_mesh_dims, + avg_scale=avg_scale, + ), + None, + ) + + # Merge consecutive same-type operations (without reordering) + result: list[_TransformInfo | _FlattenedTransformInfo] = [] + i = 0 + + while i < len(transform_infos): + info = transform_infos[i] + current_key = info._comm_type_key() + + # Only try to merge if this is a mergeable comm type + if not is_mergeable(current_key): + result.append(info) + i += 1 + continue + + # Collect consecutive transforms with mergeable src_dst_placements + # (not just same comm type - e.g., Partial->Shard(0) vs Partial->Shard(1) can't merge) + # Note: sum/avg partials can be merged since they use the same reduction + current_placements = info.src_dst_placements + group: list[_TransformInfo] = [info] + j = i + 1 + while ( + j < len(transform_infos) + and is_mergeable(transform_infos[j]._comm_type_key()) + and are_placements_mergeable( + transform_infos[j].src_dst_placements, current_placements + ) + ): + group.append(transform_infos[j]) + j += 1 + + # Try to flatten the group + flattened, failure_reason = try_create_flattened(group) + if flattened is not None: + result.append(flattened) + else: + # Can't flatten - add individually and warn once if applicable + result.extend(group) + # Warn for reasons that indicate a real optimization opportunity was missed + if failure_reason in ( + "no_flattened_mesh", + "uneven_tensor_shape", + "non_ascending_mesh_dims", + ): + mesh_dims = tuple(sorted(g.mesh_dim for g in group)) + _warn_flatten_optimization_not_possible( + device_mesh, + mesh_dims, + src_placements, + dst_placements, + len(group), + current_key, # type: ignore[arg-type] + failure_reason, + ) + + i = j + logger.debug( + "_optimize_transform_infos original: %s, optimized: %s", transform_infos, result + ) + + return result + + +# Global cache for DTensorRedistributePlanner instances +_planner_cache: dict[ + tuple[weakref.ReferenceType[DeviceMesh], TensorMeta], + "DTensorRedistributePlanner", +] = {} + + +def get_redistribute_planner( + device_mesh: DeviceMesh, + dtensor_meta: TensorMeta, +) -> "DTensorRedistributePlanner": + """ + Factory function to get or create a DTensorRedistributePlanner instance. + This function provides transparent caching of planner instances based on + device mesh and dtensor meta. Multiple calls with the same parameters + will return the same cached instance for better performance. + Args: + device_mesh: The device mesh for the planner + dtensor_meta: TensorMeta of the DTensor to redistribute + Returns: + A DTensorRedistributePlanner instance (potentially cached) + """ + if _are_we_tracing(): + return DTensorRedistributePlanner(device_mesh, dtensor_meta) + + cache_key = (weakref.ref(device_mesh), dtensor_meta) + if cache_key not in _planner_cache: + planner = DTensorRedistributePlanner(device_mesh, dtensor_meta) + _planner_cache[cache_key] = planner + + return _planner_cache[cache_key] + + +def clear_redistribute_planner_cache() -> None: + """Clear the cache of DTensorRedistributePlanner instances.""" + _planner_cache.clear() + + +class DTensorRedistributePlanner: + """ + This class is used to plan the collective calls to transform the local shard + of the DTensor from its current spec to the target spec. + Suppose there are N tensor dimensions and M mesh dimensions, the total + possible state size will be (N+2)*M*M!. + Note: Use get_redistribute_planner() factory function instead of direct + instantiation for automatic caching. + """ + + @dataclasses.dataclass(frozen=True, slots=True) + class DistState: + placements: tuple[Placement, ...] + tensor_dim_to_mesh_dim: ShardOrder + _hash: int | None = dataclasses.field( + default=None, init=False, repr=False, compare=False + ) + + def __str__(self): + return DTensorSpec.format_shard_order_str( + self.placements, + self.tensor_dim_to_mesh_dim, + ) + + def __repr__(self): + return self.__str__() + + def __post_init__(self): + # precompute hash after all attributes are set + object.__setattr__( + self, + "_hash", + self._compute_hash(), + ) + + def __hash__(self) -> int: + return self._hash if self._hash is not None else self._compute_hash() + + def _compute_hash(self) -> int: + return hash( + ( + self.placements, + self.tensor_dim_to_mesh_dim, + ) + ) + + def __eq__(self, other: object) -> bool: + if not isinstance(other, DTensorRedistributePlanner.DistState): + return False + if self._hash != other._hash: + return False + return ( + self.placements, + self.tensor_dim_to_mesh_dim, + ) == ( + other.placements, + other.tensor_dim_to_mesh_dim, + ) + + def _to_tuple(self, x): + """Convert a nested list structure to a nested tuple structure.""" + if isinstance(x, list | tuple): + return tuple(self._to_tuple(item) for item in x) + return x + + @staticmethod + def _dict_to_ShardOrder(x: dict[int, list[int]]) -> ShardOrder: + """Convert dict to ShardOrder""" + return tuple( + ShardOrderEntry(tensor_dim=key, mesh_dims=tuple(value)) + for key, value in sorted(x.items()) + if value + ) + + @staticmethod + def _ShardOrder_to_dict(x: ShardOrder) -> dict[int, list[int]]: + """Convert ShardOrder to dict with tensor dim as key""" + tensor_mesh_dim_dict = defaultdict(list) + for entry in x: + tensor_mesh_dim_dict[entry.tensor_dim] = list(entry.mesh_dims) + return tensor_mesh_dim_dict + + @staticmethod + def stringify_transform_infos( + mesh: DeviceMesh, + transform_infos: Sequence[_TransformInfo], + src_placement: tuple[Placement, ...], + src_shard_order: ShardOrder | None = None, + use_strided_shard_as_shard_order: bool = False, + ) -> str: + """ + Generate a string representation of the sequence of state transitions + (placements and shard orders) as described by the given transform_info. + + Args: + mesh: The DeviceMesh used for the redistribution. + transform_infos: A sequence of _TransformInfo objects describing each + transformation step. + src_placement: The initial tuple of Placement objects. + src_shard_order: (Optional) The initial ShardOrder representing + the mapping of tensor dimensions to mesh dimensions. If None, + the default shard order is computed from src_placement and mesh. + use_strided_shard_as_shard_order: If True, normalize _StridedShard + placements into regular Shard placements with an explicit + shard_order before stringifying. + + Returns: + A string showing the sequence of DistState transitions, separated by '->'. + """ + if len(src_placement) != mesh.ndim: + raise AssertionError( + f"dimensions mismatch {len(src_placement)} vs {mesh.ndim}" + ) + if use_strided_shard_as_shard_order: + src_placement, src_shard_order = ( + DTensorSpec._normalize_placements_into_shard_order( + src_placement, mesh, use_strided_shard_as_shard_order=True + ) + ) + if src_shard_order is None: + src_shard_order = DTensorSpec.compute_default_shard_order(src_placement) + cur_placement = list(src_placement) + shard_order_dict = DTensorRedistributePlanner._ShardOrder_to_dict( + src_shard_order + ) + cur_state = DTensorRedistributePlanner.DistState( + tuple(cur_placement), src_shard_order + ) + state_list = [ + cur_state, + ] + # Track whether each transition is flattened (for visualization) + is_flattened_list: list[bool] = [] + + for transform_info in transform_infos: + is_flattened = isinstance(transform_info, _FlattenedTransformInfo) + + _update_shard_order_and_placements( + transform_info, cur_placement, shard_order_dict + ) + + new_state = DTensorRedistributePlanner.DistState( + tuple(cur_placement), + DTensorRedistributePlanner._dict_to_ShardOrder(shard_order_dict), + ) + state_list.append(new_state) + is_flattened_list.append(is_flattened) + + # Build the trace string using '-->' for flattened transforms, '->' for regular + trace_parts = [str(state_list[0])] + for i, is_flattened in enumerate(is_flattened_list): + separator = "-->" if is_flattened else "->" + trace_parts.append(separator) + trace_parts.append(str(state_list[i + 1])) + return "".join(trace_parts) + + def __init__( + self, + device_mesh: DeviceMesh, + dtensor_meta: TensorMeta, + ) -> None: + """ + Initialize DTensorRedistributePlanner. + + Args: + device_mesh: The device mesh for this planner + dtensor_meta: TensorMeta of the DTensor to redistribute + """ + self.device_mesh = device_mesh + if not device_mesh._is_current_rank_part_of_mesh(): + raise AssertionError + if dtensor_meta is None: + raise AssertionError + self.dtensor_meta = dtensor_meta + self.tensor_dimension = len(dtensor_meta.shape) + self.strided_shard_placements_in_target: set[_StridedShard] = set() + self.partial_reduce_ops_in_target: set[str] = set() + self.setup_cost_callbacks() + + def setup_cost_callbacks( + self, + ) -> None: + """ + Set up the cost function for different collective operations. + Uses communication time estimation based on actual tensor sizes and + mesh topology for accurate cost modeling. + """ + + def state_to_spec( + state: DTensorRedistributePlanner.DistState, + ) -> DTensorSpec: + return DTensorSpec( + mesh=self.device_mesh, + placements=state.placements, + tensor_meta=self.dtensor_meta, + shard_order=state.tensor_dim_to_mesh_dim, + use_strided_shard_as_shard_order=False, + ) + + def cost_function(src_state, dst_state): + return one_step_redistribute_cost( + state_to_spec(src_state), state_to_spec(dst_state) + ) + + self.cost_function = cost_function + + def get_next_state( + self, + placements: tuple[Placement, ...], + tensor_mesh_dim_tuple: ShardOrder, + ) -> dict["DTensorRedistributePlanner.DistState", float]: + # We map tensor dimensions to device mesh axes, similar to JAX-style + # sharding representation. Notation: + # S()[] means tensor dimension + # is sharded on the listed device mesh axes, where + # is sorted by device order. + # + # To generalize to arbitrary dimensionality, we use the following notation: + # S(a)[x, ...] : tensor dimension 'a' is sharded on device mesh axes x, ... (variadic, possibly empty) + # SS(a)[x, ...] : _StridedShard on tensor dimension 'a' on device mesh axes x, ... (variadic, possibly empty) + # R[...] : replicated on the listed device mesh axes (possibly empty) + # P[...] : partial on the listed device mesh axes (possibly empty) + # The ellipsis '...' denotes a variadic wildcard, i.e., zero or more device mesh axes. + # + # Below are possible transitions from one sharding state to another. + # We use `S` for Shard, `SS` for _StridedShard, `R` for Replicate, and `P` for Partial. + # + # Case 1. Shard(a) -> Shard(b), use all-to-all (a2a), applies to: + # S(a)[..., x] -> S(b)[..., x] + # or + # S(a)[..., x, y]S(b)[..., z, k] -> S(a)[..., x]S(b)[..., z, k, y] + # where device order of 'y' > device order of 'z' and 'k' + # + # Case 2. Shard() -> Replicate(), use all-gather, applies to: + # S(a)[..., x, y, z] -> S(a)[..., x, y] + # + # Case 3. Partial() -> Replicate(), use all-reduce, applies to: + # P[..., x, y] -> P[..., y] or P[..., x] + # Note: this case can be disabled because all-reduce technically is not + # a primitive since it combines a reduce-scatter + all-gather. + # + # Case 4. Replicate() -> Shard(), use chunk, applies to: + # S(a)[..., z] -> S(a)[..., z, y] (`a` can be any tensor dim). Note that + # 'y' must be after 'z'. + # + # Case 5. Partial() -> Shard(), use reduce-scatter, applies to: + # P[..., x, y] -> P[..., x]S(a)[..., y] or P[..., x, y] -> P[..., y]S(a)[..., x] + # + # Case 6. Replicate() -> Partial(), local math op, applies to: + # R* -> P[..., x] + # + # Case 7. Shard() -> Partial("sum"), local zero-fill, generated only as + # a destination-directed transition in find_min_cost_path(), applies to: + # S(a)[..., x] -> P[..., x] + # This transition is available only for explicit redistribution. + # + # Strategy planning assigns it infinite cost. A zero communication cost + # would ignore the logical-shape allocation, zero-fill, and shard copy, + # causing the planner to over-prefer it. + # + # Transform planning assigns an explicit request zero search cost because + # it requires no communication. This difference is intentional: the + # transform planner supports strategy-selected transitions plus additional + # explicit-only transitions. + + # (TODO) Case 8. _StridedShard(a) -> Shard(b), use all-to-all (a2a), applies to: + # SS(a)[..., x] -> S(b)[..., x] + # + # Case 9. _StridedShard() -> Replicate(), use all-gather, applies to: + # SS(a)[..., x, y, z] -> SS(a)[..., x, y] + # + # (TODO) Case 10. Shard(a) -> _StridedShard(b), use all-to-all (a2a), applies to: + # S(a)[..., x] -> SS(b)[..., x] + # + # (TODO) Case 11. Partial() -> _StridedShard(), use reduce-scatter, applies to: + # P[..., x, y] -> P[..., x]SS(a)[..., y] or P[..., x, y] -> P[..., y]SS(a)[..., x] + # + # Case 12. Replicate() -> _StridedShard(), use chunk, applies to: + # R* -> SS(a)[..., x] + # + # NB: Regarding `_StridedShard``, we only allow changing `Replicate` into + # `_StridedShard` with the same tensor dim and split_factor that occurs in the + # target placement. + # + # (TODO) Verify device order impact in Partial placement. We may need to handle + # device ordering for Partial also. + + # list of [DistState, cost] + all_next_state: dict[DTensorRedistributePlanner.DistState, float] = {} + + tensor_mesh_dim_dict = DTensorRedistributePlanner._ShardOrder_to_dict( + tensor_mesh_dim_tuple + ) + cur_dist_state = self.DistState( + self._to_tuple(placements), + tensor_mesh_dim_tuple, + ) + ###################################################################### + # handle case 1: Shard(a) -> Shard(b) + # For S(a), S(b), only the last device order of S(a) and S(b) can be a2a + # interchangeably. + + # convert sparse tuple + for entry in tensor_mesh_dim_tuple: + src_tensor_dim = entry.tensor_dim + src_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim][-1] + if not isinstance(placements[src_mesh_dim], Shard): + # skip special case like `_StridedShard` + continue + for dst_tensor_dim in range(self.tensor_dimension): + if src_tensor_dim == dst_tensor_dim: + continue + # try move the last sharded device dim from + # Shard(src_tensor_dim) to Shard(dst_tensor_dim) + move_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim].pop() + tensor_mesh_dim_dict[dst_tensor_dim].append(move_mesh_dim) + new_placements = list(placements) + new_placements[move_mesh_dim] = Shard(dst_tensor_dim) + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder( + tensor_mesh_dim_dict + ), + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + # reset content for next iteration + tensor_mesh_dim_dict[src_tensor_dim].append(move_mesh_dim) + tensor_mesh_dim_dict[dst_tensor_dim].pop() + # TODO(zpcore): support discovering submesh to prevent padding when + # tensor dim is not divisible by the mesh dim. + + ###################################################################### + # handle case 2: Shard() -> Replicate() + for entry in tensor_mesh_dim_tuple: + src_tensor_dim = entry.tensor_dim + src_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim][-1] + if not isinstance(placements[src_mesh_dim], Shard): + # skip special case like `_StridedShard` + continue + move_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim].pop() + new_placements = list(placements) + new_placements[move_mesh_dim] = Replicate() + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder(tensor_mesh_dim_dict), + ) + tensor_mesh_dim_dict[src_tensor_dim].append(move_mesh_dim) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + + ###################################################################### + # handle case 3: Partial() -> Replicate() + for src_mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Partial): + continue + new_placements = list(placements) + new_placements[src_mesh_dim] = Replicate() + dist_state = self.DistState( + self._to_tuple(new_placements), tensor_mesh_dim_tuple + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + + ###################################################################### + # handle case 4: Replicate() -> Shard() + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Replicate): + continue + for dst_tensor_dim in range(self.tensor_dimension): + # try convert placement[mesh_dim] to Shard(dst_tensor_dim) + new_placements = list(placements) + new_placements[mesh_dim] = Shard(dst_tensor_dim) + tensor_mesh_dim_dict[dst_tensor_dim].append(mesh_dim) + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder( + tensor_mesh_dim_dict + ), + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + tensor_mesh_dim_dict[dst_tensor_dim].pop() + + ###################################################################### + # handle case 5: Partial() -> Shard() + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Partial): + continue + for dst_tensor_dim in range(self.tensor_dimension): + # try convert placement[mesh_dim] to Shard(dst_tensor_dim) + new_placements = list(placements) + new_placements[mesh_dim] = Shard(dst_tensor_dim) + tensor_mesh_dim_dict[dst_tensor_dim].append(mesh_dim) + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder( + tensor_mesh_dim_dict + ), + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + tensor_mesh_dim_dict[dst_tensor_dim].pop() + + ###################################################################### + # handle case 6: Replicate() -> Partial() + # Generate transitions only for reduce_ops that are present in the src/dst + # placements for this redistribution, avoiding unnecessary graph expansion. + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Replicate): + continue + for reduce_op in self.partial_reduce_ops_in_target: + new_placements = list(placements) + new_placements[mesh_dim] = Partial(reduce_op) + + # Skip if this would create mixed partial types (except sum+avg which commute) + partial_reduce_ops = { + p.reduce_op for p in new_placements if isinstance(p, Partial) + } + if len(partial_reduce_ops) > 1 and partial_reduce_ops != {"sum", "avg"}: + continue + + dist_state = self.DistState( + self._to_tuple(new_placements), tensor_mesh_dim_tuple + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + + # Additional cases handling for _StridedShard + + ###################################################################### + # TODO(zpcore): handle case 8: _StridedShard() -> Shard() on the same dim + + ###################################################################### + # handle case 9: _StridedShard() -> Replicate() + for entry in tensor_mesh_dim_tuple: + src_tensor_dim = entry.tensor_dim + src_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim][-1] + if not isinstance(placements[src_mesh_dim], _StridedShard): + continue + move_mesh_dim = tensor_mesh_dim_dict[src_tensor_dim].pop() + new_placements = list(placements) + new_placements[move_mesh_dim] = Replicate() + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder(tensor_mesh_dim_dict), + ) + tensor_mesh_dim_dict[src_tensor_dim].append(move_mesh_dim) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + + # Early exit if no StridedShard in target + if not self.strided_shard_placements_in_target: + return all_next_state + + ###################################################################### + # TODO(zpcore): handle case 10: Shard() -> _StridedShard() + + ###################################################################### + # TODO(zpcore): handle case 11: Partial() -> _StridedShard() + + ###################################################################### + # handle case 12: Replicate() -> _StridedShard() + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, Replicate): + continue + for strided_shard_obj in self.strided_shard_placements_in_target: + dst_tensor_dim = strided_shard_obj.dim + # try convert placement[mesh_dim] to strided_shard_obj + new_placements = list(placements) + new_placements[mesh_dim] = strided_shard_obj + tensor_mesh_dim_dict[dst_tensor_dim].append(mesh_dim) + dist_state = self.DistState( + self._to_tuple(new_placements), + DTensorRedistributePlanner._dict_to_ShardOrder( + tensor_mesh_dim_dict + ), + ) + all_next_state[dist_state] = self.cost_function( + cur_dist_state, + dist_state, + ) + tensor_mesh_dim_dict[dst_tensor_dim].pop() + + return all_next_state + + def _get_shard_to_partial_target_states( + self, current_state: DistState, dst_state: DistState + ) -> dict[DistState, float]: + """Generate explicit-only Shard -> Partial("sum") target steps. + + Zero weight means no communication; allocation and copy costs are omitted. + This edge is generated only when Partial("sum") is the requested target, + so path search cannot use it as an intermediate step to another layout. + """ + next_states: dict[DTensorRedistributePlanner.DistState, float] = {} + shard_order = self._ShardOrder_to_dict(current_state.tensor_dim_to_mesh_dim) + for entry in current_state.tensor_dim_to_mesh_dim: + tensor_dim = entry.tensor_dim + mesh_dim = shard_order[tensor_dim][-1] + current = current_state.placements[mesh_dim] + target = dst_state.placements[mesh_dim] + if not ( + isinstance(current, Shard) + and type(target) is Partial + and target.reduce_op == "sum" + ): + continue + + shard_order[tensor_dim].pop() + placements = list(current_state.placements) + placements[mesh_dim] = target + next_state = self.DistState( + tuple(placements), self._dict_to_ShardOrder(shard_order) + ) + shard_order[tensor_dim].append(mesh_dim) + next_states[next_state] = 0.0 + return next_states + + # TODO(zpcore): if the dst_state contains special placement like + # `_MaskPartial`, we will never reach that state. Need to support this case. + def find_min_cost_path( + self, src_state: DistState, dst_state: DistState + ) -> list["DTensorRedistributePlanner.DistState"]: + """ + Find the min cost path from src_state to dst_state using Dijkstra's + algorithm. + + Args: + src_state: The source state + dst_state: The destination state + + Returns: + A list of states representing the min cost path from src_state to + dst_state + """ + import heapq + + # priority queue (cost_key, counter, cost, state, path) for Dijkstra's + # algorithm. The sort key is a non-guarding optimization hint so that + # symbolic communication costs do not force heapq to evaluate guards. + # Counter breaks ties and avoids comparing DistState objects. + counter = 0 + pq: list[ + tuple[ + float, + int, + FloatLikeType, + DTensorRedistributePlanner.DistState, + list[DTensorRedistributePlanner.DistState], + ] + ] = [(0.0, counter, 0.0, src_state, [src_state])] + visited = set() + while pq: + _, _, cost, current_state, path = heapq.heappop(pq) + if current_state == dst_state: + return path + if current_state in visited: + continue + visited.add(current_state) + # get all possible next states and their costs + next_states = self.get_next_state( + current_state.placements, + current_state.tensor_dim_to_mesh_dim, + ) + next_states.update( + self._get_shard_to_partial_target_states(current_state, dst_state) + ) + for next_state, transition_cost in next_states.items(): + if next_state not in visited: + new_cost = cast(Any, cost) + transition_cost + new_path = path + [next_state] + counter += 1 + heapq.heappush( + pq, + ( + _redistribute_cost_sort_key(new_cost), + counter, + new_cost, + next_state, + new_path, + ), + ) + raise AssertionError( + f"No path found from src_state {src_state} to dst_state {dst_state}" + ) + + def get_logical_shape( + self, + src_state: "DTensorRedistributePlanner.DistState", + mesh_dim: int, + full_tensor_shape: tuple[int, ...], + ) -> list[IntLikeType]: + new_logical_shape: list[IntLikeType] = list(full_tensor_shape) + for entry in src_state.tensor_dim_to_mesh_dim: + tensor_dim = entry.tensor_dim + mesh_dims = entry.mesh_dims + if len(mesh_dims) <= 0: + raise AssertionError + for mdim in mesh_dims: + if mdim == mesh_dim: + continue + placement = src_state.placements[mdim] + if isinstance(placement, Shard): + new_size, _ = placement.local_shard_size_and_offset( + new_logical_shape[tensor_dim], + self.device_mesh.size(mesh_dim=mdim), + self.device_mesh._sym_get_coordinate(mdim), + ) + elif isinstance(placement, _StridedShard): + new_size, _ = placement.local_shard_size_and_offset( + new_logical_shape[tensor_dim], + self.device_mesh.size(mesh_dim=mdim), + self.device_mesh._sym_get_coordinate(mdim), + ) + else: + raise ValueError(f"Unsupported placement type: {placement}") + new_logical_shape[tensor_dim] = new_size + return new_logical_shape + + def generate_graph_based_transform_infos( + self, + src_spec: DTensorSpec, + dst_spec: DTensorSpec, + full_tensor_shape: tuple[int, ...], + ) -> list[_TransformInfo]: + # TODO(zpcore): Temporary workaround for backward compatibility where + # _StridedShard was used to encode device shard order. We should migrate + # to explicit `shard_order` instead. + def _try_normalize_spec( + spec: DTensorSpec, + ) -> tuple[tuple[Placement, ...], ShardOrder]: + if spec.use_strided_shard_as_shard_order: + new_placements, shard_order = ( + DTensorSpec._normalize_placements_into_shard_order( + spec.placements, + spec.mesh, + use_strided_shard_as_shard_order=True, + ) + ) + return new_placements, shard_order + else: + if spec.shard_order is None: + raise ValueError(f"Missing shard_order field in {spec}") + return spec.placements, spec.shard_order + + src_placements, src_shard_order = _try_normalize_spec(src_spec) + dst_placements, dst_shard_order = _try_normalize_spec(dst_spec) + + # In case _StridedShard still exists in placements, collect possible + # split_factor values in the target placements. Need those values to + # redistribute from Shard into _StridedShard. + for placement in dst_placements: + if isinstance(placement, _StridedShard): + self.strided_shard_placements_in_target.add(placement) + + # Collect Partial reduce ops from src and dst placements. These are used + # to generate R->P transitions only for reduce ops that are actually + # present in the redistribution, avoiding unnecessary graph expansion. + for placement in itertools.chain(src_placements, dst_placements): + if isinstance(placement, Partial): + self.partial_reduce_ops_in_target.add(placement.reduce_op) + + src_state = self.DistState(src_placements, src_shard_order) + dst_state = self.DistState(dst_placements, dst_shard_order) + transform_infos: list[_TransformInfo] = [] + state_path = self.find_min_cost_path(src_state, dst_state) + for cur_state, nxt_state in itertools.pairwise(state_path): + # find the mesh_dim that is different between cur_state and nxt_state + if cur_state.placements != nxt_state.placements: + update_mesh_dim = -1 + for mesh_dim, (cur_placement, nxt_placement) in enumerate( + zip(cur_state.placements, nxt_state.placements) + ): + if cur_placement != nxt_placement: + if update_mesh_dim != -1: + raise AssertionError( + "Multiple mesh_dims are different between cur_state and nxt_state" + ) + update_mesh_dim = mesh_dim + logical_shape = self.get_logical_shape( + cur_state, mesh_dim, full_tensor_shape + ) + transform_infos.append( + _TransformInfo( + mesh_dim=update_mesh_dim, + src_dst_placements=(cur_placement, nxt_placement), + logical_shape=logical_shape, + ) + ) + + return transform_infos + + def generate_greedy_transform_infos( + self, + src_spec: DTensorSpec, + dst_spec: DTensorSpec, + ) -> list[_TransformInfo]: + """ + Generate the transform infos from the source placements to the target placements. + + To transform from source to target placement it might have multiple steps, i.e. it + might decompose Si -> Sj into Si -> R -> Sj. + This would detect if there're mis-aligned/nested shardings between src/dst placements. + E.g. Suppose the redistribution to perform is (Shard(0), Shard(0)) -> (Replicate(), Shard(0)), + in this case Shard(0) -> Shard(0) for mesh dimension 1 actually needs resharding, because in + the former is a nested-sharding of a tensor already sharded dimension 0, whereas + the latter is the first sharding on tensor dimension 0. + """ + # logical shape records the logic tensor shape on the mesh dimension + # this is useful to ensure uneven sharding gets correct output shape + initial_logical_shape = list(src_spec.shape) + mesh_dims_to_logical_shape = [initial_logical_shape] + transform_infos: list[_TransformInfo] = [] + if self.device_mesh.ndim == 1: + # if device_mesh is 1D, redistribute is a simple direct + # transformation (skip if src == dst) + if src_spec.placements[0] != dst_spec.placements[0]: + transform_infos.append( + _TransformInfo( + mesh_dim=0, + src_dst_placements=( + src_spec.placements[0], + dst_spec.placements[0], + ), + logical_shape=initial_logical_shape, + ) + ) + return transform_infos + + # Handle multi-dim device mesh placement redistribution First, we need + # to build the logical shape for each mesh dim for correct allgather + # uneven shards on each mesh dim (with dynamic padding) + for i, src in enumerate(src_spec.placements): + current_logical_shape = mesh_dims_to_logical_shape[i] + if isinstance(src, Shard): + if i < self.device_mesh.ndim - 1: + # calculate and save the logical shape for this sharding + mesh_dim_size = self.device_mesh.size(mesh_dim=i) + local_shard_size, _ = src._local_shard_size_and_offset( + current_logical_shape[src.dim], + mesh_dim_size, + self.device_mesh._sym_get_coordinate(i), + ) + new_logical_shape = list(current_logical_shape) + new_logical_shape[src.dim] = local_shard_size + mesh_dims_to_logical_shape.append(new_logical_shape) + else: + mesh_dims_to_logical_shape.append(current_logical_shape) + + # Next, we need to derive the transform infos from src to dst + # placements, here we use a greedy search with step by step state + # transformations + current_placements = list(src_spec.placements) + target_placements = list(dst_spec.placements) + + if src_spec.num_shards > 1: + # If src_spec have sharding, it could potentially have sharding that + # is misaligned with dst_spec a common case of this is nested + # sharding (i.e. (S(0), S(0)) -> (R, S(0))). In those cases, we + # first traverse from inner placement to outer placement to detect + # misaligned shardings and properly replicate nested sharding first. + for mesh_dim in reversed(range(len(current_placements))): + current = current_placements[mesh_dim] + target = target_placements[mesh_dim] + # If target is not Shard, we can directly redistribute since we + # are traversing from inner to outer placements here + # TODO: extend nested sharding detection to _StridedShard + # (isinstance check and is_shard() below miss it). + # Safe today: strategies convert _StridedShard to Replicate + # on ALL mesh dims for a given reduction dim, so misaligned + # nested _StridedShard targets can't arise. + if isinstance(target, Shard): + # If target is Shard, check for nested sharding on the + # tensor dim BEFORE the current mesh_dim + shard_dim = target.dim + current_mesh_sharding, target_mesh_sharding = [], [] + for i, (s, p) in enumerate( + zip(current_placements, target_placements) + ): + if i >= mesh_dim: + break + if s.is_shard(shard_dim): + current_mesh_sharding.append(i) + if p.is_shard(shard_dim): + target_mesh_sharding.append(i) + + if current_mesh_sharding != target_mesh_sharding: + # if current/target_placements have misaligned sharding + # on the tensor dim BEFORE the current mesh_dim, we need + # to replicate the tensor on the mesh dim first to clear + # the nested sharding + target = Replicate() + + if current != target: + transform_infos.append( + _TransformInfo( + mesh_dim=mesh_dim, + src_dst_placements=(current, target), + logical_shape=mesh_dims_to_logical_shape[mesh_dim], + ) + ) + current_placements[mesh_dim] = target + + # We always traverse from outer placement to inner placement to collect + # the remaining needed transform infos (i.e. the replication from nested + # sharding might need to further perform resharding to Shard again) + for mesh_dim, (current, target) in enumerate( + zip(current_placements, target_placements) + ): + if current != target: + transform_infos.append( + _TransformInfo( + mesh_dim=mesh_dim, + src_dst_placements=(current, target), + logical_shape=mesh_dims_to_logical_shape[mesh_dim], + ) + ) + current_placements[mesh_dim] = target + return transform_infos + + +def _gen_transform_infos_non_cached( + src_spec: DTensorSpec, + dst_spec: DTensorSpec, + use_graph_based_transform: bool | None = None, +) -> list[_TransformInfo]: + device_mesh = src_spec.device_mesh + src_shard_order = src_spec.shard_order + dst_shard_order = dst_spec.shard_order + # DTensorSpec should automatically generate shard_order, and it can be () if + # no shard. + has_non_default_order = not all( + DTensorSpec.is_default_device_order(order) + for order in (src_shard_order, dst_shard_order) + ) + has_strided_shard = any( + isinstance(p, _StridedShard) + for p in (*src_spec.placements, *dst_spec.placements) + ) + + # Determine which transform strategy to use: + # 1. Non-standard device order or contains _StridedShard → always use graph-based + # 2. Global flag or explicit parameter True → use graph-based + # 3. Otherwise → use greedy + if has_non_default_order or has_strided_shard: + use_graph_based_transform = True + elif _FORCE_MIN_COST_REDISTRIBUTION_PLAN is not None: + use_graph_based_transform = _FORCE_MIN_COST_REDISTRIBUTION_PLAN + elif use_graph_based_transform is None: + use_graph_based_transform = False + if src_spec.tensor_meta is None: + raise AssertionError + drp = get_redistribute_planner( + device_mesh, + src_spec.tensor_meta, + ) + if use_graph_based_transform: + # TODO(zpcore): Temporary workaround for the case where _StridedShard + # cannot be decoded into shard order. This happens when + # use_strided_shard_as_shard_order defaults to True (e.g. in + # Redistribute.forward where the target DTensorSpec is constructed from + # raw placements without the flag), but the split_factor doesn't + # correspond to any valid product of mesh dimension sizes (e.g. sf=2 + # on a 1D mesh). A proper fix is to either pass + # use_strided_shard_as_shard_order through the Redistribute API, or + # migrate to explicit shard_order so _StridedShard is no longer + # overloaded for two purposes. + try: + transform_infos = drp.generate_graph_based_transform_infos( + src_spec, + dst_spec, + src_spec.shape, + ) + except _StridedShardNotDecodableError: + transform_infos = drp.generate_greedy_transform_infos(src_spec, dst_spec) + else: + transform_infos = drp.generate_greedy_transform_infos(src_spec, dst_spec) + return transform_infos + + +@cache +def _gen_transform_infos( + src_spec: DTensorSpec, + dst_spec: DTensorSpec, + use_graph_based_transform: bool | None = None, +) -> list[_TransformInfo]: + return _gen_transform_infos_non_cached( + src_spec, + dst_spec, + use_graph_based_transform, + ) + + +def redistribute_local_tensor( + local_tensor: torch.Tensor, + current_spec: DTensorSpec, + target_spec: DTensorSpec, + *, + async_op: bool = False, + use_graph_based_transform: bool | None = None, + # True if user explicitly called DTensor.redistribute() + is_explicit: bool = False, +) -> torch.Tensor: + """ + This redistribute the local tensor (torch.Tensor) from the current DTensorSpec to + the target DTensorSpec, which involves the necessary collective calls to transform + the local shard of the DTensor from its current spec to the target spec. + """ + + if current_spec.mesh != target_spec.mesh: + # TODO: alltoall/permute reshuffling to change device_mesh if they are not the same + raise NotImplementedError("Cross device mesh comm not supported yet!") + + if current_spec.use_strided_shard_as_shard_order is None: + raise ValueError( + "use_strided_shard_as_shard_order should be initialized in DTensorSpec.__post_init__()" + ) + + # We do not see a valid use case for mixing different partial types in the same DTensor. + # in principle it could be supported, but since nonlinear reductions (e.g. max) exist, relative ordering + # of different partials would become semantically critical. Without a motivating use case, we prohibit this. + assert_no_mixed_partial_types(current_spec.placements) + assert_no_mixed_partial_types(target_spec.placements) + + new_local_tensor = local_tensor + device_mesh = current_spec.mesh + + if not device_mesh._is_current_rank_part_of_mesh(): + # if rank is not part of mesh, we skip redistribute and simply return local_tensor, + # which should be an empty tensor + return local_tensor + + if _are_we_tracing(): + transform_infos = _gen_transform_infos_non_cached( + current_spec, + target_spec, + use_graph_based_transform, + ) + else: + transform_infos = _gen_transform_infos( + current_spec, + target_spec, + use_graph_based_transform, + ) + + # Optimize by grouping same-type collectives into flattened operations + optimized_transform_infos = _optimize_transform_infos( + transform_infos, + device_mesh, + current_spec.placements, + target_spec.placements, + ) + + debug_mode = get_active_debug_mode() + + redistribute_context = ( + debug_mode.record_redistribute_calls( # type: ignore[union-attr] + local_tensor, + current_spec.placements, + target_spec.placements, + DTensorRedistributePlanner.stringify_transform_infos( + device_mesh, + optimized_transform_infos, + current_spec.placements, + current_spec.shard_order, + current_spec.use_strided_shard_as_shard_order, + ), + is_explicit=is_explicit, + ) + if debug_mode is not None + else contextlib.nullcontext() + ) + + with redistribute_context: + for transform_info in optimized_transform_infos: + # Determine which mesh to use: flattened transforms have their own mesh + if isinstance(transform_info, _FlattenedTransformInfo): + mesh_to_use = transform_info.mesh + else: + mesh_to_use = device_mesh + i = transform_info.mesh_dim + current, target = transform_info.src_dst_placements + + # _StridedShard methods use device_mesh directly, not mesh_to_use. + # This is safe because _StridedShard.is_shard() returns False, so + # _comm_type_key() returns None and flattening is never attempted. + if isinstance(current, _StridedShard) or isinstance(target, _StridedShard): + if mesh_to_use is not device_mesh: + raise AssertionError( + "_StridedShard redistribute assumes no flattened transforms" + ) + + num_chunks = mesh_to_use.size(mesh_dim=i) + + if current == target: + # short cut, just use the original local tensor + new_local_tensor = local_tensor + continue + + if num_chunks == 1: + # short cut, if there's only one shard, we don't need to do any collective + # comm, just use the original local tensor + new_local_tensor = local_tensor + continue + + if target.is_replicate(): + # Case 1: target is Replicate + if current.is_partial(): + partial_spec = cast(Partial, current) + new_local_tensor = partial_spec._reduce_value( + local_tensor, mesh_to_use, i + ) + # For merged sum/avg partials, apply avg scaling + if ( + isinstance(transform_info, _FlattenedTransformInfo) + and transform_info.avg_scale is not None + ): + new_local_tensor = new_local_tensor / transform_info.avg_scale + elif current.is_shard(): + current_placement = cast(Shard, current) + new_local_tensor = current_placement._to_replicate_tensor( + local_tensor, mesh_to_use, i, transform_info.logical_shape + ) + elif isinstance(current, _StridedShard): + new_local_tensor = current._to_replicate_tensor( + local_tensor, device_mesh, i, transform_info.logical_shape + ) + else: + raise RuntimeError( + f"redistribute from {current} to {target} not supported yet" + ) + + elif target.is_shard(): + # Case 2: target is Shard + target_placement = cast(Shard, target) + if current.is_partial(): + partial_spec = cast(Partial, current) + new_local_tensor = partial_spec._reduce_shard_value( + local_tensor, mesh_to_use, i, target_placement + ) + # For merged sum/avg partials, apply avg scaling + if ( + isinstance(transform_info, _FlattenedTransformInfo) + and transform_info.avg_scale is not None + ): + new_local_tensor = new_local_tensor / transform_info.avg_scale + elif current.is_replicate(): + # split the tensor and return the corresponding cloned local shard + new_local_tensor = target_placement._replicate_to_shard( + local_tensor, mesh_to_use, i, mesh_to_use._sym_get_coordinate(i) + ) + elif current.is_shard(): + shard_spec = cast(Shard, current) + if shard_spec.dim != target_placement.dim: + new_local_tensor = shard_spec._to_new_shard_dim( + local_tensor, + mesh_to_use, + i, + transform_info.logical_shape, + target_placement.dim, + ) + elif isinstance(current, _StridedShard): + # _StridedShard -> Shard: go via Replicate as intermediate + replicated = current._to_replicate_tensor( + local_tensor, device_mesh, i, transform_info.logical_shape + ) + new_local_tensor = target_placement._replicate_to_shard( + replicated, + mesh_to_use, + i, + mesh_to_use._sym_get_coordinate(i), + ) + else: + raise ValueError( + f"Unexpected placement {current} for redistribute to target placement {target}" + ) + elif target.is_partial(): + if current.is_replicate(): + partial_spec = cast(Partial, target) + new_local_tensor = partial_spec._partition_value( + local_tensor, mesh_to_use, i + ) + elif _is_shard_like(current): + if ( + not isinstance(current, Shard) + or type(target) is not Partial + or target.reduce_op != "sum" + ): + raise RuntimeError( + f"redistribute from {current} to {target} not supported yet" + ) + new_local_tensor = current._to_partial_tensor( + local_tensor, + mesh_to_use, + i, + transform_info.logical_shape, + ) + else: + if current != target: + raise AssertionError( + f"Redistribution from one partial type ({current}) to another ({target}) is unsupported." + ) + # partial -> partial no op, should never hit + new_local_tensor = local_tensor + elif isinstance(target, _StridedShard): + # Case 4: target is _StridedShard + if current.is_partial(): + # Partial -> _StridedShard: reduce to Replicate, then strided shard + partial_spec = cast(Partial, current) + replicated = partial_spec._reduce_value( + local_tensor, mesh_to_use, i + ) + new_local_tensor = target._replicate_to_strided_shard( + replicated, device_mesh, i, device_mesh._sym_get_coordinate(i) + ) + elif current.is_replicate(): + # split the tensor and return the corresponding local strided shard + new_local_tensor = target._replicate_to_strided_shard( + local_tensor, device_mesh, i, device_mesh._sym_get_coordinate(i) + ) + elif current.is_shard(): + # Shard -> _StridedShard: all-gather to Replicate, then strided shard + current_placement = cast(Shard, current) + replicated = current_placement._to_replicate_tensor( + local_tensor, mesh_to_use, i, transform_info.logical_shape + ) + new_local_tensor = target._replicate_to_strided_shard( + replicated, device_mesh, i, device_mesh._sym_get_coordinate(i) + ) + elif isinstance(current, _StridedShard): + # _StridedShard -> _StridedShard: go through Replicate + # First convert to Replicate, then to _StridedShard + replicated = current._to_replicate_tensor( + local_tensor, device_mesh, i, transform_info.logical_shape + ) + new_local_tensor = target._replicate_to_strided_shard( + replicated, device_mesh, i, device_mesh._sym_get_coordinate(i) + ) + else: + raise ValueError( + f"Unexpected placement {current} for redistribute to target placement {target}" + ) + + if not async_op and isinstance( + new_local_tensor, funcol.AsyncCollectiveTensor + ): + new_local_tensor = new_local_tensor.wait() + local_tensor = new_local_tensor + return new_local_tensor + + +def _redistribute_backward( + grad_output: "dtensor.DTensor", + previous_spec: DTensorSpec, + *, + out_dtype: torch.dtype, + op_dtype: torch.dtype, + async_op: bool = False, +): + """ + Common function for redistributing a distributed tensor during backward + and twice-backward backpropagation steps. + + Args: + grad_output: The output gradient tensor. + previous_spec: DTensorSpec prior to redistribution. + out_dtype: dtype to cast the returned gradient to. Pinned by autograd + to match the dtype of the input of the node above this one. + op_dtype: dtype to run the backward collective at. + async_op: whether to perform the DTensor redistribute operation + asynchronously or not. Default: False + + Returns: + A :class:`torch.Tensor` object. + A :class:`DTensorSpec` object. + """ + if op_dtype != grad_output._local_tensor.dtype: + local_tensor = grad_output._local_tensor.to(dtype=op_dtype) + current_spec = DTensorSpec( + mesh=grad_output._spec.device_mesh, + placements=grad_output._spec.placements, + tensor_meta=TensorMeta( + shape=grad_output.shape, + stride=grad_output.stride(), + dtype=op_dtype, + ), + use_strided_shard_as_shard_order=grad_output._spec.use_strided_shard_as_shard_order, + ) + previous_spec = DTensorSpec( + mesh=previous_spec.device_mesh, + placements=previous_spec.placements, + tensor_meta=current_spec.tensor_meta, + use_strided_shard_as_shard_order=previous_spec.use_strided_shard_as_shard_order, + ) + else: + local_tensor = grad_output._local_tensor + current_spec = grad_output._spec + # skip the replicate to partial transformation when we are in backward pass + # In this case we keep the grad as replicate, this is because we don't + # want to convert the replicated gradients back to partial, although + # that's logically conform with the same layout, converting the gradients + # back to partial is actually useless as you would have to do reduce later + # which would be more expensive than keeping it replicate! + + # for backward shard -> partial, we just do shard -> replicate + # for backward replicate -> partial, we skip the transformation + # NOTE: _is_shard_like covers _StridedShard defensively; currently + # unreachable because Partial -> _StridedShard is not implemented. + normalized_placements: list[Placement] = [] + for current, target in zip(current_spec.placements, previous_spec.placements): + if (_is_shard_like(current) or current.is_replicate()) and target.is_partial(): + normalized_placements.append(Replicate()) + else: + normalized_placements.append(target) + + previous_spec = DTensorSpec( + previous_spec.device_mesh, + placements=tuple(normalized_placements), + tensor_meta=previous_spec.tensor_meta, + use_strided_shard_as_shard_order=previous_spec.use_strided_shard_as_shard_order, + ) + + output = redistribute_local_tensor( + local_tensor, + current_spec, + previous_spec, + async_op=async_op, + ) + + if output.dtype != out_dtype: + output = output.to(out_dtype) + + spec = DTensorSpec( + previous_spec.device_mesh, + tuple(normalized_placements), + tensor_meta=TensorMeta( + shape=grad_output.shape, + stride=grad_output.stride(), + dtype=output.dtype, + ), + use_strided_shard_as_shard_order=previous_spec.use_strided_shard_as_shard_order, + ) + return output, spec + + +class _BackwardDtypeConfig(TypedDict): + op_dtype: torch.dtype # dtype the backward collective runs at + out_dtype: torch.dtype # dtype the returned gradient is cast to + + +class _DtypeConfig(TypedDict): + op_dtype: torch.dtype # forward: dtype the collective runs at + out_dtype: torch.dtype # forward: dtype of the output tensor + backward_options: _BackwardDtypeConfig + + +class Redistribute(torch.autograd.Function): + @staticmethod + def forward( # type: ignore[override] + # pyre-fixme[2]: Parameter must be annotated. + ctx, + input: "dtensor.DTensor", + device_mesh: DeviceMesh, + placements: tuple[Placement, ...], + async_op: bool, + dtype_config: _DtypeConfig, + ): + ctx.async_op = async_op + bwd = dtype_config["backward_options"] + ctx.bwd_op_dtype = bwd["op_dtype"] + ctx.bwd_out_dtype = bwd["out_dtype"] + + op_dtype = dtype_config["op_dtype"] + out_dtype = dtype_config["out_dtype"] + + if op_dtype != input._local_tensor.dtype: + local_tensor = input._local_tensor.to(dtype=op_dtype) + current_spec = DTensorSpec( + mesh=device_mesh, + placements=input._spec.placements, + tensor_meta=TensorMeta( + shape=input.shape, + stride=input.stride(), + dtype=op_dtype, + ), + use_strided_shard_as_shard_order=input._spec.use_strided_shard_as_shard_order, + ) + else: + local_tensor = input._local_tensor + current_spec = input._spec + + ctx.current_spec = current_spec + + if current_spec.placements != placements: + target_spec = DTensorSpec( + device_mesh, placements, tensor_meta=current_spec.tensor_meta + ) + + output = redistribute_local_tensor( + local_tensor, + current_spec, + target_spec, + async_op=async_op, + is_explicit=True, + ) + else: + # use the same local tensor if placements are the same. + output = local_tensor + target_spec = current_spec + + if output.dtype != out_dtype: + output = output.to(out_dtype) + target_spec = DTensorSpec( + device_mesh, + target_spec.placements, + tensor_meta=TensorMeta( + shape=input.shape, + stride=input.stride(), + dtype=out_dtype, + ), + use_strided_shard_as_shard_order=target_spec.use_strided_shard_as_shard_order, + ) + + # pyrefly: ignore [bad-argument-type] + return dtensor.DTensor( + # pyrefly: ignore [bad-argument-count] + output, + target_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=input.requires_grad, + ) + + @staticmethod + def backward(ctx, grad_output: "dtensor.DTensor"): # type: ignore[override] + output_dtensor = NestedRedistribute.apply( + grad_output, + ctx.current_spec, + ctx.async_op, + ctx.bwd_op_dtype, + ctx.bwd_out_dtype, + ) + return ( + output_dtensor, + None, + None, + None, + None, + ) + + +class NestedRedistribute(torch.autograd.Function): + """ + This class is used to make the redistribution of a DTensor twice-differentiable. + This is called during the `Redistribute.forward`. + Therefore, `NestedRedistribute.forward` is called during the first backward pass, + and `NestedRedistribute.backward` is called during the second backward pass. + + Note: `NestedRedistribute.backward` is not differentiable, and therefore triple + backward is not yet supported. + """ + + @staticmethod + def forward( # type: ignore[override] + # pyre-fixme[2]: Parameter must be annotated. + ctx, + grad_output: "dtensor.DTensor", + previous_spec: DTensorSpec, + async_op: bool, + op_dtype: torch.dtype, + out_dtype: torch.dtype, + ): + # Persist op_dtype so the double-backward reuses the same collective + # dtype one level down. + ctx.async_op = async_op + ctx.original_dtype = grad_output._local_tensor.dtype + ctx.op_dtype = op_dtype + + output, spec = _redistribute_backward( + grad_output, + previous_spec, + out_dtype=out_dtype, + op_dtype=op_dtype, + async_op=async_op, + ) + + ctx.current_spec = spec + + # pyrefly: ignore [bad-argument-type] + return dtensor.DTensor( + # pyrefly: ignore [bad-argument-count] + output, + spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=grad_output.requires_grad, + ) + + @staticmethod + def backward(ctx, grad2_output: "dtensor.DTensor"): # type: ignore[override] + previous_spec = ctx.current_spec + # Reuse the same op_dtype one level down; pin out_dtype to the dtype + # of the grad we received at this level, since that's what the node + # above us expects back. + output_dtensor = NestedRedistribute.apply( + grad2_output, + previous_spec, + ctx.async_op, + ctx.op_dtype, + ctx.original_dtype, + ) + + return ( + output_dtensor, + None, + None, + None, + None, + None, + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_sharding_prop.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_sharding_prop.py new file mode 100644 index 0000000000000000000000000000000000000000..0a4e6f4e66fdb9fafd7cb8f1dae1997a3eff8864 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_sharding_prop.py @@ -0,0 +1,1119 @@ +# mypy: allow-untyped-defs +import logging +import threading +from collections.abc import Callable, Sequence +from contextlib import nullcontext +from functools import lru_cache +from itertools import chain +from typing import cast + +import torch +from torch._guards import detect_fake_mode +from torch._logging import LazyString +from torch._ops import OpOverload +from torch._subclasses import FakeTensorMode +from torch.distributed._functional_collectives import _are_we_tracing +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._decompositions import DecompShardingStrategy +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpInfo, + OpSchema, + OpSpec, + OpStrategy, + OutputSharding, + OutputSpecType, + RuntimeSchemaInfo, + StrategyType, + TupleStrategy, +) +from torch.distributed.tensor._ops.single_dim_strategy import ( + _expand_single_dim_strategy_to_mesh, + _SingleDimStrategyInfo, +) +from torch.distributed.tensor._utils import ( + compute_local_shape_and_global_offset, + compute_local_stride, + try_find_mesh_from_args, +) +from torch.distributed.tensor.placement_types import _StridedShard, Shard +from torch.utils._pytree import tree_map + + +aten = torch.ops.aten + +log = logging.getLogger(__name__) + + +def _propagate_use_strided_shard_flag( + op_strategy: OpStrategy, + op_schema: OpSchema, +) -> None: + """Propagate use_strided_shard_as_shard_order from input specs to output specs. + + When inputs carry _StridedShard with an explicit flag, all output (and input) + DTensorSpecs in the strategy that also contain _StridedShard must agree. + Strategy functions may forget to propagate the flag; this function fixes + them up centrally after the strategy is produced. + """ + _use_strided: bool | None = None + for spec in op_schema.args_spec: + if any(isinstance(p, _StridedShard) for p in spec.placements): + val = spec.use_strided_shard_as_shard_order + if _use_strided is not None and _use_strided != val: + _use_strided = True + else: + _use_strided = val + + if _use_strided is None: + return + + def _fixup(spec: DTensorSpec) -> None: + if not any(isinstance(p, _StridedShard) for p in spec.placements): + return + if spec.use_strided_shard_as_shard_order == _use_strided: + return + spec.use_strided_shard_as_shard_order = _use_strided + if _use_strided: + spec.shard_order = None # pyrefly: ignore[bad-assignment] + else: + spec.shard_order = DTensorSpec.compute_default_shard_order(spec.placements) + + for op_spec in op_strategy.strategies: + out = op_spec.output_specs + if out is not None: + if isinstance(out, DTensorSpec): + _fixup(out) + else: + for s in out: + if s is not None: + _fixup(s) + if op_spec.input_specs is not None: + for s in op_spec.input_specs: + _fixup(s) + + +def _length(obj) -> int: + if obj is None: + return 0 + if not isinstance(obj, Sequence): + return 1 + return len(obj) + + +def _get_expected_num_tensor_outputs(op: OpOverload) -> int | None: + """ + Get the expected number of tensor outputs for an operator based on its schema. + + Returns: + The number of tensor outputs expected. Returns 0 for ops that don't return tensors + (e.g., _linalg_check_errors). Returns 1 for single tensor return, and >1 for + tuple returns where each element is a tensor. Returns None for List[Tensor] + returns where the length is unknown at schema time. + """ + return_types = op._schema.returns + if len(return_types) == 0: + return 0 + + first_return = return_types[0] + if isinstance(first_return.type, torch.TensorType): + # Could be single tensor or tuple of tensors + return len(return_types) + elif isinstance(first_return.type, torch.ListType): + # List[Tensor] - we don't know the length at schema time + return None + else: + # Not a tensor return type + return 0 + + +def _validate_tensor_meta_count( + op_schema: OpSchema, + tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, +) -> None: + """ + Validate that the tensor_meta matches the expected number of outputs for the op. + + Raises AssertionError if the count doesn't match, providing a helpful error message. + """ + expected_outputs = _get_expected_num_tensor_outputs(op_schema.op) + + # Compute actual count: + # - None means 0 outputs + # - TensorMeta (single instance) means 1 output + # - Sequence of TensorMeta means len(sequence) outputs + # Note: TensorMeta is a NamedTuple (subclass of tuple), so we must check for it first + if tensor_meta is None: + actual_outputs = 0 + elif isinstance(tensor_meta, TensorMeta): + actual_outputs = 1 + else: + actual_outputs = len(tensor_meta) + + if expected_outputs is None: + # List[Tensor] return type: length unknown at schema time, but + # tensor_meta must be a list of TensorMeta. + if not isinstance(tensor_meta, list): + raise AssertionError( + f"Tensor meta for {op_schema.op} should be a list[TensorMeta] " + f"(op returns List[Tensor]), but got {type(tensor_meta).__name__}" + ) + return + + if actual_outputs != expected_outputs: + raise AssertionError( + f"Tensor meta count mismatch for {op_schema.op}: " + f"expected {expected_outputs} tensor output(s) based on op schema, " + f"but _propagate_tensor_meta returned {actual_outputs}. " + f"This usually indicates a bug in fake tensor propagation for this op." + ) + + +class LocalLRUCache(threading.local): + def __init__(self, user_function: Callable) -> None: + self.cache = lru_cache(None)(user_function) + + def __call__(self, *args, **kwargs) -> object: + # Fast path: log.handlers check is very cheap (just checking if list is non-empty) + # Only do the more expensive isEnabledFor check if handlers exist + if log.handlers and log.isEnabledFor(logging.DEBUG): + info_before = self.cache.cache_info() + result = self.cache(*args, **kwargs) + info_after = self.cache.cache_info() + cache_hit = info_after.hits > info_before.hits + op_schema = args[0] if args else None + output_spec = getattr(result, "output_spec", None) + log.debug( + "sharding_prop python cache %s: %s -> %s", + "HIT" if cache_hit else "MISS", + op_schema, + output_spec, + ) + return result + return self.cache(*args, **kwargs) + + def cache_info(self): + return self.cache.cache_info() + + def cache_clear(self): + return self.cache.cache_clear() + + +def _format_unbacked_hinting_log( + op_schema: OpSchema, + strategies: list[OpSpec], + strategy_index: int, + replacements: dict, +) -> str: + """Format log message for unbacked hinting strategy selection (only called if debug logging enabled).""" + args_spec = tuple(str(spec) for spec in op_schema.args_schema) + strat = strategies[strategy_index] + if strat.input_specs is None: + placements_in = None + else: + placements_in = tuple( + spec.format_shard_order_str(spec.placements, spec.shard_order) + for spec in strat.input_specs + ) + placements_out = tree_map( + lambda spec: spec.format_shard_order_str(spec.placements, spec.shard_order), + strat.output_specs, + is_leaf=lambda x: isinstance(x, DTensorSpec), + ) + return ( + f"Selected strategy {placements_in} -> {placements_out} " + f"for {op_schema.op} with input {args_spec}, using unbacked hints: {replacements}" + ) + + +def _select_min_redistribute_cost( + costs: list[torch.types.FloatLikeType], + strategies: list[OpSpec], + op_schema: OpSchema | None = None, +) -> int: + """ + Given a list of costs and corresponding op strategies, selects the minimum cost strategy, returning the index. + If unbacked symbols are involved, replaces them with known upper-bound values, falling back to hardcoded values. + """ + from torch.fx.experimental.symbolic_shapes import ( + free_unbacked_symbols, + is_concrete_float, + ) + from torch.utils._sympy.interp import sympy_interp + from torch.utils._sympy.numbers import int_oo + from torch.utils._sympy.reference import PythonReferenceAnalysis + + int_fallback = 8192 + free_unbacked = list(set(chain(*[free_unbacked_symbols(cost) for cost in costs]))) + + # Easy path: no unbacked shapes involved, choose min cost strategy. + # Doing the hard path for backed could also make sense? + if all(is_concrete_float(c) for c in costs) or not free_unbacked: + return costs.index(min(costs)) + + # Figure out heuristic hints for unbacked shapes. + # If available, use shape upper bound. If not, fallback to some integer (inductor size-hinting style). + shape_env = next(iter(x for x in costs if not is_concrete_float(x))).node.shape_env # type: ignore[arg-type] + replacements = {} + for sym in free_unbacked: + # TODO(laithsakka): unify with optimization_hint API + if (hint := shape_env.var_to_hint_override.get(sym)) is not None: + replacements[sym] = hint + elif (upper := shape_env.bound_sympy(sym).upper) is not int_oo: + replacements[sym] = upper + else: + replacements[sym] = int_fallback + + # Use replacements for redistribute cost hints + proxy_costs = [ + float(cost) + if is_concrete_float(cost) + else sympy_interp( + PythonReferenceAnalysis, + replacements, + cost.node.expr.xreplace(replacements), # type: ignore[arg-type] + ) + for cost in costs + ] + min_cost = min(proxy_costs) + strategy_index = proxy_costs.index(min_cost) + + if op_schema: + log.debug( + "%s", + LazyString( + _format_unbacked_hinting_log, + op_schema, + strategies, + strategy_index, + replacements, + ), + ) + return strategy_index + + +def _select_min_cost_strategy( + strategy: OpStrategy, op_schema: OpSchema | None = None +) -> OpSpec: + from torch.fx.experimental.symbolic_shapes import guard_or_false + + if len(strategy.strategies) == 1: + # short cut with only one possible OpSpec + return strategy.strategies[0] + + op_spec_costs: list[torch.types.FloatLikeType] = [] + no_redistribute_strategy_index: int = -1 + negative_cost_index: int = -1 + zero_cost_index: int = -1 + for strategy_idx, op_spec in enumerate(strategy.strategies): + if op_spec.redistribute_cost is None: + raise AssertionError("must set redistribute cost each OpSpec!") + redistribute_cost = sum(chain.from_iterable(op_spec.redistribute_cost)) + op_spec_costs.append(redistribute_cost) + + # If there are strategies with negative/zero/no redistribute cost, + # we record those indices. + # TODO: Currently this only applies to OpStrategy selection. Requires extra + # logic to make it work for TupleStrategy, if needed. + if op_schema is not None: + if guard_or_false(redistribute_cost < 0): + if ( + negative_cost_index == -1 + or redistribute_cost < op_spec_costs[negative_cost_index] + ): + negative_cost_index = strategy_idx + elif guard_or_false(redistribute_cost == 0): + needs_redistribute = False + for spec_idx, input_spec in enumerate(op_schema.args_spec): + desired_spec = ( + op_spec.output_spec + if op_spec.input_specs is None + else op_spec.input_specs[spec_idx] + ) + if input_spec.placements != desired_spec.placements: + needs_redistribute = True + break + + if not needs_redistribute: + no_redistribute_strategy_index = strategy_idx + elif zero_cost_index == -1: + zero_cost_index = strategy_idx + + # prioritize negative/zero/no redistribute cost strategies + if negative_cost_index != -1: + # If there's negative cost, we select the one with the minimal cost, + # even if this means we need to redistribute, e.g. via local chunking. + # E.g. this can happen for ops in self.op_to_shape_and_stride_idx + # when the inputs / outputs are sharded. + selected_strategy_index = negative_cost_index + elif no_redistribute_strategy_index != -1: + selected_strategy_index = no_redistribute_strategy_index + elif zero_cost_index != -1: + selected_strategy_index = zero_cost_index + else: + # default to choosing minimal redistribute cost + selected_strategy_index = _select_min_redistribute_cost( + op_spec_costs, strategy.strategies, op_schema + ) + + return strategy.strategies[selected_strategy_index] + + +class ShardingPropagator: + # Lock to protect FakeTensorMode context during tensor meta propagation. + # By default this is a no-op (nullcontext) for performance. Multi-threaded + # tests should set this to threading.Lock() to prevent race conditions + # when multiple threads enter different FakeTensorMode contexts. + _fake_mode_lock = nullcontext() + + def __init__(self) -> None: + self.op_to_rules: dict[OpOverload, Callable[[OpSchema], OutputSharding]] = {} + self.op_strategy_funcs: dict[ + OpOverload, + Callable[[OpSchema], StrategyType], + ] = {} + self.op_single_dim_strategy_funcs: dict[ + OpOverload, + _SingleDimStrategyInfo, + ] = {} + # op map to save static argnum to decide to reuse sharding prop cache or + # re-run sharding prop + self.op_to_schema_info: dict[OpOverload, RuntimeSchemaInfo] = {} + self.op_to_schema_info_for_single_dim_strategy: dict[ + OpOverload, RuntimeSchemaInfo + ] = {} + self.propagate_op_sharding = LocalLRUCache( + self.propagate_op_sharding_non_cached + ) + self.decomp_strategy = DecompShardingStrategy(self) + # op map to save indices of shape (and stride) args which may need to be + # modified in sharding prop + self.op_to_shape_and_stride_idx: dict[OpOverload, int | tuple[int, int]] = { + # new factory ops + aten.new_empty.default: 1, + aten.new_full.default: 1, + aten.new_ones.default: 1, + aten.new_zeros.default: 1, + aten.new_empty_strided.default: (1, 2), + # view ops + aten.expand.default: 1, + aten.expand_copy.default: 1, + aten.reshape.default: 1, + aten.view.default: 1, + aten.view_copy.default: 1, + aten._unsafe_view.default: 1, + aten.select_backward.default: 1, + aten.slice_backward.default: 1, + # upsample backward ops: input_size (arg 2) is the full [N,C,...] shape + aten.upsample_nearest1d_backward.default: 2, + aten.upsample_nearest2d_backward.default: 2, + aten.upsample_nearest3d_backward.default: 2, + aten._upsample_nearest_exact1d_backward.default: 2, + aten._upsample_nearest_exact2d_backward.default: 2, + aten._upsample_nearest_exact3d_backward.default: 2, + aten._upsample_bilinear2d_aa_backward.default: 2, + aten._upsample_bicubic2d_aa_backward.default: 2, + aten._upsample_lanczos2d_aa_backward.default: 2, + aten.upsample_bicubic2d_backward.default: 2, + aten.upsample_bilinear2d_backward.default: 2, + aten.upsample_linear1d_backward.default: 2, + aten.upsample_trilinear3d_backward.default: 2, + } + # ops with individual scalar shape args that need local adjustment + # maps op -> callable(input_specs, schema) -> adjusted schema + # populated by op modules (e.g. _math_ops.py) at registration time + self.op_to_scalar_shape_adjuster: dict[ + OpOverload, + Callable[[list[DTensorSpec], OpSchema], OpSchema], + ] = {} + # squeeze ops that need dim arg rewritten to only globally-singleton dims + self.squeeze_op_to_dims_variant: dict[OpOverload, OpOverload] = { + aten.squeeze.default: aten.squeeze.dims, + aten.squeeze.dim: aten.squeeze.dims, + aten.squeeze.dims: aten.squeeze.dims, + aten.squeeze_.default: aten.squeeze_.dims, + aten.squeeze_.dim: aten.squeeze_.dims, + aten.squeeze_.dims: aten.squeeze_.dims, + } + + def register_sharding_prop_rule( + self, + op_overload: OpOverload, + rule_func: Callable[[OpSchema], OutputSharding], + schema_info: RuntimeSchemaInfo | None = None, + ): + """ + Register a sharding propagation rule for an operator. + """ + self.op_to_rules[op_overload] = rule_func + if schema_info is not None: + self.op_to_schema_info[op_overload] = schema_info + + def register_single_dim_op_strategy( + self, + op_overload: OpOverload, + strategy_info: _SingleDimStrategyInfo, + schema_info: RuntimeSchemaInfo | None = None, + ): + """ + Register a strategy over a single mesh-dim, relying on infra to automatically expand to the full mesh. + """ + self.op_single_dim_strategy_funcs[op_overload] = strategy_info + if schema_info is not None: + self.op_to_schema_info_for_single_dim_strategy[op_overload] = schema_info + + def register_op_strategy( + self, + op_overload: OpOverload, + strategy_func: Callable[[OpSchema], StrategyType], + schema_info: RuntimeSchemaInfo | None = None, + ): + """ + Register a :class:`OpStrategy` generator for an operator. + + During the sharding propagation, DTensor wants to enumerate all + acceptable sharding specs (:class:`OpSpec`) for an operator, + and by "acceptable" we mean that the operator can be executed on + the ``_local_tensor`` of DTensor args/kwargs (with ``OpSpec.input_specs``) + and the output(s) constitute valid DTensor(s) (with ``OpSpec.output_specs``). + + ``strategy_func`` is the function that enumerates such acceptable specs + for the operator ``op_overload``. One general approach to write ``strategy_func`` + is, if the operator has simple arguments structure (e.g. mm, bmm), first enumerating + all sharding specs for the operands, and then filtering out the ones that + are not valid. For example, for ``mm``, the operands are two 2D tensors, and + if both ``input`` and ``mat2`` have sharding placements ``[Shard(0)]``, then this + is not an acceptable ``input_specs``. + + Once we have a way to enumerate all acceptable sharding specs, we can use each + of them to construct a :class:`OpSpec`. The ``OpSpec.input_specs`` directly comes + from the sharding spec, and the ``OpSpec.output_specs`` is therefore determined + (e.g. ``[Shard(1)]`` @ ``[Shard(0)]`` yields ``[Partial()]``). In addition, + :class:`OpSpec` also contains ``redistribute_cost`` which records the redistribution + cost from each :class:`OpSpec` in the source :class:`OpStrategy.strategies` to + the target sharding spec, for each operand. + + The ``strategy_func`` should return a :class:`OpStrategy` which contains a list of + all the :class:`OpSpec`s generated in the above. + + The optional ``schema_info`` tells which non-DTensor args/kwargs could affect the + cache and whether ``pytree`` is needed to flatten the nested args. ``static_argnum`` + marks the starting index of the non-DTensor args that should be hashed into the + sharding propagation hash key, and ``static_kwargkey`` marks the keys of the + non-DTensor kwargs that should be hashed. ``needs_pytree`` should be used when + the input arg has :class:`list` or :class:`dict` structure. + + For example, ``aten.cat.default`` op has a ``List[Tensor]`` argument ``tensors`` + and an ``int`` argument ``dim``. Because ``dim`` affects the sharding propagation + result, we want to pass ``RuntimeSchemaInfo(static_argnum=1)`` because the argument + index of ``dim`` is 1. Besides, we also want to set ``needs_pytree=True`` because + ``tensors`` needs be flattened in sharding propagation. Another example is + ``aten.histc.default``. ``histc`` has 4 arguments (self, bins, min, max) and the + last two would affect sharding propagation along with the :class:`DTensor` argument + ``self``. Since the argument index of ``min`` is 2, the `schema_info` should be + `RuntimeSchemaInfo(static_argnum=2)`. + """ + self.op_strategy_funcs[op_overload] = strategy_func + if schema_info is not None: + self.op_to_schema_info[op_overload] = schema_info + + def _propagate_tensor_meta_non_cached( + self, op_schema: OpSchema + ) -> TensorMeta | Sequence[TensorMeta | None] | None: + """ + Propagate the tensor metadata, it could either return a TensorMeta + or a list/tuple of TensorMetas + """ + if op_schema.op == aten.equal.default: + # data dependent ops can't be used for fake propagation + return None + + # NOTE: We must call the tracing in fake tensor mode so that it avoids + # materializing memory. + # NOTE: Use _fake_mode_lock to serialize access when running in + # multi-threaded tests (lock must be set to threading.Lock()). + # This is a nullcontext by default. + # NOTE: disable_proxy_modes_tracing() prevents make_fx from tracing + # into shard propagation. This op runs purely to derive output + # metadata (global shape/stride/dtype); it is not part of the real + # (local) computation. Without disabling the proxy tracer, the + # gen_fake_args() factory calls and op_schema.op(...) get recorded + # into the enclosing graph as dead code -- global-shape "shadow" nodes + # on empty_strided placeholders that no real output consumes. + from torch.fx.experimental.proxy_tensor import disable_proxy_modes_tracing + + with ShardingPropagator._fake_mode_lock: + fake_mode = detect_fake_mode() or FakeTensorMode() + with fake_mode, disable_proxy_modes_tracing(): + fake_args = op_schema.gen_fake_args() + fake_kwargs = op_schema.gen_fake_kwargs() + fake_out = op_schema.op(*fake_args, **fake_kwargs) + + if isinstance(fake_out, torch.Tensor): + return TensorMeta( + shape=fake_out.shape, stride=fake_out.stride(), dtype=fake_out.dtype + ) + + elif isinstance(fake_out, (tuple, list)): + tensor_meta_list: list[TensorMeta | None] = [] + for fake_out_item in fake_out: + if isinstance(fake_out_item, torch.Tensor): + tensor_meta_list.append( + TensorMeta( + shape=fake_out_item.shape, + stride=fake_out_item.stride(), + dtype=fake_out_item.dtype, + ) + ) + else: + tensor_meta_list.append(None) + return ( + tuple(tensor_meta_list) + if isinstance(fake_out, tuple) + else tensor_meta_list + ) + else: + # if fake is not a tensor or tuple of tensor, return as none + return None + + @lru_cache # noqa: B019 + def _propagate_tensor_meta_cached( + self, op_schema: OpSchema + ) -> TensorMeta | Sequence[TensorMeta | None] | None: + """ + Cached version of _propagate_tensor_meta_non_cached + Use _propagate_tensor_meta instead to handle dynamic shapes. + """ + return self._propagate_tensor_meta_non_cached(op_schema) + + def _propagate_tensor_meta( + self, op_schema: OpSchema + ) -> TensorMeta | Sequence[TensorMeta | None] | None: + """ + Propagate the tensor metadata, it could either return a TensorMeta + or a list/tuple of TensorMetas. Uses the cached version if not + actively tracing. Use this method instead of _propagate_tensor_meta_non_cached + """ + if _are_we_tracing(): + return self._propagate_tensor_meta_non_cached(op_schema) + else: + return self._propagate_tensor_meta_cached(op_schema) + + def _create_output_spec_with_new_tensor_meta( + self, + op: OpOverload, + output_specs: OutputSpecType, + output_tensor_meta: TensorMeta | Sequence[TensorMeta | None] | None, + ) -> OutputSpecType: + """ + Wrap the output_specs with the tensor metadata from the output. + """ + + if isinstance(output_specs, DTensorSpec): + if not isinstance(output_tensor_meta, TensorMeta): + # Either error due to ShardingPropagator or due to incorrect OutputSpec + if not isinstance(output_tensor_meta, (tuple, list)): + raise ValueError( + "ShardingPropagator error: output does not have an associated " + "TensorMeta" + ) + raise ValueError( + f"For the op {op.name()}, `output_specs` has 1 output which does " + "not equal the " + f"number of op outputs: {len(output_tensor_meta)}." + ) + return output_specs.shallow_copy_with_tensor_meta(output_tensor_meta) + elif isinstance(output_specs, (tuple, list)): + new_specs: list[DTensorSpec | None] = [] + if not isinstance(output_tensor_meta, (tuple, list)) or len( + output_specs + ) != len(output_tensor_meta): + raise ValueError( + f"For the op {op.name()}, `output_specs` has {len(output_specs)} " + "outputs which does not equal the " + f"number of op outputs {_length(output_tensor_meta)}." + ) + + # pyrefly: ignore [bad-argument-type] + for i, spec in enumerate(output_specs): + if isinstance(spec, DTensorSpec): + output_tensor_meta_i = output_tensor_meta[i] + if not isinstance(output_tensor_meta_i, TensorMeta): + # Some ops (e.g. convolution_backward, native_layer_norm_backward, + # _fused_rms_norm_backward) have an output_mask parameter that + # controls which outputs are computed. When output_mask[i] is + # False, the output at position i is None and has no TensorMeta. + if output_tensor_meta_i is None: + new_specs.append(None) + continue + else: + raise ValueError( + f"ShardingPropagator error: output {i} of {op.name()} " + "does not have an associated TensorMeta" + ) + + new_specs.append( + spec.shallow_copy_with_tensor_meta(output_tensor_meta_i) + ) + else: + new_specs.append(spec) + + return tuple(new_specs) + else: + if output_specs is not None: + raise AssertionError + return output_specs + + def _wrap_with_op_strategy(self, op_schema: OpSchema) -> OpSchema: + """ + wrap a op_schema that contains DTensorSpec to another op_schema that contains + OpStrategy/TupleStrategy, the returned op_schema is then used for sharding + strategy propagation on pytorch operators. + """ + + def spec_to_strategy(spec: object) -> object: + if isinstance(spec, DTensorSpec): + return OpStrategy([OpSpec(spec)]) + elif isinstance(spec, (list, tuple)) and len(spec) > 0: + if all(isinstance(s, DTensorSpec) for s in spec): + # tensor list create tuple strategy + tuple_strategy = [spec_to_strategy(s) for s in spec] + tuple_strategy = cast(Sequence[StrategyType], tuple_strategy) + return TupleStrategy( + tuple(tuple_strategy) + if isinstance(spec, tuple) + else tuple_strategy + ) + elif any(isinstance(s, DTensorSpec) for s in spec): + # mixed list (e.g. [DTensorSpec, None, DTensorSpec]) for + # ops like aten.index.Tensor; keep as list so pytree + # flattening can extract OpStrategy items + return [spec_to_strategy(s) for s in spec] + else: + return spec + else: + return spec + + args_op_strategy = [spec_to_strategy(i) for i in op_schema.args_schema] + + kwargs_op_strategy = { + k: spec_to_strategy(v) for k, v in op_schema.kwargs_schema.items() + } + + return OpSchema( + op=op_schema.op, + args_schema=tuple(args_op_strategy), + kwargs_schema=kwargs_op_strategy, + schema_info=op_schema.schema_info, + ) + + def propagate(self, op_info: OpInfo) -> None: + # NB: The logic here is duplicated in _propagate_op_sharding_dispatch_slow_path. + # Ideally, this function would be deleted, but there are a handful of + # one off call sites here that aren't cleaned up. + + # NOTE: schema should always be populated when calling this function, + # as it's only called after unwrap_to_op_info (create_schema=True). + if op_info.schema is None: + raise AssertionError( + "op_info.schema should not be None in propagate. " + "This function should only be called after unwrap_to_op_info." + ) + + # We cannot use an lru cache if we know that inputs will have dynamic shapes, + # because SymInts are not hashable. + # This is generally ok because this only happens during tracing in torch.compile, + # and tracing does not need to be as fast as eagermode DTensor usages. + if _are_we_tracing(): + output_sharding = self.propagate_op_sharding_non_cached(op_info.schema) + else: + output_sharding = cast( + OutputSharding, self.propagate_op_sharding(op_info.schema) + ) + op_info.output_sharding = output_sharding + + def propagate_op_sharding_non_cached(self, op_schema: OpSchema) -> OutputSharding: + """ + Propagate the sharding for an operator given the op_schema. + """ + # no-op in OSS, logs API usage metrics in meta-internal runs + torch._C._log_api_usage_once( + "torch.distributed.tensor._sharding_prop.ShardingPropagator.propogate_op_sharding_non_cached" + ) + # special case op, we don't need to propagate for local + # scalar. TODO: figure out a better way to handle this + if op_schema.op is aten._local_scalar_dense.default: + return OutputSharding(None, op_schema) + + out_tensor_meta = self._propagate_tensor_meta_non_cached(op_schema) + + single_dim_strategy_info = self.op_single_dim_strategy_funcs.get(op_schema.op) + op_strategy_func = self.op_strategy_funcs.get(op_schema.op) + decomp_exception = None + if single_dim_strategy_info is not None or op_strategy_func is not None: + # Validate that tensor_meta count matches expected outputs from op schema. + # This catches bugs in fake tensor propagation early. + if single_dim_strategy_info is not None: + _validate_tensor_meta_count(op_schema, out_tensor_meta) + """ + Given the single_dim_strategy, which is just a minimal set of valid input-output placement specifications + for the operator over a single mesh dimension, + + And the OpSchema, which includes information about the runtime input tensor placements, and the mesh, + + Combine single_dim_strategies across mesh dims, also expanding placeholders (ShardPlaceholder) to any real + sharding types in op_schema, and find the lowest cost redistribution of inputs to match a valid strategy + combination. + """ + # wrap the op_schema with op strategy for sharding strategy propagation + strategy_schema = self._wrap_with_op_strategy(op_schema) + + if single_dim_strategy_info is not None: + mesh = try_find_mesh_from_args( + op_schema.op, + (*op_schema.args_schema, *op_schema.kwargs_schema.values()), + ) + if not isinstance(mesh, DeviceMesh): + raise AssertionError("Expected to find a valid mesh") + # expand to generate the full set of strategy combinations, each one + # with a redistribute cost, and then find the min strategy over those costs. + _expanded_strategy_fn = _expand_single_dim_strategy_to_mesh( + mesh, strategy_schema, single_dim_strategy_info, out_tensor_meta + ) + op_strategy = _expanded_strategy_fn( + op_schema.op, strategy_schema.args_meta, strategy_schema.kwargs_meta + ) + else: + if op_strategy_func is None: + raise AssertionError + op_strategy = op_strategy_func(strategy_schema) + + else: + # try operator decomposition path + + op_strategy = None + if DecompShardingStrategy.has_decomp(op_schema.op): + # Ensure schema_info is registered for proper cache key computation + self.decomp_strategy.ensure_schema_info(op_schema.op) + try: + op_strategy = self.decomp_strategy.propagate_strategy( + op_schema, + out_tensor_meta, + ) + except Exception as e: + decomp_exception = e + + if op_strategy is not None: + if isinstance(op_strategy, OpStrategy): + _propagate_use_strided_shard_flag(op_strategy, op_schema) + # single Op strategy + output_strategy = _select_min_cost_strategy(op_strategy, op_schema) + + # check if we need to redistribute the input + needs_redistribute = False + # check if we want to use args value from redistribute_schema + use_val_from_redistribute_schema = False + expected_input_specs: list[DTensorSpec] = [] + + # in case where the op does not specify input_specs and output_specs + # is a DTensorSpec, we use output_specs as the spec for each DTensor + # input arg. + if output_strategy.input_specs is None: + if not isinstance(output_strategy.output_specs, DTensorSpec): + raise AssertionError + + for idx, input_spec in enumerate(op_schema.args_spec): + desired_spec = ( + output_strategy.output_spec + if output_strategy.input_specs is None + else output_strategy.input_specs[idx] + ) + expected_input_specs.append( + desired_spec.shallow_copy_with_tensor_meta( + input_spec.tensor_meta + ) + ) + if input_spec.placements != desired_spec.placements: + needs_redistribute = True + + suggestion_schema = None + if needs_redistribute: + suggestion_schema = OpSchema( + op_schema.op, tuple(expected_input_specs), {} + ) + suggestion_schema._inplace_rewrap_schema_suggestion(op_schema) + + # Rewrite shape/stride args from global to local when sharded. + # Covers view ops, new factory ops, and upsample backward ops + # whose shape args refer to global tensor dimensions. + if op_schema.op in self.op_to_shape_and_stride_idx: + if not isinstance(output_strategy.output_spec, DTensorSpec): + raise AssertionError + if any( + isinstance(p, Shard | _StridedShard) + for p in output_strategy.output_spec.placements + ): + schema = suggestion_schema or op_schema + if not isinstance(out_tensor_meta, TensorMeta): + raise AssertionError + suggestion_schema = self._adjust_shape_and_stride_args( + out_tensor_meta, schema, output_strategy.output_spec + ) + needs_redistribute = True + use_val_from_redistribute_schema = True + + # adjust individual scalar shape args (e.g. N, C, HxW in group_norm) + if op_schema.op in self.op_to_scalar_shape_adjuster: + if any( + isinstance(p, Shard | _StridedShard) + for spec in expected_input_specs + for p in spec.placements + ): + schema = suggestion_schema or op_schema + adjuster = self.op_to_scalar_shape_adjuster[op_schema.op] + suggestion_schema = adjuster(expected_input_specs, schema) + needs_redistribute = True + use_val_from_redistribute_schema = True + + # rewrite squeeze to use only globally-singleton dims + if op_schema.op in self.squeeze_op_to_dims_variant: + schema = suggestion_schema or op_schema + adjusted = self._adjust_squeeze_to_global_singletons(schema) + if adjusted is not None: + suggestion_schema = adjusted + needs_redistribute = True + use_val_from_redistribute_schema = True + + # construct output spec for the op + if op_schema.return_type_tuple_tensor_like(): + # for ops that return multiple tensors and the output_specs is not + # a tuple, we use a tuple of that single output spec as the new + # output_specs + output_specs: OutputSpecType = output_strategy.output_specs + if isinstance(output_specs, DTensorSpec): + output_specs = tuple( + # create a new DTensorSpec with the same placement as the + # output_specs in output_strategy + DTensorSpec( + mesh=output_specs.mesh, + placements=output_specs.placements, + tensor_meta=output_specs.tensor_meta, + use_strided_shard_as_shard_order=output_specs.use_strided_shard_as_shard_order, + ) + for _ in range(len(op_schema.op._schema.returns)) + ) + elif ( + op_schema.return_type_tensor() + or op_schema.return_type_list_tensor_like() + ): + output_specs = output_strategy.output_specs + else: + output_specs = None + + output_sharding = OutputSharding( + output_specs, + suggestion_schema, + needs_redistribute=needs_redistribute, + use_val_from_redistribute_schema=use_val_from_redistribute_schema, + ) + elif isinstance(op_strategy, TupleStrategy): + # tuple strategy output sharding processing + # runtime select OpSpec for each TupleStrategy input arg + selected_strategies: list[OpSpec] = [] + out_spec_list: list[DTensorSpec] = [] + for strategy in op_strategy.children: + if not isinstance(strategy, OpStrategy): + raise AssertionError + _propagate_use_strided_shard_flag(strategy, op_schema) + selected_strategy = _select_min_cost_strategy(strategy) + selected_strategies.append(selected_strategy) + if selected_strategy.output_specs is not None: + out_spec_list.append(selected_strategy.output_spec) + + needs_redistribute = False + suggestion_args: list[object] = [] + tensor_or_list_tensor_arg_idx = 0 + + for arg in op_schema.args_schema: + if ( + arg + and isinstance(arg, (list, tuple)) + and isinstance(arg[0], DTensorSpec) + ): + expected_input_spec_list: list[DTensorSpec] = [] + for idx, arg_spec in enumerate(arg): + expected_input_spec = selected_strategies[idx].input_spec( + tensor_or_list_tensor_arg_idx + ) + expected_input_spec = ( + expected_input_spec.shallow_copy_with_tensor_meta( + arg_spec.tensor_meta + ) + ) + if arg_spec.placements != expected_input_spec.placements: + needs_redistribute = True + expected_input_spec_list.append(expected_input_spec) + suggestion_args.append( + tuple(expected_input_spec_list) + if isinstance(arg, tuple) + else expected_input_spec_list + ) + tensor_or_list_tensor_arg_idx += 1 + + elif isinstance(arg, DTensorSpec): + expected_input_spec = selected_strategies[0].input_spec( + tensor_or_list_tensor_arg_idx + ) + expected_input_spec = ( + expected_input_spec.shallow_copy_with_tensor_meta( + arg.tensor_meta + ) + ) + if arg.placements != expected_input_spec.placements: + needs_redistribute = True + suggestion_args.append(expected_input_spec) + tensor_or_list_tensor_arg_idx += 1 + else: + suggestion_args.append(arg) + + suggestion_schema = None + if needs_redistribute: + suggestion_schema = OpSchema( + op_schema.op, tuple(suggestion_args), op_schema.kwargs_schema + ) + + output_sharding = OutputSharding( + tuple(out_spec_list) if out_tensor_meta is not None else None, + suggestion_schema, + needs_redistribute=needs_redistribute, + use_val_from_redistribute_schema=False, + ) + else: + raise ValueError("Unsupported op strategy type") + + # associate the output sharding with the output tensor metadata + new_output_spec = self._create_output_spec_with_new_tensor_meta( + op_schema.op, output_sharding.output_spec, out_tensor_meta + ) + output_sharding.output_spec = new_output_spec + return output_sharding + elif op_schema.op in self.op_to_rules: + # propagate the sharding with rule + sharding_prop_func = self.op_to_rules[op_schema.op] + + # step 1. there's sharding propagation rule, run + # sharding propagation to get the output sharding + try: + output_sharding = sharding_prop_func(op_schema) + except NotImplementedError as e: + raise e + except Exception as e: + raise RuntimeError( + f"Sharding propagation failed on op {op_schema}.\nError: {e}" + ) from e + + # step 2. if can't get output_spec from sharding + # propagation (i.e. no rules apply for input + # placements), we return the output sharding + # with schema suggestions, which can be used to + # decide how to do redistribute on inputs + if output_sharding.output_spec is None: + if output_sharding.redistribute_schema is None: + raise RuntimeError( + f"Sharding propagation failed on op {op_schema}!" + ) + else: + # we do auto redistribute on inputs if necessary + # run sharding propagation again with suggested schema + propagation_res = sharding_prop_func( + output_sharding.redistribute_schema + ) + # we set the output sharding with the new propagation result + # so that dispatching know both output_spec and redistribute_schema + # exist, which indicates a reshard is needed + output_sharding.output_spec = propagation_res.output_spec + output_sharding.needs_redistribute = True + + # associate the output sharding with the output tensor metadata + new_output_spec = self._create_output_spec_with_new_tensor_meta( + op_schema.op, output_sharding.output_spec, out_tensor_meta + ) + output_sharding.output_spec = new_output_spec + + return output_sharding + else: + raise NotImplementedError( + f"Operator {op_schema.op} does not have a sharding strategy registered." + ) from decomp_exception + + def _adjust_shape_and_stride_args( + self, + out_tensor_meta: TensorMeta, + schema: OpSchema, + spec: DTensorSpec, + ) -> OpSchema: + shape_stride_idx = self.op_to_shape_and_stride_idx[schema.op] + if isinstance(shape_stride_idx, tuple): + shape_idx, stride_idx = shape_stride_idx + else: + shape_idx = shape_stride_idx + stride_idx = None + + expected_input_schema = list(schema.args_schema) + # rewrite args[shape_idx] from the global output shape to the local output shape + local_shape, _ = compute_local_shape_and_global_offset( + out_tensor_meta.shape, spec.mesh, spec.placements, skip_offset=True + ) + expected_input_schema[shape_idx] = local_shape + + # adjust the stride arg for aten.new_empty_strided.default + if stride_idx: + expected_input_schema[stride_idx] = compute_local_stride( + out_tensor_meta.stride, local_shape + ) + + return OpSchema(schema.op, tuple(expected_input_schema), schema.kwargs_schema) + + def _adjust_squeeze_to_global_singletons(self, schema: OpSchema) -> OpSchema | None: + """ + Rewrite squeeze ops to squeeze.dims with only globally-singleton dims. + Fixes bug where sharded dims with local size 1 get incorrectly squeezed. + Returns None if no rewrite is needed (already squeeze.dims with correct args). + """ + from torch.fx.experimental.symbolic_shapes import guard_or_false + + input_spec = cast(DTensorSpec, schema.args_schema[0]) + tensor_meta = input_spec.tensor_meta + if tensor_meta is None: + raise RuntimeError("squeeze requires tensor metadata") + global_shape = tensor_meta.shape + ndim = len(global_shape) + + def normalize(d: int) -> int: + return d if d >= 0 else d + ndim + + def is_singleton(d: int) -> bool: + nd = normalize(d) + return 0 <= nd < ndim and guard_or_false(global_shape[nd] == 1) + + # guard_or_false: conservatively keep dims when size is symbolic/unknown + if schema.op in (aten.squeeze.default, aten.squeeze_.default): + target_dims = tuple( + i for i, s in enumerate(global_shape) if guard_or_false(s == 1) + ) + elif schema.op in (aten.squeeze.dim, aten.squeeze_.dim): + dim = normalize(schema.args_schema[1]) # type: ignore[arg-type] + target_dims = (dim,) if is_singleton(dim) else () + else: + dims = cast(Sequence[int], schema.args_schema[1]) + target_dims = tuple( # type: ignore[union-attr] + normalize(d) for d in dims if is_singleton(d) + ) + + dims_variant = self.squeeze_op_to_dims_variant[schema.op] + # Skip rewrite if already targeting the right op with the same dims + if schema.op == dims_variant and len(schema.args_schema) > 1: + existing_dims = schema.args_schema[1] + if existing_dims == target_dims: + return None + return OpSchema(dims_variant, (input_spec, target_dims), {}) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_shards_wrapper.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_shards_wrapper.py new file mode 100644 index 0000000000000000000000000000000000000000..2ef4f1d8fdb21bdd1837ce50d107f05e30dc1d27 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_shards_wrapper.py @@ -0,0 +1,362 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from typing import Any + +import torch +from torch.distributed.checkpoint.metadata import ( + ChunkStorageMetadata, + MetadataIndex, + TensorProperties, + TensorStorageMetadata, +) +from torch.distributed.checkpoint.planner import ( + TensorWriteData, + WriteItem, + WriteItemType, +) + + +aten = torch.ops.aten + + +class LocalShardsWrapper(torch.Tensor): + """ + A wrapper class to hold local shards of a DTensor. + This class is used largely for checkpointing purposes and implicitly subtypes + the _Checkpointable protocol. + """ + + __slots__ = ["_local_shards", "_storage_meta"] + _local_shards: list[torch.Tensor] + _storage_meta: TensorStorageMetadata + + @staticmethod + def __new__( + cls, local_shards: list[torch.Tensor], local_offsets: list[tuple[int, ...]] + ) -> "LocalShardsWrapper": + if not all( + tensor.device == local_shards[0].device for tensor in local_shards[1:] + ): + raise AssertionError + + # if empty shard, we create an empty tensor + if len(local_shards) == 0: + r = torch.Tensor._make_wrapper_subclass( + cls, + torch.Size([0, 0]), + ) + r._local_shards = [] + r._storage_meta = TensorStorageMetadata( + properties=TensorProperties(), + size=torch.Size([0, 0]), + chunks=[ + ChunkStorageMetadata( + offsets=torch.Size([0, 0]), sizes=torch.Size([0, 0]) + ) + ], + ) + return r + + # we calculate the total tensor size by "concat" on second tensor dimension + cat_tensor_shape = list(local_shards[0].size()) + if len(local_shards) > 1 and local_shards[0].ndim == 2: # column-wise sharding + for shard in local_shards[1:]: + cat_tensor_shape[1] += shard.size()[1] + + # in cases of sharding optimizer rowwise, we calculate total tensor size by "concat" on first tensor dimension + if len(local_shards) > 1 and local_shards[0].ndim == 1: # column-wise sharding + for shard in local_shards[1:]: + cat_tensor_shape[0] += shard.size()[0] + + wrapper_properties = TensorProperties.create_from_tensor(local_shards[0]) + wrapper_shape = torch.Size(cat_tensor_shape) + chunks_meta = [ + ChunkStorageMetadata( + offsets=torch.Size(offset), + sizes=shard.size(), + ) + for shard, offset in zip(local_shards, local_offsets) + ] + + r = torch.Tensor._make_wrapper_subclass( + cls, + torch.Size(cat_tensor_shape), + ) + r._local_shards = local_shards + r._storage_meta = TensorStorageMetadata( + properties=wrapper_properties, + size=wrapper_shape, + chunks=chunks_meta, + ) + + return r + + # necessary for ops dispatching from this subclass to its local shards + @classmethod + def __torch_dispatch__(cls, func, types, args=(), kwargs=None): # type: ignore[override] + kwargs = kwargs or {} + + dispatcher = { + torch.ops._c10d_functional.all_gather_into_tensor.default: cls.handle_all_gather_into_tensor, + torch.ops._c10d_functional.wait_tensor.default: cls.handle_wait_tensor, + aten._to_copy.default: cls.handle_to_copy, + aten.view.default: cls.handle_view, + aten.equal.default: cls.handle_equal, + aten.detach.default: cls.handle_detach, + aten.clone.default: cls.handle_clone, + aten.new_empty.default: cls.handle_new_empty, + } + + if func in dispatcher: + return dispatcher[func](args, kwargs) + else: + raise NotImplementedError( + f"{func} is not supported for LocalShardsWrapper!" + ) + + @staticmethod + def handle_all_gather_into_tensor(args, kwargs) -> torch.Tensor: + dim = args[0].local_sizes()[0][1] + cat_tensor = torch.cat( + [t.view(-1) for t in args[0].local_shards()], dim=0 + ).view(-1, dim) + return torch.ops._c10d_functional.all_gather_into_tensor.default( + cat_tensor, *args[1:], **kwargs + ) + + @staticmethod + def handle_wait_tensor(args, kwargs) -> torch.Tensor: + return torch.ops._c10d_functional.wait_tensor(args[0]) + + @staticmethod + def handle_to_copy(args, kwargs) -> torch.Tensor: + res_shards_list = [ + aten._to_copy.default(shard, *args[1:], **kwargs) + for shard in args[0].local_shards() + ] + return LocalShardsWrapper(res_shards_list, args[0].local_offsets()) + + @staticmethod + def handle_view(args, kwargs) -> "LocalShardsWrapper": + view_shape = args[1] + res_shards_list = [] + if len(args[0].local_shards()) > 1: + if args[0].local_shards()[0].ndim == 2: + if not ( + args[0].storage_metadata().size[0] == view_shape[0] + and args[0].storage_metadata().size[1] == view_shape[1] + ): + raise AssertionError + # This accounts for a DTensor quirk, when multiple shards are present on a rank, DTensor on + # init calls view_as() on the global tensor shape + # will fail because the view shape is not applicable to individual shards. + res_shards_list = [ + aten.view.default(shard, shard.shape, **kwargs) + for shard in args[0].local_shards() + ] + elif args[0].local_shards()[0].ndim == 1: + if args[0].storage_metadata().size[0] != view_shape[0]: + raise AssertionError + # This case is for optimizer sharding as regardless of sharding type, optimizer state is row wise sharded + res_shards_list = [ + aten.view.default(shard, shard.shape, **kwargs) + for shard in args[0].local_shards() + ] + else: + raise NotImplementedError("No support for view on tensors ndim > 2") + else: + # view is called per shard + res_shards_list = [ + aten.view.default(shard, args[1], **kwargs) + for shard in args[0].local_shards() + ] + return LocalShardsWrapper(res_shards_list, args[0].local_offsets()) + + @staticmethod + def handle_equal(args, kwargs) -> bool: + """ + LocalShardsWrapper equal impl also checks for equality of storage metadata + and the order of shards + """ + a, b = args[0], args[1] + if len(a.local_shards()) != len(b.local_shards()): + return False + if not all( + aten.equal.default(x, y) for x, y in zip(a.local_shards(), b.local_shards()) + ): + return False + if a.storage_metadata() != b.storage_metadata(): + return False + return True + + @staticmethod + def handle_detach(args, kwargs) -> "LocalShardsWrapper": + self_ls = args[0] + deatched_local_shards = [ + aten.detach.default(shard) for shard in self_ls.local_shards() + ] + self_ls._local_shards = deatched_local_shards + self_ls._storage_meta.properties.requires_grad = False + return self_ls + + @staticmethod + def handle_clone(args, kwargs) -> "LocalShardsWrapper": + self_ls = args[0] + desired_memory_format = kwargs.get("memory_format", None) + if desired_memory_format and desired_memory_format != torch.preserve_format: + raise NotImplementedError( + f"{desired_memory_format} is not supported for LocalShardsWrapper!" + ) + cloned_local_shards = [ + shard.clone(memory_format=desired_memory_format) + for shard in self_ls._local_shards + ] + return LocalShardsWrapper(cloned_local_shards, self_ls.local_offsets()) + + @staticmethod + def handle_new_empty(args, kwargs) -> "LocalShardsWrapper": + self_ls = args[0] + return LocalShardsWrapper( + [torch.empty_like(shard) for shard in self_ls._local_shards], + self_ls.local_offsets(), + ) + + @property + def device(self) -> torch._C.device: # type: ignore[override] + return ( + self._local_shards[0].device if self._local_shards else torch.device("meta") + ) + + @property + def is_meta(self) -> bool: # type: ignore[override] + return self._local_shards[0].is_meta if self._local_shards else True + + def is_pinned(self) -> bool: # type: ignore[override] + return self._storage_meta.properties.pin_memory + + def requires_grad_(self, requires_grad: bool = True) -> "LocalShardsWrapper": + self._storage_meta.properties.requires_grad = requires_grad + [shard.requires_grad_(requires_grad) for shard in self._local_shards] + return self + + def local_shards(self) -> list[torch.Tensor]: + """ + Returns a list of :class:`torch.Tensor' corresponding to the + local shards for this rank. Returns an empty list if the current rank + does not host any shards for this Tensor. + """ + return self._local_shards + + def local_sizes(self) -> list[torch.Size]: + """ + Returns a list of :class:`torch.Size' corresponding to the + local sizes for the shards on this rank. Returns an empty list if the current rank + does not host any shards for this Tensor. + """ + return [chunk.sizes for chunk in self._storage_meta.chunks] + + def local_offsets(self) -> list[torch.Size]: + """ + Returns a list of :class:`torch.Size' corresponding to the + local offsets for the shards on this rank. Returns an empty list if the current rank + does not host any shards for this Tensor. + """ + return [chunk.offsets for chunk in self._storage_meta.chunks] + + @property + def local_chunks(self) -> list[ChunkStorageMetadata]: + """ + Returns a :class:`list[ChunkStorageMetadata]` object corresponding to the + metadata for each tensor shard + """ + return self._storage_meta.chunks + + def storage_metadata(self) -> TensorStorageMetadata: + """ + Returns a :class:`TensorStorageMetadata` object corresponding to the + metadata for the local tensor on current rank + """ + return self._storage_meta + + def is_empty_shard(self) -> bool: + """ + Returns a :class:`bool` object indicating if the local tensor on current rank + is an empty tensor + """ + return self._storage_meta.size[0] == 0 and self._storage_meta.size[1] == 0 + + def __create_write_items__(self, fqn: str, object: Any) -> list[WriteItem]: + """ + For compatibility with DCP, we support creation of WriteItems + such that they can be saved properly. + """ + return [ + WriteItem( + index=MetadataIndex(fqn, chunks.offsets), + type=WriteItemType.SHARD, + tensor_data=TensorWriteData( + chunk=ChunkStorageMetadata( + offsets=chunks.offsets, + sizes=chunks.sizes, + ), + properties=self._storage_meta.properties, + size=object.size(), + ), + ) + for tensor, chunks in zip(self.local_shards(), self.local_chunks) + ] + + def __create_chunk_list__(self) -> list[ChunkStorageMetadata]: + """ + For compatibility with DCP, we support creation of chunk lists + such that they can be saved properly. + """ + return self._storage_meta.chunks + + def __get_tensor_shard__(self, index: MetadataIndex) -> torch.Tensor: + """ + For compatibility with DCP, we support finding shard based on index + Return a 'torch.Tensor' shard based on 'MetadataIndex'. + """ + # Fast lookup path + if index.index is not None: + if ( + len(self._local_shards) > index.index + and self._storage_meta.chunks[index.index].offsets == index.offset + ): + return self._local_shards[index.index] + + if index.offset is not None: + for shard, chunk in zip(self._local_shards, self._storage_meta.chunks): + if chunk.offsets == index.offset: + return shard + + # Empty shard case + if len(self._local_shards) == 0 and self._storage_meta.chunks[ + 0 + ].sizes == torch.Size([0, 0]): + return torch.empty(0) + + raise ValueError( + f"Could not find shard at '{index.offset}' for FQN: '{index.fqn}'" + ) + + def _get_tensor_size_bytes(self) -> int: + object_size = 0 + for shard in self.local_shards(): + object_size += shard.nelement() * shard.element_size() + return object_size + + def __hash__(self) -> int: + return id(self) + + def __repr__(self) -> str: # type: ignore[override] + return f"LocalShardsWrapper:{self._local_shards} {self._storage_meta}" + + def __str__(self) -> str: + return f"LocalShardsWrapper:{self._local_shards} {self._storage_meta}" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_tp_conv.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_tp_conv.py new file mode 100644 index 0000000000000000000000000000000000000000..6ec876baae55f8728c8f1ff2654859d51867366c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_tp_conv.py @@ -0,0 +1,314 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +# implement matrix related ops for distributed tensor +from typing import cast + +import torch +import torch.distributed as dist +import torch.distributed.tensor._api as dtensor + + +aten = torch.ops.aten + + +def _requires_data_exchange(padding, dim_map) -> bool: + # Data exchange is not need if only sharded across batch dim + if all(x == -1 for x in dim_map[1:]): + return False + # TODO: whether there requires data exchange is currently determined by padding + return padding[-1] != 0 + + +def _is_supported(input_size, kernel_size, stride, padding, dilation): + if dilation[-1] != 1: + raise RuntimeError("Dilation must be 1 for tensor parallel convolution.") + if padding[-1] != 0: + if stride[-1] != 1: + raise RuntimeError( + "Stride must be 1 when there is padding for tensor parallel convolution." + ) + if kernel_size[-1] // 2 > input_size[-1]: + raise RuntimeError( + "kernel_size[-1] // 2 should be less than or equal to input_size[-1] for tensor parallel convolution." + ) + else: + if not (input_size[-1] % stride[-1] == 0 and stride[-1] == kernel_size[-1]): + raise RuntimeError( + "It requires that input_size[-1] is divisible by stride[-1] and stride[-1] equals kernel_size[-1] " + "when there is padding for tensor parallel convolution." + ) + return True + + +def _ring_send_recv_construct(in_tensor, d1, d2, left, right, rank, size): + # dist comms and reconstruct local input tensor + send_to_right = in_tensor[..., -d1:].contiguous() + send_to_left = in_tensor[..., :d2].contiguous() + recv_from_right = torch.zeros_like(send_to_left) + recv_from_left = torch.zeros_like(send_to_right) + + send_op_right = dist.P2POp(dist.isend, send_to_right, right) + send_op_left = dist.P2POp(dist.isend, send_to_left, left) + recv_op_right = dist.P2POp(dist.irecv, recv_from_right, right) + recv_op_left = dist.P2POp(dist.irecv, recv_from_left, left) + + reqs = dist.batch_isend_irecv( + [send_op_right, send_op_left, recv_op_left, recv_op_right] + ) + for req in reqs: + req.wait() + + if rank == 0: + in_tensor = torch.cat([in_tensor, recv_from_right], dim=-1) + elif rank == size - 1: + in_tensor = torch.cat([recv_from_left, in_tensor], dim=-1) + else: + in_tensor = torch.cat([recv_from_left, in_tensor, recv_from_right], dim=-1) + + return in_tensor + + +def _ring_send_recv_aggregate(grad_in_tensor, d1, d2, left, right, rank, size): + # dist comms and aggregate gradients for edge pixels + send_to_right = grad_in_tensor[:, :, :, -d2:].contiguous() + send_to_left = grad_in_tensor[:, :, :, :d1].contiguous() + recv_from_right = torch.zeros_like(send_to_left) + recv_from_left = torch.zeros_like(send_to_right) + + send_op_right = dist.P2POp(dist.isend, send_to_right, right) + send_op_left = dist.P2POp(dist.isend, send_to_left, left) + recv_op_right = dist.P2POp(dist.irecv, recv_from_right, right) + recv_op_left = dist.P2POp(dist.irecv, recv_from_left, left) + + reqs = dist.batch_isend_irecv( + [send_op_right, send_op_left, recv_op_left, recv_op_right] + ) + for req in reqs: + req.wait() + + if rank == 0: + grad_in_tensor = grad_in_tensor[:, :, :, :-d2] + grad_in_tensor[:, :, :, -d1:] = torch.add( + grad_in_tensor[:, :, :, -d1:], recv_from_right + ) + elif rank == size - 1: + grad_in_tensor = grad_in_tensor[:, :, :, d1:] + grad_in_tensor[:, :, :, :d2] = torch.add( + grad_in_tensor[:, :, :, :d2], recv_from_left + ) + else: + grad_in_tensor = grad_in_tensor[:, :, :, d1:-d2] + grad_in_tensor[:, :, :, -d1:] = torch.add( + grad_in_tensor[:, :, :, -d1:], recv_from_right + ) + grad_in_tensor[:, :, :, :d2] = torch.add( + grad_in_tensor[:, :, :, :d2], recv_from_left + ) + + +def tp_convolution( + op_call: torch._ops.OpOverload, + local_tensor_args: tuple[object, ...], + local_tensor_kwargs: dict[str, object], + dim_map: list[int], +) -> object: + if op_call != aten.convolution.default: + raise AssertionError + if len(local_tensor_args) != 9: + raise AssertionError + + rank = dist.get_rank() + size = dist.get_world_size() + in_tensor = cast(torch.Tensor, local_tensor_args[0]) + weight = cast(torch.Tensor, local_tensor_args[1]) + stride, padding, dilation = local_tensor_args[3:6] + + if not isinstance(padding, list): + raise AssertionError + + if not _requires_data_exchange(padding, dim_map): + local_results = op_call(*local_tensor_args, **local_tensor_kwargs) + return local_results + else: + if not _is_supported(in_tensor.shape, weight.shape, stride, padding, dilation): + raise AssertionError( + "tp_convolution data exchange requires supported stride/padding/dilation" + ) + # step 0 compute the overlap pixels of the input tensor + d = weight.shape[-1] - 1 + d1 = d // 2 + d2 = d - d1 + if d1 + d2 != d: + raise AssertionError + right = (rank + 1) % size + left = (rank - 1 + size) % size + + # step1 reconstruct local input tensor + in_tensor = _ring_send_recv_construct( + in_tensor, d1, d2, left, right, rank, size + ) + + # step2 feed local input tensor to op_call + local_tensor_args_list = list(local_tensor_args) + local_tensor_args_list[0] = in_tensor + local_tensor_args = cast(tuple[object, ...], local_tensor_args_list) + local_results = op_call(*local_tensor_args, **local_tensor_kwargs) + + # step3 remove extra outputs from the results + padding_w = padding[-1] + w = local_results.size(-1) + if rank == 0: + local_results = local_results[..., : w - padding_w] + elif rank == size - 1: + local_results = local_results[..., padding_w:] + else: + local_results = local_results[..., padding_w : w - padding_w] + + return local_results + + +def tp_convolution_backward( + op_call: torch._ops.OpOverload, + local_tensor_args: tuple[object, ...], + local_tensor_kwargs: dict[str, object], + dim_map: list[int], +) -> object: + if op_call != aten.convolution_backward.default: + raise AssertionError + if len(local_tensor_args) != 11: + raise AssertionError + + rank = dist.get_rank() + size = dist.get_world_size() + grad_out_tensor = cast(torch.Tensor, local_tensor_args[0]) + in_tensor = cast(torch.Tensor, local_tensor_args[1]) + weight = cast(torch.Tensor, local_tensor_args[2]) + stride, padding, dilation = local_tensor_args[4:7] + + if not isinstance(padding, list): + raise AssertionError + + if not _requires_data_exchange(padding, dim_map): + local_results = op_call(*local_tensor_args, **local_tensor_kwargs) + return local_results + else: + if not _is_supported(in_tensor.shape, weight.shape, stride, padding, dilation): + raise AssertionError( + "tp_convolution_backward data exchange requires supported stride/padding/dilation" + ) + # step 0 compute the overlap pixels of the input tensor + d = weight.shape[3] - 1 + d1 = d // 2 + d2 = d - d1 + if d1 + d2 != d: + raise AssertionError + right = (rank + 1) % size + left = (rank - 1 + size) % size + + # step1 reconstruct local input tensor + in_tensor = _ring_send_recv_construct( + in_tensor, d1, d2, left, right, rank, size + ) + + # step2 reconstruct local gradient output tensor + padding_w = padding[1] + if rank == 0: + grad_out_tensor = torch.nn.functional.pad( + grad_out_tensor, (0, padding_w), "constant", 0 + ) + elif rank == size - 1: + grad_out_tensor = torch.nn.functional.pad( + grad_out_tensor, (padding_w, 0), "constant", 0 + ) + else: + grad_out_tensor = torch.nn.functional.pad( + grad_out_tensor, (padding_w, padding_w), "constant", 0 + ) + + # step3 feed local input tensor to op_call + local_tensor_args_list = list(local_tensor_args) + local_tensor_args_list[0] = grad_out_tensor + local_tensor_args_list[1] = in_tensor + local_tensor_args = cast(tuple[object, ...], local_tensor_args_list) + local_results = op_call(*local_tensor_args, **local_tensor_kwargs) + + # step4 aggregate gradients for edge pixels + grad_in_tensor = local_results[0] + if grad_in_tensor is not None: + grad_in_tensor = _ring_send_recv_aggregate( + grad_in_tensor, d1, d2, left, right, rank, size + ) + local_results = list(local_results) + local_results[0] = grad_in_tensor + + local_results = cast(tuple[object, ...], local_results) + + return local_results + + +def convolution_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + # extract local tensor and sharding infos to an OpInfo + op_info = dtensor.DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + + # sharding propagation + dtensor.DTensor._op_dispatcher.sharding_propagator.propagate(op_info) + output_sharding = op_info.output_sharding + if output_sharding is None: + raise AssertionError("output sharding should not be None") + output_spec = output_sharding.output_spec + if not isinstance(output_spec, dtensor.DTensorSpec): + raise AssertionError + + # local propagation + local_results = tp_convolution( + op_call, + tuple(op_info.local_args), + op_info.local_kwargs, + output_spec.dim_map, + ) + + return dtensor.DTensor._op_dispatcher.wrap(local_results, output_spec) + + +def convolution_backward_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + # Redistribute grad_output tensor to the same placement as input tensor + # pyrefly: ignore [bad-assignment] + args = list(args) + if not ( + isinstance(args[0], dtensor.DTensor) and isinstance(args[1], dtensor.DTensor) + ): + raise AssertionError + # pyrefly: ignore [unsupported-operation] + args[0] = args[0].redistribute(args[1].device_mesh, args[1].placements) + args = tuple(args) + + # extract local tensor and sharding infos to an OpInfo + op_info = dtensor.DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + + # sharding propagation + dtensor.DTensor._op_dispatcher.sharding_propagator.propagate(op_info) + output_sharding = op_info.output_sharding + if output_sharding is None: + raise AssertionError("output sharding should not be None") + if not isinstance(op_info.flat_args_schema[0], dtensor.DTensorSpec): + raise AssertionError + + # local propagation + local_results = tp_convolution_backward( + op_call, + tuple(op_info.local_args), + op_info.local_kwargs, + op_info.flat_args_schema[0].dim_map, + ) + + return dtensor.DTensor._op_dispatcher.wrap( + local_results, output_sharding.output_spec + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..fb05f685e20cfb7b6ea82cd789c693ddaa477d55 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/_utils.py @@ -0,0 +1,532 @@ +import logging +import threading +from collections.abc import Callable, Sequence +from typing import Any + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.tensor._api as dtensor +from torch._logging import LazyString +from torch._prims_common import ShapeType +from torch.distributed import RankType +from torch.distributed._local_tensor import maybe_run_for_local_tensor +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._collective_utils import redistribute_cost +from torch.distributed.tensor._dtensor_spec import DTensorSpec +from torch.distributed.tensor._op_schema import OpSchema +from torch.distributed.tensor.placement_types import ( + _is_shard_like, + _StridedShard, + _StridedShardOffsetMode, + Partial, + Placement, + Replicate, + Shard, +) + + +logger = logging.getLogger(__name__) + + +def _format_implicit_redistribution_msg(schema: OpSchema) -> str: + return f"Implicit redistribution occurred for {schema} while ExplicitRedistributionContext was active" + + +class ExplicitRedistributionContext: + """ + Within this context manager, DTensor will refuse to perform implicit redistribution, + instead raising an error. Manual calls to ``redistribute()`` are required wherever a redistribution + must occur to avoid erroring. This can be used to ensure that the user is aware of all redistribution. + + Note: it is easier to use this mode on just the forward pass of a typical DTensor program, as the backwards pass + may contain implicit redistribution calls that are not visible to the user and difficult to replace with manual + calls. Redistribution during backward can be made explicit by writing `autograd.Function`s that are no-op + during forward and perform a manual redistribution during backwards. + + enable (bool) if False, disables the context manager. Can be used nested inside an enabled region. + + strict (bool) if True, triggers on any redistribution. If False, only triggers on redistributions that perform + communication. + + mode (str) Determines what happens when ExplicitRedistributionContext triggers: + "raise": raises an exception, "warn" issues a warning + """ + + _local = threading.local() + + def __init__(self, enable: bool = True, strict: bool = False, mode="raise"): + self._enable = enable + self._strict = strict + if mode not in ("raise", "warn"): + raise RuntimeError(f"Invalid mode {mode}") + self._raise_on_redistribution = mode == "raise" + + @classmethod + def observe_redistribution( + cls, + src_spec: DTensorSpec, + dst_spec: DTensorSpec, + redistribution_msg: LazyString, + ): + if instance := getattr(cls._local, "_active", None): + allowed = True + if instance._enable: + if instance._strict: + allowed = False + else: + allowed = redistribute_cost(src_spec, dst_spec) <= 0 + if not allowed: + if instance._raise_on_redistribution: + raise RuntimeError(redistribution_msg) + else: + logger.warning(redistribution_msg) + + def __enter__(self): + self._prev = getattr(ExplicitRedistributionContext._local, "_active", None) + ExplicitRedistributionContext._local._active = self + return self + + def __exit__(self, exc_type, exc_val, exc_tb): + ExplicitRedistributionContext._local._active = self._prev + + +def compute_local_shape_and_global_offset( + global_shape: ShapeType, + mesh: DeviceMesh, + placements: Sequence[Placement], + skip_offset: bool = False, +) -> tuple[tuple[int, ...], tuple[int, ...]]: + """ + Compute the local tensor shape and the global offsets into the original tensor + of a DTensor on its current global rank. This is useful for checkpointing purpose. + + Example: + global_tensor = [[0, 1, 2, 3, 4], sharded on mesh (DP=2, TP=2) with (Shard(1), Shard(1)) + [10, 11, 12, 13, 14]] + + This table shows the return value of local_shape and global_offset for each rank. + (`local_tensor` is for illustration only). + + Note how the first coordinate of global_offset is always 0, corresponding to tensor dim 0 being replicated. + + Rank local_tensor local_shape global_offset + ------------------------------------------------------------- + 0 [[0, 1], (2, 2) (0, 0) + [10, 11]] + + 1 [[2], (2, 1) (0, 2) + [12]] + + 2 [[3], (2, 1) (0, 3) + [13]] + + 3 [[4], (2, 1) (0, 4) + [14]] + + Args: + global_shape (ShapeType): The global shape of the DTensor. + mesh (:class:`DeviceMesh`): The device mesh this DTensor is distributed on. + placements (Sequence[:class:`Placement`]]): The placements of the DTensor. + skip_offset (bool): If True, skip computing the global offsets and return an empty + tuple for global_offset. This can improve performance when only the local shape + is needed. Defaults to False. + + Return: + local_shape: the shape of the DTensor's _local_tensor on the current rank. + global_offset: a tuple of offsets for each dimension of the global tensor shape, + identifying how this shard fits into the global tensor in each dimension. If + skip_offset is True, this will be an empty tuple. + + """ + empty_offset = () + if not mesh._is_current_rank_part_of_mesh(): + # if rank not in the mesh, return empty offset + return ((0,), empty_offset) + + return _compute_local_shape_and_global_offset( + global_shape, mesh.shape, mesh._sym_get_coordinate, placements, skip_offset + ) + + +@maybe_run_for_local_tensor +def _get_shard_size_and_offsets( + curr_local_size: int, + mesh_dim_size: int, + rank: RankType, + placement: Shard | _StridedShard, + previous_offsets, + zero_global_offset: int, + skip_offset: bool, +) -> tuple[int, torch.Tensor | None]: + kwargs: dict[str, Any] = { + "curr_local_size": curr_local_size, + "num_chunks": mesh_dim_size, + "rank": rank, + } + if isinstance(placement, _StridedShard): + kwargs["offset_mode"] = ( + _StridedShardOffsetMode.NONE if skip_offset else _StridedShardOffsetMode.ALL + ) + # _StridedShard.local_shard_size_and_offset materializes the offsets list + # via .tolist() on a (potentially fake) index tensor; under FakeTensorMode + # that allocates one unbacked SymInt per element. Skip when the caller + # discards the offsets anyway. + shard_size, shard_offsets = placement._local_shard_size_and_offset(**kwargs) + if skip_offset: + return shard_size, None + if shard_size == 0: + return shard_size, torch.arange(zero_global_offset, zero_global_offset + 1) + if isinstance(placement, Shard) and not isinstance(placement, _StridedShard): + if not isinstance(shard_offsets, int): + raise AssertionError + index = torch.arange(shard_offsets, shard_offsets + shard_size) + else: + if not isinstance(shard_offsets, list): + raise AssertionError + index = torch.tensor(shard_offsets) + if previous_offsets is None: + return shard_size, index + else: + return shard_size, previous_offsets[index] + + +@maybe_run_for_local_tensor +def _get_first_offset(offsets: torch.Tensor) -> int: + return int(offsets[0]) + + +# accept 'plain data types' to enable simpler unit testing without creating device mesh +def _compute_local_shape_and_global_offset( + global_shape: ShapeType, + mesh_shape: ShapeType, + my_coordinate: list[int] | Callable[[int], RankType] | None, + placements: Sequence[Placement], + skip_offset: bool = False, +) -> tuple[tuple[int, ...], tuple[int, ...]]: + """ + Suppose you have a full tensor with size global_shape, and you have sharded + it according to placements for mesh_shape. This function returns, for a + specific coordinate my_coordinate in the device mesh: + + - The size of your local shard WITHOUT padding (i.e., if you have + an uneven split, your size might be smaller than the other entries + in your dim), and + + - Where the data for your shard begins, in the full tensor. + + This function is fairly simple if your tensor is evenly sharded; the complication + is around uneven splits. There is also some complication for handling StridedShard, + which changes the order you should apply sharding. + + Args: + global_shape (ShapeType): The global shape of the tensor. + mesh_shape (ShapeType): The shape of the device mesh. + my_coordinate (Optional[list[int]]): The coordinate of the current rank in the device mesh. + placements (Sequence[Placement]): The placements of the DTensor. + skip_offset (bool): If True, skip computing the global offsets and return an empty + tuple for global_offset. This can improve performance when only the local shape + is needed. Defaults to False. + + Returns: + tuple: A tuple containing: + - local_shape (tuple[int, ...]): The shape of the local shard on the current rank. + - global_offset (tuple[int, ...]): The offsets for each dimension identifying where + this shard begins in the global tensor. If skip_offset is True, this will be an + empty tuple. + """ + + if isinstance(my_coordinate, (list, tuple)): + _coord: list | tuple = my_coordinate + + def coordinate_lookup(dim: int) -> RankType: + return _coord[dim] + else: + if my_coordinate is None: + raise AssertionError + coordinate_lookup = my_coordinate + + local_shape = list(global_shape) + # Perform shard from left to right. For example, + # global tensor: [0, 1, 2, 3, 4, 5, 6, 7] + # placements: S(0), SS(0, split_factor=2) + # mesh_shape: (2, 2) + # After S(0), shard_dim_to_global_offsets are + # {0: [0, 1, 2, 3]} on my_coordinate [0, 0] [0, 1] + # {0: [4, 5, 6, 7]} on my_coordinate [1, 0] [1, 1] + # After SS(0, split_factor=2), shard_dim_to_global_offsets are + # {0: [0, 2]} on my_coordinate [0, 0] + # {0: [1, 3]} on my_coordinate [0, 1] + # {0: [4, 6]} on my_coordinate [1, 0] + # {0: [5, 7]} on my_coordinate [1, 1] + shard_dim_to_global_offsets = {} + for mesh_dim, placement in enumerate(placements): + if not isinstance(placement, (Shard, _StridedShard)): + continue + shard_dim = placement.dim + zero_global_offset = global_shape[shard_dim] + if shard_dim >= len(local_shape): + raise AssertionError( + f"Sharding dim {shard_dim} greater than tensor ndim {len(local_shape)}" + ) + previous_offsets = shard_dim_to_global_offsets.get(shard_dim) + shard_size, shard_offsets = _get_shard_size_and_offsets( + local_shape[shard_dim], + mesh_shape[mesh_dim], + coordinate_lookup(mesh_dim), + placement, + previous_offsets, + zero_global_offset, + skip_offset, + ) + local_shape[shard_dim] = shard_size + shard_dim_to_global_offsets[shard_dim] = shard_offsets + if skip_offset: + return tuple(local_shape), () + global_offset = [0] * len(global_shape) + for shard_dim, global_offsets in shard_dim_to_global_offsets.items(): + global_offset[shard_dim] = _get_first_offset(global_offsets) + return tuple(local_shape), tuple(global_offset) + + +compute_global_tensor_info = torch._C._DTensor_compute_global_tensor_info + + +def compute_local_tensor_info( + global_tensor: torch.Tensor, + mesh: DeviceMesh, + placements: Sequence[Placement], +) -> tuple[list[int], list[int]]: + """ + Compute the local size and stride of a DTensor from the given global tensor info. + + For example, if we have a global tensor with size (4, 8, 4) and stride (32, 1, 8). + If the DTensor placements are [Shard(2)] and world_size is 2; + then the local size is (4, 8, 2) and stride is (16, 1, 8). + + Args: + tensor (:class:`torch.Tensor`): + Global tensor which DTensor will distribute + mesh (:class:`DeviceMesh`): + Object which describes the mesh topology + of devices for the DTensor. + placements (Sequence[:class:`Placement`]): + The attribute of the DTensor that describes its layout + on the mesh topology. + + Returns: + local_shape: A List of int which specifies the size of the local tensor. + local_stride: A List of int which specifies the stride of the local tensor. + """ + local_shape = list(global_tensor.size()) + local_stride = list(global_tensor.stride()) + + for idx, placement in enumerate(placements): + mesh_dim_size = mesh.size(idx) + if _is_shard_like(placement): + if placement.dim < 0: + raise AssertionError( + "Shard placements should have negative dims normalized in " + f"the user-facing APIs: {placement}" + ) + shard_dim = placement.dim + if shard_dim >= len(local_shape): + raise AssertionError( + f"Sharding dim {shard_dim} greater than tensor ndim {len(local_shape)} " + f"for placement number {idx}." + ) + + global_dim_size = local_shape[shard_dim] + if global_dim_size % mesh_dim_size != 0: + raise AssertionError( + f"Global dim {global_dim_size} not divisible by mesh size {mesh_dim_size}" + ) + local_shape[shard_dim] = global_dim_size // mesh_dim_size + + # shrink strides that were scaled up globally + for i in range(len(local_stride)): + if ( + i != shard_dim + and local_stride[i] >= local_stride[shard_dim] * mesh_dim_size + ): + local_stride[i] = local_stride[i] // mesh_dim_size + + elif not isinstance(placement, (Replicate, Partial)): + raise RuntimeError(f"placement type {type(placement)} not supported!") + + return local_shape, local_stride + + +def compute_global_tensor_shape( + shape: torch.Size, mesh: DeviceMesh, placements: Sequence[Placement] +) -> torch.Size: + """ + Compute the global size of a DTensor from the given local tensor shape, + the mesh and placements. Different from `compute_global_tensor_info`, + which assumes sharding is even, this util allgathers local shards' shapes + from all ranks and thus can support uneven sharding. + NOTE: Currently this function only supports 1D mesh. + + Args: + shape (:class:`torch.Size`): + Shape of the local tensor + mesh (:class:`DeviceMesh`): + Object which describes the mesh topology + of devices for the DTensor. + placements (Sequence[:class:`Placement`]]): + The attribute of the DTensor that describes its layout + on the mesh topology. + + Return: + tensor_shape: Shape of the global DTensor. + """ + if len(placements) != 1: + raise NotImplementedError( + "compute_global_tensor_shape only supports 1 placement for now." + ) + + if len(placements) != mesh.ndim: + raise RuntimeError( + "Expected one placement per mesh dim, " + f"but found {len(placements)} placements and {mesh.ndim} mesh dims." + ) + + if isinstance(placements[0], Replicate): + return shape + # NOTE: isinstance(_, Shard) does not match _StridedShard; see _is_shard_like(). + elif isinstance(placements[0], Shard): + + @maybe_run_for_local_tensor + def _create_local_shape_tensor(shape): + return torch.tensor(list(shape), device=mesh.device_type) + + local_shape = _create_local_shape_tensor(shape) + gathered_shaped_tensors = [ + torch.empty_like(local_shape, device=local_shape.device) + for _ in range(mesh.size()) + ] + funcol.all_gather_inplace(gathered_shaped_tensors, local_shape, mesh) + + @maybe_run_for_local_tensor + def _validate_and_compute_global_shape(local_shape, gathered_shaped_tensors): + sharded_dim_sum = 0 + shard_dim = placements[0].dim # type: ignore[union-attr] + other_dims = [d for d in range(len(shape)) if d != shard_dim] + for shape_tensor in gathered_shaped_tensors: + if not torch.equal(local_shape[other_dims], shape_tensor[other_dims]): + raise RuntimeError( + "Non-sharded dimensions should have identical size across ranks." + ) + shape_tensor_list = shape_tensor.tolist() + sharded_dim_sum += shape_tensor_list[shard_dim] + return sharded_dim_sum + + sharded_dim_sum = _validate_and_compute_global_shape( + local_shape, gathered_shaped_tensors + ) + global_shape = list(shape) + global_shape[placements[0].dim] = sharded_dim_sum + return torch.Size(global_shape) + else: + raise NotImplementedError( + f"Placement type {type(placements[0])} not supported." + ) + + +def try_find_mesh_from_args( + op_call: torch._ops.OpOverload, args: Sequence[object] +) -> DeviceMesh: + """ + Find the device mesh object from args. + It returns None if no mesh is found. + NOTE: we can optimize this search if needed + """ + for arg in args: + if isinstance(arg, (dtensor.DTensor, DTensorSpec)): + return arg.device_mesh + elif ( + isinstance(arg, (list, tuple)) + and len(arg) > 0 + and isinstance(arg[0], (dtensor.DTensor, DTensorSpec)) + ): + return arg[0].device_mesh + + raise ValueError(f"Cannot find device mesh from args for op : {op_call}.") + + +def compute_local_stride( + global_stride: ShapeType, local_shape: ShapeType +) -> tuple[int, ...]: + """ + Compute the stride of a local tensor shard, given the global stride and local shape. + + Derives strides by preserving the memory layout (dimension ordering) implied + by the global strides, then computing contiguous strides for the local shape + in that order. Assumes the global tensor is non-overlapping and dense. + """ + ndim = len(global_stride) + # Sort dims by global stride descending to recover memory layout order. + # Stable sort preserves original dim order for ties, which only occur + # on size-1 dims where the stride value is semantically irrelevant. + perm = sorted(range(ndim), key=lambda d: global_stride[d], reverse=True) + local_strides = [0] * ndim + s = 1 + for d in reversed(perm): + local_strides[d] = s + s *= local_shape[d] + return tuple(local_strides) + + +def normalize_to_torch_size(size) -> torch.Size: # type: ignore[no-untyped-def] + """ + Unify variable types of size argument to torch.Size + Acceptable types include: + int, Sequence[int], Tuple[int], Tuple[Sequence[int]], + or torch.Size + """ + if isinstance(size, torch.Size): + return size + + if isinstance(size, int): + torch_size = [size] + elif len(size) == 1 and isinstance(size[0], Sequence): + torch_size = list(size[0]) + else: + torch_size = list(size) + return torch.Size(torch_size) + + +def assert_no_mixed_partial_types(placements: Sequence[Placement]) -> None: + """ + Assert that a placement list doesn't contain mixed Partial reduce types. + + Mixed Partial types (e.g., ``Partial("sum")`` and ``Partial("max")`` together in the + same placement list) are not supported and will raise a ``ValueError``. This restriction + exists because nonlinear reductions (e.g., max) don't commute with linear reductions + (e.g., sum), which means the relative ordering of different partial types would be + semantically critical during redistribution. Rather than introducing complex ordering + constraints, we prohibit mixing different Partial reduce types. + + Note: Partial("sum") and Partial("avg") DO commute with each other, so they can be ordered + arbitrarily, and we allow this. + + This function is called internally by public APIs like :meth:`DTensor.from_local` and + :func:`distribute_tensor` to validate placements early, before DTensor construction. + + Args: + placements (Sequence[:class:`Placement`]): A sequence of placement specifications + to validate. + + Raises: + ValueError: If the placements contain more than one distinct Partial reduce type. + """ + partial_reduce_ops: set[str] = set() + for p in placements: + if isinstance(p, Partial): + partial_reduce_ops.add(p.reduce_op) + + if len(partial_reduce_ops) > 1 and partial_reduce_ops != {"sum", "avg"}: + raise ValueError( + f"Mixed Partial reduce types are not supported in the same placement list. " + f"Found reduce ops: {partial_reduce_ops}. " + f"Please ensure all Partial placements use the same reduce operation." + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..df99f7d750034e48f9100399c18ee467c4e98f14 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__init__.py @@ -0,0 +1,70 @@ +# mypy: allow-untyped-defs +import torch._C +from torch.distributed.tensor.debug._comm_mode import CommDebugMode +from torch.distributed.tensor.debug._visualize_sharding import visualize_sharding + + +__all__ = ["CommDebugMode", "visualize_sharding"] + + +def _get_python_sharding_prop_cache_info(): + """ + Get the cache info for the Python sharding propagation cache, used for debugging purpose only. + This would return a named tuple showing hits, misses, maxsize and cursize of the sharding + propagator cache. Note that directly calling into the sharding propagator does not share cache + state with the DTensor dispatch fast path! + """ + from torch.distributed.tensor._api import DTensor + + return ( + DTensor._op_dispatcher.sharding_propagator.propagate_op_sharding.cache_info() # type:ignore[attr-defined] + ) + + +def _get_fast_path_sharding_prop_cache_stats(): + """ + Get a tuple (hits, misses) for the fast path sharding propagation cache, used for debugging + only. + """ + return torch._C._get_DTensor_sharding_propagator_cache_stats() + + +def _clear_python_sharding_prop_cache(): + """ + Clears the cache for the Python sharding propagation cache, used for debugging purpose only. + """ + from torch.distributed.tensor._api import DTensor + + return ( + DTensor._op_dispatcher.sharding_propagator.propagate_op_sharding.cache_clear() # type:ignore[attr-defined] + ) + + +def _clear_fast_path_sharding_prop_cache(): + """ + Clears the cache for the fast path sharding propagation cache, used for debugging purpose only. + """ + torch._C._clear_DTensor_sharding_propagator_cache() + + +def _clear_sharding_prop_cache(): + """ + Clears both the Python and fast path sharding propagation caches, used for debugging purpose only. + This is the recommended way to clear all sharding propagation caches. + """ + _clear_python_sharding_prop_cache() + _clear_fast_path_sharding_prop_cache() + + +def _reinit_dispatch_logger(): + """ + Resets the cached DTensor dispatch logger state so that the next DTensor + dispatch re-checks whether debug logging is enabled. Call this after + changing the log level on the ``torch.distributed.tensor._dispatch`` logger. + """ + torch._C._reinit_DTensor_dispatch_logger() + + +# Set namespace for exposed private names +CommDebugMode.__module__ = "torch.distributed.tensor.debug" +visualize_sharding.__module__ = "torch.distributed.tensor.debug" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5ee3de5c62346672097171b75fbfb2d1c4980a22 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_comm_mode.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_comm_mode.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ca361315d0430f341f99a7af7ef8cb676f511775 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_comm_mode.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_op_coverage.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_op_coverage.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..58745a7e669a76a035fe739cf758acd2724523a8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_op_coverage.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_visualize_sharding.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_visualize_sharding.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5ab5f13c9b2d3c0582ca96a6c0ddf7213da86072 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/__pycache__/_visualize_sharding.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_comm_mode.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_comm_mode.py new file mode 100644 index 0000000000000000000000000000000000000000..f4882de56d3daebb106fa6922f902137e1678a52 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_comm_mode.py @@ -0,0 +1,749 @@ +# mypy: allow-untyped-defs +import copy +import json +import re +import weakref +from collections import defaultdict +from typing import Any + +import torch +import torch.nn +from torch._guards import detect_fake_mode +from torch.autograd.graph import register_multi_grad_hook +from torch.distributed._tools.mod_tracker import ModTracker +from torch.distributed.tensor._api import DTensor +from torch.nn.modules.module import ( + register_module_forward_hook, + register_module_forward_pre_hook, + register_module_full_backward_pre_hook, +) +from torch.utils._python_dispatch import TorchDispatchMode +from torch.utils._pytree import tree_flatten + + +__all__ = ["CommDebugMode"] + +funcol_native = torch.ops._c10d_functional +funcol_py = torch.ops.c10d_functional +funcol_autograd = torch.ops._c10d_functional_autograd +c10d_ops = torch.ops.c10d + +NATIVE_TO_PY_MAPPING = { + funcol_native.all_gather_into_tensor: funcol_py.all_gather_into_tensor, + funcol_native.all_gather_into_tensor_coalesced: funcol_py.all_gather_into_tensor_coalesced, + funcol_native.all_reduce: funcol_py.all_reduce, + funcol_native.all_reduce_coalesced: funcol_py.all_reduce_coalesced, + funcol_native.all_to_all_single: funcol_py.all_to_all_single, + funcol_native.broadcast: funcol_py.broadcast, + funcol_native.reduce_scatter_tensor: funcol_py.reduce_scatter_tensor, + funcol_native.reduce_scatter_tensor_coalesced: funcol_py.reduce_scatter_tensor_coalesced, + # functional ops + funcol_autograd.all_to_all_single: funcol_py.all_to_all_single, +} + +c10d_collective_ops = { + c10d_ops._allgather_base_, + c10d_ops._reduce_scatter_base_, + c10d_ops.allgather_, + c10d_ops.allgather_coalesced_, + c10d_ops.allgather_into_tensor_coalesced_, + c10d_ops.allreduce_, + c10d_ops.allreduce_coalesced_, + c10d_ops.alltoall_, + c10d_ops.alltoall_base_, + c10d_ops.broadcast_, + c10d_ops.gather_, + c10d_ops.scatter_, + c10d_ops.reduce_, + c10d_ops.reduce_scatter_, + c10d_ops.reduce_scatter_tensor_coalesced_, +} + +trivial_ops = { + "aten.detach.default", + "aten.t.default", + "aten.view.default", + "aten._to_copy.default", + "aten.as_strided.default", + "aten.transpose.int", +} + + +class _CommModeModuleTracker(ModTracker): + """ + Inherits ModuleTracker and expands on its functionality to track the + parameters and sharding information of a model at a module-level + """ + + def __init__(self): + super().__init__() + self.module_helper_dict = {} + self.module_parameters_dict = {} + self.module_parents_dict = {} + self.register_forward_hook_handles = {} + self.parent_dict = {} + self.parent_list = [] + self.sharding_dict = {} + self.activation_checkpointing = False + self.name = "" + + def _fw_set_module_hook(self, mod, input, output): + """ + Updates the current module after module finishes running and + all other hooks are resolved + """ + + if self.is_bw: + self.activation_checkpointing = True + else: + self.activation_checkpointing = False + + if not self.activation_checkpointing: + # module is no longer parent of next modules + self.parent_list.pop() + + # set current module to previous parent module + self.name = self.parent_list[-1] + + def _fw_pre_hook(self, mod, input): + """ + This function is called before the forward pass of a module. It + collects the parameters and sharding information of a module and + stores it in a dictionary. + """ + if self.is_bw: + self.activation_checkpointing = True + else: + self.activation_checkpointing = False + + self.name = super()._get_mod_name(mod) + w_mod = weakref.ref(mod) + + # adds current sub-module to module tracker parent class + super()._get_append_fn(w_mod, self.name, False)() + + args, _ = tree_flatten(input) + tensors = [a for a in args if isinstance(a, torch.Tensor) and a.requires_grad] + if not self.is_bw and tensors: + register_multi_grad_hook( + tensors, super()._get_pop_fn(w_mod, self.name, True) + ) + + if not self.activation_checkpointing: + # contains information about module ordering and depth in the module tree + if self.name not in self.module_helper_dict: + self.module_helper_dict[self.name] = {} + + self.module_helper_dict[self.name]["module_type"] = ( + str(type(mod)).replace("<", "").replace(">", "") + ) + self.module_helper_dict[self.name]["depth"] = len(self.parents) - 1 + + for param_name, param in mod.named_parameters(recurse=False): + if self.name not in self.module_parameters_dict: + self.module_parameters_dict[self.name] = {} + + self.module_parameters_dict[self.name][param_name] = param.data + + if isinstance(param.data, DTensor): + key_name = self.name + "." + param_name + self.sharding_dict[key_name] = param.data.placements + + if "parameters" not in self.module_helper_dict[self.name]: + self.module_helper_dict[self.name]["parameters"] = {} + + self.module_helper_dict[self.name]["parameters"][param_name] = str( + param.data.placements + ) + + # used to store module's parents to ensure correctness in backward pass/checkpointing + if self.name not in self.module_parents_dict: + self.module_parents_dict[self.name] = copy.deepcopy(self.parents) + + # used to create parent-child module associations for json dumps + parent = self.parent_list[-1] + if parent not in self.parent_dict: + self.parent_dict[parent] = [] + + self.parent_dict[parent].append(self.name) + self.parent_list.append(self.name) + + if self.name in self.register_forward_hook_handles: + self.register_forward_hook_handles[self.name].remove() + self.register_forward_hook_handles[self.name] = mod.register_forward_hook( + self._fw_set_module_hook + ) + + def _fw_post_hook(self, mod, input, output): # pylint: disable=useless-parent-delegation + """ + This function is called when the forward pass of a module is called. + It updates the module tracker and removes the module from parent data + """ + + super()._fw_post_hook(mod, input, output) + + def _bw_hook(self, mod, output): + """ + This function is called when the backward pass of a module is called. It + updates the current module for backward passes + """ + self.activation_checkpointing = False + self.name = super()._get_mod_name(mod) + + def __enter__(self): + self.activation_checkpointing = False + self.module_parameters_dict.clear() + self.sharding_dict.clear() + self.parent_dict.clear() + self.parent_list = ["Global"] + self.module_helper_dict.clear() + self.module_helper_dict["Global"] = {"depth": 0} + self.module_parents_dict.clear() + self.module_parents_dict["Global"] = set() + self._fw_pre_handle = register_module_forward_pre_hook(self._fw_pre_hook) + self._fw_post_handle = register_module_forward_hook(self._fw_post_hook) + self.register_forward_hook_handles.clear() + self._bw_handle = register_module_full_backward_pre_hook(self._bw_hook) + self.name = "Global" + + def __exit__(self, *args): + super().__exit__(*args) + self._bw_handle.remove() + + # removes all forward_hook handles added in the pre-hook + for handle in self.register_forward_hook_handles.values(): + handle.remove() + + def print_paramater_info(self): + print(self.module_parameters_dict) + + def print_sharding_info(self): + for key, value in self.sharding_dict.items(): + print(key + ": " + str(value)) + + +class CommDebugMode(TorchDispatchMode): + """ + :class:`CommDebugMode` is a context manager that counts the number of + functional collectives within its context. It does this using a + ``TorchDispatchMode``. + + .. note:: Not all collectives are supported yet. + + Example usage + + .. code-block:: python + + mod = ... + comm_mode = CommDebugMode() + with comm_mode: + mod.sum().backward() + print(comm_mode.get_comm_counts()) + """ + + def __init__(self): + super().__init__() + self.supports_higher_order_operators = True + self.comm_counts: dict[Any, int] = defaultdict(int) + self.comm_module_counts = {} + self.comm_module_operation_counts = {} + self.comm_registry = set() + for native_op, py_op in NATIVE_TO_PY_MAPPING.items(): + self.comm_registry.add(native_op) + self.comm_registry.add(py_op) + + self.comm_registry.add(torch.ops._dtensor.shard_dim_alltoall) + self.advanced_module_tracker = _CommModeModuleTracker() + + def generate_json_dump(self, file_name="comm_mode_log.json", noise_level=3): + """ + Creates json file used to build browser visual + 0. prints module-level collective counts + 1. prints dTensor operations not included in trivial operations + 2. prints operations not included in trivial operations + 3. prints all operations + """ + + ( + include_DTensor_ops, + include_module_data, + include_ops, + include_trivial_ops, + ) = self._set_noise_parameters(noise_level) + + # recursively builds json data + def add_json_information(json_dict, fqn): + json_dict["fqn"] = fqn + json_dict["module_type"] = "" + json_dict["parameters"] = [] + json_dict["children"] = [] + json_dict["collectives_forward"] = [] + json_dict["collectives_backward"] = [] + json_dict["operations_forward"] = [] + json_dict["operations_backward"] = [] + + # adds module layer type and parameters, and their sharding + if ( + "module_type" in self.advanced_module_tracker.module_helper_dict[fqn] + and include_module_data + ): + json_dict["module_type"] = ( + self.advanced_module_tracker.module_helper_dict[fqn]["module_type"] + ) + + if "parameters" in self.advanced_module_tracker.module_helper_dict[fqn]: + for ( + param_name, + placement, + ) in self.advanced_module_tracker.module_helper_dict[fqn][ + "parameters" + ].items(): + json_dict["parameters"].append((param_name, placement)) + + # adds module collective information + if fqn in self.comm_module_counts: + for collective, count in self.comm_module_counts[fqn][ + "forward" + ].items(): + json_dict["collectives_forward"].append((str(collective), count)) + + for collective, count in self.comm_module_counts[fqn][ + "backward" + ].items(): + json_dict["collectives_backward"].append((str(collective), count)) + + # adds module operation information + forward_operations = [] + backward_operations = [] + checkpointing_operations = [] + + # only get operations if the minimum operation noise level is set to true + if include_DTensor_ops: + if fqn in self.comm_module_operation_counts: + ( + forward_operations, + backward_operations, + checkpointing_operations, + ) = self._get_operations_list( + self.comm_module_operation_counts[fqn] + ) + + # remove all operations who don't have DTensor inputs + if not include_ops: + forward_operations = [ + op for op in forward_operations if len(op["input_sharding"]) + ] + backward_operations = [ + op for op in backward_operations if len(op["input_sharding"]) + ] + checkpointing_operations = [ + op for op in checkpointing_operations if len(op["input_sharding"]) + ] + + # remove all operations in trivial operations set + if not include_trivial_ops: + forward_operations = [ + op + for op in forward_operations + if str(op["name"]) not in trivial_ops + ] + backward_operations = [ + op + for op in backward_operations + if str(op["name"]) not in trivial_ops + ] + checkpointing_operations = [ + op + for op in checkpointing_operations + if str(op["name"]) not in trivial_ops + ] + + # converts operation information into string format for json.dumps() + forward_operations = copy.deepcopy(forward_operations) + for op in forward_operations: + op["name"] = str(op["name"]) + + for i in range(len(op["input_sharding"])): + op["input_sharding"][i] = str(op["input_sharding"][i]) + op["input_shape"][i] = str(op["input_shape"][i]) + + backward_operations = copy.deepcopy(backward_operations) + for op in backward_operations: + op["name"] = str(op["name"]) + + for i in range(len(op["input_sharding"])): + op["input_sharding"][i] = str(op["input_sharding"][i]) + op["input_shape"][i] = str(op["input_shape"][i]) + + checkpointing_operations = copy.deepcopy(checkpointing_operations) + for op in checkpointing_operations: + op["name"] = str(op["name"]) + + for i in range(len(op["input_sharding"])): + op["input_sharding"][i] = str(op["input_sharding"][i]) + op["input_shape"][i] = str(op["input_shape"][i]) + + json_dict["operations_forward"] = forward_operations + json_dict["operations_backward"] = backward_operations + json_dict["operations_checkpointing"] = checkpointing_operations + + if fqn not in self.advanced_module_tracker.parent_dict: + return json_dict + + # recursively adds module's children + for ele in self.advanced_module_tracker.parent_dict[fqn]: + json_dict["children"].append(add_json_information({}, ele)) + + return json_dict + + json_dict: dict[str, Any] = {} + add_json_information(json_dict, "Global") + + # converts dictionary into json file + with open(file_name, "w") as json_file: + json.dump(json_dict, json_file, indent=4) + + def generate_comm_debug_tracing_table(self, noise_level=3): + """ + Generates detailed table displaying operations and collective tracing information + on a module level. Amount of information is dependent on noise_level + + 0. prints module-level collective counts + 1. prints dTensor operations not included in trivial operations, module information + 2. prints operations not included in trivial operations + 3. prints all operations + """ + + ( + include_DTensor_ops, + include_module_data, + include_ops, + include_trivial_ops, + ) = self._set_noise_parameters(noise_level) + + table = "" + for fqn in self.advanced_module_tracker.module_helper_dict: + # setting up indentations for table formatting + indent = " " * ( + 2 * self.advanced_module_tracker.module_helper_dict[fqn]["depth"] + ) + table += f"{indent}{fqn}\n" + + if include_module_data: + if ( + "module_type" + in self.advanced_module_tracker.module_helper_dict[fqn] + ): + module_type = self.advanced_module_tracker.module_helper_dict[fqn][ + "module_type" + ] + table += f"{indent}*module type: {module_type}\n" + + if "parameters" in self.advanced_module_tracker.module_helper_dict[fqn]: + table += f"{indent}*Parameter List\n" + for ( + param_name, + placement, + ) in self.advanced_module_tracker.module_helper_dict[fqn][ + "parameters" + ].items(): + table += f"{indent} *{param_name}: {placement}\n" + + indent += " " + collective_indent = " " * ( + 2 * self.advanced_module_tracker.module_helper_dict[fqn]["depth"] + 2 + ) + operation_indent = " " * ( + 2 * self.advanced_module_tracker.module_helper_dict[fqn]["depth"] + 3 + ) + + # separate the module's collective and operations by forward and backward + forward_collectives = {} + backward_collectives = {} + if fqn in self.comm_module_counts: + forward_collectives = self.comm_module_counts[fqn]["forward"] + backward_collectives = self.comm_module_counts[fqn]["backward"] + + forward_operations = [] + backward_operations = [] + checkpointing_operations = [] + + if include_DTensor_ops: + if fqn in self.comm_module_operation_counts: + ( + forward_operations, + backward_operations, + checkpointing_operations, + ) = self._get_operations_list( + self.comm_module_operation_counts[fqn] + ) + + def add_tracing_information(table, collectives_dict, operation_list): + """ + adds tracing information for module's forward or backward + """ + for collective, count in collectives_dict.items(): + table += ( + f"\033[1;33m{collective_indent}*{collective}: {count}\033[0m\n" + ) + + def add_operations( + table, operation, collective_indent, operation_indent + ): + """ + adds operation information to the table + """ + table += f"\033[1;33m{collective_indent}**{operation_name}\033[0m\n" + + if len(operation["input_shape"]): + operation_shape = operation["input_shape"] + operation_sharding = operation["input_sharding"] + operation_device_mesh = operation["device_mesh"] + + table += f"\033[1;31m{operation_indent}shape: {operation_shape}\033[0m\n" + table += f"\033[1;31m{operation_indent}sharding: {operation_sharding}\033[0m\n" + table += f"\033[1;31m{operation_indent}device mesh: {operation_device_mesh}\033[0m\n" + + return table + + for operation in operation_list: + operation_name = str(operation["name"]) + + # include all operations + if include_trivial_ops: + table = add_operations( + table, operation, collective_indent, operation_indent + ) + + # include all operations not in trivial operations + elif include_ops and operation_name not in trivial_ops: + table = add_operations( + table, operation, collective_indent, operation_indent + ) + + # only include dTensor operations not in trivial set + elif ( + include_DTensor_ops + and (operation_name not in trivial_ops) + and len(operation["input_shape"]) + ): + table = add_operations( + table, operation, collective_indent, operation_indent + ) + + return table + + if len(forward_collectives) or len(forward_operations): + table += f"{indent}FORWARD PASS\n" + table = add_tracing_information( + table, forward_collectives, forward_operations + ) + + if len(backward_collectives) or len(backward_operations): + table += f"{indent}BACKWARD PASS\n" + table = add_tracing_information( + table, backward_collectives, backward_operations + ) + + if len(checkpointing_operations): + table += f"{indent}ACTIVATION CHECKPOINTING\n" + table = add_tracing_information(table, {}, checkpointing_operations) + + return table + + def _get_operations_list(self, module_operation_counts): + forward_operations = [ + op for op in module_operation_counts["operations_list"] if not op["is_bw"] + ] + backward_operations = [ + op + for op in module_operation_counts["operations_list"] + if op["is_bw"] and not op["is_activation_checkpointing"] + ] + checkpointing_operations = [ + op + for op in module_operation_counts["operations_list"] + if op["is_activation_checkpointing"] + ] + + return forward_operations, backward_operations, checkpointing_operations + + def get_total_counts(self) -> int: + return sum(self.comm_counts.values()) + + def get_comm_counts(self) -> dict[Any, int]: + """Returns the communication counts as a dictionary. + + Returns: + Dict[Any, int]: The communication counts as a dictionary. + """ + return self.comm_counts + + def get_parameter_info(self) -> dict[str, dict[str, Any]]: + return self.advanced_module_tracker.module_parameters_dict + + def get_sharding_info(self) -> dict[str, dict[str, Any]]: + return self.advanced_module_tracker.sharding_dict + + def __enter__(self): + self.comm_counts.clear() + self.comm_module_counts.clear() + self.comm_module_counts["Global"] = {} + self.comm_module_counts["Global"]["forward"] = defaultdict(int) + self.comm_module_counts["Global"]["backward"] = defaultdict(int) + + self.comm_module_operation_counts.clear() + + super().__enter__() + self.advanced_module_tracker.__enter__() + return self + + # pyrefly: ignore [bad-override] + def __exit__(self, *args): + self.advanced_module_tracker.__exit__() + super().__exit__(*args) + + def log_comm_debug_tracing_table_to_file( + self, file_name="comm_mode_log.txt", noise_level=3 + ): + """ + Alternative to console CommDebugMode output, writes to file specified by the user + """ + ansi_escape = re.compile(r"\x1B\[[0-?]*[ -/]*[@-~]") + table = ansi_escape.sub("", self.generate_comm_debug_tracing_table(noise_level)) + + with open(file_name, "w") as log_file: + log_file.write(table) + + def _set_noise_parameters(self, noise_level): + """ + sets variables controlling what information displays based on noise level + """ + include_DTensor_ops = False + include_module_data = False + include_ops = False + include_trivial_ops = False + + if noise_level > 0: + include_DTensor_ops = True + include_module_data = True + + if noise_level > 1: + include_ops = True + + if noise_level > 2: + include_trivial_ops = True + + return ( + include_DTensor_ops, + include_module_data, + include_ops, + include_trivial_ops, + ) + + def __torch_dispatch__(self, func, types, args=(), kwargs=None): + # When running this mode with DTensor, ordinarily all modes will + # run **before** subclasses get a chance to run. + # Returning NotImplemented here gives us a chance to let DTensor + # run and desugar into comms ops, before CommDebugMode sees them. + + # Higher-order operators (e.g. run_dtensor_rng_op) don't have + # _overloadpacket and aren't collectives — just redispatch. + if isinstance(func, torch._ops.HigherOrderOperator): + kwargs = kwargs if kwargs else {} + return func(*args, **kwargs) + + # sets up operation-level collective count + if self.advanced_module_tracker.name not in self.comm_module_operation_counts: + # dictionary should hold module input and output shape, operations list and collective counter + self.comm_module_operation_counts[self.advanced_module_tracker.name] = { + "operations_list": [] + } + operation_dict = {} + operation_dict["name"] = func + + operation_dict["input_shape"] = [] + operation_dict["input_sharding"] = [] + operation_dict["device_mesh"] = "" + + # tracks if the operation is part of the backward pass + operation_dict["is_bw"] = self.advanced_module_tracker.is_bw + + # tracks if the operation is part of activation checkpointing + operation_dict["is_activation_checkpointing"] = ( + self.advanced_module_tracker.activation_checkpointing + ) + + if any(t == DTensor for t in types): + for ele in args: + if isinstance(ele, DTensor): + # saves shapes and placements of all DTensor args + operation_dict["input_shape"].append(ele.shape) + operation_dict["input_sharding"].append(ele.placements) + operation_dict["device_mesh"] = str(ele.device_mesh) + + self.comm_module_operation_counts[self.advanced_module_tracker.name][ + "operations_list" + ].append(operation_dict) + + return NotImplemented + + kwargs = kwargs if kwargs else {} + out = func(*args, **kwargs) + func_packet = func._overloadpacket + + # We have many tests that use CommDebugMode to verify the occurrence of + # collectives. These tests do so by querying comm_counts with legacy + # funcol ops as key. For the purpose of native funcol migration, we + # need these tests to work for both legacy and native funcol. To avoid + # the need to modify all tests to accommodate the two implementations, + # we make CommDebugMode translate native funcol ops into legacy funcol + # ops until the migration finishes. + + if func_packet in self.comm_registry or func_packet in c10d_collective_ops: + if func_packet in NATIVE_TO_PY_MAPPING: + func_packet = NATIVE_TO_PY_MAPPING[func_packet] + self.comm_counts[func_packet] += 1 + + key = "forward" + if self.advanced_module_tracker.is_bw: + key = "backward" + + # adds collective count to current module + if self.advanced_module_tracker.name not in self.comm_module_counts: + self.comm_module_counts[self.advanced_module_tracker.name] = {} + self.comm_module_counts[self.advanced_module_tracker.name][ + "forward" + ] = defaultdict(int) + self.comm_module_counts[self.advanced_module_tracker.name][ + "backward" + ] = defaultdict(int) + self.comm_module_counts[self.advanced_module_tracker.name][key][ + func_packet + ] += 1 + + # adds collective count to parent modules + for par in self.advanced_module_tracker.module_parents_dict[ + self.advanced_module_tracker.name + ]: + # makes sure we aren't double counting when current sub-module hasn't been removed from parents + if par != self.advanced_module_tracker.name: + if par not in self.comm_module_counts: + self.comm_module_counts[par] = {} + self.comm_module_counts[par]["forward"] = defaultdict(int) + self.comm_module_counts[par]["backward"] = defaultdict(int) + self.comm_module_counts[par][key][func_packet] += 1 + + # if tensor op uses fake tensors, return + if detect_fake_mode(args): + return out + + # add tensor operation to module operation list + self.comm_module_operation_counts[self.advanced_module_tracker.name][ + "operations_list" + ].append(operation_dict) + + return out + + def __repr__(self): + return f"CommDebugMode(get_total_counts()={self.get_total_counts()})" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_op_coverage.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_op_coverage.py new file mode 100644 index 0000000000000000000000000000000000000000..0b7a239fe521de1259d7be9af053372c87d264cd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_op_coverage.py @@ -0,0 +1,105 @@ +# mypy: allow-untyped-defs +from operator import itemgetter + +import torch +import torch.fx +import torch.nn as nn +from functorch.compile import make_boxed_func +from torch._functorch.compilers import aot_module +from torch._inductor.decomposition import select_decomp_table +from torch.distributed.tensor import DTensor + + +inductor_decomps = select_decomp_table() + +graphs: list[torch.fx.GraphModule] = [] + + +def fwd_bwd_compiler(fx_g, _): + graphs.append(fx_g) + return make_boxed_func(fx_g) + + +def get_inductor_decomp_graphs(model: nn.Module, args, kwargs): + """ + Obtain forward and backward graphs of a model with inductor decompositions using tracing and aot_module. + + Convenient util to get the fwd and bwd graphs of an arbitrary model + with inductor decompositions. Note that this would simply do tracing + with aot_module and don't ensure correctness. This is useful to track + the ops needed in DTensor. + """ + compiled_mod = aot_module( + model, fw_compiler=fwd_bwd_compiler, decompositions=inductor_decomps + ) + output = compiled_mod(*args, **kwargs) + + if output.ndim != 0: + # if output is not a scalar tensor, by default sum it in order to + # run backward + output = output.sum() + + output.backward() + + # one fwd, one bwd graph + if len(graphs) != 2: + raise AssertionError + return graphs + + +def print_op_coverage_summary(model: nn.Module, args, kwargs, *, output_csv=False): + """ + Util to print the operator coverage summary of a certain model with tabulate. + + Must have tabulate module installed. + """ + # python module required for summary + import csv + + from tabulate import tabulate + + fwd_graph, bwd_graph = get_inductor_decomp_graphs(model, args, kwargs) + + op_counts = {} + + for node in fwd_graph.graph.nodes: + if node.op == "call_function" and isinstance( + node.target, torch._ops.OpOverload + ): + if node.target not in op_counts: + op_counts[node.target] = 0 + + op_counts[node.target] += 1 + + for node in bwd_graph.graph.nodes: + if node.op == "call_function" and isinstance( + node.target, torch._ops.OpOverload + ): + if node.target not in op_counts: + op_counts[node.target] = 0 + + op_counts[node.target] += 1 + + op_infos = [] + + for op, count in op_counts.items(): + supported = op in DTensor._op_dispatcher.sharding_propagator.op_to_rules + op_infos.append([op, str(op._schema), count, supported]) + + # sort the op info base on the total count index + count_idx = 2 + op_infos.sort(key=itemgetter(count_idx), reverse=True) + + headers = ["Operator", "Schema", "Total Count", "Supported"] + print(tabulate(op_infos, headers=headers)) + + if output_csv: + # Open a CSV file for writing + with open("op_summary.csv", "w", newline="") as csv_file: + # Create a CSV writer object + csv_writer = csv.writer(csv_file) + + csv_writer.writerow(headers) + # Write each table row to the CSV file + for row in op_infos: + csv_writer.writerow(row) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_visualize_sharding.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_visualize_sharding.py new file mode 100644 index 0000000000000000000000000000000000000000..ef8897c83f667de3d97fe51e416e24180305c2a3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/debug/_visualize_sharding.py @@ -0,0 +1,227 @@ +# mypy: allow-untyped-defs +import importlib.util + +import numpy as np + +from torch._prims_common import ShapeType +from torch.distributed.tensor._utils import _compute_local_shape_and_global_offset + + +__all__ = ["visualize_sharding"] + +Color = tuple[float, float, float] + + +def _create_table( + shards: list[tuple[tuple[int, int], tuple[int, int], int]], device_kind: str = "" +): + """ + Creates a tabulate table given row and column ranges with device name + """ + from tabulate import tabulate + + # Extract unique row and column ranges + row_ranges = sorted({block[0] for block in shards}) + col_ranges = sorted({block[1] for block in shards}) + + # Create a matrix initialized with empty strings + matrix = [["" for _ in col_ranges] for _ in row_ranges] + + # Fill the matrix with values + for block in shards: + row_index = row_ranges.index(block[0]) + col_index = col_ranges.index(block[1]) + if matrix[row_index][col_index] == "": + matrix[row_index][col_index] = device_kind + ":" + str(block[2]) + else: + matrix[row_index][col_index] += "," + str(block[2]) + + # Prepare headers + row_headers = [f"Row {r[0]}-{r[1]}" for r in row_ranges] + col_headers = [f"Col {c[0]}-{c[1]}" for c in col_ranges] + + return tabulate(matrix, headers=col_headers, showindex=row_headers) + + +def make_color_iter(color_map, num_rows, num_cols): + num_colors = num_rows * num_cols + for idx in range(num_colors): + yield color_map(idx) + + +def _canonicalize_color(color: Color) -> str: + if isinstance(color, str): + return color + r, g, b = (int(a * 255) for a in color) + return f"#{r:02X}{g:02X}{b:02X}" + + +def _get_text_color(color: str) -> str: + r, g, b = map(lambda x: int(x, 16), (color[1:3], color[3:5], color[5:7])) # noqa: C417 + if (r * 0.299 + g * 0.587 + b * 0.114) > 186: + return "#000000" + return "#ffffff" + + +def _create_rich_table( + shape: ShapeType, + shards: list[tuple[tuple[int, int], tuple[int, int], int]], + device_kind: str = "", + scale: float = 1.0, + min_width: int = 9, + max_width: int = 80, +): + import matplotlib + import rich.align + import rich.box + import rich.console + import rich.padding + import rich.style + import rich.table + + dtensor_height = shape[0] + dtensor_width = shape[1] if len(shape) == 2 else 1 + + row_ranges = sorted({s[0] for s in shards}) + col_ranges = sorted({s[1] for s in shards}) + num_rows, num_cols = len(row_ranges), len(col_ranges) + + console = rich.console.Console(width=max_width) + use_color = console.color_system + color_iter = make_color_iter(matplotlib.colormaps["tab20b"], num_rows, num_cols) + + base_height = int(10 * scale) + aspect_ratio = (shape[1] if len(shape) == 2 else 1) / shape[0] + base_width = int(base_height * aspect_ratio) + height_to_width_ratio = 2.5 + + table = rich.table.Table( + show_header=False, + show_lines=not use_color, + padding=0, + highlight=not use_color, + pad_edge=False, + box=rich.box.SQUARE if not use_color else None, + ) + for row in range(num_rows): + table_row = [] + for col in range(num_cols): + entry = ( + device_kind + + ":" + + ",".join( + [ + str(device_id) + for row_range, col_range, device_id in shards + if row_range == row_ranges[row] and col_range == col_ranges[col] + ] + ) + ) + width = (col_ranges[col][1] - col_ranges[col][0]) / dtensor_width + width = int(width * base_width * height_to_width_ratio) + height = (row_ranges[row][1] - row_ranges[row][0]) / dtensor_height + height = int(height * base_height) + left_padding, remainder = divmod(width - len(entry) - 2, 2) + right_padding = left_padding + remainder + top_padding, remainder = divmod(height - 2, 2) + bottom_padding = top_padding + remainder + if use_color: + color = _canonicalize_color(next(color_iter)[:3]) + text_color = _get_text_color(color) + top_padding += 1 + bottom_padding += 1 + left_padding += 1 + right_padding += 1 + else: + color = None + text_color = None + padding = ( + max(top_padding, 0), + max(right_padding, 0), + max(bottom_padding, 0), + max(left_padding, 0), + ) + table_row.append( + rich.padding.Padding( + rich.align.Align(entry, "center", vertical="middle"), + padding, + style=rich.style.Style(bgcolor=color, color=text_color), + ) + ) + table.add_row(*table_row) + console.print(table, end="\n\n") + + +def visualize_sharding(dtensor, header="", use_rich: bool = False): + """ + Visualizes sharding in the terminal for :class:`DTensor` that are 1D or 2D. + + .. note:: This requires the ``tabulate`` package, or ``rich`` and ``matplotlib``. + No sharding info will be printed for empty tensors + """ + if dtensor.numel() == 0: # Do not print empty dtensors. + return + + if len(dtensor.shape) >= 3: + raise RuntimeError("visualize sharding supports only 1D or 2D DTensor") + + if dtensor.device_mesh.get_coordinate() is None: # current rank is not in the mesh + return + + # Only display the visualization once for each DTensor, on the rank whose + # coordinate is 0 on all dimensions. For example, if the mesh is a full mesh, + # we will only print on rank 0. + local_rank_zero_on_all_dim = all( + dtensor.device_mesh.get_local_rank(mesh_dim=dim) == 0 + for dim in range(dtensor.device_mesh.ndim) + ) + if not local_rank_zero_on_all_dim: + return + + device_coords = { + int(device_index.item()): list(coord) + for coord, device_index in np.ndenumerate( + np.array(dtensor.device_mesh.mesh.tolist()) + ) + } + + device_shard_shape_and_offsets = { + device_index: _compute_local_shape_and_global_offset( + dtensor.shape, + dtensor.device_mesh.shape, + lambda i: device_coords[device_index][i], + dtensor.placements, + ) + for device_index in device_coords + } + + # Extend shards in a 1D tensor to 2D + device_shard_shape_and_offsets = { + device_index: ( + shape if len(shape) == 2 else (shape[0], 1), + offset if len(offset) == 2 else (offset[0], 0), + ) + for device_index, (shape, offset) in device_shard_shape_and_offsets.items() + } + + shards = [ + ( + (offset[0], offset[0] + shape[0] - 1), + (offset[1], offset[1] + shape[1] - 1), + device_index, + ) + for device_index, (shape, offset) in device_shard_shape_and_offsets.items() + ] + + if ( + importlib.util.find_spec("rich") + and importlib.util.find_spec("matplotlib") + and use_rich + ): + _create_rich_table( + dtensor.shape, shards, device_kind=dtensor.device_mesh.device_type + ) + elif importlib.util.find_spec("tabulate"): + print(_create_table(shards, device_kind=dtensor.device_mesh.device_type)) + else: + raise ValueError("`visualize_sharding` requires either `rich` or `tabulate`.") diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/device_mesh.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/device_mesh.py new file mode 100644 index 0000000000000000000000000000000000000000..ca59ded5eb52bc0a3878e76077ad2879df4bf499 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/device_mesh.py @@ -0,0 +1,9 @@ +from torch.distributed.device_mesh import ( # noqa: F401 + _get_device_handle, + _mesh_resources, + DeviceMesh, + init_device_mesh, +) + + +__all__ = ["init_device_mesh", "DeviceMesh"] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..0012040d74a3e0caaf23a71c138681b9c372e591 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__init__.py @@ -0,0 +1,34 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from collections.abc import Iterator +from contextlib import contextmanager + +from torch.distributed.tensor._api import DTensor +from torch.distributed.tensor.experimental._attention import context_parallel +from torch.distributed.tensor.experimental._func_map import local_map +from torch.distributed.tensor.experimental._register_sharding import register_sharding + + +__all__ = ["context_parallel", "implicit_replication", "local_map", "register_sharding"] + + +@contextmanager +def implicit_replication() -> Iterator[None]: + """ + This context manager allows :class:`DTensor` to implicitly treat all non-DTensors (``torch.Tensor``) + in the program be replicate :class:`DTensor` s during the operator computation. + + .. warning:: This might possible lead to incorrect results if ``torch.Tensor`` s are not replicated + in practice, please use it at your discretion. + """ + try: + DTensor._op_dispatcher._allow_implicit_replication = True + yield + finally: + DTensor._op_dispatcher._allow_implicit_replication = False + + +# Set namespace for exposed private names +context_parallel.__module__ = "torch.distributed.tensor.experimental" +implicit_replication.__module__ = "torch.distributed.tensor.experimental" +local_map.__module__ = "torch.distributed.tensor.experimental" +register_sharding.__module__ = "torch.distributed.tensor.experimental" diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..40382dc926fde49de21fea231920c8e174fd3de3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_attention.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_attention.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c432e7f6bbfbe6affdf74d52f9613fb9233b770a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_attention.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_func_map.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_func_map.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..876975348b99c80682eb2dc00c9f17f78acb95be Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_func_map.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_register_sharding.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_register_sharding.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..847014a12e8cb616021d0c681acf7c6960c14f53 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_register_sharding.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_tp_transform.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_tp_transform.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c6ab79a369bc607f1989954ac269c757b0f92856 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/__pycache__/_tp_transform.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_attention.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_attention.py new file mode 100644 index 0000000000000000000000000000000000000000..f238739ddd5cf4f8e120f1e6a0337f0cfc8cc58d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_attention.py @@ -0,0 +1,44 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +# Backward compatibility stub - this module has been moved to _context_parallel/_attention.py + +from ._context_parallel._attention import ( + _CausalBehavior, + _context_parallel_shard, + _ContextParallel, + _cp_options, + _disable_context_parallel_dispatcher, + _enable_context_parallel_dispatcher, + _is_causal_behavior, + _RotateMethod, + _templated_ring_attention, + context_parallel, + context_parallel_unshard, + set_rotate_method, +) +from ._context_parallel._load_balancer import ( + _HeadTailLoadBalancer, + _LoadBalancer, + _PerDocumentHeadTailLoadBalancer, + _PTRRLoadBalancer, +) + + +# TODO(fegin): add deprecation message once the final interfaces are concluded. +__all__ = [ + "_CausalBehavior", + "_context_parallel_shard", + "_ContextParallel", + "_cp_options", + "_disable_context_parallel_dispatcher", + "_enable_context_parallel_dispatcher", + "_is_causal_behavior", + "_RotateMethod", + "_templated_ring_attention", + "context_parallel", + "context_parallel_unshard", + "set_rotate_method", + "_HeadTailLoadBalancer", + "_LoadBalancer", + "_PerDocumentHeadTailLoadBalancer", + "_PTRRLoadBalancer", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..009255631796fc56b6607ae46b6ae4f91589e83b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__init__.py @@ -0,0 +1,46 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +# Context Parallel components + +from ._attention import ( + _CausalBehavior, + _context_parallel_shard, + _ContextParallel, + _cp_options, + _disable_context_parallel_dispatcher, + _enable_context_parallel_dispatcher, + _is_causal_behavior, + _RotateMethod, + context_parallel, + context_parallel_unshard, + set_rotate_method, +) +from ._cp_custom_ops import flex_cp_allgather +from ._load_balancer import ( + _HeadTailLoadBalancer, + _LoadBalancer, + _PerDocumentHeadTailLoadBalancer, + _PTRRLoadBalancer, +) + + +__all__ = [ + # From _attention + "_CausalBehavior", + "_context_parallel_shard", + "_ContextParallel", + "_cp_options", + "_disable_context_parallel_dispatcher", + "_enable_context_parallel_dispatcher", + "_is_causal_behavior", + "_RotateMethod", + "context_parallel", + "context_parallel_unshard", + "set_rotate_method", + # From _cp_custom_ops + "flex_cp_allgather", + # From _load_balancer + "_HeadTailLoadBalancer", + "_LoadBalancer", + "_PerDocumentHeadTailLoadBalancer", + "_PTRRLoadBalancer", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..60f1acfbd685552c9f30422bc306f1ed46dc17b9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_attention.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_attention.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..515848b8862ca25c6273a1b2ad1677372f61d682 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_attention.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_cp_custom_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_cp_custom_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f84b714ca0db9ab47c26b0ea413b8d152976c01d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_cp_custom_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_load_balancer.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_load_balancer.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c89a70e7de3ea4c233e21f36e4db54ea552a5986 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_load_balancer.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_sharding_rules.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_sharding_rules.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9d97ed3fee3891a4de1655cb84c71af88ab2c4de Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/__pycache__/_sharding_rules.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_attention.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_attention.py new file mode 100644 index 0000000000000000000000000000000000000000..a5e692489c74af4838a8603797a5581e51ea78eb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_attention.py @@ -0,0 +1,1728 @@ +import contextlib +import itertools +import logging +import types +from abc import ABC, abstractmethod +from collections.abc import Callable, Generator, Mapping, Sequence +from dataclasses import dataclass +from enum import auto, Enum +from functools import partial +from typing import Any, cast, Protocol, TypeAlias + +import torch +import torch.distributed as dist +import torch.distributed._functional_collectives as ft_c +import torch.distributed.distributed_c10d as c10d +import torch.nn as nn +import torch.nn.functional as F +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor import distribute_tensor, DTensor, Shard +from torch.distributed.tensor.parallel import ParallelStyle +from torch.nn.attention.flex_attention import ( + _mask_mod_signature, + BlockMask, + create_block_mask, +) +from torch.utils._pytree import tree_flatten, tree_unflatten + +from ._cp_custom_ops import flex_cp_allgather +from ._load_balancer import _create_default_load_balancer, _LoadBalancer + + +__all__ = [ + "_CausalBehavior", + "_context_parallel_shard", + "_ContextParallel", + "_cp_options", + "_disable_context_parallel_dispatcher", + "_enable_context_parallel_dispatcher", + "_is_causal_behavior", + "_RotateMethod", + "context_parallel", + "context_parallel_unshard", + "set_rotate_method", +] + + +class _CausalBehavior(Enum): + SKIP = None + NOT_IS_CAUSAL = False + IS_CAUSAL = True + + +class _RotateMethod(Enum): + ALL_TO_ALL = auto() + ALL_GATHER = auto() + + +aten = torch.ops.aten +logger = logging.getLogger(__name__) + + +class _DispatchMode(Enum): + MONKEY_PATCH = auto() + MODULE_WRAPPER = auto() + + +_dispatch_mode: _DispatchMode = _DispatchMode.MONKEY_PATCH + + +@dataclass +class _ContextParallelOptions: + # Whether to upcast parameters and gradients to float32 to avoid accumulation + # errors. It is likely this is always True, but we currently keep this variable + # for experimental purposes. + convert_to_f32: bool = True + enable_load_balance: bool = True + rotate_method: _RotateMethod = _RotateMethod.ALL_GATHER + + +_cp_options = _ContextParallelOptions() + + +def _is_causal_behavior( + rank: int, world_size: int, i: int, is_causal: bool +) -> _CausalBehavior: + """ + Calculate is_causal behavior for each KV block. The attention can either be + calculated in full, not at all or with the causal mask applied. + """ + if not is_causal: + return _CausalBehavior.NOT_IS_CAUSAL + + if i == 0: + return _CausalBehavior.IS_CAUSAL + + source_rank = (rank - i) % world_size + if source_rank < rank or _cp_options.enable_load_balance: + return _CausalBehavior.NOT_IS_CAUSAL + else: + return _CausalBehavior.SKIP + + +def _maybe_wait(tensor: torch.Tensor) -> torch.Tensor: + """ + When tracing the code, the result tensor is not an AsyncCollectiveTensor, + so we cannot call ``wait()``. + """ + if isinstance(tensor, ft_c.AsyncCollectiveTensor): + return tensor.wait() + return tensor + + +def _partial_update( + original: torch.Tensor, + new: torch.Tensor, + dim: int, + n_chunks: int, + idx: int, + add: bool, +) -> torch.Tensor: + """ + This API partially updates a chunk of ``original`` tensor. The ``original`` + tensor will be first chunked along ``dim`` dimension, then the ``idx`` chunk + will be updated with ``new``. If ``add`` is True, the chunk will be added + with ``new``, otherwise the chunk will be replaced by ``new``. + + The result is a tensor that is the same size as ``original``. + """ + chunks = list(original.chunk(n_chunks, dim=dim)) + if chunks[idx].shape != new.shape: + raise AssertionError((original.shape, new.shape, idx)) + if add: + chunks[idx] += new + else: + chunks[idx] = new + return torch.cat(chunks, dim=dim) + + +class _SDPAMerger: + """A class to help merge the local SDPA result.""" + + def __init__(self, convert_to_f32: bool, seq_dim: int): + self._seq_dim = seq_dim + self._out: torch.Tensor | None = None + self._lse: torch.Tensor | None = None + self._should_lse_squeeze = False + self._convert_to_f32 = convert_to_f32 + self._out_dtype = torch.float32 + self._lse_dtype = torch.float32 + + def _merge_one( + self, block_out: torch.Tensor, block_lse: torch.Tensor, partial: bool + ) -> None: + # The cuDNN backend preserves the last dimension for LSE. + # Apply unsqueeze only if the input does not already have + # the required dimensionality. + if len(block_lse.shape) < len(block_out.shape): + block_lse = block_lse.unsqueeze(dim=-1) + self._should_lse_squeeze = True + if len(block_lse.shape) != len(block_out.shape): + raise AssertionError + + if self._lse is None: + self._lse = block_lse + self._out = block_out + else: + ROUND_ROBIN_CYCLE = 2 + if self._lse is None: + raise AssertionError + if self._out is None: + raise AssertionError + lse = ( + self._lse.chunk(ROUND_ROBIN_CYCLE, dim=self._seq_dim)[1] + if partial + else self._lse + ) + out = ( + self._out.chunk(ROUND_ROBIN_CYCLE, dim=self._seq_dim)[1] + if partial + else self._out + ) + + # The algorithm from + # github.com/zhuzilin/ring-flash-attention/pull/34#issuecomment-2076126795 + # gives a relatively stable result. + out = out - F.sigmoid(block_lse - lse) * (out - block_out) + lse = lse - F.logsigmoid(lse - block_lse) + if partial: + self._lse = _partial_update( + self._lse, + lse, + dim=self._seq_dim, + n_chunks=ROUND_ROBIN_CYCLE, + idx=1, + add=False, + ) + self._out = _partial_update( + self._out, + out, + dim=self._seq_dim, + n_chunks=ROUND_ROBIN_CYCLE, + idx=1, + add=False, + ) + else: + self._lse = lse + self._out = out + + def step(self, out: torch.Tensor, lse: torch.Tensor, partial: bool) -> None: + self._out_dtype = out.dtype + self._lse_dtype = lse.dtype + + if self._convert_to_f32: + out = out.to(torch.float32) + lse = lse.to(torch.float32) + + self._merge_one(out, lse, partial) + + def results(self) -> tuple[torch.Tensor, torch.Tensor]: + if self._out is None: + raise AssertionError + if self._lse is None: + raise AssertionError + out = self._out.to(self._out_dtype) + if self._should_lse_squeeze: + lse = self._lse.squeeze(-1).to(self._lse_dtype) + else: + lse = self._lse.to(self._lse_dtype) + return out, lse + + +class _AttentionOp(Protocol): + def __call__( + self, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + **kwargs: object, + ) -> tuple[torch.Tensor, ...]: ... + + +class _RingRotater(ABC): + @abstractmethod + def __init__(self, pg: dist.ProcessGroup, seq_dim: int) -> None: ... + + @abstractmethod + def exchange_buffers(self, curr_buffer: torch.Tensor) -> None: ... + + @abstractmethod + def next_buffer(self) -> torch.Tensor: ... + + +class _AllToAllRotater(_RingRotater): + """Use all_to_all to send the kv to the next rank.""" + + def __init__(self, pg: dist.ProcessGroup, seq_dim: int) -> None: + self._pg = pg + self._seq_dim = seq_dim + self._buffer: torch.Tensor | None = None + + def exchange_buffers(self, curr_buffer: torch.Tensor) -> None: + curr_buffer = curr_buffer.contiguous() + size = dist.get_world_size(self._pg) + dsts = list(range(1, size)) + [0] + self._buffer = ft_c.permute_tensor(curr_buffer, dsts, self._pg) + + def next_buffer(self) -> torch.Tensor: + if self._buffer is None: + raise AssertionError + return _maybe_wait(self._buffer) + + +class _AllGatherRotater(_RingRotater): + """ + Allgather the kv and return only the required kv. + Only one communication will be done. + """ + + def __init__(self, pg: dist.ProcessGroup, seq_dim: int) -> None: + self._pg = pg + self._seq_dim = seq_dim + self._aggregated_buffer: torch.Tensor | None = None + self._idx = 0 + + def exchange_buffers(self, curr_buffer: torch.Tensor) -> None: + # We only need to perform allgather once. + self._idx += 1 + if self._aggregated_buffer is None: + self._aggregated_buffer = ft_c.all_gather_single( + curr_buffer.contiguous(), gather_dim=0, group=self._pg + ) + + def next_buffer(self) -> torch.Tensor: + rank = dist.get_rank(self._pg) + idx = rank - self._idx + + if self._aggregated_buffer is None: + raise AssertionError + self._aggregated_buffer = _maybe_wait(self._aggregated_buffer) + return self._aggregated_buffer.chunk(dist.get_world_size(self._pg))[idx] + + +def _create_rotater( + pg: dist.ProcessGroup, seq_dim: int, method: _RotateMethod | None = None +) -> _RingRotater: + if method is None: + method = _cp_options.rotate_method + + if method == _RotateMethod.ALL_TO_ALL: + return _AllToAllRotater(pg, seq_dim) + elif method == _RotateMethod.ALL_GATHER: + return _AllGatherRotater(pg, seq_dim) + else: + raise NotImplementedError(f"Unknown method {method}") + + +def _templated_ring_attention( + group: dist.ProcessGroup, + seq_dim: int, + op: _AttentionOp, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + is_causal: bool = False, + **kwargs: object, +) -> tuple[torch.Tensor, ...]: + """ + A generalized ring attention implementation that can support multiple attention ops. + + Note [Context parallelism load balance algorithm for causal masking] + ===================== + This explanation uses an example to illustrate the CP algorithm with causal + masking. + + Consider a scenario where the sequence length of q, k, and v is 4 (e.g., + q = (q0, q1, q2, q3)), and there are two ranks. For simplicity, we will discuss + only q and k, as v follows the same pattern as k. + + The diagram below represents a complete QK^T operation without parallelism. + The `****` entries indicate that the result is not required due to causal + masking (e.g., q0k1 is marked as `****`). + + +----+------------------------+ + | | k0 k1 k2 k3 | + +----+------------------------+ + | q0 | q0k0, ****, ****, **** | + | q1 | q1k0, q1k1, ****, **** | + | q2 | q2k0, q2k1, q2k2, **** | + | q3 | q3k0, q3k1, q3k2, q3k3 | + +----+------------------------+ + + ### No Load Balance: + + In this scenario, each rank owns a local chunk of q, k, and v, with each chunk + containing two elements. Rank0 is responsible for managing (q0, q1) and (k0, k1), + while rank1 manages (q2, q3) and (k2, k3). + + First Iteration: Both rank0 and rank1 perform SDPA with their local qkv pairs. + Causal masking is enabled as some results are not required (e.g., q0k1). + + Second Iteration: Local queries remain the same, but local kv pairs are exchanged. + Rank0 now has (q0, q1) and (k2, k3); rank1 has (q2, q3) and (k0, k1). Rank0 performs + no computation, while rank1 computes locally without causal masking since all results + (q2k0, q2k1, q3k0, q3k1) are needed. + + ### Round-robin Load Balance: + + In this setup, each rank owns two local chunks of q, k, and v, with each chunk + containing one element. Rank0 manages (q0, q3) and (k0, k3); Rank1 manages (q1, q2) + and (k1, k2). Although the local chunks are not consecutive, they are concatenated to + enable SDPA to be performed in a single call for each step. Consequently, the chunk() + function may be required to prepare the correct q, k, and v configurations. + + First Iteration: Both ranks perform SDPA with their local qkv pairs, similar to the + no-load-balance case. This iteration corresponds to the `if` of the + (`if, `elif`, `else`) in the implementation. + + Second Iteration: Rank0 now has (q0, q3) and (k1, k2); rank1 has (q1, q2) and + (k0, k3). For rank0, no computation is needed for q0. However, computations for + q3k1 and q3k2 are required, so only q3 is used for SDPA. This corresponds to the + `else` of the (`if`, `elif`, `else`) in the implementation. + For rank1, k3 is not needed for q1 and q2, so only k0 is used for SDPA. This + corresponds to the `elif` of (`if`, `elif`, `else`) in the implementation. + + Parameters + ---------- + op: + The attention op to use + *args: + additional args are passed to the op + **kwargs: + additional kwargs are passed to the op + + Returns + ------- + out: + The merged attention output + softmax_lse: + The logsumexp of the merged attention output + """ + if is_causal and (query.size(2) != key.size(2)): + raise NotImplementedError( + "is_causal requires the same query and context sequence lengths" + ) + if not is_causal and _cp_options.enable_load_balance: + raise RuntimeError("Load balancing requires `is_causal=True`.") + + if not isinstance(group, dist.ProcessGroup): + raise AssertionError("process group must be single dimension") + rank = dist.get_rank(group) + size = dist.get_world_size(group) + + next_kv = None + + # Without making key and value contiguous(), the loss curve is bad. + # TODO(fegin): figure out why this is a requirement since SDPA does not have + # this requirement. + key = key.contiguous() + value = value.contiguous() + + sdpa_merger = _SDPAMerger(_cp_options.convert_to_f32, seq_dim=seq_dim) + + saved_rest: list[Any] | None = None + out: torch.Tensor + logsumexp: torch.Tensor + + rotater = _create_rotater(group, 2) + + for i in range(size): + if i > 0: + # Wait for the kv from the (cp_rank - 1) rank. + next_kv = rotater.next_buffer() + key = next_kv[: key.numel()].reshape(key.shape) + value = next_kv[key.numel() :].reshape(value.shape) + + if i < (size - 1): + # Send the k, v to the next rank + next_kv = torch.cat([key.flatten(), value.flatten()]) + next_kv = rotater.exchange_buffers(next_kv) + + is_causal_behavior = _is_causal_behavior( + rank=rank, world_size=size, i=i, is_causal=is_causal + ) + + # For a detailed understanding of the load balancing algorithm, see + # Note [Context parallelism load balance algorithm for causal masking] + if is_causal_behavior == _CausalBehavior.SKIP: + # If i > rank and load balancing is not turned on. + continue + + if i == 0 or (not _cp_options.enable_load_balance or not is_causal): + # When local balance is enabled, we still need to do SDPA with + # the both local chunks of q, k, v for the first iteration. + q, k, v, partial = (query, key, value, False) + elif i <= rank: + # Round-robin load balancing case, and i <= rank. + # We need to do SDPA with only the first local chunk of k, v. + # Note that q, k, v each contains two local chunks. + ROUND_ROBIN_CYCLE = 2 + q, k, v, partial = ( + query, + key.chunk(ROUND_ROBIN_CYCLE, dim=2)[0], + value.chunk(ROUND_ROBIN_CYCLE, dim=2)[0], + False, + ) + else: + # Round-robin load balancing case, and i > rank. + # We need to do SDPA with only the second half of q, and update + # only the second part of logsumexp. So partial is True. + # Note that q, k, v each contains two chunks. + q, k, v, partial = query.chunk(2, dim=2)[1], key, value, True + + # See https://github.com/pytorch/pytorch/blob/release/2.4/aten/src/ATen/native/native_functions.yaml#L14695 + # for the SDPA kernel definitions. + out, logsumexp, *rest = op( + q, + k, + v, + is_causal=is_causal_behavior.value, + **kwargs, + ) + if saved_rest is None: + saved_rest = rest + sdpa_merger.step(out, logsumexp, partial) + + if saved_rest is None: + raise AssertionError("No SDPA op was executed in ring attention forward") + return *sdpa_merger.results(), *saved_rest + + +def _templated_ring_attention_backward( + group: dist.ProcessGroup, + seq_dim: int, + op: _AttentionOp, + grad_out: torch.Tensor, + grad_out_name: str, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + out: torch.Tensor, + logsumexp: torch.Tensor, + is_causal: bool, + **kwargs: Any, +) -> tuple[torch.Tensor, ...]: + """This API implements the backward pass of the ring attention.""" + if not is_causal and _cp_options.enable_load_balance: + raise RuntimeError("Load balancing requires `is_causal=True`.") + rank = dist.get_rank(group) + size = dist.get_world_size(group) + next_kv = None + next_grad_kv = None + rest: list[Any] + grad_query_, grad_key_, grad_value_ = None, None, None + + accum_dtype = torch.float32 if _cp_options.convert_to_f32 else query.dtype + grad_query = torch.zeros_like(query, dtype=accum_dtype) + grad_key = torch.zeros_like(key, dtype=accum_dtype) + grad_value = torch.zeros_like(value, dtype=accum_dtype) + + key = key.contiguous() + value = value.contiguous() + kv_rotater = _create_rotater(group, 2) + dkv_rotater = _create_rotater(group, 2, method=_RotateMethod.ALL_TO_ALL) + for i in range(size): + if i > 0: + # Wait for the kv from the (cp_rank - 1) rank. + buffer = kv_rotater.next_buffer() + pointer = 0 + key = buffer[pointer : pointer + key.numel()].reshape(key.shape) + pointer += key.numel() + value = buffer[pointer : pointer + value.numel()].reshape(value.shape) + pointer += value.numel() + + if i != size - 1: + # Send the kv to the next rank. + next_kv = torch.cat([key.flatten(), value.flatten()]) + kv_rotater.exchange_buffers(next_kv) + + is_causal_behavior = _is_causal_behavior( + rank=rank, world_size=size, i=i, is_causal=is_causal + ) + + if is_causal_behavior != _CausalBehavior.SKIP: + if i == 0 or (not _cp_options.enable_load_balance or not is_causal): + # We need to do SDPA with the full local q, k, v. + q, k, v, out_, dout, lse = (query, key, value, out, grad_out, logsumexp) + elif i <= rank: + # Round-robin load balancing case, and i <= rank. + # We need to do SDPA with only the first half of k, v. + # Note that q, k, v each contains two chunks. + q, k, v, out_, dout, lse = ( + query, + key.chunk(2, dim=seq_dim)[0], + value.chunk(2, dim=seq_dim)[0], + out, + grad_out, + logsumexp, + ) + else: + # Round-robin load balancing case, and i > rank. + # We need to do SDPA with only the second half of q. + # Note that q, k, v each contains two chunks. + q, k, v, out_, dout, lse = ( + query.chunk(2, dim=seq_dim)[1], + key, + value, + out.chunk(2, dim=seq_dim)[1], + grad_out.chunk(2, dim=seq_dim)[1], + # Need to make logsumexp contiguous, otherwise there will + # be numerical error. + logsumexp.chunk(2, dim=seq_dim)[1].contiguous(), + ) + + kwargs[grad_out_name] = dout + iter_kwargs = kwargs + if _cp_options.enable_load_balance and i > 0: + iter_kwargs = dict(kwargs) + if "max_q" in iter_kwargs: + iter_kwargs["max_q"] = q.shape[seq_dim] + if "max_k" in iter_kwargs: + iter_kwargs["max_k"] = k.shape[seq_dim] + # See https://github.com/pytorch/pytorch/blob/release/2.4/aten/src/ATen/native/native_functions.yaml#L14695 + # for the SDPA kernel definitions. + grad_query_, grad_key_, grad_value_, *rest = op( + query=q, + key=k, + value=v, + out=out_, + logsumexp=lse, + is_causal=is_causal_behavior.value, + **iter_kwargs, + ) + else: + grad_query_ = torch.zeros_like(query, dtype=accum_dtype) + grad_key_ = torch.zeros_like(key, dtype=accum_dtype) + grad_value_ = torch.zeros_like(value, dtype=accum_dtype) + + ROUND_ROBIN_CYCLE = 2 + if i == 0: + grad_key += grad_key_ + grad_value += grad_value_ + else: + pointer = 0 + # Wait for the kv gradient from (cp_rank - 1) rank. + next_grad_kv = dkv_rotater.next_buffer() + grad_key = next_grad_kv[pointer : pointer + grad_key.numel()].reshape( + grad_key.shape + ) + pointer += grad_key.numel() + grad_value = next_grad_kv[pointer : pointer + grad_value.numel()].reshape( + grad_value.shape + ) + + if i <= rank and _cp_options.enable_load_balance: + grad_key = _partial_update( + grad_key, + grad_key_, + dim=seq_dim, + n_chunks=ROUND_ROBIN_CYCLE, + idx=0, + add=True, + ) + grad_value = _partial_update( + grad_value, + grad_value_, + dim=seq_dim, + n_chunks=ROUND_ROBIN_CYCLE, + idx=0, + add=True, + ) + else: + grad_key += grad_key_ + grad_value += grad_value_ + + next_grad_kv = torch.cat([grad_key.flatten(), grad_value.flatten()]) + # Send the grad key and grad value to the next rank. + dkv_rotater.exchange_buffers(next_grad_kv) + + if i <= rank or not _cp_options.enable_load_balance: + grad_query += grad_query_ + else: + grad_query = _partial_update( + grad_query, + grad_query_, + dim=seq_dim, + n_chunks=ROUND_ROBIN_CYCLE, + idx=1, + add=True, + ) + + if grad_key_ is None: + raise AssertionError + if grad_value_ is None: + raise AssertionError + grad_query = grad_query.to(query.dtype) + next_grad_kv = dkv_rotater.next_buffer().to(key.dtype) + grad_key = next_grad_kv[: grad_key.numel()].reshape(grad_key.shape) + grad_value = next_grad_kv[grad_key.numel() :].reshape(grad_value.shape) + return ( + grad_query, + grad_key, + grad_value, + # pyrefly: ignore [unbound-name] + *rest, + ) + + +def _scaled_dot_product_ring_flash_attention( + mesh: DeviceMesh, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + dropout_p: float = 0.0, + is_causal: bool = False, + return_debug_mask: bool = False, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + if return_debug_mask: + raise NotImplementedError("return_debug_mask is not supported yet") + + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention( + group, + seq_dim, + aten._scaled_dot_product_flash_attention, + query=query, + key=key, + value=value, + is_causal=is_causal, + dropout_p=dropout_p, + scale=scale, + ) + + +def _scaled_dot_product_ring_efficient_attention( + mesh: DeviceMesh, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attn_bias: torch.Tensor | None = None, + compute_log_sumexp: bool = True, + dropout_p: float = 0.0, + is_causal: bool = False, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + if attn_bias is not None: + raise NotImplementedError("attn_bias is not supported yet") + + if not compute_log_sumexp: + # CP requires compute_log_sumexp to be True because it always merges LSE + compute_log_sumexp = True + + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention( + group, + seq_dim, + aten._scaled_dot_product_efficient_attention, + query=query, + key=key, + value=value, + is_causal=is_causal, + attn_bias=attn_bias, + dropout_p=dropout_p, + scale=scale, + compute_log_sumexp=compute_log_sumexp, + ) + + +def _scaled_dot_product_ring_cudnn_attention( + mesh: DeviceMesh, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attn_bias: torch.Tensor | None = None, + compute_log_sumexp: bool = True, + dropout_p: float = 0.0, + is_causal: bool = False, + return_debug_mask: bool = False, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + if attn_bias is not None: + raise NotImplementedError("attn_bias is not supported yet") + + if not compute_log_sumexp: + # CP requires compute_log_sumexp to be True because it always merges LSE + compute_log_sumexp = True + + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention( + group, + seq_dim, + aten._scaled_dot_product_cudnn_attention, + query=query, + key=key, + value=value, + attn_bias=attn_bias, + compute_log_sumexp=compute_log_sumexp, + dropout_p=dropout_p, + is_causal=is_causal, + return_debug_mask=return_debug_mask, + scale=scale, + ) + + +def _scaled_dot_product_ring_flash_attention_backward( + mesh: DeviceMesh, + grad_out: torch.Tensor, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + out: torch.Tensor, + logsumexp: torch.Tensor, + cum_seq_q: torch.Tensor, + cum_seq_k: torch.Tensor, + max_q: int, + max_k: int, + dropout_p: float, + is_causal: bool, + philox_seed: torch.Tensor, + philox_offset: torch.Tensor, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention_backward( + group, + seq_dim, + aten._scaled_dot_product_flash_attention_backward.default, + grad_out=grad_out, + grad_out_name="grad_out", + query=query, + key=key, + value=value, + out=out, + logsumexp=logsumexp, + is_causal=is_causal, + cum_seq_q=cum_seq_q, + cum_seq_k=cum_seq_k, + max_q=max_q, + max_k=max_k, + dropout_p=dropout_p, + philox_seed=philox_seed, + philox_offset=philox_offset, + scale=scale, + ) + + +def _scaled_dot_product_ring_efficient_attention_backward( + mesh: DeviceMesh, + grad_out: torch.Tensor, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + bias: torch.Tensor, + out: torch.Tensor, + logsumexp: torch.Tensor, + philox_seed: torch.Tensor, + philox_offset: torch.Tensor, + dropout_p: float, + grad_input_mask: tuple[bool, ...], + is_causal: bool = False, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention_backward( + group, + seq_dim, + aten._scaled_dot_product_efficient_attention_backward.default, + grad_out=grad_out, + grad_out_name="grad_out_", + query=query, + key=key, + value=value, + attn_bias=bias, + out=out, + logsumexp=logsumexp, + philox_seed=philox_seed, + philox_offset=philox_offset, + dropout_p=dropout_p, + grad_input_mask=grad_input_mask, + is_causal=is_causal, + scale=scale, + ) + + +def _scaled_dot_product_ring_cudnn_attention_backward( + mesh: DeviceMesh, + grad_out: torch.Tensor, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + out: torch.Tensor, + logsumexp: torch.Tensor, + philox_seed: torch.Tensor, + philox_offset: torch.Tensor, + attn_bias: torch.Tensor, + cum_seq_q: torch.Tensor, + cum_seq_k: torch.Tensor, + max_q: int, + max_k: int, + dropout_p: float, + is_causal: bool, + *, + scale: float | None = None, +) -> tuple[torch.Tensor, ...]: + # TODO: remove this hardcoding + seq_dim = 2 + group = mesh.get_group() + return _templated_ring_attention_backward( + group, + seq_dim, + aten._scaled_dot_product_cudnn_attention_backward.default, + grad_out=grad_out, + grad_out_name="grad_out", + query=query, + key=key, + value=value, + out=out, + logsumexp=logsumexp, + philox_seed=philox_seed, + philox_offset=philox_offset, + attn_bias=attn_bias, + cum_seq_q=cum_seq_q, + cum_seq_k=cum_seq_k, + max_q=max_q, + max_k=max_k, + dropout_p=dropout_p, + is_causal=is_causal, + scale=scale, + ) + + +def _sdpa_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + # extract local tensor and sharding infos to a OpInfo + op_info = DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + logger.debug("Dispatching op_call: %s", op_info.schema or op_call) + + # sharding propagation + # TODO: remove the context parallel strategy from the default propagation + # rule. Either figure out how to dynamically enable it or just don't call + # propagate. + DTensor._op_dispatcher.sharding_propagator.propagate(op_info) + output_sharding = op_info.output_sharding + if output_sharding is None: + raise AssertionError("output sharding should not be None") + if output_sharding.needs_redistribute: + raise AssertionError("inputs need to be redistributed") + + call_maps: dict[torch._ops.OpOverload, Callable] = { + aten._scaled_dot_product_flash_attention.default: _scaled_dot_product_ring_flash_attention, + aten._scaled_dot_product_efficient_attention.default: _scaled_dot_product_ring_efficient_attention, + aten._scaled_dot_product_cudnn_attention.default: _scaled_dot_product_ring_cudnn_attention, + aten._scaled_dot_product_flash_attention_backward.default: _scaled_dot_product_ring_flash_attention_backward, + aten._scaled_dot_product_efficient_attention_backward.default: _scaled_dot_product_ring_efficient_attention_backward, + aten._scaled_dot_product_cudnn_attention_backward.default: _scaled_dot_product_ring_cudnn_attention_backward, + } + if op_call in call_maps: + local_results = call_maps[op_call]( + op_info.compute_mesh, + *op_info.local_args, # type: ignore[arg-type] + **op_info.local_kwargs, # type: ignore[arg-type] + ) + else: + raise NotImplementedError( + "CP only supports flash attention and memory efficient attention now." + ) + + return DTensor._op_dispatcher.wrap(local_results, output_sharding.output_spec) + + +custom_ops = { + aten._scaled_dot_product_flash_attention.default: _sdpa_handler, + aten._scaled_dot_product_flash_attention_backward.default: _sdpa_handler, + aten._scaled_dot_product_efficient_attention.default: _sdpa_handler, + aten._scaled_dot_product_efficient_attention_backward.default: _sdpa_handler, + aten._scaled_dot_product_cudnn_attention.default: _sdpa_handler, + aten._scaled_dot_product_cudnn_attention_backward.default: _sdpa_handler, +} +existing_custom_ops = DTensor._op_dispatcher._custom_op_handlers + + +ArgsType = tuple[Any, ...] +KwargsType = dict[str, Any] +InputFnType = Callable[[nn.Module | None, ArgsType, KwargsType, DeviceMesh], Any] +OutputFnType = Callable[[nn.Module | None, Any, Any, DeviceMesh], Any] + +_replaced_functions: dict[Callable, tuple[str, Callable]] = {} + + +def _distribute_function( + fn: Callable, + fn_module: types.ModuleType, + device_mesh: DeviceMesh, + input_fn: InputFnType, + output_fn: OutputFnType, +) -> None: + """ + A helper function to replace a function with a distributed version by + using the monkey patching approach. + + This function is for the CP internal usage only. + """ + + def wrapper( + target_fn: Callable, input_fn: InputFnType, output_fn: OutputFnType + ) -> Callable: + def inner_fn(*args: ArgsType, **kwargs: KwargsType) -> Any: + args, kwargs = input_fn(None, args, kwargs, device_mesh) + outputs = target_fn(*args, **kwargs) + return output_fn(None, (args, kwargs), outputs, device_mesh) + + return inner_fn + + global _replaced_functions + + if fn in _replaced_functions: + return + + wrapper_fn = wrapper(fn, input_fn, output_fn) + setattr(fn_module, fn.__name__, wrapper_fn) + _replaced_functions[wrapper_fn] = (fn.__name__, fn) + + +def _restore_function(fn: Callable, fn_module: types.ModuleType) -> None: + """Restore the function that is replaced by _distribute_function.""" + if fn not in _replaced_functions: + return + + original_name, original_fn = _replaced_functions[fn] + setattr(fn_module, original_name, original_fn) + + +def _enable_cp_dtensor_dispatcher() -> None: + """Enables DTensor dispatcher to dispatch SDPA to CP.""" + # Enable custom op handlers for CP + DTensor._op_dispatcher._custom_op_handlers = { + **existing_custom_ops, + **custom_ops, + } + # Register CP-specific sharding rules + from ._sharding_rules import register_cp_sharding_rules + + register_cp_sharding_rules() + + +def _disable_cp_dtensor_dispatcher() -> None: + """Disables DTensor dispatcher to dispatch SDPA to CP.""" + # Restore original custom op handlers + DTensor._op_dispatcher._custom_op_handlers = existing_custom_ops + + # TODO: unregister_cp_sharding_rules(clear_the_cache=True) will cause + # all DTensor sharding propagation cache being invalidated. It is not + # easy to achieve selectively invalidating lru cache without rewriting + # the sharding propagation wrapper. + + from ._sharding_rules import unregister_cp_sharding_rules + + unregister_cp_sharding_rules(clear_the_cache=False) + + +def _enable_context_parallel_dispatcher_impl(seq_dim: int, mesh: DeviceMesh) -> None: + sdpa_cp = _ContextParallel( + seq_dim=seq_dim, + attention_type=_ContextParallel.AttentionType.SDPA, + ) + + if _dispatch_mode == _DispatchMode.MONKEY_PATCH: + _distribute_function( + F.scaled_dot_product_attention, + F, + mesh, + sdpa_cp.sdpa_input_fn, + sdpa_cp.sdpa_output_fn, + ) + _enable_cp_dtensor_dispatcher() + elif _dispatch_mode == _DispatchMode.MODULE_WRAPPER: + _enable_cp_dtensor_dispatcher() + else: + raise ValueError(f"Unknown dispatch mode: {_dispatch_mode}") + + +def _disable_context_parallel_dispatcher_impl() -> None: + if _dispatch_mode == _DispatchMode.MONKEY_PATCH: + _restore_function(F.scaled_dot_product_attention, F) + elif _dispatch_mode == _DispatchMode.MODULE_WRAPPER: + pass + else: + raise NotImplementedError(f"Unknown dispatch mode: {_dispatch_mode}") + + _disable_cp_dtensor_dispatcher() + + +_compiled_create_block_mask = None + + +def _context_parallel_buffers( + mesh: DeviceMesh, + buffers: list[torch.Tensor | BlockMask], + buffer_seq_dims: list[int], + load_balancer: _LoadBalancer | None = None, +) -> list[torch.Tensor | BlockMask]: + """ + Shard the buffers along the sequence dimensions according to CP rules. + Args: + mesh (:class:`DeviceMesh`): the device mesh for the context parallelism. + buffers (List[torch.Tensor]): the buffers to be sharded. + seq_dims (List[int]): the sequence dimensions of ``buffers``. This list + must have the same length as ``buffers``. + load_balancer (Optional[:class:`_LoadBalancer`]): an optional `_LoadBalancer` + object. If this argument is `None`, it means the `buffers` need no + rearrangement before being sharded. If this argument is a `_LoadBalancer` + object, call its `_generate_indices(restore=False)` to generate the + rearrangement indices such that each shard of `buffer[rearrange_idx]` is + well-balanced (i.e., having close sparsities). + + Returns: + List[torch.Tensor]: the sharded buffers. + + Note: + For `_context_parallel_shard` we require a non-None `load_balancer` object to be + explicitly passed if load-balancing is needed. + """ + # generate the index tensor for rearranging the buffer if a load-balance + # is available + load_balance_indices = load_balancer._generate_indices() if load_balancer else None + if not (load_balance_indices is None or load_balance_indices.ndim == 2): + raise AssertionError( + "load balance index expects shape (1, seq_len) or (B, seq_len) " + f"but got {load_balance_indices.shape}." + ) + + new_buffers = [] + sharded_buffer: torch.Tensor | BlockMask + for buffer, seq_dim in zip(buffers, buffer_seq_dims): + if isinstance(buffer, torch.Tensor): + # NOTE: assuming batch dim is 0 + + if load_balance_indices is not None: + # TODO: we should explicitly ask users to unsqueeze the batch dim. + # But this is a BC breaking ask. + # However, what we have done today is also not very safe. + idx_batch_size = load_balance_indices.size(0) + data_batch_size = buffer.size(0) if seq_dim > 0 else 1 + + if idx_batch_size != 1 and idx_batch_size != data_batch_size: + raise ValueError( + "Cannot rearrange buffer: " + f"load_balance_indices has shape {load_balance_indices.shape}, " + f"but buffer has shape {buffer.shape}." + ) + + if seq_dim == 0: + # buffer has shape [seq_len] or [seq_len, ...] + # Just use the first (and only) batch of indices + buffer = torch.index_select( + buffer, dim=0, index=load_balance_indices[0] + ) + else: + indices = load_balance_indices + if idx_batch_size == 1: + size = [data_batch_size] + list(indices.size())[1:] + indices = indices.expand(*size) + + # load_balance_indices that has shape [B, seq_len] where: + # - dim 0 corresponds to buffer dim 0 (batch) + # - dim 1 corresponds to buffer dim seq_dim + # Need to insert dimensions for all dims between 0 and seq_dim, + # and all dims after seq_dim. + + # Insert dimensions between batch (dim 0) and seq_dim + for i in range(1, seq_dim): + indices = indices.unsqueeze(i) + + # Insert dimensions after seq_dim + for _ in range(seq_dim + 1, buffer.ndim): + indices = indices.unsqueeze(-1) + + # Expand to match buffer's shape + indices = indices.expand(buffer.shape) + + buffer = torch.gather(buffer, dim=seq_dim, index=indices) + + # use DTensor to shard the buffer on sequence dimension, + # retain the local tensor + sharded_buffer = distribute_tensor( + buffer, mesh, [Shard(seq_dim)], src_data_rank=None + ).to_local() + elif isinstance(buffer, BlockMask): + sharded_buffer = _create_cp_block_mask( + mask_mod=buffer.mask_mod, + B=buffer.kv_num_blocks.shape[0], + H=buffer.kv_num_blocks.shape[1], + Q_LEN=buffer.seq_lengths[0], + KV_LEN=buffer.seq_lengths[1], + device_mesh=mesh, + load_balancer=load_balancer, + ) + else: + raise ValueError(f"Unknown buffer type: {type(buffer)}") + + new_buffers.append(sharded_buffer) + + return new_buffers + + +def _create_cp_block_mask( + mask_mod: _mask_mod_signature, + B: int, + H: int, + Q_LEN: int, + KV_LEN: int, + device_mesh: DeviceMesh, + load_balancer: _LoadBalancer | None = None, +) -> BlockMask: + """ + Creates a specialized BlockMask for Context Parallel FlexAttention. + + This function creates a BlockMask that enables computation of attention results + for sharded Q attending to global KV. The mask appropriately handles the query + index offset required when each rank operates on a shard of the query sequence + while accessing the full key-value sequence. + + The function internally rewrites the provided mask_mod function to translate local + query indices to global query indices, ensuring that the masking logic is applied + correctly across the distributed computation. + + Args: + mask_mod (Callable): Mask function that operates on global attention indices. + B (int): Batch size. + H (int): Number of query heads. + Q_LEN (int): Global sequence length of the query. + KV_LEN (int): Global sequence length of the key/value. + device_mesh (DeviceMesh): Device mesh used for context parallelism. + load_balancer (Optional[:class:`_LoadBalancer`]): The load-balancer used to rearrange + QKV before sharding. This will be used to modify the block_mask generated. + + Returns: + BlockMask: A block mask configured for the local query shard that can be used + with flex_attention() for the given cp_mesh. + + Raises: + NotImplementedError: If Q_LEN is not divisible by (CP world size * BLOCK_SIZE). + + Warning: + Currently requires Q_LEN to be divisible by CP mesh world size * BLOCK_SIZE + (BLOCK_SIZE defaults to 128). This constraint exists because the BlockMask + must handle both padding and offsets correctly. For example, if Q_LEN is 384, + CP world size is 2, and BLOCK_SIZE is 128, the local Q_LEN would be 192. In + such cases, both rank0 and rank1 would have paddings in their local BlockMasks. + Support for padding in this scenario is planned for future work. + + """ + + from torch.nn.attention.flex_attention import _DEFAULT_SPARSE_BLOCK_SIZE + + if Q_LEN % (device_mesh.size() * _DEFAULT_SPARSE_BLOCK_SIZE) != 0: + raise NotImplementedError( + f"Q_LEN {Q_LEN} is not divisible by CP mesh world size {device_mesh.size()} * " + f"BLOCK_SIZE {_DEFAULT_SPARSE_BLOCK_SIZE}. This is not supported yet. " + ) + + global _compiled_create_block_mask + if _compiled_create_block_mask is None: + _compiled_create_block_mask = torch.compile( + create_block_mask, dynamic=False, fullgraph=True + ) + compiled_create_block_mask = _compiled_create_block_mask + + def _rewrite_mask_mod( + mask_mod: _mask_mod_signature, + rank: int, + block_size: int, + local_q_size: int, + qkv_rearrange_indices: torch.Tensor | None = None, + ) -> _mask_mod_signature: + if not (qkv_rearrange_indices is None or qkv_rearrange_indices.ndim == 2): + raise AssertionError( + "load balance index expects shape (1, seq_len) or (B, seq_len) " + f"but got {qkv_rearrange_indices.shape}." + ) + + def qkv_idx_restore( + b: torch.Tensor, idx_post_rearrange: torch.Tensor + ) -> torch.Tensor: + if qkv_rearrange_indices is not None: + if ( + qkv_rearrange_indices.size(0) == 1 + ): # identical load-balance in batch + idx_pre_rearrange = qkv_rearrange_indices[0][idx_post_rearrange] + else: + idx_pre_rearrange = qkv_rearrange_indices[b][idx_post_rearrange] + else: + idx_pre_rearrange = idx_post_rearrange + + return idx_pre_rearrange + + def local_q_idx_to_q_idx(local_q_idx: torch.Tensor) -> torch.Tensor: + # calculate local block_idx and block_offset + local_blk_idx, local_blk_offset = ( + local_q_idx // block_size, + local_q_idx % block_size, + ) + # NOTE: load balancing is not used + local_num_blocks = local_q_size // block_size + blk_idx = local_num_blocks * rank + local_blk_idx + return blk_idx * block_size + local_blk_offset + + return lambda b, h, q_idx, kv_idx: mask_mod( + b, + h, + qkv_idx_restore(b, local_q_idx_to_q_idx(q_idx)), + qkv_idx_restore(b, kv_idx), + ) + + cp_rank = device_mesh.get_local_rank() + cp_group_size = device_mesh.size() + load_balancer = load_balancer or _create_default_load_balancer( + Q_LEN, cp_group_size, device_mesh.device_type + ) + Q_SHARD_LEN = Q_LEN // cp_group_size + block_size = _DEFAULT_SPARSE_BLOCK_SIZE + + rearrange_indices = ( + load_balancer._generate_indices(restore=False) if load_balancer else None + ) + block_mask = compiled_create_block_mask( + _rewrite_mask_mod( + mask_mod, + cp_rank, + block_size, + Q_SHARD_LEN, + qkv_rearrange_indices=rearrange_indices, + ), + B, + H, + Q_SHARD_LEN, + KV_LEN, + device=device_mesh.device_type, + BLOCK_SIZE=(block_size, block_size), + ) + return block_mask + + +##################### +# Experimental APIs +##################### + + +class _ContextParallel(ParallelStyle): + class AttentionType(Enum): + FLEX = "flex_attention" + SDPA = "scaled_dot_product_attention" + + def __init__( + self, + seq_dim: int, + attention_type: AttentionType, + ) -> None: + super().__init__() + self.seq_dim = seq_dim + self.attention_type = attention_type + + def _apply(self, module: nn.Module, mesh: DeviceMesh) -> nn.Module: + if self.attention_type == self.AttentionType.FLEX: + module.register_forward_pre_hook( + partial(self.flex_input_fn, mesh=mesh), with_kwargs=True + ) + return module + elif self.attention_type == self.AttentionType.SDPA: + module.register_forward_pre_hook( + partial(self.sdpa_input_fn, mesh=mesh), + with_kwargs=True, + ) + module.register_forward_hook(partial(self.sdpa_output_fn, mesh=mesh)) + return module + else: + raise ValueError(f"Unknown attention type: {self.attention_type}") + + def flex_input_fn( + self, module: nn.Module | None, args: Any, kwargs: Any, mesh: DeviceMesh + ) -> Any: + # We don't care about other args, and these argument order must be consistent + # with the signature of flex_attention. + expected_arg_names = ("query", "key", "value") + args_list = list(args) + for idx, name in enumerate(expected_arg_names): + if idx >= len(args): + args_list.append(kwargs.pop(name, None)) + + query, key, value = args_list[: len(expected_arg_names)] + if not isinstance(query, torch.Tensor): + raise AssertionError + if not isinstance(key, torch.Tensor): + raise AssertionError + if not isinstance(value, torch.Tensor): + raise AssertionError + + key = key.contiguous() + value = value.contiguous() + + global_key, global_value = flex_cp_allgather( + key, value, self.seq_dim, c10d._get_process_group_name(mesh.get_group()) + ) + args_list[1] = global_key + args_list[2] = global_value + + for idx in range(len(args), len(expected_arg_names)): + kwargs[expected_arg_names[idx]] = args_list[idx] + args_list = args_list[: len(args)] + return tuple(args_list), kwargs + + def sdpa_input_fn( + self, + module: nn.Module | None, + args: tuple[Any, ...], + kwargs: dict[str, Any], + mesh: DeviceMesh, + ) -> tuple[tuple[Any, ...], dict[str, Any]]: + placement = [Shard(self.seq_dim)] + all_args = [] + + for arg in itertools.chain(args, kwargs.values()): + if isinstance(arg, torch.Tensor): + if isinstance(arg, DTensor): + if arg._spec.placements != placement: + raise AssertionError + else: + arg = DTensor.from_local(arg, mesh, placement, run_check=False) + + all_args.append(arg) + + new_args = tuple(all_args[0 : len(args)]) + new_kwargs = dict(zip(kwargs.keys(), all_args[len(args) :])) + return new_args, new_kwargs + + def sdpa_output_fn( + self, module: nn.Module | None, inputs: Any, outputs: Any, mesh: DeviceMesh + ) -> Any: + new_outputs = [] + for output in [outputs] if isinstance(outputs, torch.Tensor) else outputs: + output = output.to_local() if isinstance(output, DTensor) else output + new_outputs.append(output) + + if isinstance(outputs, torch.Tensor): + return new_outputs[0] + + return tuple(new_outputs) + + +CPBuffer: TypeAlias = torch.Tensor | BlockMask +CPBufferContainer: TypeAlias = Sequence[CPBuffer] | Mapping[str, CPBuffer] +CPBufferSeqDims: TypeAlias = Sequence[int] | Mapping[str, int] + + +def _context_parallel_shard( + mesh: DeviceMesh, + buffers: CPBufferContainer, + seq_dims: CPBufferSeqDims, + load_balancer: _LoadBalancer | None = None, +) -> list[torch.Tensor | BlockMask]: + """ + Shard the buffers along the specified sequence dimensions (`seq_dims`), so that each + rank retains only its corresponding shard according to the provided `mesh`. If a + `load_balancer` is provided, the buffers will be rearranged by the load balancer + before sharding to improve load balance. Buffers can be either tensors or `BlockMask` + objects. If a buffer is a `BlockMask`, its sharding dimension is determined by the + `BlockMask` implementation, and the corresponding `seq_dim` is ignored. + + Note: + For `_context_parallel_shard`, a non-None `load_balancer` must be explicitly passed + if load balancing is required. + + Args: + mesh (DeviceMesh): The device mesh used for context parallelism. + buffers (List[torch.Tensor | BlockMask]): Buffers whose usage depends on the sequence + dimension. Examples include input batches, labels, and positional embedding buffers. + These buffers must be sharded along the sequence dimension to ensure correctness. + seq_dims (List[int]): The sequence dimensions for each buffer in `buffers`. Must have + the same length as `buffers`. + load_balancer (Optional[_LoadBalancer]): An optional load balancer object. If provided, + it rearranges the buffers before sharding to achieve better load balance. If not + provided, no rearrangement is performed. + + Returns: + List[torch.Tensor | BlockMask]: The sharded buffers, each corresponding to the local + shard for the current rank. + """ + # TODO: these global variables are going to bite us someday. + # We will have to remove them soon. + # For the new API, we only support the module wrapper mode. + global _dispatch_mode + _dispatch_mode = _DispatchMode.MODULE_WRAPPER + global _cp_options + if load_balancer is not None: + _cp_options.enable_load_balance = True + else: + _cp_options.enable_load_balance = False + + if len(buffers) != len(seq_dims): + raise ValueError( + "`seq_dims` must have the same number of elements as `buffers`." + ) + + flat_buffers, spec = tree_flatten(buffers) + flat_seq_dims, _ = tree_flatten(seq_dims) + if len(flat_buffers) != len(flat_seq_dims): + raise ValueError("`seq_dims` must have the pytree structure as `buffers`.") + + if isinstance(flat_buffers[0], torch.Tensor): + device = flat_buffers[0].device + else: + device = flat_buffers[0].kv_num_blocks.device + for buffer in flat_buffers: + if isinstance(buffer, torch.Tensor): + if device != buffer.device: + raise AssertionError("All buffers must be on the same device") + else: + if device != buffer.kv_num_blocks.device: + raise AssertionError("All buffers must be on the same device") + + flat_sharded_buffers = _context_parallel_buffers( + mesh, flat_buffers, flat_seq_dims, load_balancer + ) + + return tree_unflatten(flat_sharded_buffers, spec) + + +def _enable_context_parallel_dispatcher() -> None: + """ + Enable the context parallel dispatcher. This API is experimental and subject to change. + """ + _enable_cp_dtensor_dispatcher() + + +def _disable_context_parallel_dispatcher() -> None: + """ + Disable the context parallel dispatcher. This API is experimental and subject to change. + """ + _disable_cp_dtensor_dispatcher() + + +##################################################### +# Current public APIs, but are also subject to change +##################################################### +@contextlib.contextmanager +@torch.no_grad() +def context_parallel( + mesh: DeviceMesh, + *, + buffers: list[torch.Tensor] | None = None, + buffer_seq_dims: list[int] | None = None, + no_restore_buffers: set[torch.Tensor] | None = None, +) -> Generator[None, None, None]: + """ + + ``context_parallel`` is an experimental API to enable context + parallelism (CP). This API performs two actions: 1) patch the SDPA + (``torch.nn.functional.scaled_dot_product_attention``) with the CP-enabled + one, 2) shard ``buffers`` along the sequence dimension and each rank will + preserve the corresponding shard according ``mesh``. + + Args: + mesh (:class:`DeviceMesh`): the device mesh for the context parallelism. + buffers (Optional[List[torch.Tensor]]): buffers that the usage depend + on the sequence dimension. Examples are input batch, labels and + positional embedding buffers. These buffers must be sharded along + the sequence dimension to ensure the accuracy. The sharding will + happen in-place, the buffer's shape will change within the context. + The buffers will be restored after the context finishes. + ``no_restore_buffers`` can be used to specify which buffers don't + need to be restored. Note that ``buffers`` should not contain any + nn.Parameter. + buffer_seq_dims (Optional[List[int]]): the sequence dimensions of ``buffers``. + no_restore_buffers (Optional[Set[torch.Tensor]]): buffers in these set + won't be restored after the context exits. This set must be a subset + of ``buffers``. If the buffers won't be used after the context exits, + these buffers can be put in this list to avoid extra restore time. + + .. warning:: + `torch.distributed.tensor.experimental.context_parallel` is a + prototype feature in PyTorch. The API is subject to change. + """ + # For the legacy API, we only support the monkey-patch mode. + # We will deprecate this API once the new API is widely used. + global _dispatch_mode + _dispatch_mode = _DispatchMode.MONKEY_PATCH + + buffers = [] if buffers is None else buffers + buffer_seq_dims = [] if buffer_seq_dims is None else buffer_seq_dims + no_restore_buffers = set() if no_restore_buffers is None else no_restore_buffers + + if len(buffers) != len(buffer_seq_dims): + raise ValueError( + "`seq_dims` must have the same number of elements as `buffers`." + ) + + for buffer in no_restore_buffers: + # Cannot use `if not buffer in buffers` which will incur tensor comparison. + if not any(b is buffer for b in buffers): + raise ValueError("`no_restore_buffers` must be a subset of `buffers`.") + + original_buffers = [None if b in no_restore_buffers else b.clone() for b in buffers] + + device = buffers[0].device + seq_length = buffers[0].shape[buffer_seq_dims[0]] + cp_world_size = mesh.size() + + # If `enable_load_balance` is True, the default Head-tail load balancer + # (:class:`_HeadTailLoadBalancer`) is used to rearrange the buffers before + # sharding. Otherwise, we don't do any load-balance rearrange by passing + # `None` to `_context_parallel_shard()`. + old_enable_load_balance = _cp_options.enable_load_balance + load_balancer = _create_default_load_balancer(seq_length, cp_world_size, device) + _cp_options.enable_load_balance = load_balancer is not None + shards = _context_parallel_buffers( + mesh, + cast(list[torch.Tensor | BlockMask], buffers), + buffer_seq_dims, + load_balancer, + ) + for buffer, shard in zip(buffers, shards): + if not isinstance(shard, torch.Tensor): + raise AssertionError("ContextParallel only supports Tensor") + shard = shard.clone() + buffer.resize_(shard.shape) + buffer.copy_(shard) + + _enable_context_parallel_dispatcher_impl(seq_dim=2, mesh=mesh) + try: + yield + finally: + _disable_context_parallel_dispatcher_impl() + _cp_options.enable_load_balance = old_enable_load_balance + + for buffer, original_buffer in zip(buffers, original_buffers, strict=True): + if original_buffer is not None: + buffer.resize_(original_buffer.shape) + buffer.copy_(original_buffer) + + +@torch.no_grad() +def context_parallel_unshard( + mesh: DeviceMesh, + buffers: list[torch.Tensor], + seq_dims: list[int], + load_balancer: _LoadBalancer | None = None, +) -> list[torch.Tensor]: + """ + Unshard the tensors (e.g., output) that are sharded due to context parallelism. + + Args: + mesh (:class:`DeviceMesh`): the device mesh for the context parallelism. + buffers (List[torch.Tensor]): the buffers to be unsharded. + seq_dims (List[int]): the sequence dimensions of ``buffers``. This list + must have the same length as ``buffers``. + load_balancer (Optional[:class:`_Loadbalancer`]): an optional `_LoadBalancer` + object. If this argument is `None`, it means the `buffers` were not + rearranged when being sharded and there's no need to put it back to order + after unsharding. If this argument is a `_LoadBalancer` object, call + its `_generate_indices(restore=True)` to generate the restore indices such + that `unsharded[restore_idx]` is the original buffer. + + Returns: + List[torch.Tensor]: the unsharded buffers. + + Note: + For `context_parallel_unshard` we require not-None `load_balancer` object be + explicitly passed if flex_attention() is to be used and load-balancing is needed. + This is different from the case of SDPA though we strongly suggest users follow + the same convention. + """ + device = buffers[0].device + cp_world_size = mesh.size() + seq_length = buffers[0].shape[seq_dims[0]] * cp_world_size + + # If users don't pass in a `load_balancer`: + # - if `enable_load_balance` is True, we use the default round-robin + # load balancer. + # - if `enable_load_balance` is False, we don't do any load balancing + # by passing in `None` as `restore_indices`. + load_balancer = load_balancer or _create_default_load_balancer( + seq_length, cp_world_size, device + ) + restore_indices = ( + load_balancer._generate_indices(restore=True) if load_balancer else None + ) + + if not (restore_indices is None or restore_indices.ndim == 2): + raise AssertionError( + "load balance restore index expects shape (1, seq_len) or (B, seq_len) " + f"but got {restore_indices.shape}." + ) + unsharded_buffers = [] + for b, dim in zip(buffers, seq_dims): + b = b.contiguous() + unsharded_b = _maybe_wait(ft_c.all_gather_single(b, dim, mesh)) + + if restore_indices is not None: + # NOTE: assuming batch dim is 0 + idx_batch_size = restore_indices.size(0) + data_batch_size = unsharded_b.size(0) + if idx_batch_size != 1 and idx_batch_size != data_batch_size: + raise ValueError( + "Cannot restore buffer: " + f"restore_indices has shape {restore_indices.shape}, " + f"but unsharded_b has shape {unsharded_b.shape}." + ) + + for i in range(data_batch_size): + index = ( + restore_indices[0] # identical load-balance in batch + if idx_batch_size == 1 + else restore_indices[i] + ) + unsharded_b_batch_i = torch.index_select( + unsharded_b[i], dim=dim - 1, index=index + ) + unsharded_b[i] = unsharded_b_batch_i + + unsharded_buffers.append(unsharded_b) + + return unsharded_buffers + + +def set_rotate_method(rotate_method: str) -> None: + """ + Context Parallel SDPA requires the rotation of kv shards. Users can call this + API to specify which rotation method to use. "alltoall" shuffles the kv shards + using all-to-all collective. While "allgather" gathers the kv shards using + all-gather collective after the first sub-SDPA computation. If this API has not + been called, the default rotate method is "allgather". + + Args: + rotate_method (str): the rotate method to use. Currently only supports + "allgather" and "alltoall". If a different string other than these two + is passed in, the function will raise an error. + + Returns: + None + """ + logger.info("Note that FlexAttention CP doesn't support alltoall yet.") + if rotate_method == "allgather": + _cp_options.rotate_method = _RotateMethod.ALL_GATHER + elif rotate_method == "alltoall": + _cp_options.rotate_method = _RotateMethod.ALL_TO_ALL + else: + raise NotImplementedError( + "Context Parallel does not support " + f"using {rotate_method} for kv shards rotation" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_cp_custom_ops.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_cp_custom_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..33ae73f8a8583a39915d1f71cba902175162e4bd --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_cp_custom_ops.py @@ -0,0 +1,88 @@ +from typing import Any + +import torch +import torch.distributed._functional_collectives as funcol +import torch.distributed.distributed_c10d as c10d + + +@torch.library.custom_op("cplib::flex_cp_allgather", mutates_args=()) +def flex_cp_allgather( + k: torch.Tensor, v: torch.Tensor, seq_dim: int, pg_name: c10d.GroupName +) -> tuple[torch.Tensor, torch.Tensor]: + k = k.contiguous() + v = v.contiguous() + k = funcol.all_gather_single(k, seq_dim, pg_name) + v = funcol.all_gather_single(v, seq_dim, pg_name) + if isinstance(k, funcol.AsyncCollectiveTensor): + k = k.wait() + if isinstance(v, funcol.AsyncCollectiveTensor): + v = v.wait() + return k, v + + +@flex_cp_allgather.register_fake +def _( + k: torch.Tensor, v: torch.Tensor, seq_dim: int, pg_name: c10d.GroupName +) -> tuple[torch.Tensor, torch.Tensor]: + shape_k = list(k.shape) + shape_v = list(v.shape) + shape_k[seq_dim] *= c10d._get_group_size_by_name(pg_name) + shape_v[seq_dim] *= c10d._get_group_size_by_name(pg_name) + new_k = torch.empty(shape_k, dtype=k.dtype, device=k.device) + new_v = torch.empty(shape_v, dtype=v.dtype, device=v.device) + return new_k, new_v + + +@torch.library.custom_op("cplib::flex_cp_allgather_backward", mutates_args=()) +def flex_cp_allgather_backward( + grad_full_k: torch.Tensor, + grad_full_v: torch.Tensor, + seq_dim: int, + pg_name: c10d.GroupName, +) -> tuple[torch.Tensor, torch.Tensor]: + grad_k = funcol.reduce_scatter_single(grad_full_k, "sum", seq_dim, pg_name) + if isinstance(grad_k, funcol.AsyncCollectiveTensor): + grad_k = grad_k.wait() + grad_v = funcol.reduce_scatter_single(grad_full_v, "sum", seq_dim, pg_name) + if isinstance(grad_v, funcol.AsyncCollectiveTensor): + grad_v = grad_v.wait() + + return grad_k, grad_v + + +@flex_cp_allgather_backward.register_fake +def _( + grad_full_k: torch.Tensor, + grad_full_v: torch.Tensor, + seq_dim: int, + pg_name: c10d.GroupName, +) -> tuple[torch.Tensor, torch.Tensor]: + shape_k = list(grad_full_k.shape) + shape_v = list(grad_full_v.shape) + shape_k[seq_dim] //= c10d._get_group_size_by_name(pg_name) + shape_v[seq_dim] //= c10d._get_group_size_by_name(pg_name) + new_grad_k = torch.empty( + shape_k, dtype=grad_full_k.dtype, device=grad_full_k.device + ) + new_grad_v = torch.empty( + shape_v, dtype=grad_full_v.dtype, device=grad_full_v.device + ) + return new_grad_k, new_grad_v + + +def _flex_cp_allgather_backward( + ctx: Any, grad_full_k: torch.Tensor, grad_full_v: torch.Tensor +) -> tuple[torch.Tensor, torch.Tensor, None, None]: + grad_k, grad_v = flex_cp_allgather_backward( + grad_full_k, grad_full_v, ctx.seq_dim, ctx.pg_name + ) + return grad_k, grad_v, None, None + + +def _flex_cp_setup_context(ctx: Any, inputs: Any, output: Any) -> None: + _, _, ctx.seq_dim, ctx.pg_name = inputs + + +flex_cp_allgather.register_autograd( + _flex_cp_allgather_backward, setup_context=_flex_cp_setup_context +) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_load_balancer.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_load_balancer.py new file mode 100644 index 0000000000000000000000000000000000000000..6af6583801ad244c07544cf8d37c54db99bd6a57 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_load_balancer.py @@ -0,0 +1,483 @@ +# this file contains the `_LoadBalancer` class and its family of implementation +# for different load-balancing strategies in tensor sharding. +import functools +from abc import ABC, abstractmethod + +import torch +from torch import Tensor +from torch.nn.attention.flex_attention import BlockMask + + +# make it private since it's still a prototype +class _LoadBalancer(ABC): + @abstractmethod + def _generate_indices(self, restore: bool = False) -> Tensor | None: + """ + Generate indices for load balancing. + Args: + restore (bool): + + Returns: + The generated indices of shape `(1, seq_len)` if the load-balancing is + identical within the batch, or `(batch_size, seq_len)` if the load-balancing + should vary within the batch. + + Warning: + For Multi-Head Attention, we require the masks over the head dimension are identical + (i.e. the return value of `_generate_indices()` does not have `heads` dimension). + + Example: + Here is the causal mask for attention where q_len == kv_len == 8: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 0, 0, 0, 0, 0] + Q_index [1, 1, 1, 1, 0, 0, 0, 0] + [1, 1, 1, 1, 1, 0, 0, 0] + [1, 1, 1, 1, 1, 1, 0, 0] + [1, 1, 1, 1, 1, 1, 1, 0] + [1, 1, 1, 1, 1, 1, 1, 1] + + This mask matrix also represents the computation required to compute + the masked Q @ K^T by: + - mask[i, j] == 1: the computation of Q[i, :] dot K[j, :] is required + - mask[i, j] == 0: the computation should be skipped + + Therefore the number of 1s in matrix represents the amount of computation + required. + + Assume we want to distribute this Q @ K^T computation to 2 devices, then + the matrix is also distributed as: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 0, 0, 0, 0, 0] rank 0 + [1, 1, 1, 1, 0, 0, 0, 0] + Q_index ------------------------ + [1, 1, 1, 1, 1, 0, 0, 0] + [1, 1, 1, 1, 1, 1, 0, 0] rank 1 + [1, 1, 1, 1, 1, 1, 1, 0] + [1, 1, 1, 1, 1, 1, 1, 1] + + An imbalance of computation is observed on these 2 ranks and this could make + rank 1 the straggler when performing Context Parallel. In order to balance + the computation, we need to rearrange the QKV tensors before sharding in such a + way that the result mask matrix is evenly distributed over devices and each + rank has the number of 1s as close as possible. + + This method defines the strategy of how to rearrange the QKV tensor for better + load-balance: + - when `restore == False`, this method returns an indices tensor `rearrange_idx` + such that Q[rearrange_idx] is the desired Q tensor after rearranging. + - when `restore == True`, this method returns an indices tensor `restore_idx` + such that Q[rearrange_idx][restore_idx] == Q, i.e. restoring the rearranged tensor + back to the original status before rearranging. + + Sharding contract: + After rearranging, Context Parallel shards `Q[rearrange_idx]` with + `Shard(seq_dim)`, which follows `torch.chunk` semantics: the sequence is + split into `world_size` contiguous, equal-sized chunks and chunk `r` is + assigned to rank `r`. Implementations must therefore lay out + `rearrange_idx` so that each such contiguous chunk is well-balanced -- + i.e. group each rank's positions together (rank-major), not interleaved + by any other axis such as document. + """ + + +class _HeadTailLoadBalancer(_LoadBalancer): + def __init__(self, seq_length: int, world_size: int, device: str | torch.device): + self.seq_length = seq_length + self.world_size = world_size + self.device = device + + def _generate_indices(self, restore: bool = False) -> Tensor: + """ + Generate head-and-tail load balancing indices or restore indices. + Args: + restore: + If True, generate restore indices that map head-and-tail rearranged + positions back to original positions. If False, generate load + balance indices that rearrange original positions to head-and-tail pattern. + + Returns: + The generated indices of shape `(1, seq_len)` because the load-balancing is + identical within the batch. + + Warning: + For Multi-Head Attention, we require the masks over the head dimension are identical + (i.e. the return value of `_generate_indices()` does not have `heads` dimension). + + Example: + Here is the causal mask for attention where q_len == kv_len == 8: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 0, 0, 0, 0, 0] + Q_index [1, 1, 1, 1, 0, 0, 0, 0] + [1, 1, 1, 1, 1, 0, 0, 0] + [1, 1, 1, 1, 1, 1, 0, 0] + [1, 1, 1, 1, 1, 1, 1, 0] + [1, 1, 1, 1, 1, 1, 1, 1] + + Head-tail load-balance strategy rearranges the Q tensor by combining + Q[0:k] (on seq dim) and Q[-k:] for rank 0, Q[k:2k] and Q[-2k:-k] for + rank 1, and so on. In python code it looks like: + + k = Q.size(0) // (2 * cp_world_size) + for rank in range(cp_world_size): + reordered_Q[rank * 2 * k : (rank + 1) * 2 * k] = torch.cat( + (Q[rank * k : (rank + 1) * k], Q[-(rank + 1) * k : -rank * k]) + ) + + This can also be done by tensor slicing. For the above example, the indices + tensor for slicing is: + slice_indices = Tensor([0, 7, 1, 6, 2, 5, 3, 4]) + + After reordering QKV using the `slice_indices`, the corresponding mask matrix + distributing over 2 devices becomes well-balanced: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 1, 1, 1, 1, 1] + [1, 1, 0, 0, 0, 0, 0, 0] rank 0 + [1, 1, 1, 1, 1, 1, 1, 0] + Q_index ------------------------ + [1, 1, 1, 0, 0, 0, 0, 0] + [1, 1, 1, 1, 1, 1, 0, 0] rank 1 + [1, 1, 1, 1, 0, 0, 0, 0] + [1, 1, 1, 1, 1, 0, 0, 0] + + To restore the reordering and putting the tensor back, slicing op can do the + trick with a `restore_indices` such that: + slice_indices[restore_indices] == Tensor([0, 1, 2, ...]) + + In this way, `reordered_Q[restore_indices]` will just be the original Q. + """ + seq_length = self.seq_length + world_size = self.world_size + if seq_length % (world_size * 2) != 0: + raise AssertionError + chunk_size = seq_length // (world_size * 2) + + # Split sequence into 2*world_size chunks, then pair chunk r with + # chunk (2*world_size - 1 - r) for each rank. + indices = torch.arange(seq_length, dtype=torch.int, device=self.device) + chunks = indices.view(world_size * 2, chunk_size) + head_idx = torch.arange(world_size, device=self.device) + tail_idx = 2 * world_size - 1 - head_idx + paired = torch.stack([chunks[head_idx], chunks[tail_idx]], dim=1) + all_indices_tensor = paired.reshape(-1) + + if restore: + all_indices_tensor = torch.argsort(all_indices_tensor) + + return all_indices_tensor.unsqueeze(0) # add batch dim + + +class _PerDocumentHeadTailLoadBalancer(_LoadBalancer): + def __init__( + self, + seq_length_per_doc: list[list[int]], + world_size: int, + device: str | torch.device, + ): + """ + `seq_length_per_doc` has size (B, seq_len) if the load-balancing should vary + within the batch. Otherwise `seq_length_per_doc` should have size (1, seq_len). + """ + self.seq_length_per_doc = seq_length_per_doc + self.world_size = world_size + self.device = device + + def _generate_indices(self, restore: bool = False) -> Tensor: + """ + Generate the per-document head-and-tail rearrange indices so that after rearranging + the input is load-balanced in per-document head-and-tail style. + + Args: + restore: + If True, generate restore indices that map per-document head-and-tail + rearranged positions back to original positions. If False, generate load + balance indices that rearrange original positions to per-document + head-and-tail pattern. + + Returns: + The generated indices of shape `(batch_size, seq_len)` if the load-balancing + should vary within the batch. Otherwise, it should have shape `(1, seq_len)`. + + Warning: + For Multi-Head Attention, we require the masks over the head dimension are identical + (i.e. `seq_length_per_doc` must have size (B, seq_len) or (1, seq_len)). + + Example: + Here is the document causal mask for attention where q_len == kv_len == 16: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0] + Q_index [0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1] + + The per-document head-and-tail load-balancer will apply head-and-tail + reordering within each document. After load-balancing for context-parallel + on 2 devices, the above mask matrix will look like this: + KV_index + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0] + Q_index [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1] + ------------------------------------------------ + [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0] + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0] + """ + return torch.stack( + [ + self._generate_indices_for_batch(seq_lengths, restore) + for seq_lengths in self.seq_length_per_doc + ] + ) + + def _generate_indices_for_batch(self, seq_length_per_doc, restore) -> Tensor: # type: ignore[no-untyped-def] + world_size = self.world_size + device = self.device + if not all( + seq_length % (2 * world_size) == 0 for seq_length in seq_length_per_doc + ): + raise AssertionError + + # For each document, split it into 2 * world_size equal chunks and pair + # chunk r with chunk (2 * world_size - 1 - r), so row r holds rank r's + # head+tail chunks (the same strategy as _HeadTailLoadBalancer). Stacking + # the per-document rows rank-wise (dim=1) and flattening lays the indices + # out rank-major, so a contiguous per-rank shard gives each rank a + # head+tail slice of every document. A document-major layout would instead + # let the cut fall mid-document, imbalancing work for mixed-length docs. + head_idx = torch.arange(world_size, device=device) + tail_idx = 2 * world_size - 1 - head_idx + per_doc_rank_chunks = [] + document_start_idx = 0 + for seq_length in seq_length_per_doc: + chunk_length = seq_length // (2 * world_size) + chunks = torch.arange( + document_start_idx, + document_start_idx + seq_length, + device=device, + ).view(2 * world_size, chunk_length) + # (world_size, 2, chunk_length) -> (world_size, 2 * chunk_length): + # row r is rank r's head chunk followed by its tail chunk. + paired = torch.stack([chunks[head_idx], chunks[tail_idx]], dim=1) + per_doc_rank_chunks.append(paired.reshape(world_size, -1)) + document_start_idx += seq_length + + # (world_size, seq_len / world_size) flattened rank-major. + indices_tensor = torch.cat(per_doc_rank_chunks, dim=1).reshape(-1) + if restore: + indices_tensor = torch.argsort(indices_tensor) + + return indices_tensor + + +class _PTRRLoadBalancer(_LoadBalancer): + """ + Processing-Time based Round-Robin (PTRR) load balancer. This load balancer should + only be used for flex_attention() since it leverages `BlockMask`. + """ + + def __init__( + self, + block_mask: BlockMask, + world_size: int, + ): + """ + `block_mask` must have shape (B, 1, seq_len, seq_len) or (1, 1, seq_len, seq_len). + """ + self.block_mask = block_mask + self.world_size = world_size + + @staticmethod + def ptrr_scheduling(process_time: Tensor, group_size: int) -> Tensor: + """ + Separate the tasks into `group_size` groups using PTRR scheduling. + process_time: + 1D tensor of size n, where n is the number of tasks. The value + is the process time of the task. Size `n` must be divisible by + `group_size`. + group_size: + the number of groups + + Returns: + tasks_in_group (list[list[int]]): + A collection of list[int] and each list should have size `n // group_size` + (`group_size` lists in total). Each element is an index in the input + `process_time` (i.e. [0, len(process_time) - 1]). + + Example: + process_time = [9, 14, 2, 20, 10, 15, 8, 14, 16, 19, 15, 3, 12, 1, 12, 10] + tasks_in_group = [ + [3, 12, 13, 14], # values = [1, 12, 12, 20], sum = 45 + [2, 4, 7, 9], # values = [2, 10, 14, 19], sum = 45 + [1, 8, 11, 15], # values = [14, 16, 3, 10], sum = 43 + [0, 5, 6, 10] # values = [9, 15, 8, 15], sum = 47 + ] + """ + if process_time.ndim != 1: + raise AssertionError + + num_tasks = process_time.size(0) + + if num_tasks % group_size != 0: + raise NotImplementedError( + f"num_tasks {num_tasks} must be divisible by group_size {group_size}" + ) + + device = process_time.device + _, sorted_indices_descending = torch.sort( + process_time, descending=True, stable=True + ) # if process time is tied, the order is preserved + sorted_indices_descending_reversed = torch.flip( + sorted_indices_descending.view(-1, group_size), dims=[1] + ).view(-1) + tasks_in_group = torch.where( + torch.arange(num_tasks, device=device) // group_size % 2 == 0, + sorted_indices_descending, + sorted_indices_descending_reversed, + ) + tasks_in_group = tasks_in_group.view(-1, group_size).transpose( + 0, 1 + ) # (group_size, n // group_size) + + # sort each group. This step should not have impact on correctness + # nor execution run time, but it helps users visualize the mask + tasks_in_group, _ = torch.sort(tasks_in_group, dim=1) + return tasks_in_group + + def _generate_indices(self, restore: bool = False) -> Tensor: + """ + Generate the PTRR reorder indices of shape `(1, seq_len)` or `(batch_size, seq_len)`. + + Args: + restore: + If True, generate restore indices that map Processing-Time based Round-Robin + (PTRR) rearranged positions back to original positions. If False, generate + load balance indices that rearrange original positions to PTRR pattern. + + Returns: + The generated indices of shape `(1, seq_len)` if the load-balancing is + identical within the batch (i.e. `BlockMask.shape[0] == 1`), or + `(batch_size, seq_len)` if the load-balancing should vary within the batch. + + Warning: + For Multi-Head Attention, we require the masks over the head dimension are identical + (i.e. `self.block_mask` must have shape (B, 1, seq_len, seq_len) or (1, 1, seq_len, seq_len)). + + Example: + Here is the document causal mask for attention where q_len == kv_len == 16 * BLOCK_SIZE + (each entry is a block): + KV_index + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 1 + [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 2 + [1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 3 + [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 4 + [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 1 + [0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 2 + [0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 3 + Q_index [0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 4 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0] -> row value = 5 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0] -> row value = 6 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0] -> row value = 7 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0] -> row value = 8 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0] -> row value = 1 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0] -> row value = 2 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0] -> row value = 3 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1] -> row value = 4 + + The reorder indices will be: [2, 3, 5, 6, 8, 11, 12, 13, 0, 1, 4, 7, 9, 10, 14, 15] and + the mask matrix will look like: + KV_index + [1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 3 + [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 4 + [0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 2 + [0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 3 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0] -> row value = 5 rank 0 (sum=28) + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0] -> row value = 8 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0] -> row value = 1 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0] -> row value = 2 + ------------------------------------------------ + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 1 + [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 2 + [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 1 + [0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0] -> row value = 4 + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0] -> row value = 6 rank 1 (sum=28) + [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0] -> row value = 7 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0] -> row value = 3 + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1] -> row value = 4 + """ + block_mask = self.block_mask + kv_num_blocks = block_mask.kv_num_blocks + full_kv_num_blocks = block_mask.full_kv_num_blocks + non_sparse_kv_num_blocks = ( + kv_num_blocks + full_kv_num_blocks + if full_kv_num_blocks is not None + else kv_num_blocks + ) + B, H, Q = non_sparse_kv_num_blocks.shape + # requirement: the masking is identical across heads (i.e. H == 1 in BlockMask) + non_sparse_kv_num_blocks = non_sparse_kv_num_blocks.view(-1, Q) # (B, Q_BLK) + + batch_ptrr = torch.vmap( + functools.partial( + _PTRRLoadBalancer.ptrr_scheduling, + group_size=self.world_size, + ) + ) + ptrr_indices = batch_ptrr( + non_sparse_kv_num_blocks + ) # (B, group_size, num_blks_in_group) + ptrr_indices = ptrr_indices.reshape(B, -1) # (B, num_blocks) + + # NOTE: only support the case where the qkv block size are equal + q_blk_size, kv_blk_size = block_mask.BLOCK_SIZE + if q_blk_size != kv_blk_size: + raise AssertionError("for now only support q_blk_size == kv_blk_size") + + indices = torch.arange( + q_blk_size * ptrr_indices.size(1), device=ptrr_indices.device + ).view(-1, q_blk_size) # (NUM_BLOCKS, BLOCK_SIZE) + indices = indices[ptrr_indices].view(B, -1) # (B, qkv_size) + + if restore: + # pyrefly: ignore[missing-argument] + indices = torch.vmap(torch.argsort)(indices) + + return indices + + +def _create_default_load_balancer( + seq_length: int, world_size: int, device: str | torch.device +) -> _LoadBalancer | None: + from ._attention import _cp_options + + if _cp_options.enable_load_balance and seq_length % (world_size * 2) == 0: + return _HeadTailLoadBalancer(seq_length, world_size, device) + else: + return None diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_sharding_rules.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_sharding_rules.py new file mode 100644 index 0000000000000000000000000000000000000000..5769342a2a890f6a2f13fcb88b929aaa54899da2 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_context_parallel/_sharding_rules.py @@ -0,0 +1,338 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +""" +Context Parallelism sharding rules for scaled_dot_product attention operators. + +The sharding rules for CP cannot be embedded by default because Shard(2) is not +a valid sharding for SDPA without CP enabled. This module provides utilities to +dynamically install Shard(2) sharding rules when CP is activated. +""" + +from contextlib import contextmanager + +import torch +from torch.distributed.tensor._dtensor_spec import TensorMeta +from torch.distributed.tensor._op_schema import ArgsType, KwargsType, RuntimeSchemaInfo +from torch.distributed.tensor._ops.single_dim_strategy import ( + _ShardingPlaceholder, + register_single_dim_strategy, +) +from torch.distributed.tensor.debug import ( + _clear_fast_path_sharding_prop_cache, + _clear_python_sharding_prop_cache, +) +from torch.distributed.tensor.placement_types import Placement, Replicate + + +aten = torch.ops.aten + +SEQ_DIM = 2 +SingleDimPlacementList = list[Placement | _ShardingPlaceholder | None] + + +@contextmanager +def _single_dim_strategy_context(op_overload, strategy_func, schema_info=None): + """Temporarily install a single-dim strategy for Context Parallelism.""" + from torch.distributed.tensor import DTensor + + propagator = DTensor._op_dispatcher.sharding_propagator + _origin_single_dim_strategy = None + _origin_op_strategy_schema = None + try: + if op_overload in propagator.op_single_dim_strategy_funcs: + _origin_single_dim_strategy = propagator.op_single_dim_strategy_funcs[ + op_overload + ] + if op_overload in propagator.op_to_schema_info: + _origin_op_strategy_schema = propagator.op_to_schema_info[op_overload] + + register_single_dim_strategy(op_overload, schema_info=schema_info)( + strategy_func + ) + yield (_origin_single_dim_strategy, _origin_op_strategy_schema) + finally: + if _origin_single_dim_strategy is None: + if op_overload in propagator.op_single_dim_strategy_funcs: + del propagator.op_single_dim_strategy_funcs[op_overload] + else: + propagator.op_single_dim_strategy_funcs[op_overload] = ( + _origin_single_dim_strategy + ) + + if _origin_op_strategy_schema is None: + if op_overload in propagator.op_to_schema_info: + del propagator.op_to_schema_info[op_overload] + else: + propagator.op_to_schema_info[op_overload] = _origin_op_strategy_schema + + # Ideally, we should clear the cache, but it is too expensive. + # _clear_python_sharding_prop_cache() + # _clear_fast_path_sharding_prop_cache() + + +def _cp_sharding() -> _ShardingPlaceholder: + return _ShardingPlaceholder(SEQ_DIM) + + +# ==================== Flash Attention Strategies ==================== + + +def _scaled_dot_product_flash_attention_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + return_debug_mask = len(args_schema) >= 6 and args_schema[5] + debug_attn_mask_sharding: Placement | _ShardingPlaceholder = ( + _cp_sharding() if return_debug_mask else Replicate() + ) + + return [ + [ + _cp_sharding(), # output + _cp_sharding(), # logsumexp + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + Replicate(), # rng_state + None, # unused + debug_attn_mask_sharding, + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + ] + ] + + +def _scaled_dot_product_flash_attention_backward_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + num_tensor_inputs = sum(isinstance(arg, TensorMeta) for arg in args_schema) + cp_strategy: SingleDimPlacementList = [ + _cp_sharding(), # grad_q + _cp_sharding(), # grad_k + _cp_sharding(), # grad_v + _cp_sharding(), # grad_output + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + _cp_sharding(), # output + _cp_sharding(), # logsumexp + ] + cp_strategy.extend([Replicate()] * (num_tensor_inputs - 6)) + return [cp_strategy] + + +# ==================== Efficient Attention Strategies ==================== + + +def _scaled_dot_product_efficient_attention_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + has_attn_bias = args_schema[3] is not None + + cp_strategy: SingleDimPlacementList = [ + _cp_sharding(), # output + _cp_sharding(), # logsumexp + None, # philox_seed + None, # philox_offset + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + ] + if has_attn_bias: + cp_strategy.append(Replicate()) + return [cp_strategy] + + +def _scaled_dot_product_efficient_attention_backward_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + has_attn_bias = args_schema[4] is not None + + cp_strategy: SingleDimPlacementList = [ + _cp_sharding(), # grad_q + _cp_sharding(), # grad_k + _cp_sharding(), # grad_v + _ShardingPlaceholder(1) if has_attn_bias else None, # grad_bias + _cp_sharding(), # grad_output + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + _cp_sharding(), # output + _cp_sharding(), # logsumexp + ] + if has_attn_bias: + cp_strategy.insert(8, _ShardingPlaceholder(1)) # attn_bias + cp_strategy.extend([Replicate(), Replicate()]) + return [cp_strategy] + + +# ==================== cuDNN Attention Strategies ==================== + + +def _scaled_dot_product_cudnn_attention_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + attn_bias_meta = args_schema[3] + compute_log_sumexp = args_schema[4] + return_debug_mask = len(args_schema) >= 8 and args_schema[7] + has_attn_bias = attn_bias_meta is not None + + logsumexp_sharding: Placement | _ShardingPlaceholder = ( + _cp_sharding() if compute_log_sumexp else Replicate() + ) + debug_attn_mask_sharding = _cp_sharding() if return_debug_mask else None + + cp_strategy: SingleDimPlacementList = [ + _cp_sharding(), # output + logsumexp_sharding, + None, # cum_seq_q + None, # cum_seq_k + None, # max_q + None, # max_k + None, # philox_seed + None, # philox_offset + debug_attn_mask_sharding, + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + ] + if has_attn_bias: + cp_strategy.append(Replicate()) + return [cp_strategy] + + +def _scaled_dot_product_cudnn_attention_backward_cp_single_dim_strategy( + op: torch._ops.OpOverload, + args_schema: ArgsType, + kwargs_schema: KwargsType, +) -> list[SingleDimPlacementList]: + if len(args_schema) < 15: + raise AssertionError(f"Expected at least 15 args, got {len(args_schema)}") + + for arg_index in range(6): + arg = args_schema[arg_index] + if not isinstance(arg, TensorMeta): + raise AssertionError(f"Expected TensorMeta, got {type(arg)}") + + philox_placements: list[Placement] = [] + for arg_index in (6, 7): + arg = args_schema[arg_index] + if isinstance(arg, TensorMeta): + philox_placements.append(Replicate()) + elif not isinstance(arg, torch.Tensor): + raise AssertionError(f"Expected TensorMeta or Tensor, got {type(arg)}") + + has_attn_bias = args_schema[8] is not None + if has_attn_bias and not isinstance(args_schema[8], (TensorMeta, torch.Tensor)): + raise AssertionError( + f"Expected TensorMeta or Tensor, got {type(args_schema[8])}" + ) + + cum_seq_placements: list[None] = [] + for arg_index in (9, 10): + arg = args_schema[arg_index] + if isinstance(arg, TensorMeta): + cum_seq_placements.append(None) + elif arg is None or isinstance(arg, torch.Tensor): + pass + else: + raise AssertionError(f"Expected TensorMeta or Tensor, got {type(arg)}") + + cp_sharding: SingleDimPlacementList = [ + _cp_sharding(), # grad_q + _cp_sharding(), # grad_k + _cp_sharding(), # grad_v + _cp_sharding(), # grad_out + _cp_sharding(), # q + _cp_sharding(), # k + _cp_sharding(), # v + _cp_sharding(), # output + _cp_sharding(), # logsumexp + ] + cp_sharding.extend(philox_placements) + if has_attn_bias and isinstance(args_schema[8], TensorMeta): + cp_sharding.append(_cp_sharding()) + cp_sharding.extend(cum_seq_placements) + + return [cp_sharding] + + +# Store context managers and original strategies +_cp_strategy_contexts = {} +_original_strategies = {} + + +def register_cp_sharding_rules(): + """Register Context Parallelism sharding rules for all scaled_dot_product ops.""" + global _cp_strategy_contexts, _original_strategies + + # If already registered, don't register again + if _cp_strategy_contexts: + return + + # Define ops and their corresponding CP strategy functions + cp_strategies = [ + ( + aten._scaled_dot_product_flash_attention.default, + _scaled_dot_product_flash_attention_cp_single_dim_strategy, + RuntimeSchemaInfo(5), + ), + ( + aten._scaled_dot_product_flash_attention_backward.default, + _scaled_dot_product_flash_attention_backward_cp_single_dim_strategy, + None, + ), + ( + aten._scaled_dot_product_efficient_attention.default, + _scaled_dot_product_efficient_attention_cp_single_dim_strategy, + RuntimeSchemaInfo(4), + ), + ( + aten._scaled_dot_product_efficient_attention_backward.default, + _scaled_dot_product_efficient_attention_backward_cp_single_dim_strategy, + None, + ), + ( + aten._scaled_dot_product_cudnn_attention.default, + _scaled_dot_product_cudnn_attention_cp_single_dim_strategy, + RuntimeSchemaInfo(4), + ), + ( + aten._scaled_dot_product_cudnn_attention_backward.default, + _scaled_dot_product_cudnn_attention_backward_cp_single_dim_strategy, + None, + ), + ] + + # Register each strategy + for op_overload, strategy_func, schema_info in cp_strategies: + ctx = _single_dim_strategy_context(op_overload, strategy_func, schema_info) + orig_funcs, orig_schema = ctx.__enter__() + _cp_strategy_contexts[op_overload] = ctx + _original_strategies[op_overload] = (orig_funcs, orig_schema) + + +def unregister_cp_sharding_rules(clear_the_cache=False): + """Unregister Context Parallelism sharding rules and restore original strategies.""" + global _cp_strategy_contexts, _original_strategies + + # Exit all context managers + for ctx in _cp_strategy_contexts.values(): + ctx.__exit__(None, None, None) + + if clear_the_cache: + _clear_fast_path_sharding_prop_cache() + _clear_python_sharding_prop_cache() + + _cp_strategy_contexts = {} + _original_strategies = {} diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_func_map.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_func_map.py new file mode 100644 index 0000000000000000000000000000000000000000..663e3bc4369ce357ab71d48be09d9f995772fe39 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_func_map.py @@ -0,0 +1,535 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import functools +from collections.abc import Callable, Sequence +from typing import Any + +import torch +import torch.distributed as dist + + +if dist._is_spmd_types_available(): + import spmd_types as spmd + +from torch.distributed._functional_collectives import AsyncCollectiveTensor +from torch.distributed.tensor import DeviceMesh, DTensor +from torch.distributed.tensor.placement_types import ( + Partial, + Placement, + Replicate, + Shard, +) + + +try: + from torch.utils import _cxx_pytree as pytree +except ImportError: + from torch.utils import _pytree as pytree # type: ignore[no-redef] + + +__all__ = ["local_map"] + +PlacementType = Sequence[Placement] | None +InputPlacements = tuple[PlacementType, ...] | None +OutputPlacements = PlacementType | tuple[PlacementType, ...] + + +def _placements_to_spmd_type( + placements: PlacementType, + grad_placements: PlacementType, + device_mesh: DeviceMesh, +) -> dict: + """Convert DTensor placements to a dict of MeshAxis -> spmd local type. + + Mapping: Shard -> V, Replicate -> R (or I if grad is Replicate), Partial -> P. + Validates that grad_placements are compatible with forward placements. + """ + result = {} + for dim_idx, placement in enumerate(placements): # pyrefly: ignore + axis = spmd.MeshAxis.of(device_mesh.get_group(dim_idx)) + grad_p = grad_placements[dim_idx] if grad_placements is not None else None + + if type(placement) is Shard: + fwd_type = spmd.V + elif type(placement) is Replicate: + fwd_type = spmd.I if type(grad_p) is Replicate else spmd.R + elif type(placement) is Partial: + fwd_type = spmd.P + else: + raise ValueError( + f"local_map(spmd_types=True) does not support placement type " + f"{type(placement).__name__}: {placement}" + ) + + if grad_p is not None: + actual_grad_p = _dtensor_placement_if_compatible( + fwd_type.backward_type(), # pyrefly: ignore [missing-attribute] + grad_p, + ) + if actual_grad_p is None or type(actual_grad_p) is not type(grad_p): + raise ValueError( + "local_map(spmd_types=True) cannot represent this " + "forward/backward placement pair:\n" + f"in_placements={placement}, in_grad_placements={grad_p}.\n\n" + "The local_map boundary assigns one SPMD type to each " + "local tensor, which also fixes the expected backward " + f"placement. For in_placements={placement}, valid grad " + f"placements are {_valid_grad_placements(placement)}.\n\n" + "If the forward and backward layouts intentionally " + "diverge in a way not represented by one SPMD type, " + "expand the local_map region to include that divergence, " + "so the local_map region boundary respects tied FWD-BWD " + "typing." + ) + + result[axis] = fwd_type + return result + + +def _annotate_spmd_types( + flat_local_args: list, + in_placements: InputPlacements, + in_grad_placements: InputPlacements, + device_mesh: DeviceMesh, +) -> None: + """Annotate unwrapped local tensors with spmd_types inferred from placements. + + Assumes local_map has already validated in_placements and unwrapped DTensors. + """ + for idx, local_arg in enumerate(flat_local_args): + if not isinstance(local_arg, torch.Tensor): + continue + if in_placements is None or in_placements[idx] is None: + continue + grad_placements = ( + in_grad_placements[idx] if in_grad_placements is not None else None + ) + spmd_type = _placements_to_spmd_type( + in_placements[idx], grad_placements, device_mesh + ) + spmd.assert_type(local_arg, spmd_type) + + +def _valid_grad_placements(placement: Placement) -> str: + if type(placement) is Shard: + return "Shard (spmd.V)" + elif type(placement) is Replicate: + return "Partial (spmd.R) or Replicate (spmd.I)" + elif type(placement) is Partial: + return "Replicate (spmd.P)" + raise ValueError( + f"local_map(spmd_types=True) does not support placement type {type(placement).__name__}" + ) + + +def _dtensor_placement_if_compatible( + local_type, placement: Placement +) -> Placement | None: + if local_type == spmd.V: + if type(placement) in (Shard, Partial): + return placement + return None + return spmd.spmd_type_to_dtensor_placement(local_type) + + +def _out_spmd_types_to_grad_placements( + flat_out: Sequence[Any], + out_placements_tuple: tuple[PlacementType, ...], + device_mesh: DeviceMesh, +) -> tuple[PlacementType, ...]: + """Validate output SPMD types and return backward grad placements.""" + grad_out_placements: list[PlacementType] = [] + for out, spec in zip(flat_out, out_placements_tuple, strict=True): + if spec is not None and not isinstance(out, torch.Tensor): + raise ValueError( + f"out_placements specifies {spec} but the corresponding " + f"output is {type(out).__name__}, not a Tensor" + ) + if spec is None: + if isinstance(out, torch.Tensor): + raise ValueError( + "out_placements has None for a Tensor output. Provide " + "placements for Tensor outputs and use None only for " + "non-Tensor outputs." + ) + grad_out_placements.append(None) + continue + + # validate out type + actual_type = spmd.get_local_type(out) + if not actual_type: + raise ValueError( + "Output tensor has no spmd_types annotation but out_placements " + "expects one. Ensure the function's output is derived from " + "annotated inputs or is explicitly annotated." + ) + for dim_idx, placement in enumerate(spec): # pyrefly: ignore + axis = spmd.MeshAxis.of(device_mesh.get_group(dim_idx)) + actual = actual_type.get(axis) + if actual is None: + if device_mesh.mesh_dim_names is None: + mesh_dims: Sequence[Any] = tuple(range(device_mesh.ndim)) + else: + mesh_dims = device_mesh.mesh_dim_names + annotated_dims = [ + mesh_dims[i] + for i in range(device_mesh.ndim) + if spmd.MeshAxis.of(device_mesh.get_group(i)) in actual_type + ] + raise ValueError( + "Output tensor has no spmd_types annotation on " + f"DeviceMesh dimension {mesh_dims[dim_idx]} " + f"but out_placements expects {placement}. " + "Actual annotations on this DeviceMesh are: " + f"{annotated_dims}" + ) + actual_placement = _dtensor_placement_if_compatible(actual, placement) + if actual_placement is None or type(actual_placement) is not type( + placement + ): + raise ValueError( + f"Output tensor placement mismatch on {axis}: " + f"out_placements={placement} but spmd_types inferred " + f"spmd.{actual.name}" + ) + + # converts out types -> out bwd types -> to DTensor placements, e.g. + # P (or V) -> R -> Replicate + # R -> P -> Partial + # I -> I -> Replicate + # V -> V -> Shard (shard dim consulted from out_placements) + grad_spec = [] + for dim_idx, placement in enumerate(spec): + axis = spmd.MeshAxis.of(device_mesh.get_group(dim_idx)) + if actual_type[axis] == spmd.V and type(placement) is Partial: + grad_placement = Replicate() + else: + grad_placement = _dtensor_placement_if_compatible( + actual_type[axis].backward_type(), placement + ) + if grad_placement is None: + raise AssertionError("expected grad_placement to be not None") + grad_spec.append(grad_placement) + grad_out_placements.append(tuple(grad_spec)) + + return tuple(grad_out_placements) + + +def local_map( + func: Callable | None = None, + out_placements: OutputPlacements = None, + in_placements: InputPlacements = None, + in_grad_placements: InputPlacements = None, + device_mesh: DeviceMesh | None = None, + *, + redistribute_inputs: bool = False, + spmd_types: bool = False, +): + """ + :meth:`local_map` is an experimental API that allows users to pass :class:`DTensor` s + to a function that is written to be applied on ``torch.Tensor`` s. It is done by extracting + the local components of :class:`DTensor`, call the function, and wrap the outputs to + :class:`DTensor` according to the ``out_placements``. + + Args: + func (Callable): the function to be applied on each local shard of + :class:`DTensor` s. + out_placements (Union[`PlacementType`, Tuple[`PlacementType`, ...]]): + the desired placements of the :class:`DTensor` s in ``func``'s flattened output. + If the flattened ``output`` is a single value, the ``out_placements`` should be + of type `PlacementType`. Otherwise if the flattened ``output`` has multiple + values, the ``out_placements`` should be a tuple of `PlacementType` values 1:1 + mapping to the flattened ``output``. + Besides, for :class:`Tensor` output, we use `PlacementType` as its + placements (a `Tuple[Placement]` value). For non-Tensor output, the `PlacementType` + should be `None`. + Note that the only exception is when no :class:`DTensor` argument is passed + in. In this case, even if `out_placements` is not `None`, the result function + should ignore the desired placements because the function is not running with + :class:`DTensor` s. + in_placements (Tuple[`PlacementType`, ...], optional): + the required placements of the :class:`DTensor` s in the flattened inputs of ``func``. + If ``in_placements`` is specified, :meth:`local_map` would examine whether the + placements of each :class:`DTensor` argument is the same as the required + placements or not. If the placements are not the same and + ``redistribute_inputs`` is ``False``, an exception will be raised. Otherwise if + ``redistribute_inputs`` is ``True``, the argument will be first redistributed to + the required sharding placements before passing its local tensor to ``func``. + The only exception is when required placements are not ``None`` and the + argument is a :class:`torch.Tensor`. In this case, the placements examination + will be skipped and the argument will be directly passed to ``func``. + If ``in_placements`` is ``None``, no placements examination will be performed. + Default: None + in_grad_placements (Tuple[`PlacementType`, ...], optional): + the placements hint of the :class:`DTensor` s gradient corresponds + to the flattened input DTensor. This argument is the hint that user + can give to :meth:`to_local` in case the gradient layout of the + local tensor input does not match its :class:`DTensor` input layout. + If not specified, we will assume the gradient layout of the local + tensor input remains the same as the original :class:`DTensor` input + and use that for gradient computation. Default: None. + device_mesh (:class:`DeviceMesh`, optional): + the device mesh that the output :class:`DTensor` s are placed on. If not + specified, this will be inferred from the first input :class:`DTensor`'s device + mesh. Default: None. + + Keyword Args: + redistribute_inputs (bool, optional): + the bool value indicating whether to reshard the input :class:`DTensor` s when + their placements are different from the required input placements. If this + value is ``False`` and some :class:`DTensor` input has a different placement, + an exception will be raised. Default: False. + spmd_types (bool, optional): + if ``True``, enable runtime SPMD type checking for ``func``. Local + tensors are annotated with SPMD types (V, R, I, P) inferred from + ``in_placements`` and ``in_grad_placements``, and ``func`` is executed + under strict type checking. This helps surface distributed correctness + bugs such as whether a replicated local tensor's gradient is + Partial or Replicate. + See the ``spmd_types`` local SPMD types documentation for the + meaning of V, R, I, and P and their backward types: + https://github.com/meta-pytorch/spmd_types/blob/main/docs/local_spmd_types.md + For Replicate input placements, if a gradient placement is not provided, + we default to Replicate (R), instead of Invariant (I). Requires the + ``spmd_types`` package to be installed. Default: False. + + Returns: + A ``Callable`` that applies ``func`` to each local shard of the input :class:`DTensor` + and returns a :class:`DTensor` constructed from the return value of ``func``. + + Raises: + AssertionError: For any non-DTensor output, we require its corresponding + output placement in ``out_placements`` be None. An AssertionError will be raised + if this is not the case. + + ValueError: If ``redistribute_inputs=False`` but the input :class:`DTensor` needs + a redistribution according to ``in_placements``. + + Example: + >>> # xdoctest: +SKIP("distributed") + >>> def mm_allreduce_forward(device_mesh, W, X): + >>> partial_sum_tensor = torch.mm(W, X) + >>> reduced_tensor = funcol.all_reduce(partial_sum_tensor, "sum", device_mesh) + >>> return reduced_tensor + >>> + >>> W = torch.randn(12, 8, requires_grad=False) + >>> X = torch.randn(8, 16, requires_grad=False) + >>> Y = torch.mm(W, X) + >>> row_wise = [Shard(0)] # row-wise sharding placements on 1-d mesh + >>> col_wise = [Shard(1)] # col-wise sharding placements on 1-d mesh + >>> + >>> # local_mm_allreduce_forward is the function wrapped with DTensor/Tensor conversion + >>> local_mm_allreduce_forward = local_map( + >>> mm_allreduce_forward, + >>> out_placements=[Replicate()], + >>> in_placements=[col_wise, row_wise], + >>> device_mesh=device_mesh, + >>> ) + >>> + >>> W_dt = distribute_tensor( + ... W, device_mesh, (col_wise) + ... ) # col-wisely sharded W tensor + >>> X_dt = distribute_tensor( + ... X, device_mesh, (row_wise) + ... ) # row-wisely sharded X tensor + >>> Y_dt = local_mm_allreduce_forward( + ... device_mesh, W_dt, X_dt + ... ) # apply local_mm_allreduce_forward to DTensors + + .. note:: This API is currently experimental and subject to change + """ + + if func is None: + # decorator mode + def decorated(func): + return local_map( + func=func, + out_placements=out_placements, + in_placements=in_placements, + in_grad_placements=in_grad_placements, + device_mesh=device_mesh, + redistribute_inputs=redistribute_inputs, + spmd_types=spmd_types, + ) + + return decorated + + return functools.partial( + _local_map_wrapped, + func, + out_placements, + in_placements, + in_grad_placements, + device_mesh, + redistribute_inputs, + spmd_types, + ) + + +def _local_map_wrapped( + func: Callable, + out_placements: OutputPlacements, + in_placements: InputPlacements, + in_grad_placements: InputPlacements, + device_mesh: DeviceMesh | None, + redistribute_inputs: bool, + enable_spmd_types: bool, + *args, + **kwargs, +): + # process input args + flat_args, args_spec = pytree.tree_flatten(args) + if in_placements is not None: + if len(in_placements) != len(flat_args): + raise AssertionError( + f"in_placements length {len(in_placements)} does not match the number " + f"of input args {len(flat_args)}!" + ) + + # we assume every DTensor object is placed on the same device mesh + flat_local_args = [] + seen_dtensor_arg = False + for idx, arg in enumerate(flat_args): + if isinstance(arg, DTensor): + # TODO: the current code doesn't consider the uneven sharding case + # Need to think about what the consequence is when the input DTensor + # is uneven sharded. + if device_mesh is None: # infer device mesh from the DTensor arg + device_mesh = arg.device_mesh + + # this function is applied to at least one DTensor argument + seen_dtensor_arg = True + + if in_placements is not None: + spec = in_placements[idx] + if spec is None: + raise AssertionError( + f"DTensor input {arg} expects placements but received {spec}!" + ) + + if not isinstance(spec, tuple): + spec = tuple(spec) + + if arg.placements != spec: + if redistribute_inputs: + # redistribute to input placements + arg = arg.redistribute(placements=spec) + else: + raise ValueError( + f"arg {arg} in local_map has a mismatched placements: " + f"arg placements is {arg.placements} but the input " + f"placements is {spec}! " + "If redistribute_inputs is wanted, set " + "redistribute_inputs=True to local_map." + ) + + if in_grad_placements is not None: + spec = in_grad_placements[idx] + if spec is None: + raise AssertionError( + f"DTensor input {arg} expects in grad placements but received {spec}!" + ) + if not isinstance(spec, tuple): + spec = tuple(spec) + local_arg = arg.to_local(grad_placements=spec) + else: + local_arg = arg.to_local() + + if isinstance(local_arg, AsyncCollectiveTensor): + local_arg = local_arg.wait() + + flat_local_args.append(local_arg) + else: + # Non-Tensor input must have None in `in_placements` + if in_placements is not None and not isinstance(arg, torch.Tensor): + spec = in_placements[idx] + if spec is not None: + raise AssertionError( + f"Non-Tensor input {arg} expects None placements " + f"but received {spec}!" + ) + + flat_local_args.append(arg) + + # pyrefly: ignore [bad-argument-type] + local_args = pytree.tree_unflatten(flat_local_args, args_spec) + + if enable_spmd_types and seen_dtensor_arg: + if not dist._is_spmd_types_available(): + raise RuntimeError( + "spmd_types=True requires the spmd_types package to be installed" + ) + if device_mesh is None: + raise AssertionError("expected device_mesh to be not None") + _annotate_spmd_types( + flat_local_args, in_placements, in_grad_placements, device_mesh + ) + from spmd_types._checker import typecheck + + with ( + spmd.set_current_mesh(device_mesh), + typecheck(strict_mode="strict"), + ): + out = func(*local_args, **kwargs) + else: + out = func(*local_args, **kwargs) + + if seen_dtensor_arg: + # process output to be DTensor if we've seen DTensor inputs + flat_out, out_spec = pytree.tree_flatten(out) + + flat_dist_out = [] + out_placements_tuple = ( + out_placements if isinstance(out_placements, tuple) else (out_placements,) + ) + if len(flat_out) != len(out_placements_tuple): + raise AssertionError( + "local_map requires one PlacementType be provided for each output value," + f" received {len(out_placements_tuple)} out_placements but" + f" {len(flat_out)} is expected!" + ) + + grad_out_placements = None + if enable_spmd_types: + if device_mesh is None: + raise AssertionError("expected device_mesh to be not None") + grad_out_placements = _out_spmd_types_to_grad_placements( + flat_out, + out_placements_tuple, # pyrefly: ignore [bad-argument-type] + device_mesh, + ) + + for idx, (out, spec) in enumerate(zip(flat_out, out_placements_tuple)): + if isinstance(out, torch.Tensor): + if isinstance(out, DTensor): + raise AssertionError( + f"torch.Tensor output expected but received {type(out)}: {out}" + ) + grad_placements = ( + grad_out_placements[idx] + if grad_out_placements is not None + else None + ) + + flat_dist_out.append( + # pyrefly: ignore [bad-argument-type] + DTensor.from_local( + out, + device_mesh, + spec, # pyrefly: ignore[bad-argument-type] + run_check=False, + grad_placements=grad_placements, + ) + ) + else: + if spec is not None: + raise AssertionError( + f"Non-tensor output {out} expects None placements but received {spec}!" + ) + + flat_dist_out.append(out) + + # pyrefly: ignore [bad-argument-type] + return pytree.tree_unflatten(flat_dist_out, out_spec) + else: + return out diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_register_sharding.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_register_sharding.py new file mode 100644 index 0000000000000000000000000000000000000000..7b365dcf286d03be9628c5f909682bcd0a818f7e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_register_sharding.py @@ -0,0 +1,136 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +from collections.abc import Callable, Sequence +from functools import partial + +import torch +from torch._ops import OpOverload +from torch.distributed.tensor import DTensor +from torch.distributed.tensor._op_schema import ( + OpSchema, + OpStrategy, + PlacementList, + RuntimeSchemaInfo, + StrategyType, + TupleStrategy, +) +from torch.distributed.tensor._ops.utils import expand_to_full_mesh_op_strategy + + +__all__ = ["register_sharding"] + + +def register_sharding(op: OpOverload | list[OpOverload]): + """ + :meth:`register_sharding` is an experimental API that allows users to register sharding + strategies for an operator when the tensor inputs and outputs are DTensor. + It can be useful when: (1) there doesn't exist a default sharding strategy for ``op``, + e.g. when ``op`` is a custom operator that is not supported by :class:`DTensor`; (2) + when users would like to overwrite default sharding strategies of existing operators. + + Args: + op (Union[OpOverload, List[OpOverload]]): + An op or a list of ops to register the customized sharding function. + + Returns: + A function decorator which can be used to wrap a function that defines the sharding + strategy for the operator specified in ``op``. The defined sharding strategy will be + registered to DTensor and will override the default sharding strategy if DTensor has + already implemented the operator. The customized sharding function takes the same inputs + as the original op (except that if an arg is a :class:`torch.Tensor`, it will be + replaced by a tensor-like object that DTensor uses internally). The function should + return a sequence of 2-tuples, each specifying acceptable output placements and its + corresponding input placements. + + Example: + >>> # xdoctest: +SKIP("distributed") + >>> @register_sharding(aten._softmax.default) + >>> def custom_softmax_sharding(x, dim, half_to_float): + >>> softmax_dim = dim if dim >= 0 else dim + x.ndim + >>> acceptable_shardings = [] + >>> + >>> all_replicate = ([Replicate()], [Replicate(), None, None]) + >>> acceptable_shardings.append(all_replicate) + >>> + >>> for sharding_dim in range(x.ndim): + >>> if sharding_dim != softmax_dim: + >>> all_sharded = ( + >>> [Shard(sharding_dim)], + >>> [Shard(sharding_dim), None, None], + >>> ) + >>> acceptable_shardings.append(all_sharded) + >>> + >>> return acceptable_shardings + + .. note:: This API is currently experimental and subject to change + """ + + def custom_strategy( + custom_sharding_fn: Callable[ + ..., Sequence[tuple[PlacementList, PlacementList]] + ], + op_schema: OpSchema, + ) -> StrategyType: + def strategy_to_spec(strategy: object) -> object: + if isinstance(strategy, OpStrategy): + # take the output spec from the first strategy + return strategy.strategies[0].output_spec + elif isinstance(strategy, TupleStrategy): + return tuple(strategy_to_spec(s) for s in strategy.children) + else: + return strategy + + mesh = op_schema.get_mesh_from_args() + + args_schema = tuple(strategy_to_spec(i) for i in op_schema.args_schema) + kwargs_schema = { + k: strategy_to_spec(v) for k, v in op_schema.kwargs_schema.items() + } + + acceptable_shardings = custom_sharding_fn(*args_schema, **kwargs_schema) + + single_mesh_dim_strategies: list[PlacementList] = [] + for output_specs, input_specs in acceptable_shardings: + single_mesh_dim_strategies.append(output_specs + input_specs) + + # TODO: handle out variant ops + return expand_to_full_mesh_op_strategy( + mesh, + op_schema, + single_mesh_dim_strategies, + input_index=len(op_schema.op._schema.returns), + inplace_op=op_schema.is_inplace_op(), + ) + + def wrapper(custom_sharding_fn): + def derive_schema_info(op): + # NOTE: without user directly providing RuntimeSchemaInfo, for now + # we create it in a conservative fashion as follows: + # 1. let static_argnum be the first int argument + # 2. let static_kwargkey include all the int type kwargs + # 3. always set needs_pytree=True + static_argnum = 100 + static_kwargkey: list[str] = [] + for i, arg in enumerate(op._schema.arguments): + if isinstance(arg.type, torch.IntType) or ( + isinstance(arg.type, torch.OptionalType) + and isinstance(arg.type.getElementType(), torch.IntType) + ): + static_argnum = min(i, static_argnum) + if arg.kwarg_only: + static_kwargkey.append(arg.name) + return RuntimeSchemaInfo( + static_argnum, static_kwargkey or None, needs_pytree=True + ) + + overloads = op if isinstance(op, list) else [op] + for overload in overloads: + DTensor._op_dispatcher.sharding_propagator.register_op_strategy( + overload, + partial(custom_strategy, custom_sharding_fn), + derive_schema_info(overload), + ) + + return custom_sharding_fn + + return wrapper diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_tp_transform.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_tp_transform.py new file mode 100644 index 0000000000000000000000000000000000000000..adde891c4b68c0198568702d16f3b011a7a309ed --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/experimental/_tp_transform.py @@ -0,0 +1,576 @@ +# mypy: allow-untyped-defs +import copy +import operator +from collections.abc import Sequence +from typing import Any, cast + +import torch +from torch._subclasses.fake_tensor import FakeTensor +from torch.distributed.tensor import DeviceMesh, distribute_tensor, DTensor +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._op_schema import ( + OpSchema, + OpSpec, + OutputSharding, + OutputSpecType, +) +from torch.distributed.tensor._redistribute import redistribute_local_tensor +from torch.distributed.tensor.parallel.style import ColwiseParallel, ParallelStyle +from torch.distributed.tensor.placement_types import Placement, Replicate, Shard +from torch.export import ExportedProgram +from torch.export.exported_program import ExportGraphSignature +from torch.fx import GraphModule +from torch.fx.experimental.proxy_tensor import make_fx +from torch.fx.node import Node +from torch.fx.passes.infra.pass_base import PassBase, PassResult +from torch.fx.passes.shape_prop import _extract_tensor_metadata +from torch.utils import _pytree as pytree + + +__all__ = ["tensor_parallel_transformation"] + +aten = torch.ops.aten + + +def tensor_parallel_transformation( + exported_program: ExportedProgram, + rank: int, + world_size: int, + device_type: str, + parallel_strategies: dict[str, ParallelStyle], +) -> ExportedProgram: + """ + The entry point function to perform graph transformations on an exported program + to transform a single-device graph into a tensor parallel graph. + + .. warning:: + This API is experimental and subject to change. + """ + + gm = exported_program.graph_module + sig = copy.deepcopy(exported_program.graph_signature) + state_dict = copy.copy(exported_program.state_dict) + + with gm._set_replace_hook(sig.get_replace_hook()): + res = _TensorParallelTransformPass( + rank, + world_size, + device_type, + state_dict, + exported_program.graph_signature, + parallel_strategies, + )(gm) + if res is None: + raise AssertionError + gm = res.graph_module + + return exported_program._update(gm, sig, state_dict=state_dict) + + +class _TensorParallelTransformPass(PassBase): + """ + This pass is responsible for transforming a single-device graph into a tensor parallel + graph. It will mark the OpSpec of each node in the graph, partition the graph into + distributed graph, then shard the parameters/buffers accordingly. + """ + + def __init__( + self, + rank: int, + world_size: int, + device_type: str, + state_dict: dict[str, torch.Tensor], + graph_signature: ExportGraphSignature, + parallel_strategies: dict[str, ParallelStyle], + ) -> None: + super().__init__() + self.rank = rank + self.mesh = DeviceMesh(device_type, torch.arange(world_size)) + self.state_dict: dict[str, torch.Tensor] = state_dict + self.graph_signature = graph_signature + self.parallel_strategies = parallel_strategies + + def call(self, graph_module) -> PassResult: + gm = copy.deepcopy(graph_module) + + parameter_placements = _generate_parameter_and_buffer_placements( + list(self.state_dict.keys()), self.parallel_strategies + ) + placement_strategies = _mark_sharding( + gm, self.graph_signature, self.mesh, parameter_placements + ) + _partitioner(gm) + _shard_state_dict( + self.state_dict, placement_strategies, self.graph_signature, self.mesh + ) + return PassResult(gm, True) + + +def _generate_parameter_and_buffer_placements( + params_and_buffers: list[str], + parallel_strategies: dict[str, ParallelStyle], +) -> dict[str, Placement]: + """ + Build parameter placements based on the give parallel style of linear layers. + """ + parameter_placements: dict[str, Placement] = {} + for linear_fqn, parallel_style in parallel_strategies.items(): + weight_fqn = f"{linear_fqn}.weight" + bias_fqn = f"{linear_fqn}.bias" + if weight_fqn not in params_and_buffers: + raise AssertionError + parameter_placements[weight_fqn] = ( + Shard(0) if parallel_style == ColwiseParallel else Shard(1) + ) + if bias_fqn in params_and_buffers: + parameter_placements[bias_fqn] = ( + Shard(0) if parallel_style == ColwiseParallel else Replicate() + ) + return parameter_placements + + +def _mark_tensor_parallel_shardings( + gm: GraphModule, + graph_signature: ExportGraphSignature, + mesh: DeviceMesh, + parameter_placements: dict[str, Placement], +) -> dict[Node, OpSpec]: + """ + Mark the placement strategies of the parameter and buffer placeholder nodes. + """ + placement_strategies: dict[Node, OpSpec] = {} + num_params_and_buffers = len(graph_signature.inputs_to_parameters) + len( + graph_signature.inputs_to_buffers + ) + placeholder_idx: int = 0 + for node in gm.graph.nodes: + if node.op == "placeholder": + if placeholder_idx < num_params_and_buffers: + fqn: str = _get_input_node_fqn(node.name, graph_signature) + placement: Placement = ( + parameter_placements[fqn] + if fqn in parameter_placements + else Replicate() + ) + placement_strategies[node] = _create_placement_strategy( + node, + mesh, + placements=(placement,), + ) + placeholder_idx += 1 + else: + placement_strategies[node] = _create_placement_strategy( + node, + mesh, + placements=(Replicate(),), + ) + return placement_strategies + + +def _get_input_node_fqn(input_name: str, graph_signature: ExportGraphSignature) -> str: + """ + Return the FQN of an input node. + """ + if input_name in graph_signature.inputs_to_parameters: + return graph_signature.inputs_to_parameters[input_name] + elif input_name in graph_signature.inputs_to_buffers: + return graph_signature.inputs_to_buffers[input_name] + else: + raise ValueError( + f"{input_name} not found in inputs_to_parameters or inputs_to_buffers" + ) + + +def _mark_sharding( + gm: GraphModule, + graph_signature: ExportGraphSignature, + mesh: DeviceMesh, + parameter_placements: dict[str, Placement], +) -> dict[Node, OpSpec]: + """ + Mark the sharding strategy for each node in the graph module. + """ + placement_strategies: dict[Node, OpSpec] = _mark_tensor_parallel_shardings( + gm, + graph_signature, + mesh, + parameter_placements, + ) + + for node in gm.graph.nodes: + if node.op == "placeholder": + if node not in placement_strategies: + placement_strategies[node] = _create_placement_strategy( + node, mesh, placements=(Replicate(),) + ) + node.meta["sharding"] = placement_strategies[node] + elif node.op == "call_function": + if node.target is operator.getitem: + input_nodes = node.all_input_nodes + if len(input_nodes) != 1: + raise AssertionError( + f"non-compute op only support one input now, found node: {node} " + f"with length of inputs: {len(node.args)}" + ) + arg_strategy = placement_strategies[input_nodes[0]] + placement_strategies[node] = _create_placement_strategy( + node, + mesh, + placements=arg_strategy.output_spec.placements, + input_specs=_get_input_node_specs(node, placement_strategies), + ) + node.meta["sharding"] = placement_strategies[node] + else: + op_schema = _get_op_schema(node, placement_strategies) + + # get DTensor specs for inputs and outputs + sharding_propagator = DTensor._op_dispatcher.sharding_propagator + if ( + op_schema.op not in sharding_propagator.op_strategy_funcs + and op_schema.op not in sharding_propagator.op_to_rules + and op_schema.op + not in sharding_propagator.op_single_dim_strategy_funcs + ): + # Mark all as replicated + output_sharding = _generate_default_output_sharding( + node, + mesh, + op_schema, + ) + else: + output_sharding = DTensor._op_dispatcher.sharding_propagator.propagate_op_sharding( # type: ignore[assignment] + op_schema, + ) + placement_strategies[node] = OpSpec( + # pyrefly: ignore [bad-argument-type] + output_specs=_get_output_spec_from_output_sharding(output_sharding), + # pyrefly: ignore [missing-attribute] + input_specs=output_sharding.redistribute_schema.args_spec + # pyrefly: ignore [missing-attribute] + if output_sharding.redistribute_schema is not None + else _get_input_node_specs(node, placement_strategies), + ) + node.meta["sharding"] = placement_strategies[node] + elif node.op == "output": + node.meta["sharding"] = None + else: + raise RuntimeError(f"op code {node.op} not supported") + return placement_strategies + + +def _get_output_spec_from_output_sharding( + output_sharding: OutputSharding, +) -> DTensorSpec: + """ + Util function to extract output spec from output sharding. + """ + if isinstance(output_sharding.output_spec, DTensorSpec): + return output_sharding.output_spec + else: + # For ops that return multiple outputs, the outputs should have the same output spec + if not isinstance(output_sharding.output_spec, Sequence): + raise AssertionError + if output_sharding.output_spec[0] is None: + raise AssertionError + output_sharding.output_spec[0].tensor_meta = None + return output_sharding.output_spec[0] + + +def _create_placement_strategy( + node: Node, + mesh: DeviceMesh, + placements: tuple[Placement, ...], + input_specs: Sequence[DTensorSpec] | None = None, +) -> OpSpec: + """ + Util function to construct an OpSpec for a given node. + """ + placement = OpSpec( + input_specs=input_specs, + output_specs=DTensorSpec( + mesh=mesh, + placements=placements, + ), + ) + _populate_tensor_meta(node, placement.output_specs) + return placement + + +def _populate_tensor_meta(node: Node, output_spec: OutputSpecType) -> None: + """ + Util function to populate tensor meta of output_spec based on node metadata. + """ + if isinstance(node.meta["val"], Sequence): + if not isinstance(output_spec, Sequence): + raise AssertionError + for spec, fake_tensor in zip(output_spec, node.meta["val"]): + if spec is None: + raise AssertionError + spec.tensor_meta = TensorMeta( + shape=fake_tensor.shape, + stride=fake_tensor.stride(), + dtype=fake_tensor.dtype, + ) + else: + if not isinstance(output_spec, DTensorSpec): + raise AssertionError + output_spec.tensor_meta = TensorMeta( + shape=node.meta["val"].shape, + stride=node.meta["val"].stride(), + dtype=node.meta["val"].dtype, + ) + + +def _generate_default_output_sharding( + node: Node, + mesh: DeviceMesh, + op_schema: OpSchema, +) -> OutputSharding: + """ + Util function to create a default output sharding that suggests Replicate placement for both args and outputs. + """ + + def update_arg_spec(arg_spec: DTensorSpec) -> DTensorSpec: + return DTensorSpec( + mesh=arg_spec.mesh, + placements=(Replicate(),), + tensor_meta=arg_spec.tensor_meta, + ) + + new_op_schema = OpSchema( + op=op_schema.op, + args_schema=pytree.tree_map_only( + DTensorSpec, update_arg_spec, op_schema.args_schema + ), + kwargs_schema=op_schema.kwargs_schema, + ) + + def create_output_spec(tensor: FakeTensor) -> DTensorSpec: + return DTensorSpec( + mesh=mesh, + placements=(Replicate(),), + tensor_meta=TensorMeta( + shape=tensor.shape, + stride=tensor.stride(), + dtype=tensor.dtype, + ), + ) + + return OutputSharding( + output_spec=pytree.tree_map_only( + FakeTensor, create_output_spec, node.meta["val"] + ), + redistribute_schema=new_op_schema, + needs_redistribute=True, + ) + + +def _partitioner(gm: torch.fx.GraphModule) -> torch.fx.GraphModule: + """ + Graph partitioner that partitions the single device graph + to distributed graph + """ + for node in gm.graph.nodes: + node_sharding = node.meta["sharding"] + if node.op == "placeholder": + out_spec = node_sharding.output_spec + local_val = _partition_val(node.meta["val"], out_spec) + # update node value + node.meta["val"] = local_val + elif node.op == "call_function": + out_spec = node_sharding.output_spec + # check if there's misaligned sharding, insert reshard if there is + expected_input_specs = node_sharding.input_specs + for idx, input_arg in enumerate(node.all_input_nodes): + input_arg_sharding = input_arg.meta["sharding"] + input_arg_spec = input_arg_sharding.output_spec + desired_spec = ( + out_spec + if expected_input_specs is None + else expected_input_specs[idx] + ) + if input_arg_spec != desired_spec: + _insert_reshard_gm( + gm, node, input_arg, input_arg_spec, desired_spec + ) + # convert output val to its local component + output_val = node.meta["val"] + node.meta["val"] = _partition_val(output_val, out_spec) + elif node.op == "output": + for input_arg in node.all_input_nodes: + # input args of output should be Replicate, otherwise redistribution is needed. + input_args_to_check: Sequence[Node] = ( + input_arg if isinstance(input_arg, Sequence) else [input_arg] + ) + for arg in input_args_to_check: + arg_sharding = arg.meta["sharding"] + arg_spec = arg_sharding.output_spec + desired_spec = copy.copy(arg_spec) + desired_spec.placements = (Replicate(),) + if arg_spec != desired_spec: + _insert_reshard_gm(gm, node, arg, arg_spec, desired_spec) + else: + raise RuntimeError(f"op code {node} not supported") + + _clean_up_graph_metadata(gm) + gm.graph.lint() + gm.recompile() + return gm + + +def _partition_val(val: Any, spec: DTensorSpec) -> Any: + """ + util function to convert a full tensor val to its local component + """ + if isinstance(val, torch.Tensor): + local_shard = val + if val.ndim == 0: + # If it's already a scalar tensor, it is already local, we don't + # need to do anything + return local_shard + + for idx, placement in enumerate(spec.placements): + # NOTE: is_shard() does not match _StridedShard; see _is_shard_like(). + if placement.is_shard(): + placement = cast(Shard, placement) + num_chunks = spec.mesh.size(mesh_dim=idx) + my_coord = spec.mesh.get_coordinate() + if my_coord is None: + raise AssertionError("current rank not in mesh!") + my_coord_on_mesh_dim = my_coord[idx] + local_shard = placement._select_split_tensor( + local_shard, + num_chunks, + my_coord_on_mesh_dim, + with_padding=False, + contiguous=True, + clone=False, + ) + return local_shard + elif isinstance(val, (list, tuple)): + return val.__class__(_partition_val(v, spec) for v in val) + else: + raise RuntimeError(f"val type {type(val)} not supported") + + +def _insert_reshard_gm( + gm: torch.fx.GraphModule, + node: Node, + input_arg: Node, + input_arg_spec: DTensorSpec, + desired_spec: DTensorSpec, +) -> None: + """ + Transform the graph for tensor redistribution. + """ + input_arg_spec.tensor_meta = input_arg.meta["tensor_meta"] + desired_spec.tensor_meta = input_arg.meta["tensor_meta"] + input_arg_tensor = input_arg.meta["val"] + + # insert reshard operation + def reshard_fn(local_tensor: torch.Tensor) -> torch.Tensor: + return redistribute_local_tensor( + local_tensor, + input_arg_spec, + desired_spec, + ) + + reshard_gm = make_fx(reshard_fn)(input_arg_tensor) + reshard_gm_nodes = list(reshard_gm.graph.nodes) + input_node = reshard_gm_nodes[0] + with gm.graph.inserting_before(node): + # copy nn_module_stack metadata for output, all-reduce nodes + for reshard_node in reshard_gm.graph.nodes: + if reshard_node.op not in ["placeholder", "output"]: + reshard_node.meta["nn_module_stack"] = ( + copy.copy(input_arg.meta["nn_module_stack"]) + if input_arg.op != "placeholder" + else copy.copy(node.meta["nn_module_stack"]) + ) + output_node = gm.graph.graph_copy( + reshard_gm.graph, + val_map={ + input_node: input_arg, + }, + ) + node.replace_input_with(input_arg, output_node) # type: ignore[arg-type] + + +def _clean_up_graph_metadata(gm: torch.fx.GraphModule) -> None: + """ + Clean up the graph by removing sharding and partitioning related metadata + """ + for node in gm.graph.nodes: + if "sharding" in node.meta: + del node.meta["sharding"] + if "val" in node.meta and isinstance(node.meta["val"], torch.Tensor): + local_tensor_meta = _extract_tensor_metadata(node.meta["val"]) + node.meta["tensor_meta"] = local_tensor_meta + + +def _get_input_node_specs( + node: Node, placement_strategies: dict[Node, OpSpec] +) -> tuple[DTensorSpec, ...]: + """ + Get the input specs of a node. + """ + input_specs_list: list[DTensorSpec] = [] + for input_arg in node.all_input_nodes: + if input_arg in placement_strategies: + output_spec = placement_strategies[input_arg].output_specs + if not isinstance(output_spec, DTensorSpec): + raise AssertionError + input_specs_list.append(output_spec) + else: + raise ValueError(f"{input_arg} does not have output_spec populated.") + return tuple(input_specs_list) + + +def _get_op_schema(node: Node, placement_strategies: dict[Node, OpSpec]) -> OpSchema: + """ + Util function to construct the operator schema of a node. + """ + args_schema_list = pytree.tree_map_only( + Node, lambda arg: placement_strategies[arg].output_specs, node.args + ) + op_schema = OpSchema( + op=cast(torch._ops.OpOverload, node.target), + args_schema=tuple(args_schema_list), + kwargs_schema=cast(dict[str, object], node.kwargs), + ) + return op_schema + + +def _shard_state_dict( + state_dict: dict[str, torch.Tensor], + placement_strategies: dict[Node, OpSpec], + graph_signature: ExportGraphSignature, + mesh: DeviceMesh, +) -> None: + """ + Inplace partition the weights based on the OpSpec + """ + for node, op_spec in placement_strategies.items(): + if node.op != "placeholder": + continue + if node.name in graph_signature.inputs_to_parameters: + fqn = graph_signature.inputs_to_parameters[node.name] + elif node.name in graph_signature.inputs_to_buffers: + fqn = graph_signature.inputs_to_buffers[node.name] + else: + continue + if fqn not in state_dict: + raise AssertionError(f"{fqn} not found in state dict: {state_dict.keys()}") + + original_param = state_dict[fqn] + dtensor_param = distribute_tensor( + original_param, + mesh, + op_spec.output_spec.placements, + ) + local_param = dtensor_param.to_local() + state_dict[fqn] = ( + torch.nn.Parameter(local_param) + if isinstance(original_param, torch.nn.Parameter) + else local_param + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..5e4881de43874ab238b1cfbe6003c9a8751f0c3b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__init__.py @@ -0,0 +1,25 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from torch.distributed.tensor.parallel.api import parallelize_module +from torch.distributed.tensor.parallel.loss import loss_parallel +from torch.distributed.tensor.parallel.style import ( + ColwiseParallel, + ParallelStyle, + PrepareModuleInput, + PrepareModuleInputOutput, + PrepareModuleOutput, + RowwiseParallel, + SequenceParallel, +) + + +__all__ = [ + "ColwiseParallel", + "ParallelStyle", + "PrepareModuleInput", + "PrepareModuleInputOutput", + "PrepareModuleOutput", + "RowwiseParallel", + "SequenceParallel", + "parallelize_module", + "loss_parallel", +] diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..705b5a66e3087441cf6ca88e6586e54f5cdcd3c5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/_data_parallel_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/_data_parallel_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4197e22abb9b83e9329952e7a866c982300be7cf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/_data_parallel_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/api.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..251fc27b4ae1bc2701f9dee4f6d0afe1aeec82c8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/api.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/ddp.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/ddp.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..718e54d116d42bfb8e925abab6efa392174efd55 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/ddp.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/fsdp.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/fsdp.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..02a9e25e317e1f78a29fc87084f48999e3f5a648 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/fsdp.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/input_reshard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/input_reshard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..16106f0354b1d93e7997855a2d8cf88ab14936f7 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/input_reshard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/loss.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/loss.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2606046330f09b62da2a8e642317903f999bdacc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/loss.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/style.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/style.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7ff6a48df28504183066373540d6c0e8bae7496e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/__pycache__/style.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/_data_parallel_utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/_data_parallel_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..735b74e099478ebc606d68b3099f721c59874297 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/_data_parallel_utils.py @@ -0,0 +1,51 @@ +from functools import partial +from typing import no_type_check + +import torch +from torch.distributed._functional_collectives import AsyncCollectiveTensor +from torch.distributed.tensor import DTensor +from torch.distributed.tensor._dtensor_spec import DTensorSpec + + +@no_type_check +def sync_grad_hook(grad, *, device_handle=None, compute_stream=None): + if isinstance(grad, AsyncCollectiveTensor): + if compute_stream is not None: + with device_handle.stream(compute_stream): + grad = grad.wait() + else: + grad = grad.wait() + + return grad + + +def _flatten_tensor( + tensor: torch.Tensor, +) -> tuple[torch.Tensor, DTensorSpec | None]: + if isinstance(tensor, DTensor): + tensor._local_tensor.requires_grad_() + return tensor._local_tensor, tensor._spec + return tensor, None + + +@no_type_check +def _unflatten_tensor(tensor, spec, *, device_handle=None, compute_stream=None): + # unflatten would mainly be called every time FSDP allgather parameters. + result = DTensor.from_local( + tensor, + spec.mesh, + spec.placements, + run_check=False, + shape=spec.shape, + stride=spec.stride, + ) + if tensor.requires_grad: + # only register the hook if the tensor requires grad + tensor.register_hook( + partial( + sync_grad_hook, + device_handle=device_handle, + compute_stream=compute_stream, + ) + ) + return result diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/api.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/api.py new file mode 100644 index 0000000000000000000000000000000000000000..ededb94b0b5e8e554756f5a6577b190faa15f47c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/api.py @@ -0,0 +1,142 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +import warnings +from fnmatch import fnmatch + +import torch +import torch.nn as nn +from torch.distributed.device_mesh import _mesh_resources, DeviceMesh +from torch.distributed.tensor.parallel.style import ParallelStyle + + +__all__ = ["parallelize_module"] + + +def parallelize_module( # type: ignore[return] + module: nn.Module, + device_mesh: DeviceMesh | None = None, + parallelize_plan: ParallelStyle | dict[str, ParallelStyle] | None = None, + *, + src_data_rank: int | None = 0, +) -> nn.Module: + """ + Apply Tensor Parallelism in PyTorch by parallelizing modules or sub-modules based on a user-specified plan. + + We parallelize module or sub_modules based on a parallelize_plan. The parallelize_plan contains + :class:`ParallelStyle`, which indicates how user wants the module or sub_module + to be parallelized. + + User can also specify different parallel style per module fully qualified name (FQN). + + Note that ``parallelize_module`` only accepts a 1-D :class:`DeviceMesh`, if you have a 2-D or N-D :class:`DeviceMesh`, + slice the DeviceMesh to a 1-D sub DeviceMesh first then pass to this API(i.e. ``device_mesh[\"tp\"]``) + + Args: + module (:class:`nn.Module`): + Module to be parallelized. + device_mesh (:class:`DeviceMesh`, optional): + Object which describes the mesh topology of devices for the DTensor. + If not specified, the call must be under a DeviceMesh context. + parallelize_plan (Union[:class:`ParallelStyle`, Dict[str, :class:`ParallelStyle`]], optional): + The plan used to parallelize the module. It can be either a + :class:`ParallelStyle` object which contains how we prepare + input/output for Tensor Parallelism or it can be a dict of module + FQN and its corresponding :class:`ParallelStyle` object. If not + specified, the call will do nothing at the moment. + Keyword args: + src_data_rank (int, optional): the rank of the source data for the logical/global tensor, it is used by + :meth:`distribute_tensor` to scatter/broadcast the shards/replicas to other ranks. By default, + we use ``group_rank=0`` on each DeviceMesh dimension as the source data to preserve the single-device + semantic. If passing ``None`` explicitly, :meth:`parallelize_module` simply uses its local data instead + of trying to preserve the single-device semantic via scatter/broadcast. Default: 0 + Return: + A :class:`nn.Module` object parallelized. + + Example:: + >>> # xdoctest: +SKIP("distributed") + >>> from torch.distributed.tensor.parallel import parallelize_module, ColwiseParallel + >>> from torch.distributed.device_mesh import init_device_mesh + >>> + >>> # Define the module. + >>> m = Model(...) + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> m = parallelize_module(m, tp_mesh, {"w1": ColwiseParallel(), "w2": RowwiseParallel()}) + >>> + + .. note:: For complex module architecture like Attention, MLP layers, we recommend composing + different ParallelStyles together (i.e. ``ColwiseParallel`` and ``RowwiseParallel``) and pass + as a parallelize_plan, to achieve the desired sharding computation. + """ + torch._C._log_api_usage_once("torch.distributed.tensor.parallel.parallelize_module") + + device_mesh = device_mesh or _mesh_resources.get_current_mesh() + + if parallelize_plan is None: + warnings.warn( + "No parallelize_plan is provided and auto-parallel is not supported " + "at the moment, so this parallelize_module call will do nothing.", + stacklevel=2, + ) + return module + + # note: The RNG tracker will be initialized in distribute_tensor() call if it hasn't + # been initialized. + + if isinstance(parallelize_plan, ParallelStyle): + parallelize_plan.src_data_rank = src_data_rank + return parallelize_plan._apply(module, device_mesh) + elif isinstance(parallelize_plan, dict): + for module_path, parallelize_style in parallelize_plan.items(): + if module_path == "": + # shortcut: empty string means to apply the plan to the current module + parallelize_module(module, device_mesh, parallelize_style) + continue + + path_splits = module_path.split(".") + # Instead of blindly popping tokens, first check the match, + # we only consume/pop the token if we found a match. + token = path_splits[0] + + matched_children = list( + filter( + # `t[0]` is child name + lambda t: fnmatch(t[0], token), + module.named_children(), + ) + ) + if not matched_children: + # No match at this level. Log a warning and process next plan entry. + warnings.warn( + f"Parallelize plan key '{module_path}' could not be resolved: " + f"no submodule matching token '{token}' in module {module}, " + f"skipping this plan entry.", + stacklevel=2, + ) + continue + + # Now that we have a match, we can consume the token. + path_splits.pop(0) + # apply the plan to all matched submodules + for _, submodule in matched_children: + if path_splits: + # we haven't reached the leaf, apply in dict style + leaf_path = ".".join(path_splits) # rest of the path after `token` + parallelize_module( + submodule, + device_mesh, + {leaf_path: parallelize_style}, + src_data_rank=src_data_rank, + ) + else: + # otherwise, directly apply style to this submodule + parallelize_module( + submodule, + device_mesh, + parallelize_style, + src_data_rank=src_data_rank, + ) + return module + else: + raise TypeError( # pyre-ignore[7] + "Expect Union[ParallelStyle, Dict[str, ParallelStyle]] for" + f" parallelize_plan, {type(parallelize_plan)} found!" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/ddp.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/ddp.py new file mode 100644 index 0000000000000000000000000000000000000000..a1331f8bd8c87bf326e52351afbfaba56616c1d9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/ddp.py @@ -0,0 +1,105 @@ +# mypy: allow-untyped-defs +from typing import Any + +import torch.nn as nn +from torch.distributed.tensor.parallel._data_parallel_utils import ( + _flatten_tensor, + _unflatten_tensor, +) + + +__all__ = [] # type: ignore[var-annotated] + + +def _get_submodule_n_params(module: nn.Module, path: str): + """ + Get submodule and the direct path of parameter from the module + """ + if "." in path: + path_list = path.split(".") + parent_module_path = ".".join(path_list[:-1]) + module = module.get_submodule(parent_module_path) + path = path_list[-1] + return module, path + + +def _update_module_param(param_list: list[tuple[nn.Module, str, nn.Parameter]]): + """ + Update parameters within the module + """ + for item in param_list: + parent_module, module_path, t = item + if not hasattr(parent_module, module_path): + raise AssertionError + delattr(parent_module, module_path) + setattr(parent_module, module_path, t) + + +def _reconstruct_dtensor(module: nn.Module, _input: Any): + """ + Reconstruct DTensor parameters from local tensors + """ + param_list = [] + # TODO: To add perf optimizations to this iterations + for name, t in module.named_parameters(): + if hasattr(t, "_st_info"): + dtensor = _unflatten_tensor(t, t._st_info) + param_list.append((*_get_submodule_n_params(module, name), dtensor)) + _update_module_param(param_list) # type: ignore[arg-type] + + +def _localize_dtensor( + module: nn.Module, *_: Any, ignored_params: set[nn.Parameter] | None = None +): + """ + Convert DTensor parameters to local tensors + """ + if ignored_params is None: + ignored_params = set() + param_list = [] + for name, param in module.named_parameters(): + if param in ignored_params: + continue + t, sharding_info = _flatten_tensor(param) + if sharding_info is not None: + t = nn.Parameter(t) + t._st_info = sharding_info # type: ignore[attr-defined] + param_list.append((*_get_submodule_n_params(module, name), t)) + _update_module_param(param_list) # type: ignore[arg-type] + + +def _pre_dp_module_transform(module: nn.Module): + """ + Enable the composability between Tensor Parallelism (TP) and Data + Parallelism(DP) in PyTorch when using DDP. We need to convert Parameters which + are DTensors to local tensors before wrapping with data parallelism API. + We then register two hooks, one for converting local tensors back to DTensor + preforward and one to convert DTensors back to tensors after Forward. By + integrating this way, we avoid any special handling of DTensor parameters by DDP + and get DTensor's gradients propagated back to DP, e.g. gradient buckets of DDP. + + For now, this API only works with ``DistributedDataParallel``. It will later support + other DP methods such as FSDP. + + Args: + module (:class:`nn.Module`): + Module which has been applied TP on. + + Example:: + >>> # xdoctest: +SKIP("distributed") + >>> from torch.distributed.tensor.parallel import parallelize_module, PairwiseParallel + >>> from torch.nn.parallel import DistributedDataParallel as DDP + >>> from torch.distributed.tensor.parallel.ddp import pre_dp_module_transform + >>> + >>> # Define the module. + >>> m = module(...) + >>> parallelize_module(m, PairwiseParallel()) + >>> m = pre_dp_module_transform(m) + >>> m = DDP(m) + >>> + """ + + _localize_dtensor(module, None, None) + # TODO: To add test cases and ensure that it works for nested modules + module.register_forward_pre_hook(_reconstruct_dtensor) + module.register_forward_hook(_localize_dtensor) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/fsdp.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/fsdp.py new file mode 100644 index 0000000000000000000000000000000000000000..c9e9be8d86b547cf627e70518bf2f773b59b18ae --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/fsdp.py @@ -0,0 +1,400 @@ +# mypy: allow-untyped-defs +import copy +from typing import Any, cast + +import torch +import torch.distributed as dist +import torch.distributed._shard.sharding_spec as shard_spec +import torch.distributed.distributed_c10d as c10d +from torch.distributed._shard.sharded_tensor import ( + Shard, + ShardedTensor, + ShardedTensorMetadata, + TensorProperties, +) +from torch.distributed._shard.sharding_spec import ShardMetadata +from torch.distributed._shard.sharding_spec.chunk_sharding_spec import ChunkShardingSpec +from torch.distributed.fsdp._common_utils import _set_fsdp_flattened +from torch.distributed.fsdp._fsdp_extensions import FSDPExtensions +from torch.distributed.fsdp._shard_utils import _create_chunk_sharded_tensor +from torch.distributed.remote_device import _remote_device +from torch.distributed.tensor import DeviceMesh, DTensor, Replicate, Shard as DShard +from torch.distributed.tensor.parallel._data_parallel_utils import ( + _flatten_tensor, + _unflatten_tensor, +) + + +__all__ = ["DTensorExtensions"] + + +def _get_box(tensor: DTensor) -> tuple[torch.Size, torch.Size]: + device_mesh = tensor.device_mesh + if device_mesh.ndim != 1: + raise AssertionError("Only 1D DeviceMeshes currently handled") + + placement = tensor.placements[0] + offsets = [0] * len(tensor.size()) + num_chunks = device_mesh.size(mesh_dim=0) + + # NOTE: is_shard() does not match _StridedShard; see _is_shard_like(). + if tensor.placements[0].is_shard(): + shard_dim = cast(DShard, placement).dim + chunk_size = tensor.size(shard_dim) // num_chunks + offsets[shard_dim] = chunk_size + + return (torch.Size(offsets), tensor._local_tensor.size()) + + +def _get_box_for(tensor: DTensor, idx: int) -> tuple[torch.Size, torch.Size]: + offsets, size = _get_box(tensor) + return (torch.Size([val * idx for val in offsets]), size) + + +def _get_local_box(tensor: DTensor) -> tuple[torch.Size, torch.Size]: + device_mesh = tensor.device_mesh + coord = device_mesh.get_coordinate() + if coord is None: + raise AssertionError + return _get_box_for(tensor, coord[0]) + + +def _create_shard_md_from_dt(dt: DTensor, current_rank: int) -> ShardMetadata: + mesh = dt.device_mesh + if mesh.ndim != 1: + raise AssertionError("Only 1D DeviceMeshes currently handled") + + offsets, sizes = _get_local_box(dt) + return ShardMetadata( + shard_offsets=list(offsets), + shard_sizes=list(sizes), + placement=f"rank:{current_rank}/{dt._local_tensor.device}", + ) + + +def _create_sharded_tensor_md_from_dt( + dt: DTensor, dt_pg: c10d.ProcessGroup +) -> ShardedTensorMetadata: + # This is where it gets tricky, we have to produce a ShardedTensor that has full coverage + # and yet has only one valid shard for the current rank. + + shards_md = [] + my_rank = dist.get_rank(dt_pg) + scapegoat_rank = 0 if my_rank > 0 else 1 + + # NOTE: is_shard() does not match _StridedShard; see _is_shard_like(). + if dt.placements[0].is_shard(): + shard_count = dt_pg.size() + else: + shard_count = 1 + + for i in range(shard_count): + offsets, sizes = _get_box_for(dt, i) + shards_md.append( + ShardMetadata( + shard_offsets=list(offsets), + shard_sizes=list(sizes), + placement=( + f"rank:{scapegoat_rank if i > 0 else my_rank}/{dt._local_tensor.device}" + ), + ) + ) + + return ShardedTensorMetadata( + shards_metadata=shards_md, + size=dt.size(), + tensor_properties=TensorProperties( + dtype=dt.dtype, + layout=dt.layout, + requires_grad=dt.requires_grad, + # ignore memory_format and pin_memory as those are not supported by DT + ), + ) + + +def _get_dt_pg(dt: DTensor) -> c10d.ProcessGroup: + mesh = dt.device_mesh + if mesh.ndim != 1: + raise AssertionError("Only 1D DeviceMeshes currently handled") + return mesh.get_group() + + +def _rewrite_spec_if_needed( + spec: shard_spec.ShardingSpec, tensor: torch.Tensor, rank: int +) -> shard_spec.ShardingSpec: + """ + Rewrite ``spec`` to match the device of ``tensor``. + + FSDP.sharded_optim_state_dict sneakily ships optimizer state to CPU so if the original ShardingSpec + produces CUDA metadata, ST construction bombs. + """ + if not isinstance(spec, ChunkShardingSpec): + return spec + + # let's see if we need + rewrite = False + for p in spec.placements: + p = cast(_remote_device, p) + if p.rank() == rank and p.device() != tensor.device: + rewrite = True + break + if rewrite: + spec = copy.deepcopy(spec) + # pyrefly: ignore [missing-attribute] + for i, placement in enumerate(spec.placements): + placement = cast(_remote_device, placement) + if placement.rank() == rank and placement.device() != tensor.device: + # pyrefly: ignore [missing-attribute] + spec.placements[i] = _remote_device(f"rank:{rank}/{tensor.device}") + + return spec + + +def _chunk_tensor( + tensor: torch.Tensor, + rank: int, + world_size: int, + num_devices_per_node: int, + pg: dist.ProcessGroup, +) -> torch.Tensor: + if type(tensor) is ShardedTensor: + if len(tensor.local_shards()) != 1: + raise AssertionError + + inner_param = tensor.local_tensor() + inner_st = _create_chunk_sharded_tensor( + inner_param, + rank, + world_size, + num_devices_per_node, + pg, + ) + + outer_local_shard = tensor.local_shards()[0] + shards: list[Shard] = [ + Shard(inner_st, copy.deepcopy(outer_local_shard.metadata)) + ] + st_meta = copy.deepcopy(tensor.metadata()) + st_meta.tensor_properties.requires_grad = False + + st_outer = ShardedTensor._init_from_local_shards_and_global_metadata( + shards, + sharded_tensor_metadata=st_meta, + process_group=tensor._process_group, + init_rrefs=False, + ) + return st_outer + elif type(tensor) is DTensor: + device_mesh = tensor.device_mesh + if device_mesh.ndim != 1: + raise AssertionError("Only 1D DeviceMeshes currently handled") + + inner_param = tensor._local_tensor + + inner_st = _create_chunk_sharded_tensor( + inner_param, + rank, + world_size, + torch.accelerator.device_count(), + pg, + ) + + dt_pg = _get_dt_pg(tensor) + # We do this differently here, we create a ST with no local shards then patch it + shards = [ + Shard(inner_st, _create_shard_md_from_dt(tensor, dist.get_rank(dt_pg))) + ] + + st_meta = _create_sharded_tensor_md_from_dt(tensor, dt_pg) + st_meta.tensor_properties.requires_grad = False + + st_outer = ShardedTensor._init_from_local_shards_and_global_metadata( + shards, + sharded_tensor_metadata=st_meta, + process_group=dt_pg, + init_rrefs=False, + ) + + return st_outer + else: + return _create_chunk_sharded_tensor( + tensor, + rank, + world_size, + num_devices_per_node, + pg, + ) + + +def _chunk_dtensor( + tensor: torch.Tensor, + rank: int, + device_mesh: DeviceMesh, +) -> DTensor: + """ + Shard a tensor to chunks along the first dimension. + + The local rank will gets its corresponding chunk as the local tensor to create a DTensor. + """ + root_mesh = device_mesh._get_root_mesh() if device_mesh is not None else None + if root_mesh is None: + raise RuntimeError("No parent device_mesh is found for FSDP device_mesh.") + if root_mesh.ndim < 2: + raise RuntimeError( + f"Found parent device_mesh of ndim={root_mesh.ndim},", + "but meshes must be at least 2D.", + ) + + # We need to explicitly call .detach() to return a new tensor detached from the current graph. + tensor = tensor.detach().clone() + + # When a layer is not involved in TP, then the tensor will not be a DTensor. + # e.g. When a layer is not specified in the parallelize_plan, TP will have no effect on the layer. + # e.g. When you do PairwiseParallel on a 3 layer model, TP will have no effect on the third layer. + if isinstance(tensor, torch.Tensor) and not isinstance(tensor, DTensor): + # For tensors, it is replicated across tp dimension and sharded across FSDP dimension. + # TP is the inner dimension and FSDP is the outer dimension. + # Therefore, shard placements for tensor is (Shard(0), Replicate()). + replicate_placements = [Replicate() for _ in range(root_mesh.ndim)] + shard_placements = [Replicate() for _ in range(root_mesh.ndim)] + shard_placements[0] = DShard(0) # type: ignore[call-overload] + + return DTensor.from_local( + tensor, root_mesh, replicate_placements, run_check=False + ).redistribute( + device_mesh=root_mesh, + placements=shard_placements, + ) + + else: + tp_placements = tensor.placements + tp_placement = tp_placements[0] + + tensor = tensor.to_local() + + # For DTensors, it is sharded across tp dimension first and then sharded across FSDP dimension. + # TP is the inner dimension and FSDP is the outer dimension. + # Therefore, shard placements for tensor is (Shard(0), tp_placement). + # For higher dimensional meshes, it is replicated across other dimensions. For example, with + # HSDP the shard placements for tensor is (Replicate, Shard(0), tp_placement). + replicate_placements = [Replicate() for _ in range(root_mesh.ndim)] + replicate_placements[-1] = tp_placement # type: ignore[call-overload] + shard_placements = [Replicate() for i in range(root_mesh.ndim)] # type: ignore[misc] + shard_placements[-2] = DShard(0) # type: ignore[call-overload] + shard_placements[-1] = tp_placement # type: ignore[call-overload] + + return DTensor.from_local( + tensor, root_mesh, replicate_placements, run_check=False + ).redistribute( + device_mesh=root_mesh, + placements=shard_placements, + ) + + +def _pre_load_state_dict( + tensor: torch.Tensor, +) -> tuple[torch.Tensor, list[Shard]]: + shards = cast(ShardedTensor, tensor).local_shards() + if len(shards) == 1 and type(shards[0].tensor) is ShardedTensor: + inner_tensor = shards[0].tensor + shards = inner_tensor.local_shards() # pyre-ignore[16] + tensor = inner_tensor + + return (tensor, shards if len(shards) > 0 else []) + + +def _all_gather_dtensor( + tensor: DTensor, + parent_mesh: DeviceMesh | None, +) -> torch.Tensor: + """All gather a DTensor in its FSDP dimension and return the local tensor.""" + if parent_mesh != tensor.device_mesh: + raise AssertionError + + placements = list(copy.deepcopy(tensor.placements)) + # FSDP + TP: [Shard(0), tp_placement] -> [Replicate(), tp_placement] + # HSDP + TP: [Replicate(), Shard(0), tp_placement] -> [Replicate(), Replicate(), tp_placement] + for i in range(len(placements) - 1): + placements[i] = Replicate() + tensor = tensor.redistribute( + device_mesh=tensor.device_mesh, + placements=placements, + ) + + return tensor.to_local() + + +class DTensorExtensions(FSDPExtensions): + """ + DTensorExtension is the TensorFlattener extension needed for 2D FSDP + TP. + + This is the implementation for FSDPExtensions defined in + https://github.com/pytorch/pytorch/blob/main/torch/distributed/fsdp/_fsdp_extensions.py + """ + + def __init__(self, device_handle) -> None: + super().__init__() + self.compute_stream = None + self.device_handle = device_handle + # we have to use the dynamo disable this way to disable dynamo as the decorator way would + # trigger build failure with torch deploy... + self.post_unflatten_transform = torch._dynamo.disable( # type: ignore[method-assign] + self.post_unflatten_transform + ) + + def pre_flatten_transform( + self, + tensor: torch.Tensor, + ) -> tuple[torch.Tensor, Any | None]: + return _flatten_tensor(tensor) + + def post_unflatten_transform( + self, tensor: torch.Tensor, param_extension: Any + ) -> torch.Tensor: + stream = self.compute_stream or self.device_handle.current_stream() + with self.device_handle.stream(stream): + # runtime we put the unflattened tensor call on the compute stream since + # the unflattened tensor might contain computations in fwd/bwd where we + # need to sync properly. + # TODO: this is a short term fix and we should make the get_unflat_views + # directly happen in the compute stream. + result = _unflatten_tensor( + tensor, + param_extension, + device_handle=self.device_handle, + compute_stream=self.compute_stream, + ) + _set_fsdp_flattened(result) + return result + + def chunk_tensor( + self, + tensor: torch.Tensor, + rank: int, + world_size: int, + num_devices_per_node: int, + pg: dist.ProcessGroup, + device: torch.device | None = None, + ) -> torch.Tensor: + return _chunk_tensor(tensor, rank, world_size, num_devices_per_node, pg) + + def chunk_dtensor( + self, + tensor: torch.Tensor, + rank: int, + device_mesh: DeviceMesh, + ) -> torch.Tensor: + return _chunk_dtensor(tensor, rank, device_mesh) + + def pre_load_state_dict_transform( + self, + tensor: torch.Tensor, + ) -> tuple[torch.Tensor, list[Shard]]: + return _pre_load_state_dict(tensor) + + def all_gather_dtensor( + self, + tensor: DTensor, + parent_mesh: DeviceMesh | None, + ) -> torch.Tensor: + return _all_gather_dtensor(tensor, parent_mesh) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/input_reshard.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/input_reshard.py new file mode 100644 index 0000000000000000000000000000000000000000..b3c47ddae8460b339202c154e21e36ab81a0b8c4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/input_reshard.py @@ -0,0 +1,107 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates +from functools import partial +from typing import Any + +import torch +from torch.distributed.tensor import DeviceMesh, DTensor, Replicate, Shard +from torch.distributed.tensor.placement_types import _is_shard_like + + +__all__ = [ + "input_reshard", +] + + +def input_reshard( + module: torch.nn.Module, + tp_device_mesh: DeviceMesh, + input_reshard_dim: int | None = None, +) -> torch.nn.Module: + """ + Register hooks to an nn.Module for input resharding, enabling sharding and restoration during backward computation. + + Register hooks to an nn.Module with input resharding so that we can shard + per the given `tp_device_mesh` and `input_reshard_dim` and restore the + input back when recomputing the activations in the backward. The reason + why we can do this is that for Tensor Parallel(TP), the input are same + across all TP ranks. + + Args: + module (:class:`nn.Module`): + Module to be registered with input resharding. + tp_device_mesh (:class:`DeviceMesh`): + Object which describes the mesh topology + of devices for Tensor Parallel. + input_reshard_dim (Optional[int]): + The dimension of where we perform the sharding + of input. If set None, there is no sharding of input. + Default: None + + Return: + A :class:`nn.Module` object registered with TP input resharding. + """ + if input_reshard_dim is None: + return module + + cx: torch.autograd.graph.saved_tensors_hooks | None = None + + def input_reshard_forward_pre_hook(_: torch.nn.Module, _i: tuple[Any, ...]) -> None: + saved_tensor_hooks = torch.autograd.graph.saved_tensors_hooks( + partial(_pack_hook_tp, tp_device_mesh, input_reshard_dim), + partial(_unpack_hook_tp, tp_device_mesh, input_reshard_dim), + ) + saved_tensor_hooks.__enter__() + nonlocal cx + cx = saved_tensor_hooks # type: ignore[name-defined] + + def input_reshard_backward_hook( + _: torch.nn.Module, _i: tuple[Any, ...], _o: Any + ) -> Any: + nonlocal cx + cx.__exit__() # type: ignore[name-defined, union-attr] + + module.register_forward_pre_hook(input_reshard_forward_pre_hook) + module.register_forward_hook(input_reshard_backward_hook) + return module + + +def _pack_hook_tp(mesh: DeviceMesh, input_reshard_dim: int, x: torch.Tensor) -> Any: + """Hook function called after FWD to shard input.""" + if isinstance(x, DTensor) and all(p.is_replicate() for p in x._spec.placements): + return x.redistribute(device_mesh=mesh, placements=[Shard(input_reshard_dim)]) + elif ( + not isinstance(x, DTensor) + and isinstance(x, torch.Tensor) + and x.numel() >= mesh.size() + ): + return ( + DTensor.from_local(x, device_mesh=mesh) + .redistribute(device_mesh=mesh, placements=[Shard(input_reshard_dim)]) + .to_local() + ) + else: + return x + + +def _unpack_hook_tp(mesh: DeviceMesh, input_reshard_dim: int, x: Any) -> torch.Tensor: + """Hook function called before activation recomputing in BWD to restore input.""" + if ( + isinstance(x, DTensor) + and len(x._spec.placements) == 1 + and _is_shard_like(x._spec.placements[0]) + ): + return x.redistribute(device_mesh=mesh, placements=[Replicate()]) + elif ( + not isinstance(x, DTensor) + and isinstance(x, torch.Tensor) + and x.numel() >= mesh.size() + ): + return ( + DTensor.from_local( + x, device_mesh=mesh, placements=[Shard(input_reshard_dim)] + ) + .redistribute(device_mesh=mesh, placements=[Replicate()]) + .to_local() + ) + else: + return x diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/loss.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/loss.py new file mode 100644 index 0000000000000000000000000000000000000000..e846058716dae68e70e7ae4e221db42b5939da76 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/loss.py @@ -0,0 +1,614 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +import contextlib +from typing import cast + +import torch +import torch._prims_common as utils +import torch.distributed._functional_collectives as funcol +import torch.distributed.distributed_c10d as c10d +from torch import Tensor +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor import DTensor, Partial, Replicate, Shard +from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta +from torch.distributed.tensor._ops._embedding_ops import _MaskPartial +from torch.distributed.tensor._ops._math_ops import ( + _skip_dim, + Reduction, + replicate_reduction_dims, +) +from torch.distributed.tensor._ops.utils import normalize_dim +from torch.distributed.tensor.placement_types import Placement + + +aten = torch.ops.aten + + +__all__ = ["loss_parallel"] + + +@contextlib.contextmanager +def loss_parallel(): + """ + A context manager that enables loss parallelism, where efficient parallelized loss computation + can be performed when the input is sharded on the class dimension. Currently only the cross-entropy + loss is supported. + + Within this context manager, one can use :func:`~torch.nn.functional.cross_entropy` or + :class:`~torch.nn.CrossEntropyLoss` as usual, with the following assumptions on the input parameters. + The corresponding ``backward()`` call, if any, also needs to happen under this context manager. + + Both one-dimensional and multi-dimensional ``DeviceMesh`` are supported. On a + multi-dimensional mesh, exactly one mesh dim must shard the class dimension — + that dim is treated as the "TP" dim and can be in any position (not necessarily + last). The remaining mesh dims may be ``Shard`` (e.g. batch-sharded for DP/CP) + or ``Replicate``. + + Args: + input (:class:`DTensor`): + Input logits. Assumed to be sharded on the class dimension on exactly + one mesh dim. + target (Union[:class:`torch.Tensor`, :class:`DTensor`]): + Must be ground truth class indices (class probabilities currently not + supported). Its placements are derived from ``input``'s placements by + (1) replacing the TP dim's ``Shard(class_dim)`` with ``Replicate`` and + (2) shifting any other ``Shard(d)`` with ``d > class_dim`` down by one + (since the class tensor dim is removed from the output):: + + input tensor shape input placements target shape target placements + ---------------------- -------------------------- --------------- ------------------------ + (batch, class) (Shard(0), Shard(1)) (batch,) (Shard(0), Replicate()) + (batch, class) (Replicate(), Shard(1)) (batch,) (Replicate(), Replicate()) + (batch, class, seq) (Shard(2), Shard(1)) (batch, seq) (Shard(1), Replicate()) + (batch, class, seq) (Shard(0), Shard(1)) (batch, seq) (Shard(0), Replicate()) + + A plain :class:`torch.Tensor` is only accepted when the derived target + placements are all ``Replicate`` (e.g. on a one-dimensional mesh); + otherwise a :class:`DTensor` must be passed explicitly so that the + intended sharding is unambiguous. + weight (Union[:class:`torch.Tensor`, :class:`DTensor`], optional): + If given, assumed to be replicated across the ``DeviceMesh``. + label_smoothing: + Currently not supported. + + Returns: + A :class:`DTensor` whose placements depend on ``reduction``: + + - ``reduction="none"`` — per-sample loss; inherits the ``target`` + placements (``Shard`` on batch-parallel dims, ``Replicate`` on TP). + - ``reduction="sum"`` — scalar; ``Replicate`` on the TP dim (the custom + all-reduce makes each rank's local value globally correct over TP), + ``Partial("sum")`` on every other ``Shard`` mesh dim so that the + cross-rank reduction happens lazily on materialization/redistribution, + and ``Replicate`` on ``Replicate`` mesh dims. + - ``reduction="mean"`` — only supported on a one-dimensional mesh; returns + a fully replicated :class:`DTensor`. + + On a one-dimensional mesh all three cases simplify to a fully replicated + :class:`DTensor` (the original behavior). + + .. note:: + ``reduction="mean"`` is only supported on a one-dimensional ``DeviceMesh``. + On a multi-dimensional mesh the per-rank division by ``total_weight`` uses a + local count, so aggregating across non-TP dims does not yield the correct + global mean; use ``reduction="sum"`` or ``"none"`` instead and divide by the + global count yourself if needed. + + Example: + A sharded DTensor is manually created here to showcase the usage. + In practice, it is usually the output of a TP module. + + >>> # xdoctest: +SKIP("distributed") + >>> from torch.distributed.tensor.parallel import loss_parallel + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> device_mesh = init_device_mesh("cuda", (8,)) + >>> input = torch.randn(4, 16, device="cuda", requires_grad=True) + >>> dist_input = distribute_tensor(input, device_mesh, placements=[Shard(1)]) + >>> target = torch.randint(16, (4,), device="cuda") + >>> with loss_parallel(): + >>> loss = F.cross_entropy(dist_input, target, reduction="mean") + >>> loss.backward() + >>> ... + """ + _enable_custom_loss_ops() + + yield + + _disable_custom_loss_ops() + + +# Return the mesh_dim whose placement is Shard(dim). Exactly one such mesh dim +# (the "TP" dim) is required; other mesh dims must be Shard on a different +# tensor dim or Replicate. Partial on a non-TP dim is rejected so that +# reduction="none" cannot silently accept garbage (replicate_reduction_dims +# would otherwise rewrite Partial -> Replicate in target_placements). +def _find_all_reduce_mesh_dim(placements: tuple[Placement, ...], dim: int) -> int: + shard_mesh_dims = [i for i, p in enumerate(placements) if p.is_shard(dim)] + if len(shard_mesh_dims) != 1: + raise ValueError( + f"loss_parallel() requires exactly one mesh dim to shard tensor " + f"dimension {dim}, but got placements {placements} with " + f"{len(shard_mesh_dims)} such mesh dim(s)." + ) + mesh_dim = shard_mesh_dims[0] + for i, p in enumerate(placements): + if i == mesh_dim: + continue + if not (p.is_shard() or p.is_replicate()): + raise ValueError( + f"loss_parallel() expects non-TP mesh dims to be " + f"Shard or Replicate, got {p} at dim {i}." + ) + return mesh_dim + + +def _cast_to_dtensor( + tensor, placements: tuple[Placement, ...], mesh: DeviceMesh +) -> DTensor: + if isinstance(tensor, DTensor): + if tensor.placements == placements: + return tensor + else: + raise RuntimeError(f"Expected {placements} but got {tensor.placements}.") + elif isinstance(tensor, torch.Tensor): + if any(p.is_shard() for p in placements): + # A plain torch.Tensor with Shard placements is ambiguous: it could + # be a full global tensor the user expects us to shard, or an + # already-local slice. Require a DTensor so the intent is explicit. + raise ValueError( + f"loss_parallel() requires a DTensor (not a plain torch.Tensor) " + f"when the derived placements {placements} contain Shard — this " + f"happens on a multi-dimensional mesh with a batch-sharded non-TP " + f"dim. Wrap the tensor with DTensor.from_local or distribute_tensor " + f"to make the sharding explicit." + ) + return DTensor.from_local( + tensor, device_mesh=mesh, placements=placements, run_check=False + ) + else: + raise TypeError(f"Unsupported type {type(tensor)}") + + +def _propagate_tensor_meta( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> TensorMeta: + op_info = DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) + if op_info.schema is None: + raise AssertionError( + "op_info.schema should not be None after unwrap_to_op_info" + ) + tensor_meta = DTensor._op_dispatcher.sharding_propagator._propagate_tensor_meta( + op_info.schema + ) + if isinstance(tensor_meta, TensorMeta): + return tensor_meta + elif isinstance(tensor_meta, tuple): + # pyrefly: ignore [bad-return] + return tensor_meta[0] + else: + raise RuntimeError(f"Unexpected tensor meta type: {type(tensor_meta)}.") + + +# NOTE: The implementation follows torch._decomp.decomposition._log_softmax, +# with all_reduce manually inserted to perform distributed computation. +def _log_softmax(x, dim, half_to_float, mesh, mesh_dim): + if half_to_float: + if x.dtype != torch.half: + raise AssertionError + computation_dtype, result_dtype = utils.elementwise_dtypes( + x, type_promotion_kind=utils.ELEMENTWISE_TYPE_PROMOTION_KIND.DEFAULT + ) + x = x.to(dtype=computation_dtype, memory_format=torch.contiguous_format) + if x.numel() == 0: + shifted = x + else: + x_max = torch.amax(x, dim, keepdim=True) + x_max = funcol.all_reduce( + x_max, reduceOp=c10d.ReduceOp.MAX.name, group=(mesh, mesh_dim) + ) + shifted = x - x_max + shifted_sumexp = torch.sum(torch.exp(shifted), dim, keepdim=True) + shifted_sumexp = funcol.all_reduce( + shifted_sumexp, reduceOp=c10d.ReduceOp.SUM.name, group=(mesh, mesh_dim) + ) + shifted_logsumexp = torch.log(shifted_sumexp) + result = shifted - shifted_logsumexp + if not half_to_float: + result = result.to(result_dtype) + return result + + +def _log_softmax_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + x = cast(DTensor, args[0]) + dim = cast(int, args[1]) + half_to_float = cast(bool, args[2]) + + spec = x._spec + dim = normalize_dim(dim, x.dim()) + mesh_dim = _find_all_reduce_mesh_dim(spec.placements, dim) + + output_tensor_meta = _propagate_tensor_meta(op_call, args, kwargs) + + res = _log_softmax(x._local_tensor, dim, half_to_float, spec.mesh, mesh_dim) + + res_spec = DTensorSpec( + spec.mesh, + spec.placements, + tensor_meta=output_tensor_meta, + ) + + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + res, + res_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=res.requires_grad, + ) + + +# NOTE: As explained below at _nll_loss_and_log_softmax_backward, the +# _log_softmax_backward_handler does not actually do any computation. +def _log_softmax_backward_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + grad_output = cast(DTensor, args[0]) + input_dtype = cast(torch.dtype, args[3]) + return grad_output.to(input_dtype) + + +# NOTE: The implementation follows torch._decomp.decomposition._nll_loss_forward, +# with customized communication inserted to perform distributed computation. +def _nll_loss_forward( + x: Tensor, + target: Tensor, + weight: Tensor | None, + local_weight: Tensor | None, + reduction: int, + ignore_index: int, + input_shape: torch.Size, + channel_dim: int, + mesh: DeviceMesh, + mesh_dim: int, +) -> tuple[Tensor, Tensor]: + n_dims = x.dim() + channel_dim = 1 + if n_dims < 2: + channel_dim = 0 + + def _weight_view(weight: Tensor) -> Tensor: + if n_dims > 1: + shape = [ + 1, + ] * n_dims + shape[channel_dim] = weight.shape[0] + w = weight.view(shape) + else: + w = weight + return w + + if weight is not None: + w = _weight_view(weight) + if local_weight is None: + raise AssertionError + local_w = _weight_view(local_weight) + x = x * local_w + safe_target = torch.where(target != ignore_index, target, 0) + safe_target_ = safe_target.unsqueeze(channel_dim) + + # The following code block is a distributed version of + # result = -torch.gather(self, channel_dim, safe_target_).squeeze(channel_dim) + partial_placement = _MaskPartial(offset_shape=input_shape, offset_dim=channel_dim) + safe_target_partial_ = partial_placement._partition_value( + safe_target_, mesh, mesh_dim + ) + result_partial = torch.gather(x, channel_dim, safe_target_partial_) + # an all_reduce happens here + result_reduced = partial_placement._reduce_value(result_partial, mesh, mesh_dim) + result = -result_reduced.squeeze(channel_dim) + + result = torch.where(target != ignore_index, result, 0) + + if reduction == Reduction.NONE.value and n_dims > 1: + total_weight = x.new_full((), 0.0) + return result, total_weight + + if weight is not None: + new_shape = list(x.shape) + new_shape[channel_dim] = -1 + # pyrefly: ignore [unbound-name] + w = w.expand(new_shape) + wsum = torch.gather(w, channel_dim, safe_target_).squeeze(channel_dim) + wsum = torch.where(target != ignore_index, wsum, 0) + total_weight = wsum.sum() + else: + total_weight = (target != ignore_index).sum().to(x) + + # On an N-D mesh, result.sum() is a per-rank sum over the local batch; + # the caller emits Partial("sum") on non-TP Shard dims so the cross-rank + # reduction happens when the output is later materialized/redistributed. + # reduction="mean" is rejected in the handler for N-D mesh because + # total_weight is a per-rank local count. + if reduction == Reduction.SUM.value: + result = result.sum() + elif reduction == Reduction.MEAN.value: + result = result.sum() / total_weight + + return result, total_weight + + +def _nll_loss_forward_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + x = cast(DTensor, args[0]) + target = args[1] + weight = args[2] + reduction = cast(int, args[3]) + ignore_index = cast(int, args[4]) + + channel_dim = 1 if x.dim() >= 2 else 0 + spec = x._spec + mesh_dim = _find_all_reduce_mesh_dim(spec.placements, channel_dim) + + # Check user input: if target and weight are not DTensors, convert them to DTensors; + # if they are DTensors, check that they have the desired placements. + target_placements = _skip_dim( + replicate_reduction_dims(spec.placements, [channel_dim]), channel_dim + ) + all_replicate_placements = (Replicate(),) * spec.mesh.ndim + target = _cast_to_dtensor(target, target_placements, spec.mesh) + local_weight = None + if weight is not None: + weight = _cast_to_dtensor(weight, all_replicate_placements, spec.mesh) + # For local computation, both (replicated) weight and (sharded) local_weight + # are needed in _nll_loss_forward(). local_weight is generated here using + # DTensor API, without incurring any communication. + sharded_placements = [ + Shard(0) if i == mesh_dim else Replicate() for i in range(spec.mesh.ndim) + ] + local_weight = weight.redistribute(spec.mesh, sharded_placements)._local_tensor + if local_weight.shape[0] != x._local_tensor.shape[channel_dim]: + raise AssertionError + + if reduction == Reduction.NONE.value: + output_placements = target_placements + else: + # reduction="mean" is not yet correct on multi-dimensional mesh: the + # local division by total_weight uses a per-rank count, so summing + # across non-TP dims does not yield the global mean. See the NOTE in + # _nll_loss_forward. + if reduction == Reduction.MEAN.value and spec.mesh.ndim > 1: + raise NotImplementedError( + "loss_parallel() with reduction='mean' is only supported on " + "one-dimensional DeviceMesh; got mesh with ndim=" + f"{spec.mesh.ndim}. Use reduction='sum' or 'none' instead." + ) + # On the shard dim (TP), the custom all-reduce produces a globally + # correct result -> Replicate. On other dims (e.g. DP/CP), each rank + # only reduced its local batch -> Partial(sum). + out_placements_list: list[Placement] = [] + for i, p in enumerate(spec.placements): + if i == mesh_dim: + out_placements_list.append(Replicate()) + elif p.is_shard(): + out_placements_list.append(Partial()) + else: + out_placements_list.append(p) + output_placements = tuple(out_placements_list) + + # tensor inputs to _propagate_tensor_meta need to be DTensors + # pyrefly: ignore [bad-assignment] + args = list(args) + # pyrefly: ignore [unsupported-operation] + args[1], args[2] = target, weight + output_tensor_meta = _propagate_tensor_meta(op_call, tuple(args), kwargs) + + result, total_weight = _nll_loss_forward( + x._local_tensor, + target._local_tensor, + weight._local_tensor if weight is not None else None, + local_weight, + reduction, + ignore_index, + x.shape, + channel_dim, + spec.mesh, + mesh_dim, + ) + out_spec = DTensorSpec(spec.mesh, output_placements, tensor_meta=output_tensor_meta) + + return ( + # pyrefly: ignore [bad-argument-type] + DTensor( + # pyrefly: ignore [bad-argument-count] + result, + out_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=result.requires_grad, + ), + total_weight, + ) + + +# NOTE: The backward computation of cross_entropy goes through two steps: +# backward for nll_loss and then backward for log_softmax. In loss parallel, +# the two steps are fused into the following function (called by _nll_loss_backward_handler) +# to avoid communication when target contains class indices not class probabilities. +# Also note that the _log_softmax_backward_handler does not perform computation. +# The implementation resembles _nll_loss_backward and _log_softmax_backward_data +# from torch._decomp.decomposition. +def _nll_loss_and_log_softmax_backward( + grad_output: Tensor, + x: Tensor, + target: Tensor, + weight: Tensor | None, + reduction: int, + ignore_index: int, + total_weight: Tensor, + input_shape: torch.Size, + channel_dim: int, + mesh: DeviceMesh, + mesh_dim: int, +) -> Tensor: + channel_dim = 0 if x.dim() < 2 else 1 + if reduction == Reduction.MEAN.value: + grad_output = grad_output / total_weight + + target = target.unsqueeze(channel_dim) + safe_target = torch.where(target != ignore_index, target, 0) + grad_input = torch.zeros_like(x) + + # The following code block is a distributed version of + # grad_input = torch.scatter(grad_input, channel_dim, safe_target, -1.0) + partial_placement = _MaskPartial(offset_shape=input_shape, offset_dim=channel_dim) + safe_target = safe_target.squeeze(channel_dim).flatten() + masked_safe_target = partial_placement._partition_value(safe_target, mesh, mesh_dim) + # only update grad_input to -1 if not masked + if partial_placement.mask_buffer.data is None: + raise AssertionError + grad_update = partial_placement.mask_buffer.data.to(grad_input.dtype) - 1.0 + arange_1d = torch.arange( + masked_safe_target.shape[0], device=masked_safe_target.device + ) + # The first two cases with x.dim() <= 2 are for aten.nll_loss_backward.default; + # the last case is for aten.nll_loss2d_backward.default. + if x.dim() == 1: + grad_input[masked_safe_target] = grad_update + elif x.dim() == 2: + grad_input[arange_1d, masked_safe_target] = grad_update + else: + grad_input_t = grad_input.transpose(channel_dim, -1) + intermidate_shape = grad_input_t.shape + grad_input_2d = grad_input_t.reshape(-1, x.shape[channel_dim]) + grad_input_2d[arange_1d, masked_safe_target] = grad_update + grad_input = grad_input_2d.view(intermidate_shape).transpose(channel_dim, -1) + + if grad_input.dim() > grad_output.dim() > 0: + grad_output = grad_output.unsqueeze(channel_dim) + + if weight is not None: + new_shape = [1 for _ in range(x.dim())] + new_shape[channel_dim] = weight.shape[0] + weight = weight.reshape(new_shape) + # In order for fused computation to work, the following line is rewritten. + # grad_output = grad_output * weight + new_shape = list(x.shape) + new_shape[channel_dim] = -1 + w = weight.expand(new_shape) + w_target = torch.gather(w, channel_dim, target) + grad_output = grad_output * w_target + + grad_output = torch.where(target != ignore_index, grad_output, 0) + + # NOTE: Instead of directly returning the grad_input as grad_output for log_softmax, + # here we perform backward computation for log_softmax altogether to avoid the + # otherwise extra all_gather communication. + # return grad_input * grad_output + return (grad_input + torch.exp(x)) * grad_output + + +def _nll_loss_backward_handler( + op_call: torch._ops.OpOverload, + args: tuple[object, ...], + kwargs: dict[str, object], +) -> object: + grad_output = cast(DTensor, args[0]) + x = cast(DTensor, args[1]) + target = args[2] + weight = args[3] + reduction = cast(int, args[4]) + ignore_index = cast(int, args[5]) + total_weight = cast(Tensor, args[6]) + + channel_dim = 1 if x.dim() >= 2 else 0 + spec = x._spec + mesh_dim = _find_all_reduce_mesh_dim(spec.placements, channel_dim) + + # if target and weight are not DTensors, convert them to DTensors + target_placements = _skip_dim( + replicate_reduction_dims(spec.placements, [channel_dim]), channel_dim + ) + all_replicate_placements = (Replicate(),) * spec.mesh.ndim + target = _cast_to_dtensor(target, target_placements, spec.mesh) + if weight is not None: + weight = _cast_to_dtensor(weight, all_replicate_placements, spec.mesh) + + # For reduction="none", autograd may deliver grad_output with different + # placements than the forward output (e.g. Replicate on a batch-sharded dim), + # causing a shape mismatch with the local batch. Redistribute it to match + # target_placements — which is what the forward output uses for reduction="none". + # For reduction="sum"/"mean" the forward output is a 0-D scalar, so a placement + # mismatch (e.g. Replicate vs. Partial("sum")) cannot produce a shape mismatch + # against x._local_tensor and no redistribute is needed here. + if reduction == Reduction.NONE.value: + grad_output = grad_output.redistribute(spec.mesh, target_placements) + + # tensor inputs to _propagate_tensor_meta need to be DTensors + # pyrefly: ignore [bad-assignment] + args = list(args) + # pyrefly: ignore [unsupported-operation] + args[0] = grad_output + # pyrefly: ignore [unsupported-operation] + args[2], args[3] = target, weight + # pyrefly: ignore [unsupported-operation] + args[6] = _cast_to_dtensor(total_weight, all_replicate_placements, spec.mesh) + output_tensor_meta = _propagate_tensor_meta(op_call, tuple(args), kwargs) + + result = _nll_loss_and_log_softmax_backward( + grad_output._local_tensor, + x._local_tensor, + target._local_tensor, + weight._local_tensor if weight is not None else None, + reduction, + ignore_index, + total_weight, + x.shape, + channel_dim, + spec.mesh, + mesh_dim, + ) + # the output sharding is the same as input sharding: Shard(channel_dim) on mesh_dim + out_spec = DTensorSpec( + spec.mesh, + spec.placements, + tensor_meta=output_tensor_meta, + ) + + # pyrefly: ignore [bad-argument-type] + return DTensor( + # pyrefly: ignore [bad-argument-count] + result, + out_spec, + # pyrefly: ignore [unexpected-keyword] + requires_grad=result.requires_grad, + ) + + +customized_loss_ops = { + aten._log_softmax.default: _log_softmax_handler, + aten._log_softmax_backward_data.default: _log_softmax_backward_handler, + aten.nll_loss_forward.default: _nll_loss_forward_handler, + aten.nll_loss2d_forward.default: _nll_loss_forward_handler, + aten.nll_loss_backward.default: _nll_loss_backward_handler, + aten.nll_loss2d_backward.default: _nll_loss_backward_handler, +} + + +def _enable_custom_loss_ops(): + DTensor._op_dispatcher._custom_op_handlers.update(customized_loss_ops) + + +def _disable_custom_loss_ops(): + for custom_op in customized_loss_ops: + DTensor._op_dispatcher._custom_op_handlers.pop(custom_op) diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/style.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/style.py new file mode 100644 index 0000000000000000000000000000000000000000..22169c2a4e428c30b1f9c1c444f1b72f585ccb1f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/parallel/style.py @@ -0,0 +1,823 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates +from abc import ABC, abstractmethod +from functools import partial +from typing import Any + +import torch +import torch.nn as nn +from torch.distributed.tensor import ( + DeviceMesh, + distribute_module, + distribute_tensor, + DTensor, + Replicate, + Shard, +) +from torch.distributed.tensor.placement_types import Placement + + +__all__ = [ + "ParallelStyle", + "RowwiseParallel", + "SequenceParallel", + "ColwiseParallel", + "PrepareModuleInput", + "PrepareModuleInputOutput", + "PrepareModuleOutput", +] + + +class ParallelStyle(ABC): + """ + The parallel style contract defines how the module or submodule should be parallelized. + + It only defines the ``apply`` method for ``parallelize_module`` to use, this allows maximum + flexibility for different kind of style implementations. + """ + + src_data_rank: int | None = 0 + + @abstractmethod + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: ... + + +class ColwiseParallel(ParallelStyle): + """ + Partition a compatible nn.Module in a column-wise fashion. Currently supports nn.Linear and nn.Embedding. + Users can compose it together with RowwiseParallel to achieve the sharding of more complicated modules. + (i.e. MLP, Attention) + + Keyword Args: + input_layouts (Placement, optional): + The DTensor layout of input tensor for the nn.Module, this is used to annotate the input tensor to + become a DTensor. If not specified, we assume the input tensor to be replicated. + output_layouts (Placement, optional): + The DTensor layout of the output for the nn.Module, this is used to ensure the output of the nn.Module + with the user desired layout. If not specified, the output tensor is sharded on the last dimension. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module output, default: True. + Returns: + A :class:`ParallelStyle` object that represents Colwise sharding of the nn.Module. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, ColwiseParallel + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> m = Model(...) # m is a nn.Module that contains a "w1" nn.Linear submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # By default, the input of the "w1" Linear will be converted to Replicated DTensor + >>> # and the output of "w1" will return :class:`torch.Tensor` that shards on the last dim. + >>> + >>> sharded_mod = parallelize_module(m, tp_mesh, {"w1": ColwiseParallel()}) + >>> ... + + .. note:: By default ``ColwiseParallel`` output is sharded on the last dimension if the ``output_layouts`` not + specified, if there're operators that require specific tensor shape (i.e. before the paired ``RowwiseParallel``), + keep in mind that if the output is sharded the operator might need to be adjusted to the sharded size. + """ + + def __init__( + self, + *, + input_layouts: Placement | None = None, + output_layouts: Placement | None = None, + use_local_output: bool = True, + ): + super().__init__() + self.input_layouts = (input_layouts or Replicate(),) + self.output_layouts = (output_layouts or Shard(-1),) + # colwise linear runtime sharding (desired sharding): + # 1. requires replicate input + # 2. shard output on last dim + self.desired_input_layouts = (Replicate(),) + self.use_local_output = use_local_output + + @staticmethod + def _prepare_input_fn( + input_layouts, desired_input_layouts, mod, inputs, device_mesh + ): + # TODO: figure out dynamo support for instance method and switch this to instance method + + # annotate module input placements/sharding with input_layouts + input_tensor = inputs[0] + if not isinstance(input_tensor, DTensor): + input_tensor = DTensor.from_local( + input_tensor, + device_mesh, + input_layouts, + run_check=False, + ) + + # transform the input layouts to the desired layouts of ColwiseParallel + if input_layouts != desired_input_layouts: + input_tensor = input_tensor.redistribute( + placements=desired_input_layouts, async_op=True + ) + return input_tensor + + def _partition_linear_fn(self, name, module, device_mesh): + # colwise shard weight/bias to Shard(0), weight be Shard(0) + # means Colwise as Linear is input * weight^T + bias, where + # weight would become Shard(1) + for name, param in module.named_parameters(): + dist_param = nn.Parameter( + distribute_tensor( + param, device_mesh, [Shard(0)], src_data_rank=self.src_data_rank + ), + requires_grad=param.requires_grad, + ) + module.register_parameter(name, dist_param) + + def _partition_embedding_fn(self, name, module, device_mesh): + # colwise shard embedding.weight is straight forward as Shard(1) + for name, param in module.named_parameters(): + dist_param = nn.Parameter( + distribute_tensor( + param, device_mesh, [Shard(1)], src_data_rank=self.src_data_rank + ), + requires_grad=param.requires_grad, + ) + module.register_parameter(name, dist_param) + + @staticmethod + def _prepare_output_fn(output_layouts, use_local_output, mod, outputs, device_mesh): + # outputs is a shard on last dimension DTensor, i.e. Shard(-1) + if outputs.placements != output_layouts: + outputs = outputs.redistribute(placements=output_layouts, async_op=True) + # back to local tensor + return outputs.to_local() if use_local_output else outputs + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + if isinstance(module, nn.Linear): + partition_fn = self._partition_linear_fn + elif isinstance(module, nn.Embedding): + partition_fn = self._partition_embedding_fn + else: + raise NotImplementedError( + "ColwiseParallel currently only support nn.Linear and nn.Embedding!" + ) + + return distribute_module( + module, + device_mesh, + partition_fn, + partial( + self._prepare_input_fn, + self.input_layouts, + self.desired_input_layouts, + ), + partial( + self._prepare_output_fn, self.output_layouts, self.use_local_output + ), + ) + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + tmpstr += f"input_layouts={self.input_layouts}, " + tmpstr += f"output_layouts={self.output_layouts}, " + tmpstr += f"use_local_output={self.use_local_output}" + tmpstr += ")" + return tmpstr + + +class RowwiseParallel(ParallelStyle): + """ + Partition a compatible nn.Module in a row-wise fashion. Currently supports nn.Linear and nn.Embedding. + Users can compose it with ColwiseParallel to achieve the sharding of more complicated modules. + (i.e. MLP, Attention) + + Keyword Args: + input_layouts (Placement, optional): + The DTensor layout of input tensor for the nn.Module, this is used to annotate the input tensor to + become a DTensor. If not specified, we assume the input tensor to be sharded on the last dimension. + output_layouts (Placement, optional): + The DTensor layout of the output for the nn.Module, this is used to ensure the output of the nn.Module + with the user desired layout. If not specified, the output tensor is replicated. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module output, default: True. + Returns: + A :class:`ParallelStyle` object that represents Rowwise sharding of the nn.Module. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, RowwiseParallel + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> m = Model(...) # m is a nn.Module that contains a "w2" nn.Linear submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # By default, the input of the "w2" Linear will be converted to DTensor that shards on the last dim + >>> # and the output of "w2" will return a replicated :class:`torch.Tensor`. + >>> + >>> sharded_mod = parallelize_module(m, tp_mesh, {"w2": RowwiseParallel()}), + >>> ... + """ + + def __init__( + self, + *, + input_layouts: Placement | None = None, + output_layouts: Placement | None = None, + use_local_output: bool = True, + ): + super().__init__() + self.input_layouts = (input_layouts or Shard(-1),) + self.output_layouts = (output_layouts or Replicate(),) + self.use_local_output = use_local_output + + @staticmethod + def _prepare_input_fn( + input_layouts, desired_input_layouts, mod, inputs, device_mesh + ): + input_tensor = inputs[0] + if not isinstance(input_tensor, DTensor): + input_tensor = DTensor.from_local( + input_tensor, + device_mesh, + input_layouts, + run_check=False, + ) + + if input_layouts != desired_input_layouts: + input_tensor = input_tensor.redistribute( + placements=desired_input_layouts, async_op=True + ) + return input_tensor + + def _partition_linear_fn(self, name, module, device_mesh): + # Rowwise shard weight to Shard(1), bias to Replicate(), weight be Shard(1) + # means Rowwise as nn.Linear is input * weight^T + bias, where + # weight would become Shard(0) + module.register_parameter( + "weight", + nn.Parameter( + distribute_tensor( + module.weight, + device_mesh, + [Shard(1)], + src_data_rank=self.src_data_rank, + ), + requires_grad=module.weight.requires_grad, + ), + ) + if getattr(module, "bias", None) is not None: + # The Linear module has bias + module.register_parameter( + "bias", + nn.Parameter( + distribute_tensor( + module.bias, + device_mesh, + [Replicate()], + src_data_rank=self.src_data_rank, + ), + requires_grad=module.bias.requires_grad, + ), + ) + + def _partition_embedding_fn(self, name, module, device_mesh): + # rowwise shard embedding.weight is Shard(0) + for name, param in module.named_parameters(): + dist_param = nn.Parameter( + distribute_tensor( + param, device_mesh, [Shard(0)], src_data_rank=self.src_data_rank + ), + requires_grad=param.requires_grad, + ) + module.register_parameter(name, dist_param) + + @staticmethod + def _prepare_output_fn(output_layouts, use_local_output, mod, outputs, device_mesh): + # Rowwise sharding produces partial output, depending on output layouts: + # 1. to replicate -> allreduce + # 2. to shard -> reduce_scatter + if outputs.placements != output_layouts: + outputs = outputs.redistribute(placements=output_layouts, async_op=True) + # back to local tensor if use_local_output is True + return outputs.to_local() if use_local_output else outputs + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + if isinstance(module, nn.Linear): + partition_fn = self._partition_linear_fn + # rowwise linear runtime sharding requires input tensor shard on last dim + self.desired_input_layouts: tuple[Placement, ...] = (Shard(-1),) + elif isinstance(module, nn.Embedding): + partition_fn = self._partition_embedding_fn + # rowwise embedding runtime sharding requires input tensor replicated + self.desired_input_layouts = (Replicate(),) + else: + raise NotImplementedError( + "RowwiseParallel currently only support nn.Linear and nn.Embedding!" + ) + + return distribute_module( + module, + device_mesh, + partition_fn, + partial( + self._prepare_input_fn, + self.input_layouts, + self.desired_input_layouts, + ), + partial( + self._prepare_output_fn, self.output_layouts, self.use_local_output + ), + ) + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + tmpstr += f"input_layouts={self.input_layouts}, " + tmpstr += f"output_layouts={self.output_layouts}, " + tmpstr += f"use_local_output={self.use_local_output}" + tmpstr += ")" + return tmpstr + + +class SequenceParallel(ParallelStyle): + """ + SequenceParallel replicates a compatible ``nn.Module`` parameters and runs the sharded computation with + input sharded on the sequence dimension. This currently supports ``nn.LayerNorm``, ``nn.Dropout``, and the + `RMSNorm python implementation `__ + + This style implements the operation that is described in the paper + `Reducing Activation Recomputation in Large Transformer Models `__ + + If the input passed in to this ``nn.Module`` is a :class:`torch.Tensor`, it assumes that the input is already sharded + on the sequence dimension and converts the input to a :class:`DTensor` sharded on the sequence dimension. If the input + passed in to this ``nn.Module`` is already a :class:`DTensor` but is not sharded on the sequence dimension, it would + redistribute the input to be sharded on the sequence dimension. + + The output of the ``nn.Module`` will be sharded on the sequence dimension. + + Keyword Args: + sequence_dim (int, optional): + The sequence dimension of the input tensor for the ``nn.Module``, this is used to annotate the input tensor to + become a DTensor that is sharded on the sequence dimension, default: 1. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module output, default: False. + Returns: + A :class:`ParallelStyle` object that represents Sequence Parallel of the ``nn.Module``. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, SequenceParallel + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> m = Model(...) # m is a nn.Module that contains a "norm" nn.LayerNorm submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # By default, the input of the "norm" will be converted to DTensor that shards on the sequence dim + >>> # and the output of "norm" will return a sharded on sequence dimension :class:`DTensor`. + >>> + >>> sharded_mod = parallelize_module(m, tp_mesh, {"norm": SequenceParallel()}), + >>> ... + + .. note:: SequenceParallel style assumes ones initialization if there are weights in the nn.Module (i.e. + ``nn.LayerNorm`` or ``RMSNorm``, and they by default have ones initialization). If you have custom + inits for the weights on those modules, you need to broadcast the weights before/after parallelizing + to ensure that they are replicated. + """ + + def __init__(self, *, sequence_dim: int = 1, use_local_output: bool = False): + super().__init__() + self.sequence_sharding = (Shard(sequence_dim),) + self.use_local_output = use_local_output + + def _replicate_module_fn( + self, name: str, module: nn.Module, device_mesh: DeviceMesh + ): + for p_name, param in module.named_parameters(): + # simple replication with fixed ones_ init from LayerNorm/RMSNorm, which allow + # us to simply just use from_local + replicated_param = torch.nn.Parameter( + DTensor.from_local(param, device_mesh, [Replicate()], run_check=False) + ) + module.register_parameter(p_name, replicated_param) + + @staticmethod + def _prepare_input_fn(sequence_sharding, mod, inputs, device_mesh): + input_tensor = inputs[0] + if isinstance(input_tensor, DTensor): + # if the passed in input DTensor is not sharded on the sequence dim, we need to redistribute it + if input_tensor.placements != sequence_sharding: + input_tensor = input_tensor.redistribute( + placements=sequence_sharding, async_op=True + ) + return input_tensor + elif isinstance(input_tensor, torch.Tensor): + # assume the input passed in already sharded on the sequence dim and create the DTensor + return DTensor.from_local( + input_tensor, device_mesh, sequence_sharding, run_check=False + ) + else: + raise ValueError( + f"expecting input of {mod} to be a torch.Tensor or DTensor, but got {input_tensor}" + ) + + @staticmethod + def _prepare_output_fn(use_local_output, mod, outputs, device_mesh): + return outputs.to_local() if use_local_output else outputs + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + return distribute_module( + module, + device_mesh, + self._replicate_module_fn, + partial(self._prepare_input_fn, self.sequence_sharding), + partial(self._prepare_output_fn, self.use_local_output), + ) + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + if len(self.sequence_sharding) == 1: + tmpstr += f"sequence_dim={self.sequence_sharding[0].dim}, " + tmpstr += f"use_local_output={self.use_local_output}" + tmpstr += ")" + return tmpstr + + +class PrepareModuleInput(ParallelStyle): + """ + Configure the nn.Module's inputs to convert the input tensors of the nn.Module to DTensors at runtime according to + ``input_layouts``, and perform layout redistribution according to the ``desired_input_layouts``. + + Keyword Args: + input_layouts (Union[Placement, Tuple[Optional[Placement]]]): + The DTensor layouts of input tensors for the nn.Module, this is used to convert the input tensors to + DTensors. If some inputs are not torch.Tensor or no need to convert to DTensors, ``None`` need to be specified + as a placeholder. default: None. + desired_input_layouts (Union[Placement, Tuple[Optional[Placement]]]): + The desired DTensor layout of input tensors for the nn.Module, this is used to ensure the inputs of the nn.Module + have the desired DTensor layouts. This argument needs to have the same length with ``input_layouts``. default: None. + input_kwarg_layouts (Dict[str, Placement]): + The DTensor layouts of input kwargs for the nn.Module, this is used to convert the input kwarg tensors to DTensors. + default: None + desired_input_kwarg_layouts: (Dict[str, Placement]): + The desired DTensor layout of input kwargs for the nn.Module, this is used to ensure the inputs of the nn.Module + have the desired DTensor layouts. default: None. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module inputs, default: False. + Returns: + A :class:`ParallelStyle` object that prepares the sharding layouts of the nn.Module's inputs. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, PrepareModuleInput + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> block = TransformerBlock(...) # block is a nn.Module that contains an "attn" Attention submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # According to the style specified below, the first input of attn will be annotated to Sharded DTensor + >>> # and then redistributed to Replicated DTensor. + >>> parallelize_module( + >>> block, # this can be a submodule or module + >>> tp_mesh, + >>> parallelize_plan={ + >>> "attn": PrepareModuleInput( + >>> input_layouts=(Shard(0), None, None, ...), + >>> desired_input_layouts=(Replicate(), None, None, ...) + >>> ), + >>> } + >>> ) + """ + + def __init__( + self, + *, + input_layouts: Placement | tuple[Placement | None, ...] | None = None, + desired_input_layouts: Placement | tuple[Placement | None, ...] | None = None, + input_kwarg_layouts: dict[str, Placement] | None = None, + desired_input_kwarg_layouts: dict[str, Placement] | None = None, + use_local_output: bool = False, + ): + self.input_layouts = ( + (input_layouts,) if isinstance(input_layouts, Placement) else input_layouts + ) + self.desired_input_layouts = ( + (desired_input_layouts,) + if isinstance(desired_input_layouts, Placement) + else desired_input_layouts + ) + self.use_local_output = use_local_output + if self.input_layouts is not None: + if self.desired_input_layouts is None: + raise AssertionError("desired module inputs should not be None!") + if len(self.input_layouts) != len(self.desired_input_layouts): + raise AssertionError( + "input_layouts and desired_input_layouts should have same length!" + ) + self.with_kwargs = input_kwarg_layouts is not None + self.input_kwarg_layouts = input_kwarg_layouts or {} + self.desired_input_kwarg_layouts = desired_input_kwarg_layouts or {} + if self.with_kwargs: + if len(self.input_kwarg_layouts) != len(self.desired_input_kwarg_layouts): + raise AssertionError( + "input_kwarg_layouts and desired_input_kwarg_layouts should have same length!" + ) + + def _prepare_input_arg( + self, + input: Any, + mesh: DeviceMesh, + input_layout: Placement | None, + desired_layout: Placement | None, + ): + if input_layout is not None: + if isinstance(input, DTensor): + # TODO: re-enable the check once we fix the compile path + # assert inp.placements[0] == input_layout + dt_inp = input + else: + if not isinstance(input, torch.Tensor): + raise AssertionError("expecting input to be a torch.Tensor!") + dt_inp = DTensor.from_local( + input, mesh, (input_layout,), run_check=False + ) + + if desired_layout is not None and input_layout != desired_layout: + dt_inp = dt_inp.redistribute(placements=(desired_layout,)) + + return dt_inp.to_local() if self.use_local_output else dt_inp + else: + return input + + def _prepare_input_fn(self, inputs, device_mesh): + if self.input_layouts is None: + return inputs + prepared_inputs = [] + if not isinstance(inputs, tuple): + inputs = (inputs,) + if len(inputs) != len(self.input_layouts): + raise ValueError("module inputs and input_layouts should have same length!") + + if self.desired_input_layouts is None: + raise AssertionError("desired module inputs should not be None!") + + for inp, input_layout, desired_layout in zip( + inputs, self.input_layouts, self.desired_input_layouts + ): + prepared_inputs.append( + self._prepare_input_arg(inp, device_mesh, input_layout, desired_layout) + ) + return tuple(prepared_inputs) + + def _prepare_input_kwarg_fn(self, inputs, kwarg_inputs, device_mesh): + prepared_arg_inputs = self._prepare_input_fn(inputs, device_mesh) + prepared_kwarg_inputs = {} + for kwarg_key in kwarg_inputs: + kwarg_val = kwarg_inputs[kwarg_key] + input_layout = self.input_kwarg_layouts.get(kwarg_key) + desired_input_layout = self.desired_input_kwarg_layouts.get(kwarg_key) + + prepared_kwarg_inputs[kwarg_key] = self._prepare_input_arg( + kwarg_val, device_mesh, input_layout, desired_input_layout + ) + + return (prepared_arg_inputs, prepared_kwarg_inputs) + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + if self.with_kwargs: + module.register_forward_pre_hook( + lambda _, inputs, kwargs: self._prepare_input_kwarg_fn( + inputs, kwargs, device_mesh + ), + with_kwargs=True, + ) # type: ignore[misc] + else: + module.register_forward_pre_hook( + lambda _, inputs: self._prepare_input_fn(inputs, device_mesh) + ) # type: ignore[misc, call-arg] + return module + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + tmpstr += f"input_layouts={self.input_layouts}, " + tmpstr += f"desired_input_layouts={self.desired_input_layouts}, " + tmpstr += f"input_kwarg_layouts={self.input_kwarg_layouts}, " + tmpstr += f"desired_input_kwarg_layouts={self.desired_input_kwarg_layouts}, " + tmpstr += f"use_local_output={self.use_local_output}" + tmpstr += ")" + return tmpstr + + +class PrepareModuleOutput(ParallelStyle): + """ + Configure the nn.Module's outputs to convert the output tensors of the nn.Module to DTensors at runtime according to + ``output_layouts``, and perform layout redistribution according to the ``desired_output_layouts``. + + Keyword Args: + output_layouts (Union[Placement, Tuple[Placement]]): + The DTensor layouts of output tensors for the nn.Module, this is used to convert the output tensors to + DTensors if they are :class:`torch.Tensor`. If some outputs are not torch.Tensor or no need to convert to DTensors, + ``None`` need to be specified as a placeholder. + desired_output_layouts (Union[Placement, Tuple[Placement]]): + The desired DTensor layouts of output tensors for the nn.Module, this is used to ensure the outputs of the nn.Module + have the desired DTensor layouts. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module outputs, default: True. + Returns: + A ParallelStyle object that prepares the sharding layouts of the nn.Module's outputs. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, PrepareModuleOutput + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> block = TransformerBlock(...) # block is a nn.Module that contains an "attn" Attention submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # According to the style specified below, the output of the TransformerBlock will be converted to Replicated DTensor + >>> # and then redistributed to Sharded DTensor. + >>> parallelize_module( + >>> block, # this can be a submodule or module + >>> tp_mesh, + >>> parallelize_plan = PrepareModuleOutput( + >>> output_layouts=Replicate(), + >>> desired_output_layouts=Shard(0) + >>> ) + >>> ) + """ + + def __init__( + self, + *, + output_layouts: Placement | tuple[Placement | None, ...], + desired_output_layouts: Placement | tuple[Placement, ...], + use_local_output: bool = True, + ): + self.output_layouts = ( + (output_layouts,) + if isinstance(output_layouts, Placement) + else output_layouts + ) + self.desired_output_layouts = ( + (desired_output_layouts,) + if isinstance(desired_output_layouts, Placement) + else desired_output_layouts + ) + self.use_local_output = use_local_output + if len(self.output_layouts) != len(self.desired_output_layouts): + raise AssertionError( + "output_layouts and desired_output_layouts should have same length!" + ) + + def _prepare_out_fn(self, outputs, device_mesh): + prepared_outputs = [] + if not isinstance(outputs, tuple): + outputs = (outputs,) + if len(outputs) != len(self.output_layouts): + raise ValueError( + "module outputs and output_layouts should have same length!" + ) + + for out, out_layout, desired_out_layout in zip( + outputs, self.output_layouts, self.desired_output_layouts + ): + if out_layout is not None: + if isinstance(out, DTensor): + # TODO: re-enable the check once we fix the compile path + # assert out.placements[0] == out_layout + dt_out = out + else: + dt_out = DTensor.from_local( + out, device_mesh, (out_layout,), run_check=False + ) + + if out_layout != desired_out_layout: + dt_out = dt_out.redistribute(placements=(desired_out_layout,)) + prepared_outputs.append( + dt_out.to_local() if self.use_local_output else dt_out + ) + else: + prepared_outputs.append(out) + if len(prepared_outputs) == 1: + return prepared_outputs[0] + else: + return tuple(prepared_outputs) + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + module.register_forward_hook( + lambda _, inputs, outputs: self._prepare_out_fn(outputs, device_mesh) + ) # type: ignore[misc, call-arg] + return module + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + tmpstr += f"output_layouts={self.output_layouts}, " + tmpstr += f"desired_output_layouts={self.desired_output_layouts}, " + tmpstr += f"use_local_output={self.use_local_output}" + tmpstr += ")" + return tmpstr + + +class PrepareModuleInputOutput(ParallelStyle): + """ + Configure the nn.Module's inputs (and outputs) to convert the input tensors (and output tensors, respectively) of the nn.Module + to DTensors at runtime according to ``input_layouts`` (and output_layouts, respectively), and perform layout redistribution + according to the ``desired_input_layouts`` (and ``desired_output_layouts``, respectively). This is a combination of + :class:`PrepareModuleInput` and :class:`PrepareModuleOutput`. + + Keyword Args: + input_layouts (Union[Placement, Tuple[Optional[Placement]]]): + The DTensor layouts of input tensors for the nn.Module, this is used to convert the input tensors to + DTensors. If some inputs are not torch.Tensor or no need to convert to DTensors, ``None`` need to be specified + as a placeholder. default: None. + desired_input_layouts (Union[Placement, Tuple[Optional[Placement]]]): + The desired DTensor layout of input tensors for the nn.Module, this is used to ensure the inputs of the nn.Module + have the desired DTensor layouts. This argument needs to have the same length with ``input_layouts``. default: None. + input_kwarg_layouts (Dict[str, Placement]): + The DTensor layouts of input kwargs for the nn.Module, this is used to convert the input kwarg tensors to DTensors. + default: None + desired_input_kwarg_layouts: (Dict[str, Placement]): + The desired DTensor layout of input kwargs for the nn.Module, this is used to ensure the inputs of the nn.Module + have the desired DTensor layouts. default: None. + use_local_input (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module inputs, default: False. + output_layouts (Union[Placement, Tuple[Placement]]): + The DTensor layouts of output tensors for the nn.Module, this is used to convert the output tensors to + DTensors if they are :class:`torch.Tensor`. If some outputs are not torch.Tensor or no need to convert to DTensors, + ``None`` need to be specified as a placeholder. + desired_output_layouts (Union[Placement, Tuple[Placement]]): + The desired DTensor layouts of output tensors for the nn.Module, this is used to ensure the outputs of the nn.Module + have the desired DTensor layouts. + use_local_output (bool, optional): + Whether to use local :class:`torch.Tensor` instead of :class:`DTensor` for the module outputs, default: True. + Returns: + A :class:`ParallelStyle` object that prepares the sharding layouts of the nn.Module's inputs and outputs. + + Example:: + >>> # xdoctest: +SKIP(failing) + >>> from torch.distributed.tensor.parallel import parallelize_module, PrepareModuleInputOutput + >>> from torch.distributed.device_mesh import init_device_mesh + >>> ... + >>> block = TransformerBlock(...) # block is a nn.Module that contains an "attn" Attention submodule + >>> tp_mesh = init_device_mesh("cuda", (8,)) + >>> + >>> # According to the style specified below, the first input of attn will be annotated as Sharded DTensor + >>> # and then redistributed to Replicated DTensor, and the output of the TransformerBlock will be annotated + >>> # as Replicated DTensor and then redistributed to Sharded DTensor. + >>> parallelize_module( + >>> block, # this can be a submodule or module + >>> tp_mesh, + >>> parallelize_plan={ + >>> "attn": PrepareModuleInputOutput( + >>> input_layouts=(Shard(0), None, None, ...), + >>> desired_input_layouts=(Replicate(), None, None, ...), + >>> output_layouts=Replicate(), + >>> desired_output_layouts=Shard(0), + >>> ), + >>> } + >>> ) + """ + + def __init__( + self, + *, + input_layouts: Placement | tuple[Placement | None, ...] | None = None, + desired_input_layouts: Placement | tuple[Placement | None, ...] | None = None, + input_kwarg_layouts: dict[str, Placement] | None = None, + desired_input_kwarg_layouts: dict[str, Placement] | None = None, + use_local_input: bool = False, + output_layouts: Placement | tuple[Placement | None, ...], + desired_output_layouts: Placement | tuple[Placement, ...], + use_local_output: bool = True, + ): + self.prepare_module_input = PrepareModuleInput( + input_layouts=input_layouts, + desired_input_layouts=desired_input_layouts, + input_kwarg_layouts=input_kwarg_layouts, + desired_input_kwarg_layouts=desired_input_kwarg_layouts, + use_local_output=use_local_input, + ) + self.prepare_module_output = PrepareModuleOutput( + output_layouts=output_layouts, + desired_output_layouts=desired_output_layouts, + use_local_output=use_local_output, + ) + + def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module: + self.prepare_module_input._apply(module, device_mesh) + self.prepare_module_output._apply(module, device_mesh) + + return module + + def __repr__(self) -> str: + tmpstr = self.__class__.__name__ + "(" + tmpstr += f"input_layouts={self.prepare_module_input.input_layouts}, " + tmpstr += ( + f"desired_input_layouts={self.prepare_module_input.desired_input_layouts}, " + ) + tmpstr += ( + f"input_kwarg_layouts={self.prepare_module_input.input_kwarg_layouts}, " + ) + tmpstr += f"desired_input_kwarg_layouts={self.prepare_module_input.desired_input_kwarg_layouts}, " + tmpstr += f"use_local_input={self.prepare_module_input.use_local_output}, " + tmpstr += f"output_layouts={self.prepare_module_output.output_layouts}, " + tmpstr += f"desired_output_layouts={self.prepare_module_output.desired_output_layouts}, " + tmpstr += f"use_local_output={self.prepare_module_output.use_local_output}" + tmpstr += ")" + return tmpstr diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/tensor/placement_types.py b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/placement_types.py new file mode 100644 index 0000000000000000000000000000000000000000..b7f0b7504bdd159e8e9222b35457544d9edfae73 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/tensor/placement_types.py @@ -0,0 +1,2084 @@ +# mypy: allow-untyped-defs +# Copyright (c) Meta Platforms, Inc. and affiliates + +import functools +from collections.abc import Sequence +from dataclasses import dataclass, field +from enum import IntEnum +from typing import Any, cast, TypeGuard, TypeVar + +import torch +import torch._C +import torch.distributed._functional_collectives as funcol +from torch._C._distributed import Placement +from torch.distributed import RankType +from torch.distributed._local_tensor import maybe_run_for_local_tensor +from torch.distributed.device_mesh import DeviceMesh +from torch.distributed.tensor._collective_utils import ( + fill_empty_tensor_to_shards, + mesh_broadcast, + mesh_scatter, + pad_tensor, + shard_dim_alltoall, + unpad_tensor, +) +from torch.distributed.tensor._ops._mask_buffer import MaskBuffer +from torch.types import IntLikeType + + +__all__ = ["Placement", "Shard", "Replicate", "Partial"] + +_RankTypeT = TypeVar("_RankTypeT", bound=RankType) +_STRIDED_SHARD_BASE_SPLIT_FACTOR: Any = torch._C._distributed.StridedShard.split_factor + + +# Appease TestPublicBindings.test_correct_module_names +Placement.__module__ = "torch.distributed.tensor.placement_types" + + +def _raise_error(method_name, _): + raise RuntimeError( + f"Placement method '{method_name}' should not be called as " + "it should be overridden by the subclass" + ) + + +Placement.__eq__ = functools.partial(_raise_error, method_name="__eq__") +Placement.__hash__ = functools.partial(_raise_error, method_name="__hash__") +Placement.__fx_repr__ = functools.partial(_raise_error, method_name="__fx_repr__") + + +class _StridedShardOffsetMode(IntEnum): + FIRST = 0 + ALL = 1 + NONE = 2 + + +def _explicit_or_backed_hint(value: IntLikeType) -> int | None: + from torch.fx.experimental.symbolic_shapes import ( + free_unbacked_symbols, + optimization_hint, + ) + + if isinstance(value, int): + return int(value) + if not isinstance(value, torch.SymInt): + try: + return int(value) + except (TypeError, ValueError): + return None + shape_env = getattr(value.node, "shape_env", None) + if shape_env is None: + return None + unbacked_symbols = free_unbacked_symbols(value.node.expr) + if unbacked_symbols and not unbacked_symbols.issubset( + shape_env.var_to_hint_override.keys() + ): + return None + return int(optimization_hint(value, fallback=None)) + + +def _guarded_hint_int(value: IntLikeType, *, reason: str) -> int: + if isinstance(value, int): + return int(value) + if not isinstance(value, torch.SymInt): + return int(value) + hint = _explicit_or_backed_hint(value) + if hint is None: + raise RuntimeError( + f"Cannot specialize symbolic {reason} without a hint: {value}" + ) + _defer_symbolic_equality_assert(value, hint, reason=reason) + return hint + + +def _defer_symbolic_equality_assert( + value: IntLikeType, hint: int, *, reason: str +) -> None: + if not isinstance(value, torch.SymInt): + return + shape_env = getattr(value.node, "shape_env", None) + if shape_env is None: + return + import sympy + + from torch.fx.experimental.symbolic_shapes import ( + free_unbacked_symbols, + RuntimeAssert, + ) + from torch.utils._traceback import CapturedTraceback + + expr = sympy.Eq(value.node.expr, hint) + cands = sorted( + free_unbacked_symbols(expr), + key=lambda symbol: int(symbol.name[1:]), + ) + key = cands[-1] if cands else None + runtime_asserts = shape_env.deferred_runtime_asserts.setdefault(key, []) + if any(runtime_assert.expr == expr for runtime_assert in runtime_asserts): + return + runtime_asserts.append( + RuntimeAssert( + expr, + f"Runtime assertion failed: symbolic {reason} {value.node.expr} == {hint}", + CapturedTraceback.extract(skip=1), + ) + ) + shape_env.num_deferred_runtime_asserts += 1 + shape_env._update_version_counter() + + +def _split_factor_key(value: IntLikeType) -> object: + if isinstance(value, torch.SymInt): + return value.node.expr + return int(value) + + +def _sym_mod(left: IntLikeType, right: IntLikeType) -> Any: + return cast(Any, left) % cast(Any, right) + + +def _hint_proves_even_shard(size: IntLikeType, num_chunks: IntLikeType) -> bool: + from torch.fx.experimental.symbolic_shapes import guard_or_false + + if guard_or_false(_sym_mod(size, num_chunks) == 0): + return True + size_hint = _explicit_or_backed_hint(size) + chunk_hint = _explicit_or_backed_hint(num_chunks) + if ( + size_hint is None + or chunk_hint is None + or chunk_hint == 0 + or size_hint % chunk_hint != 0 + ): + return False + + # Padding decisions require a Python branch. Only explicit hint overrides + # can justify skipping padding for otherwise data-dependent sharding. + torch._check(_sym_mod(size, num_chunks) == 0) + return True + + +class Shard(torch._C._distributed.Shard): + """ + The ``Shard(dim)`` placement describes the DTensor sharding on tensor dimension + ``dim`` over a corresponding ``DeviceMesh`` dimension, where each rank on the + DeviceMesh dimension only holds a shard/piece of the global Tensor. The + ``Shard(dim)`` placement follows the ``torch.chunk(dim)`` semantic, where the + last few shards on the DeviceMesh dimension might be empty when the tensor dimension + is not evenly divisible on the DeviceMesh dimension. The ``Shard`` placement can be + used by all DTensor APIs (i.e. distribute_tensor, from_local, etc.) + + Args: + dim (int): The tensor dimension that describes the DTensor is sharded over its + corresponding DeviceMesh dimension. + + .. warning:: sharding on a tensor dimension where the tensor dimension size is not + evenly divisible on a DeviceMesh dimension is currently experimental and subject to change. + + .. note:: When checking whether a placement is shard-like, use + :func:`_is_shard_like` instead of ``isinstance(p, Shard)`` to also + match :class:`_StridedShard`. + """ + + def _split_tensor( + self, + tensor: torch.Tensor, + num_chunks: int, + *, + with_padding: bool = True, + contiguous: bool = True, + ) -> tuple[list[torch.Tensor], list[int]]: + """ + This function uses torch.chunk to split a tensor into num_chunks shards along + the Shard placement dimension, and return a list of shards with their pad sizes. + + Keyword args: + with_padding (bool, optional): when True, we pad the tensor on the last + few ranks before calling the collectives (i.e. scatter/all_gather, etc.). + This is because collectives usually require equal size tensor inputs + """ + return self._split_tensor_helper( + tensor, num_chunks, with_padding, contiguous, self.dim + ) + + @staticmethod + def _split_tensor_helper( + tensor: torch.Tensor, + num_chunks: int, + with_padding: bool, + contiguous: bool, + dim: int, + ) -> tuple[list[torch.Tensor], list[int]]: + if dim > tensor.ndim: + raise AssertionError( + f"Sharding dim {dim} greater than tensor ndim {tensor.ndim}" + ) + + # chunk tensor over dimension `dim` into n slices + tensor_list = Shard._custom_chunk(tensor, num_chunks, dim=dim) + + # compute the chunk size inline with ``torch.chunk`` to calculate padding + full_chunk_size = (tensor.size(dim) + num_chunks - 1) // num_chunks + + shard_list: list[torch.Tensor] = [] + pad_sizes: list[int] = [] + for shard in tensor_list: + if with_padding: + pad_size = Shard._get_shard_pad_size(full_chunk_size, shard, dim) + shard = pad_tensor(shard, dim, pad_size) + pad_sizes.append(pad_size) + if contiguous: + shard = shard.contiguous() + shard_list.append(shard) + return shard_list, pad_sizes + + @maybe_run_for_local_tensor + def _select_split_tensor( + self, + tensor: torch.Tensor, + num_chunks: int, + index: RankType, + *, + with_padding: bool = True, + contiguous: bool = True, + clone: bool = True, + ) -> torch.Tensor: + """ + Like _split_tensor() but only returns a single shard at the given index. + + This function splits a tensor into num_chunks shards along the Shard placement + dimension and returns only the shard at the specified index. + + Keyword args: + with_padding (bool, optional): when True, we pad the tensor on the last + few ranks before calling the collectives (i.e. scatter/all_gather, etc.). + This is because collectives usually require equal size tensor inputs. + contiguous (bool, optional): when True, the returned shard is made contiguous. + clone (bool, optional): when True, the returned shard is cloned. + """ + # We don't handle SymInt with_padding yet (because that requires extra + # work based on the shard) + if isinstance(index, int) or with_padding: + shards, _ = self._split_tensor( + tensor, num_chunks, with_padding=with_padding, contiguous=False + ) + result = shards[index] + if clone: + result = result.clone() + elif contiguous: + result = result.contiguous() + return result + + # For the SymInt implementation just compute the value for the tensor we + # want rather than computing all of them. + + if self.dim > tensor.ndim: + raise AssertionError( + f"Sharding dim {self.dim} greater than tensor ndim {tensor.ndim}" + ) + + # chunk tensor over dimension `dim` into n slices + dim_size = tensor.size(self.dim) + split_size = (dim_size + num_chunks - 1) // num_chunks + # each split is split_size except (maybe) the last one... + last_split = dim_size - split_size * (num_chunks - 1) + + start = split_size * index + length = torch.sym_ite(index == num_chunks - 1, last_split, split_size) + result = torch.narrow(tensor, self.dim, start, length) + if clone: + result = result.clone() + elif contiguous: + result = result.contiguous() + return result + + @staticmethod + def _custom_chunk( + tensor: torch.Tensor, num_chunks: IntLikeType, dim: int + ) -> list[torch.Tensor]: + """ + Returns list of tensor chunks along dim. + Uses torch.chunk in eager mode, but torch.narrow under tracing to be unbacked-symint safe. + Also handles uneven/zero-sharding cases. + """ + from torch.distributed._functional_collectives import _are_we_tracing + from torch.fx.experimental.symbolic_shapes import has_free_unbacked_symbols + + if tensor.dim() <= 0: + raise AssertionError(f"Expected tensor.dim() > 0, got {tensor.dim()}") + num_chunks_int = _guarded_hint_int(num_chunks, reason="chunk count") + if num_chunks_int <= 0: + raise AssertionError(f"Expected num_chunks > 0, got {num_chunks}") + + # TODO(pianpwk): remove the unbacked symbols check and fix AsyncTP pattern matching + # for test_micro_pipeline_tp.py. + if not _are_we_tracing() or not has_free_unbacked_symbols(tensor): + tensor_list = list(torch.chunk(tensor, num_chunks_int, dim=dim)) + return fill_empty_tensor_to_shards( + tensor_list, dim, num_chunks_int - len(tensor_list) + ) + else: + dim_size = tensor.size(dim) + if _hint_proves_even_shard(dim_size, num_chunks): + chunk_size = dim_size // num_chunks + return [ + tensor.narrow(dim, chunk_size * i, chunk_size) + for i in range(num_chunks_int) + ] + + split_size = (dim_size + num_chunks - 1) // num_chunks + chunks = [] + for i in range(num_chunks_int): + start = torch.sym_min(split_size * i, dim_size) + end = torch.sym_min(split_size * (i + 1), dim_size) + chunks.append(tensor.narrow(dim, start, end - start)) + return chunks + + @staticmethod + @maybe_run_for_local_tensor + def local_shard_size_and_offset( + curr_local_size: IntLikeType, + num_chunks: int, + rank: _RankTypeT, + ) -> tuple[_RankTypeT, _RankTypeT]: + """ + Given the size of the current local tensor (which may already be sharded on some dimensions), + computes the new local shard size and offset given the desired number of chunks + (num_chunks is generally equal to the size of the current sharding dim). + + Note: new local shard offset is relative to the current sharded tensor, not the global tensor. + See `_utils.compute_local_shape_and_global_offset` for computing global offset. + + Returns (new local shard size, offset) + + """ + from torch.fx.experimental.symbolic_shapes import guard_or_false + + # Compute the chunk size inline with ``torch.chunk`` + # Assume uneven sharding as general case for unbacked sizes. + if guard_or_false(curr_local_size % num_chunks == 0): + full_chunk_size = curr_local_size // num_chunks + # pyrefly: ignore[bad-assignment] # pyrefly bug? + shard_starting_idx: _RankTypeT = full_chunk_size * rank + return full_chunk_size, shard_starting_idx # pyrefly: ignore[bad-return] + + # uneven sharding case + full_chunk_size = (curr_local_size + num_chunks - 1) // num_chunks + # pyrefly: ignore[bad-assignment] # pyrefly bug? + shard_starting_idx: _RankTypeT = full_chunk_size * rank + + shard_end_idx = torch.sym_min( + curr_local_size, shard_starting_idx + full_chunk_size + ) + local_shard_size = torch.sym_max(0, shard_end_idx - shard_starting_idx) + return local_shard_size, torch.sym_min(curr_local_size, shard_starting_idx) # type: ignore[return-value] + + def _local_shard_size_and_offset( + self, + curr_local_size: int, + num_chunks: int, + rank: RankType, + ) -> tuple[int, RankType]: + # pyrefly: ignore [bad-argument-type, bad-return] + return Shard.local_shard_size_and_offset(curr_local_size, num_chunks, rank) + + @staticmethod + @maybe_run_for_local_tensor + def _maybe_unpad_tensor_with_sizes( + dim, local_tensor, pad_sizes, mesh_dim_local_rank, make_contiguous + ) -> torch.Tensor: + from torch.fx.experimental.symbolic_shapes import guard_or_true + + # Only unpad if the local_tensor was padded on the dimension. + # Assume padding (uneven sharding) as general case for unbacked sizes. + if guard_or_true(pad_sizes[mesh_dim_local_rank] > 0): + local_tensor = unpad_tensor( + local_tensor, dim, pad_sizes[mesh_dim_local_rank] + ) + if make_contiguous: + local_tensor = local_tensor.contiguous() + return local_tensor + + def _shard_tensor( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + ) -> torch.Tensor: + """ + Shard and scatter a tensor on a mesh dimension (use coordinate 0 on the + mesh dimension as source of truth). + + Create the local tensor for this rank following the given Shard + placement. If src_data_rank is None, perform only local splitting. + Otherwise, additionally scatter data from src_data_rank. Unlike + ``_split_tensor``, which supports uneven sharding via padding, this + method requires the tensor dimension to be evenly divisible by the + number of chunks (mesh dimension size). + """ + my_coordinate = mesh.get_coordinate() + num_chunks = mesh.size(mesh_dim=mesh_dim) + + if my_coordinate is None: + # if rank is not part of mesh, we simply return an empty tensor + return tensor.new_empty(0, requires_grad=tensor.requires_grad) + + mesh_dim_local_rank = my_coordinate[mesh_dim] + + if src_data_rank is None: + # src_data_rank specified as None explicitly means to skip the + # communications, simply split + return self._select_split_tensor( + tensor, + num_chunks, + mesh_dim_local_rank, + with_padding=False, + contiguous=True, + ) + + scatter_list, pad_sizes = self._split_tensor( + tensor, num_chunks, with_padding=True, contiguous=True + ) + + it = iter(scatter_list) + first = next(it) + # Tensors in the scatter list are expected to have the same shape because + # split is requested with padding. + if not all(first.shape == v.shape for v in it): + raise AssertionError + + output = torch.empty_like(first) + + # perform scatter from the src_data_rank as data source when it is not None + mesh_scatter( + output, scatter_list, mesh, mesh_dim=mesh_dim, group_src=src_data_rank + ) + + return Shard._maybe_unpad_tensor_with_sizes( + self.dim, output, pad_sizes, mesh_dim_local_rank, True + ) + + @classmethod + def _make_shard_tensor( + cls, + dim: int, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + ) -> torch.Tensor: + shard_placement = cls(dim) + return shard_placement._shard_tensor(tensor, mesh, mesh_dim, src_data_rank) + + def _reduce_shard_tensor( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + reduce_op: str, + mesh_dim: int, + ) -> torch.Tensor: + """ + reduce and scatter a tensor on a mesh dimension + """ + from torch.fx.experimental.symbolic_shapes import guard_or_true + + my_coordinate = mesh.get_coordinate() + num_chunks = mesh.size(mesh_dim=mesh_dim) + + if my_coordinate is None: + # if rank is not part of mesh, we simply return local_tensor, + # which should be an empty tensor + return tensor + + # Assume padding (uneven sharding) as general case for unbacked sizes, + # unless an optimization hint can prove the even-shard branch. + is_padded = not _hint_proves_even_shard( + tensor.size(self.dim), num_chunks + ) and guard_or_true(tensor.size(self.dim) % num_chunks != 0) + pad_sizes = None + if is_padded: + scattered_list, pad_sizes = self._split_tensor( + tensor, num_chunks, with_padding=True, contiguous=True + ) + tensor = torch.cat(scattered_list, dim=self.dim) + elif not tensor.is_contiguous(): + tensor = tensor.contiguous() + + output = funcol.reduce_scatter_single( + tensor, reduce_op, scatter_dim=self.dim, group=(mesh, mesh_dim) + ) + + if is_padded: + if pad_sizes is None: + raise AssertionError + output = Shard._maybe_unpad_tensor_with_sizes( + self.dim, output, pad_sizes, mesh._sym_get_coordinate(mesh_dim), False + ) + return output + + @maybe_run_for_local_tensor + def _maybe_pad_tensor( + self, + local_tensor: torch.Tensor, + logical_dim_size: IntLikeType, + num_chunks: int, + ) -> torch.Tensor: + from torch.fx.experimental.symbolic_shapes import guard_or_true + + # Assume padding (uneven sharding) as general case for unbacked sizes, + # unless an optimization hint can prove the even-shard branch. + is_padded = not _hint_proves_even_shard( + logical_dim_size, num_chunks + ) and guard_or_true(logical_dim_size % num_chunks != 0) + + if is_padded: + full_chunk_size = (logical_dim_size + num_chunks - 1) // num_chunks + pad_size = full_chunk_size - local_tensor.size(self.dim) + local_tensor = pad_tensor(local_tensor, self.dim, pad_size) # type: ignore[bad-argument-type] + + if not local_tensor.is_contiguous(): + local_tensor = local_tensor.contiguous() + + return local_tensor + + @maybe_run_for_local_tensor + def _maybe_unpad_tensor( + self, + local_tensor: torch.Tensor, + logical_dim_size: IntLikeType, + num_chunks: int, + ) -> torch.Tensor: + from torch.fx.experimental.symbolic_shapes import guard_or_false, guard_or_true + + # Assume padding (uneven sharding) as general case for unbacked sizes, + # unless an optimization hint can prove the even-shard branch. + is_padded = not _hint_proves_even_shard( + logical_dim_size, num_chunks + ) and guard_or_true(logical_dim_size % num_chunks != 0) + + if is_padded: + full_chunk_size = (logical_dim_size + num_chunks - 1) // num_chunks + unpad_size = full_chunk_size * num_chunks - logical_dim_size # type: ignore[possibly-undefined] + local_tensor = unpad_tensor(local_tensor, self.dim, unpad_size) # type: ignore[bad-argument-type] + + # Bind derived symbolic sizes (e.g. 2*(s//2)) back to the original + # symbol - needed for correct shape propagation and dynamo generation + if local_tensor.size(self.dim) is not logical_dim_size: + orig_size = local_tensor.size(self.dim) + torch._check(orig_size >= logical_dim_size) + local_tensor = local_tensor.narrow(self.dim, 0, logical_dim_size) + + # Safety check: the narrow should never change the concrete size. + # Use guard_or_false so we don't trigger data-dependent guards + # on unbacked symints. + if guard_or_false(local_tensor.size(self.dim) != orig_size): + raise RuntimeError( + f"narrow unexpectedly changed concrete size on dim {self.dim}: " + f"{orig_size} -> {local_tensor.size(self.dim)}" + ) + + # narrow/unpad on a non-leading dim leaves the tensor non-contiguous, + # which leaks out to callers that expect a Replicate() result to have + # a contiguous local tensor (e.g. downstream view ops in TP linear). + # Use unconditional .contiguous() to avoid guarding on symbolic strides + # under torch.compile with unbacked symints. + local_tensor = local_tensor.contiguous() + + return local_tensor + + def _to_replicate_tensor( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + current_logical_shape: Sequence[IntLikeType], + ) -> torch.Tensor: + """ + This function all_gather all shards and return a tensor that + is replicated on the previously sharded mesh dimension + """ + num_chunks = mesh.size(mesh_dim=mesh_dim) + logical_dim_size = current_logical_shape[self.dim] + + local_tensor = self._maybe_pad_tensor( + local_tensor, logical_dim_size, num_chunks + ) + + result = funcol.all_gather_single( + local_tensor, + gather_dim=self.dim, + group=(mesh, mesh_dim), + ) + + result = self._maybe_unpad_tensor(result, logical_dim_size, num_chunks) + + return result + + def _replicate_to_shard( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + shard_index: IntLikeType, + ) -> torch.Tensor: + """ + transform from replicated tensor to a sharded tensor on + the current rank, which would perform a local chunk + """ + num_chunks = mesh.size(mesh_dim=mesh_dim) + return self._select_split_tensor( + local_tensor, + num_chunks, + shard_index, + with_padding=False, + clone=True, + ) + + def _to_partial_tensor( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + current_logical_shape: Sequence[IntLikeType], + ) -> torch.Tensor: + """Embed a local shard in a zero-filled tensor pending sum reduction.""" + num_chunks = mesh.size(mesh_dim=mesh_dim) + logical_dim_size = current_logical_shape[self.dim] + local_shard_size, local_offset = self.local_shard_size_and_offset( + logical_dim_size, + num_chunks, + mesh._sym_get_coordinate(mesh_dim), + ) + torch._check(local_tensor.size(self.dim) == local_shard_size) + output_shape: list[IntLikeType] = list(local_tensor.shape) + output_shape[self.dim] = logical_dim_size + output = local_tensor.new_zeros(output_shape) + offset = torch.scalar_tensor( + local_offset, dtype=torch.int64, device=local_tensor.device + ) + indices = torch.arange( + local_shard_size, dtype=torch.int64, device=local_tensor.device + ) + return output.index_add(self.dim, indices + offset, local_tensor) + + @staticmethod + @maybe_run_for_local_tensor + def _get_shard_pad_size( + full_size: int, local_tensor: torch.Tensor, dim: int + ) -> int: + """ + Get the padding size of the local tensor on the shard dimension. + """ + return full_size - local_tensor.size(dim) + + @staticmethod + def _compute_padding_info( + current_logical_shape: Sequence[IntLikeType], + num_chunks: int, + old_shard_dim: int, + new_shard_dim: int, + ) -> tuple[bool, IntLikeType, int, bool, IntLikeType, int]: + from torch.fx.experimental.symbolic_shapes import guard_or_true + + results = [] + for shard_dim in [old_shard_dim, new_shard_dim]: + dim_logical_size = current_logical_shape[shard_dim] + dim_padding = guard_or_true(dim_logical_size % num_chunks != 0) + dim_full_chunk_size = (dim_logical_size + num_chunks - 1) // num_chunks + results.append((dim_padding, dim_logical_size, dim_full_chunk_size)) + + return results[0] + results[1] # type: ignore[bad-return] + + @staticmethod + @maybe_run_for_local_tensor + def _pad_for_new_shard_dim( + current_logical_shape: Sequence[IntLikeType], + local_tensor: torch.Tensor, + num_chunks: int, + old_shard_dim: int, + new_shard_dim: int, + ) -> torch.Tensor: + ( + old_dim_padding, + _, + old_dim_full_chunk_size, + new_dim_padding, + _, + new_dim_full_chunk_size, + ) = Shard._compute_padding_info( + current_logical_shape, num_chunks, old_shard_dim, new_shard_dim + ) + + if old_dim_padding: + old_dim_pad_size = Shard._get_shard_pad_size( + old_dim_full_chunk_size, local_tensor, old_shard_dim + ) + local_tensor = pad_tensor(local_tensor, old_shard_dim, old_dim_pad_size) + if new_dim_padding: + new_dim_pad_size = Shard._get_shard_pad_size( + new_dim_full_chunk_size * num_chunks, local_tensor, new_shard_dim + ) + local_tensor = pad_tensor(local_tensor, new_shard_dim, new_dim_pad_size) + + if not local_tensor.is_contiguous(): + local_tensor = local_tensor.contiguous() + return local_tensor + + @staticmethod + @maybe_run_for_local_tensor + def _unpad_for_new_shard_dim( + current_logical_shape: Sequence[IntLikeType], + local_tensor: torch.Tensor, + num_chunks: int, + old_shard_dim: int, + new_shard_dim: int, + local_rank: int, + ) -> torch.Tensor: + ( + old_dim_padding, + _, + old_dim_full_chunk_size, + new_dim_padding, + new_dim_logical_size, + new_dim_full_chunk_size, + ) = Shard._compute_padding_info( + current_logical_shape, num_chunks, old_shard_dim, new_shard_dim + ) + + if old_dim_padding: + old_dim_unpad_size = ( + old_dim_full_chunk_size * num_chunks + - current_logical_shape[old_shard_dim] # type: ignore[possibly-undefined] + ) + local_tensor = unpad_tensor(local_tensor, old_shard_dim, old_dim_unpad_size) # type: ignore[possibly-undefined] + + if new_dim_padding: + local_shard_size_on_new_dim = Shard.local_shard_size_and_offset( + new_dim_logical_size, num_chunks, local_rank + )[0] + new_dim_unpad_size = new_dim_full_chunk_size - local_shard_size_on_new_dim # type: ignore[possibly-undefined] + local_tensor = unpad_tensor(local_tensor, new_shard_dim, new_dim_unpad_size) # type: ignore[possibly-undefined] + + return local_tensor + + def _to_new_shard_dim( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + current_logical_shape: Sequence[IntLikeType], + new_shard_dim: int, + ) -> torch.Tensor: + """ + transform from existing sharded tensor to a new sharded tensor on + that shard on a new dimension, which performs an alltoall + """ + my_coordinate = mesh.get_coordinate() + if my_coordinate is None: + # if rank is not part of mesh, we simply return local_tensor, + # which should be an empty tensor + return local_tensor + + num_chunks = mesh.size(mesh_dim=mesh_dim) + + local_tensor = Shard._pad_for_new_shard_dim( + current_logical_shape, local_tensor, num_chunks, self.dim, new_shard_dim + ) + + new_tensor = shard_dim_alltoall( + local_tensor, self.dim, new_shard_dim, mesh, mesh_dim + ) + + new_tensor = Shard._unpad_for_new_shard_dim( + current_logical_shape, + new_tensor, + num_chunks, + self.dim, + new_shard_dim, + my_coordinate[mesh_dim], + ) + + return new_tensor + + def __hash__(self) -> int: + return hash(self.dim) + + def __repr__(self) -> str: + """ + machine readable representation of the Shard placement + """ + return f"Shard(dim={self.dim})" + + def __fx_repr__(self): + """ + Returns FX-evaluable repr and required globals for Shard placement. + Needed for passing this type as an opaque object input to a custom op. + """ + return f"torch.distributed.tensor.placement_types.Shard(dim={self.dim})", {} + + def __str__(self) -> str: + """human readable representation of the Shard placement""" + return f"S({self.dim})" + + +class _StridedShard(torch._C._distributed.StridedShard): + """ + _StridedShard was originally introduced to support 2D FSDP2 + TP sharding where the tensor + is sharded on the TP mesh dimension first, then sharded on the FSDP mesh dimension. + We call this right-to-left sharding which is the opposite of the default + left-to-right sharding. See the example below:: + + tensor shape: [8, 8] + mesh: [[0, 1], [2, 3]], names=("dp", "tp") + placements: [Shard(0), Shard(0)] + + The default sharding behavior shards the tensor on "dp" mesh dimension first then + "tp" dimension. The sharding result will be:: + + Rank | Mesh Coordinate | Shard Index + ------------------------------------------------ + 0 | (0, 0) | 0 (row 0-1) + 1 | (0, 1) | 1 (row 2-3) + 2 | (1, 0) | 2 (row 4-5) + 3 | (1, 1) | 3 (row 6-7) + + While the FSDP2 + TP sharding behavior does the opposite: it shards the tensor on + "tp" mesh dim first then "dp" dim. This right-to-left sharding will produce the + result:: + + Rank | Mesh Coordinate | Shard Index + ------------------------------------------------ + 0 | (0, 0) | 0 (row 0-1) + 1 | (0, 1) | 2 (row 4-5) + 2 | (1, 0) | 1 (row 2-3) + 3 | (1, 1) | 3 (row 6-7) + + The consequence is, any attempt to redistribute this DTensor to a full replica will + produce a wrong result because the shard-to-replicate redistribution always happens + right-to-left, regardless it's left-to-right sharding or right-to-left. To address + this, we use _StridedShard placement to make this right-to-left sharding compatible + with our left-to-right convention on both tensor distribution and redistribution. + + Now with _StridedShard, the right-to-left sharding above can be represented as:: + + tensor shape: [8, 8] + mesh: [[0, 1], [2, 3]], names=("dp", "tp") + placements: [_StridedShard(0, split_factor=2), Shard(0)] + + And a left-to-right processing of `placements` will produce the same result, which is + different from using the `Shard` placement:: + + Rank | Mesh Coordinate | Shard Index + ------------------------------------------------ + 0 | (0, 0) | 0 (row 0-1) + 1 | (0, 1) | 2 (row 4-5) + 2 | (1, 0) | 1 (row 2-3) + 3 | (1, 1) | 3 (row 6-7) + + The argument `split_factor` is the number of existing shards over the tensor sharding + dimension before processing the _StridedShard placement, as if the sharding happened + right-to-left. In the example above, the tensor should first be sharded on the "tp" + dimension into 2 shards before being sharded on the "dp" dimension. Therefore, the + `split_factor` of the _StridedShard placement on "dp" dim is 2. + """ + + def __init__( + self, + dim: int, + *, + split_factor: IntLikeType = 1, + sf: IntLikeType | None = None, + ) -> None: + # The C++ placement stores the unroll count as int64_t, but view + # propagation may derive the split factor from symbolic shape metadata. + # Preserve that symbolic value on the Python wrapper and guard the + # concrete unroll count used by list-based lowering. + if sf is not None: + if split_factor != 1: + raise TypeError("_StridedShard received both split_factor and sf") + split_factor = sf + split_factor_int = _guarded_hint_int( + split_factor, reason="_StridedShard split_factor" + ) + super().__init__(dim, split_factor=split_factor_int) + if isinstance(split_factor, torch.SymInt): + self._symbolic_split_factor = split_factor + + @property + def split_factor(self) -> IntLikeType: # pyrefly: ignore [bad-override] + symbolic_split_factor = getattr(self, "_symbolic_split_factor", None) + if symbolic_split_factor is not None: + return symbolic_split_factor + return int(_STRIDED_SHARD_BASE_SPLIT_FACTOR.__get__(self, type(self))) + + def _split_factor_int(self) -> int: + return int(_STRIDED_SHARD_BASE_SPLIT_FACTOR.__get__(self, type(self))) + + def __eq__(self, other: object) -> bool: + return ( + isinstance(other, _StridedShard) + and self.dim == other.dim + and _split_factor_key(self.split_factor) + == _split_factor_key(other.split_factor) + ) + + def __hash__(self) -> int: + return hash((self.dim, _split_factor_key(self.split_factor))) + + def __repr__(self) -> str: + """ + machine readable representation of the _StridedShard placement + """ + return f"_StridedShard(dim={self.dim}, sf={self.split_factor})" + + def __str__(self) -> str: + """human readable representation of the _StridedShard placement""" + return f"_S({self.dim}, {self.split_factor})" + + @staticmethod + @maybe_run_for_local_tensor + def _select_shard(shards: list[torch.Tensor], shard_index) -> torch.Tensor: + return shards[shard_index].clone() + + def __fx_repr__(self): + """ + Returns FX-evaluable repr and required globals for Shard placement. + Needed for passing this type as an opaque object input to a custom op. + """ + # FX repr cannot refer to this placement's private SymInt object. The + # constructor emitted a guard tying the symbolic split factor to this + # concrete unroll count before the placement can reach codegen. + return ( + f"torch.distributed.tensor.placement_types._StridedShard(dim={self.dim}, sf={self._split_factor_int()})", + {}, + ) + + @classmethod + def _make_shard_tensor( + cls, + dim: int, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + split_factor: IntLikeType = 1, + ) -> torch.Tensor: + strided_shard_placement = cls(dim=dim, split_factor=split_factor) + return strided_shard_placement._shard_tensor( + tensor, mesh, mesh_dim, src_data_rank + ) + + def _shard_tensor( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + ) -> torch.Tensor: + """ + Shard and scatter a tensor on a mesh dimension (use coordinate 0 on the + mesh dimension as source of truth). + + Create the local tensor for this rank following the given StridedShard + placement. If src_data_rank is None, perform only local splitting. + Otherwise, additionally scatter data from src_data_rank. Unlike + ``_split_tensor``, which supports uneven sharding via padding, this + method requires the tensor dimension to be evenly divisible by the + number of chunks (mesh dimension size). + """ + my_coordinate = mesh.get_coordinate() + num_chunks = mesh.size(mesh_dim=mesh_dim) + + if my_coordinate is None: + # if rank is not part of mesh, we simply return an empty tensor + return tensor.new_empty(0, requires_grad=tensor.requires_grad) + + mesh_dim_local_rank = my_coordinate[mesh_dim] + + if src_data_rank is None: + # src_data_rank specified as None explicitly means to skip the + # communications, simply split + scatter_list, _ = self._split_tensor( + tensor, num_chunks, with_padding=False, contiguous=True + ) + + return self._select_shard(scatter_list, mesh_dim_local_rank) + + scatter_list, pad_sizes = self._split_tensor( + tensor, num_chunks, with_padding=True, contiguous=True + ) + + it = iter(scatter_list) + first = next(it) + # Tensors in the scatter list are expected to have the same shape because + # split is requested with padding. + if not all(first.shape == v.shape for v in it): + raise AssertionError + + output = torch.empty_like(first) + + # perform scatter from the src_data_rank as data source when it is not None + mesh_scatter( + output, scatter_list, mesh, mesh_dim=mesh_dim, group_src=src_data_rank + ) + + return Shard._maybe_unpad_tensor_with_sizes( + self.dim, output, pad_sizes, mesh_dim_local_rank, True + ) + + def _split_tensor( + self, + tensor: torch.Tensor, + num_chunks: int, + *, + with_padding: bool = True, + contiguous: bool = True, + ) -> tuple[list[torch.Tensor], list[int]]: + if self.dim > tensor.ndim: + raise AssertionError( + f"Sharding dim {self.dim} greater than tensor ndim {tensor.ndim}" + ) + + # Essentially _StridedShard express the right-to-left sharding in the + # reversed order. Here we perform first_split as the virtual "right" sharding, + # and then second_split as the virtual "left" sharding, and finally assemble + # results in the transposed left-first order. + + # First split: chunk into split_factor pieces + split_factor = self.split_factor + split_factor_int = self._split_factor_int() + first_split = Shard._custom_chunk(tensor, split_factor, dim=self.dim) + + # Second split: chunk each piece into num_chunks pieces + second_split = [] + for s in first_split: + second_split.append(Shard._custom_chunk(s, num_chunks, dim=self.dim)) + + shard_list: list[torch.Tensor] = [] + for i in range(num_chunks): + shard = torch.cat( + [second_split[j][i] for j in range(split_factor_int)], + dim=self.dim, + ) + if contiguous: + shard = shard.contiguous() + shard_list.append(shard) + + pad_sizes: list[int] = [] + if with_padding: + # The amount of padding is determined by the local chunk with the + # largest size. Avoid this Python max when padding is disabled so + # tracing paths can keep unbacked sizes symbolic. + max_chunk_size = max([shard.size(self.dim) for shard in shard_list]) + pad_sizes = [max_chunk_size - shard.size(self.dim) for shard in shard_list] + + return shard_list, pad_sizes + + @maybe_run_for_local_tensor + def _select_split_tensor( + self, + tensor: torch.Tensor, + num_chunks: int, + index: IntLikeType, + *, + with_padding: bool = True, + contiguous: bool = True, + clone: bool = True, + ) -> torch.Tensor: + """ + Like _split_tensor() but only returns a single shard at the given index. + + This function splits a tensor into num_chunks shards along the _StridedShard + placement dimension and returns only the shard at the specified index. + + Keyword args: + with_padding (bool, optional): when True, we pad the tensor on the last + few ranks before calling the collectives (i.e. scatter/all_gather, etc.). + This is because collectives usually require equal size tensor inputs. + contiguous (bool, optional): when True, the returned shard is made contiguous. + clone (bool, optional): when True, the returned shard is cloned. + """ + shards, _ = self._split_tensor( + tensor, num_chunks, with_padding=with_padding, contiguous=False + ) + result = shards[index] + if clone: + result = result.clone() + elif contiguous: + result = result.contiguous() + return result + + def _to_replicate_tensor( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + current_logical_shape: Sequence[IntLikeType], + ) -> torch.Tensor: + """ + Replay the replicate-to-shard process to understand how to stitch shards back. + + This method performs all_gather to collect all shards and then reconstructs + the original replicated tensor by handling padding, unpadding, and reordering. + + Example: + Consider a 1D input tensor [0, 1, 2, 3, 4, 5, 6, 7, 8] with 9 elements. + Using _StridedShard(dim=0, split_factor=2) and num_chunks=4: + + Preparation (via _split_tensor, before _to_replicate_tensor is called): + _split_tensor produces 4 shards with strided indices: + - First split (split_factor=2): [0,1,2,3,4] and [5,6,7,8] + - Second split (num_chunks=4) on each piece: + [0,1,2,3,4] (5 elements) -> [[0,1], [2,3], [4], []] + [5,6,7,8] (4 elements) -> [[5], [6], [7], [8]] + - Transpose and concatenate for each chunk: + Chunk 0: [0,1] + [5] = [0,1,5] (indices) + Chunk 1: [2,3] + [6] = [2,3,6] (indices) + Chunk 2: [4] + [7] = [4,7] (indices) + Chunk 3: [] + [8] = [8] (indices) + + So we get shards with values: + Rank 0: [0, 1, 5] (size=3) + Rank 1: [2, 3, 6] (size=3) + Rank 2: [4, 7] (size=2) + Rank 3: [8] (size=1) + + These shards are the `local_tensor` input to _to_replicate_tensor + on each rank. Each rank only has its own shard when this function + is called. + + Step 1: Pad all shards to max_chunk_size=3: + Rank 0: [0, 1, 5] (no padding needed) + Rank 1: [2, 3, 6] (no padding needed) + Rank 2: [4, 7, P] (padded with 1 element) + Rank 3: [8, P, P] (padded with 2 elements) + + Step 2: all_gather produces concatenated padded tensor: + [0, 1, 5, | 2, 3, 6, | 4, 7, P, | 8, P, P] + chunk 0 chunk 1 chunk 2 chunk 3 + (pos 0-2) (pos 3-5) (pos 6-8) (pos 9-11) + + Step 3: Compute select_indices to extract valid elements and reorder: + sharded_indices = [[0,1,5], [2,3,6], [4,7], [8]] + padded_positions: + chunk 0: base=0 -> [0, 1, 2] (positions of [0,1,5] in gathered) + chunk 1: base=3 -> [3, 4, 5] (positions of [2,3,6] in gathered) + chunk 2: base=6 -> [6, 7] (positions of [4,7] in gathered) + chunk 3: base=9 -> [9] (position of [8] in gathered) + + permutation = cat(sharded_indices) = [0, 1, 5, 2, 3, 6, 4, 7, 8] + select_positions = cat(padded_positions) = [0, 1, 2, 3, 4, 5, 6, 7, 9] + + inv_permutation = argsort(permutation) + permutation[0]=0 -> inv_permutation[0]=0 + permutation[1]=1 -> inv_permutation[1]=1 + permutation[2]=5 -> inv_permutation[5]=2 + permutation[3]=2 -> inv_permutation[2]=3 + permutation[4]=3 -> inv_permutation[3]=4 + permutation[5]=6 -> inv_permutation[6]=5 + permutation[6]=4 -> inv_permutation[4]=6 + permutation[7]=7 -> inv_permutation[7]=7 + permutation[8]=8 -> inv_permutation[8]=8 + => inv_permutation = [0, 1, 3, 4, 6, 2, 5, 7, 8] + + select_indices = select_positions[inv_permutation] + = [0, 1, 2, 3, 4, 5, 6, 7, 9][inv_permutation] + For original position 0: select_indices[0] = select_positions[0] = 0 + For original position 1: select_indices[1] = select_positions[1] = 1 + For original position 2: select_indices[2] = select_positions[3] = 3 + For original position 3: select_indices[3] = select_positions[4] = 4 + For original position 4: select_indices[4] = select_positions[6] = 6 + For original position 5: select_indices[5] = select_positions[2] = 2 + For original position 6: select_indices[6] = select_positions[5] = 5 + For original position 7: select_indices[7] = select_positions[7] = 7 + For original position 8: select_indices[8] = select_positions[8] = 9 + => select_indices = [0, 1, 3, 4, 6, 2, 5, 7, 9] + + Step 4: index_select from gathered tensor using select_indices: + gathered = [0, 1, 5, 2, 3, 6, 4, 7, P, 8, P, P] + result = gathered[select_indices] + = gathered[[0, 1, 3, 4, 6, 2, 5, 7, 9]] + = [0, 1, 2, 3, 4, 5, 6, 7, 8] + + The result is the original replicated tensor [0, 1, 2, 3, 4, 5, 6, 7, 8]. + """ + num_chunks = mesh.size(mesh_dim=mesh_dim) + logical_dim_size = current_logical_shape[self.dim] + + # indices_tensor is 1D torch.arange(logical_dim_size) unsqueezed + # so that we can reuse self._split_tensor which splits on self.dim + shape = [1] * self.dim + [logical_dim_size] + # pyrefly: ignore [no-matching-overload] + indices_tensor = torch.arange( + logical_dim_size, device=local_tensor.device + ).view(shape) + + sharded_indices, _ = self._split_tensor( + indices_tensor, + num_chunks, + with_padding=False, + contiguous=False, + ) + # squeeze back to 1D indices tensor + sharded_indices = [shard.view(-1) for shard in sharded_indices] + + # First chunk should be one of those biggest chunks. Compute the size + # from the logical shape rather than the fake arange split so padding + # depends only on the original symbolic dimension and its hint. + max_chunk_size, _ = self.local_shard_size_and_offset( + logical_dim_size, + num_chunks, + rank=0, + offset_mode=_StridedShardOffsetMode.ALL, + ) + local_pad_size = max_chunk_size - local_tensor.size(self.dim) + local_tensor_padded = pad_tensor(local_tensor, self.dim, local_pad_size) + + if not local_tensor_padded.is_contiguous(): + local_tensor_padded = local_tensor_padded.contiguous() + + replicate_tensor_permuted_padded = funcol.all_gather_single( + local_tensor_padded, + gather_dim=self.dim, + group=(mesh, mesh_dim), + ) + if isinstance(replicate_tensor_permuted_padded, funcol.AsyncCollectiveTensor): + replicate_tensor_permuted_padded = replicate_tensor_permuted_padded.wait() + + # After all_gather, the tensor is [chunk0, chunk1 (may padded), ...]. + # Each chunk may have padding at the end. Use a single index_select to + # both extract non-padding data and reorder to the original positions. + # + # Build select_indices where select_indices[original_pos] = position in + # the padded tensor that holds the element for original_pos. + padded_positions: list[torch.Tensor] = [] + for i, shard in enumerate(sharded_indices): + base_offset = i * max_chunk_size + positions = cast( + torch.Tensor, + base_offset + + torch.arange( + shard.numel(), + device=local_tensor.device, + ), + ) + padded_positions.append(positions) + + # Permutation ends up containing strided indices because we create it by + # chunking over a particular dimension of an N-D shaped arange tensor. + permutation = torch.cat(sharded_indices) + # Choose the position by skipping padding indices from + # replicate_tensor_permuted_padded. + select_positions = torch.cat(padded_positions) + + inv_permutation = torch.argsort(permutation) + select_indices = select_positions.index_select(0, inv_permutation) + + replicate_tensor = torch.index_select( + replicate_tensor_permuted_padded, self.dim, select_indices + ) + logical_shape = cast(list[IntLikeType], list(replicate_tensor.shape)) + logical_shape[self.dim] = current_logical_shape[self.dim] + replicate_tensor = replicate_tensor.as_strided( + logical_shape, + torch._prims_common.make_contiguous_strides_for( + cast(list[int], logical_shape) + ), + ) + + return replicate_tensor.contiguous() + + def _replicate_to_strided_shard( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + shard_index: IntLikeType, + ) -> torch.Tensor: + """ + Transform from replicated tensor to a strided-sharded tensor on the current rank. + + This performs a local chunking operation using the _StridedShard pattern, + where the tensor is split according to the strided sharding semantics + (interleaved pieces based on split_factor). + + Args: + local_tensor: The replicated tensor on this rank. + mesh: The device mesh over which the tensor is distributed. + mesh_dim: The mesh dimension for the sharding. + shard_index: The index of the shard to select (typically the rank's + coordinate on the mesh dimension). + + Returns: + The local strided shard for this rank. + """ + num_chunks = mesh.size(mesh_dim=mesh_dim) + return self._select_split_tensor( + local_tensor, + num_chunks, + shard_index, + with_padding=False, + clone=True, + ) + + @staticmethod + @maybe_run_for_local_tensor + def _local_shard_size(sharded_indices: list[torch.Tensor], rank: RankType) -> int: + return len(sharded_indices[rank]) + + def _local_shard_size_and_offset( + self, + curr_local_size: int, + num_chunks: int, + rank: RankType, + offset_mode: _StridedShardOffsetMode = _StridedShardOffsetMode.FIRST, + ) -> tuple[int, int | list[int] | None]: + return self.local_shard_size_and_offset( + curr_local_size, num_chunks, rank, offset_mode + ) + + @maybe_run_for_local_tensor + def local_shard_size_and_offset( + self, + curr_local_size: IntLikeType, + num_chunks: int, + rank: RankType, + offset_mode: _StridedShardOffsetMode = _StridedShardOffsetMode.FIRST, + ) -> tuple[int, list[int] | int | None]: + """ + Compute the local shard size and offset(s) for a _StridedShard placement. + + Unlike the regular Shard placement which produces contiguous offsets, _StridedShard + produces non-contiguous (strided) offsets due to the right-to-left sharding semantics. + This method computes the actual indices that belong to the local shard. + + Args: + self (_StridedShard): The _StridedShard placement instance. + curr_local_size (int): The current size of the tensor dimension to be sharded. + num_chunks (int): Number of chunks to split the dimension into (typically the mesh dimension size). + rank (RankType): The rank index to compute the shard for. + offset_mode (_StridedShardOffsetMode): Controls offset materialization. + FIRST returns only the first offset as an int, ALL returns all offsets + as a list, and NONE skips offset materialization. Defaults to FIRST. + + Returns: + tuple: A tuple containing: + - local_shard_size (int): The number of elements in the local shard + for this rank. + - offset (int | list[int] | None): Determined by offset_mode. FIRST + returns the first offset as an int, ALL returns a list of all offsets, + and NONE returns None. + """ + offset_mode = _StridedShardOffsetMode(offset_mode) + from torch.fx.experimental.symbolic_shapes import free_unbacked_symbols + + if ( + offset_mode is not _StridedShardOffsetMode.FIRST + and isinstance(curr_local_size, torch.SymInt) + and free_unbacked_symbols(curr_local_size.node.expr) + ): + split_factor = self.split_factor + split_factor_int = self._split_factor_int() + offset = None if offset_mode is _StridedShardOffsetMode.NONE else [] + if _hint_proves_even_shard(curr_local_size, split_factor * num_chunks): + # pyrefly: ignore[bad-return] + return ( + curr_local_size // num_chunks, + offset, + ) + + local_shard_size: IntLikeType = 0 + first_split_size = (curr_local_size + split_factor - 1) // split_factor + for split_idx in range(split_factor_int): + first_start = torch.sym_min( + first_split_size * split_idx, curr_local_size + ) + first_end = torch.sym_min( + first_split_size * (split_idx + 1), curr_local_size + ) + first_len = torch.sym_max(0, first_end - first_start) + second_split_size = (first_len + num_chunks - 1) // num_chunks + second_start = torch.sym_min(second_split_size * rank, first_len) + second_end = torch.sym_min(second_split_size * (rank + 1), first_len) + local_shard_size = local_shard_size + torch.sym_max( + 0, second_end - second_start + ) + return local_shard_size, offset # pyrefly: ignore[bad-return] + + # indices_tensor is 1D torch.arange(logical_dim_size) unsqueezed + # so that we can reuse self._split_tensor which splits on self.dim + shape = [1] * self.dim + [curr_local_size] + # pyrefly: ignore [no-matching-overload] + indices_tensor = torch.arange( + curr_local_size, + ).view(shape) + + sharded_indices, _ = self._split_tensor( + indices_tensor, + num_chunks, + with_padding=False, + contiguous=False, + ) + # squeeze back to 1D indices tensor + sharded_indices = [shard.view(-1) for shard in sharded_indices] + + local_shard_size = _StridedShard._local_shard_size(sharded_indices, rank) + if offset_mode is _StridedShardOffsetMode.NONE: + # Avoid .tolist() which creates unbacked SymInts under FakeTensorMode. + return local_shard_size, None + if local_shard_size > 0: + offsets = sharded_indices[rank].tolist() + else: + offsets = [] + + if offset_mode is _StridedShardOffsetMode.FIRST: + # Always return an int for consistency across ranks. + # For empty shards, return -1 as an invalid offset indicator. + offsets = offsets[0] if len(offsets) > 0 else -1 + + return local_shard_size, offsets + + @staticmethod + def _compute_padding_info( + logical_size_on_dim: int, + num_chunks: int, + shard_dim: int, + split_factor: int = 1, + ) -> tuple[bool, int]: + """ + Compute padding information for _StridedShard collective operations. + + This method calculates whether padding is needed and the maximum chunk size + for collective operations (e.g., all-to-all) involving _StridedShard tensors. + + For _StridedShard with split_factor > 1, the tensor undergoes two-level splitting: + 1. First level: split into ``split_factor`` pieces + 2. Second level: each piece is split into ``num_chunks`` pieces + + The resulting shards are interleaved, so padding must account for both levels. + Unlike ``Shard._compute_padding_info`` where at most one partition (the last) + requires padding, _StridedShard can have multiple trailing partitions that + need padding due to the interleaved structure. + + When split_factor=1, this behaves identically to regular Shard padding logic. + + Args: + logical_size_on_dim: The logical shape size of the tensor ``shard_dim``. + num_chunks: Number of chunks to split into (typically the mesh dim size). + shard_dim: The dimension along which the tensor is sharded. + split_factor: The number of pre-existing splits from right-to-left sharding. + Defaults to 1 (no strided sharding effect). + + Returns: + A tuple of (needs_padding_on_dim, max_chunk_size): + - needs_padding_on_dim: Whether padding is required on the shard dimension. + - max_chunk_size: The maximum chunk size per rank after both levels of splitting. + """ + + def _ceil_div(a: int, b: int) -> int: + return (a + b - 1) // b + + if split_factor != 1: + # Computing padding info for StridedShard tensor dim + # First level: split into split_factor pieces + first_chunk_size = _ceil_div(logical_size_on_dim, split_factor) + num_full_first_chunks = logical_size_on_dim // first_chunk_size + remainder = logical_size_on_dim - num_full_first_chunks * first_chunk_size + + # Determine if padding is needed: + # - remainder > 0 means one chunk has partial size + # - empty first-level chunks means some ranks get nothing at each level + # - first_chunk_size % num_chunks != 0 means uneven second-level split + # (e.g., 6 rows, sf=2, 8 ranks: first_chunk=3, but only 3 of 8 ranks + # get data per first-level chunk, so ranks 3-7 have size-0 tensors) + has_partial_chunk = remainder > 0 + num_empty_chunks = ( + split_factor - num_full_first_chunks - (1 if has_partial_chunk else 0) + ) + has_uneven_second_level = first_chunk_size % num_chunks != 0 + needs_padding_on_dim = ( + has_partial_chunk or num_empty_chunks > 0 or has_uneven_second_level + ) + + # Second level: each first-level chunk is split into num_chunks pieces + # Calculate the per-rank chunk size after both levels of splitting + max_chunk_size = ( + _ceil_div(first_chunk_size, num_chunks) * num_full_first_chunks + ) + if has_partial_chunk: + max_chunk_size += _ceil_div(remainder, num_chunks) + else: + # Compute padding info for normal shard, no split_factor impact. + needs_padding_on_dim = logical_size_on_dim % num_chunks != 0 + max_chunk_size = _ceil_div(logical_size_on_dim, num_chunks) + + return ( + needs_padding_on_dim, + max_chunk_size, + ) + + @maybe_run_for_local_tensor + def _pad_for_new_shard_dim( + self, + current_logical_shape: list[int], + local_tensor: torch.Tensor, + num_chunks: int, + old_shard_dim: int, + new_shard_dim: int, + split_factor: int, + ) -> torch.Tensor: + """ + Pad the local tensor before alltoall for _StridedShard's _to_new_shard_dim operation. + + For _StridedShard, the padding must account for the split_factor since each rank + holds split_factor interleaved pieces. The full chunk size per rank is + split_factor * single_chunk_size. + """ + ( + old_dim_padding, + old_dim_max_chunk_size, + ) = _StridedShard._compute_padding_info( + current_logical_shape[old_shard_dim], + num_chunks, + old_shard_dim, + split_factor, + ) + ( + new_dim_padding, + new_dim_max_chunk_size, + ) = _StridedShard._compute_padding_info( + current_logical_shape[new_shard_dim], + num_chunks, + new_shard_dim, + ) + + if old_dim_padding: + target_size = old_dim_max_chunk_size + old_dim_pad_size = target_size - local_tensor.size(old_shard_dim) + if old_dim_pad_size > 0: + local_tensor = pad_tensor(local_tensor, old_shard_dim, old_dim_pad_size) + + if new_dim_padding: + target_total_size = new_dim_max_chunk_size * num_chunks + new_dim_pad_size = target_total_size - local_tensor.size(new_shard_dim) + if new_dim_pad_size > 0: + local_tensor = pad_tensor(local_tensor, new_shard_dim, new_dim_pad_size) + + if not local_tensor.is_contiguous(): + local_tensor = local_tensor.contiguous() + return local_tensor + + @maybe_run_for_local_tensor + def _unpad_for_new_shard_dim( + self, + current_logical_shape: list[int], + local_tensor: torch.Tensor, + num_chunks: int, + old_shard_dim: int, + new_shard_dim: int, + split_factor: int, + local_rank: int, + ) -> torch.Tensor: + """ + Unpad the local tensor after alltoall for _StridedShard's + _to_new_shard_dim operation. + + This uses the same index_select approach as _to_replicate_tensor to both + extract correct elements (removing padding) and reorder them. The key insight + is that padding can only be at the end, so the first shard has the max size. + """ + old_dim_logical_size = current_logical_shape[old_shard_dim] + new_dim_logical_size = current_logical_shape[new_shard_dim] + ( + new_dim_padding, + new_dim_max_chunk_size, + ) = _StridedShard._compute_padding_info( + current_logical_shape[new_shard_dim], + num_chunks, + new_shard_dim, + ) + + # Build sharded indices to understand the strided pattern + shape = [1] * old_shard_dim + [old_dim_logical_size] + indices_tensor = torch.arange( + old_dim_logical_size, device=local_tensor.device + ).view(shape) + sharded_indices, _ = self._split_tensor( + indices_tensor, num_chunks, with_padding=False, contiguous=False + ) + sharded_indices = [shard.view(-1) for shard in sharded_indices] + + max_chunk_size = len(sharded_indices[0]) + + # Compute positions in the padded local_tensor for each actual element. + # Each shard i starts at position i * max_chunk_size in local_tensor. + padded_positions = [] + for i, shard in enumerate(sharded_indices): + base_offset = i * max_chunk_size + positions = base_offset + torch.arange( + len(shard), device=local_tensor.device + ) + padded_positions.append(positions) + + permutation = torch.cat(sharded_indices) + select_positions = torch.cat(padded_positions) + + inv_permutation = torch.argsort(permutation) + select_indices = select_positions.index_select(0, inv_permutation) + local_tensor = torch.index_select(local_tensor, old_shard_dim, select_indices) + + if new_dim_padding: + # copy from Shard._unpad_for_new_shard_dim + local_shard_size_on_new_dim = Shard.local_shard_size_and_offset( + new_dim_logical_size, num_chunks, local_rank + )[0] + new_dim_unpad_size = new_dim_max_chunk_size - local_shard_size_on_new_dim # type: ignore[possibly-undefined] + local_tensor = unpad_tensor(local_tensor, new_shard_dim, new_dim_unpad_size) # type: ignore[possibly-undefined] + return local_tensor + + def _to_new_shard_dim( + self, + local_tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + current_logical_shape: list[int], + new_shard_dim: int, + ) -> torch.Tensor: + """ + Transform from existing _StridedShard tensor to a new _StridedShard tensor that + shards on a new dimension, which performs an alltoall. + + This is analogous to Shard._to_new_shard_dim but accounts for the split_factor + in padding calculations since _StridedShard represents interleaved sharding. + """ + my_coordinate = mesh.get_coordinate() + if my_coordinate is None: + # if rank is not part of mesh, we simply return local_tensor, + # which should be an empty tensor + return local_tensor + + num_chunks = mesh.size(mesh_dim=mesh_dim) + + local_tensor = self._pad_for_new_shard_dim( + current_logical_shape, + local_tensor, + num_chunks, + self.dim, + new_shard_dim, + self._split_factor_int(), + ) + + new_tensor = shard_dim_alltoall( + local_tensor, self.dim, new_shard_dim, mesh, mesh_dim + ) + new_tensor = self._unpad_for_new_shard_dim( + current_logical_shape, + new_tensor, + num_chunks, + self.dim, + new_shard_dim, + self._split_factor_int(), + my_coordinate[mesh_dim], + ) + + return new_tensor + + +def _is_shard_like(p: "Placement") -> TypeGuard[Shard | _StridedShard]: + """Check if a placement is Shard or _StridedShard. + + Use this instead of ``isinstance(p, Shard)`` to avoid silently missing + ``_StridedShard``. When ``_StridedShard`` is unified with ``Shard`` + (see TODO on the class), this helper can be collapsed to a single + ``isinstance`` check. + """ + return isinstance(p, Shard | _StridedShard) + + +class Replicate(torch._C._distributed.Replicate): + """ + The ``Replicate()`` placement describes the DTensor replicating on a corresponding + ``DeviceMesh`` dimension, where each rank on the DeviceMesh dimension holds a + replica of the global Tensor. The ``Replicate`` placement can be used by all + DTensor APIs (i.e. ``distribute_tensor``, ``DTensor.from_local``, etc.) + """ + + def __hash__(self) -> int: + # every replicate placement is the same + return -1 + + def __repr__(self) -> str: + """ + machine readable representation of the Replicate placement + """ + return "Replicate()" + + def __fx_repr__(self): + """ + Returns FX-evaluable repr and required globals for Replicate placement. + Needed for passing this type as an opaque object input to a custom op. + """ + return "torch.distributed.tensor.placement_types.Replicate()", {} + + def __str__(self) -> str: + """ + human readable representation of the Replicate placement + """ + return "R" + + @classmethod + def _make_replicate_tensor( + cls, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + ) -> torch.Tensor: + """ + Replicate (broadcast) a torch.Tensor on a mesh dimension (use + the first coordinate on the mesh dimension as source of truth) + """ + my_coordinate = mesh.get_coordinate() + if my_coordinate is None: + # if rank is not part of mesh, we simply return an empty tensor + return tensor.new_empty(0, requires_grad=tensor.requires_grad) + + tensor = tensor.contiguous() + + if src_data_rank is not None: + # perform broadcast from the src_data_rank as data source when it is not None + mesh_broadcast(tensor, mesh, mesh_dim=mesh_dim, group_src=src_data_rank) + return tensor + + def _replicate_tensor( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + src_data_rank: int | None = 0, + ) -> torch.Tensor: + return Replicate._make_replicate_tensor(tensor, mesh, mesh_dim, src_data_rank) + + +class Partial(torch._C._distributed.Partial): + # reduce_ops that distribute over addition, enabling per-input linearity + # for bilinear ops like mm: reduce_op(A_i @ B) = reduce_op(A_i) @ B + LINEAR_REDUCE_OPS: tuple[str, ...] = ("sum", "avg") + ALL_REDUCE_OPS: tuple[str, ...] = ("sum", "avg", "min", "max", "product") + + """ + The ``Partial(reduce_op)`` placement describes the DTensor that is pending + reduction on a specified ``DeviceMesh`` dimension, where each rank on the + DeviceMesh dimension holds the partial value of the global Tensor. User can + redistribute the ``Partial`` DTensor to a ``Replicate`` or ``Shard(dim)`` + placement on the specified ``DeviceMesh`` dimension using ``redistribute``, + which would trigger necessary communication operations under the hood (i.e. + ``allreduce``, ``reduce_scatter``). + + Args: + reduce_op (str, optional): The reduction op to be used for the partial DTensor + to produce Replicated/Sharded DTensor. Corresponds to the reduce operations + supported by ``torch.distributed.ReduceOp``. Default: "sum". + + Supported values: + + * ``"sum"``: Element-wise sum across all ranks. + * ``"avg"``: Element-wise average across all ranks. + * ``"min"``: Element-wise minimum across all ranks. + * ``"max"``: Element-wise maximum across all ranks. + * ``"product"``: Element-wise product across all ranks. + * ``"band"``: Bitwise AND across all ranks (integer tensors only). + * ``"bor"``: Bitwise OR across all ranks (integer tensors only). + * ``"bxor"``: Bitwise XOR across all ranks (integer tensors only). + + .. note:: The ``Partial`` placement can be generated as a result of DTensor + operators, by ``DTensor.from_local``, or by redistributing a ``Shard`` + placement to ``Partial("sum")``. + """ + + def _reduce_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + # Partial placement contract #1: + # _reduce_value: reduce the value of the tensor on the mesh dimension + return funcol.all_reduce( + tensor, reduceOp=self.reduce_op, group=(mesh, mesh_dim) + ) + + def _reduce_shard_value( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + shard_spec: Placement, + ) -> torch.Tensor: + # Partial placement contract #2: + # _reduce_shard_value: reduce_scatter the value of the tensor over the mesh dimension + shard_spec = cast(Shard, shard_spec) + return shard_spec._reduce_shard_tensor(tensor, mesh, self.reduce_op, mesh_dim) + + def _partition_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + """ + Partition a replicated tensor to create partial values for Replicate → Partial. + + This is the conjugate operation of _reduce_value. The partition operation + must satisfy the invariant that applying _reduce_value to the partitioned + values recovers the original replicated value (modulo floating-point error). + + Mathematical analysis by reduce_op: + + * "sum": partition(v) = v / n, then sum([v/n] * n) = v + Introduces floating-point error from the division and summation. + Error grows with n (the number of ranks). + + * "avg": partition(v) = v, then avg([v] * n) = v + Numerically exact (averaging identical values). + + * "min": partition(v) = v, then min([v] * n) = v + Numerically exact. + + * "max": partition(v) = v, then max([v] * n) = v + Numerically exact. + + * "product": Would need partition(v) = v^(1/n), but n-th root is not exact + for general values and undefined for negative values with even n. + NOT SUPPORTED. + + * "band"/"bor"/"bxor": Bitwise operations have no well-defined inverse + that partitions a value such that reducing recovers the original. + NOT SUPPORTED. + """ + num_chunks = mesh.size(mesh_dim=mesh_dim) + if self.reduce_op == "sum": + return tensor / num_chunks + elif self.reduce_op in ("avg", "min", "max"): + return tensor + else: + raise ValueError( + f"Replicate to Partial({self.reduce_op}) conversion is not supported." + ) + + def __hash__(self) -> int: + return 1 + hash(self.reduce_op) + + def __repr__(self) -> str: + """ + machine readable representation of the Partial placement + """ + return f"Partial({self.reduce_op})" + + def __fx_repr__(self): + """ + Returns FX-evaluable repr and required globals for Partial placement. + Needed for passing this type as an input to a custom op. + """ + return ( + f"torch.distributed.tensor.placement_types.Partial({self.reduce_op!r})", + {}, + ) + + def __str__(self) -> str: + """ + human readable representation of the Partial placement + """ + return f"P({self.reduce_op})" + + +# We keep the old _Partial name for a while for BC reason +_Partial = Partial + + +@dataclass(frozen=True) +class _MaskPartial(Partial): + """ + A partial mask placement devised for rowwise sharded embedding op, where we need + to mask and adjust the indices to the local embedding shard, embedding masking + is a special type of the Partial placement + + NOTE: the lifecycle of this _MaskPartial placement follows the corresponding DTensor + lifecycle, i.e. the indices_mask would only be alive during the lifetime of the DTensor. + """ + + mask_buffer: MaskBuffer = field(default_factory=MaskBuffer) + + # required fields for computing the local offset and deriving the mask + offset_shape: torch.Size | None = None + offset_dim: int = 0 + + def __init__( + self, + reduce_op=None, + mask_buffer=None, + offset_shape=None, + offset_dim=0, + *args, + **kwargs, + ): + super().__init__(reduce_op) + if mask_buffer is None: + mask_buffer = MaskBuffer() + object.__setattr__(self, "mask_buffer", mask_buffer) + object.__setattr__(self, "offset_shape", offset_shape) + object.__setattr__(self, "offset_dim", offset_dim) + + @staticmethod + @maybe_run_for_local_tensor + def _mask_tensor( + tensor: torch.Tensor, + local_offset_on_dim: IntLikeType, + local_shard_size: IntLikeType, + ) -> tuple[torch.Tensor, torch.Tensor]: + # Build the input mask and save it for the current partial placement + # this is so that the output of embedding op can reuse the same partial + # placement saved mask to perform mask + reduction + mask = (tensor < local_offset_on_dim) | ( + tensor >= local_offset_on_dim + local_shard_size + ) + # mask the input tensor + # pyrefly: ignore [unsupported-operation] + masked_tensor = tensor.clone() - local_offset_on_dim + masked_tensor[mask] = 0 + # pyrefly: ignore [bad-return] + return mask, masked_tensor + + def _partition_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + if not mesh._is_current_rank_part_of_mesh(): + raise AssertionError("rank is not part of mesh") + # override parent logic to perform partial mask for embedding + num_chunks = mesh.size(mesh_dim) + # get local shard size and offset on the embedding_dim + if self.offset_shape is None: + raise AssertionError("offset_shape needs to be set for _MaskPartial") + local_shard_size, local_offset_on_dim = Shard.local_shard_size_and_offset( + self.offset_shape[self.offset_dim], + num_chunks, + mesh._sym_get_coordinate(mesh_dim), + ) + mask, masked_tensor = _MaskPartial._mask_tensor( + tensor, local_offset_on_dim, local_shard_size + ) + # materialize the mask buffer to be used for reduction + self.mask_buffer.materialize_mask(mask) + return masked_tensor + + def _reduce_value( + self, tensor: torch.Tensor, mesh: DeviceMesh, mesh_dim: int + ) -> torch.Tensor: + # by the time we need reduction, we should have already saved the mask + if self.mask_buffer.data is None: + raise AssertionError + + # apply the mask to the tensor that pending reduction + self.mask_buffer.apply_mask(tensor) + + # clear the mask buffer + self.mask_buffer.release_mask() + + # perform sum reduction + return funcol.all_reduce( + tensor, reduceOp=self.reduce_op, group=(mesh, mesh_dim) + ) + + def _reduce_shard_value( + self, + tensor: torch.Tensor, + mesh: DeviceMesh, + mesh_dim: int, + shard_spec: Placement, + ) -> torch.Tensor: + # by the time we need reduction, we should have already saved the mask + if self.mask_buffer.data is None: + raise AssertionError + + # apply the mask to the tensor that pending reduction + self.mask_buffer.apply_mask(tensor) + + # clear the mask buffer + self.mask_buffer.release_mask() + + # call reduce_shard_tensor of the shard_spec. + shard_spec = cast(Shard, shard_spec) + return shard_spec._reduce_shard_tensor(tensor, mesh, self.reduce_op, mesh_dim) + + def __eq__(self, other: object) -> bool: + if not isinstance(other, _MaskPartial): + return False + + return ( + self.reduce_op == other.reduce_op + and self.offset_shape == other.offset_shape + and self.offset_dim == other.offset_dim + and self.mask_buffer is other.mask_buffer + ) + + def __hash__(self) -> int: + return 1 + hash( + ( + self.reduce_op, + self.offset_shape, + self.offset_dim, + id(self.mask_buffer), + ) + ) + + def __repr__(self) -> str: + """ + machine readable representation of the _MaskPartial placement + """ + return f"_MaskPartial(reduce_op={self.reduce_op}, offset_shape={self.offset_shape}, offset_dim={self.offset_dim})" + + def __str__(self) -> str: + """ + human readable representation of the _MaskPartial placement + """ + return f"MaskP({self.reduce_op}, {self.offset_shape}, {self.offset_dim})" + + def __fx_repr__(self): + """ + Returns FX-evaluable repr and required globals for Partial placement. + Needed for passing this type as an input to a custom op. + """ + return ( + f"torch.distributed.tensor.placement_types.MaskPartial(reduce_op={self.reduce_op}, offset_shape={self.offset_shape}, offset_dim={self.offset_dim})", + {}, + ) + + +def _register_placements_as_opaque(): + from torch._library.opaque_object import MemberType, register_custom_class + + allowed_members = { + "is_shard": MemberType.USE_REAL, + "is_partial": MemberType.USE_REAL, + "is_replicate": MemberType.USE_REAL, + "__eq__": MemberType.USE_REAL, + } + register_custom_class(Placement, typ="constant", members=allowed_members) + register_custom_class( + Shard, typ="constant", members=allowed_members | {"dim": MemberType.USE_REAL} + ) + register_custom_class(Replicate, typ="constant", members=allowed_members) + register_custom_class( + Partial, + typ="constant", + members=allowed_members | {"reduce_op": MemberType.USE_REAL}, + ) + register_custom_class( + _StridedShard, + typ="constant", + members=allowed_members | {"dim": MemberType.USE_REAL}, + ) + register_custom_class( + _MaskPartial, + typ="constant", + members=allowed_members | {"reduce_op": MemberType.USE_REAL}, + ) + + +_register_placements_as_opaque() diff --git a/.venv/lib/python3.12/site-packages/torch/distributed/utils.py b/.venv/lib/python3.12/site-packages/torch/distributed/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..890889a3a2adfa574ecdec1743166d2b6563307b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributed/utils.py @@ -0,0 +1,394 @@ +# mypy: allow-untyped-defs +import dataclasses +import traceback +from collections import OrderedDict +from collections.abc import Callable, Container +from typing import Any, Optional, overload, TypeVar + +import torch +import torch.distributed as dist +from torch import nn +from torch.nn.utils.rnn import PackedSequence + + +__all__ = [] # type: ignore[var-annotated] + + +def _pack_kwargs(*args: Any, **kwargs: Any) -> tuple[tuple[Any, ...], tuple[str, ...]]: + """ + Turn argument list into separate key list and value list (unpack_kwargs does the opposite). + + Inspiration: https://github.com/facebookresearch/fairscale/blob/eeb6684/fairscale/internal/containers.py#L70 + Usage:: + + kwarg_keys, flat_args = pack_kwargs(1, 2, a=3, b=4) + assert kwarg_keys == ("a", "b") + assert flat_args == (1, 2, 3, 4) + args, kwargs = unpack_kwargs(kwarg_keys, flat_args) + assert args == (1, 2) + assert kwargs == {"a": 3, "b": 4} + Returns: + Tuple[Tuple[Any, ...], Tuple[str, ...]]: The first tuple element gives + gives both positional args and kwarg values, where the positional args + proceed kwarg values and kwarg values are ordered consistently with the + kwarg keys. The second tuple element gives the kwarg keys. + The second tuple element's length is at most the first tuple element's length. + """ + kwarg_keys: list[str] = [] + flat_args: list[Any] = list(args) + for k, v in kwargs.items(): + kwarg_keys.append(k) + flat_args.append(v) + + return tuple(flat_args), tuple(kwarg_keys) + + +def _cast_forward_inputs( + dtype: torch.dtype | None, + *args: Any, + **kwargs: Any, +) -> tuple[Any, Any]: + """ + Cast floating point tensors in ``args`` and ``kwargs`` to ``input_dtype``. + + This respects the existing ``requires_grad`` on the tensors. + """ + if dtype is None: + return args, kwargs + + def cast_fn(x: torch.Tensor) -> torch.Tensor: + if not torch.is_floating_point(x) or x.dtype == dtype: + return x + + return x.to(dtype) + + return (_apply_to_tensors(cast_fn, args), _apply_to_tensors(cast_fn, kwargs)) + + +def _unpack_kwargs( + flat_args: tuple[Any, ...], kwarg_keys: tuple[str, ...] +) -> tuple[tuple[Any, ...], dict[str, Any]]: + """See _pack_kwargs.""" + if len(kwarg_keys) > len(flat_args): + raise AssertionError(f"too many keys {len(kwarg_keys)} vs. {len(flat_args)}") + if len(kwarg_keys) == 0: + return flat_args, {} + args = flat_args[: -len(kwarg_keys)] + kwargs = dict(zip(kwarg_keys, flat_args[-len(kwarg_keys) :])) + return args, kwargs + + +S = TypeVar("S", dict, list, tuple) +T = TypeVar("T", torch.Tensor, PackedSequence) + + +@overload +def _recursive_to( + inputs: S, target_device: torch.device, use_side_stream_for_tensor_copies: bool +) -> list[S]: ... + + +@overload +def _recursive_to( + inputs: T, target_device: torch.device, use_side_stream_for_tensor_copies: bool +) -> tuple[T]: ... + + +def _recursive_to(inputs, target_device, use_side_stream_for_tensor_copies): + r"""Recursively moves input to the target_device.""" + + def to_map(obj): + if isinstance(obj, (torch.Tensor, PackedSequence)): + device = obj.data.device if isinstance(obj, PackedSequence) else obj.device + if device == target_device: + return (obj,) + if not use_side_stream_for_tensor_copies: + return (obj.to(target_device),) + else: + # If the custom module is not registered to torch, stream is not used for acceleration + if device.type == "cpu": + return (obj.to(target_device),) + + from torch.nn.parallel._functions import _get_stream + + # Perform CPU -> target_device copies in a background stream. This code is + # motivated from similar logic in torch/nn/parallel/_functions.py + stream = _get_stream(target_device) + with stream: + output = obj.to(target_device) + # synchronize with the copy stream + with torch.accelerator.device_index(target_device.index): + current_stream = torch.accelerator.current_stream() + # Sync the current stream with the copy stream + current_stream.wait_stream(stream) + # Ensure tensor memory is not reused until work on + # main stream is complete + if isinstance(obj, PackedSequence): + output.data.record_stream(current_stream) # type: ignore[arg-type] + else: + if not isinstance(output, torch.Tensor): + raise AssertionError("output must be a torch.Tensor") + output.record_stream(current_stream) # type: ignore[arg-type] + return (output,) + + from torch.nn.parallel.scatter_gather import _is_namedtuple + + def _handle_container(obj, elements, make_container): + """Handle container types with object identity preservation.""" + # pyrefly: ignore [bad-argument-type] + mapped = list(map(to_map, elements)) + # Preserve object identity when all elements are unchanged (single-device case) + if all(len(m) == 1 for m in mapped): + transformed = [m[0] for m in mapped] + if all(t is o for t, o in zip(transformed, elements)): + return [obj] + # pyrefly: ignore [no-matching-overload] + return [make_container(transformed)] + # pyrefly: ignore [no-matching-overload] + return [make_container(args) for args in zip(*mapped)] + + if _is_namedtuple(obj): + return _handle_container(obj, obj, lambda x: type(obj)(*x)) + if isinstance(obj, tuple) and len(obj) > 0: + return _handle_container(obj, obj, tuple) + if isinstance(obj, list) and len(obj) > 0: + return _handle_container(obj, obj, list) + if isinstance(obj, dict) and len(obj) > 0: + keys = list(obj.keys()) + return _handle_container( + obj, obj.values(), lambda v: type(obj)(zip(keys, v)) + ) + return [obj] + + # Avoid reference cycle + try: + res = to_map(inputs) + finally: + to_map = None # type: ignore[assignment] + return res + + +def _p_assert(cond: Any, s: str, raise_assertion_error: bool = True) -> None: + """Alternate to ``assert`` when in the backward context to print the error message ``s`` since otherwise, it is swallowed.""" + if not cond: + print(s) + traceback.print_stack() + if raise_assertion_error: + raise AssertionError(s) + + +def _alloc_storage(tensor: torch.Tensor, size: torch.Size) -> None: + """ + Allocate storage for ``tensor`` with the given size. + + Returns: + bool: ``True`` if this method allocated storage and ``False`` if the + storage was already allocated. + """ + with torch.no_grad(): + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + already_allocated = tensor._typed_storage()._size() == size.numel() + if not already_allocated: + tensor_storage_size = tensor._typed_storage()._size() + _p_assert( + tensor_storage_size == 0, + "Tensor storage should have been resized to be 0 but got PLACEHOLDER", + ) + tensor._typed_storage()._resize_(size.numel()) + + +def _free_storage(tensor: torch.Tensor): + """ + Frees the underlying storage of ``tensor``. + + Returns: + bool: ``True`` if the method freed the storage and ``False`` if the + storage was already freed. + """ + with torch.no_grad(): + if not torch.distributed._functional_collectives.is_torchdynamo_compiling(): + already_freed = tensor._typed_storage()._size() == 0 + if not already_freed: + _p_assert( + tensor.storage_offset() == 0, + "Freeing a tensor's storage is unsafe when it is not the sole occupant\n" + f"storage offset: {tensor.storage_offset()}\n" + f"storage size: {tensor._typed_storage()._size()}\n" + f"tensor shape: {tensor.shape}", + ) + tensor._typed_storage()._resize_(0) + + +Q = TypeVar("Q") +R = TypeVar("R", dict, list, tuple, set, OrderedDict, PackedSequence, Any) + + +@overload +def _apply_to_tensors( + fn: Callable[[torch.Tensor], Q], container: torch.Tensor +) -> Q: ... + + +@overload +def _apply_to_tensors(fn: Callable[[torch.Tensor], Any], container: R) -> R: ... + + +def _apply_to_tensors(fn, container): + """Recursively apply to all tensor in different kinds of container types.""" + + def apply(x): + from torch.nn.parallel.scatter_gather import _is_namedtuple + + if isinstance(x, torch.Tensor): + return fn(x) + elif hasattr(x, "__dataclass_fields__"): + dc = dataclasses.replace(x) + changes = { + f.name: apply(getattr(dc, f.name)) for f in dataclasses.fields(dc) + } + return dataclasses.replace(dc, **changes) + elif isinstance(x, OrderedDict): + od = x.__class__() + for key, value in x.items(): + od[key] = apply(value) + return od + elif isinstance(x, PackedSequence): + apply(x.data) + return x + elif isinstance(x, dict): + return {key: apply(value) for key, value in x.items()} + elif _is_namedtuple(x): + res = (apply(el) for el in x) + return type(x)(*res) + elif isinstance(x, (list, tuple, set)): + return type(x)(apply(el) for el in x) + else: + return x + + return apply(container) + + +def _to_kwargs( + inputs: tuple[Any, ...], + kwargs: dict[str, Any] | None, + target_device: torch.device, + use_side_stream_for_tensor_copies: bool, +) -> tuple[tuple[Any, ...], tuple[dict[str, Any], ...]]: + moved_inputs = ( + _recursive_to(inputs, target_device, use_side_stream_for_tensor_copies) + if inputs + else [] + ) + moved_kwargs = ( + _recursive_to(kwargs, target_device, use_side_stream_for_tensor_copies) + if kwargs + else [] + ) + if len(moved_inputs) < len(moved_kwargs): + moved_inputs.extend([() for _ in range(len(moved_kwargs) - len(inputs))]) + elif len(moved_kwargs) < len(moved_inputs): + moved_kwargs.extend([{} for _ in range(len(moved_inputs) - len(moved_kwargs))]) + return tuple(moved_inputs), tuple(moved_kwargs) + + +def _verify_param_shape_across_processes( + process_group: dist.ProcessGroup, + tensors: list[torch.Tensor], + logger: Optional["dist.Logger"] = None, +): + return dist._verify_params_across_processes(process_group, tensors, logger) + + +def _sync_module_states( + module: nn.Module, + process_group: dist.ProcessGroup, + broadcast_bucket_size: int, + src: int, + params_and_buffers_to_ignore: Container[str], + broadcast_buffers: bool = True, +) -> None: + """ + Sync ``module``'s parameters and buffers state. + + Syncs ``module``'s parameters and buffers state so that all ranks contain + the same module state across all ranks. Note that this API assumes that all + parameter shapes are consistent before running the synchronization. This can + be checked with ``_verify_param_shape_across_processes``. + """ + module_states: list[torch.Tensor] = [] + for name, param in module.named_parameters(): + if name not in params_and_buffers_to_ignore: + module_states.append(param.detach()) + + if broadcast_buffers: + for name, buffer in module.named_buffers(): + if name not in params_and_buffers_to_ignore: + module_states.append(buffer.detach()) + + _sync_params_and_buffers(process_group, module_states, broadcast_bucket_size, src) + + +def _sync_params_and_buffers( + process_group: dist.ProcessGroup, + module_states: list[torch.Tensor], + broadcast_bucket_size: int, + src: int, +) -> None: + """Synchronize ``module_states`` (list of tensors) across all processes by broadcasting them from rank 0.""" + if len(module_states) > 0: + dist._broadcast_coalesced( + process_group, module_states, broadcast_bucket_size, src + ) + + +def _replace_by_prefix( + state_dict: dict[str, Any], + old_prefix: str, + new_prefix: str, +) -> None: + """ + Replace all keys that match a given old_prefix with a new_prefix (in-place). + + Usage:: + + state_dict = {"layer.xyz": torch.tensor(1)} + replace_by_prefix_(state_dict, "layer.", "module.layer.") + assert state_dict == {"module.layer.xyz": torch.tensor(1)} + """ + if old_prefix == new_prefix: + raise ValueError("old_prefix and new_prefix must be distinct") + for key in list(state_dict.keys()): + if not key.startswith(old_prefix): + continue + new_key = new_prefix + key[len(old_prefix) :] + state_dict[new_key] = state_dict[key] + del state_dict[key] + + +def _data_ptr_allocated(tensor: torch.Tensor) -> bool: + return tensor.untyped_storage().data_ptr() > 0 + + +def _get_root_modules(modules: list[nn.Module]) -> list[nn.Module]: + """ + Returns the modules in ``modules`` that are root modules (i.e. + parent-less) with respect to the set ``modules``. In other words, these + are the modules in ``modules`` that are the not child of any other + module in ``modules``. + """ + root_modules: list[nn.Module] = [] + module_to_modules: dict[nn.Module, set[nn.Module]] = { + module: set(module.modules()) for module in modules + } + for candidate_module in modules: + is_root_module = True + for module, _modules in module_to_modules.items(): + is_child_module = ( + candidate_module is not module and candidate_module in _modules + ) + if is_child_module: + is_root_module = False + break + if is_root_module: + root_modules.append(candidate_module) + return root_modules diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__init__.py b/.venv/lib/python3.12/site-packages/torch/distributions/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..79407620fc9c4f3c94ec42a041a384bac05e89d9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/__init__.py @@ -0,0 +1,174 @@ +r""" +The ``distributions`` package contains parameterizable probability distributions +and sampling functions. This allows the construction of stochastic computation +graphs and stochastic gradient estimators for optimization. This package +generally follows the design of the `TensorFlow Distributions`_ package. + +.. _`TensorFlow Distributions`: + https://arxiv.org/abs/1711.10604 + +It is not possible to directly backpropagate through random samples. However, +there are two main methods for creating surrogate functions that can be +backpropagated through. These are the score function estimator/likelihood ratio +estimator/REINFORCE and the pathwise derivative estimator. REINFORCE is commonly +seen as the basis for policy gradient methods in reinforcement learning, and the +pathwise derivative estimator is commonly seen in the reparameterization trick +in variational autoencoders. Whilst the score function only requires the value +of samples :math:`f(x)`, the pathwise derivative requires the derivative +:math:`f'(x)`. The next sections discuss these two in a reinforcement learning +example. For more details see +`Gradient Estimation Using Stochastic Computation Graphs`_ . + +.. _`Gradient Estimation Using Stochastic Computation Graphs`: + https://arxiv.org/abs/1506.05254 + +Score function +^^^^^^^^^^^^^^ + +When the probability density function is differentiable with respect to its +parameters, we only need :meth:`~torch.distributions.Distribution.sample` and +:meth:`~torch.distributions.Distribution.log_prob` to implement REINFORCE: + +.. math:: + + \Delta\theta = \alpha r \nabla_\theta \log \pi_\theta(a \mid s) + +where :math:`\theta` are the parameters, :math:`\alpha` is the learning rate, +:math:`r` is the reward and :math:`\pi_\theta(a \mid s)` is the probability of +taking action :math:`a` in state :math:`s` given policy :math:`\pi_\theta`. + +In practice we would sample an action from the output of a network, apply this +action in an environment, and then use ``log_prob`` to construct an equivalent +loss function. Note that we use a negative because optimizers use gradient +descent, whilst the rule above assumes gradient ascent. With a categorical +policy, the code for implementing REINFORCE would be as follows:: + + probs = policy_network(state) + # Note that this is equivalent to what used to be called multinomial + m = Categorical(probs) + action = m.sample() + next_state, reward = env.step(action) + loss = -m.log_prob(action) * reward + loss.backward() + +Pathwise derivative +^^^^^^^^^^^^^^^^^^^ + +The other way to implement these stochastic/policy gradients would be to use the +reparameterization trick from the +:meth:`~torch.distributions.Distribution.rsample` method, where the +parameterized random variable can be constructed via a parameterized +deterministic function of a parameter-free random variable. The reparameterized +sample therefore becomes differentiable. The code for implementing the pathwise +derivative would be as follows:: + + params = policy_network(state) + m = Normal(*params) + # Any distribution with .has_rsample == True could work based on the application + action = m.rsample() + next_state, reward = env.step(action) # Assuming that reward is differentiable + loss = -reward + loss.backward() +""" + +from . import transforms +from .bernoulli import Bernoulli +from .beta import Beta +from .binomial import Binomial +from .categorical import Categorical +from .cauchy import Cauchy +from .chi2 import Chi2 +from .constraint_registry import biject_to, transform_to +from .continuous_bernoulli import ContinuousBernoulli +from .dirichlet import Dirichlet +from .distribution import Distribution +from .exp_family import ExponentialFamily +from .exponential import Exponential +from .fishersnedecor import FisherSnedecor +from .gamma import Gamma +from .generalized_pareto import GeneralizedPareto +from .geometric import Geometric +from .gumbel import Gumbel +from .half_cauchy import HalfCauchy +from .half_normal import HalfNormal +from .independent import Independent +from .inverse_gamma import InverseGamma +from .kl import _add_kl_info, kl_divergence, register_kl +from .kumaraswamy import Kumaraswamy +from .laplace import Laplace +from .lkj_cholesky import LKJCholesky +from .log_normal import LogNormal +from .logistic_normal import LogisticNormal +from .lowrank_multivariate_normal import LowRankMultivariateNormal +from .mixture_same_family import MixtureSameFamily +from .multinomial import Multinomial +from .multivariate_normal import MultivariateNormal +from .negative_binomial import NegativeBinomial +from .normal import Normal +from .one_hot_categorical import OneHotCategorical, OneHotCategoricalStraightThrough +from .pareto import Pareto +from .poisson import Poisson +from .relaxed_bernoulli import RelaxedBernoulli +from .relaxed_categorical import RelaxedOneHotCategorical +from .studentT import StudentT +from .transformed_distribution import TransformedDistribution +from .transforms import * # noqa: F403 +from .uniform import Uniform +from .von_mises import VonMises +from .weibull import Weibull +from .wishart import Wishart + + +_add_kl_info() +del _add_kl_info + +__all__ = [ + "Bernoulli", + "Beta", + "Binomial", + "Categorical", + "Cauchy", + "Chi2", + "ContinuousBernoulli", + "Dirichlet", + "Distribution", + "Exponential", + "ExponentialFamily", + "FisherSnedecor", + "Gamma", + "GeneralizedPareto", + "Geometric", + "Gumbel", + "HalfCauchy", + "HalfNormal", + "Independent", + "InverseGamma", + "Kumaraswamy", + "LKJCholesky", + "Laplace", + "LogNormal", + "LogisticNormal", + "LowRankMultivariateNormal", + "MixtureSameFamily", + "Multinomial", + "MultivariateNormal", + "NegativeBinomial", + "Normal", + "OneHotCategorical", + "OneHotCategoricalStraightThrough", + "Pareto", + "RelaxedBernoulli", + "RelaxedOneHotCategorical", + "StudentT", + "Poisson", + "Uniform", + "VonMises", + "Weibull", + "Wishart", + "TransformedDistribution", + "biject_to", + "kl_divergence", + "register_kl", + "transform_to", +] +__all__.extend(transforms.__all__) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7d5a6d92757d1f3059cfbc1374a6937d3a048c8b Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/bernoulli.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/bernoulli.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d18b194b0faa97533f1d2c344f658fda497a0e80 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/bernoulli.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/beta.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/beta.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d4375b4bdec060b5ddbc215bdd5e366d4ce528b3 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/beta.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/binomial.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/binomial.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..593dc1523912b8117bb188db0b1fe7d3c4fe03ac Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/binomial.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/categorical.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/categorical.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ce29442cac95523e9642c0ba2776d6a98ce74914 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/categorical.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/cauchy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/cauchy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..71f8bccb8a1c9ccac10a5e2b4e781ca7684a7132 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/cauchy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/chi2.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/chi2.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..90c95d2e54cda4a4c65e8290fd7179a01262ecdf Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/chi2.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraint_registry.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraint_registry.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1f3edec9d752e39026acd5a569256c7ba9c98294 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraint_registry.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraints.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraints.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..43f891be4a0e51054a6692e91cf0518edbd54e1d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/constraints.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/continuous_bernoulli.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/continuous_bernoulli.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c7541f61db0fb75e58847bcab57bdc5e37b044b9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/continuous_bernoulli.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/dirichlet.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/dirichlet.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..745b7f5133195f7bb764f6995fea2c155f1a8503 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/dirichlet.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/distribution.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/distribution.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..87c6e97951fc74da5939f7369cea1542340c0991 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/distribution.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exp_family.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exp_family.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8c8e32703be5a7d6f1c9c0a224d19e2020efab3a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exp_family.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exponential.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exponential.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..145359a1f0933ce33dbd6fc89acf4804041ab9ec Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/exponential.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/fishersnedecor.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/fishersnedecor.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d8f63ea028481b8a89546c76214162f61fcae1da Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/fishersnedecor.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gamma.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gamma.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2707790610a1843a8e9aabf6b5de7c8e371555eb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gamma.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/generalized_pareto.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/generalized_pareto.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2f639f491135ad0a8519d9ac5c49febbacab3df8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/generalized_pareto.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/geometric.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/geometric.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4d63218176617deef0f21f63121033dcb49c995a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/geometric.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gumbel.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gumbel.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..56d67dea495653fb7ecdb4809e8aa1cd476039f9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/gumbel.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_cauchy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_cauchy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d8ada1118c17f1bd3c60ea902fd27a49ce560a40 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_cauchy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f9a276f37b3f9fc9a728cb26722cf4575332c142 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/half_normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/independent.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/independent.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1199a718655b32b2f42caff5888d0057c192dacb Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/independent.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/inverse_gamma.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/inverse_gamma.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f7e9ba3343aa7fd22b47faef945c59a898316292 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/inverse_gamma.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kl.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kl.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..137975ef56a1f4974e2fe1119e453a34a5d7861f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kl.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kumaraswamy.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kumaraswamy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0dd24dbc2806af6cbe41380f3a7929d2f9a1aaa9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/kumaraswamy.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/laplace.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/laplace.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ad3e728726f757c6498d0b26aa9eaccb142a678e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/laplace.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lkj_cholesky.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lkj_cholesky.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d8455dc8e3173ca305e7b52dd59e0f31f189132d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lkj_cholesky.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/log_normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/log_normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..50a1fc6afa71d43a1a3fa121b4184bd23d8ea26a Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/log_normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/logistic_normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/logistic_normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..05a7f9c87e335508ec354a602876f5b4443605f6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/logistic_normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lowrank_multivariate_normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lowrank_multivariate_normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..258cb50aa9d97ec0dcba6656c7efafd46aed910f Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/lowrank_multivariate_normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/mixture_same_family.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/mixture_same_family.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c8cfbc3d165e987d4a2ac2be9f118cd7eed47f22 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/mixture_same_family.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multinomial.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multinomial.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e18c4b02f4cf17b8f6801c072413e070d20f8a68 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multinomial.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multivariate_normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multivariate_normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ffa7675826704f3767226a61192be8f5f1e8a111 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/multivariate_normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/negative_binomial.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/negative_binomial.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1d124e6936aa16991822116a8c314fb900214838 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/negative_binomial.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/normal.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/normal.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..37b99aaae317c9467addc635b14992c0ca0099e6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/normal.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/one_hot_categorical.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/one_hot_categorical.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f91561042914a9d77d8098547d953de7b9fdb417 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/one_hot_categorical.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/pareto.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/pareto.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a940175adc5b73fd4239f3ffc6d22c2cf13fc6b2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/pareto.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/poisson.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/poisson.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4cce4cb6a07669c3fd49d5edae78ccd439f3c435 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/poisson.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_bernoulli.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_bernoulli.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..15d28a9ada3136649f24bbf1967f03157028fe93 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_bernoulli.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_categorical.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_categorical.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..f9998f1ddfe483d4385dd4535b2755c97d151217 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/relaxed_categorical.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/studentT.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/studentT.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2ff2e00f6b03376311bb281ac05c3d7db9c8e486 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/studentT.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transformed_distribution.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transformed_distribution.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..27c6f8160cf4ca5d514fb757fa8b996adc47f9df Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transformed_distribution.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transforms.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transforms.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..949c1efb5361d2fe650a3a4277423ddb85b4d59d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/transforms.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/uniform.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/uniform.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..fef9a34fa1b4fbebb440a3f265b59d7617d326da Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/uniform.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7d3dce7fccfa61851be98a000cfb899cd43d795d Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/von_mises.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/von_mises.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..135219e715ac486d65fda0719df735bb3e6cd2b6 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/von_mises.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/weibull.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/weibull.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ebc20c29b5aa71a98580a732bce9e2aa6da712e8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/weibull.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/wishart.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/wishart.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7db3b2b759ed908b40694363fd0c41fee2aa0553 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/distributions/__pycache__/wishart.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/bernoulli.py b/.venv/lib/python3.12/site-packages/torch/distributions/bernoulli.py new file mode 100644 index 0000000000000000000000000000000000000000..e5254eadba50faa3d3b869ceccb43b08bffeda87 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/bernoulli.py @@ -0,0 +1,145 @@ +# mypy: allow-untyped-defs + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import ( + broadcast_all, + lazy_property, + logits_to_probs, + probs_to_logits, +) +from torch.nn.functional import binary_cross_entropy_with_logits +from torch.types import _Number, Number + + +__all__ = ["Bernoulli"] + + +class Bernoulli(ExponentialFamily): + r""" + Creates a Bernoulli distribution parameterized by :attr:`probs` + or :attr:`logits` (but not both). + + Samples are binary (0 or 1). They take the value `1` with probability `p` + and `0` with probability `1 - p`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Bernoulli(torch.tensor([0.3])) + >>> m.sample() # 30% chance 1; 70% chance 0 + tensor([ 0.]) + + Args: + probs (Number, Tensor): the probability of sampling `1` + logits (Number, Tensor): the log-odds of sampling `1` + validate_args (bool, optional): whether to validate arguments, None by default + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.unit_interval, "logits": constraints.real} + support = constraints.boolean + has_enumerate_support = True + _mean_carrier_measure = 0 + + def __init__( + self, + probs: Tensor | Number | None = None, + logits: Tensor | Number | None = None, + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + is_scalar = isinstance(probs, _Number) + # pyrefly: ignore [read-only] + (self.probs,) = broadcast_all(probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + is_scalar = isinstance(logits, _Number) + # pyrefly: ignore [read-only] + (self.logits,) = broadcast_all(logits) + self._param = self.probs if probs is not None else self.logits + if is_scalar: + batch_shape = torch.Size() + else: + batch_shape = self._param.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Bernoulli, _instance) + batch_shape = torch.Size(batch_shape) + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + new._param = new.logits + super(Bernoulli, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + @property + def mean(self) -> Tensor: + return self.probs + + @property + def mode(self) -> Tensor: + mode = (self.probs >= 0.5).to(self.probs) + mode[self.probs == 0.5] = nan + return mode + + @property + def variance(self) -> Tensor: + return self.probs * (1 - self.probs) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits, is_binary=True) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + with torch.no_grad(): + return torch.bernoulli(self.probs.expand(shape)) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + logits, value = broadcast_all(self.logits, value) + return -binary_cross_entropy_with_logits(logits, value, reduction="none") + + def entropy(self): + return binary_cross_entropy_with_logits( + self.logits, self.probs, reduction="none" + ) + + def enumerate_support(self, expand=True): + values = torch.arange(2, dtype=self._param.dtype, device=self._param.device) + values = values.view((-1,) + (1,) * len(self._batch_shape)) + if expand: + values = values.expand((-1,) + self._batch_shape) + return values + + @property + def _natural_params(self) -> tuple[Tensor]: + return (torch.logit(self.probs),) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x): + return torch.log1p(torch.exp(x)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/beta.py b/.venv/lib/python3.12/site-packages/torch/distributions/beta.py new file mode 100644 index 0000000000000000000000000000000000000000..a69009f7ae43d0fe82f09367dcc9261ca5e03491 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/beta.py @@ -0,0 +1,118 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.dirichlet import Dirichlet +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Beta"] + + +class Beta(ExponentialFamily): + r""" + Beta distribution parameterized by :attr:`concentration1` and :attr:`concentration0`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Beta(torch.tensor([0.5]), torch.tensor([0.5])) + >>> m.sample() # Beta distributed with concentration concentration1 and concentration0 + tensor([ 0.1046]) + + Args: + concentration1 (float or Tensor): 1st concentration parameter of the distribution + (often referred to as alpha) + concentration0 (float or Tensor): 2nd concentration parameter of the distribution + (often referred to as beta) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "concentration1": constraints.positive, + "concentration0": constraints.positive, + } + support = constraints.unit_interval + has_rsample = True + + def __init__( + self, + concentration1: Tensor | float, + concentration0: Tensor | float, + validate_args: bool | None = None, + ) -> None: + if isinstance(concentration1, _Number) and isinstance(concentration0, _Number): + concentration1_concentration0 = torch.tensor( + [float(concentration1), float(concentration0)] + ) + else: + concentration1, concentration0 = broadcast_all( + concentration1, concentration0 + ) + concentration1_concentration0 = torch.stack( + [concentration1, concentration0], -1 + ) + self._dirichlet = Dirichlet( + concentration1_concentration0, validate_args=validate_args + ) + super().__init__(self._dirichlet._batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Beta, _instance) + batch_shape = torch.Size(batch_shape) + new._dirichlet = self._dirichlet.expand(batch_shape) + super(Beta, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + return self.concentration1 / (self.concentration1 + self.concentration0) + + @property + def mode(self) -> Tensor: + return self._dirichlet.mode[..., 0] + + @property + def variance(self) -> Tensor: + total = self.concentration1 + self.concentration0 + return self.concentration1 * self.concentration0 / (total.pow(2) * (total + 1)) + + def rsample(self, sample_shape: _size = ()) -> Tensor: + return self._dirichlet.rsample(sample_shape).select(-1, 0) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + heads_tails = torch.stack([value, 1.0 - value], -1) + return self._dirichlet.log_prob(heads_tails) + + def entropy(self): + return self._dirichlet.entropy() + + @property + def concentration1(self) -> Tensor: + result = self._dirichlet.concentration[..., 0] + if isinstance(result, _Number): + return torch.tensor([result]) + else: + return result + + @property + def concentration0(self) -> Tensor: + result = self._dirichlet.concentration[..., 1] + if isinstance(result, _Number): + return torch.tensor([result]) + else: + return result + + @property + def _natural_params(self) -> tuple[Tensor, Tensor]: + return (self.concentration1, self.concentration0) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x, y): + return torch.lgamma(x) + torch.lgamma(y) - torch.lgamma(x + y) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/binomial.py b/.venv/lib/python3.12/site-packages/torch/distributions/binomial.py new file mode 100644 index 0000000000000000000000000000000000000000..f9cd0a3ee1a5e3f24f727eac6e4fea0023f138cb --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/binomial.py @@ -0,0 +1,182 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import ( + broadcast_all, + lazy_property, + logits_to_probs, + probs_to_logits, +) + + +__all__ = ["Binomial"] + + +def _clamp_by_zero(x): + # works like clamp(x, min=0) but has grad at 0 is 0.5 + return (x.clamp(min=0) + x - x.clamp(max=0)) / 2 + + +class Binomial(Distribution): + r""" + Creates a Binomial distribution parameterized by :attr:`total_count` and + either :attr:`probs` or :attr:`logits` (but not both). :attr:`total_count` must be + broadcastable with :attr:`probs`/:attr:`logits`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Binomial(100, torch.tensor([0 , .2, .8, 1])) + >>> x = m.sample() + tensor([ 0., 22., 71., 100.]) + + >>> m = Binomial(torch.tensor([[5.], [10.]]), torch.tensor([0.5, 0.8])) + >>> x = m.sample() + tensor([[ 4., 5.], + [ 7., 6.]]) + + Args: + total_count (int or Tensor): number of Bernoulli trials + probs (Tensor): Event probabilities + logits (Tensor): Event log-odds + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "total_count": constraints.nonnegative_integer, + "probs": constraints.unit_interval, + "logits": constraints.real, + } + has_enumerate_support = True + + def __init__( + self, + total_count: Tensor | int = 1, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + ( + self.total_count, + # pyrefly: ignore [read-only] + self.probs, + ) = broadcast_all(total_count, probs) + self.total_count = self.total_count.type_as(self.probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + ( + self.total_count, + # pyrefly: ignore [read-only] + self.logits, + ) = broadcast_all(total_count, logits) + self.total_count = self.total_count.type_as(self.logits) + + self._param = self.probs if probs is not None else self.logits + batch_shape = self._param.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Binomial, _instance) + batch_shape = torch.Size(batch_shape) + new.total_count = self.total_count.expand(batch_shape) + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + new._param = new.logits + super(Binomial, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + @constraints.dependent_property(is_discrete=True, event_dim=0) + # pyrefly: ignore [bad-override] + def support(self): + return constraints.integer_interval(0, self.total_count) + + @property + def mean(self) -> Tensor: + return self.total_count * self.probs + + @property + def mode(self) -> Tensor: + return ((self.total_count + 1) * self.probs).floor().clamp(max=self.total_count) + + @property + def variance(self) -> Tensor: + return self.total_count * self.probs * (1 - self.probs) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits, is_binary=True) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + with torch.no_grad(): + return torch.binomial( + self.total_count.expand(shape), self.probs.expand(shape) + ) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + log_factorial_n = torch.lgamma(self.total_count + 1) + log_factorial_k = torch.lgamma(value + 1) + log_factorial_nmk = torch.lgamma(self.total_count - value + 1) + # k * log(p) + (n - k) * log(1 - p) = k * (log(p) - log(1 - p)) + n * log(1 - p) + # (case logit < 0) = k * logit - n * log1p(e^logit) + # (case logit > 0) = k * logit - n * (log(p) - log(1 - p)) + n * log(p) + # = k * logit - n * logit - n * log1p(e^-logit) + # (merge two cases) = k * logit - n * max(logit, 0) - n * log1p(e^-|logit|) + normalize_term = ( + self.total_count * _clamp_by_zero(self.logits) + + self.total_count * torch.log1p(torch.exp(-torch.abs(self.logits))) + - log_factorial_n + ) + return ( + value * self.logits - log_factorial_k - log_factorial_nmk - normalize_term + ) + + def entropy(self): + total_count = int(self.total_count.max()) + if not self.total_count.min() == total_count: + raise NotImplementedError( + "Inhomogeneous total count not supported by `entropy`." + ) + + log_prob = self.log_prob(self.enumerate_support(False)) + return -(torch.exp(log_prob) * log_prob).sum(0) + + def enumerate_support(self, expand=True): + total_count = int(self.total_count.max()) + if not self.total_count.min() == total_count: + raise NotImplementedError( + "Inhomogeneous total count not supported by `enumerate_support`." + ) + values = torch.arange( + 1 + total_count, dtype=self._param.dtype, device=self._param.device + ) + values = values.view((-1,) + (1,) * len(self._batch_shape)) + if expand: + values = values.expand((-1,) + self._batch_shape) + return values diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/categorical.py b/.venv/lib/python3.12/site-packages/torch/distributions/categorical.py new file mode 100644 index 0000000000000000000000000000000000000000..2c0fa35acbba5748fe7166c2396690e2cc168866 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/categorical.py @@ -0,0 +1,171 @@ +# mypy: allow-untyped-defs + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import lazy_property, logits_to_probs, probs_to_logits + + +__all__ = ["Categorical"] + + +class Categorical(Distribution): + r""" + Creates a categorical distribution parameterized by either :attr:`probs` or + :attr:`logits` (but not both). + + .. note:: + It is equivalent to the distribution that :func:`torch.multinomial` + samples from. + + Samples are integers from :math:`\{0, \ldots, K-1\}` where `K` is ``probs.size(-1)``. + + If `probs` is 1-dimensional with length-`K`, each element is the relative probability + of sampling the class at that index. + + If `probs` is N-dimensional, the first N-1 dimensions are treated as a batch of + relative probability vectors. + + .. note:: The `probs` argument must be non-negative, finite and have a non-zero sum, + and it will be normalized to sum to 1 along the last dimension. :attr:`probs` + will return this normalized value. + The `logits` argument will be interpreted as unnormalized log probabilities + and can therefore be any real number. It will likewise be normalized so that + the resulting probabilities sum to 1 along the last dimension. :attr:`logits` + will return this normalized value. + + See also: :func:`torch.multinomial` + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Categorical(torch.tensor([ 0.25, 0.25, 0.25, 0.25 ])) + >>> m.sample() # equal probability of 0, 1, 2, 3 + tensor(3) + + Args: + probs (Tensor): event probabilities + logits (Tensor): event log probabilities (unnormalized) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.simplex, "logits": constraints.real_vector} + has_enumerate_support = True + + def __init__( + self, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + if probs.dim() < 1: + raise ValueError("`probs` parameter must be at least one-dimensional.") + # pyrefly: ignore [read-only] + self.probs = probs / probs.sum(-1, keepdim=True) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + if logits.dim() < 1: + raise ValueError("`logits` parameter must be at least one-dimensional.") + # Normalize + # pyrefly: ignore [read-only] + self.logits = logits - logits.logsumexp(dim=-1, keepdim=True) + self._param = self.probs if probs is not None else self.logits + self._num_events = self._param.size()[-1] + batch_shape = ( + self._param.size()[:-1] if self._param.ndimension() > 1 else torch.Size() + ) + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Categorical, _instance) + batch_shape = torch.Size(batch_shape) + param_shape = batch_shape + torch.Size((self._num_events,)) + if "probs" in self.__dict__: + new.probs = self.probs.expand(param_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(param_shape) + new._param = new.logits + new._num_events = self._num_events + super(Categorical, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + @constraints.dependent_property(is_discrete=True, event_dim=0) + # pyrefly: ignore [bad-override] + def support(self): + return constraints.integer_interval(0, self._num_events - 1) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + @property + def mean(self) -> Tensor: + return torch.full( + self._extended_shape(), + nan, + dtype=self.probs.dtype, + device=self.probs.device, + ) + + @property + def mode(self) -> Tensor: + return self.probs.argmax(dim=-1) + + @property + def variance(self) -> Tensor: + return torch.full( + self._extended_shape(), + nan, + dtype=self.probs.dtype, + device=self.probs.device, + ) + + def sample(self, sample_shape=torch.Size()): + if not isinstance(sample_shape, torch.Size): + sample_shape = torch.Size(sample_shape) + probs_2d = self.probs.reshape(-1, self._num_events) + samples_2d = torch.multinomial(probs_2d, sample_shape.numel(), True).T + return samples_2d.reshape(self._extended_shape(sample_shape)) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + value = value.long().unsqueeze(-1) + value, log_pmf = torch.broadcast_tensors(value, self.logits) + value = value[..., :1] + return log_pmf.gather(-1, value).squeeze(-1) + + def entropy(self): + min_real = torch.finfo(self.logits.dtype).min + logits = torch.clamp(self.logits, min=min_real) + p_log_p = logits * self.probs + return -p_log_p.sum(-1) + + def enumerate_support(self, expand=True): + num_events = self._num_events + values = torch.arange(num_events, dtype=torch.long, device=self._param.device) + values = values.view((-1,) + (1,) * len(self._batch_shape)) + if expand: + values = values.expand((-1,) + self._batch_shape) + return values diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/cauchy.py b/.venv/lib/python3.12/site-packages/torch/distributions/cauchy.py new file mode 100644 index 0000000000000000000000000000000000000000..222a264fc5987fd54520500f3ec476c0c8851d99 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/cauchy.py @@ -0,0 +1,99 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import inf, nan, Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Cauchy"] + + +class Cauchy(Distribution): + r""" + Samples from a Cauchy (Lorentz) distribution. The distribution of the ratio of + independent normally distributed random variables with means `0` follows a + Cauchy distribution. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Cauchy(torch.tensor([0.0]), torch.tensor([1.0])) + >>> m.sample() # sample from a Cauchy distribution with loc=0 and scale=1 + tensor([ 2.3214]) + + Args: + loc (float or Tensor): mode or median of the distribution. + scale (float or Tensor): half width at half maximum. + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + support = constraints.real + has_rsample = True + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.loc, self.scale = broadcast_all(loc, scale) + if isinstance(loc, _Number) and isinstance(scale, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.loc.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Cauchy, _instance) + batch_shape = torch.Size(batch_shape) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + super(Cauchy, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + return torch.full( + self._extended_shape(), nan, dtype=self.loc.dtype, device=self.loc.device + ) + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def variance(self) -> Tensor: + return torch.full( + self._extended_shape(), inf, dtype=self.loc.dtype, device=self.loc.device + ) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + eps = self.loc.new(shape).cauchy_() + return self.loc + eps * self.scale + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + return ( + -math.log(math.pi) + - self.scale.log() + - (((value - self.loc) / self.scale) ** 2).log1p() + ) + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return torch.atan((value - self.loc) / self.scale) / math.pi + 0.5 + + def icdf(self, value): + return torch.tan(math.pi * (value - 0.5)) * self.scale + self.loc + + def entropy(self): + return math.log(4 * math.pi) + self.scale.log() diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/chi2.py b/.venv/lib/python3.12/site-packages/torch/distributions/chi2.py new file mode 100644 index 0000000000000000000000000000000000000000..28579499f38d12936c6d2258a5340a936cb31a86 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/chi2.py @@ -0,0 +1,42 @@ +# mypy: allow-untyped-defs + +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.gamma import Gamma + + +__all__ = ["Chi2"] + + +class Chi2(Gamma): + r""" + Creates a Chi-squared distribution parameterized by shape parameter :attr:`df`. + This is exactly equivalent to ``Gamma(alpha=0.5*df, beta=0.5)`` + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Chi2(torch.tensor([1.0])) + >>> m.sample() # Chi2 distributed with shape df=1 + tensor([ 0.1046]) + + Args: + df (float or Tensor): shape parameter of the distribution + """ + + arg_constraints = {"df": constraints.positive} + + def __init__( + self, + df: Tensor | float, + validate_args: bool | None = None, + ) -> None: + super().__init__(0.5 * df, 0.5, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Chi2, _instance) + return super().expand(batch_shape, new) + + @property + def df(self) -> Tensor: + return self.concentration * 2 diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/constraint_registry.py b/.venv/lib/python3.12/site-packages/torch/distributions/constraint_registry.py new file mode 100644 index 0000000000000000000000000000000000000000..8907e5b467abf400f806e70197f70f526b93b5f7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/constraint_registry.py @@ -0,0 +1,291 @@ +# mypy: allow-untyped-defs +r""" +PyTorch provides two global :class:`ConstraintRegistry` objects that link +:class:`~torch.distributions.constraints.Constraint` objects to +:class:`~torch.distributions.transforms.Transform` objects. These objects both +input constraints and return transforms, but they have different guarantees on +bijectivity. + +1. ``biject_to(constraint)`` looks up a bijective + :class:`~torch.distributions.transforms.Transform` from ``constraints.real`` + to the given ``constraint``. The returned transform is guaranteed to have + ``.bijective = True`` and should implement ``.log_abs_det_jacobian()``. +2. ``transform_to(constraint)`` looks up a not-necessarily bijective + :class:`~torch.distributions.transforms.Transform` from ``constraints.real`` + to the given ``constraint``. The returned transform is not guaranteed to + implement ``.log_abs_det_jacobian()``. + +The ``transform_to()`` registry is useful for performing unconstrained +optimization on constrained parameters of probability distributions, which are +indicated by each distribution's ``.arg_constraints`` dict. These transforms often +overparameterize a space in order to avoid rotation; they are thus more +suitable for coordinate-wise optimization algorithms like Adam:: + + loc = torch.zeros(100, requires_grad=True) + unconstrained = torch.zeros(100, requires_grad=True) + scale = transform_to(Normal.arg_constraints["scale"])(unconstrained) + loss = -Normal(loc, scale).log_prob(data).sum() + +The ``biject_to()`` registry is useful for Hamiltonian Monte Carlo, where +samples from a probability distribution with constrained ``.support`` are +propagated in an unconstrained space, and algorithms are typically rotation +invariant.:: + + dist = Exponential(rate) + unconstrained = torch.zeros(100, requires_grad=True) + sample = biject_to(dist.support)(unconstrained) + potential_energy = -dist.log_prob(sample).sum() + +.. note:: + + An example where ``transform_to`` and ``biject_to`` differ is + ``constraints.simplex``: ``transform_to(constraints.simplex)`` returns a + :class:`~torch.distributions.transforms.SoftmaxTransform` that simply + exponentiates and normalizes its inputs; this is a cheap and mostly + coordinate-wise operation appropriate for algorithms like SVI. In + contrast, ``biject_to(constraints.simplex)`` returns a + :class:`~torch.distributions.transforms.StickBreakingTransform` that + bijects its input down to a one-fewer-dimensional space; this a more + expensive less numerically stable transform but is needed for algorithms + like HMC. + +The ``biject_to`` and ``transform_to`` objects can be extended by user-defined +constraints and transforms using their ``.register()`` method either as a +function on singleton constraints:: + + transform_to.register(my_constraint, my_transform) + +or as a decorator on parameterized constraints:: + + @transform_to.register(MyConstraintClass) + def my_factory(constraint): + assert isinstance(constraint, MyConstraintClass) + return MyTransform(constraint.param1, constraint.param2) + +You can create your own registry by creating a new :class:`ConstraintRegistry` +object. +""" + +from torch.distributions import constraints, transforms +from torch.types import _Number + + +__all__ = [ + "ConstraintRegistry", + "biject_to", + "transform_to", +] + + +class ConstraintRegistry: + """ + Registry to link constraints to transforms. + """ + + def __init__(self): + self._registry = {} + super().__init__() + + def register(self, constraint, factory=None): + """ + Registers a :class:`~torch.distributions.constraints.Constraint` + subclass in this registry. Usage:: + + @my_registry.register(MyConstraintClass) + def construct_transform(constraint): + assert isinstance(constraint, MyConstraint) + return MyTransform(constraint.arg_constraints) + + Args: + constraint (subclass of :class:`~torch.distributions.constraints.Constraint`): + A subclass of :class:`~torch.distributions.constraints.Constraint`, or + a singleton object of the desired class. + factory (Callable): A callable that inputs a constraint object and returns + a :class:`~torch.distributions.transforms.Transform` object. + """ + # Support use as decorator. + if factory is None: + return lambda factory: self.register(constraint, factory) + + # Support calling on singleton instances. + if isinstance(constraint, constraints.Constraint): + constraint = type(constraint) + + if not isinstance(constraint, type) or not issubclass( + constraint, constraints.Constraint + ): + raise TypeError( + f"Expected constraint to be either a Constraint subclass or instance, but got {constraint}" + ) + + self._registry[constraint] = factory + return factory + + def __call__(self, constraint): + """ + Looks up a transform to constrained space, given a constraint object. + Usage:: + + constraint = Normal.arg_constraints["scale"] + scale = transform_to(constraint)(torch.zeros(1)) # constrained + u = transform_to(constraint).inv(scale) # unconstrained + + Args: + constraint (:class:`~torch.distributions.constraints.Constraint`): + A constraint object. + + Returns: + A :class:`~torch.distributions.transforms.Transform` object. + + Raises: + `NotImplementedError` if no transform has been registered. + """ + # Look up by Constraint subclass. + try: + factory = self._registry[type(constraint)] + except KeyError: + raise NotImplementedError( + f"Cannot transform {type(constraint).__name__} constraints" + ) from None + return factory(constraint) + + +biject_to = ConstraintRegistry() +transform_to = ConstraintRegistry() + + +################################################################################ +# Registration Table +################################################################################ + + +@biject_to.register(constraints.real) +@transform_to.register(constraints.real) +def _transform_to_real(constraint): + return transforms.identity_transform + + +@biject_to.register(constraints.independent) +def _biject_to_independent(constraint): + base_transform = biject_to(constraint.base_constraint) + return transforms.IndependentTransform( + base_transform, constraint.reinterpreted_batch_ndims + ) + + +@transform_to.register(constraints.independent) +def _transform_to_independent(constraint): + base_transform = transform_to(constraint.base_constraint) + return transforms.IndependentTransform( + base_transform, constraint.reinterpreted_batch_ndims + ) + + +@biject_to.register(constraints.positive) +@biject_to.register(constraints.nonnegative) +@transform_to.register(constraints.positive) +@transform_to.register(constraints.nonnegative) +def _transform_to_positive(constraint): + return transforms.ExpTransform() + + +@biject_to.register(constraints.greater_than) +@biject_to.register(constraints.greater_than_eq) +@transform_to.register(constraints.greater_than) +@transform_to.register(constraints.greater_than_eq) +def _transform_to_greater_than(constraint): + return transforms.ComposeTransform( + [ + transforms.ExpTransform(), + transforms.AffineTransform(constraint.lower_bound, 1), + ] + ) + + +@biject_to.register(constraints.less_than) +@transform_to.register(constraints.less_than) +def _transform_to_less_than(constraint): + return transforms.ComposeTransform( + [ + transforms.ExpTransform(), + transforms.AffineTransform(constraint.upper_bound, -1), + ] + ) + + +@biject_to.register(constraints.interval) +@biject_to.register(constraints.half_open_interval) +@transform_to.register(constraints.interval) +@transform_to.register(constraints.half_open_interval) +def _transform_to_interval(constraint): + # Handle the special case of the unit interval. + lower_is_0 = ( + isinstance(constraint.lower_bound, _Number) and constraint.lower_bound == 0 + ) + upper_is_1 = ( + isinstance(constraint.upper_bound, _Number) and constraint.upper_bound == 1 + ) + if lower_is_0 and upper_is_1: + return transforms.SigmoidTransform() + + loc = constraint.lower_bound + scale = constraint.upper_bound - constraint.lower_bound + return transforms.ComposeTransform( + [transforms.SigmoidTransform(), transforms.AffineTransform(loc, scale)] + ) + + +@biject_to.register(constraints.simplex) +def _biject_to_simplex(constraint): + return transforms.StickBreakingTransform() + + +@transform_to.register(constraints.simplex) +def _transform_to_simplex(constraint): + return transforms.SoftmaxTransform() + + +# TODO define a bijection for LowerCholeskyTransform +@transform_to.register(constraints.lower_cholesky) +def _transform_to_lower_cholesky(constraint): + return transforms.LowerCholeskyTransform() + + +@transform_to.register(constraints.positive_definite) +@transform_to.register(constraints.positive_semidefinite) +def _transform_to_positive_definite(constraint): + return transforms.PositiveDefiniteTransform() + + +@biject_to.register(constraints.corr_cholesky) +@transform_to.register(constraints.corr_cholesky) +def _transform_to_corr_cholesky(constraint): + return transforms.CorrCholeskyTransform() + + +@biject_to.register(constraints.cat) +def _biject_to_cat(constraint): + return transforms.CatTransform( + [biject_to(c) for c in constraint.cseq], constraint.dim, constraint.lengths + ) + + +@transform_to.register(constraints.cat) +def _transform_to_cat(constraint): + return transforms.CatTransform( + [transform_to(c) for c in constraint.cseq], constraint.dim, constraint.lengths + ) + + +@biject_to.register(constraints.stack) +def _biject_to_stack(constraint): + return transforms.StackTransform( + [biject_to(c) for c in constraint.cseq], constraint.dim + ) + + +@transform_to.register(constraints.stack) +def _transform_to_stack(constraint): + return transforms.StackTransform( + [transform_to(c) for c in constraint.cseq], constraint.dim + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/constraints.py b/.venv/lib/python3.12/site-packages/torch/distributions/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..162c41b94daaf1b51506cad3285abafeb4851d25 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/constraints.py @@ -0,0 +1,761 @@ +# mypy: allow-untyped-defs + +from collections.abc import Callable +from typing import Any + + +r""" +The following constraints are implemented: + +- ``constraints.boolean`` +- ``constraints.cat`` +- ``constraints.corr_cholesky`` +- ``constraints.dependent`` +- ``constraints.greater_than(lower_bound)`` +- ``constraints.greater_than_eq(lower_bound)`` +- ``constraints.independent(constraint, reinterpreted_batch_ndims)`` +- ``constraints.integer_interval(lower_bound, upper_bound)`` +- ``constraints.interval(lower_bound, upper_bound)`` +- ``constraints.less_than(upper_bound)`` +- ``constraints.lower_cholesky`` +- ``constraints.lower_triangular`` +- ``constraints.MixtureSameFamilyConstraint(base_constraint)`` +- ``constraints.multinomial`` +- ``constraints.nonnegative`` +- ``constraints.nonnegative_integer`` +- ``constraints.one_hot`` +- ``constraints.positive_integer`` +- ``constraints.positive`` +- ``constraints.positive_semidefinite`` +- ``constraints.positive_definite`` +- ``constraints.real_vector`` +- ``constraints.real`` +- ``constraints.simplex`` +- ``constraints.symmetric`` +- ``constraints.stack`` +- ``constraints.square`` +- ``constraints.symmetric`` +- ``constraints.unit_interval`` +""" + +import torch + + +__all__ = [ + "Constraint", + "boolean", + "cat", + "corr_cholesky", + "dependent", + "dependent_property", + "greater_than", + "greater_than_eq", + "independent", + "integer_interval", + "interval", + "half_open_interval", + "is_dependent", + "less_than", + "lower_cholesky", + "lower_triangular", + "MixtureSameFamilyConstraint", + "multinomial", + "nonnegative", + "nonnegative_integer", + "one_hot", + "positive", + "positive_semidefinite", + "positive_definite", + "positive_integer", + "real", + "real_vector", + "simplex", + "square", + "stack", + "symmetric", + "unit_interval", +] + + +class Constraint: + """ + Abstract base class for constraints. + + A constraint object represents a region over which a variable is valid, + e.g. within which a variable can be optimized. + + Attributes: + is_discrete (bool): Whether constrained space is discrete. + Defaults to False. + event_dim (int): Number of rightmost dimensions that together define + an event. The :meth:`check` method will remove this many dimensions + when computing validity. + """ + + is_discrete = False # Default to continuous. + event_dim = 0 # Default to univariate. + + def check(self, value): + """ + Returns a byte tensor of ``sample_shape + batch_shape`` indicating + whether each event in value satisfies this constraint. + """ + raise NotImplementedError + + def __repr__(self): + return self.__class__.__name__[1:] + "()" + + +class _Dependent(Constraint): + """ + Placeholder for variables whose support depends on other variables. + These variables obey no simple coordinate-wise constraints. + + Args: + is_discrete (bool): Optional value of ``.is_discrete`` in case this + can be computed statically. If not provided, access to the + ``.is_discrete`` attribute will raise a NotImplementedError. + event_dim (int): Optional value of ``.event_dim`` in case this + can be computed statically. If not provided, access to the + ``.event_dim`` attribute will raise a NotImplementedError. + """ + + def __init__(self, *, is_discrete=NotImplemented, event_dim=NotImplemented): + self._is_discrete = is_discrete + self._event_dim = event_dim + super().__init__() + + @property + def is_discrete(self) -> bool: # type: ignore[override] + if self._is_discrete is NotImplemented: + raise NotImplementedError(".is_discrete cannot be determined statically") + return self._is_discrete + + @property + def event_dim(self) -> int: # type: ignore[override] + if self._event_dim is NotImplemented: + raise NotImplementedError(".event_dim cannot be determined statically") + return self._event_dim + + def __call__(self, *, is_discrete=NotImplemented, event_dim=NotImplemented): + """ + Support for syntax to customize static attributes:: + + constraints.dependent(is_discrete=True, event_dim=1) + """ + if is_discrete is NotImplemented: + is_discrete = self._is_discrete + if event_dim is NotImplemented: + event_dim = self._event_dim + return _Dependent(is_discrete=is_discrete, event_dim=event_dim) + + def check(self, x): + raise ValueError("Cannot determine validity of dependent constraint") + + +def is_dependent(constraint): + """ + Checks if ``constraint`` is a ``_Dependent`` object. + + Args: + constraint : A ``Constraint`` object. + + Returns: + ``bool``: True if ``constraint`` can be refined to the type ``_Dependent``, False otherwise. + + Examples: + >>> import torch + >>> from torch.distributions import Bernoulli + >>> from torch.distributions.constraints import is_dependent + + >>> dist = Bernoulli(probs=torch.tensor([0.6], requires_grad=True)) + >>> constraint1 = dist.arg_constraints["probs"] + >>> constraint2 = dist.arg_constraints["logits"] + + >>> for constraint in [constraint1, constraint2]: + >>> if is_dependent(constraint): + >>> continue + """ + return isinstance(constraint, _Dependent) + + +class _DependentProperty(property, _Dependent): + """ + Decorator that extends @property to act like a `Dependent` constraint when + called on a class and act like a property when called on an object. + + Example:: + + class Uniform(Distribution): + def __init__(self, low, high): + self.low = low + self.high = high + + @constraints.dependent_property(is_discrete=False, event_dim=0) + def support(self): + return constraints.interval(self.low, self.high) + + Args: + fn (Callable): The function to be decorated. + is_discrete (bool): Optional value of ``.is_discrete`` in case this + can be computed statically. If not provided, access to the + ``.is_discrete`` attribute will raise a NotImplementedError. + event_dim (int): Optional value of ``.event_dim`` in case this + can be computed statically. If not provided, access to the + ``.event_dim`` attribute will raise a NotImplementedError. + """ + + def __init__( + self, + fn: Callable[..., Any] | None = None, + *, + is_discrete: bool | None = NotImplemented, + event_dim: int | None = NotImplemented, + ) -> None: + super().__init__(fn) + self._is_discrete = is_discrete + self._event_dim = event_dim + + def __call__(self, fn: Callable[..., Any]) -> "_DependentProperty": # type: ignore[override] + """ + Support for syntax to customize static attributes:: + + @constraints.dependent_property(is_discrete=True, event_dim=1) + def support(self): ... + """ + return _DependentProperty( + fn, is_discrete=self._is_discrete, event_dim=self._event_dim + ) + + +class _IndependentConstraint(Constraint): + """ + Wraps a constraint by aggregating over ``reinterpreted_batch_ndims``-many + dims in :meth:`check`, so that an event is valid only if all its + independent entries are valid. + """ + + def __init__(self, base_constraint, reinterpreted_batch_ndims): + if not isinstance(base_constraint, Constraint): + raise AssertionError( + f"base_constraint must be a Constraint, got {type(base_constraint).__name__}" + ) + if not isinstance(reinterpreted_batch_ndims, int): + raise AssertionError( + f"reinterpreted_batch_ndims must be an int, got {type(reinterpreted_batch_ndims).__name__}" + ) + if reinterpreted_batch_ndims < 0: + raise AssertionError( + f"reinterpreted_batch_ndims must be >= 0, got {reinterpreted_batch_ndims}" + ) + self.base_constraint = base_constraint + self.reinterpreted_batch_ndims = reinterpreted_batch_ndims + super().__init__() + + @property + def is_discrete(self) -> bool: # type: ignore[override] + return self.base_constraint.is_discrete + + @property + def event_dim(self) -> int: # type: ignore[override] + return self.base_constraint.event_dim + self.reinterpreted_batch_ndims + + def check(self, value): + result = self.base_constraint.check(value) + if result.dim() < self.reinterpreted_batch_ndims: + expected = self.base_constraint.event_dim + self.reinterpreted_batch_ndims + raise ValueError( + f"Expected value.dim() >= {expected} but got {value.dim()}" + ) + result = result.reshape( + result.shape[: result.dim() - self.reinterpreted_batch_ndims] + (-1,) + ) + result = result.all(-1) + return result + + def __repr__(self): + return f"{self.__class__.__name__[1:]}({repr(self.base_constraint)}, {self.reinterpreted_batch_ndims})" + + +class MixtureSameFamilyConstraint(Constraint): + """ + Constraint for the :class:`~torch.distributions.MixtureSameFamily` + distribution that adds back the rightmost batch dimension before + performing the validity check with the component distribution + constraint. + + Args: + base_constraint: The ``Constraint`` object of + the component distribution of + the :class:`~torch.distributions.MixtureSameFamily` distribution. + """ + + def __init__(self, base_constraint): + if not isinstance(base_constraint, Constraint): + raise AssertionError( + f"base_constraint must be a Constraint, got {type(base_constraint).__name__}" + ) + self.base_constraint = base_constraint + super().__init__() + + @property + def is_discrete(self) -> bool: # type: ignore[override] + return self.base_constraint.is_discrete + + @property + def event_dim(self) -> int: # type: ignore[override] + return self.base_constraint.event_dim + + def check(self, value): + """ + Check validity of ``value`` as a possible outcome of sampling + the :class:`~torch.distributions.MixtureSameFamily` distribution. + """ + unsqueezed_value = value.unsqueeze(-1 - self.event_dim) + result = self.base_constraint.check(unsqueezed_value) + if value.dim() < self.event_dim: + raise ValueError( + f"Expected value.dim() >= {self.event_dim} but got {value.dim()}" + ) + num_dim_to_keep = value.dim() - self.event_dim + result = result.reshape(result.shape[:num_dim_to_keep] + (-1,)) + result = result.all(-1) + return result + + def __repr__(self): + return f"{self.__class__.__name__}({repr(self.base_constraint)})" + + +class _Boolean(Constraint): + """ + Constrain to the two values `{0, 1}`. + """ + + is_discrete = True + + def check(self, value): + return (value == 0) | (value == 1) + + +class _OneHot(Constraint): + """ + Constrain to one-hot vectors. + """ + + is_discrete = True + event_dim = 1 + + def check(self, value): + is_boolean = (value == 0) | (value == 1) + is_normalized = value.sum(-1).eq(1) + return is_boolean.all(-1) & is_normalized + + +class _IntegerInterval(Constraint): + """ + Constrain to an integer interval `[lower_bound, upper_bound]`. + """ + + is_discrete = True + + def __init__(self, lower_bound, upper_bound): + self.lower_bound = lower_bound + self.upper_bound = upper_bound + super().__init__() + + def check(self, value): + return ( + (value % 1 == 0) & (self.lower_bound <= value) & (value <= self.upper_bound) + ) + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += ( + f"(lower_bound={self.lower_bound}, upper_bound={self.upper_bound})" + ) + return fmt_string + + +class _IntegerLessThan(Constraint): + """ + Constrain to an integer interval `(-inf, upper_bound]`. + """ + + is_discrete = True + + def __init__(self, upper_bound): + self.upper_bound = upper_bound + super().__init__() + + def check(self, value): + return (value % 1 == 0) & (value <= self.upper_bound) + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += f"(upper_bound={self.upper_bound})" + return fmt_string + + +class _IntegerGreaterThan(Constraint): + """ + Constrain to an integer interval `[lower_bound, inf)`. + """ + + is_discrete = True + + def __init__(self, lower_bound): + self.lower_bound = lower_bound + super().__init__() + + def check(self, value): + return (value % 1 == 0) & (value >= self.lower_bound) + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += f"(lower_bound={self.lower_bound})" + return fmt_string + + +class _Real(Constraint): + """ + Trivially constrain to the extended real line `[-inf, inf]`. + """ + + def check(self, value): + return value == value # False for NANs. + + +class _GreaterThan(Constraint): + """ + Constrain to a real half line `(lower_bound, inf]`. + """ + + def __init__(self, lower_bound): + self.lower_bound = lower_bound + super().__init__() + + def check(self, value): + return self.lower_bound < value + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += f"(lower_bound={self.lower_bound})" + return fmt_string + + +class _GreaterThanEq(Constraint): + """ + Constrain to a real half line `[lower_bound, inf)`. + """ + + def __init__(self, lower_bound): + self.lower_bound = lower_bound + super().__init__() + + def check(self, value): + return self.lower_bound <= value + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += f"(lower_bound={self.lower_bound})" + return fmt_string + + +class _LessThan(Constraint): + """ + Constrain to a real half line `[-inf, upper_bound)`. + """ + + def __init__(self, upper_bound): + self.upper_bound = upper_bound + super().__init__() + + def check(self, value): + return value < self.upper_bound + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += f"(upper_bound={self.upper_bound})" + return fmt_string + + +class _Interval(Constraint): + """ + Constrain to a real interval `[lower_bound, upper_bound]`. + """ + + def __init__(self, lower_bound, upper_bound): + self.lower_bound = lower_bound + self.upper_bound = upper_bound + super().__init__() + + def check(self, value): + return (self.lower_bound <= value) & (value <= self.upper_bound) + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += ( + f"(lower_bound={self.lower_bound}, upper_bound={self.upper_bound})" + ) + return fmt_string + + +class _HalfOpenInterval(Constraint): + """ + Constrain to a real interval `[lower_bound, upper_bound)`. + """ + + def __init__(self, lower_bound, upper_bound): + self.lower_bound = lower_bound + self.upper_bound = upper_bound + super().__init__() + + def check(self, value): + return (self.lower_bound <= value) & (value < self.upper_bound) + + def __repr__(self): + fmt_string = self.__class__.__name__[1:] + fmt_string += ( + f"(lower_bound={self.lower_bound}, upper_bound={self.upper_bound})" + ) + return fmt_string + + +class _Simplex(Constraint): + """ + Constrain to the unit simplex in the innermost (rightmost) dimension. + Specifically: `x >= 0` and `x.sum(-1) == 1`. + """ + + event_dim = 1 + + def check(self, value): + return torch.all(value >= 0, dim=-1) & ((value.sum(-1) - 1).abs() < 1e-6) + + +class _Multinomial(Constraint): + """ + Constrain to nonnegative integer values summing to at most an upper bound. + + Note due to limitations of the Multinomial distribution, this currently + checks the weaker condition ``value.sum(-1) <= upper_bound``. In the future + this may be strengthened to ``value.sum(-1) == upper_bound``. + """ + + is_discrete = True + event_dim = 1 + + def __init__(self, upper_bound): + self.upper_bound = upper_bound + + def check(self, x): + return (x >= 0).all(dim=-1) & (x.sum(dim=-1) <= self.upper_bound) + + +class _LowerTriangular(Constraint): + """ + Constrain to lower-triangular square matrices. + """ + + event_dim = 2 + + def check(self, value): + value_tril = value.tril() + return (value_tril == value).view(value.shape[:-2] + (-1,)).min(-1)[0] + + +class _LowerCholesky(Constraint): + """ + Constrain to lower-triangular square matrices with positive diagonals. + """ + + event_dim = 2 + + def check(self, value): + value_tril = value.tril() + lower_triangular = ( + (value_tril == value).view(value.shape[:-2] + (-1,)).min(-1)[0] + ) + + positive_diagonal = (value.diagonal(dim1=-2, dim2=-1) > 0).min(-1)[0] + return lower_triangular & positive_diagonal + + +class _CorrCholesky(Constraint): + """ + Constrain to lower-triangular square matrices with positive diagonals and each + row vector being of unit length. + """ + + event_dim = 2 + + def check(self, value): + tol = ( + torch.finfo(value.dtype).eps * value.size(-1) * 10 + ) # 10 is an adjustable fudge factor + row_norm = torch.linalg.norm(value.detach(), dim=-1) + unit_row_norm = (row_norm - 1.0).abs().le(tol).all(dim=-1) + return _LowerCholesky().check(value) & unit_row_norm + + +class _Square(Constraint): + """ + Constrain to square matrices. + """ + + event_dim = 2 + + def check(self, value): + return torch.full( + size=value.shape[:-2], + fill_value=(value.shape[-2] == value.shape[-1]), + dtype=torch.bool, + device=value.device, + ) + + +class _Symmetric(_Square): + """ + Constrain to Symmetric square matrices. + """ + + def check(self, value): + square_check = super().check(value) + if not square_check.all(): + return square_check + return torch.isclose(value, value.mT, atol=1e-6).all(-2).all(-1) + + +class _PositiveSemidefinite(_Symmetric): + """ + Constrain to positive-semidefinite matrices. + """ + + def check(self, value): + sym_check = super().check(value) + if not sym_check.all(): + return sym_check + return torch.linalg.eigvalsh(value).ge(0).all(-1) + + +class _PositiveDefinite(_Symmetric): + """ + Constrain to positive-definite matrices. + """ + + def check(self, value): + sym_check = super().check(value) + if not sym_check.all(): + return sym_check + return torch.linalg.cholesky_ex(value).info.eq(0) + + +class _Cat(Constraint): + """ + Constraint functor that applies a sequence of constraints + `cseq` at the submatrices at dimension `dim`, + each of size `lengths[dim]`, in a way compatible with :func:`torch.cat`. + """ + + def __init__(self, cseq, dim=0, lengths=None): + if not all(isinstance(c, Constraint) for c in cseq): + raise AssertionError("All elements of cseq must be Constraint instances") + self.cseq = list(cseq) + if lengths is None: + lengths = [1] * len(self.cseq) + self.lengths = list(lengths) + if len(self.lengths) != len(self.cseq): + raise AssertionError( + f"lengths ({len(self.lengths)}) must match cseq ({len(self.cseq)})" + ) + self.dim = dim + super().__init__() + + @property + def is_discrete(self) -> bool: # type: ignore[override] + return any(c.is_discrete for c in self.cseq) + + @property + def event_dim(self) -> int: # type: ignore[override] + return max(c.event_dim for c in self.cseq) + + def check(self, value): + if not (-value.dim() <= self.dim < value.dim()): + raise AssertionError( + f"dim {self.dim} out of range for value with {value.dim()} dimensions" + ) + checks = [] + start = 0 + for constr, length in zip(self.cseq, self.lengths): + v = value.narrow(self.dim, start, length) + checks.append(constr.check(v)) + start = start + length # avoid += for jit compat + return torch.cat(checks, self.dim) + + +class _Stack(Constraint): + """ + Constraint functor that applies a sequence of constraints + `cseq` at the submatrices at dimension `dim`, + in a way compatible with :func:`torch.stack`. + """ + + def __init__(self, cseq, dim=0): + if not all(isinstance(c, Constraint) for c in cseq): + raise AssertionError("All elements of cseq must be Constraint instances") + self.cseq = list(cseq) + self.dim = dim + super().__init__() + + @property + def is_discrete(self) -> bool: # type: ignore[override] + return any(c.is_discrete for c in self.cseq) + + @property + def event_dim(self) -> int: # type: ignore[override] + dim = max(c.event_dim for c in self.cseq) + if self.dim + dim < 0: + dim += 1 + return dim + + def check(self, value): + if not (-value.dim() <= self.dim < value.dim()): + raise AssertionError( + f"dim {self.dim} out of range for value with {value.dim()} dimensions" + ) + vs = [value.select(self.dim, i) for i in range(value.size(self.dim))] + return torch.stack( + [constr.check(v) for v, constr in zip(vs, self.cseq)], self.dim + ) + + +# Public interface. +dependent = _Dependent() +dependent_property = _DependentProperty +independent = _IndependentConstraint +boolean = _Boolean() +one_hot = _OneHot() +nonnegative_integer = _IntegerGreaterThan(0) +positive_integer = _IntegerGreaterThan(1) +integer_interval = _IntegerInterval +real = _Real() +real_vector = independent(real, 1) +positive = _GreaterThan(0.0) +nonnegative = _GreaterThanEq(0.0) +greater_than = _GreaterThan +greater_than_eq = _GreaterThanEq +less_than = _LessThan +multinomial = _Multinomial +unit_interval = _Interval(0.0, 1.0) +interval = _Interval +half_open_interval = _HalfOpenInterval +simplex = _Simplex() +lower_triangular = _LowerTriangular() +lower_cholesky = _LowerCholesky() +corr_cholesky = _CorrCholesky() +square = _Square() +symmetric = _Symmetric() +positive_semidefinite = _PositiveSemidefinite() +positive_definite = _PositiveDefinite() +cat = _Cat +stack = _Stack diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/continuous_bernoulli.py b/.venv/lib/python3.12/site-packages/torch/distributions/continuous_bernoulli.py new file mode 100644 index 0000000000000000000000000000000000000000..28828e1620be8e7cc77bab83972ee04915d6e70d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/continuous_bernoulli.py @@ -0,0 +1,250 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import ( + broadcast_all, + clamp_probs, + lazy_property, + logits_to_probs, + probs_to_logits, +) +from torch.nn.functional import binary_cross_entropy_with_logits +from torch.types import _Number, _size, Number + + +__all__ = ["ContinuousBernoulli"] + + +class ContinuousBernoulli(ExponentialFamily): + r""" + Creates a continuous Bernoulli distribution parameterized by :attr:`probs` + or :attr:`logits` (but not both). + + The distribution is supported in [0, 1] and parameterized by 'probs' (in + (0,1)) or 'logits' (real-valued). Note that, unlike the Bernoulli, 'probs' + does not correspond to a probability and 'logits' does not correspond to + log-odds, but the same names are used due to the similarity with the + Bernoulli. See [1] for more details. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = ContinuousBernoulli(torch.tensor([0.3])) + >>> m.sample() + tensor([ 0.2538]) + + Args: + probs (Number, Tensor): (0,1) valued parameters + logits (Number, Tensor): real valued parameters whose sigmoid matches 'probs' + + [1] The continuous Bernoulli: fixing a pervasive error in variational + autoencoders, Loaiza-Ganem G and Cunningham JP, NeurIPS 2019. + https://arxiv.org/abs/1907.06845 + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.unit_interval, "logits": constraints.real} + support = constraints.unit_interval + _mean_carrier_measure = 0 + has_rsample = True + + def __init__( + self, + probs: Tensor | Number | None = None, + logits: Tensor | Number | None = None, + lims: tuple[float, float] = (0.499, 0.501), + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + is_scalar = isinstance(probs, _Number) + # pyrefly: ignore [read-only] + (self.probs,) = broadcast_all(probs) + # validate 'probs' here if necessary as it is later clamped for numerical stability + # close to 0 and 1, later on; otherwise the clamped 'probs' would always pass + if validate_args is not None: + if not self.arg_constraints["probs"].check(self.probs).all(): + raise ValueError("The parameter probs has invalid values") + # pyrefly: ignore [read-only] + self.probs = clamp_probs(self.probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + is_scalar = isinstance(logits, _Number) + # pyrefly: ignore [read-only] + (self.logits,) = broadcast_all(logits) + self._param = self.probs if probs is not None else self.logits + if is_scalar: + batch_shape = torch.Size() + else: + batch_shape = self._param.size() + self._lims = lims + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(ContinuousBernoulli, _instance) + new._lims = self._lims + batch_shape = torch.Size(batch_shape) + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + new._param = new.logits + super(ContinuousBernoulli, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + def _outside_unstable_region(self): + return torch.max( + torch.le(self.probs, self._lims[0]), torch.gt(self.probs, self._lims[1]) + ) + + def _cut_probs(self): + return torch.where( + self._outside_unstable_region(), + self.probs, + self._lims[0] * torch.ones_like(self.probs), + ) + + def _cont_bern_log_norm(self): + """computes the log normalizing constant as a function of the 'probs' parameter""" + cut_probs = self._cut_probs() + cut_probs_below_half = torch.where( + torch.le(cut_probs, 0.5), cut_probs, torch.zeros_like(cut_probs) + ) + cut_probs_above_half = torch.where( + torch.ge(cut_probs, 0.5), cut_probs, torch.ones_like(cut_probs) + ) + log_norm = torch.log( + torch.abs(torch.log1p(-cut_probs) - torch.log(cut_probs)) + ) - torch.where( + torch.le(cut_probs, 0.5), + torch.log1p(-2.0 * cut_probs_below_half), + torch.log(2.0 * cut_probs_above_half - 1.0), + ) + x = torch.pow(self.probs - 0.5, 2) + taylor = math.log(2.0) + (4.0 / 3.0 + 104.0 / 45.0 * x) * x + return torch.where(self._outside_unstable_region(), log_norm, taylor) + + @property + def mean(self) -> Tensor: + cut_probs = self._cut_probs() + mus = cut_probs / (2.0 * cut_probs - 1.0) + 1.0 / ( + torch.log1p(-cut_probs) - torch.log(cut_probs) + ) + x = self.probs - 0.5 + taylor = 0.5 + (1.0 / 3.0 + 16.0 / 45.0 * torch.pow(x, 2)) * x + return torch.where(self._outside_unstable_region(), mus, taylor) + + @property + def stddev(self) -> Tensor: + return torch.sqrt(self.variance) + + @property + def variance(self) -> Tensor: + cut_probs = self._cut_probs() + vars = cut_probs * (cut_probs - 1.0) / torch.pow( + 1.0 - 2.0 * cut_probs, 2 + ) + 1.0 / torch.pow(torch.log1p(-cut_probs) - torch.log(cut_probs), 2) + x = torch.pow(self.probs - 0.5, 2) + taylor = 1.0 / 12.0 - (1.0 / 15.0 - 128.0 / 945.0 * x) * x + return torch.where(self._outside_unstable_region(), vars, taylor) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return clamp_probs(logits_to_probs(self.logits, is_binary=True)) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + u = torch.rand(shape, dtype=self.probs.dtype, device=self.probs.device) + with torch.no_grad(): + return self.icdf(u) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + u = torch.rand(shape, dtype=self.probs.dtype, device=self.probs.device) + return self.icdf(u) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + logits, value = broadcast_all(self.logits, value) + return ( + -binary_cross_entropy_with_logits(logits, value, reduction="none") + + self._cont_bern_log_norm() + ) + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + cut_probs = self._cut_probs() + cdfs = ( + torch.pow(cut_probs, value) * torch.pow(1.0 - cut_probs, 1.0 - value) + + cut_probs + - 1.0 + ) / (2.0 * cut_probs - 1.0) + unbounded_cdfs = torch.where(self._outside_unstable_region(), cdfs, value) + return torch.where( + torch.le(value, 0.0), + torch.zeros_like(value), + torch.where(torch.ge(value, 1.0), torch.ones_like(value), unbounded_cdfs), + ) + + def icdf(self, value): + cut_probs = self._cut_probs() + return torch.where( + self._outside_unstable_region(), + ( + torch.log1p(-cut_probs + value * (2.0 * cut_probs - 1.0)) + - torch.log1p(-cut_probs) + ) + / (torch.log(cut_probs) - torch.log1p(-cut_probs)), + value, + ) + + def entropy(self): + log_probs0 = torch.log1p(-self.probs) + log_probs1 = torch.log(self.probs) + return ( + self.mean * (log_probs0 - log_probs1) + - self._cont_bern_log_norm() + - log_probs0 + ) + + @property + def _natural_params(self) -> tuple[Tensor]: + return (self.logits,) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x): + """computes the log normalizing constant as a function of the natural parameter""" + out_unst_reg = torch.max( + torch.le(x, self._lims[0] - 0.5), torch.gt(x, self._lims[1] - 0.5) + ) + cut_nat_params = torch.where( + out_unst_reg, x, (self._lims[0] - 0.5) * torch.ones_like(x) + ) + log_norm = torch.log( + torch.abs(torch.special.expm1(cut_nat_params)) + ) - torch.log(torch.abs(cut_nat_params)) + taylor = 0.5 * x + torch.pow(x, 2) / 24.0 - torch.pow(x, 4) / 2880.0 + return torch.where(out_unst_reg, log_norm, taylor) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/dirichlet.py b/.venv/lib/python3.12/site-packages/torch/distributions/dirichlet.py new file mode 100644 index 0000000000000000000000000000000000000000..82011e88fea6ddc312ae5679ab7900b8c76d2daf --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/dirichlet.py @@ -0,0 +1,138 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.autograd import Function +from torch.autograd.function import once_differentiable +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.types import _size + + +__all__ = ["Dirichlet"] + + +# This helper is exposed for testing. +def _Dirichlet_backward(x, concentration, grad_output): + total = concentration.sum(-1, True).expand_as(concentration) + grad = torch._dirichlet_grad(x, concentration, total) + return grad * (grad_output - (x * grad_output).sum(-1, True)) + + +class _Dirichlet(Function): + @staticmethod + # pyrefly: ignore [bad-override] + def forward(ctx, concentration): + x = torch._sample_dirichlet(concentration) + ctx.save_for_backward(x, concentration) + return x + + @staticmethod + @once_differentiable + # pyrefly: ignore [bad-override] + def backward(ctx, grad_output): + x, concentration = ctx.saved_tensors + return _Dirichlet_backward(x, concentration, grad_output) + + +class Dirichlet(ExponentialFamily): + r""" + Creates a Dirichlet distribution parameterized by concentration :attr:`concentration`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Dirichlet(torch.tensor([0.5, 0.5])) + >>> m.sample() # Dirichlet distributed with concentration [0.5, 0.5] + tensor([ 0.1046, 0.8954]) + + Args: + concentration (Tensor): concentration parameter of the distribution + (often referred to as alpha) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "concentration": constraints.independent(constraints.positive, 1) + } + support = constraints.simplex + has_rsample = True + + def __init__( + self, + concentration: Tensor, + validate_args: bool | None = None, + ) -> None: + if concentration.dim() < 1: + raise ValueError( + "`concentration` parameter must be at least one-dimensional." + ) + self.concentration = concentration + batch_shape, event_shape = concentration.shape[:-1], concentration.shape[-1:] + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Dirichlet, _instance) + batch_shape = torch.Size(batch_shape) + new.concentration = self.concentration.expand(batch_shape + self.event_shape) + super(Dirichlet, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + def rsample(self, sample_shape: _size = ()) -> Tensor: + shape = self._extended_shape(sample_shape) + concentration = self.concentration.expand(shape) + return _Dirichlet.apply(concentration) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + return ( + torch.xlogy(self.concentration - 1.0, value).sum(-1) + + torch.lgamma(self.concentration.sum(-1)) + - torch.lgamma(self.concentration).sum(-1) + ) + + @property + def mean(self) -> Tensor: + return self.concentration / self.concentration.sum(-1, True) + + @property + def mode(self) -> Tensor: + concentrationm1 = (self.concentration - 1).clamp(min=0.0) + mode = concentrationm1 / concentrationm1.sum(-1, True) + mask = (self.concentration < 1).all(dim=-1) + mode[mask] = torch.nn.functional.one_hot( + mode[mask].argmax(dim=-1), concentrationm1.shape[-1] + ).to(mode) + return mode + + @property + def variance(self) -> Tensor: + con0 = self.concentration.sum(-1, True) + return ( + self.concentration + * (con0 - self.concentration) + / (con0.pow(2) * (con0 + 1)) + ) + + def entropy(self): + k = self.concentration.size(-1) + a0 = self.concentration.sum(-1) + return ( + torch.lgamma(self.concentration).sum(-1) + - torch.lgamma(a0) + - (k - a0) * torch.digamma(a0) + - ((self.concentration - 1.0) * torch.digamma(self.concentration)).sum(-1) + ) + + @property + def _natural_params(self) -> tuple[Tensor]: + return (self.concentration,) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x): + return x.lgamma().sum(-1) - torch.lgamma(x.sum(-1)) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/distribution.py b/.venv/lib/python3.12/site-packages/torch/distributions/distribution.py new file mode 100644 index 0000000000000000000000000000000000000000..f8115518f0f6c3fa2d58be6082087dad92881b0b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/distribution.py @@ -0,0 +1,348 @@ +# mypy: allow-untyped-defs +import warnings +from typing_extensions import deprecated + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.utils import lazy_property +from torch.types import _size + + +__all__ = ["Distribution"] + + +class Distribution: + r""" + Distribution is the abstract base class for probability distributions. + + Args: + batch_shape (torch.Size): The shape over which parameters are batched. + event_shape (torch.Size): The shape of a single sample (without batching). + validate_args (bool, optional): Whether to validate arguments. Default: None. + """ + + has_rsample = False + has_enumerate_support = False + _validate_args = __debug__ + + @staticmethod + def set_default_validate_args(value: bool) -> None: + """ + Sets whether validation is enabled or disabled. + + The default behavior mimics Python's ``assert`` statement: validation + is on by default, but is disabled if Python is run in optimized mode + (via ``python -O``). Validation may be expensive, so you may want to + disable it once a model is working. + + Args: + value (bool): Whether to enable validation. + """ + if value not in [True, False]: + raise ValueError + Distribution._validate_args = value + + def __init__( + self, + batch_shape: torch.Size = torch.Size(), + event_shape: torch.Size = torch.Size(), + validate_args: bool | None = None, + ) -> None: + self._batch_shape = batch_shape + self._event_shape = event_shape + if validate_args is not None: + self._validate_args = validate_args + if self._validate_args: + try: + arg_constraints = self.arg_constraints + except NotImplementedError: + arg_constraints = {} + warnings.warn( + f"{self.__class__} does not define `arg_constraints`. " + + "Please set `arg_constraints = {}` or initialize the distribution " + + "with `validate_args=False` to turn off validation.", + stacklevel=2, + ) + for param, constraint in arg_constraints.items(): + if constraints.is_dependent(constraint): + continue # skip constraints that cannot be checked + if param not in self.__dict__ and isinstance( + getattr(type(self), param), lazy_property + ): + continue # skip checking lazily-constructed args + value = getattr(self, param) + valid = constraint.check(value) + if not torch._is_all_true(valid): + raise ValueError( + f"Expected parameter {param} " + f"({type(value).__name__} of shape {tuple(value.shape)}) " + f"of distribution {repr(self)} " + f"to satisfy the constraint {repr(constraint)}, " + f"but found invalid values:\n{value}" + ) + super().__init__() + + def expand(self, batch_shape: _size, _instance=None): + """ + Returns a new distribution instance (or populates an existing instance + provided by a derived class) with batch dimensions expanded to + `batch_shape`. This method calls :class:`~torch.Tensor.expand` on + the distribution's parameters. As such, this does not allocate new + memory for the expanded distribution instance. Additionally, + this does not repeat any args checking or parameter broadcasting in + `__init__.py`, when an instance is first created. + + Args: + batch_shape (torch.Size): the desired expanded size. + _instance: new instance provided by subclasses that + need to override `.expand`. + + Returns: + New distribution instance with batch dimensions expanded to + `batch_shape`. + """ + raise NotImplementedError + + @property + def batch_shape(self) -> torch.Size: + """ + Returns the shape over which parameters are batched. + """ + return self._batch_shape + + @property + def event_shape(self) -> torch.Size: + """ + Returns the shape of a single sample (without batching). + """ + return self._event_shape + + @property + def arg_constraints(self) -> dict[str, constraints.Constraint]: + """ + Returns a dictionary from argument names to + :class:`~torch.distributions.constraints.Constraint` objects that + should be satisfied by each argument of this distribution. Args that + are not tensors need not appear in this dict. + """ + raise NotImplementedError + + @property + def support(self) -> constraints.Constraint | None: + """ + Returns a :class:`~torch.distributions.constraints.Constraint` object + representing this distribution's support. + """ + raise NotImplementedError + + @property + def mean(self) -> Tensor: + """ + Returns the mean of the distribution. + """ + raise NotImplementedError + + @property + def mode(self) -> Tensor: + """ + Returns the mode of the distribution. + """ + raise NotImplementedError(f"{self.__class__} does not implement mode") + + @property + def variance(self) -> Tensor: + """ + Returns the variance of the distribution. + """ + raise NotImplementedError + + @property + def stddev(self) -> Tensor: + """ + Returns the standard deviation of the distribution. + """ + return self.variance.sqrt() + + def sample(self, sample_shape: _size = torch.Size()) -> Tensor: + """ + Generates a sample_shape shaped sample or sample_shape shaped batch of + samples if the distribution parameters are batched. + """ + with torch.no_grad(): + return self.rsample(sample_shape) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + """ + Generates a sample_shape shaped reparameterized sample or sample_shape + shaped batch of reparameterized samples if the distribution parameters + are batched. + """ + raise NotImplementedError + + @deprecated( + "`sample_n(n)` will be deprecated. Use `sample((n,))` instead.", + category=FutureWarning, + ) + def sample_n(self, n: int) -> Tensor: + """ + Generates n samples or n batches of samples if the distribution + parameters are batched. + """ + return self.sample(torch.Size((n,))) + + def log_prob(self, value: Tensor) -> Tensor: + """ + Returns the log of the probability density/mass function evaluated at + `value`. + + Args: + value (Tensor): + """ + raise NotImplementedError + + def cdf(self, value: Tensor) -> Tensor: + """ + Returns the cumulative density/mass function evaluated at + `value`. + + Args: + value (Tensor): + """ + raise NotImplementedError + + def icdf(self, value: Tensor) -> Tensor: + """ + Returns the inverse cumulative density/mass function evaluated at + `value`. + + Args: + value (Tensor): + """ + raise NotImplementedError + + def enumerate_support(self, expand: bool = True) -> Tensor: + """ + Returns tensor containing all values supported by a discrete + distribution. The result will enumerate over dimension 0, so the shape + of the result will be `(cardinality,) + batch_shape + event_shape` + (where `event_shape = ()` for univariate distributions). + + Note that this enumerates over all batched tensors in lock-step + `[[0, 0], [1, 1], ...]`. With `expand=False`, enumeration happens + along dim 0, but with the remaining batch dimensions being + singleton dimensions, `[[0], [1], ..`. + + To iterate over the full Cartesian product use + `itertools.product(m.enumerate_support())`. + + Args: + expand (bool): whether to expand the support over the + batch dims to match the distribution's `batch_shape`. + + Returns: + Tensor iterating over dimension 0. + """ + raise NotImplementedError + + def entropy(self) -> Tensor: + """ + Returns entropy of distribution, batched over batch_shape. + + Returns: + Tensor of shape batch_shape. + """ + raise NotImplementedError + + def perplexity(self) -> Tensor: + """ + Returns perplexity of distribution, batched over batch_shape. + + Returns: + Tensor of shape batch_shape. + """ + return torch.exp(self.entropy()) + + def _extended_shape(self, sample_shape: _size = torch.Size()) -> torch.Size: + """ + Returns the size of the sample returned by the distribution, given + a `sample_shape`. Note, that the batch and event shapes of a distribution + instance are fixed at the time of construction. If this is empty, the + returned shape is upcast to (1,). + + Args: + sample_shape (torch.Size): the size of the sample to be drawn. + """ + if not isinstance(sample_shape, torch.Size): + sample_shape = torch.Size(sample_shape) + return torch.Size(sample_shape + self._batch_shape + self._event_shape) + + def _validate_sample(self, value: Tensor) -> None: + """ + Argument validation for distribution methods such as `log_prob`, + `cdf` and `icdf`. The rightmost dimensions of a value to be + scored via these methods must agree with the distribution's batch + and event shapes. + + Args: + value (Tensor): the tensor whose log probability is to be + computed by the `log_prob` method. + Raises + ValueError: when the rightmost dimensions of `value` do not match the + distribution's batch and event shapes. + """ + if not isinstance(value, torch.Tensor): + raise ValueError("The value argument to log_prob must be a Tensor") + + event_dim_start = len(value.size()) - len(self._event_shape) + if value.size()[event_dim_start:] != self._event_shape: + raise ValueError( + f"The right-most size of value must match event_shape: {value.size()} vs {self._event_shape}." + ) + + actual_shape = value.size() + expected_shape = self._batch_shape + self._event_shape + for i, j in zip(reversed(actual_shape), reversed(expected_shape)): + if i != 1 and j != 1 and i != j: + raise ValueError( + f"Value is not broadcastable with batch_shape+event_shape: {actual_shape} vs {expected_shape}." + ) + try: + support = self.support + except NotImplementedError: + warnings.warn( + f"{self.__class__} does not define `support` to enable " + + "sample validation. Please initialize the distribution with " + + "`validate_args=False` to turn off validation.", + stacklevel=2, + ) + return + if support is None: + raise AssertionError("support is unexpectedly None") + valid = support.check(value) + if not torch._is_all_true(valid): + raise ValueError( + "Expected value argument " + f"({type(value).__name__} of shape {tuple(value.shape)}) " + f"to be within the support ({repr(support)}) " + f"of the distribution {repr(self)}, " + f"but found invalid values:\n{value}" + ) + + def _get_checked_instance(self, cls, _instance=None): + if _instance is None and type(self).__init__ != cls.__init__: + raise NotImplementedError( + f"Subclass {self.__class__.__name__} of {cls.__name__} that defines a custom __init__ method " + "must also define a custom .expand() method." + ) + return self.__new__(type(self)) if _instance is None else _instance + + def __repr__(self) -> str: + param_names = [k for k in self.arg_constraints if k in self.__dict__] + args_string = ", ".join( + [ + f"{p}: {self.__dict__[p] if self.__dict__[p].numel() == 1 else self.__dict__[p].size()}" + for p in param_names + ] + ) + return self.__class__.__name__ + "(" + args_string + ")" diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/exp_family.py b/.venv/lib/python3.12/site-packages/torch/distributions/exp_family.py new file mode 100644 index 0000000000000000000000000000000000000000..a9de981091b1f67a5cc7672df2781668d548fcc3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/exp_family.py @@ -0,0 +1,66 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions.distribution import Distribution + + +__all__ = ["ExponentialFamily"] + + +class ExponentialFamily(Distribution): + r""" + ExponentialFamily is the abstract base class for probability distributions belonging to an + exponential family, whose probability mass/density function is defined below + + .. math:: + + p_{F}(x; \theta) = \exp(\langle t(x), \theta\rangle - F(\theta) + k(x)) + + where :math:`\theta` denotes the natural parameters, :math:`t(x)` denotes the sufficient statistic, + :math:`F(\theta)` is the log normalizer function for a given family and :math:`k(x)` is the carrier + measure. + + Note: + This class is an intermediary between the `Distribution` class and distributions which belong + to an exponential family mainly to check the correctness of the `.entropy()` and analytic KL + divergence methods. We use this class to compute the entropy and KL divergence using the AD + framework and Bregman divergences (courtesy of: Frank Nielsen and Richard Nock, Entropies and + Cross-entropies of Exponential Families). + """ + + @property + def _natural_params(self) -> tuple[Tensor, ...]: + """ + Abstract method for natural parameters. Returns a tuple of Tensors based + on the distribution + """ + raise NotImplementedError + + def _log_normalizer(self, *natural_params): + """ + Abstract method for log normalizer function. Returns a log normalizer based on + the distribution and input + """ + raise NotImplementedError + + @property + def _mean_carrier_measure(self) -> float: + """ + Abstract method for expected carrier measure, which is required for computing + entropy. + """ + raise NotImplementedError + + def entropy(self): + """ + Method to compute the entropy using Bregman divergence of the log normalizer. + """ + result: Tensor | float = -self._mean_carrier_measure + nparams = [p.detach().requires_grad_() for p in self._natural_params] + lg_normal = self._log_normalizer(*nparams) + gradients = torch.autograd.grad(lg_normal.sum(), nparams, create_graph=True) + result += lg_normal + for np, g in zip(nparams, gradients): + result -= (np * g).reshape(self._batch_shape + (-1,)).sum(-1) + return result diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/exponential.py b/.venv/lib/python3.12/site-packages/torch/distributions/exponential.py new file mode 100644 index 0000000000000000000000000000000000000000..0eadf705f16a51aa25540058375b9ed116b06ed4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/exponential.py @@ -0,0 +1,94 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Exponential"] + + +class Exponential(ExponentialFamily): + r""" + Creates a Exponential distribution parameterized by :attr:`rate`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Exponential(torch.tensor([1.0])) + >>> m.sample() # Exponential distributed with rate=1 + tensor([ 0.1046]) + + Args: + rate (float or Tensor): rate = 1 / scale of the distribution + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"rate": constraints.positive} + support = constraints.nonnegative + has_rsample = True + _mean_carrier_measure = 0 + + @property + def mean(self) -> Tensor: + return self.rate.reciprocal() + + @property + def mode(self) -> Tensor: + return torch.zeros_like(self.rate) + + @property + def stddev(self) -> Tensor: + return self.rate.reciprocal() + + @property + def variance(self) -> Tensor: + return self.rate.pow(-2) + + def __init__( + self, + rate: Tensor | float, + validate_args: bool | None = None, + ) -> None: + (self.rate,) = broadcast_all(rate) + batch_shape = torch.Size() if isinstance(rate, _Number) else self.rate.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Exponential, _instance) + batch_shape = torch.Size(batch_shape) + new.rate = self.rate.expand(batch_shape) + super(Exponential, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + return self.rate.new(shape).exponential_() / self.rate + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + return self.rate.log() - self.rate * value + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return 1 - torch.exp(-self.rate * value) + + def icdf(self, value): + return -torch.log1p(-value) / self.rate + + def entropy(self): + return 1.0 - torch.log(self.rate) + + @property + def _natural_params(self) -> tuple[Tensor]: + return (-self.rate,) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x): + return -torch.log(-x) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/fishersnedecor.py b/.venv/lib/python3.12/site-packages/torch/distributions/fishersnedecor.py new file mode 100644 index 0000000000000000000000000000000000000000..9c81ad3690eefc9c89ed252646a1741c6705613d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/fishersnedecor.py @@ -0,0 +1,107 @@ +# mypy: allow-untyped-defs + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.gamma import Gamma +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["FisherSnedecor"] + + +class FisherSnedecor(Distribution): + r""" + Creates a Fisher-Snedecor distribution parameterized by :attr:`df1` and :attr:`df2`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = FisherSnedecor(torch.tensor([1.0]), torch.tensor([2.0])) + >>> m.sample() # Fisher-Snedecor-distributed with df1=1 and df2=2 + tensor([ 0.2453]) + + Args: + df1 (float or Tensor): degrees of freedom parameter 1 + df2 (float or Tensor): degrees of freedom parameter 2 + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"df1": constraints.positive, "df2": constraints.positive} + support = constraints.positive + has_rsample = True + + def __init__( + self, + df1: Tensor | float, + df2: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.df1, self.df2 = broadcast_all(df1, df2) + self._gamma1 = Gamma(self.df1 * 0.5, self.df1) + self._gamma2 = Gamma(self.df2 * 0.5, self.df2) + + if isinstance(df1, _Number) and isinstance(df2, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.df1.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(FisherSnedecor, _instance) + batch_shape = torch.Size(batch_shape) + new.df1 = self.df1.expand(batch_shape) + new.df2 = self.df2.expand(batch_shape) + new._gamma1 = self._gamma1.expand(batch_shape) + new._gamma2 = self._gamma2.expand(batch_shape) + super(FisherSnedecor, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + df2 = self.df2.clone(memory_format=torch.contiguous_format) + df2[df2 <= 2] = nan + return df2 / (df2 - 2) + + @property + def mode(self) -> Tensor: + mode = (self.df1 - 2) / self.df1 * self.df2 / (self.df2 + 2) + mode[self.df1 <= 2] = nan + return mode + + @property + def variance(self) -> Tensor: + df2 = self.df2.clone(memory_format=torch.contiguous_format) + df2[df2 <= 4] = nan + return ( + 2 + * df2.pow(2) + * (self.df1 + df2 - 2) + / (self.df1 * (df2 - 2).pow(2) * (df2 - 4)) + ) + + def rsample(self, sample_shape: _size = torch.Size(())) -> Tensor: + shape = self._extended_shape(sample_shape) + # X1 ~ Gamma(df1 / 2, 1 / df1), X2 ~ Gamma(df2 / 2, 1 / df2) + # Y = df2 * df1 * X1 / (df1 * df2 * X2) = X1 / X2 ~ F(df1, df2) + X1 = self._gamma1.rsample(sample_shape).view(shape) + X2 = self._gamma2.rsample(sample_shape).view(shape) + tiny = torch.finfo(X2.dtype).tiny + X2.clamp_(min=tiny) + Y = X1 / X2 + Y.clamp_(min=tiny) + return Y + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + ct1 = self.df1 * 0.5 + ct2 = self.df2 * 0.5 + ct3 = self.df1 / self.df2 + t1 = (ct1 + ct2).lgamma() - ct1.lgamma() - ct2.lgamma() + t2 = ct1 * ct3.log() + (ct1 - 1) * torch.log(value) + t3 = (ct1 + ct2) * torch.log1p(ct3 * value) + return t1 + t2 - t3 diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/gamma.py b/.venv/lib/python3.12/site-packages/torch/distributions/gamma.py new file mode 100644 index 0000000000000000000000000000000000000000..a3e70de24330ad27418fd64a2b120062697d1239 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/gamma.py @@ -0,0 +1,132 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Gamma"] + + +def _standard_gamma(concentration, generator=None): + return torch._standard_gamma(concentration, generator=generator) + + +class Gamma(ExponentialFamily): + r""" + Creates a Gamma distribution parameterized by shape :attr:`concentration` and :attr:`rate`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Gamma(torch.tensor([1.0]), torch.tensor([1.0])) + >>> m.sample() # Gamma distributed with concentration=1 and rate=1 + tensor([ 0.1046]) + + Args: + concentration (float or Tensor): shape parameter of the distribution + (often referred to as alpha) + rate (float or Tensor): rate parameter of the distribution + (often referred to as beta), rate = 1 / scale + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "concentration": constraints.positive, + "rate": constraints.positive, + } + support = constraints.nonnegative + has_rsample = True + _mean_carrier_measure = 0 + + @property + def mean(self) -> Tensor: + return self.concentration / self.rate + + @property + def mode(self) -> Tensor: + return ((self.concentration - 1) / self.rate).clamp(min=0) + + @property + def variance(self) -> Tensor: + return self.concentration / self.rate.pow(2) + + def __init__( + self, + concentration: Tensor | float, + rate: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.concentration, self.rate = broadcast_all(concentration, rate) + if isinstance(concentration, _Number) and isinstance(rate, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.concentration.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Gamma, _instance) + batch_shape = torch.Size(batch_shape) + new.concentration = self.concentration.expand(batch_shape) + new.rate = self.rate.expand(batch_shape) + super(Gamma, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def sample( + self, + sample_shape: _size = torch.Size(), + *, + generator: torch.Generator | None = None, + ) -> Tensor: + with torch.no_grad(): + return self.rsample(sample_shape, generator=generator) + + def rsample( + self, + sample_shape: _size = torch.Size(), + generator: torch.Generator | None = None, + ) -> Tensor: + shape = self._extended_shape(sample_shape) + value = _standard_gamma( + self.concentration.expand(shape), generator=generator + ) / self.rate.expand(shape) + value.detach().clamp_( + min=torch.finfo(value.dtype).tiny + ) # do not record in autograd graph + return value + + def log_prob(self, value): + value = torch.as_tensor(value, dtype=self.rate.dtype, device=self.rate.device) + if self._validate_args: + self._validate_sample(value) + return ( + torch.xlogy(self.concentration, self.rate) + + torch.xlogy(self.concentration - 1, value) + - self.rate * value + - torch.lgamma(self.concentration) + ) + + def entropy(self): + return ( + self.concentration + - torch.log(self.rate) + + torch.lgamma(self.concentration) + + (1.0 - self.concentration) * torch.digamma(self.concentration) + ) + + @property + def _natural_params(self) -> tuple[Tensor, Tensor]: + return (self.concentration - 1, -self.rate) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x, y): + return torch.lgamma(x + 1) + (x + 1) * torch.log(-y.reciprocal()) + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return torch.special.gammainc(self.concentration, self.rate * value) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/generalized_pareto.py b/.venv/lib/python3.12/site-packages/torch/distributions/generalized_pareto.py new file mode 100644 index 0000000000000000000000000000000000000000..6b5dabf542dd291be4b828aadfd2aa91225ff8d9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/generalized_pareto.py @@ -0,0 +1,152 @@ +# mypy: allow-untyped-defs +import math +from numbers import Number, Real + +import torch +from torch import inf, nan +from torch.distributions import constraints, Distribution +from torch.distributions.utils import broadcast_all + + +__all__ = ["GeneralizedPareto"] + + +class GeneralizedPareto(Distribution): + r""" + Creates a Generalized Pareto distribution parameterized by :attr:`loc`, :attr:`scale`, and :attr:`concentration`. + + The Generalized Pareto distribution is a family of continuous probability distributions on the real line. + Special cases include Exponential (when :attr:`loc` = 0, :attr:`concentration` = 0), Pareto (when :attr:`concentration` > 0, + :attr:`loc` = :attr:`scale` / :attr:`concentration`), and Uniform (when :attr:`concentration` = -1). + + This distribution is often used to model the tails of other distributions. This implementation is based on the + implementation in TensorFlow Probability. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = GeneralizedPareto(torch.tensor([0.1]), torch.tensor([2.0]), torch.tensor([0.4])) + >>> m.sample() # sample from a Generalized Pareto distribution with loc=0.1, scale=2.0, and concentration=0.4 + tensor([ 1.5623]) + + Args: + loc (float or Tensor): Location parameter of the distribution + scale (float or Tensor): Scale parameter of the distribution + concentration (float or Tensor): Concentration parameter of the distribution + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "loc": constraints.real, + "scale": constraints.positive, + "concentration": constraints.real, + } + has_rsample = True + + def __init__(self, loc, scale, concentration, validate_args=None): + self.loc, self.scale, self.concentration = broadcast_all( + loc, scale, concentration + ) + if ( + isinstance(loc, Number) + and isinstance(scale, Number) + and isinstance(concentration, Number) + ): + batch_shape = torch.Size() + else: + batch_shape = self.loc.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(GeneralizedPareto, _instance) + batch_shape = torch.Size(batch_shape) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + new.concentration = self.concentration.expand(batch_shape) + super(GeneralizedPareto, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def rsample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + u = torch.rand(shape, dtype=self.loc.dtype, device=self.loc.device) + return self.icdf(u) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + z = self._z(value) + eq_zero = torch.isclose(self.concentration, torch.tensor(0.0)) + safe_conc = torch.where( + eq_zero, torch.ones_like(self.concentration), self.concentration + ) + y = 1 / safe_conc + torch.ones_like(z) + where_nonzero = torch.where(y == 0, y, y * torch.log1p(safe_conc * z)) + log_scale = ( + math.log(self.scale) if isinstance(self.scale, Real) else self.scale.log() + ) + return -log_scale - torch.where(eq_zero, z, where_nonzero) + + def log_survival_function(self, value): + if self._validate_args: + self._validate_sample(value) + z = self._z(value) + eq_zero = torch.isclose(self.concentration, torch.tensor(0.0)) + safe_conc = torch.where( + eq_zero, torch.ones_like(self.concentration), self.concentration + ) + where_nonzero = -torch.log1p(safe_conc * z) / safe_conc + return torch.where(eq_zero, -z, where_nonzero) + + def log_cdf(self, value): + return torch.log1p(-torch.exp(self.log_survival_function(value))) + + def cdf(self, value): + return torch.exp(self.log_cdf(value)) + + def icdf(self, value): + loc = self.loc + scale = self.scale + concentration = self.concentration + eq_zero = torch.isclose(concentration, torch.zeros_like(concentration)) + safe_conc = torch.where(eq_zero, torch.ones_like(concentration), concentration) + logu = torch.log1p(-value) + where_nonzero = loc + scale / safe_conc * torch.expm1(-safe_conc * logu) + where_zero = loc - scale * logu + return torch.where(eq_zero, where_zero, where_nonzero) + + def _z(self, x): + return (x - self.loc) / self.scale + + @property + def mean(self): + concentration = self.concentration + valid = concentration < 1 + safe_conc = torch.where(valid, concentration, 0.5) + result = self.loc + self.scale / (1 - safe_conc) + return torch.where(valid, result, nan) + + @property + def variance(self): + concentration = self.concentration + valid = concentration < 0.5 + safe_conc = torch.where(valid, concentration, 0.25) + result = self.scale**2 / ((1 - safe_conc) ** 2 * (1 - 2 * safe_conc)) + return torch.where(valid, result, nan) + + def entropy(self): + ans = torch.log(self.scale) + self.concentration + 1 + return torch.broadcast_to(ans, self._batch_shape) + + @property + def mode(self): + return self.loc + + @constraints.dependent_property(is_discrete=False, event_dim=0) + # pyrefly: ignore [bad-override] + def support(self): + lower = self.loc + upper = torch.where( + self.concentration < 0, lower - self.scale / self.concentration, inf + ) + return constraints.interval(lower, upper) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/geometric.py b/.venv/lib/python3.12/site-packages/torch/distributions/geometric.py new file mode 100644 index 0000000000000000000000000000000000000000..6a3b54642e430050aad3ba4918936682a3692a2a --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/geometric.py @@ -0,0 +1,144 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import ( + broadcast_all, + lazy_property, + logits_to_probs, + probs_to_logits, +) +from torch.nn.functional import binary_cross_entropy_with_logits +from torch.types import _Number, Number + + +__all__ = ["Geometric"] + + +class Geometric(Distribution): + r""" + Creates a Geometric distribution parameterized by :attr:`probs`, + where :attr:`probs` is the probability of success of Bernoulli trials. + + .. math:: + + P(X=k) = (1-p)^{k} p, k = 0, 1, ... + + .. note:: + :func:`torch.distributions.geometric.Geometric` :math:`(k+1)`-th trial is the first success + hence draws samples in :math:`\{0, 1, \ldots\}`, whereas + :func:`torch.Tensor.geometric_` `k`-th trial is the first success hence draws samples in :math:`\{1, 2, \ldots\}`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Geometric(torch.tensor([0.3])) + >>> m.sample() # underlying Bernoulli has 30% chance 1; 70% chance 0 + tensor([ 2.]) + + Args: + probs (Number, Tensor): the probability of sampling `1`. Must be in range (0, 1] + logits (Number, Tensor): the log-odds of sampling `1`. + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.unit_interval, "logits": constraints.real} + support = constraints.nonnegative_integer + + def __init__( + self, + probs: Tensor | Number | None = None, + logits: Tensor | Number | None = None, + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + # pyrefly: ignore [read-only] + (self.probs,) = broadcast_all(probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + # pyrefly: ignore [read-only] + (self.logits,) = broadcast_all(logits) + probs_or_logits = probs if probs is not None else logits + if isinstance(probs_or_logits, _Number): + batch_shape = torch.Size() + else: + if probs_or_logits is None: + raise AssertionError("probs_or_logits is unexpectedly None") + batch_shape = probs_or_logits.size() + super().__init__(batch_shape, validate_args=validate_args) + if self._validate_args and probs is not None: + # Add an extra check beyond unit_interval + value = self.probs + valid = value > 0 + if not valid.all(): + invalid_value = value.data[~valid] + raise ValueError( + "Expected parameter probs " + f"({type(value).__name__} of shape {tuple(value.shape)}) " + f"of distribution {repr(self)} " + f"to be positive but found invalid values:\n{invalid_value}" + ) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Geometric, _instance) + batch_shape = torch.Size(batch_shape) + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + super(Geometric, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + return 1.0 / self.probs - 1.0 + + @property + def mode(self) -> Tensor: + return torch.zeros_like(self.probs) + + @property + def variance(self) -> Tensor: + return (1.0 / self.probs - 1.0) / self.probs + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits, is_binary=True) + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + tiny = torch.finfo(self.probs.dtype).tiny + with torch.no_grad(): + if torch._C._get_tracing_state(): + # [JIT WORKAROUND] lack of support for .uniform_() + u = torch.rand(shape, dtype=self.probs.dtype, device=self.probs.device) + u = u.clamp(min=tiny) + else: + u = self.probs.new(shape).uniform_(tiny, 1) + return (u.log() / (-self.probs).log1p()).floor() + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + value, probs = broadcast_all(value, self.probs) + probs = probs.clone(memory_format=torch.contiguous_format) + probs[(probs == 1) & (value == 0)] = 0 + return value * (-probs).log1p() + self.probs.log() + + def entropy(self): + return ( + binary_cross_entropy_with_logits(self.logits, self.probs, reduction="none") + / self.probs + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/gumbel.py b/.venv/lib/python3.12/site-packages/torch/distributions/gumbel.py new file mode 100644 index 0000000000000000000000000000000000000000..762efc3bff6e86e216946dfa0ddd22cb589d1392 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/gumbel.py @@ -0,0 +1,91 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AffineTransform, ExpTransform +from torch.distributions.uniform import Uniform +from torch.distributions.utils import broadcast_all, euler_constant +from torch.types import _Number + + +__all__ = ["Gumbel"] + + +class Gumbel(TransformedDistribution): + r""" + Samples from a Gumbel Distribution. + + Examples:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Gumbel(torch.tensor([1.0]), torch.tensor([2.0])) + >>> m.sample() # sample from Gumbel distribution with loc=1, scale=2 + tensor([ 1.0124]) + + Args: + loc (float or Tensor): Location parameter of the distribution + scale (float or Tensor): Scale parameter of the distribution + """ + + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + # pyrefly: ignore [bad-override] + support = constraints.real + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.loc, self.scale = broadcast_all(loc, scale) + finfo = torch.finfo(self.loc.dtype) + if isinstance(loc, _Number) and isinstance(scale, _Number): + base_dist = Uniform(finfo.tiny, 1 - finfo.eps, validate_args=validate_args) + else: + base_dist = Uniform( + torch.full_like(self.loc, finfo.tiny), + torch.full_like(self.loc, 1 - finfo.eps), + validate_args=validate_args, + ) + transforms = [ + ExpTransform().inv, + AffineTransform(loc=0, scale=-torch.ones_like(self.scale)), + ExpTransform().inv, + AffineTransform(loc=loc, scale=-self.scale), + ] + super().__init__(base_dist, transforms, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Gumbel, _instance) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + return super().expand(batch_shape, _instance=new) + + # Explicitly defining the log probability function for Gumbel due to precision issues + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + y = (self.loc - value) / self.scale + return (y - y.exp()) - self.scale.log() + + @property + def mean(self) -> Tensor: + return self.loc + self.scale * euler_constant + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def stddev(self) -> Tensor: + return (math.pi / math.sqrt(6)) * self.scale + + @property + def variance(self) -> Tensor: + return self.stddev.pow(2) + + def entropy(self): + return self.scale.log() + (1 + euler_constant) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/half_cauchy.py b/.venv/lib/python3.12/site-packages/torch/distributions/half_cauchy.py new file mode 100644 index 0000000000000000000000000000000000000000..6f79c762558e22c310f18b8ecce89b38b5f424c7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/half_cauchy.py @@ -0,0 +1,92 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import inf, Tensor +from torch.distributions import constraints +from torch.distributions.cauchy import Cauchy +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AbsTransform + + +__all__ = ["HalfCauchy"] + + +class HalfCauchy(TransformedDistribution): + r""" + Creates a half-Cauchy distribution parameterized by `scale` where:: + + X ~ Cauchy(0, scale) + Y = |X| ~ HalfCauchy(scale) + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = HalfCauchy(torch.tensor([1.0])) + >>> m.sample() # half-cauchy distributed with scale=1 + tensor([ 2.3214]) + + Args: + scale (float or Tensor): scale of the full Cauchy distribution + """ + + arg_constraints = {"scale": constraints.positive} + # pyrefly: ignore [bad-override] + support = constraints.nonnegative + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: Cauchy + + def __init__( + self, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + base_dist = Cauchy(0, scale, validate_args=False) + super().__init__(base_dist, AbsTransform(), validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(HalfCauchy, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def scale(self) -> Tensor: + return self.base_dist.scale + + @property + def mean(self) -> Tensor: + return torch.full( + self._extended_shape(), + math.inf, + dtype=self.scale.dtype, + device=self.scale.device, + ) + + @property + def mode(self) -> Tensor: + return torch.zeros_like(self.scale) + + @property + def variance(self) -> Tensor: + return self.base_dist.variance + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + value = torch.as_tensor( + value, dtype=self.base_dist.scale.dtype, device=self.base_dist.scale.device + ) + log_prob = self.base_dist.log_prob(value) + math.log(2) + log_prob = torch.where(value >= 0, log_prob, -inf) + return log_prob + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return 2 * self.base_dist.cdf(value) - 1 + + def icdf(self, prob): + return self.base_dist.icdf((prob + 1) / 2) + + def entropy(self): + return self.base_dist.entropy() - math.log(2) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/half_normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/half_normal.py new file mode 100644 index 0000000000000000000000000000000000000000..9d7324a6bf5d48844c896da473766bef3f1bd6e7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/half_normal.py @@ -0,0 +1,84 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import inf, Tensor +from torch.distributions import constraints +from torch.distributions.normal import Normal +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AbsTransform + + +__all__ = ["HalfNormal"] + + +class HalfNormal(TransformedDistribution): + r""" + Creates a half-normal distribution parameterized by `scale` where:: + + X ~ Normal(0, scale) + Y = |X| ~ HalfNormal(scale) + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = HalfNormal(torch.tensor([1.0])) + >>> m.sample() # half-normal distributed with scale=1 + tensor([ 0.1046]) + + Args: + scale (float or Tensor): scale of the full Normal distribution + """ + + arg_constraints = {"scale": constraints.positive} + # pyrefly: ignore [bad-override] + support = constraints.nonnegative + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: Normal + + def __init__( + self, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + base_dist = Normal(0, scale, validate_args=False) + super().__init__(base_dist, AbsTransform(), validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(HalfNormal, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def scale(self) -> Tensor: + return self.base_dist.scale + + @property + def mean(self) -> Tensor: + return self.scale * math.sqrt(2 / math.pi) + + @property + def mode(self) -> Tensor: + return torch.zeros_like(self.scale) + + @property + def variance(self) -> Tensor: + return self.scale.pow(2) * (1 - 2 / math.pi) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + log_prob = self.base_dist.log_prob(value) + math.log(2) + log_prob = torch.where(value >= 0, log_prob, -inf) + return log_prob + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return 2 * self.base_dist.cdf(value) - 1 + + def icdf(self, prob): + return self.base_dist.icdf((prob + 1) / 2) + + def entropy(self): + return self.base_dist.entropy() - math.log(2) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/independent.py b/.venv/lib/python3.12/site-packages/torch/distributions/independent.py new file mode 100644 index 0000000000000000000000000000000000000000..33d86e35731e953942884257c90a0e52a1e198f8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/independent.py @@ -0,0 +1,139 @@ +# mypy: allow-untyped-defs +from typing import Generic, TypeVar + +import torch +from torch import Size, Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import _sum_rightmost +from torch.types import _size + + +__all__ = ["Independent"] + + +D = TypeVar("D", bound=Distribution) + + +class Independent(Distribution, Generic[D]): + r""" + Reinterprets some of the batch dims of a distribution as event dims. + + This is mainly useful for changing the shape of the result of + :meth:`log_prob`. For example to create a diagonal Normal distribution with + the same shape as a Multivariate Normal distribution (so they are + interchangeable), you can:: + + >>> from torch.distributions.multivariate_normal import MultivariateNormal + >>> from torch.distributions.normal import Normal + >>> loc = torch.zeros(3) + >>> scale = torch.ones(3) + >>> mvn = MultivariateNormal(loc, scale_tril=torch.diag(scale)) + >>> [mvn.batch_shape, mvn.event_shape] + [torch.Size([]), torch.Size([3])] + >>> normal = Normal(loc, scale) + >>> [normal.batch_shape, normal.event_shape] + [torch.Size([3]), torch.Size([])] + >>> diagn = Independent(normal, 1) + >>> [diagn.batch_shape, diagn.event_shape] + [torch.Size([]), torch.Size([3])] + + Args: + base_distribution (torch.distributions.distribution.Distribution): a + base distribution + reinterpreted_batch_ndims (int): the number of batch dims to + reinterpret as event dims + """ + + arg_constraints: dict[str, constraints.Constraint] = {} + base_dist: D + + def __init__( + self, + base_distribution: D, + reinterpreted_batch_ndims: int, + validate_args: bool | None = None, + ) -> None: + if reinterpreted_batch_ndims > len(base_distribution.batch_shape): + raise ValueError( + "Expected reinterpreted_batch_ndims <= len(base_distribution.batch_shape), " + f"actual {reinterpreted_batch_ndims} vs {len(base_distribution.batch_shape)}" + ) + shape: Size = base_distribution.batch_shape + base_distribution.event_shape + event_dim: int = reinterpreted_batch_ndims + len(base_distribution.event_shape) + batch_shape = shape[: len(shape) - event_dim] + event_shape = shape[len(shape) - event_dim :] + self.base_dist = base_distribution + self.reinterpreted_batch_ndims = reinterpreted_batch_ndims + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Independent, _instance) + batch_shape = torch.Size(batch_shape) + new.base_dist = self.base_dist.expand( + batch_shape + self.event_shape[: self.reinterpreted_batch_ndims] + ) + new.reinterpreted_batch_ndims = self.reinterpreted_batch_ndims + super(Independent, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + @property + def has_rsample(self) -> bool: # type: ignore[override] + return self.base_dist.has_rsample + + @property + def has_enumerate_support(self) -> bool: # type: ignore[override] + if self.reinterpreted_batch_ndims > 0: + return False + return self.base_dist.has_enumerate_support + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def support(self): + result = self.base_dist.support + if self.reinterpreted_batch_ndims: + result = constraints.independent(result, self.reinterpreted_batch_ndims) + return result + + @property + def mean(self) -> Tensor: + return self.base_dist.mean + + @property + def mode(self) -> Tensor: + return self.base_dist.mode + + @property + def variance(self) -> Tensor: + return self.base_dist.variance + + def sample(self, sample_shape=torch.Size()) -> Tensor: + return self.base_dist.sample(sample_shape) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + return self.base_dist.rsample(sample_shape) + + def log_prob(self, value): + log_prob = self.base_dist.log_prob(value) + return _sum_rightmost(log_prob, self.reinterpreted_batch_ndims) + + def entropy(self): + entropy = self.base_dist.entropy() + return _sum_rightmost(entropy, self.reinterpreted_batch_ndims) + + def enumerate_support(self, expand=True): + if self.reinterpreted_batch_ndims > 0: + raise NotImplementedError( + "Enumeration over cartesian product is not implemented" + ) + return self.base_dist.enumerate_support(expand=expand) + + def __repr__(self): + return ( + self.__class__.__name__ + + f"({self.base_dist}, {self.reinterpreted_batch_ndims})" + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/inverse_gamma.py b/.venv/lib/python3.12/site-packages/torch/distributions/inverse_gamma.py new file mode 100644 index 0000000000000000000000000000000000000000..5dfd50378b544b1acad6e61a103a6215c5d48dc4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/inverse_gamma.py @@ -0,0 +1,92 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.gamma import Gamma +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import PowerTransform + + +__all__ = ["InverseGamma"] + + +class InverseGamma(TransformedDistribution): + r""" + Creates an inverse gamma distribution parameterized by :attr:`concentration` and :attr:`rate` + where:: + + X ~ Gamma(concentration, rate) + Y = 1 / X ~ InverseGamma(concentration, rate) + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = InverseGamma(torch.tensor([2.0]), torch.tensor([3.0])) + >>> m.sample() + tensor([ 1.2953]) + + Args: + concentration (float or Tensor): shape parameter of the distribution + (often referred to as alpha) + rate (float or Tensor): rate = 1 / scale of the distribution + (often referred to as beta) + """ + + arg_constraints = { + "concentration": constraints.positive, + "rate": constraints.positive, + } + # pyrefly: ignore [bad-override] + support = constraints.positive + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: Gamma + + def __init__( + self, + concentration: Tensor | float, + rate: Tensor | float, + validate_args: bool | None = None, + ) -> None: + base_dist = Gamma(concentration, rate, validate_args=validate_args) + neg_one = -base_dist.rate.new_ones(()) + super().__init__( + base_dist, PowerTransform(neg_one), validate_args=validate_args + ) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(InverseGamma, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def concentration(self) -> Tensor: + return self.base_dist.concentration + + @property + def rate(self) -> Tensor: + return self.base_dist.rate + + @property + def mean(self) -> Tensor: + result = self.rate / (self.concentration - 1) + return torch.where(self.concentration > 1, result, torch.inf) + + @property + def mode(self) -> Tensor: + return self.rate / (self.concentration + 1) + + @property + def variance(self) -> Tensor: + result = self.rate.square() / ( + (self.concentration - 1).square() * (self.concentration - 2) + ) + return torch.where(self.concentration > 2, result, torch.inf) + + def entropy(self): + return ( + self.concentration + + self.rate.log() + + self.concentration.lgamma() + - (1 + self.concentration) * self.concentration.digamma() + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/kl.py b/.venv/lib/python3.12/site-packages/torch/distributions/kl.py new file mode 100644 index 0000000000000000000000000000000000000000..b8ea3890180e7f53aac2d57c9f4a1c103891c171 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/kl.py @@ -0,0 +1,974 @@ +# mypy: allow-untyped-defs +import math +import warnings +from collections.abc import Callable +from functools import total_ordering + +import torch +from torch import inf, Tensor + +from .bernoulli import Bernoulli +from .beta import Beta +from .binomial import Binomial +from .categorical import Categorical +from .cauchy import Cauchy +from .continuous_bernoulli import ContinuousBernoulli +from .dirichlet import Dirichlet +from .distribution import Distribution +from .exp_family import ExponentialFamily +from .exponential import Exponential +from .gamma import Gamma +from .geometric import Geometric +from .gumbel import Gumbel +from .half_normal import HalfNormal +from .independent import Independent +from .laplace import Laplace +from .lowrank_multivariate_normal import ( + _batch_lowrank_logdet, + _batch_lowrank_mahalanobis, + LowRankMultivariateNormal, +) +from .multivariate_normal import _batch_mahalanobis, MultivariateNormal +from .normal import Normal +from .one_hot_categorical import OneHotCategorical +from .pareto import Pareto +from .poisson import Poisson +from .transformed_distribution import TransformedDistribution +from .uniform import Uniform +from .utils import _sum_rightmost, euler_constant as _euler_gamma + + +_KL_REGISTRY: dict[ + tuple[type, type], Callable +] = {} # Source of truth mapping a few general (type, type) pairs to functions. +_KL_MEMOIZE: dict[ + tuple[type, type], Callable +] = {} # Memoized version mapping many specific (type, type) pairs to functions. + +__all__ = ["register_kl", "kl_divergence"] + + +def register_kl(type_p, type_q): + """ + Decorator to register a pairwise function with :meth:`kl_divergence`. + Usage:: + + @register_kl(Normal, Normal) + def kl_normal_normal(p, q): + # insert implementation here + + Lookup returns the most specific (type,type) match ordered by subclass. If + the match is ambiguous, a `RuntimeWarning` is raised. For example to + resolve the ambiguous situation:: + + @register_kl(BaseP, DerivedQ) + def kl_version1(p, q): ... + @register_kl(DerivedP, BaseQ) + def kl_version2(p, q): ... + + you should register a third most-specific implementation, e.g.:: + + register_kl(DerivedP, DerivedQ)(kl_version1) # Break the tie. + + Args: + type_p (type): A subclass of :class:`~torch.distributions.Distribution`. + type_q (type): A subclass of :class:`~torch.distributions.Distribution`. + """ + if not isinstance(type_p, type) and issubclass(type_p, Distribution): + raise TypeError( + f"Expected type_p to be a Distribution subclass but got {type_p}" + ) + if not isinstance(type_q, type) and issubclass(type_q, Distribution): + raise TypeError( + f"Expected type_q to be a Distribution subclass but got {type_q}" + ) + + def decorator(fun): + _KL_REGISTRY[type_p, type_q] = fun + _KL_MEMOIZE.clear() # reset since lookup order may have changed + return fun + + return decorator + + +@total_ordering +class _Match: + __slots__ = ["types"] + + def __init__(self, *types): + self.types = types + + def __eq__(self, other): + return self.types == other.types + + def __le__(self, other): + for x, y in zip(self.types, other.types): + if not issubclass(x, y): + return False + if x is not y: + break + return True + + +def _dispatch_kl(type_p, type_q): + """ + Find the most specific approximate match, assuming single inheritance. + """ + matches = [ + (super_p, super_q) + for super_p, super_q in _KL_REGISTRY + if issubclass(type_p, super_p) and issubclass(type_q, super_q) + ] + if not matches: + return NotImplemented + # Check that the left- and right- lexicographic orders agree. + # mypy isn't smart enough to know that _Match implements __lt__ + # see: https://github.com/python/typing/issues/760#issuecomment-710670503 + left_p, left_q = min(_Match(*m) for m in matches).types # type: ignore[type-var] + right_q, right_p = min(_Match(*reversed(m)) for m in matches).types # type: ignore[type-var] + left_fun = _KL_REGISTRY[left_p, left_q] + right_fun = _KL_REGISTRY[right_p, right_q] + if left_fun is not right_fun: + warnings.warn( + f"Ambiguous kl_divergence({type_p.__name__}, {type_q.__name__}). " + f"Please register_kl({left_p.__name__}, {right_q.__name__})", + RuntimeWarning, + stacklevel=2, + ) + return left_fun + + +def _infinite_like(tensor): + """ + Helper function for obtaining infinite KL Divergence throughout + """ + return torch.full_like(tensor, inf) + + +def _x_log_x(tensor): + """ + Utility function for calculating x log x + """ + return torch.special.xlogy(tensor, tensor) # produces correct result for x=0 + + +def _batch_trace_XXT(bmat): + """ + Utility function for calculating the trace of XX^{T} with X having arbitrary trailing batch dimensions + """ + n = bmat.size(-1) + m = bmat.size(-2) + flat_trace = bmat.reshape(-1, m * n).pow(2).sum(-1) + return flat_trace.reshape(bmat.shape[:-2]) + + +def kl_divergence(p: Distribution, q: Distribution) -> Tensor: + r""" + Compute Kullback-Leibler divergence :math:`KL(p \| q)` between two distributions. + + .. math:: + + KL(p \| q) = \int p(x) \log\frac {p(x)} {q(x)} \,dx + + Args: + p (Distribution): A :class:`~torch.distributions.Distribution` object. + q (Distribution): A :class:`~torch.distributions.Distribution` object. + + Returns: + Tensor: A batch of KL divergences of shape `batch_shape`. + + Raises: + NotImplementedError: If the distribution types have not been registered via + :meth:`register_kl`. + """ + try: + fun = _KL_MEMOIZE[type(p), type(q)] + except KeyError: + fun = _dispatch_kl(type(p), type(q)) + _KL_MEMOIZE[type(p), type(q)] = fun + if fun is NotImplemented: + raise NotImplementedError( + f"No KL(p || q) is implemented for p type {p.__class__.__name__} and q type {q.__class__.__name__}" + ) + return fun(p, q) + + +################################################################################ +# KL Divergence Implementations +################################################################################ + +# Same distributions + + +@register_kl(Bernoulli, Bernoulli) +def _kl_bernoulli_bernoulli(p, q): + t1 = p.probs * ( + torch.nn.functional.softplus(-q.logits) + - torch.nn.functional.softplus(-p.logits) + ) + t1[q.probs == 0] = inf + t1[p.probs == 0] = 0 + t2 = (1 - p.probs) * ( + torch.nn.functional.softplus(q.logits) - torch.nn.functional.softplus(p.logits) + ) + t2[q.probs == 1] = inf + t2[p.probs == 1] = 0 + return t1 + t2 + + +@register_kl(Beta, Beta) +def _kl_beta_beta(p, q): + sum_params_p = p.concentration1 + p.concentration0 + sum_params_q = q.concentration1 + q.concentration0 + t1 = q.concentration1.lgamma() + q.concentration0.lgamma() + (sum_params_p).lgamma() + t2 = p.concentration1.lgamma() + p.concentration0.lgamma() + (sum_params_q).lgamma() + t3 = (p.concentration1 - q.concentration1) * torch.digamma(p.concentration1) + t4 = (p.concentration0 - q.concentration0) * torch.digamma(p.concentration0) + t5 = (sum_params_q - sum_params_p) * torch.digamma(sum_params_p) + return t1 - t2 + t3 + t4 + t5 + + +@register_kl(Binomial, Binomial) +def _kl_binomial_binomial(p, q): + # from https://math.stackexchange.com/questions/2214993/ + # kullback-leibler-divergence-for-binomial-distributions-p-and-q + if (p.total_count < q.total_count).any(): + raise NotImplementedError( + "KL between Binomials where q.total_count > p.total_count is not implemented" + ) + kl = p.total_count * ( + p.probs * (p.logits - q.logits) + (-p.probs).log1p() - (-q.probs).log1p() + ) + inf_idxs = p.total_count > q.total_count + kl[inf_idxs] = _infinite_like(kl[inf_idxs]) + return kl + + +@register_kl(Categorical, Categorical) +def _kl_categorical_categorical(p, q): + t = p.probs * (p.logits - q.logits) + t[(q.probs == 0).expand_as(t)] = inf + t[(p.probs == 0).expand_as(t)] = 0 + return t.sum(-1) + + +@register_kl(ContinuousBernoulli, ContinuousBernoulli) +def _kl_continuous_bernoulli_continuous_bernoulli(p, q): + t1 = p.mean * (p.logits - q.logits) + t2 = p._cont_bern_log_norm() + torch.log1p(-p.probs) + t3 = -q._cont_bern_log_norm() - torch.log1p(-q.probs) + return t1 + t2 + t3 + + +@register_kl(Dirichlet, Dirichlet) +def _kl_dirichlet_dirichlet(p, q): + # From https://statproofbook.github.io/P/dir-kl.html + sum_p_concentration = p.concentration.sum(-1) + sum_q_concentration = q.concentration.sum(-1) + t1 = sum_p_concentration.lgamma() - sum_q_concentration.lgamma() + t2 = (p.concentration.lgamma() - q.concentration.lgamma()).sum(-1) + t3 = p.concentration - q.concentration + t4 = p.concentration.digamma() - sum_p_concentration.digamma().unsqueeze(-1) + return t1 - t2 + (t3 * t4).sum(-1) + + +@register_kl(Exponential, Exponential) +def _kl_exponential_exponential(p, q): + rate_ratio = q.rate / p.rate + t1 = -rate_ratio.log() + return t1 + rate_ratio - 1 + + +@register_kl(ExponentialFamily, ExponentialFamily) +def _kl_expfamily_expfamily(p, q): + if type(p) is not type(q): + raise NotImplementedError( + "The cross KL-divergence between different exponential families cannot \ + be computed using Bregman divergences" + ) + p_nparams = [np.detach().requires_grad_() for np in p._natural_params] + q_nparams = q._natural_params + lg_normal = p._log_normalizer(*p_nparams) + gradients = torch.autograd.grad(lg_normal.sum(), p_nparams, create_graph=True) + result = q._log_normalizer(*q_nparams) - lg_normal + for pnp, qnp, g in zip(p_nparams, q_nparams, gradients): + term = (qnp - pnp) * g + result -= _sum_rightmost(term, len(q.event_shape)) + return result + + +@register_kl(Gamma, Gamma) +def _kl_gamma_gamma(p, q): + t1 = q.concentration * (p.rate / q.rate).log() + t2 = torch.lgamma(q.concentration) - torch.lgamma(p.concentration) + t3 = (p.concentration - q.concentration) * torch.digamma(p.concentration) + t4 = (q.rate - p.rate) * (p.concentration / p.rate) + return t1 + t2 + t3 + t4 + + +@register_kl(Gumbel, Gumbel) +def _kl_gumbel_gumbel(p, q): + ct1 = p.scale / q.scale + ct2 = q.loc / q.scale + ct3 = p.loc / q.scale + t1 = -ct1.log() - ct2 + ct3 + t2 = ct1 * _euler_gamma + t3 = torch.exp(ct2 + (1 + ct1).lgamma() - ct3) + return t1 + t2 + t3 - (1 + _euler_gamma) + + +@register_kl(Geometric, Geometric) +def _kl_geometric_geometric(p, q): + return -p.entropy() - torch.log1p(-q.probs) / p.probs - q.logits + + +@register_kl(HalfNormal, HalfNormal) +def _kl_halfnormal_halfnormal(p, q): + return _kl_normal_normal(p.base_dist, q.base_dist) + + +@register_kl(Laplace, Laplace) +def _kl_laplace_laplace(p, q): + # From http://www.mast.queensu.ca/~communications/Papers/gil-msc11.pdf + scale_ratio = p.scale / q.scale + loc_abs_diff = (p.loc - q.loc).abs() + t1 = -scale_ratio.log() + t2 = loc_abs_diff / q.scale + t3 = scale_ratio * torch.exp(-loc_abs_diff / p.scale) + return t1 + t2 + t3 - 1 + + +@register_kl(LowRankMultivariateNormal, LowRankMultivariateNormal) +def _kl_lowrankmultivariatenormal_lowrankmultivariatenormal(p, q): + if p.event_shape != q.event_shape: + raise ValueError( + "KL-divergence between two Low Rank Multivariate Normals with\ + different event shapes cannot be computed" + ) + + term1 = _batch_lowrank_logdet( + q._unbroadcasted_cov_factor, q._unbroadcasted_cov_diag, q._capacitance_tril + ) - _batch_lowrank_logdet( + p._unbroadcasted_cov_factor, p._unbroadcasted_cov_diag, p._capacitance_tril + ) + term3 = _batch_lowrank_mahalanobis( + q._unbroadcasted_cov_factor, + q._unbroadcasted_cov_diag, + q.loc - p.loc, + q._capacitance_tril, + ) + # Expands term2 according to + # inv(qcov) @ pcov = [inv(qD) - inv(qD) @ qW @ inv(qC) @ qW.T @ inv(qD)] @ (pW @ pW.T + pD) + # = [inv(qD) - A.T @ A] @ (pD + pW @ pW.T) + qWt_qDinv = q._unbroadcasted_cov_factor.mT / q._unbroadcasted_cov_diag.unsqueeze(-2) + A = torch.linalg.solve_triangular(q._capacitance_tril, qWt_qDinv, upper=False) + term21 = (p._unbroadcasted_cov_diag / q._unbroadcasted_cov_diag).sum(-1) + term22 = _batch_trace_XXT( + p._unbroadcasted_cov_factor * q._unbroadcasted_cov_diag.rsqrt().unsqueeze(-1) + ) + term23 = _batch_trace_XXT(A * p._unbroadcasted_cov_diag.sqrt().unsqueeze(-2)) + term24 = _batch_trace_XXT(A.matmul(p._unbroadcasted_cov_factor)) + term2 = term21 + term22 - term23 - term24 + return 0.5 * (term1 + term2 + term3 - p.event_shape[0]) + + +@register_kl(MultivariateNormal, LowRankMultivariateNormal) +def _kl_multivariatenormal_lowrankmultivariatenormal(p, q): + if p.event_shape != q.event_shape: + raise ValueError( + "KL-divergence between two (Low Rank) Multivariate Normals with\ + different event shapes cannot be computed" + ) + + term1 = _batch_lowrank_logdet( + q._unbroadcasted_cov_factor, q._unbroadcasted_cov_diag, q._capacitance_tril + ) - 2 * p._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum(-1) + term3 = _batch_lowrank_mahalanobis( + q._unbroadcasted_cov_factor, + q._unbroadcasted_cov_diag, + q.loc - p.loc, + q._capacitance_tril, + ) + # Expands term2 according to + # inv(qcov) @ pcov = [inv(qD) - inv(qD) @ qW @ inv(qC) @ qW.T @ inv(qD)] @ p_tril @ p_tril.T + # = [inv(qD) - A.T @ A] @ p_tril @ p_tril.T + qWt_qDinv = q._unbroadcasted_cov_factor.mT / q._unbroadcasted_cov_diag.unsqueeze(-2) + A = torch.linalg.solve_triangular(q._capacitance_tril, qWt_qDinv, upper=False) + term21 = _batch_trace_XXT( + p._unbroadcasted_scale_tril * q._unbroadcasted_cov_diag.rsqrt().unsqueeze(-1) + ) + term22 = _batch_trace_XXT(A.matmul(p._unbroadcasted_scale_tril)) + term2 = term21 - term22 + return 0.5 * (term1 + term2 + term3 - p.event_shape[0]) + + +@register_kl(LowRankMultivariateNormal, MultivariateNormal) +def _kl_lowrankmultivariatenormal_multivariatenormal(p, q): + if p.event_shape != q.event_shape: + raise ValueError( + "KL-divergence between two (Low Rank) Multivariate Normals with\ + different event shapes cannot be computed" + ) + + term1 = 2 * q._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum( + -1 + ) - _batch_lowrank_logdet( + p._unbroadcasted_cov_factor, p._unbroadcasted_cov_diag, p._capacitance_tril + ) + term3 = _batch_mahalanobis(q._unbroadcasted_scale_tril, (q.loc - p.loc)) + # Expands term2 according to + # inv(qcov) @ pcov = inv(q_tril @ q_tril.T) @ (pW @ pW.T + pD) + combined_batch_shape = torch._C._infer_size( + q._unbroadcasted_scale_tril.shape[:-2], p._unbroadcasted_cov_factor.shape[:-2] + ) + n = p.event_shape[0] + q_scale_tril = q._unbroadcasted_scale_tril.expand(combined_batch_shape + (n, n)) + p_cov_factor = p._unbroadcasted_cov_factor.expand( + combined_batch_shape + (n, p.cov_factor.size(-1)) + ) + p_cov_diag = torch.diag_embed(p._unbroadcasted_cov_diag.sqrt()).expand( + combined_batch_shape + (n, n) + ) + term21 = _batch_trace_XXT( + torch.linalg.solve_triangular(q_scale_tril, p_cov_factor, upper=False) + ) + term22 = _batch_trace_XXT( + torch.linalg.solve_triangular(q_scale_tril, p_cov_diag, upper=False) + ) + term2 = term21 + term22 + return 0.5 * (term1 + term2 + term3 - p.event_shape[0]) + + +@register_kl(MultivariateNormal, MultivariateNormal) +def _kl_multivariatenormal_multivariatenormal(p, q): + # From https://en.wikipedia.org/wiki/Multivariate_normal_distribution#Kullback%E2%80%93Leibler_divergence + if p.event_shape != q.event_shape: + raise ValueError( + "KL-divergence between two Multivariate Normals with\ + different event shapes cannot be computed" + ) + + half_term1 = q._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum( + -1 + ) - p._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum(-1) + combined_batch_shape = torch._C._infer_size( + q._unbroadcasted_scale_tril.shape[:-2], p._unbroadcasted_scale_tril.shape[:-2] + ) + n = p.event_shape[0] + q_scale_tril = q._unbroadcasted_scale_tril.expand(combined_batch_shape + (n, n)) + p_scale_tril = p._unbroadcasted_scale_tril.expand(combined_batch_shape + (n, n)) + term2 = _batch_trace_XXT( + torch.linalg.solve_triangular(q_scale_tril, p_scale_tril, upper=False) + ) + term3 = _batch_mahalanobis(q._unbroadcasted_scale_tril, (q.loc - p.loc)) + return half_term1 + 0.5 * (term2 + term3 - n) + + +@register_kl(Normal, Normal) +def _kl_normal_normal(p, q): + var_ratio = (p.scale / q.scale).pow(2) + t1 = ((p.loc - q.loc) / q.scale).pow(2) + return 0.5 * (var_ratio + t1 - 1 - var_ratio.log()) + + +@register_kl(OneHotCategorical, OneHotCategorical) +def _kl_onehotcategorical_onehotcategorical(p, q): + return _kl_categorical_categorical(p._categorical, q._categorical) + + +@register_kl(Pareto, Pareto) +def _kl_pareto_pareto(p, q): + # From http://www.mast.queensu.ca/~communications/Papers/gil-msc11.pdf + scale_ratio = p.scale / q.scale + alpha_ratio = q.alpha / p.alpha + t1 = q.alpha * scale_ratio.log() + t2 = -alpha_ratio.log() + result = t1 + t2 + alpha_ratio - 1 + result[p.support.lower_bound < q.support.lower_bound] = inf + return result + + +@register_kl(Poisson, Poisson) +def _kl_poisson_poisson(p, q): + return p.rate * (p.rate.log() - q.rate.log()) - (p.rate - q.rate) + + +@register_kl(TransformedDistribution, TransformedDistribution) +def _kl_transformed_transformed(p, q): + if p.transforms != q.transforms: + raise NotImplementedError + if p.event_shape != q.event_shape: + raise NotImplementedError + return kl_divergence(p.base_dist, q.base_dist) + + +@register_kl(Uniform, Uniform) +def _kl_uniform_uniform(p, q): + result = ((q.high - q.low) / (p.high - p.low)).log() + result[(q.low > p.low) | (q.high < p.high)] = inf + return result + + +# Different distributions +@register_kl(Bernoulli, Poisson) +def _kl_bernoulli_poisson(p, q): + return -p.entropy() - (p.probs * q.rate.log() - q.rate) + + +@register_kl(Beta, ContinuousBernoulli) +def _kl_beta_continuous_bernoulli(p, q): + return ( + -p.entropy() + - p.mean * q.logits + - torch.log1p(-q.probs) + - q._cont_bern_log_norm() + ) + + +@register_kl(Beta, Pareto) +def _kl_beta_infinity(p, q): + return _infinite_like(p.concentration1) + + +@register_kl(Beta, Exponential) +def _kl_beta_exponential(p, q): + return ( + -p.entropy() + - q.rate.log() + + q.rate * (p.concentration1 / (p.concentration1 + p.concentration0)) + ) + + +@register_kl(Beta, Gamma) +def _kl_beta_gamma(p, q): + t1 = -p.entropy() + t2 = q.concentration.lgamma() - q.concentration * q.rate.log() + t3 = (q.concentration - 1) * ( + p.concentration1.digamma() - (p.concentration1 + p.concentration0).digamma() + ) + t4 = q.rate * p.concentration1 / (p.concentration1 + p.concentration0) + return t1 + t2 - t3 + t4 + + +# TODO: Add Beta-Laplace KL Divergence + + +@register_kl(Beta, Normal) +def _kl_beta_normal(p, q): + E_beta = p.concentration1 / (p.concentration1 + p.concentration0) + var_normal = q.scale.pow(2) + t1 = -p.entropy() + t2 = 0.5 * (var_normal * 2 * math.pi).log() + t3 = ( + E_beta * (1 - E_beta) / (p.concentration1 + p.concentration0 + 1) + + E_beta.pow(2) + ) * 0.5 + t4 = q.loc * E_beta + t5 = q.loc.pow(2) * 0.5 + return t1 + t2 + (t3 - t4 + t5) / var_normal + + +@register_kl(Beta, Uniform) +def _kl_beta_uniform(p, q): + result = -p.entropy() + (q.high - q.low).log() + result[(q.low > p.support.lower_bound) | (q.high < p.support.upper_bound)] = inf + return result + + +# Note that the KL between a ContinuousBernoulli and Beta has no closed form + + +@register_kl(ContinuousBernoulli, Pareto) +def _kl_continuous_bernoulli_infinity(p, q): + return _infinite_like(p.probs) + + +@register_kl(ContinuousBernoulli, Exponential) +def _kl_continuous_bernoulli_exponential(p, q): + return -p.entropy() - torch.log(q.rate) + q.rate * p.mean + + +# Note that the KL between a ContinuousBernoulli and Gamma has no closed form +# TODO: Add ContinuousBernoulli-Laplace KL Divergence + + +@register_kl(ContinuousBernoulli, Normal) +def _kl_continuous_bernoulli_normal(p, q): + t1 = -p.entropy() + t2 = 0.5 * (math.log(2.0 * math.pi) + torch.square(q.loc / q.scale)) + torch.log( + q.scale + ) + t3 = (p.variance + torch.square(p.mean) - 2.0 * q.loc * p.mean) / ( + 2.0 * torch.square(q.scale) + ) + return t1 + t2 + t3 + + +@register_kl(ContinuousBernoulli, Uniform) +def _kl_continuous_bernoulli_uniform(p, q): + result = -p.entropy() + (q.high - q.low).log() + return torch.where( + torch.max( + torch.ge(q.low, p.support.lower_bound), + torch.le(q.high, p.support.upper_bound), + ), + torch.ones_like(result) * inf, + result, + ) + + +@register_kl(Exponential, Beta) +@register_kl(Exponential, ContinuousBernoulli) +@register_kl(Exponential, Pareto) +@register_kl(Exponential, Uniform) +def _kl_exponential_infinity(p, q): + return _infinite_like(p.rate) + + +@register_kl(Exponential, Gamma) +def _kl_exponential_gamma(p, q): + ratio = q.rate / p.rate + t1 = -q.concentration * torch.log(ratio) + return ( + t1 + + ratio + + q.concentration.lgamma() + + q.concentration * _euler_gamma + - (1 + _euler_gamma) + ) + + +@register_kl(Exponential, Gumbel) +def _kl_exponential_gumbel(p, q): + scale_rate_prod = p.rate * q.scale + loc_scale_ratio = q.loc / q.scale + t1 = scale_rate_prod.log() - 1 + t2 = torch.exp(loc_scale_ratio) * scale_rate_prod / (scale_rate_prod + 1) + t3 = scale_rate_prod.reciprocal() + return t1 - loc_scale_ratio + t2 + t3 + + +# TODO: Add Exponential-Laplace KL Divergence + + +@register_kl(Exponential, Normal) +def _kl_exponential_normal(p, q): + var_normal = q.scale.pow(2) + rate_sqr = p.rate.pow(2) + t1 = 0.5 * torch.log(rate_sqr * var_normal * 2 * math.pi) + t2 = rate_sqr.reciprocal() + t3 = q.loc / p.rate + t4 = q.loc.pow(2) * 0.5 + return t1 - 1 + (t2 - t3 + t4) / var_normal + + +@register_kl(Gamma, Beta) +@register_kl(Gamma, ContinuousBernoulli) +@register_kl(Gamma, Pareto) +@register_kl(Gamma, Uniform) +def _kl_gamma_infinity(p, q): + return _infinite_like(p.concentration) + + +@register_kl(Gamma, Exponential) +def _kl_gamma_exponential(p, q): + return -p.entropy() - q.rate.log() + q.rate * p.concentration / p.rate + + +@register_kl(Gamma, Gumbel) +def _kl_gamma_gumbel(p, q): + beta_scale_prod = p.rate * q.scale + loc_scale_ratio = q.loc / q.scale + t1 = ( + (p.concentration - 1) * p.concentration.digamma() + - p.concentration.lgamma() + - p.concentration + ) + t2 = beta_scale_prod.log() + p.concentration / beta_scale_prod + t3 = ( + torch.exp(loc_scale_ratio) + * (1 + beta_scale_prod.reciprocal()).pow(-p.concentration) + - loc_scale_ratio + ) + return t1 + t2 + t3 + + +# TODO: Add Gamma-Laplace KL Divergence + + +@register_kl(Gamma, Normal) +def _kl_gamma_normal(p, q): + var_normal = q.scale.pow(2) + beta_sqr = p.rate.pow(2) + t1 = ( + 0.5 * torch.log(beta_sqr * var_normal * 2 * math.pi) + - p.concentration + - p.concentration.lgamma() + ) + t2 = 0.5 * (p.concentration.pow(2) + p.concentration) / beta_sqr + t3 = q.loc * p.concentration / p.rate + t4 = 0.5 * q.loc.pow(2) + return ( + t1 + + (p.concentration - 1) * p.concentration.digamma() + + (t2 - t3 + t4) / var_normal + ) + + +@register_kl(Gumbel, Beta) +@register_kl(Gumbel, ContinuousBernoulli) +@register_kl(Gumbel, Exponential) +@register_kl(Gumbel, Gamma) +@register_kl(Gumbel, Pareto) +@register_kl(Gumbel, Uniform) +def _kl_gumbel_infinity(p, q): + return _infinite_like(p.loc) + + +# TODO: Add Gumbel-Laplace KL Divergence + + +@register_kl(Gumbel, Normal) +def _kl_gumbel_normal(p, q): + param_ratio = p.scale / q.scale + t1 = (param_ratio / math.sqrt(2 * math.pi)).log() + t2 = (math.pi * param_ratio * 0.5).pow(2) / 3 + t3 = ((p.loc + p.scale * _euler_gamma - q.loc) / q.scale).pow(2) * 0.5 + return -t1 + t2 + t3 - (_euler_gamma + 1) + + +@register_kl(Laplace, Beta) +@register_kl(Laplace, ContinuousBernoulli) +@register_kl(Laplace, Exponential) +@register_kl(Laplace, Gamma) +@register_kl(Laplace, Pareto) +@register_kl(Laplace, Uniform) +def _kl_laplace_infinity(p, q): + return _infinite_like(p.loc) + + +@register_kl(Laplace, Normal) +def _kl_laplace_normal(p, q): + var_normal = q.scale.pow(2) + scale_sqr_var_ratio = p.scale.pow(2) / var_normal + t1 = 0.5 * torch.log(2 * scale_sqr_var_ratio / math.pi) + t2 = 0.5 * p.loc.pow(2) + t3 = p.loc * q.loc + t4 = 0.5 * q.loc.pow(2) + return -t1 + scale_sqr_var_ratio + (t2 - t3 + t4) / var_normal - 1 + + +@register_kl(Normal, Beta) +@register_kl(Normal, ContinuousBernoulli) +@register_kl(Normal, Exponential) +@register_kl(Normal, Gamma) +@register_kl(Normal, Pareto) +@register_kl(Normal, Uniform) +def _kl_normal_infinity(p, q): + return _infinite_like(p.loc) + + +@register_kl(Normal, Gumbel) +def _kl_normal_gumbel(p, q): + mean_scale_ratio = p.loc / q.scale + var_scale_sqr_ratio = (p.scale / q.scale).pow(2) + loc_scale_ratio = q.loc / q.scale + t1 = var_scale_sqr_ratio.log() * 0.5 + t2 = mean_scale_ratio - loc_scale_ratio + t3 = torch.exp(-mean_scale_ratio + 0.5 * var_scale_sqr_ratio + loc_scale_ratio) + return -t1 + t2 + t3 - (0.5 * (1 + math.log(2 * math.pi))) + + +@register_kl(Normal, Laplace) +def _kl_normal_laplace(p, q): + loc_diff = p.loc - q.loc + scale_ratio = p.scale / q.scale + loc_diff_scale_ratio = loc_diff / p.scale + t1 = torch.log(scale_ratio) + t2 = ( + math.sqrt(2 / math.pi) * p.scale * torch.exp(-0.5 * loc_diff_scale_ratio.pow(2)) + ) + t3 = loc_diff * torch.erf(math.sqrt(0.5) * loc_diff_scale_ratio) + return -t1 + (t2 + t3) / q.scale - (0.5 * (1 + math.log(0.5 * math.pi))) + + +@register_kl(Pareto, Beta) +@register_kl(Pareto, ContinuousBernoulli) +@register_kl(Pareto, Uniform) +def _kl_pareto_infinity(p, q): + return _infinite_like(p.scale) + + +@register_kl(Pareto, Exponential) +def _kl_pareto_exponential(p, q): + scale_rate_prod = p.scale * q.rate + t1 = (p.alpha / scale_rate_prod).log() + t2 = p.alpha.reciprocal() + t3 = p.alpha * scale_rate_prod / (p.alpha - 1) + result = t1 - t2 + t3 - 1 + result[p.alpha <= 1] = inf + return result + + +@register_kl(Pareto, Gamma) +def _kl_pareto_gamma(p, q): + common_term = p.scale.log() + p.alpha.reciprocal() + t1 = p.alpha.log() - common_term + t2 = q.concentration.lgamma() - q.concentration * q.rate.log() + t3 = (1 - q.concentration) * common_term + t4 = q.rate * p.alpha * p.scale / (p.alpha - 1) + result = t1 + t2 + t3 + t4 - 1 + result[p.alpha <= 1] = inf + return result + + +# TODO: Add Pareto-Laplace KL Divergence + + +@register_kl(Pareto, Normal) +def _kl_pareto_normal(p, q): + var_normal = 2 * q.scale.pow(2) + common_term = p.scale / (p.alpha - 1) + t1 = (math.sqrt(2 * math.pi) * q.scale * p.alpha / p.scale).log() + t2 = p.alpha.reciprocal() + t3 = p.alpha * common_term.pow(2) / (p.alpha - 2) + t4 = (p.alpha * common_term - q.loc).pow(2) + result = t1 - t2 + (t3 + t4) / var_normal - 1 + result[p.alpha <= 2] = inf + return result + + +@register_kl(Poisson, Bernoulli) +@register_kl(Poisson, Binomial) +def _kl_poisson_infinity(p, q): + return _infinite_like(p.rate) + + +@register_kl(Uniform, Beta) +def _kl_uniform_beta(p, q): + common_term = p.high - p.low + t1 = torch.log(common_term) + t2 = ( + (q.concentration1 - 1) + * (_x_log_x(p.high) - _x_log_x(p.low) - common_term) + / common_term + ) + t3 = ( + (q.concentration0 - 1) + * (_x_log_x(1 - p.high) - _x_log_x(1 - p.low) + common_term) + / common_term + ) + t4 = ( + q.concentration1.lgamma() + + q.concentration0.lgamma() + - (q.concentration1 + q.concentration0).lgamma() + ) + result = t3 + t4 - t1 - t2 + result[(p.high > q.support.upper_bound) | (p.low < q.support.lower_bound)] = inf + return result + + +@register_kl(Uniform, ContinuousBernoulli) +def _kl_uniform_continuous_bernoulli(p, q): + result = ( + -p.entropy() + - p.mean * q.logits + - torch.log1p(-q.probs) + - q._cont_bern_log_norm() + ) + return torch.where( + torch.max( + torch.ge(p.high, q.support.upper_bound), + torch.le(p.low, q.support.lower_bound), + ), + torch.ones_like(result) * inf, + result, + ) + + +@register_kl(Uniform, Exponential) +def _kl_uniform_exponetial(p, q): + result = q.rate * (p.high + p.low) / 2 - ((p.high - p.low) * q.rate).log() + result[p.low < q.support.lower_bound] = inf + return result + + +@register_kl(Uniform, Gamma) +def _kl_uniform_gamma(p, q): + common_term = p.high - p.low + t1 = common_term.log() + t2 = q.concentration.lgamma() - q.concentration * q.rate.log() + t3 = ( + (1 - q.concentration) + * (_x_log_x(p.high) - _x_log_x(p.low) - common_term) + / common_term + ) + t4 = q.rate * (p.high + p.low) / 2 + result = -t1 + t2 + t3 + t4 + result[p.low < q.support.lower_bound] = inf + return result + + +@register_kl(Uniform, Gumbel) +def _kl_uniform_gumbel(p, q): + common_term = q.scale / (p.high - p.low) + high_loc_diff = (p.high - q.loc) / q.scale + low_loc_diff = (p.low - q.loc) / q.scale + t1 = common_term.log() + 0.5 * (high_loc_diff + low_loc_diff) + t2 = common_term * (torch.exp(-high_loc_diff) - torch.exp(-low_loc_diff)) + return t1 - t2 + + +# TODO: Uniform-Laplace KL Divergence + + +@register_kl(Uniform, Normal) +def _kl_uniform_normal(p, q): + common_term = p.high - p.low + t1 = (math.sqrt(math.pi * 2) * q.scale / common_term).log() + t2 = (common_term).pow(2) / 12 + t3 = ((p.high + p.low - 2 * q.loc) / 2).pow(2) + return t1 + 0.5 * (t2 + t3) / q.scale.pow(2) + + +@register_kl(Uniform, Pareto) +def _kl_uniform_pareto(p, q): + support_uniform = p.high - p.low + t1 = (q.alpha * q.scale.pow(q.alpha) * (support_uniform)).log() + t2 = (_x_log_x(p.high) - _x_log_x(p.low) - support_uniform) / support_uniform + result = t2 * (q.alpha + 1) - t1 + result[p.low < q.support.lower_bound] = inf + return result + + +@register_kl(Independent, Independent) +def _kl_independent_independent(p, q): + if p.reinterpreted_batch_ndims != q.reinterpreted_batch_ndims: + raise NotImplementedError + result = kl_divergence(p.base_dist, q.base_dist) + return _sum_rightmost(result, p.reinterpreted_batch_ndims) + + +@register_kl(Cauchy, Cauchy) +def _kl_cauchy_cauchy(p, q): + # From https://arxiv.org/abs/1905.10965 + t1 = ((p.scale + q.scale).pow(2) + (p.loc - q.loc).pow(2)).log() + t2 = (4 * p.scale * q.scale).log() + return t1 - t2 + + +def _add_kl_info(): + """Appends a list of implemented KL functions to the doc for kl_divergence.""" + rows = [ + "KL divergence is currently implemented for the following distribution pairs:" + ] + for p, q in sorted( + _KL_REGISTRY, key=lambda p_q: (p_q[0].__name__, p_q[1].__name__) + ): + rows.append( + f"* :class:`~torch.distributions.{p.__name__}` and :class:`~torch.distributions.{q.__name__}`" + ) + kl_info = "\n\t".join(rows) + if kl_divergence.__doc__: + # pyrefly: ignore [missing-attribute] + kl_divergence.__doc__ += kl_info diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/kumaraswamy.py b/.venv/lib/python3.12/site-packages/torch/distributions/kumaraswamy.py new file mode 100644 index 0000000000000000000000000000000000000000..9ad6620968a4e96ef3fa528c299804c70feb80d3 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/kumaraswamy.py @@ -0,0 +1,107 @@ +# mypy: allow-untyped-defs + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AffineTransform, PowerTransform +from torch.distributions.uniform import Uniform +from torch.distributions.utils import broadcast_all, euler_constant + + +__all__ = ["Kumaraswamy"] + + +def _moments(a, b, n): + """ + Computes nth moment of Kumaraswamy using torch.lgamma + """ + arg1 = 1 + n / a + log_value = torch.lgamma(arg1) + torch.lgamma(b) - torch.lgamma(arg1 + b) + return b * torch.exp(log_value) + + +class Kumaraswamy(TransformedDistribution): + r""" + Samples from a Kumaraswamy distribution. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Kumaraswamy(torch.tensor([1.0]), torch.tensor([1.0])) + >>> m.sample() # sample from a Kumaraswamy distribution with concentration alpha=1 and beta=1 + tensor([ 0.1729]) + + Args: + concentration1 (float or Tensor): 1st concentration parameter of the distribution + (often referred to as alpha) + concentration0 (float or Tensor): 2nd concentration parameter of the distribution + (often referred to as beta) + """ + + arg_constraints = { + "concentration1": constraints.positive, + "concentration0": constraints.positive, + } + # pyrefly: ignore [bad-override] + support = constraints.unit_interval + has_rsample = True + + def __init__( + self, + concentration1: Tensor | float, + concentration0: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.concentration1, self.concentration0 = broadcast_all( + concentration1, concentration0 + ) + base_dist = Uniform( + torch.full_like(self.concentration0, 0), + torch.full_like(self.concentration0, 1), + validate_args=validate_args, + ) + transforms = [ + PowerTransform(exponent=self.concentration0.reciprocal()), + AffineTransform(loc=1.0, scale=-1.0), + PowerTransform(exponent=self.concentration1.reciprocal()), + ] + # pyrefly: ignore [bad-argument-type] + super().__init__(base_dist, transforms, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Kumaraswamy, _instance) + new.concentration1 = self.concentration1.expand(batch_shape) + new.concentration0 = self.concentration0.expand(batch_shape) + return super().expand(batch_shape, _instance=new) + + @property + def mean(self) -> Tensor: + return _moments(self.concentration1, self.concentration0, 1) + + @property + def mode(self) -> Tensor: + # Evaluate in log-space for numerical stability. + log_mode = ( + self.concentration0.reciprocal() * (-self.concentration0).log1p() + - (-self.concentration0 * self.concentration1).log1p() + ) + log_mode[(self.concentration0 < 1) | (self.concentration1 < 1)] = nan + return log_mode.exp() + + @property + def variance(self) -> Tensor: + return _moments(self.concentration1, self.concentration0, 2) - torch.pow( + self.mean, 2 + ) + + def entropy(self): + t1 = 1 - self.concentration1.reciprocal() + t0 = 1 - self.concentration0.reciprocal() + H0 = torch.digamma(self.concentration0 + 1) + euler_constant + return ( + t0 + + t1 * H0 + - torch.log(self.concentration1) + - torch.log(self.concentration0) + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/laplace.py b/.venv/lib/python3.12/site-packages/torch/distributions/laplace.py new file mode 100644 index 0000000000000000000000000000000000000000..791a29f03ba5b79f84d6c2c674b68d1da67692b9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/laplace.py @@ -0,0 +1,104 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Laplace"] + + +class Laplace(Distribution): + r""" + Creates a Laplace distribution parameterized by :attr:`loc` and :attr:`scale`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Laplace(torch.tensor([0.0]), torch.tensor([1.0])) + >>> m.sample() # Laplace distributed with loc=0, scale=1 + tensor([ 0.1046]) + + Args: + loc (float or Tensor): mean of the distribution + scale (float or Tensor): scale of the distribution + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + support = constraints.real + has_rsample = True + + @property + def mean(self) -> Tensor: + return self.loc + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def variance(self) -> Tensor: + return 2 * self.scale.pow(2) + + @property + def stddev(self) -> Tensor: + return (2**0.5) * self.scale + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.loc, self.scale = broadcast_all(loc, scale) + if isinstance(loc, _Number) and isinstance(scale, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.loc.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Laplace, _instance) + batch_shape = torch.Size(batch_shape) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + super(Laplace, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + finfo = torch.finfo(self.loc.dtype) + if torch._C._get_tracing_state(): + # [JIT WORKAROUND] lack of support for .uniform_() + u = torch.rand(shape, dtype=self.loc.dtype, device=self.loc.device) * 2 - 1 + return self.loc - self.scale * u.sign() * torch.log1p( + -u.abs().clamp(min=finfo.tiny) + ) + u = self.loc.new(shape).uniform_(finfo.eps - 1, 1) + # TODO: If we ever implement tensor.nextafter, below is what we want ideally. + # u = self.loc.new(shape).uniform_(self.loc.nextafter(-.5, 0), .5) + return self.loc - self.scale * u.sign() * torch.log1p(-u.abs()) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + return -torch.log(2 * self.scale) - torch.abs(value - self.loc) / self.scale + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return 0.5 - 0.5 * (value - self.loc).sign() * torch.expm1( + -(value - self.loc).abs() / self.scale + ) + + def icdf(self, value): + term = value - 0.5 + return self.loc - self.scale * (term).sign() * torch.log1p(-2 * term.abs()) + + def entropy(self): + return 1 + torch.log(2 * self.scale) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/lkj_cholesky.py b/.venv/lib/python3.12/site-packages/torch/distributions/lkj_cholesky.py new file mode 100644 index 0000000000000000000000000000000000000000..d25fa35c62ac48c63eb5597930c1ca71e5b8932e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/lkj_cholesky.py @@ -0,0 +1,152 @@ +# mypy: allow-untyped-defs +""" +This closely follows the implementation in NumPyro (https://github.com/pyro-ppl/numpyro). + +Original copyright notice: + +# Copyright: Contributors to the Pyro project. +# SPDX-License-Identifier: Apache-2.0 +""" + +import math + +import torch +from torch import Tensor +from torch.distributions import Beta, constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all + + +__all__ = ["LKJCholesky"] + + +class LKJCholesky(Distribution): + r""" + LKJ distribution for lower Cholesky factor of correlation matrices. + The distribution is controlled by ``concentration`` parameter :math:`\eta` + to make the probability of the correlation matrix :math:`M` generated from + a Cholesky factor proportional to :math:`\det(M)^{\eta - 1}`. Because of that, + when ``concentration == 1``, we have a uniform distribution over Cholesky + factors of correlation matrices:: + + L ~ LKJCholesky(dim, concentration) + X = L @ L' ~ LKJCorr(dim, concentration) + + Note that this distribution samples the + Cholesky factor of correlation matrices and not the correlation matrices + themselves and thereby differs slightly from the derivations in [1] for + the `LKJCorr` distribution. For sampling, this uses the Onion method from + [1] Section 3. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> l = LKJCholesky(3, 0.5) + >>> l.sample() # l @ l.T is a sample of a correlation 3x3 matrix + tensor([[ 1.0000, 0.0000, 0.0000], + [ 0.3516, 0.9361, 0.0000], + [-0.1899, 0.4748, 0.8593]]) + + Args: + dimension (dim): dimension of the matrices + concentration (float or Tensor): concentration/shape parameter of the + distribution (often referred to as eta) + + **References** + + [1] `Generating random correlation matrices based on vines and extended onion method` (2009), + Daniel Lewandowski, Dorota Kurowicka, Harry Joe. + Journal of Multivariate Analysis. 100. 10.1016/j.jmva.2009.04.008 + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"concentration": constraints.positive} + support = constraints.corr_cholesky + + def __init__( + self, + dim: int, + concentration: Tensor | float = 1.0, + validate_args: bool | None = None, + ) -> None: + if dim < 2: + raise ValueError( + f"Expected dim to be an integer greater than or equal to 2. Found dim={dim}." + ) + self.dim = dim + (self.concentration,) = broadcast_all(concentration) + batch_shape = self.concentration.size() + event_shape = torch.Size((dim, dim)) + # This is used to draw vectorized samples from the beta distribution in Sec. 3.2 of [1]. + marginal_conc = self.concentration + 0.5 * (self.dim - 2) + offset = torch.arange( + self.dim - 1, + dtype=self.concentration.dtype, + device=self.concentration.device, + ) + offset = torch.cat([offset.new_zeros((1,)), offset]) + beta_conc1 = offset + 0.5 + beta_conc0 = marginal_conc.unsqueeze(-1) - 0.5 * offset + self._beta = Beta(beta_conc1, beta_conc0) + super().__init__(batch_shape, event_shape, validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(LKJCholesky, _instance) + batch_shape = torch.Size(batch_shape) + new.dim = self.dim + new.concentration = self.concentration.expand(batch_shape) + new._beta = self._beta.expand(batch_shape + (self.dim,)) + super(LKJCholesky, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + def sample(self, sample_shape=torch.Size()): + # This uses the Onion method, but there are a few differences from [1] Sec. 3.2: + # - This vectorizes the for loop and also works for heterogeneous eta. + # - Same algorithm generalizes to n=1. + # - The procedure is simplified since we are sampling the cholesky factor of + # the correlation matrix instead of the correlation matrix itself. As such, + # we only need to generate `w`. + y = self._beta.sample(sample_shape).unsqueeze(-1) + u_normal = torch.randn( + self._extended_shape(sample_shape), dtype=y.dtype, device=y.device + ).tril(-1) + u_hypersphere = u_normal / u_normal.norm(dim=-1, keepdim=True) + # Replace NaNs in first row + u_hypersphere[..., 0, :].fill_(0.0) + w = torch.sqrt(y) * u_hypersphere + # Fill diagonal elements; clamp for numerical stability + eps = torch.finfo(w.dtype).tiny + diag_elems = torch.clamp(1 - torch.sum(w**2, dim=-1), min=eps).sqrt() + w += torch.diag_embed(diag_elems) + return w + + def log_prob(self, value): + # See: https://mc-stan.org/docs/2_25/functions-reference/cholesky-lkj-correlation-distribution.html + # The probability of a correlation matrix is proportional to + # determinant ** (concentration - 1) = prod(L_ii ^ 2(concentration - 1)) + # Additionally, the Jacobian of the transformation from Cholesky factor to + # correlation matrix is: + # prod(L_ii ^ (D - i)) + # So the probability of a Cholesky factor is proportional to + # prod(L_ii ^ (2 * concentration - 2 + D - i)) = prod(L_ii ^ order_i) + # with order_i = 2 * concentration - 2 + D - i + if self._validate_args: + self._validate_sample(value) + diag_elems = value.diagonal(dim1=-1, dim2=-2)[..., 1:] + order = torch.arange(2, self.dim + 1, device=self.concentration.device) + order = 2 * (self.concentration - 1).unsqueeze(-1) + self.dim - order + unnormalized_log_pdf = torch.sum(order * diag_elems.log(), dim=-1) + # Compute normalization constant (page 1999 of [1]) + dm1 = self.dim - 1 + alpha = self.concentration + 0.5 * dm1 + denominator = torch.lgamma(alpha) * dm1 + numerator = torch.mvlgamma(alpha - 0.5, dm1) + # pi_constant in [1] is D * (D - 1) / 4 * log(pi) + # pi_constant in multigammaln is (D - 1) * (D - 2) / 4 * log(pi) + # hence, we need to add a pi_constant = (D - 1) * log(pi) / 2 + pi_constant = 0.5 * dm1 * math.log(math.pi) + normalize_term = pi_constant + numerator - denominator + return unnormalized_log_pdf - normalize_term diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/log_normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/log_normal.py new file mode 100644 index 0000000000000000000000000000000000000000..259875d0bb48c4e8eb59f1c1ff90f8f62ff8c406 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/log_normal.py @@ -0,0 +1,75 @@ +# mypy: allow-untyped-defs + +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.normal import Normal +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import ExpTransform + + +__all__ = ["LogNormal"] + + +class LogNormal(TransformedDistribution): + r""" + Creates a log-normal distribution parameterized by + :attr:`loc` and :attr:`scale` where:: + + X ~ Normal(loc, scale) + Y = exp(X) ~ LogNormal(loc, scale) + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = LogNormal(torch.tensor([0.0]), torch.tensor([1.0])) + >>> m.sample() # log-normal distributed with mean=0 and stddev=1 + tensor([ 0.1046]) + + Args: + loc (float or Tensor): mean of log of distribution + scale (float or Tensor): standard deviation of log of the distribution + """ + + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + # pyrefly: ignore [bad-override] + support = constraints.positive + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: Normal + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + base_dist = Normal(loc, scale, validate_args=validate_args) + super().__init__(base_dist, ExpTransform(), validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(LogNormal, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def loc(self) -> Tensor: + return self.base_dist.loc + + @property + def scale(self) -> Tensor: + return self.base_dist.scale + + @property + def mean(self) -> Tensor: + return (self.loc + self.scale.pow(2) / 2).exp() + + @property + def mode(self) -> Tensor: + return (self.loc - self.scale.square()).exp() + + @property + def variance(self) -> Tensor: + scale_sq = self.scale.pow(2) + return scale_sq.expm1() * (2 * self.loc + scale_sq).exp() + + def entropy(self): + return self.base_dist.entropy() + self.loc diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/logistic_normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/logistic_normal.py new file mode 100644 index 0000000000000000000000000000000000000000..117c83ae07f30ae1e0621ae4243ea3812ddeec4c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/logistic_normal.py @@ -0,0 +1,67 @@ +# mypy: allow-untyped-defs + +from torch import Tensor +from torch.distributions import constraints, Independent +from torch.distributions.normal import Normal +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import StickBreakingTransform + + +__all__ = ["LogisticNormal"] + + +class LogisticNormal(TransformedDistribution): + r""" + Creates a logistic-normal distribution parameterized by :attr:`loc` and :attr:`scale` + that define the base `Normal` distribution transformed with the + `StickBreakingTransform` such that:: + + X ~ LogisticNormal(loc, scale) + Y = log(X / (1 - X.cumsum(-1)))[..., :-1] ~ Normal(loc, scale) + + Args: + loc (float or Tensor): mean of the base distribution + scale (float or Tensor): standard deviation of the base distribution + + Example:: + + >>> # logistic-normal distributed with mean=(0, 0, 0) and stddev=(1, 1, 1) + >>> # of the base Normal distribution + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = LogisticNormal(torch.tensor([0.0] * 3), torch.tensor([1.0] * 3)) + >>> m.sample() + tensor([ 0.7653, 0.0341, 0.0579, 0.1427]) + + """ + + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + # pyrefly: ignore [bad-override] + support = constraints.simplex + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: Independent[Normal] + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + base_dist = Normal(loc, scale, validate_args=validate_args) + if not base_dist.batch_shape: + base_dist = base_dist.expand([1]) + super().__init__( + base_dist, StickBreakingTransform(), validate_args=validate_args + ) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(LogisticNormal, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def loc(self) -> Tensor: + return self.base_dist.base_dist.loc + + @property + def scale(self) -> Tensor: + return self.base_dist.base_dist.scale diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/lowrank_multivariate_normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/lowrank_multivariate_normal.py new file mode 100644 index 0000000000000000000000000000000000000000..75ef4ef5dc7dc637ca0e07dbdc159200cabbe4b0 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/lowrank_multivariate_normal.py @@ -0,0 +1,252 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.multivariate_normal import _batch_mahalanobis, _batch_mv +from torch.distributions.utils import _standard_normal, lazy_property +from torch.types import _size + + +__all__ = ["LowRankMultivariateNormal"] + + +def _batch_capacitance_tril(W, D): + r""" + Computes Cholesky of :math:`I + W.T @ inv(D) @ W` for a batch of matrices :math:`W` + and a batch of vectors :math:`D`. + """ + m = W.size(-1) + Wt_Dinv = W.mT / D.unsqueeze(-2) + K = torch.matmul(Wt_Dinv, W).contiguous() + K.view(-1, m * m)[:, :: m + 1] += 1 # add identity matrix to K + return torch.linalg.cholesky(K) + + +def _batch_lowrank_logdet(W, D, capacitance_tril): + r""" + Uses "matrix determinant lemma":: + log|W @ W.T + D| = log|C| + log|D|, + where :math:`C` is the capacitance matrix :math:`I + W.T @ inv(D) @ W`, to compute + the log determinant. + """ + return 2 * capacitance_tril.diagonal(dim1=-2, dim2=-1).log().sum(-1) + D.log().sum( + -1 + ) + + +def _batch_lowrank_mahalanobis(W, D, x, capacitance_tril): + r""" + Uses "Woodbury matrix identity":: + inv(W @ W.T + D) = inv(D) - inv(D) @ W @ inv(C) @ W.T @ inv(D), + where :math:`C` is the capacitance matrix :math:`I + W.T @ inv(D) @ W`, to compute the squared + Mahalanobis distance :math:`x.T @ inv(W @ W.T + D) @ x`. + """ + Wt_Dinv = W.mT / D.unsqueeze(-2) + Wt_Dinv_x = _batch_mv(Wt_Dinv, x) + mahalanobis_term1 = (x.pow(2) / D).sum(-1) + mahalanobis_term2 = _batch_mahalanobis(capacitance_tril, Wt_Dinv_x) + return mahalanobis_term1 - mahalanobis_term2 + + +class LowRankMultivariateNormal(Distribution): + r""" + Creates a multivariate normal distribution with covariance matrix having a low-rank form + parameterized by :attr:`cov_factor` and :attr:`cov_diag`:: + + covariance_matrix = cov_factor @ cov_factor.T + cov_diag + + Example: + >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_LAPACK) + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = LowRankMultivariateNormal( + ... torch.zeros(2), torch.tensor([[1.0], [0.0]]), torch.ones(2) + ... ) + >>> m.sample() # normally distributed with mean=`[0,0]`, cov_factor=`[[1],[0]]`, cov_diag=`[1,1]` + tensor([-0.2102, -0.5429]) + + Args: + loc (Tensor): mean of the distribution with shape `batch_shape + event_shape` + cov_factor (Tensor): factor part of low-rank form of covariance matrix with shape + `batch_shape + event_shape + (rank,)` + cov_diag (Tensor): diagonal part of low-rank form of covariance matrix with shape + `batch_shape + event_shape` + + Note: + The computation for determinant and inverse of covariance matrix is avoided when + `cov_factor.shape[1] << cov_factor.shape[0]` thanks to `Woodbury matrix identity + `_ and + `matrix determinant lemma `_. + Thanks to these formulas, we just need to compute the determinant and inverse of + the small size "capacitance" matrix:: + + capacitance = I + cov_factor.T @ inv(cov_diag) @ cov_factor + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "loc": constraints.real_vector, + "cov_factor": constraints.independent(constraints.real, 2), + "cov_diag": constraints.independent(constraints.positive, 1), + } + support = constraints.real_vector + has_rsample = True + + def __init__( + self, + loc: Tensor, + cov_factor: Tensor, + cov_diag: Tensor, + validate_args: bool | None = None, + ) -> None: + if loc.dim() < 1: + raise ValueError("loc must be at least one-dimensional.") + event_shape = loc.shape[-1:] + if cov_factor.dim() < 2: + raise ValueError( + "cov_factor must be at least two-dimensional, " + "with optional leading batch dimensions" + ) + if cov_factor.shape[-2:-1] != event_shape: + raise ValueError( + f"cov_factor must be a batch of matrices with shape {event_shape[0]} x m" + ) + if cov_diag.shape[-1:] != event_shape: + raise ValueError( + f"cov_diag must be a batch of vectors with shape {event_shape}" + ) + + loc_ = loc.unsqueeze(-1) + cov_diag_ = cov_diag.unsqueeze(-1) + try: + loc_, self.cov_factor, cov_diag_ = torch.broadcast_tensors( + loc_, cov_factor, cov_diag_ + ) + except RuntimeError as e: + raise ValueError( + f"Incompatible batch shapes: loc {loc.shape}, cov_factor {cov_factor.shape}, cov_diag {cov_diag.shape}" + ) from e + self.loc = loc_[..., 0] + self.cov_diag = cov_diag_[..., 0] + batch_shape = self.loc.shape[:-1] + + self._unbroadcasted_cov_factor = cov_factor + self._unbroadcasted_cov_diag = cov_diag + self._capacitance_tril = _batch_capacitance_tril(cov_factor, cov_diag) + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(LowRankMultivariateNormal, _instance) + batch_shape = torch.Size(batch_shape) + loc_shape = batch_shape + self.event_shape + new.loc = self.loc.expand(loc_shape) + new.cov_diag = self.cov_diag.expand(loc_shape) + new.cov_factor = self.cov_factor.expand(loc_shape + self.cov_factor.shape[-1:]) + new._unbroadcasted_cov_factor = self._unbroadcasted_cov_factor + new._unbroadcasted_cov_diag = self._unbroadcasted_cov_diag + new._capacitance_tril = self._capacitance_tril + super(LowRankMultivariateNormal, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + return self.loc + + @property + def mode(self) -> Tensor: + return self.loc + + @lazy_property + def variance(self) -> Tensor: # type: ignore[override] + return ( + self._unbroadcasted_cov_factor.pow(2).sum(-1) + self._unbroadcasted_cov_diag + ).expand(self._batch_shape + self._event_shape) + + @lazy_property + def scale_tril(self) -> Tensor: + # The following identity is used to increase the numerical computation stability + # for Cholesky decomposition (see http://www.gaussianprocess.org/gpml/, Section 3.4.3): + # W @ W.T + D = D1/2 @ (I + D-1/2 @ W @ W.T @ D-1/2) @ D1/2 + # The matrix "I + D-1/2 @ W @ W.T @ D-1/2" has eigenvalues bounded from below by 1, + # hence it is well-conditioned and safe to take Cholesky decomposition. + n = self._event_shape[0] + cov_diag_sqrt_unsqueeze = self._unbroadcasted_cov_diag.sqrt().unsqueeze(-1) + Dinvsqrt_W = self._unbroadcasted_cov_factor / cov_diag_sqrt_unsqueeze + K = torch.matmul(Dinvsqrt_W, Dinvsqrt_W.mT).contiguous() + K.view(-1, n * n)[:, :: n + 1] += 1 # add identity matrix to K + scale_tril = cov_diag_sqrt_unsqueeze * torch.linalg.cholesky(K) + return scale_tril.expand( + self._batch_shape + self._event_shape + self._event_shape + ) + + @lazy_property + def covariance_matrix(self) -> Tensor: + covariance_matrix = torch.matmul( + self._unbroadcasted_cov_factor, self._unbroadcasted_cov_factor.mT + ) + torch.diag_embed(self._unbroadcasted_cov_diag) + return covariance_matrix.expand( + self._batch_shape + self._event_shape + self._event_shape + ) + + @lazy_property + def precision_matrix(self) -> Tensor: + # We use "Woodbury matrix identity" to take advantage of low rank form:: + # inv(W @ W.T + D) = inv(D) - inv(D) @ W @ inv(C) @ W.T @ inv(D) + # where :math:`C` is the capacitance matrix. + Wt_Dinv = ( + self._unbroadcasted_cov_factor.mT + / self._unbroadcasted_cov_diag.unsqueeze(-2) + ) + A = torch.linalg.solve_triangular(self._capacitance_tril, Wt_Dinv, upper=False) + precision_matrix = ( + torch.diag_embed(self._unbroadcasted_cov_diag.reciprocal()) - A.mT @ A + ) + return precision_matrix.expand( + self._batch_shape + self._event_shape + self._event_shape + ) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + W_shape = shape[:-1] + self.cov_factor.shape[-1:] + eps_W = _standard_normal(W_shape, dtype=self.loc.dtype, device=self.loc.device) + eps_D = _standard_normal(shape, dtype=self.loc.dtype, device=self.loc.device) + return ( + self.loc + + _batch_mv(self._unbroadcasted_cov_factor, eps_W) + + self._unbroadcasted_cov_diag.sqrt() * eps_D + ) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + diff = value - self.loc + M = _batch_lowrank_mahalanobis( + self._unbroadcasted_cov_factor, + self._unbroadcasted_cov_diag, + diff, + self._capacitance_tril, + ) + log_det = _batch_lowrank_logdet( + self._unbroadcasted_cov_factor, + self._unbroadcasted_cov_diag, + self._capacitance_tril, + ) + return -0.5 * (self._event_shape[0] * math.log(2 * math.pi) + log_det + M) + + def entropy(self): + log_det = _batch_lowrank_logdet( + self._unbroadcasted_cov_factor, + self._unbroadcasted_cov_diag, + self._capacitance_tril, + ) + H = 0.5 * (self._event_shape[0] * (1.0 + math.log(2 * math.pi)) + log_det) + if len(self._batch_shape) == 0: + return H + else: + return H.expand(self._batch_shape) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/mixture_same_family.py b/.venv/lib/python3.12/site-packages/torch/distributions/mixture_same_family.py new file mode 100644 index 0000000000000000000000000000000000000000..d3d1d5039c56cdab81b349ed3358efcb3980504e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/mixture_same_family.py @@ -0,0 +1,223 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import Categorical, constraints +from torch.distributions.constraints import MixtureSameFamilyConstraint +from torch.distributions.distribution import Distribution + + +__all__ = ["MixtureSameFamily"] + + +class MixtureSameFamily(Distribution): + r""" + The `MixtureSameFamily` distribution implements a (batch of) mixture + distribution where all components are from different parameterizations of + the same distribution type. It is parameterized by a `Categorical` + "selecting distribution" (over `k` components) and a component + distribution, i.e., a `Distribution` with a rightmost batch shape + (equal to `[k]`) which indexes each (batch of) component. + + Examples:: + + >>> # xdoctest: +SKIP("undefined vars") + >>> # Construct Gaussian Mixture Model in 1D consisting of 5 equally + >>> # weighted normal distributions + >>> mix = D.Categorical(torch.ones(5,)) + >>> comp = D.Normal(torch.randn(5,), torch.rand(5,)) + >>> gmm = MixtureSameFamily(mix, comp) + + >>> # Construct Gaussian Mixture Model in 2D consisting of 5 equally + >>> # weighted bivariate normal distributions + >>> mix = D.Categorical(torch.ones(5,)) + >>> comp = D.Independent(D.Normal( + ... torch.randn(5,2), torch.rand(5,2)), 1) + >>> gmm = MixtureSameFamily(mix, comp) + + >>> # Construct a batch of 3 Gaussian Mixture Models in 2D each + >>> # consisting of 5 random weighted bivariate normal distributions + >>> mix = D.Categorical(torch.rand(3,5)) + >>> comp = D.Independent(D.Normal( + ... torch.randn(3,5,2), torch.rand(3,5,2)), 1) + >>> gmm = MixtureSameFamily(mix, comp) + + Args: + mixture_distribution: `torch.distributions.Categorical`-like + instance. Manages the probability of selecting components. + The number of categories must match the rightmost batch + dimension of the `component_distribution`. Must have either + scalar `batch_shape` or `batch_shape` matching + `component_distribution.batch_shape[:-1]` + component_distribution: `torch.distributions.Distribution`-like + instance. Right-most batch dimension indexes component. + """ + + arg_constraints: dict[str, constraints.Constraint] = {} + has_rsample = False + + def __init__( + self, + mixture_distribution: Categorical, + component_distribution: Distribution, + validate_args: bool | None = None, + ) -> None: + self._mixture_distribution = mixture_distribution + self._component_distribution = component_distribution + + if not isinstance(self._mixture_distribution, Categorical): + raise ValueError( + " The Mixture distribution needs to be an " + " instance of torch.distributions.Categorical" + ) + + if not isinstance(self._component_distribution, Distribution): + raise ValueError( + "The Component distribution needs to be an " + "instance of torch.distributions.Distribution" + ) + + # Check that batch size matches + mdbs = self._mixture_distribution.batch_shape + cdbs = self._component_distribution.batch_shape[:-1] + for size1, size2 in zip(reversed(mdbs), reversed(cdbs)): + if size1 != 1 and size2 != 1 and size1 != size2: + raise ValueError( + f"`mixture_distribution.batch_shape` ({mdbs}) is not " + "compatible with `component_distribution." + f"batch_shape`({cdbs})" + ) + + # Check that the number of mixture components matches + km = self._mixture_distribution.logits.shape[-1] + kc = self._component_distribution.batch_shape[-1] + if km is not None and kc is not None and km != kc: + raise ValueError( + f"`mixture_distribution component` ({km}) does not" + " equal `component_distribution.batch_shape[-1]`" + f" ({kc})" + ) + self._num_component = km + + event_shape = self._component_distribution.event_shape + self._event_ndims = len(event_shape) + super().__init__( + # pyrefly: ignore [bad-argument-type] + batch_shape=cdbs, + event_shape=event_shape, + validate_args=validate_args, + ) + + def expand(self, batch_shape, _instance=None): + batch_shape = torch.Size(batch_shape) + batch_shape_comp = batch_shape + (self._num_component,) + new = self._get_checked_instance(MixtureSameFamily, _instance) + new._component_distribution = self._component_distribution.expand( + batch_shape_comp + ) + new._mixture_distribution = self._mixture_distribution.expand(batch_shape) + new._num_component = self._num_component + new._event_ndims = self._event_ndims + event_shape = new._component_distribution.event_shape + super(MixtureSameFamily, new).__init__( + batch_shape=batch_shape, event_shape=event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def support(self): + return MixtureSameFamilyConstraint(self._component_distribution.support) + + @property + def mixture_distribution(self) -> Categorical: + return self._mixture_distribution + + @property + def component_distribution(self) -> Distribution: + return self._component_distribution + + @property + def mean(self) -> Tensor: + probs = self._pad_mixture_dimensions(self.mixture_distribution.probs) + return torch.sum( + probs * self.component_distribution.mean, dim=-1 - self._event_ndims + ) # [B, E] + + @property + def variance(self) -> Tensor: + # Law of total variance: Var(Y) = E[Var(Y|X)] + Var(E[Y|X]) + probs = self._pad_mixture_dimensions(self.mixture_distribution.probs) + mean_cond_var = torch.sum( + probs * self.component_distribution.variance, dim=-1 - self._event_ndims + ) + var_cond_mean = torch.sum( + probs * (self.component_distribution.mean - self._pad(self.mean)).pow(2.0), + dim=-1 - self._event_ndims, + ) + return mean_cond_var + var_cond_mean + + def cdf(self, x): + x = self._pad(x) + cdf_x = self.component_distribution.cdf(x) + mix_prob = self.mixture_distribution.probs + + return torch.sum(cdf_x * mix_prob, dim=-1) + + def log_prob(self, x): + if self._validate_args: + self._validate_sample(x) + x = self._pad(x) + log_prob_x = self.component_distribution.log_prob(x) # [S, B, k] + log_mix_prob = torch.log_softmax( + self.mixture_distribution.logits, dim=-1 + ) # [B, k] + return torch.logsumexp(log_prob_x + log_mix_prob, dim=-1) # [S, B] + + def sample(self, sample_shape=torch.Size()): + with torch.no_grad(): + sample_len = len(sample_shape) + batch_len = len(self.batch_shape) + gather_dim = sample_len + batch_len + es = self.event_shape + + # mixture samples [n, B] + mix_sample = self.mixture_distribution.sample(sample_shape) + mix_shape = mix_sample.shape + + # component samples [n, B, k, E] + comp_samples = self.component_distribution.sample(sample_shape) + + # Gather along the k dimension + mix_sample_r = mix_sample.reshape( + mix_shape + torch.Size([1] * (len(es) + 1)) + ) + mix_sample_r = mix_sample_r.repeat( + torch.Size([1] * len(mix_shape)) + torch.Size([1]) + es + ) + + samples = torch.gather(comp_samples, gather_dim, mix_sample_r) + return samples.squeeze(gather_dim) + + def _pad(self, x): + return x.unsqueeze(-1 - self._event_ndims) + + def _pad_mixture_dimensions(self, x): + dist_batch_ndims = len(self.batch_shape) + cat_batch_ndims = len(self.mixture_distribution.batch_shape) + pad_ndims = 0 if cat_batch_ndims == 1 else dist_batch_ndims - cat_batch_ndims + xs = x.shape + x = x.reshape( + xs[:-1] + + torch.Size(pad_ndims * [1]) + + xs[-1:] + + torch.Size(self._event_ndims * [1]) + ) + return x + + def __repr__(self): + args_string = ( + f"\n {self.mixture_distribution},\n {self.component_distribution}" + ) + return "MixtureSameFamily" + "(" + args_string + ")" diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/multinomial.py b/.venv/lib/python3.12/site-packages/torch/distributions/multinomial.py new file mode 100644 index 0000000000000000000000000000000000000000..320e640b842e5720c7d2e12876ea04aab19f9108 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/multinomial.py @@ -0,0 +1,148 @@ +# mypy: allow-untyped-defs + +import torch +from torch import inf, Tensor +from torch.distributions import Categorical, constraints +from torch.distributions.binomial import Binomial +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all + + +__all__ = ["Multinomial"] + + +class Multinomial(Distribution): + r""" + Creates a Multinomial distribution parameterized by :attr:`total_count` and + either :attr:`probs` or :attr:`logits` (but not both). The innermost dimension of + :attr:`probs` indexes over categories. All other dimensions index over batches. + + Note that :attr:`total_count` need not be specified if only :meth:`log_prob` is + called (see example below) + + .. note:: The `probs` argument must be non-negative, finite and have a non-zero sum, + and it will be normalized to sum to 1 along the last dimension. :attr:`probs` + will return this normalized value. + The `logits` argument will be interpreted as unnormalized log probabilities + and can therefore be any real number. It will likewise be normalized so that + the resulting probabilities sum to 1 along the last dimension. :attr:`logits` + will return this normalized value. + + - :meth:`sample` requires a single shared `total_count` for all + parameters and samples. + - :meth:`log_prob` allows different `total_count` for each parameter and + sample. + + Example:: + + >>> # xdoctest: +SKIP("FIXME: found invalid values") + >>> m = Multinomial(100, torch.tensor([ 1., 1., 1., 1.])) + >>> x = m.sample() # equal probability of 0, 1, 2, 3 + tensor([ 21., 24., 30., 25.]) + + >>> Multinomial(probs=torch.tensor([1., 1., 1., 1.])).log_prob(x) + tensor([-4.1338]) + + Args: + total_count (int): number of trials + probs (Tensor): event probabilities + logits (Tensor): event log probabilities (unnormalized) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.simplex, "logits": constraints.real_vector} + total_count: int + + @property + def mean(self) -> Tensor: + return self.probs * self.total_count + + @property + def variance(self) -> Tensor: + return self.total_count * self.probs * (1 - self.probs) + + def __init__( + self, + total_count: int = 1, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if not isinstance(total_count, int): + raise NotImplementedError("inhomogeneous total_count is not supported") + self.total_count = total_count + self._categorical = Categorical(probs=probs, logits=logits) + self._binomial = Binomial(total_count=total_count, probs=self.probs) + batch_shape = self._categorical.batch_shape + event_shape = self._categorical.param_shape[-1:] + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Multinomial, _instance) + batch_shape = torch.Size(batch_shape) + new.total_count = self.total_count + new._categorical = self._categorical.expand(batch_shape) + super(Multinomial, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._categorical._new(*args, **kwargs) + + @constraints.dependent_property(is_discrete=True, event_dim=1) + # pyrefly: ignore [bad-override] + def support(self): + return constraints.multinomial(self.total_count) + + @property + def logits(self) -> Tensor: + return self._categorical.logits + + @property + def probs(self) -> Tensor: + return self._categorical.probs + + @property + def param_shape(self) -> torch.Size: + return self._categorical.param_shape + + def sample(self, sample_shape=torch.Size()): + sample_shape = torch.Size(sample_shape) + samples = self._categorical.sample( + torch.Size((self.total_count,)) + sample_shape + ) + # samples.shape is (total_count, sample_shape, batch_shape), need to change it to + # (sample_shape, batch_shape, total_count) + shifted_idx = list(range(samples.dim())) + shifted_idx.append(shifted_idx.pop(0)) + samples = samples.permute(*shifted_idx) + counts = samples.new(self._extended_shape(sample_shape)).zero_() + counts.scatter_add_(-1, samples, torch.ones_like(samples)) + return counts.type_as(self.probs) + + def entropy(self): + n = torch.tensor(self.total_count) + + cat_entropy = self._categorical.entropy() + term1 = n * cat_entropy - torch.lgamma(n + 1) + + support = self._binomial.enumerate_support(expand=False)[1:] + binomial_probs = torch.exp(self._binomial.log_prob(support)) + weights = torch.lgamma(support + 1) + term2 = (binomial_probs * weights).sum([0, -1]) + + return term1 + term2 + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + logits, value = broadcast_all(self.logits, value) + logits = logits.clone(memory_format=torch.contiguous_format) + log_factorial_n = torch.lgamma(value.sum(-1) + 1) + log_factorial_xs = torch.lgamma(value + 1).sum(-1) + logits[(value == 0) & (logits == -inf)] = 0 + log_powers = (logits * value).sum(-1) + return log_factorial_n - log_factorial_xs + log_powers diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/multivariate_normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/multivariate_normal.py new file mode 100644 index 0000000000000000000000000000000000000000..e3432c1226c4d235fa7a77d93d5afaed42f5ad53 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/multivariate_normal.py @@ -0,0 +1,274 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import _standard_normal, lazy_property +from torch.types import _size + + +__all__ = ["MultivariateNormal"] + + +def _batch_mv(bmat, bvec): + r""" + Performs a batched matrix-vector product, with compatible but different batch shapes. + + This function takes as input `bmat`, containing :math:`n \times n` matrices, and + `bvec`, containing length :math:`n` vectors. + + Both `bmat` and `bvec` may have any number of leading dimensions, which correspond + to a batch shape. They are not necessarily assumed to have the same batch shape, + just ones which can be broadcasted. + """ + return torch.matmul(bmat, bvec.unsqueeze(-1)).squeeze(-1) + + +def _batch_mahalanobis(bL, bx): + r""" + Computes the squared Mahalanobis distance :math:`\mathbf{x}^\top\mathbf{M}^{-1}\mathbf{x}` + for a factored :math:`\mathbf{M} = \mathbf{L}\mathbf{L}^\top`. + + Accepts batches for both bL and bx. They are not necessarily assumed to have the same batch + shape, but `bL` one should be able to be broadcasted to `bx` one. + """ + n = bx.size(-1) + bx_batch_shape = bx.shape[:-1] + + # Assume that bL.shape = (i, 1, n, n), bx.shape = (..., i, j, n), + # we are going to make bx have shape (..., 1, j, i, 1, n) to apply batched tri.solve + bx_batch_dims = len(bx_batch_shape) + bL_batch_dims = bL.dim() - 2 + outer_batch_dims = bx_batch_dims - bL_batch_dims + old_batch_dims = outer_batch_dims + bL_batch_dims + new_batch_dims = outer_batch_dims + 2 * bL_batch_dims + # Reshape bx with the shape (..., 1, i, j, 1, n) + bx_new_shape = bx.shape[:outer_batch_dims] + for sL, sx in zip(bL.shape[:-2], bx.shape[outer_batch_dims:-1]): + bx_new_shape += (sx // sL, sL) + bx_new_shape += (n,) + bx = bx.reshape(bx_new_shape) + # Permute bx to make it have shape (..., 1, j, i, 1, n) + permute_dims = ( + list(range(outer_batch_dims)) + + list(range(outer_batch_dims, new_batch_dims, 2)) + + list(range(outer_batch_dims + 1, new_batch_dims, 2)) + + [new_batch_dims] + ) + bx = bx.permute(permute_dims) + + flat_L = bL.reshape(-1, n, n) # shape = b x n x n + flat_x = bx.reshape(-1, flat_L.size(0), n) # shape = c x b x n + flat_x_swap = flat_x.permute(1, 2, 0) # shape = b x n x c + M_swap = ( + torch.linalg.solve_triangular(flat_L, flat_x_swap, upper=False).pow(2).sum(-2) + ) # shape = b x c + M = M_swap.t() # shape = c x b + + # Now we revert the above reshape and permute operators. + permuted_M = M.reshape(bx.shape[:-1]) # shape = (..., 1, j, i, 1) + permute_inv_dims = list(range(outer_batch_dims)) + for i in range(bL_batch_dims): + permute_inv_dims += [outer_batch_dims + i, old_batch_dims + i] + reshaped_M = permuted_M.permute(permute_inv_dims) # shape = (..., 1, i, j, 1) + return reshaped_M.reshape(bx_batch_shape) + + +def _precision_to_scale_tril(P): + # Ref: https://nbviewer.jupyter.org/gist/fehiepsi/5ef8e09e61604f10607380467eb82006#Precision-to-scale_tril + Lf = torch.linalg.cholesky(torch.flip(P, (-2, -1))) + L_inv = torch.transpose(torch.flip(Lf, (-2, -1)), -2, -1) + Id = torch.eye(P.shape[-1], dtype=P.dtype, device=P.device) + L = torch.linalg.solve_triangular(L_inv, Id, upper=False) + return L + + +class MultivariateNormal(Distribution): + r""" + Creates a multivariate normal (also called Gaussian) distribution + parameterized by a mean vector and a covariance matrix. + + The multivariate normal distribution can be parameterized either + in terms of a positive definite covariance matrix :math:`\mathbf{\Sigma}` + or a positive definite precision matrix :math:`\mathbf{\Sigma}^{-1}` + or a lower-triangular matrix :math:`\mathbf{L}` with positive-valued + diagonal entries, such that + :math:`\mathbf{\Sigma} = \mathbf{L}\mathbf{L}^\top`. This triangular matrix + can be obtained via e.g. Cholesky decomposition of the covariance. + + Example: + + >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_LAPACK) + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = MultivariateNormal(torch.zeros(2), torch.eye(2)) + >>> m.sample() # normally distributed with mean=`[0,0]` and covariance_matrix=`I` + tensor([-0.2102, -0.5429]) + + Args: + loc (Tensor): mean of the distribution + covariance_matrix (Tensor): positive-definite covariance matrix + precision_matrix (Tensor): positive-definite precision matrix + scale_tril (Tensor): lower-triangular factor of covariance, with positive-valued diagonal + + Note: + Only one of :attr:`covariance_matrix` or :attr:`precision_matrix` or + :attr:`scale_tril` can be specified. + + Using :attr:`scale_tril` will be more efficient: all computations internally + are based on :attr:`scale_tril`. If :attr:`covariance_matrix` or + :attr:`precision_matrix` is passed instead, it is only used to compute + the corresponding lower triangular matrices using a Cholesky decomposition. + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "loc": constraints.real_vector, + "covariance_matrix": constraints.positive_definite, + "precision_matrix": constraints.positive_definite, + "scale_tril": constraints.lower_cholesky, + } + support = constraints.real_vector + has_rsample = True + + def __init__( + self, + loc: Tensor, + covariance_matrix: Tensor | None = None, + precision_matrix: Tensor | None = None, + scale_tril: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if loc.dim() < 1: + raise ValueError("loc must be at least one-dimensional.") + if (covariance_matrix is not None) + (scale_tril is not None) + ( + precision_matrix is not None + ) != 1: + raise ValueError( + "Exactly one of covariance_matrix or precision_matrix or scale_tril may be specified." + ) + + if scale_tril is not None: + if scale_tril.dim() < 2: + raise ValueError( + "scale_tril matrix must be at least two-dimensional, " + "with optional leading batch dimensions" + ) + batch_shape = torch.broadcast_shapes(scale_tril.shape[:-2], loc.shape[:-1]) + # pyrefly: ignore [read-only] + self.scale_tril = scale_tril.expand(batch_shape + (-1, -1)) + elif covariance_matrix is not None: + if covariance_matrix.dim() < 2: + raise ValueError( + "covariance_matrix must be at least two-dimensional, " + "with optional leading batch dimensions" + ) + batch_shape = torch.broadcast_shapes( + covariance_matrix.shape[:-2], loc.shape[:-1] + ) + # pyrefly: ignore [read-only] + self.covariance_matrix = covariance_matrix.expand(batch_shape + (-1, -1)) + else: + if precision_matrix is None: + raise AssertionError("precision_matrix is unexpectedly None") + if precision_matrix.dim() < 2: + raise ValueError( + "precision_matrix must be at least two-dimensional, " + "with optional leading batch dimensions" + ) + batch_shape = torch.broadcast_shapes( + precision_matrix.shape[:-2], loc.shape[:-1] + ) + # pyrefly: ignore [read-only] + self.precision_matrix = precision_matrix.expand(batch_shape + (-1, -1)) + self.loc = loc.expand(batch_shape + (-1,)) + + event_shape = self.loc.shape[-1:] + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + if scale_tril is not None: + self._unbroadcasted_scale_tril = scale_tril + elif covariance_matrix is not None: + self._unbroadcasted_scale_tril = torch.linalg.cholesky(covariance_matrix) + else: # precision_matrix is not None + self._unbroadcasted_scale_tril = _precision_to_scale_tril(precision_matrix) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(MultivariateNormal, _instance) + batch_shape = torch.Size(batch_shape) + loc_shape = batch_shape + self.event_shape + cov_shape = batch_shape + self.event_shape + self.event_shape + new.loc = self.loc.expand(loc_shape) + new._unbroadcasted_scale_tril = self._unbroadcasted_scale_tril + if "covariance_matrix" in self.__dict__: + new.covariance_matrix = self.covariance_matrix.expand(cov_shape) + if "scale_tril" in self.__dict__: + new.scale_tril = self.scale_tril.expand(cov_shape) + if "precision_matrix" in self.__dict__: + new.precision_matrix = self.precision_matrix.expand(cov_shape) + super(MultivariateNormal, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + @lazy_property + def scale_tril(self) -> Tensor: + return self._unbroadcasted_scale_tril.expand( + self._batch_shape + self._event_shape + self._event_shape + ) + + @lazy_property + def covariance_matrix(self) -> Tensor: + return torch.matmul( + self._unbroadcasted_scale_tril, self._unbroadcasted_scale_tril.mT + ).expand(self._batch_shape + self._event_shape + self._event_shape) + + @lazy_property + def precision_matrix(self) -> Tensor: + return torch.cholesky_inverse(self._unbroadcasted_scale_tril).expand( + self._batch_shape + self._event_shape + self._event_shape + ) + + @property + def mean(self) -> Tensor: + return self.loc + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def variance(self) -> Tensor: + return ( + self._unbroadcasted_scale_tril.pow(2) + .sum(-1) + .expand(self._batch_shape + self._event_shape) + ) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + eps = _standard_normal(shape, dtype=self.loc.dtype, device=self.loc.device) + return self.loc + _batch_mv(self._unbroadcasted_scale_tril, eps) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + diff = value - self.loc + M = _batch_mahalanobis(self._unbroadcasted_scale_tril, diff) + half_log_det = ( + self._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum(-1) + ) + return -0.5 * (self._event_shape[0] * math.log(2 * math.pi) + M) - half_log_det + + def entropy(self): + half_log_det = ( + self._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1).log().sum(-1) + ) + H = 0.5 * self._event_shape[0] * (1.0 + math.log(2 * math.pi)) + half_log_det + if len(self._batch_shape) == 0: + return H + else: + return H.expand(self._batch_shape) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/negative_binomial.py b/.venv/lib/python3.12/site-packages/torch/distributions/negative_binomial.py new file mode 100644 index 0000000000000000000000000000000000000000..d1d4e9f4374b0acc0c614bb2513c91317ce4e93e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/negative_binomial.py @@ -0,0 +1,150 @@ +# mypy: allow-untyped-defs + +import torch +import torch.nn.functional as F +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.gamma import Gamma +from torch.distributions.utils import ( + broadcast_all, + lazy_property, + logits_to_probs, + probs_to_logits, +) + + +__all__ = ["NegativeBinomial"] + + +class NegativeBinomial(Distribution): + r""" + Creates a Negative Binomial distribution, i.e. distribution + of the number of successful independent and identical Bernoulli trials + before :attr:`total_count` failures are achieved. The probability + of success of each Bernoulli trial is :attr:`probs`. + + Args: + total_count (float or Tensor): non-negative number of negative Bernoulli + trials to stop, although the distribution is still valid for real + valued count + probs (Tensor): Event probabilities of success in the half open interval [0, 1) + logits (Tensor): Event log-odds for probabilities of success + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "total_count": constraints.greater_than_eq(0), + "probs": constraints.half_open_interval(0.0, 1.0), + "logits": constraints.real, + } + support = constraints.nonnegative_integer + + def __init__( + self, + total_count: Tensor | float, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + ( + self.total_count, + # pyrefly: ignore [read-only] + self.probs, + ) = broadcast_all(total_count, probs) + self.total_count = self.total_count.type_as(self.probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + ( + self.total_count, + # pyrefly: ignore [read-only] + self.logits, + ) = broadcast_all(total_count, logits) + self.total_count = self.total_count.type_as(self.logits) + + self._param = self.probs if probs is not None else self.logits + batch_shape = self._param.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(NegativeBinomial, _instance) + batch_shape = torch.Size(batch_shape) + new.total_count = self.total_count.expand(batch_shape) + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + new._param = new.logits + super(NegativeBinomial, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + @property + def mean(self) -> Tensor: + return self.total_count * torch.exp(self.logits) + + @property + def mode(self) -> Tensor: + return ((self.total_count - 1) * self.logits.exp()).floor().clamp(min=0.0) + + @property + def variance(self) -> Tensor: + return self.mean / torch.sigmoid(-self.logits) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits, is_binary=True) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + @lazy_property + def _gamma(self) -> Gamma: + # Note we avoid validating because self.total_count can be zero. + return Gamma( + concentration=self.total_count, + rate=torch.exp(-self.logits), + validate_args=False, + ) + + def sample(self, sample_shape=torch.Size()): + with torch.no_grad(): + rate = self._gamma.sample(sample_shape=sample_shape) + return torch.poisson(rate) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + + log_unnormalized_prob = self.total_count * F.logsigmoid( + -self.logits + ) + value * F.logsigmoid(self.logits) + + log_normalization = ( + -torch.lgamma(self.total_count + value) + + torch.lgamma(1.0 + value) + + torch.lgamma(self.total_count) + ) + # The case self.total_count == 0 and value == 0 has probability 1 but + # lgamma(0) is infinite. Handle this case separately using a function + # that does not modify tensors in place to allow Jit compilation. + log_normalization = log_normalization.masked_fill( + self.total_count + value == 0.0, 0.0 + ) + + return log_unnormalized_prob - log_normalization diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/normal.py b/.venv/lib/python3.12/site-packages/torch/distributions/normal.py new file mode 100644 index 0000000000000000000000000000000000000000..6c50cdfc29f73d7c2fd578bc33892782b4c8151c --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/normal.py @@ -0,0 +1,122 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import _standard_normal, broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Normal"] + + +class Normal(ExponentialFamily): + r""" + Creates a normal (also called Gaussian) distribution parameterized by + :attr:`loc` and :attr:`scale`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Normal(torch.tensor([0.0]), torch.tensor([1.0])) + >>> m.sample() # normally distributed with loc=0 and scale=1 + tensor([ 0.1046]) + + Args: + loc (float or Tensor): mean of the distribution (often referred to as mu) + scale (float or Tensor): standard deviation of the distribution + (often referred to as sigma) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"loc": constraints.real, "scale": constraints.positive} + support = constraints.real + has_rsample = True + _mean_carrier_measure = 0 + + @property + def mean(self) -> Tensor: + return self.loc + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def stddev(self) -> Tensor: + return self.scale + + @property + def variance(self) -> Tensor: + return self.stddev.pow(2) + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.loc, self.scale = broadcast_all(loc, scale) + if isinstance(loc, _Number) and isinstance(scale, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.loc.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Normal, _instance) + batch_shape = torch.Size(batch_shape) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + super(Normal, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + with torch.no_grad(): + return torch.normal(self.loc.expand(shape), self.scale.expand(shape)) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + eps = _standard_normal(shape, dtype=self.loc.dtype, device=self.loc.device) + return self.loc + eps * self.scale + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + # compute the variance + var = self.scale**2 + log_scale = ( + math.log(self.scale) + if isinstance(self.scale, _Number) + else self.scale.log() + ) + return ( + -((value - self.loc) ** 2) / (2 * var) + - log_scale + - math.log(math.sqrt(2 * math.pi)) + ) + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + return 0.5 * ( + 1 + torch.erf((value - self.loc) * self.scale.reciprocal() / math.sqrt(2)) + ) + + def icdf(self, value): + return self.loc + self.scale * torch.erfinv(2 * value - 1) * math.sqrt(2) + + def entropy(self): + return 0.5 + 0.5 * math.log(2 * math.pi) + torch.log(self.scale) + + @property + def _natural_params(self) -> tuple[Tensor, Tensor]: + return (self.loc / self.scale.pow(2), -0.5 * self.scale.pow(2).reciprocal()) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x, y): + return -0.25 * x.pow(2) / y + 0.5 * torch.log(-math.pi / y) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/one_hot_categorical.py b/.venv/lib/python3.12/site-packages/torch/distributions/one_hot_categorical.py new file mode 100644 index 0000000000000000000000000000000000000000..4323e666cdd17307253607278bb8613ee93bda4f --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/one_hot_categorical.py @@ -0,0 +1,143 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.categorical import Categorical +from torch.distributions.distribution import Distribution +from torch.types import _size + + +__all__ = ["OneHotCategorical", "OneHotCategoricalStraightThrough"] + + +class OneHotCategorical(Distribution): + r""" + Creates a one-hot categorical distribution parameterized by :attr:`probs` or + :attr:`logits`. + + Samples are one-hot coded vectors of size ``probs.size(-1)``. + + .. note:: The `probs` argument must be non-negative, finite and have a non-zero sum, + and it will be normalized to sum to 1 along the last dimension. :attr:`probs` + will return this normalized value. + The `logits` argument will be interpreted as unnormalized log probabilities + and can therefore be any real number. It will likewise be normalized so that + the resulting probabilities sum to 1 along the last dimension. :attr:`logits` + will return this normalized value. + + See also: :func:`torch.distributions.Categorical` for specifications of + :attr:`probs` and :attr:`logits`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = OneHotCategorical(torch.tensor([ 0.25, 0.25, 0.25, 0.25 ])) + >>> m.sample() # equal probability of 0, 1, 2, 3 + tensor([ 0., 0., 0., 1.]) + + Args: + probs (Tensor): event probabilities + logits (Tensor): event log probabilities (unnormalized) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.simplex, "logits": constraints.real_vector} + support = constraints.one_hot + has_enumerate_support = True + + def __init__( + self, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + self._categorical = Categorical(probs, logits) + batch_shape = self._categorical.batch_shape + event_shape = self._categorical.param_shape[-1:] + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(OneHotCategorical, _instance) + batch_shape = torch.Size(batch_shape) + new._categorical = self._categorical.expand(batch_shape) + super(OneHotCategorical, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._categorical._new(*args, **kwargs) + + @property + def _param(self) -> Tensor: + return self._categorical._param + + @property + def probs(self) -> Tensor: + return self._categorical.probs + + @property + def logits(self) -> Tensor: + return self._categorical.logits + + @property + def mean(self) -> Tensor: + return self._categorical.probs + + @property + def mode(self) -> Tensor: + probs = self._categorical.probs + mode = probs.argmax(dim=-1) + return torch.nn.functional.one_hot(mode, num_classes=probs.shape[-1]).to(probs) + + @property + def variance(self) -> Tensor: + return self._categorical.probs * (1 - self._categorical.probs) + + @property + def param_shape(self) -> torch.Size: + return self._categorical.param_shape + + def sample(self, sample_shape=torch.Size()): + sample_shape = torch.Size(sample_shape) + probs = self._categorical.probs + num_events = self._categorical._num_events + indices = self._categorical.sample(sample_shape) + return torch.nn.functional.one_hot(indices, num_events).to(probs) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + indices = value.max(-1)[1] + return self._categorical.log_prob(indices) + + def entropy(self): + return self._categorical.entropy() + + def enumerate_support(self, expand=True): + n = self.event_shape[0] + values = torch.eye(n, dtype=self._param.dtype, device=self._param.device) + values = values.view((n,) + (1,) * len(self.batch_shape) + (n,)) + if expand: + values = values.expand((n,) + self.batch_shape + (n,)) + return values + + +class OneHotCategoricalStraightThrough(OneHotCategorical): + r""" + Creates a reparameterizable :class:`OneHotCategorical` distribution based on the straight- + through gradient estimator from [1]. + + [1] Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation + (Bengio et al., 2013) + """ + + has_rsample = True + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + samples = self.sample(sample_shape) + probs = self._categorical.probs # cached via @lazy_property + return samples + (probs - probs.detach()) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/pareto.py b/.venv/lib/python3.12/site-packages/torch/distributions/pareto.py new file mode 100644 index 0000000000000000000000000000000000000000..eb833f32ec13f96941a803572f4f321c85ee250b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/pareto.py @@ -0,0 +1,74 @@ +from typing import Optional + +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exponential import Exponential +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AffineTransform, ExpTransform +from torch.distributions.utils import broadcast_all +from torch.types import _size + + +__all__ = ["Pareto"] + + +class Pareto(TransformedDistribution): + r""" + Samples from a Pareto Type 1 distribution. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Pareto(torch.tensor([1.0]), torch.tensor([1.0])) + >>> m.sample() # sample from a Pareto distribution with scale=1 and alpha=1 + tensor([ 1.5623]) + + Args: + scale (float or Tensor): Scale parameter of the distribution + alpha (float or Tensor): Shape parameter of the distribution + """ + + arg_constraints = {"alpha": constraints.positive, "scale": constraints.positive} + + def __init__( + self, + scale: Tensor | float, + alpha: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.scale, self.alpha = broadcast_all(scale, alpha) + base_dist = Exponential(self.alpha, validate_args=validate_args) + transforms = [ExpTransform(), AffineTransform(loc=0, scale=self.scale)] + # pyrefly: ignore [bad-argument-type] + super().__init__(base_dist, transforms, validate_args=validate_args) + + def expand( + self, batch_shape: _size, _instance: Optional["Pareto"] = None + ) -> "Pareto": + new = self._get_checked_instance(Pareto, _instance) + new.scale = self.scale.expand(batch_shape) + new.alpha = self.alpha.expand(batch_shape) + return super().expand(batch_shape, _instance=new) + + @property + def mean(self) -> Tensor: + # mean is inf for alpha <= 1 + a = self.alpha.clamp(min=1) + return a * self.scale / (a - 1) + + @property + def mode(self) -> Tensor: + return self.scale + + @property + def variance(self) -> Tensor: + # var is inf for alpha <= 2 + a = self.alpha.clamp(min=2) + return self.scale.pow(2) * a / ((a - 1).pow(2) * (a - 2)) + + @constraints.dependent_property(is_discrete=False, event_dim=0) + def support(self) -> constraints.Constraint: + return constraints.greater_than_eq(self.scale) + + def entropy(self) -> Tensor: + return (self.scale / self.alpha).log() + (1 + self.alpha.reciprocal()) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/poisson.py b/.venv/lib/python3.12/site-packages/torch/distributions/poisson.py new file mode 100644 index 0000000000000000000000000000000000000000..7fad283b05dcd2068c914480c9cf56c524e237f8 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/poisson.py @@ -0,0 +1,87 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.utils import broadcast_all +from torch.types import _Number, Number + + +__all__ = ["Poisson"] + + +class Poisson(ExponentialFamily): + r""" + Creates a Poisson distribution parameterized by :attr:`rate`, the rate parameter. + + Samples are nonnegative integers, with a pmf given by + + .. math:: + \mathrm{rate}^k \frac{e^{-\mathrm{rate}}}{k!} + + Example:: + + >>> # xdoctest: +SKIP("poisson_cpu not implemented for 'Long'") + >>> m = Poisson(torch.tensor([4])) + >>> m.sample() + tensor([ 3.]) + + Args: + rate (Number, Tensor): the rate parameter + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"rate": constraints.nonnegative} + support = constraints.nonnegative_integer + + @property + def mean(self) -> Tensor: + return self.rate + + @property + def mode(self) -> Tensor: + return self.rate.floor() + + @property + def variance(self) -> Tensor: + return self.rate + + def __init__( + self, + rate: Tensor | Number, + validate_args: bool | None = None, + ) -> None: + (self.rate,) = broadcast_all(rate) + if isinstance(rate, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.rate.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Poisson, _instance) + batch_shape = torch.Size(batch_shape) + new.rate = self.rate.expand(batch_shape) + super(Poisson, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def sample(self, sample_shape=torch.Size()): + shape = self._extended_shape(sample_shape) + with torch.no_grad(): + return torch.poisson(self.rate.expand(shape)) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + rate, value = broadcast_all(self.rate, value) + return value.xlogy(rate) - rate - (value + 1).lgamma() + + @property + def _natural_params(self) -> tuple[Tensor]: + return (torch.log(self.rate),) + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x): + return torch.exp(x) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_bernoulli.py b/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_bernoulli.py new file mode 100644 index 0000000000000000000000000000000000000000..e98af3c39919a381ac81b9528aba595a3ea18d1d --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_bernoulli.py @@ -0,0 +1,174 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import SigmoidTransform +from torch.distributions.utils import ( + broadcast_all, + clamp_probs, + lazy_property, + logits_to_probs, + probs_to_logits, +) +from torch.types import _Number, _size, Number + + +__all__ = ["LogitRelaxedBernoulli", "RelaxedBernoulli"] + + +class LogitRelaxedBernoulli(Distribution): + r""" + Creates a LogitRelaxedBernoulli distribution parameterized by :attr:`probs` + or :attr:`logits` (but not both), which is the logit of a RelaxedBernoulli + distribution. + + Samples are logits of values in (0, 1). See [1] for more details. + + Args: + temperature (Tensor): relaxation temperature + probs (Number, Tensor): the probability of sampling `1` + logits (Number, Tensor): the log-odds of sampling `1` + + [1] The Concrete Distribution: A Continuous Relaxation of Discrete Random + Variables (Maddison et al., 2017) + + [2] Categorical Reparametrization with Gumbel-Softmax + (Jang et al., 2017) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.unit_interval, "logits": constraints.real} + support = constraints.real + + def __init__( + self, + temperature: Tensor, + probs: Tensor | Number | None = None, + logits: Tensor | Number | None = None, + validate_args: bool | None = None, + ) -> None: + self.temperature = temperature + if (probs is None) == (logits is None): + raise ValueError( + "Either `probs` or `logits` must be specified, but not both." + ) + if probs is not None: + is_scalar = isinstance(probs, _Number) + # pyrefly: ignore [read-only] + (self.probs,) = broadcast_all(probs) + else: + if logits is None: + raise AssertionError("logits is unexpectedly None") + is_scalar = isinstance(logits, _Number) + # pyrefly: ignore [read-only] + (self.logits,) = broadcast_all(logits) + self._param = self.probs if probs is not None else self.logits + if is_scalar: + batch_shape = torch.Size() + else: + batch_shape = self._param.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(LogitRelaxedBernoulli, _instance) + batch_shape = torch.Size(batch_shape) + new.temperature = self.temperature + if "probs" in self.__dict__: + new.probs = self.probs.expand(batch_shape) + new._param = new.probs + if "logits" in self.__dict__: + new.logits = self.logits.expand(batch_shape) + new._param = new.logits + super(LogitRelaxedBernoulli, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._param.new(*args, **kwargs) + + @lazy_property + def logits(self) -> Tensor: + return probs_to_logits(self.probs, is_binary=True) + + @lazy_property + def probs(self) -> Tensor: + return logits_to_probs(self.logits, is_binary=True) + + @property + def param_shape(self) -> torch.Size: + return self._param.size() + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + probs = clamp_probs(self.probs.expand(shape)) + uniforms = clamp_probs( + torch.rand(shape, dtype=probs.dtype, device=probs.device) + ) + return ( + uniforms.log() - (-uniforms).log1p() + probs.log() - (-probs).log1p() + ) / self.temperature + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + logits, value = broadcast_all(self.logits, value) + diff = logits - value.mul(self.temperature) + return self.temperature.log() + diff - 2 * diff.exp().log1p() + + +class RelaxedBernoulli(TransformedDistribution): + r""" + Creates a RelaxedBernoulli distribution, parametrized by + :attr:`temperature`, and either :attr:`probs` or :attr:`logits` + (but not both). This is a relaxed version of the `Bernoulli` distribution, + so the values are in (0, 1), and has reparametrizable samples. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = RelaxedBernoulli(torch.tensor([2.2]), + ... torch.tensor([0.1, 0.2, 0.3, 0.99])) + >>> m.sample() + tensor([ 0.2951, 0.3442, 0.8918, 0.9021]) + + Args: + temperature (Tensor): relaxation temperature + probs (Number, Tensor): the probability of sampling `1` + logits (Number, Tensor): the log-odds of sampling `1` + """ + + arg_constraints = {"probs": constraints.unit_interval, "logits": constraints.real} + # pyrefly: ignore [bad-override] + support = constraints.unit_interval + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: LogitRelaxedBernoulli + + def __init__( + self, + temperature: Tensor, + probs: Tensor | Number | None = None, + logits: Tensor | Number | None = None, + validate_args: bool | None = None, + ) -> None: + base_dist = LogitRelaxedBernoulli(temperature, probs, logits) + super().__init__(base_dist, SigmoidTransform(), validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(RelaxedBernoulli, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def temperature(self) -> Tensor: + return self.base_dist.temperature + + @property + def logits(self) -> Tensor: + return self.base_dist.logits + + @property + def probs(self) -> Tensor: + return self.base_dist.probs diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_categorical.py b/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_categorical.py new file mode 100644 index 0000000000000000000000000000000000000000..c8bd78d24cd2ab1f0f4cfdbc55713cc18595e2e1 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/relaxed_categorical.py @@ -0,0 +1,163 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.categorical import Categorical +from torch.distributions.distribution import Distribution +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import ExpTransform +from torch.distributions.utils import broadcast_all, clamp_probs +from torch.types import _size + + +__all__ = ["ExpRelaxedCategorical", "RelaxedOneHotCategorical"] + + +class ExpRelaxedCategorical(Distribution): + r""" + Creates a ExpRelaxedCategorical parameterized by + :attr:`temperature`, and either :attr:`probs` or :attr:`logits` (but not both). + Returns the log of a point in the simplex. Based on the interface to + :class:`OneHotCategorical`. + + Implementation based on [1]. + + See also: :func:`torch.distributions.OneHotCategorical` + + Args: + temperature (Tensor): relaxation temperature + probs (Tensor): event probabilities + logits (Tensor): unnormalized log probability for each event + + [1] The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables + (Maddison et al., 2017) + + [2] Categorical Reparametrization with Gumbel-Softmax + (Jang et al., 2017) + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"probs": constraints.simplex, "logits": constraints.real_vector} + support = ( + constraints.real_vector + ) # The true support is actually a submanifold of this. + has_rsample = True + + def __init__( + self, + temperature: Tensor, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + self._categorical = Categorical(probs, logits) + self.temperature = temperature + batch_shape = self._categorical.batch_shape + event_shape = self._categorical.param_shape[-1:] + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(ExpRelaxedCategorical, _instance) + batch_shape = torch.Size(batch_shape) + new.temperature = self.temperature + new._categorical = self._categorical.expand(batch_shape) + super(ExpRelaxedCategorical, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + def _new(self, *args, **kwargs): + return self._categorical._new(*args, **kwargs) + + @property + def param_shape(self) -> torch.Size: + return self._categorical.param_shape + + @property + def logits(self) -> Tensor: + return self._categorical.logits + + @property + def probs(self) -> Tensor: + return self._categorical.probs + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + uniforms = clamp_probs( + torch.rand(shape, dtype=self.logits.dtype, device=self.logits.device) + ) + gumbels = -((-(uniforms.log())).log()) + scores = (self.logits + gumbels) / self.temperature + return scores - scores.logsumexp(dim=-1, keepdim=True) + + def log_prob(self, value): + K = self._categorical._num_events + if self._validate_args: + self._validate_sample(value) + logits, value = broadcast_all(self.logits, value) + log_scale = torch.full_like( + self.temperature, float(K) + ).lgamma() - self.temperature.log().mul(-(K - 1)) + score = logits - value.mul(self.temperature) + score = (score - score.logsumexp(dim=-1, keepdim=True)).sum(-1) + return score + log_scale + + +class RelaxedOneHotCategorical(TransformedDistribution): + r""" + Creates a RelaxedOneHotCategorical distribution parametrized by + :attr:`temperature`, and either :attr:`probs` or :attr:`logits`. + This is a relaxed version of the :class:`OneHotCategorical` distribution, so + its samples are on simplex, and are reparametrizable. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = RelaxedOneHotCategorical(torch.tensor([2.2]), + ... torch.tensor([0.1, 0.2, 0.3, 0.4])) + >>> m.sample() + tensor([ 0.1294, 0.2324, 0.3859, 0.2523]) + + Args: + temperature (Tensor): relaxation temperature + probs (Tensor): event probabilities + logits (Tensor): unnormalized log probability for each event + """ + + arg_constraints = {"probs": constraints.simplex, "logits": constraints.real_vector} + # pyrefly: ignore [bad-override] + support = constraints.simplex + has_rsample = True + # pyrefly: ignore [bad-override] + base_dist: ExpRelaxedCategorical + + def __init__( + self, + temperature: Tensor, + probs: Tensor | None = None, + logits: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + base_dist = ExpRelaxedCategorical( + temperature, probs, logits, validate_args=validate_args + ) + super().__init__(base_dist, ExpTransform(), validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(RelaxedOneHotCategorical, _instance) + return super().expand(batch_shape, _instance=new) + + @property + def temperature(self) -> Tensor: + return self.base_dist.temperature + + @property + def logits(self) -> Tensor: + return self.base_dist.logits + + @property + def probs(self) -> Tensor: + return self.base_dist.probs diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/studentT.py b/.venv/lib/python3.12/site-packages/torch/distributions/studentT.py new file mode 100644 index 0000000000000000000000000000000000000000..ea2db5549e8095297949c770a3735537e3e9f306 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/studentT.py @@ -0,0 +1,127 @@ +# mypy: allow-untyped-defs +import math + +import torch +from torch import inf, nan, Tensor +from torch.distributions import Chi2, constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import _standard_normal, broadcast_all +from torch.types import _size + + +__all__ = ["StudentT"] + + +class StudentT(Distribution): + r""" + Creates a Student's t-distribution parameterized by degree of + freedom :attr:`df`, mean :attr:`loc` and scale :attr:`scale`. + + Example:: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = StudentT(torch.tensor([2.0])) + >>> m.sample() # Student's t-distributed with degrees of freedom=2 + tensor([ 0.1046]) + + Args: + df (float or Tensor): degrees of freedom + loc (float or Tensor): mean of the distribution + scale (float or Tensor): scale of the distribution + """ + + # pyrefly: ignore [bad-override] + arg_constraints = { + "df": constraints.positive, + "loc": constraints.real, + "scale": constraints.positive, + } + support = constraints.real + has_rsample = True + + @property + def mean(self) -> Tensor: + m = self.loc.clone(memory_format=torch.contiguous_format) + m[self.df <= 1] = nan + return m + + @property + def mode(self) -> Tensor: + return self.loc + + @property + def variance(self) -> Tensor: + m = self.df.clone(memory_format=torch.contiguous_format) + m[self.df > 2] = ( + self.scale[self.df > 2].pow(2) + * self.df[self.df > 2] + / (self.df[self.df > 2] - 2) + ) + m[(self.df <= 2) & (self.df > 1)] = inf + m[self.df <= 1] = nan + return m + + def __init__( + self, + df: Tensor | float, + loc: Tensor | float = 0.0, + scale: Tensor | float = 1.0, + validate_args: bool | None = None, + ) -> None: + self.df, self.loc, self.scale = broadcast_all(df, loc, scale) + self._chi2 = Chi2(self.df) + batch_shape = self.df.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(StudentT, _instance) + batch_shape = torch.Size(batch_shape) + new.df = self.df.expand(batch_shape) + new.loc = self.loc.expand(batch_shape) + new.scale = self.scale.expand(batch_shape) + new._chi2 = self._chi2.expand(batch_shape) + super(StudentT, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + # NOTE: This does not agree with scipy implementation as much as other distributions. + # (see https://github.com/fritzo/notebooks/blob/master/debug-student-t.ipynb). Using DoubleTensor + # parameters seems to help. + + # X ~ Normal(0, 1) + # Z ~ Chi2(df) + # Y = X / sqrt(Z / df) ~ StudentT(df) + shape = self._extended_shape(sample_shape) + X = _standard_normal(shape, dtype=self.df.dtype, device=self.df.device) + Z = self._chi2.rsample(sample_shape) + Y = X * torch.rsqrt(Z / self.df) + return self.loc + self.scale * Y + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + y = (value - self.loc) / self.scale + Z = ( + self.scale.log() + + 0.5 * self.df.log() + + 0.5 * math.log(math.pi) + + torch.lgamma(0.5 * self.df) + - torch.lgamma(0.5 * (self.df + 1.0)) + ) + return -0.5 * (self.df + 1.0) * torch.log1p(y**2.0 / self.df) - Z + + def entropy(self): + lbeta = ( + torch.lgamma(0.5 * self.df) + + math.lgamma(0.5) + - torch.lgamma(0.5 * (self.df + 1)) + ) + return ( + self.scale.log() + + 0.5 + * (self.df + 1) + * (torch.digamma(0.5 * (self.df + 1)) - torch.digamma(0.5 * self.df)) + + 0.5 * self.df.log() + + lbeta + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/transformed_distribution.py b/.venv/lib/python3.12/site-packages/torch/distributions/transformed_distribution.py new file mode 100644 index 0000000000000000000000000000000000000000..838e149f018e6507a0ce8e91c5873c06380584b7 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/transformed_distribution.py @@ -0,0 +1,226 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.independent import Independent +from torch.distributions.transforms import ComposeTransform, Transform +from torch.distributions.utils import _sum_rightmost +from torch.types import _size + + +__all__ = ["TransformedDistribution"] + + +class TransformedDistribution(Distribution): + r""" + Extension of the Distribution class, which applies a sequence of Transforms + to a base distribution. Let f be the composition of transforms applied:: + + X ~ BaseDistribution + Y = f(X) ~ TransformedDistribution(BaseDistribution, f) + log p(Y) = log p(X) + log |det (dX/dY)| + + Note that the ``.event_shape`` of a :class:`TransformedDistribution` is the + maximum shape of its base distribution and its transforms, since transforms + can introduce correlations among events. + + An example for the usage of :class:`TransformedDistribution` would be:: + + # Building a Logistic Distribution + # X ~ Uniform(0, 1) + # f = a + b * logit(X) + # Y ~ f(X) ~ Logistic(a, b) + base_distribution = Uniform(0, 1) + transforms = [SigmoidTransform().inv, AffineTransform(loc=a, scale=b)] + logistic = TransformedDistribution(base_distribution, transforms) + + For more examples, please look at the implementations of + :class:`~torch.distributions.gumbel.Gumbel`, + :class:`~torch.distributions.half_cauchy.HalfCauchy`, + :class:`~torch.distributions.half_normal.HalfNormal`, + :class:`~torch.distributions.log_normal.LogNormal`, + :class:`~torch.distributions.pareto.Pareto`, + :class:`~torch.distributions.weibull.Weibull`, + :class:`~torch.distributions.relaxed_bernoulli.RelaxedBernoulli` and + :class:`~torch.distributions.relaxed_categorical.RelaxedOneHotCategorical` + """ + + arg_constraints: dict[str, constraints.Constraint] = {} + + def __init__( + self, + base_distribution: Distribution, + transforms: Transform | list[Transform], + validate_args: bool | None = None, + ) -> None: + if isinstance(transforms, Transform): + self.transforms = [ + transforms, + ] + elif isinstance(transforms, list): + if not all(isinstance(t, Transform) for t in transforms): + raise ValueError( + "transforms must be a Transform or a list of Transforms" + ) + self.transforms = transforms + else: + raise ValueError( + f"transforms must be a Transform or list, but was {transforms}" + ) + + # Reshape base_distribution according to transforms. + base_shape = base_distribution.batch_shape + base_distribution.event_shape + base_event_dim = len(base_distribution.event_shape) + transform = ComposeTransform(self.transforms) + if len(base_shape) < transform.domain.event_dim: + raise ValueError( + f"base_distribution needs to have shape with size at least {transform.domain.event_dim}, but got {base_shape}." + ) + forward_shape = transform.forward_shape(base_shape) + expanded_base_shape = transform.inverse_shape(forward_shape) + if base_shape != expanded_base_shape: + base_batch_shape = expanded_base_shape[ + : len(expanded_base_shape) - base_event_dim + ] + base_distribution = base_distribution.expand(base_batch_shape) + reinterpreted_batch_ndims = transform.domain.event_dim - base_event_dim + if reinterpreted_batch_ndims > 0: + base_distribution = Independent( + base_distribution, reinterpreted_batch_ndims + ) + self.base_dist = base_distribution + + # Compute shapes. + transform_change_in_event_dim = ( + transform.codomain.event_dim - transform.domain.event_dim + ) + event_dim = max( + transform.codomain.event_dim, # the transform is coupled + base_event_dim + transform_change_in_event_dim, # the base dist is coupled + ) + if len(forward_shape) < event_dim: + raise AssertionError( + f"forward_shape length {len(forward_shape)} must be >= event_dim {event_dim}" + ) + cut = len(forward_shape) - event_dim + batch_shape = forward_shape[:cut] + event_shape = forward_shape[cut:] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(TransformedDistribution, _instance) + batch_shape = torch.Size(batch_shape) + shape = batch_shape + self.event_shape + for t in reversed(self.transforms): + shape = t.inverse_shape(shape) + base_batch_shape = shape[: len(shape) - len(self.base_dist.event_shape)] + new.base_dist = self.base_dist.expand(base_batch_shape) + new.transforms = self.transforms + super(TransformedDistribution, new).__init__( + batch_shape, self.event_shape, validate_args=False + ) + new._validate_args = self._validate_args + return new + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def support(self): + if not self.transforms: + return self.base_dist.support + support = self.transforms[-1].codomain + if len(self.event_shape) > support.event_dim: + support = constraints.independent( + support, len(self.event_shape) - support.event_dim + ) + return support + + @property + def has_rsample(self) -> bool: # type: ignore[override] + return self.base_dist.has_rsample + + def sample(self, sample_shape=torch.Size()): + """ + Generates a sample_shape shaped sample or sample_shape shaped batch of + samples if the distribution parameters are batched. Samples first from + base distribution and applies `transform()` for every transform in the + list. + """ + with torch.no_grad(): + x = self.base_dist.sample(sample_shape) + for transform in self.transforms: + x = transform(x) + return x + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + """ + Generates a sample_shape shaped reparameterized sample or sample_shape + shaped batch of reparameterized samples if the distribution parameters + are batched. Samples first from base distribution and applies + `transform()` for every transform in the list. + """ + x = self.base_dist.rsample(sample_shape) + for transform in self.transforms: + x = transform(x) + return x + + def log_prob(self, value): + """ + Scores the sample by inverting the transform(s) and computing the score + using the score of the base distribution and the log abs det jacobian. + """ + if self._validate_args: + self._validate_sample(value) + event_dim = len(self.event_shape) + log_prob: Tensor | float = 0.0 + y = value + for transform in reversed(self.transforms): + x = transform.inv(y) + event_dim += transform.domain.event_dim - transform.codomain.event_dim + log_prob = log_prob - _sum_rightmost( + transform.log_abs_det_jacobian(x, y), + event_dim - transform.domain.event_dim, + ) + y = x + + log_prob = log_prob + _sum_rightmost( + self.base_dist.log_prob(y), event_dim - len(self.base_dist.event_shape) + ) + return log_prob + + def _monotonize_cdf(self, value): + """ + This conditionally flips ``value -> 1-value`` to ensure :meth:`cdf` is + monotone increasing. + """ + sign = 1 + for transform in self.transforms: + sign = sign * transform.sign + if isinstance(sign, int) and sign == 1: + return value + return sign * (value - 0.5) + 0.5 + + def cdf(self, value): + """ + Computes the cumulative distribution function by inverting the + transform(s) and computing the score of the base distribution. + """ + for transform in self.transforms[::-1]: + value = transform.inv(value) + if self._validate_args: + self.base_dist._validate_sample(value) + value = self.base_dist.cdf(value) + value = self._monotonize_cdf(value) + return value + + def icdf(self, value): + """ + Computes the inverse cumulative distribution function using + transform(s) and computing the score of the base distribution. + """ + value = self._monotonize_cdf(value) + value = self.base_dist.icdf(value) + for transform in self.transforms: + value = transform(value) + return value diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/transforms.py b/.venv/lib/python3.12/site-packages/torch/distributions/transforms.py new file mode 100644 index 0000000000000000000000000000000000000000..ec97962ce299320dad48bd716729bfaf212ee8e4 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/transforms.py @@ -0,0 +1,1362 @@ +# mypy: allow-untyped-defs +import functools +import math +import operator +from collections.abc import Sequence + +import torch +import torch.nn.functional as F +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import ( + _sum_rightmost, + broadcast_all, + lazy_property, + tril_matrix_to_vec, + vec_to_tril_matrix, +) +from torch.nn.functional import pad, softplus +from torch.types import _Number + + +__all__ = [ + "AbsTransform", + "AffineTransform", + "CatTransform", + "ComposeTransform", + "CorrCholeskyTransform", + "CumulativeDistributionTransform", + "ExpTransform", + "IndependentTransform", + "LowerCholeskyTransform", + "PositiveDefiniteTransform", + "PowerTransform", + "ReshapeTransform", + "SigmoidTransform", + "SoftplusTransform", + "TanhTransform", + "SoftmaxTransform", + "StackTransform", + "StickBreakingTransform", + "Transform", + "identity_transform", +] + + +class Transform: + """ + Abstract class for invertible transformations with computable log + det jacobians. They are primarily used in + :class:`torch.distributions.TransformedDistribution`. + + Caching is useful for transforms whose inverses are either expensive or + numerically unstable. Note that care must be taken with memoized values + since the autograd graph may be reversed. For example while the following + works with or without caching:: + + y = t(x) + t.log_abs_det_jacobian(x, y).backward() # x will receive gradients. + + However the following will error when caching due to dependency reversal:: + + y = t(x) + z = t.inv(y) + grad(z.sum(), [y]) # error because z is x + + Derived classes should implement one or both of :meth:`_call` or + :meth:`_inverse`. Derived classes that set `bijective=True` should also + implement :meth:`log_abs_det_jacobian`. + + Args: + cache_size (int): Size of cache. If zero, no caching is done. If one, + the latest single value is cached. Only 0 and 1 are supported. + + Attributes: + domain (:class:`~torch.distributions.constraints.Constraint`): + The constraint representing valid inputs to this transform. + codomain (:class:`~torch.distributions.constraints.Constraint`): + The constraint representing valid outputs to this transform + which are inputs to the inverse transform. + bijective (bool): Whether this transform is bijective. A transform + ``t`` is bijective iff ``t.inv(t(x)) == x`` and + ``t(t.inv(y)) == y`` for every ``x`` in the domain and ``y`` in + the codomain. Transforms that are not bijective should at least + maintain the weaker pseudoinverse properties + ``t(t.inv(t(x)) == t(x)`` and ``t.inv(t(t.inv(y))) == t.inv(y)``. + sign (int or Tensor): For bijective univariate transforms, this + should be +1 or -1 depending on whether transform is monotone + increasing or decreasing. + """ + + bijective = False + domain: constraints.Constraint + codomain: constraints.Constraint + + def __init__(self, cache_size: int = 0) -> None: + self._cache_size = cache_size + self._inv: Transform | None = None + if cache_size == 0: + pass # default behavior + elif cache_size == 1: + self._cached_x_y = None, None + else: + raise ValueError("cache_size must be 0 or 1") + super().__init__() + + def __getstate__(self): + state = self.__dict__.copy() + state["_inv"] = None + return state + + @property + def event_dim(self) -> int: + if self.domain.event_dim == self.codomain.event_dim: + return self.domain.event_dim + raise ValueError("Please use either .domain.event_dim or .codomain.event_dim") + + @property + def inv(self) -> "Transform": + """ + Returns the inverse :class:`Transform` of this transform. + This should satisfy ``t.inv.inv is t``. + """ + inv = self._inv + if inv is None: + inv = _InverseTransform(self) + self._inv = inv + return inv + + @property + def sign(self) -> int: + """ + Returns the sign of the determinant of the Jacobian, if applicable. + In general this only makes sense for bijective transforms. + """ + raise NotImplementedError + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + if type(self).__init__ is Transform.__init__: + return type(self)(cache_size=cache_size) + raise NotImplementedError(f"{type(self)}.with_cache is not implemented") + + def __eq__(self, other): + return self is other + + def __ne__(self, other): + # Necessary for Python2 + return not self.__eq__(other) + + def __call__(self, x): + """ + Computes the transform `x => y`. + """ + if self._cache_size == 0: + return self._call(x) + x_old, y_old = self._cached_x_y + if x is x_old: + return y_old + y = self._call(x) + self._cached_x_y = x, y + return y + + def _inv_call(self, y): + """ + Inverts the transform `y => x`. + """ + if self._cache_size == 0: + return self._inverse(y) + x_old, y_old = self._cached_x_y + if y is y_old: + return x_old + x = self._inverse(y) + self._cached_x_y = x, y + return x + + def _call(self, x): + """ + Abstract method to compute forward transformation. + """ + raise NotImplementedError + + def _inverse(self, y): + """ + Abstract method to compute inverse transformation. + """ + raise NotImplementedError + + def log_abs_det_jacobian(self, x, y): + """ + Computes the log det jacobian `log |dy/dx|` given input and output. + """ + raise NotImplementedError + + def __repr__(self): + return self.__class__.__name__ + "()" + + def forward_shape(self, shape): + """ + Infers the shape of the forward computation, given the input shape. + Defaults to preserving shape. + """ + return shape + + def inverse_shape(self, shape): + """ + Infers the shapes of the inverse computation, given the output shape. + Defaults to preserving shape. + """ + return shape + + +class _InverseTransform(Transform): + """ + Inverts a single :class:`Transform`. + This class is private; please instead use the ``Transform.inv`` property. + """ + + def __init__(self, transform: Transform) -> None: + super().__init__(cache_size=transform._cache_size) + self._inv: Transform | None = transform + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def domain(self): + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv.codomain + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def codomain(self): + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv.domain + + @property + def bijective(self) -> bool: # type: ignore[override] + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv.bijective + + @property + def sign(self) -> int: + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv.sign + + @property + def inv(self) -> Transform: + return self._inv # pyrefly: ignore[bad-return] + + def with_cache(self, cache_size=1): + if self._inv is None: + raise AssertionError("_inv must not be None") + return self.inv.with_cache(cache_size).inv + + def __eq__(self, other): + if not isinstance(other, _InverseTransform): + return False + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv == other._inv + + def __repr__(self): + return f"{self.__class__.__name__}({repr(self._inv)})" + + def __call__(self, x): + if self._inv is None: + raise AssertionError("_inv must not be None") + return self._inv._inv_call(x) + + def log_abs_det_jacobian(self, x, y): + if self._inv is None: + raise AssertionError("_inv must not be None") + return -self._inv.log_abs_det_jacobian(y, x) + + def forward_shape(self, shape): + return self._inv.inverse_shape(shape) # pyrefly: ignore[missing-attribute] + + def inverse_shape(self, shape): + return self._inv.forward_shape(shape) # pyrefly: ignore[missing-attribute] + + +class ComposeTransform(Transform): + """ + Composes multiple transforms in a chain. + The transforms being composed are responsible for caching. + + Args: + parts (list of :class:`Transform`): A list of transforms to compose. + cache_size (int): Size of cache. If zero, no caching is done. If one, + the latest single value is cached. Only 0 and 1 are supported. + """ + + def __init__(self, parts: list[Transform], cache_size: int = 0) -> None: + if cache_size: + parts = [part.with_cache(cache_size) for part in parts] + super().__init__(cache_size=cache_size) + self.parts = parts + + def __eq__(self, other): + if not isinstance(other, ComposeTransform): + return False + return self.parts == other.parts + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def domain(self): + if not self.parts: + return constraints.real + domain = self.parts[0].domain + # Adjust event_dim to be maximum among all parts. + event_dim = self.parts[-1].codomain.event_dim + for part in reversed(self.parts): + event_dim += part.domain.event_dim - part.codomain.event_dim + event_dim = max(event_dim, part.domain.event_dim) + if event_dim < domain.event_dim: + raise AssertionError( + f"event_dim {event_dim} must be >= domain.event_dim {domain.event_dim}" + ) + if event_dim > domain.event_dim: + domain = constraints.independent(domain, event_dim - domain.event_dim) + return domain + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def codomain(self): + if not self.parts: + return constraints.real + codomain = self.parts[-1].codomain + # Adjust event_dim to be maximum among all parts. + event_dim = self.parts[0].domain.event_dim + for part in self.parts: + event_dim += part.codomain.event_dim - part.domain.event_dim + event_dim = max(event_dim, part.codomain.event_dim) + if event_dim < codomain.event_dim: + raise AssertionError( + f"event_dim {event_dim} must be >= codomain.event_dim {codomain.event_dim}" + ) + if event_dim > codomain.event_dim: + codomain = constraints.independent(codomain, event_dim - codomain.event_dim) + return codomain + + @lazy_property + def bijective(self) -> bool: # type: ignore[override] + return all(p.bijective for p in self.parts) + + @lazy_property + def sign(self) -> int: # type: ignore[override] + sign = 1 + for p in self.parts: + sign = sign * p.sign + return sign + + @property + def inv(self) -> Transform: + inv = self._inv + if inv is None: + inv = ComposeTransform([p.inv for p in reversed(self.parts)]) + self._inv = inv + inv._inv = self + return inv + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return ComposeTransform(self.parts, cache_size=cache_size) + + def __call__(self, x): + for part in self.parts: + x = part(x) + return x + + def log_abs_det_jacobian(self, x, y): + if not self.parts: + return torch.zeros_like(x) + + # Compute intermediates. This will be free if parts[:-1] are all cached. + xs = [x] + for part in self.parts[:-1]: + xs.append(part(xs[-1])) + xs.append(y) + + terms = [] + event_dim = self.domain.event_dim + for part, x, y in zip(self.parts, xs[:-1], xs[1:]): + terms.append( + _sum_rightmost( + part.log_abs_det_jacobian(x, y), event_dim - part.domain.event_dim + ) + ) + event_dim += part.codomain.event_dim - part.domain.event_dim + return functools.reduce(operator.add, terms) + + def forward_shape(self, shape): + for part in self.parts: + shape = part.forward_shape(shape) + return shape + + def inverse_shape(self, shape): + for part in reversed(self.parts): + shape = part.inverse_shape(shape) + return shape + + def __repr__(self): + fmt_string = self.__class__.__name__ + "(\n " + fmt_string += ",\n ".join([p.__repr__() for p in self.parts]) + fmt_string += "\n)" + return fmt_string + + +identity_transform = ComposeTransform([]) + + +class IndependentTransform(Transform): + """ + Wrapper around another transform to treat + ``reinterpreted_batch_ndims``-many extra of the right most dimensions as + dependent. This has no effect on the forward or backward transforms, but + does sum out ``reinterpreted_batch_ndims``-many of the rightmost dimensions + in :meth:`log_abs_det_jacobian`. + + Args: + base_transform (:class:`Transform`): A base transform. + reinterpreted_batch_ndims (int): The number of extra rightmost + dimensions to treat as dependent. + """ + + def __init__( + self, + base_transform: Transform, + reinterpreted_batch_ndims: int, + cache_size: int = 0, + ) -> None: + super().__init__(cache_size=cache_size) + self.base_transform = base_transform.with_cache(cache_size) + self.reinterpreted_batch_ndims = reinterpreted_batch_ndims + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return IndependentTransform( + self.base_transform, self.reinterpreted_batch_ndims, cache_size=cache_size + ) + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def domain(self): + return constraints.independent( + self.base_transform.domain, self.reinterpreted_batch_ndims + ) + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def codomain(self): + return constraints.independent( + self.base_transform.codomain, self.reinterpreted_batch_ndims + ) + + @property + def bijective(self) -> bool: # type: ignore[override] + return self.base_transform.bijective + + @property + def sign(self) -> int: + return self.base_transform.sign + + def _call(self, x): + if x.dim() < self.domain.event_dim: + raise ValueError("Too few dimensions on input") + return self.base_transform(x) + + def _inverse(self, y): + if y.dim() < self.codomain.event_dim: + raise ValueError("Too few dimensions on input") + return self.base_transform.inv(y) + + def log_abs_det_jacobian(self, x, y): + result = self.base_transform.log_abs_det_jacobian(x, y) + result = _sum_rightmost(result, self.reinterpreted_batch_ndims) + return result + + def __repr__(self): + return f"{self.__class__.__name__}({repr(self.base_transform)}, {self.reinterpreted_batch_ndims})" + + def forward_shape(self, shape): + return self.base_transform.forward_shape(shape) + + def inverse_shape(self, shape): + return self.base_transform.inverse_shape(shape) + + +class ReshapeTransform(Transform): + """ + Unit Jacobian transform to reshape the rightmost part of a tensor. + + Note that ``in_shape`` and ``out_shape`` must have the same number of + elements, just as for :meth:`torch.Tensor.reshape`. + + Arguments: + in_shape (torch.Size): The input event shape. + out_shape (torch.Size): The output event shape. + cache_size (int): Size of cache. If zero, no caching is done. If one, + the latest single value is cached. Only 0 and 1 are supported. (Default 0.) + """ + + bijective = True + + def __init__( + self, + in_shape: torch.Size, + out_shape: torch.Size, + cache_size: int = 0, + ) -> None: + self.in_shape = torch.Size(in_shape) + self.out_shape = torch.Size(out_shape) + if self.in_shape.numel() != self.out_shape.numel(): + raise ValueError("in_shape, out_shape have different numbers of elements") + super().__init__(cache_size=cache_size) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def domain(self): + return constraints.independent(constraints.real, len(self.in_shape)) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def codomain(self): + return constraints.independent(constraints.real, len(self.out_shape)) + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return ReshapeTransform(self.in_shape, self.out_shape, cache_size=cache_size) + + def _call(self, x): + batch_shape = x.shape[: x.dim() - len(self.in_shape)] + return x.reshape(batch_shape + self.out_shape) + + def _inverse(self, y): + batch_shape = y.shape[: y.dim() - len(self.out_shape)] + return y.reshape(batch_shape + self.in_shape) + + def log_abs_det_jacobian(self, x, y): + batch_shape = x.shape[: x.dim() - len(self.in_shape)] + return x.new_zeros(batch_shape) + + def forward_shape(self, shape): + if len(shape) < len(self.in_shape): + raise ValueError("Too few dimensions on input") + cut = len(shape) - len(self.in_shape) + if shape[cut:] != self.in_shape: + raise ValueError( + f"Shape mismatch: expected {shape[cut:]} but got {self.in_shape}" + ) + return shape[:cut] + self.out_shape + + def inverse_shape(self, shape): + if len(shape) < len(self.out_shape): + raise ValueError("Too few dimensions on input") + cut = len(shape) - len(self.out_shape) + if shape[cut:] != self.out_shape: + raise ValueError( + f"Shape mismatch: expected {shape[cut:]} but got {self.out_shape}" + ) + return shape[:cut] + self.in_shape + + +class ExpTransform(Transform): + r""" + Transform via the mapping :math:`y = \exp(x)`. + """ + + domain = constraints.real + codomain = constraints.positive + bijective = True + sign = +1 + + def __eq__(self, other): + return isinstance(other, ExpTransform) + + def _call(self, x): + return x.exp() + + def _inverse(self, y): + return y.log() + + def log_abs_det_jacobian(self, x, y): + return x + + +class PowerTransform(Transform): + r""" + Transform via the mapping :math:`y = x^{\text{exponent}}`. + """ + + domain = constraints.positive + codomain = constraints.positive + bijective = True + + def __init__(self, exponent: Tensor, cache_size: int = 0) -> None: + super().__init__(cache_size=cache_size) + (self.exponent,) = broadcast_all(exponent) + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return PowerTransform(self.exponent, cache_size=cache_size) + + @lazy_property + def sign(self) -> int: # type: ignore[override] + return self.exponent.sign() # type: ignore[return-value] + + def __eq__(self, other): + if not isinstance(other, PowerTransform): + return False + return self.exponent.eq(other.exponent).all().item() + + def _call(self, x): + return x.pow(self.exponent) + + def _inverse(self, y): + return y.pow(1 / self.exponent) + + def log_abs_det_jacobian(self, x, y): + return (self.exponent * y / x).abs().log() + + def forward_shape(self, shape): + return torch.broadcast_shapes(shape, getattr(self.exponent, "shape", ())) + + def inverse_shape(self, shape): + return torch.broadcast_shapes(shape, getattr(self.exponent, "shape", ())) + + +def _clipped_sigmoid(x): + finfo = torch.finfo(x.dtype) + return torch.clamp(torch.sigmoid(x), min=finfo.tiny, max=1.0 - finfo.eps) + + +class SigmoidTransform(Transform): + r""" + Transform via the mapping :math:`y = \frac{1}{1 + \exp(-x)}` and :math:`x = \text{logit}(y)`. + """ + + domain = constraints.real + codomain = constraints.unit_interval + bijective = True + sign = +1 + + def __eq__(self, other): + return isinstance(other, SigmoidTransform) + + def _call(self, x): + return _clipped_sigmoid(x) + + def _inverse(self, y): + finfo = torch.finfo(y.dtype) + y = y.clamp(min=finfo.tiny, max=1.0 - finfo.eps) + return y.log() - (-y).log1p() + + def log_abs_det_jacobian(self, x, y): + return -F.softplus(-x) - F.softplus(x) + + +class SoftplusTransform(Transform): + r""" + Transform via the mapping :math:`\text{Softplus}(x) = \log(1 + \exp(x))`. + The implementation reverts to the linear function when :math:`x > 20`. + """ + + domain = constraints.real + codomain = constraints.positive + bijective = True + sign = +1 + + def __eq__(self, other): + return isinstance(other, SoftplusTransform) + + def _call(self, x): + return softplus(x) + + def _inverse(self, y): + return (-y).expm1().neg().log() + y + + def log_abs_det_jacobian(self, x, y): + return -softplus(-x) + + +class TanhTransform(Transform): + r""" + Transform via the mapping :math:`y = \tanh(x)`. + + It is equivalent to + + .. code-block:: python + + ComposeTransform( + [ + AffineTransform(0.0, 2.0), + SigmoidTransform(), + AffineTransform(-1.0, 2.0), + ] + ) + + However this might not be numerically stable, thus it is recommended to use `TanhTransform` + instead. + + Note that one should use `cache_size=1` when it comes to `NaN/Inf` values. + + """ + + domain = constraints.real + codomain = constraints.interval(-1.0, 1.0) + bijective = True + sign = +1 + + def __eq__(self, other): + return isinstance(other, TanhTransform) + + def _call(self, x): + return x.tanh() + + def _inverse(self, y): + # We do not clamp to the boundary here as it may degrade the performance of certain algorithms. + # one should use `cache_size=1` instead + return torch.atanh(y) + + def log_abs_det_jacobian(self, x, y): + # We use a formula that is more numerically stable, see details in the following link + # https://github.com/tensorflow/probability/blob/master/tensorflow_probability/python/bijectors/tanh.py#L69-L80 + return 2.0 * (math.log(2.0) - x - softplus(-2.0 * x)) + + +class AbsTransform(Transform): + r"""Transform via the mapping :math:`y = |x|`.""" + + domain = constraints.real + codomain = constraints.positive + + def __eq__(self, other): + return isinstance(other, AbsTransform) + + def _call(self, x): + return x.abs() + + def _inverse(self, y): + return y + + +class AffineTransform(Transform): + r""" + Transform via the pointwise affine mapping :math:`y = \text{loc} + \text{scale} \times x`. + + Args: + loc (Tensor or float): Location parameter. + scale (Tensor or float): Scale parameter. + event_dim (int): Optional size of `event_shape`. This should be zero + for univariate random variables, 1 for distributions over vectors, + 2 for distributions over matrices, etc. + """ + + bijective = True + + def __init__( + self, + loc: Tensor | float, + scale: Tensor | float, + event_dim: int = 0, + cache_size: int = 0, + ) -> None: + super().__init__(cache_size=cache_size) + self.loc = loc + self.scale = scale + self._event_dim = event_dim + + @property + def event_dim(self) -> int: + return self._event_dim + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def domain(self): + if self.event_dim == 0: + return constraints.real + return constraints.independent(constraints.real, self.event_dim) + + @constraints.dependent_property(is_discrete=False) + # pyrefly: ignore [bad-override] + def codomain(self): + if self.event_dim == 0: + return constraints.real + return constraints.independent(constraints.real, self.event_dim) + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return AffineTransform( + self.loc, self.scale, self.event_dim, cache_size=cache_size + ) + + def __eq__(self, other): + if not isinstance(other, AffineTransform): + return False + + if isinstance(self.loc, _Number) and isinstance(other.loc, _Number): + if self.loc != other.loc: + return False + else: + if not (self.loc == other.loc).all().item(): # type: ignore[union-attr] + return False + + if isinstance(self.scale, _Number) and isinstance(other.scale, _Number): + if self.scale != other.scale: + return False + else: + if not (self.scale == other.scale).all().item(): # type: ignore[union-attr] + return False + + return True + + @property + def sign(self) -> Tensor | int: # type: ignore[override] + if isinstance(self.scale, _Number): + return 1 if float(self.scale) > 0 else -1 if float(self.scale) < 0 else 0 + return self.scale.sign() + + def _call(self, x): + return self.loc + self.scale * x + + def _inverse(self, y): + return (y - self.loc) / self.scale + + def log_abs_det_jacobian(self, x, y): + shape = x.shape + scale = self.scale + if isinstance(scale, _Number): + result = torch.full_like(x, math.log(abs(scale))) + else: + result = torch.abs(scale).log() + if self.event_dim: + result_size = result.size()[: -self.event_dim] + (-1,) + result = result.view(result_size).sum(-1) + shape = shape[: -self.event_dim] + return result.expand(shape) + + def forward_shape(self, shape): + return torch.broadcast_shapes( + shape, getattr(self.loc, "shape", ()), getattr(self.scale, "shape", ()) + ) + + def inverse_shape(self, shape): + return torch.broadcast_shapes( + shape, getattr(self.loc, "shape", ()), getattr(self.scale, "shape", ()) + ) + + +class CorrCholeskyTransform(Transform): + r""" + Transforms an unconstrained real vector :math:`x` with length :math:`D*(D-1)/2` into the + Cholesky factor of a D-dimension correlation matrix. This Cholesky factor is a lower + triangular matrix with positive diagonals and unit Euclidean norm for each row. + The transform is processed as follows: + + 1. First we convert x into a lower triangular matrix in row order. + 2. For each row :math:`X_i` of the lower triangular part, we apply a *signed* version of + class :class:`StickBreakingTransform` to transform :math:`X_i` into a + unit Euclidean length vector using the following steps: + - Scales into the interval :math:`(-1, 1)` domain: :math:`r_i = \tanh(X_i)`. + - Transforms into an unsigned domain: :math:`z_i = r_i^2`. + - Applies :math:`s_i = StickBreakingTransform(z_i)`. + - Transforms back into signed domain: :math:`y_i = sign(r_i) * \sqrt{s_i}`. + """ + + domain = constraints.real_vector + codomain = constraints.corr_cholesky + bijective = True + + def _call(self, x): + x = torch.tanh(x) + eps = torch.finfo(x.dtype).eps + x = x.clamp(min=-1 + eps, max=1 - eps) + r = vec_to_tril_matrix(x, diag=-1) + # apply stick-breaking on the squared values + # Note that y = sign(r) * sqrt(z * z1m_cumprod) + # = (sign(r) * sqrt(z)) * sqrt(z1m_cumprod) = r * sqrt(z1m_cumprod) + z = r**2 + z1m_cumprod_sqrt = (1 - z).sqrt().cumprod(-1) + # Diagonal elements must be 1. + r = r + torch.eye(r.shape[-1], dtype=r.dtype, device=r.device) + y = r * pad(z1m_cumprod_sqrt[..., :-1], [1, 0], value=1) + return y + + def _inverse(self, y): + # inverse stick-breaking + # See: https://mc-stan.org/docs/2_18/reference-manual/cholesky-factors-of-correlation-matrices-1.html + y_cumsum = 1 - torch.cumsum(y * y, dim=-1) + y_cumsum_shifted = pad(y_cumsum[..., :-1], [1, 0], value=1) + y_vec = tril_matrix_to_vec(y, diag=-1) + y_cumsum_vec = tril_matrix_to_vec(y_cumsum_shifted, diag=-1) + t = y_vec / (y_cumsum_vec).sqrt() + # inverse of tanh + x = (t.log1p() - t.neg().log1p()) / 2 + return x + + def log_abs_det_jacobian(self, x, y, intermediates=None): + # Because domain and codomain are two spaces with different dimensions, determinant of + # Jacobian is not well-defined. We return `log_abs_det_jacobian` of `x` and the + # flattened lower triangular part of `y`. + + # See: https://mc-stan.org/docs/2_18/reference-manual/cholesky-factors-of-correlation-matrices-1.html + y1m_cumsum = 1 - (y * y).cumsum(dim=-1) + # by taking diagonal=-2, we don't need to shift z_cumprod to the right + # also works for 2 x 2 matrix + y1m_cumsum_tril = tril_matrix_to_vec(y1m_cumsum, diag=-2) + stick_breaking_logdet = 0.5 * (y1m_cumsum_tril).log().sum(-1) + tanh_logdet = -2 * (x + softplus(-2 * x) - math.log(2.0)).sum(dim=-1) + return stick_breaking_logdet + tanh_logdet + + def forward_shape(self, shape): + # Reshape from (..., N) to (..., D, D). + if len(shape) < 1: + raise ValueError("Too few dimensions on input") + N = shape[-1] + D = round((0.25 + 2 * N) ** 0.5 + 0.5) + if D * (D - 1) // 2 != N: + raise ValueError("Input is not a flattened lower-diagonal number") + return shape[:-1] + (D, D) + + def inverse_shape(self, shape): + # Reshape from (..., D, D) to (..., N). + if len(shape) < 2: + raise ValueError("Too few dimensions on input") + if shape[-2] != shape[-1]: + raise ValueError("Input is not square") + D = shape[-1] + N = D * (D - 1) // 2 + return shape[:-2] + (N,) + + +class SoftmaxTransform(Transform): + r""" + Transform from unconstrained space to the simplex via :math:`y = \exp(x)` then + normalizing. + + This is not bijective and cannot be used for HMC. However this acts mostly + coordinate-wise (except for the final normalization), and thus is + appropriate for coordinate-wise optimization algorithms. + """ + + domain = constraints.real_vector + codomain = constraints.simplex + + def __eq__(self, other): + return isinstance(other, SoftmaxTransform) + + def _call(self, x): + logprobs = x + probs = (logprobs - logprobs.max(-1, True)[0]).exp() + return probs / probs.sum(-1, True) + + def _inverse(self, y): + probs = y + return probs.log() + + def forward_shape(self, shape): + if len(shape) < 1: + raise ValueError("Too few dimensions on input") + return shape + + def inverse_shape(self, shape): + if len(shape) < 1: + raise ValueError("Too few dimensions on input") + return shape + + +class StickBreakingTransform(Transform): + """ + Transform from unconstrained space to the simplex of one additional + dimension via a stick-breaking process. + + This transform arises as an iterated sigmoid transform in a stick-breaking + construction of the `Dirichlet` distribution: the first logit is + transformed via sigmoid to the first probability and the probability of + everything else, and then the process recurses. + + This is bijective and appropriate for use in HMC; however it mixes + coordinates together and is less appropriate for optimization. + """ + + domain = constraints.real_vector + codomain = constraints.simplex + bijective = True + + def __eq__(self, other): + return isinstance(other, StickBreakingTransform) + + def _call(self, x): + offset = x.shape[-1] + 1 - x.new_ones(x.shape[-1]).cumsum(-1) + z = _clipped_sigmoid(x - offset.log()) + z_cumprod = (1 - z).cumprod(-1) + y = pad(z, [0, 1], value=1) * pad(z_cumprod, [1, 0], value=1) + return y + + def _inverse(self, y): + y_crop = y[..., :-1] + offset = y.shape[-1] - y.new_ones(y_crop.shape[-1]).cumsum(-1) + sf = 1 - y_crop.cumsum(-1) + # we clamp to make sure that sf is positive which sometimes does not + # happen when y[-1] ~ 0 or y[:-1].sum() ~ 1 + sf = torch.clamp(sf, min=torch.finfo(y.dtype).tiny) + x = y_crop.log() - sf.log() + offset.log() + return x + + def log_abs_det_jacobian(self, x, y): + offset = x.shape[-1] + 1 - x.new_ones(x.shape[-1]).cumsum(-1) + x = x - offset.log() + # use the identity 1 - sigmoid(x) = exp(-x) * sigmoid(x) + detJ = (-x + F.logsigmoid(x) + y[..., :-1].log()).sum(-1) + return detJ + + def forward_shape(self, shape): + if len(shape) < 1: + raise ValueError("Too few dimensions on input") + return shape[:-1] + (shape[-1] + 1,) + + def inverse_shape(self, shape): + if len(shape) < 1: + raise ValueError("Too few dimensions on input") + return shape[:-1] + (shape[-1] - 1,) + + +class LowerCholeskyTransform(Transform): + """ + Transform from unconstrained matrices to lower-triangular matrices with + nonnegative diagonal entries. + + This is useful for parameterizing positive definite matrices in terms of + their Cholesky factorization. + """ + + domain = constraints.independent(constraints.real, 2) + codomain = constraints.lower_cholesky + + def __eq__(self, other): + return isinstance(other, LowerCholeskyTransform) + + def _call(self, x): + return x.tril(-1) + x.diagonal(dim1=-2, dim2=-1).exp().diag_embed() + + def _inverse(self, y): + return y.tril(-1) + y.diagonal(dim1=-2, dim2=-1).log().diag_embed() + + +class PositiveDefiniteTransform(Transform): + """ + Transform from unconstrained matrices to positive-definite matrices. + """ + + domain = constraints.independent(constraints.real, 2) + codomain = constraints.positive_definite + + def __eq__(self, other): + return isinstance(other, PositiveDefiniteTransform) + + def _call(self, x): + x = LowerCholeskyTransform()(x) + return x @ x.mT + + def _inverse(self, y): + y = torch.linalg.cholesky(y) + return LowerCholeskyTransform().inv(y) + + +class CatTransform(Transform): + """ + Transform functor that applies a sequence of transforms `tseq` + component-wise to each submatrix at `dim`, of length `lengths[dim]`, + in a way compatible with :func:`torch.cat`. + + Example:: + + x0 = torch.cat([torch.range(1, 10), torch.range(1, 10)], dim=0) + x = torch.cat([x0, x0], dim=0) + t0 = CatTransform([ExpTransform(), identity_transform], dim=0, lengths=[10, 10]) + t = CatTransform([t0, t0], dim=0, lengths=[20, 20]) + y = t(x) + """ + + transforms: list[Transform] + + def __init__( + self, + tseq: Sequence[Transform], + dim: int = 0, + lengths: Sequence[int] | None = None, + cache_size: int = 0, + ) -> None: + if not all(isinstance(t, Transform) for t in tseq): + raise AssertionError("All elements of tseq must be Transform instances") + if cache_size: + tseq = [t.with_cache(cache_size) for t in tseq] + super().__init__(cache_size=cache_size) + self.transforms = list(tseq) + if lengths is None: + lengths = [1] * len(self.transforms) + self.lengths = list(lengths) + if len(self.lengths) != len(self.transforms): + raise AssertionError( + f"lengths ({len(self.lengths)}) must match transforms ({len(self.transforms)})" + ) + self.dim = dim + + @lazy_property + def event_dim(self) -> int: # type: ignore[override] + return max(t.event_dim for t in self.transforms) + + @lazy_property + def length(self) -> int: + return sum(self.lengths) + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return CatTransform(self.transforms, self.dim, self.lengths, cache_size) + + def _call(self, x): + if not (-x.dim() <= self.dim < x.dim()): + raise AssertionError( + f"dim {self.dim} out of range for tensor with {x.dim()} dimensions" + ) + if x.size(self.dim) != self.length: + raise AssertionError( + f"x.size({self.dim}) = {x.size(self.dim)} must equal length {self.length}" + ) + yslices = [] + start = 0 + for trans, length in zip(self.transforms, self.lengths): + xslice = x.narrow(self.dim, start, length) + yslices.append(trans(xslice)) + start = start + length # avoid += for jit compat + return torch.cat(yslices, dim=self.dim) + + def _inverse(self, y): + if not (-y.dim() <= self.dim < y.dim()): + raise AssertionError( + f"dim {self.dim} out of range for tensor with {y.dim()} dimensions" + ) + if y.size(self.dim) != self.length: + raise AssertionError( + f"y.size({self.dim}) = {y.size(self.dim)} must equal length {self.length}" + ) + xslices = [] + start = 0 + for trans, length in zip(self.transforms, self.lengths): + yslice = y.narrow(self.dim, start, length) + xslices.append(trans.inv(yslice)) + start = start + length # avoid += for jit compat + return torch.cat(xslices, dim=self.dim) + + def log_abs_det_jacobian(self, x, y): + if not (-x.dim() <= self.dim < x.dim()): + raise AssertionError( + f"dim {self.dim} out of range for x with {x.dim()} dimensions" + ) + if x.size(self.dim) != self.length: + raise AssertionError( + f"x.size({self.dim}) = {x.size(self.dim)} must equal length {self.length}" + ) + if not (-y.dim() <= self.dim < y.dim()): + raise AssertionError( + f"dim {self.dim} out of range for y with {y.dim()} dimensions" + ) + if y.size(self.dim) != self.length: + raise AssertionError( + f"y.size({self.dim}) = {y.size(self.dim)} must equal length {self.length}" + ) + logdetjacs = [] + start = 0 + for trans, length in zip(self.transforms, self.lengths): + xslice = x.narrow(self.dim, start, length) + yslice = y.narrow(self.dim, start, length) + logdetjac = trans.log_abs_det_jacobian(xslice, yslice) + if trans.event_dim < self.event_dim: + logdetjac = _sum_rightmost(logdetjac, self.event_dim - trans.event_dim) + logdetjacs.append(logdetjac) + start = start + length # avoid += for jit compat + # Decide whether to concatenate or sum. + dim = self.dim + if dim >= 0: + dim = dim - x.dim() + dim = dim + self.event_dim + if dim < 0: + return torch.cat(logdetjacs, dim=dim) + else: + return sum(logdetjacs) + + @property + def bijective(self) -> bool: # type: ignore[override] + return all(t.bijective for t in self.transforms) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def domain(self): + return constraints.cat( + [t.domain for t in self.transforms], self.dim, self.lengths + ) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def codomain(self): + return constraints.cat( + [t.codomain for t in self.transforms], self.dim, self.lengths + ) + + +class StackTransform(Transform): + """ + Transform functor that applies a sequence of transforms `tseq` + component-wise to each submatrix at `dim` + in a way compatible with :func:`torch.stack`. + + Example:: + + x = torch.stack([torch.range(1, 10), torch.range(1, 10)], dim=1) + t = StackTransform([ExpTransform(), identity_transform], dim=1) + y = t(x) + """ + + transforms: list[Transform] + + def __init__( + self, tseq: Sequence[Transform], dim: int = 0, cache_size: int = 0 + ) -> None: + if not all(isinstance(t, Transform) for t in tseq): + raise AssertionError("All elements of tseq must be Transform instances") + if cache_size: + tseq = [t.with_cache(cache_size) for t in tseq] + super().__init__(cache_size=cache_size) + self.transforms = list(tseq) + self.dim = dim + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return StackTransform(self.transforms, self.dim, cache_size) + + def _slice(self, z): + return [z.select(self.dim, i) for i in range(z.size(self.dim))] + + def _call(self, x): + if not (-x.dim() <= self.dim < x.dim()): + raise AssertionError( + f"dim {self.dim} out of range for tensor with {x.dim()} dimensions" + ) + if x.size(self.dim) != len(self.transforms): + raise AssertionError( + f"x.size({self.dim}) = {x.size(self.dim)} must equal len(transforms) {len(self.transforms)}" + ) + yslices = [] + for xslice, trans in zip(self._slice(x), self.transforms): + yslices.append(trans(xslice)) + return torch.stack(yslices, dim=self.dim) + + def _inverse(self, y): + if not (-y.dim() <= self.dim < y.dim()): + raise AssertionError( + f"dim {self.dim} out of range for tensor with {y.dim()} dimensions" + ) + if y.size(self.dim) != len(self.transforms): + raise AssertionError( + f"y.size({self.dim}) = {y.size(self.dim)} must equal len(transforms) {len(self.transforms)}" + ) + xslices = [] + for yslice, trans in zip(self._slice(y), self.transforms): + xslices.append(trans.inv(yslice)) + return torch.stack(xslices, dim=self.dim) + + def log_abs_det_jacobian(self, x, y): + if not (-x.dim() <= self.dim < x.dim()): + raise AssertionError( + f"dim {self.dim} out of range for x with {x.dim()} dimensions" + ) + if x.size(self.dim) != len(self.transforms): + raise AssertionError( + f"x.size({self.dim}) = {x.size(self.dim)} must equal len(transforms) {len(self.transforms)}" + ) + if not (-y.dim() <= self.dim < y.dim()): + raise AssertionError( + f"dim {self.dim} out of range for y with {y.dim()} dimensions" + ) + if y.size(self.dim) != len(self.transforms): + raise AssertionError( + f"y.size({self.dim}) = {y.size(self.dim)} must equal len(transforms) {len(self.transforms)}" + ) + logdetjacs = [] + yslices = self._slice(y) + xslices = self._slice(x) + for xslice, yslice, trans in zip(xslices, yslices, self.transforms): + logdetjacs.append(trans.log_abs_det_jacobian(xslice, yslice)) + return torch.stack(logdetjacs, dim=self.dim) + + @property + def bijective(self) -> bool: # type: ignore[override] + return all(t.bijective for t in self.transforms) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def domain(self): + return constraints.stack([t.domain for t in self.transforms], self.dim) + + @constraints.dependent_property + # pyrefly: ignore [bad-override] + def codomain(self): + return constraints.stack([t.codomain for t in self.transforms], self.dim) + + +class CumulativeDistributionTransform(Transform): + """ + Transform via the cumulative distribution function of a probability distribution. + + Args: + distribution (Distribution): Distribution whose cumulative distribution function to use for + the transformation. + + Example:: + + # Construct a Gaussian copula from a multivariate normal. + base_dist = MultivariateNormal( + loc=torch.zeros(2), + scale_tril=LKJCholesky(2).sample(), + ) + transform = CumulativeDistributionTransform(Normal(0, 1)) + copula = TransformedDistribution(base_dist, [transform]) + """ + + bijective = True + codomain = constraints.unit_interval + sign = +1 + + def __init__(self, distribution: Distribution, cache_size: int = 0) -> None: + super().__init__(cache_size=cache_size) + self.distribution = distribution + + @property + def domain(self) -> constraints.Constraint | None: # type: ignore[override] + return self.distribution.support + + def _call(self, x): + return self.distribution.cdf(x) + + def _inverse(self, y): + return self.distribution.icdf(y) + + def log_abs_det_jacobian(self, x, y): + return self.distribution.log_prob(x) + + def with_cache(self, cache_size=1): + if self._cache_size == cache_size: + return self + return CumulativeDistributionTransform(self.distribution, cache_size=cache_size) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/uniform.py b/.venv/lib/python3.12/site-packages/torch/distributions/uniform.py new file mode 100644 index 0000000000000000000000000000000000000000..2e3fbcf7ab6d4e31f90437faaa3983a3cd79f88b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/uniform.py @@ -0,0 +1,108 @@ +# mypy: allow-untyped-defs + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all +from torch.types import _Number, _size + + +__all__ = ["Uniform"] + + +class Uniform(Distribution): + r""" + Generates uniformly distributed random samples from the half-open interval + ``[low, high)``. + + Example:: + + >>> m = Uniform(torch.tensor([0.0]), torch.tensor([5.0])) + >>> m.sample() # uniformly distributed in the range [0.0, 5.0) + >>> # xdoctest: +SKIP + tensor([ 2.3418]) + + Args: + low (float or Tensor): lower range (inclusive). + high (float or Tensor): upper range (exclusive). + """ + + has_rsample = True + + @property + def arg_constraints(self): + # TODO allow (loc,scale) parameterization to allow independent constraints. + return { + "low": constraints.less_than(self.high), + "high": constraints.greater_than(self.low), + } + + @property + def mean(self) -> Tensor: + return (self.high + self.low) / 2 + + @property + def mode(self) -> Tensor: + return nan * self.high + + @property + def stddev(self) -> Tensor: + return (self.high - self.low) / 12**0.5 + + @property + def variance(self) -> Tensor: + return (self.high - self.low).pow(2) / 12 + + def __init__( + self, + low: Tensor | float, + high: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.low, self.high = broadcast_all(low, high) + + if isinstance(low, _Number) and isinstance(high, _Number): + batch_shape = torch.Size() + else: + batch_shape = self.low.size() + super().__init__(batch_shape, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Uniform, _instance) + batch_shape = torch.Size(batch_shape) + new.low = self.low.expand(batch_shape) + new.high = self.high.expand(batch_shape) + super(Uniform, new).__init__(batch_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @constraints.dependent_property(is_discrete=False, event_dim=0) + # pyrefly: ignore [bad-override] + def support(self): + return constraints.interval(self.low, self.high) + + def rsample(self, sample_shape: _size = torch.Size()) -> Tensor: + shape = self._extended_shape(sample_shape) + rand = torch.rand(shape, dtype=self.low.dtype, device=self.low.device) + return self.low + rand * (self.high - self.low) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + lb = self.low.le(value).type_as(self.low) + ub = self.high.gt(value).type_as(self.low) + return torch.log(lb.mul(ub)) - torch.log(self.high - self.low) + + def cdf(self, value): + if self._validate_args: + self._validate_sample(value) + result = (value - self.low) / (self.high - self.low) + return result.clamp(min=0, max=1) + + def icdf(self, value): + result = value * (self.high - self.low) + self.low + return result + + def entropy(self): + return torch.log(self.high - self.low) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/utils.py b/.venv/lib/python3.12/site-packages/torch/distributions/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..e14fa35c1007f0425f76bb6d205c1a722bd9af67 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/utils.py @@ -0,0 +1,222 @@ +from collections.abc import Callable, Sequence +from functools import update_wrapper +from typing import Any, Final, Generic, overload, TypeVar + +import torch +import torch.nn.functional as F +from torch import SymInt, Tensor +from torch.overrides import is_tensor_like +from torch.types import _dtype, _Number, Device, Number + + +euler_constant: Final[float] = 0.57721566490153286060 # Euler Mascheroni Constant + +__all__ = [ + "broadcast_all", + "logits_to_probs", + "clamp_probs", + "probs_to_logits", + "lazy_property", + "tril_matrix_to_vec", + "vec_to_tril_matrix", +] + + +# FIXME: Use (*values: *Ts) -> tuple[Tensor for T in Ts] if Mapping-Type is ever added. +# See https://github.com/python/typing/issues/1216#issuecomment-2126153831 +def broadcast_all(*values: Tensor | Number) -> tuple[Tensor, ...]: + r""" + Given a list of values (possibly containing numbers), returns a list where each + value is broadcasted based on the following rules: + + - `torch.*Tensor` instances are broadcasted as per :ref:`broadcasting-semantics`. + - Number instances (scalars) are upcast to tensors having + the same size and type as the first tensor passed to `values`. If all the + values are scalars, then they are upcasted to scalar Tensors. + + Args: + values (list of `Number`, `torch.*Tensor` or objects implementing __torch_function__) + + Raises: + ValueError: if any of the values is not a `Number` instance, + a `torch.*Tensor` instance, or an instance implementing __torch_function__ + """ + if not all(is_tensor_like(v) or isinstance(v, _Number) for v in values): + raise ValueError( + "Input arguments must all be instances of Number, " + "torch.Tensor or objects implementing __torch_function__." + ) + if not all(is_tensor_like(v) for v in values): + options: dict[str, Any] = dict(dtype=torch.get_default_dtype()) + for value in values: + if isinstance(value, torch.Tensor): + options = dict(dtype=value.dtype, device=value.device) + break + new_values = [ + v if is_tensor_like(v) else torch.tensor(v, **options) for v in values + ] + return torch.broadcast_tensors(*new_values) + return torch.broadcast_tensors(*values) + + +def _standard_normal( + shape: Sequence[int | SymInt], + dtype: _dtype | None, + device: Device | None, +) -> Tensor: + if torch._C._get_tracing_state(): + # [JIT WORKAROUND] lack of support for .normal_() + return torch.normal( + torch.zeros(shape, dtype=dtype, device=device), + torch.ones(shape, dtype=dtype, device=device), + ) + return torch.empty(shape, dtype=dtype, device=device).normal_() + + +def _sum_rightmost(value: Tensor, dim: int) -> Tensor: + r""" + Sum out ``dim`` many rightmost dimensions of a given tensor. + + Args: + value (Tensor): A tensor of ``.dim()`` at least ``dim``. + dim (int): The number of rightmost dims to sum out. + """ + if dim == 0: + return value + required_shape = value.shape[:-dim] + (-1,) + return value.reshape(required_shape).sum(-1) + + +def logits_to_probs(logits: Tensor, is_binary: bool = False) -> Tensor: + r""" + Converts a tensor of logits into probabilities. Note that for the + binary case, each value denotes log odds, whereas for the + multi-dimensional case, the values along the last dimension denote + the log probabilities (possibly unnormalized) of the events. + """ + if is_binary: + return torch.sigmoid(logits) + return F.softmax(logits, dim=-1) + + +def clamp_probs(probs: Tensor) -> Tensor: + """Clamps the probabilities to be in the open interval `(0, 1)`. + + The probabilities would be clamped between `eps` and `1 - eps`, + and `eps` would be the smallest representable positive number for the input data type. + + Args: + probs (Tensor): A tensor of probabilities. + + Returns: + Tensor: The clamped probabilities. + + Examples: + >>> probs = torch.tensor([0.0, 0.5, 1.0]) + >>> clamp_probs(probs) + tensor([1.1921e-07, 5.0000e-01, 1.0000e+00]) + + >>> probs = torch.tensor([0.0, 0.5, 1.0], dtype=torch.float64) + >>> clamp_probs(probs) + tensor([2.2204e-16, 5.0000e-01, 1.0000e+00], dtype=torch.float64) + + """ + eps = torch.finfo(probs.dtype).eps + return probs.clamp(min=eps, max=1 - eps) + + +def probs_to_logits(probs: Tensor, is_binary: bool = False) -> Tensor: + r""" + Converts a tensor of probabilities into logits. For the binary case, + this denotes the probability of occurrence of the event indexed by `1`. + For the multi-dimensional case, the values along the last dimension + denote the probabilities of occurrence of each of the events. + """ + ps_clamped = clamp_probs(probs) + if is_binary: + return torch.log(ps_clamped) - torch.log1p(-ps_clamped) + return torch.log(ps_clamped) + + +T = TypeVar("T", contravariant=True) +R = TypeVar("R", covariant=True) + + +class lazy_property(Generic[T, R]): + r""" + Used as a decorator for lazy loading of class attributes. This uses a + non-data descriptor that calls the wrapped method to compute the property on + first call; thereafter replacing the wrapped method into an instance + attribute. + """ + + def __init__(self, wrapped: Callable[[T], R]) -> None: + self.wrapped: Callable[[T], R] = wrapped + update_wrapper(self, wrapped) # type:ignore[arg-type] + + @overload + def __get__( + self, instance: None, obj_type: Any = None + ) -> "_lazy_property_and_property[T, R]": ... + + @overload + def __get__(self, instance: T, obj_type: Any = None) -> R: ... + + def __get__( + self, instance: T | None, obj_type: Any = None + ) -> "R | _lazy_property_and_property[T, R]": + if instance is None: + return _lazy_property_and_property(self.wrapped) + with torch.enable_grad(): + value = self.wrapped(instance) + setattr(instance, self.wrapped.__name__, value) + return value + + +class _lazy_property_and_property(lazy_property[T, R], property): + """We want lazy properties to look like multiple things. + + * property when Sphinx autodoc looks + * lazy_property when Distribution validate_args looks + """ + + def __init__(self, wrapped: Callable[[T], R]) -> None: + property.__init__(self, wrapped) + + +def tril_matrix_to_vec(mat: Tensor, diag: int = 0) -> Tensor: + r""" + Convert a `D x D` matrix or a batch of matrices into a (batched) vector + which comprises of lower triangular elements from the matrix in row order. + """ + n = mat.shape[-1] + if not torch._C._get_tracing_state() and (diag < -n or diag >= n): + raise ValueError(f"diag ({diag}) provided is outside [{-n}, {n - 1}].") + arange = torch.arange(n, device=mat.device) + tril_mask = arange < arange.view(-1, 1) + (diag + 1) + vec = mat[..., tril_mask] + return vec + + +def vec_to_tril_matrix(vec: Tensor, diag: int = 0) -> Tensor: + r""" + Convert a vector or a batch of vectors into a batched `D x D` + lower triangular matrix containing elements from the vector in row order. + """ + # +ve root of D**2 + (1+2*diag)*D - |diag| * (diag+1) - 2*vec.shape[-1] = 0 + n = ( + -(1 + 2 * diag) + + ((1 + 2 * diag) ** 2 + 8 * vec.shape[-1] + 4 * abs(diag) * (diag + 1)) ** 0.5 + ) / 2 + eps = torch.finfo(vec.dtype).eps + if not torch._C._get_tracing_state() and (round(n) - n > eps): + raise ValueError( + f"The size of last dimension is {vec.shape[-1]} which cannot be expressed as " + + "the lower triangular part of a square D x D matrix." + ) + n = round(n.item()) if isinstance(n, torch.Tensor) else round(n) + mat = vec.new_zeros(vec.shape[:-1] + torch.Size((n, n))) + arange = torch.arange(n, device=vec.device) + tril_mask = arange < arange.view(-1, 1) + (diag + 1) + mat[..., tril_mask] = vec + return mat diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/von_mises.py b/.venv/lib/python3.12/site-packages/torch/distributions/von_mises.py new file mode 100644 index 0000000000000000000000000000000000000000..552b5c5f6662a9ce6607cc61c52f86caa1d48c04 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/von_mises.py @@ -0,0 +1,221 @@ +# mypy: allow-untyped-defs +import math + +import torch +import torch.jit +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.distribution import Distribution +from torch.distributions.utils import broadcast_all, lazy_property + + +__all__ = ["VonMises"] + + +def _eval_poly(y, coef): + coef = list(coef) + result = coef.pop() + while coef: + result = coef.pop() + y * result + return result + + +_I0_COEF_SMALL = [ + 1.0, + 3.5156229, + 3.0899424, + 1.2067492, + 0.2659732, + 0.360768e-1, + 0.45813e-2, +] +_I0_COEF_LARGE = [ + 0.39894228, + 0.1328592e-1, + 0.225319e-2, + -0.157565e-2, + 0.916281e-2, + -0.2057706e-1, + 0.2635537e-1, + -0.1647633e-1, + 0.392377e-2, +] +_I1_COEF_SMALL = [ + 0.5, + 0.87890594, + 0.51498869, + 0.15084934, + 0.2658733e-1, + 0.301532e-2, + 0.32411e-3, +] +_I1_COEF_LARGE = [ + 0.39894228, + -0.3988024e-1, + -0.362018e-2, + 0.163801e-2, + -0.1031555e-1, + 0.2282967e-1, + -0.2895312e-1, + 0.1787654e-1, + -0.420059e-2, +] + +_COEF_SMALL = [_I0_COEF_SMALL, _I1_COEF_SMALL] +_COEF_LARGE = [_I0_COEF_LARGE, _I1_COEF_LARGE] + + +def _log_modified_bessel_fn(x, order=0): + """ + Returns ``log(I_order(x))`` for ``x > 0``, + where `order` is either 0 or 1. + """ + if order != 0 and order != 1: + raise AssertionError(f"order must be 0 or 1, got {order}") + + # compute small solution + y = x / 3.75 + y = y * y + small = _eval_poly(y, _COEF_SMALL[order]) + if order == 1: + small = x.abs() * small + small = small.log() + + # compute large solution + y = 3.75 / x + large = x - 0.5 * x.log() + _eval_poly(y, _COEF_LARGE[order]).log() + + result = torch.where(x < 3.75, small, large) + return result + + +@torch.jit.script_if_tracing +def _rejection_sample(loc, concentration, proposal_r, x): + done = torch.zeros(x.shape, dtype=torch.bool, device=loc.device) + # pyrefly: ignore [bad-assignment, missing-attribute] + while not done.all(): + u = torch.rand((3,) + x.shape, dtype=loc.dtype, device=loc.device) + u1, u2, u3 = u.unbind() + z = torch.cos(math.pi * u1) + f = (1 + proposal_r * z) / (proposal_r + z) + c = concentration * (proposal_r - f) + accept = ((c * (2 - c) - u2) > 0) | ((c / u2).log() + 1 - c >= 0) + if accept.any(): + # pyrefly: ignore [no-matching-overload] + x = torch.where(accept, (u3 - 0.5).sign() * f.acos(), x) + done = done | accept + return (x + math.pi + loc) % (2 * math.pi) - math.pi + + +class VonMises(Distribution): + """ + A circular von Mises distribution. + + This implementation uses polar coordinates. The ``loc`` and ``value`` args + can be any real number (to facilitate unconstrained optimization), but are + interpreted as angles modulo 2 pi. + + Example:: + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = VonMises(torch.tensor([1.0]), torch.tensor([1.0])) + >>> m.sample() # von Mises distributed with loc=1 and concentration=1 + tensor([1.9777]) + + :param torch.Tensor loc: an angle in radians. + :param torch.Tensor concentration: concentration parameter + """ + + # pyrefly: ignore [bad-override] + arg_constraints = {"loc": constraints.real, "concentration": constraints.positive} + support = constraints.real + has_rsample = False + + def __init__( + self, + loc: Tensor, + concentration: Tensor, + validate_args: bool | None = None, + ) -> None: + self.loc, self.concentration = broadcast_all(loc, concentration) + batch_shape = self.loc.shape + event_shape = torch.Size() + super().__init__(batch_shape, event_shape, validate_args) + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + log_prob = self.concentration * torch.cos(value - self.loc) + log_prob = ( + log_prob + - math.log(2 * math.pi) + - _log_modified_bessel_fn(self.concentration, order=0) + ) + return log_prob + + @lazy_property + def _loc(self) -> Tensor: + return self.loc.to(torch.double) + + @lazy_property + def _concentration(self) -> Tensor: + return self.concentration.to(torch.double) + + @lazy_property + def _proposal_r(self) -> Tensor: + kappa = self._concentration + tau = 1 + (1 + 4 * kappa**2).sqrt() + rho = (tau - (2 * tau).sqrt()) / (2 * kappa) + _proposal_r = (1 + rho**2) / (2 * rho) + # second order Taylor expansion around 0 for small kappa + _proposal_r_taylor = 1 / kappa + kappa + return torch.where(kappa < 1e-5, _proposal_r_taylor, _proposal_r) + + @torch.no_grad() + def sample(self, sample_shape=torch.Size()): + """ + The sampling algorithm for the von Mises distribution is based on the + following paper: D.J. Best and N.I. Fisher, "Efficient simulation of the + von Mises distribution." Applied Statistics (1979): 152-157. + + Sampling is always done in double precision internally to avoid a hang + in _rejection_sample() for small values of the concentration, which + starts to happen for single precision around 1e-4 (see issue #88443). + """ + shape = self._extended_shape(sample_shape) + x = torch.empty(shape, dtype=self._loc.dtype, device=self.loc.device) + return _rejection_sample( + self._loc, self._concentration, self._proposal_r, x + ).to(self.loc.dtype) + + def expand(self, batch_shape, _instance=None): + try: + return super().expand(batch_shape) + except NotImplementedError: + validate_args = self.__dict__.get("_validate_args") + loc = self.loc.expand(batch_shape) + concentration = self.concentration.expand(batch_shape) + return type(self)(loc, concentration, validate_args=validate_args) + + @property + def mean(self) -> Tensor: + """ + The provided mean is the circular one. + """ + return self.loc + + @property + def mode(self) -> Tensor: + return self.loc + + @lazy_property + def variance(self) -> Tensor: # type: ignore[override] + """ + The provided variance is the circular one. + """ + return ( + 1 + - ( + _log_modified_bessel_fn(self.concentration, order=1) + - _log_modified_bessel_fn(self.concentration, order=0) + ).exp() + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/weibull.py b/.venv/lib/python3.12/site-packages/torch/distributions/weibull.py new file mode 100644 index 0000000000000000000000000000000000000000..22f48e2feeec5db9baa1cd0e530f2c1eddc50f4b --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/weibull.py @@ -0,0 +1,96 @@ +# mypy: allow-untyped-defs + +import torch +from torch import Tensor +from torch.distributions import constraints +from torch.distributions.exponential import Exponential +from torch.distributions.gumbel import euler_constant +from torch.distributions.transformed_distribution import TransformedDistribution +from torch.distributions.transforms import AffineTransform, PowerTransform +from torch.distributions.utils import broadcast_all + + +__all__ = ["Weibull"] + + +class Weibull(TransformedDistribution): + r""" + Samples from a two-parameter Weibull distribution. + + Example: + + >>> # xdoctest: +IGNORE_WANT("non-deterministic") + >>> m = Weibull(torch.tensor([1.0]), torch.tensor([1.0])) + >>> m.sample() # sample from a Weibull distribution with scale=1, concentration=1 + tensor([ 0.4784]) + + Args: + scale (float or Tensor): Scale parameter of distribution (lambda). + concentration (float or Tensor): Concentration parameter of distribution (k/shape). + validate_args (bool, optional): Whether to validate arguments. Default: None. + """ + + arg_constraints = { + "scale": constraints.positive, + "concentration": constraints.positive, + } + # pyrefly: ignore [bad-override] + support = constraints.positive + + def __init__( + self, + scale: Tensor | float, + concentration: Tensor | float, + validate_args: bool | None = None, + ) -> None: + self.scale, self.concentration = broadcast_all(scale, concentration) + self.concentration_reciprocal = self.concentration.reciprocal() + base_dist = Exponential( + torch.ones_like(self.scale), validate_args=validate_args + ) + transforms = [ + PowerTransform(exponent=self.concentration_reciprocal), + AffineTransform(loc=0, scale=self.scale), + ] + # pyrefly: ignore [bad-argument-type] + super().__init__(base_dist, transforms, validate_args=validate_args) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Weibull, _instance) + new.scale = self.scale.expand(batch_shape) + new.concentration = self.concentration.expand(batch_shape) + new.concentration_reciprocal = new.concentration.reciprocal() + base_dist = self.base_dist.expand(batch_shape) + transforms = [ + PowerTransform(exponent=new.concentration_reciprocal), + AffineTransform(loc=0, scale=new.scale), + ] + super(Weibull, new).__init__(base_dist, transforms, validate_args=False) + new._validate_args = self._validate_args + return new + + @property + def mean(self) -> Tensor: + return self.scale * torch.exp(torch.lgamma(1 + self.concentration_reciprocal)) + + @property + def mode(self) -> Tensor: + return ( + self.scale + * ((self.concentration - 1) / self.concentration) + ** self.concentration.reciprocal() + ) + + @property + def variance(self) -> Tensor: + return self.scale.pow(2) * ( + torch.exp(torch.lgamma(1 + 2 * self.concentration_reciprocal)) + - torch.exp(2 * torch.lgamma(1 + self.concentration_reciprocal)) + ) + + def entropy(self): + return ( + euler_constant * (1 - self.concentration_reciprocal) + + torch.log(self.scale * self.concentration_reciprocal) + + 1 + ) diff --git a/.venv/lib/python3.12/site-packages/torch/distributions/wishart.py b/.venv/lib/python3.12/site-packages/torch/distributions/wishart.py new file mode 100644 index 0000000000000000000000000000000000000000..2f5813cfaa2d0379e00b3c010c9fe376b25d2ea9 --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/distributions/wishart.py @@ -0,0 +1,351 @@ +# mypy: allow-untyped-defs +import math +import warnings + +import torch +from torch import nan, Tensor +from torch.distributions import constraints +from torch.distributions.exp_family import ExponentialFamily +from torch.distributions.multivariate_normal import _precision_to_scale_tril +from torch.distributions.utils import lazy_property +from torch.types import _Number, _size, Number + + +__all__ = ["Wishart"] + +_log_2 = math.log(2) + + +def _mvdigamma(x: Tensor, p: int) -> Tensor: + if not x.gt((p - 1) / 2).all(): + raise AssertionError("Wrong domain for multivariate digamma function.") + return torch.digamma( + x.unsqueeze(-1) + - torch.arange(p, dtype=x.dtype, device=x.device).div(2).expand(x.shape + (-1,)) + ).sum(-1) + + +def _clamp_above_eps(x: Tensor) -> Tensor: + # We assume positive input for this function + return x.clamp(min=torch.finfo(x.dtype).eps) + + +class Wishart(ExponentialFamily): + r""" + Creates a Wishart distribution parameterized by a symmetric positive definite matrix :math:`\Sigma`, + or its Cholesky decomposition :math:`\mathbf{\Sigma} = \mathbf{L}\mathbf{L}^\top` + + Example: + >>> # xdoctest: +SKIP("FIXME: scale_tril must be at least two-dimensional") + >>> m = Wishart(torch.Tensor([2]), covariance_matrix=torch.eye(2)) + >>> m.sample() # Wishart distributed with mean=`df * I` and + >>> # variance(x_ij)=`df` for i != j and variance(x_ij)=`2 * df` for i == j + + Args: + df (float or Tensor): real-valued parameter larger than the (dimension of Square matrix) - 1 + covariance_matrix (Tensor): positive-definite covariance matrix + precision_matrix (Tensor): positive-definite precision matrix + scale_tril (Tensor): lower-triangular factor of covariance, with positive-valued diagonal + Note: + Only one of :attr:`covariance_matrix` or :attr:`precision_matrix` or + :attr:`scale_tril` can be specified. + Using :attr:`scale_tril` will be more efficient: all computations internally + are based on :attr:`scale_tril`. If :attr:`covariance_matrix` or + :attr:`precision_matrix` is passed instead, it is only used to compute + the corresponding lower triangular matrices using a Cholesky decomposition. + 'torch.distributions.LKJCholesky' is a restricted Wishart distribution.[1] + + **References** + + [1] Wang, Z., Wu, Y. and Chu, H., 2018. `On equivalence of the LKJ distribution and the restricted Wishart distribution`. + [2] Sawyer, S., 2007. `Wishart Distributions and Inverse-Wishart Sampling`. + [3] Anderson, T. W., 2003. `An Introduction to Multivariate Statistical Analysis (3rd ed.)`. + [4] Odell, P. L. & Feiveson, A. H., 1966. `A Numerical Procedure to Generate a Sample Covariance Matrix`. JASA, 61(313):199-203. + [5] Ku, Y.-C. & Bloomfield, P., 2010. `Generating Random Wishart Matrices with Fractional Degrees of Freedom in OX`. + """ + + support = constraints.positive_definite + has_rsample = True + _mean_carrier_measure = 0 + + @property + def arg_constraints(self): + return { + "covariance_matrix": constraints.positive_definite, + "precision_matrix": constraints.positive_definite, + "scale_tril": constraints.lower_cholesky, + "df": constraints.greater_than(self.event_shape[-1] - 1), + } + + def __init__( + self, + df: Tensor | Number, + covariance_matrix: Tensor | None = None, + precision_matrix: Tensor | None = None, + scale_tril: Tensor | None = None, + validate_args: bool | None = None, + ) -> None: + if ( + (covariance_matrix is not None) + + (scale_tril is not None) + + (precision_matrix is not None) + ) != 1: + raise AssertionError( + "Exactly one of covariance_matrix or precision_matrix or scale_tril may be specified." + ) + + param = next( + p + for p in (covariance_matrix, precision_matrix, scale_tril) + if p is not None + ) + + if param.dim() < 2: + raise ValueError( + "scale_tril must be at least two-dimensional, with optional leading batch dimensions" + ) + + if isinstance(df, _Number): + batch_shape = torch.Size(param.shape[:-2]) + self.df = torch.tensor(df, dtype=param.dtype, device=param.device) + else: + batch_shape = torch.broadcast_shapes(param.shape[:-2], df.shape) + self.df = df.expand(batch_shape) + event_shape = param.shape[-2:] + + if self.df.le(event_shape[-1] - 1).any(): + raise ValueError( + f"Value of df={df} expected to be greater than ndim - 1 = {event_shape[-1] - 1}." + ) + + if scale_tril is not None: + # pyrefly: ignore [read-only] + self.scale_tril = param.expand(batch_shape + (-1, -1)) + elif covariance_matrix is not None: + # pyrefly: ignore [read-only] + self.covariance_matrix = param.expand(batch_shape + (-1, -1)) + elif precision_matrix is not None: + # pyrefly: ignore [read-only] + self.precision_matrix = param.expand(batch_shape + (-1, -1)) + + if self.df.lt(event_shape[-1]).any(): + warnings.warn( + "Low df values detected. Singular samples are highly likely to occur for ndim - 1 < df < ndim.", + stacklevel=2, + ) + + # pyrefly: ignore [bad-argument-type] + super().__init__(batch_shape, event_shape, validate_args=validate_args) + self._batch_dims = [-(x + 1) for x in range(len(self._batch_shape))] + + if scale_tril is not None: + self._unbroadcasted_scale_tril = scale_tril + elif covariance_matrix is not None: + self._unbroadcasted_scale_tril = torch.linalg.cholesky(covariance_matrix) + else: # precision_matrix is not None + self._unbroadcasted_scale_tril = _precision_to_scale_tril(precision_matrix) + + # Chi2 distribution is needed for Bartlett decomposition sampling + self._dist_chi2 = torch.distributions.chi2.Chi2( + df=( + self.df.unsqueeze(-1) + - torch.arange( + self._event_shape[-1], + dtype=self._unbroadcasted_scale_tril.dtype, + device=self._unbroadcasted_scale_tril.device, + ).expand(batch_shape + (-1,)) + ) + ) + + def expand(self, batch_shape, _instance=None): + new = self._get_checked_instance(Wishart, _instance) + batch_shape = torch.Size(batch_shape) + cov_shape = batch_shape + self.event_shape + new._unbroadcasted_scale_tril = self._unbroadcasted_scale_tril.expand(cov_shape) + new.df = self.df.expand(batch_shape) + + new._batch_dims = [-(x + 1) for x in range(len(batch_shape))] + + if "covariance_matrix" in self.__dict__: + new.covariance_matrix = self.covariance_matrix.expand(cov_shape) + if "scale_tril" in self.__dict__: + new.scale_tril = self.scale_tril.expand(cov_shape) + if "precision_matrix" in self.__dict__: + new.precision_matrix = self.precision_matrix.expand(cov_shape) + + # Chi2 distribution is needed for Bartlett decomposition sampling + new._dist_chi2 = torch.distributions.chi2.Chi2( + df=( + new.df.unsqueeze(-1) + - torch.arange( + self.event_shape[-1], + dtype=new._unbroadcasted_scale_tril.dtype, + device=new._unbroadcasted_scale_tril.device, + ).expand(batch_shape + (-1,)) + ) + ) + + super(Wishart, new).__init__(batch_shape, self.event_shape, validate_args=False) + new._validate_args = self._validate_args + return new + + @lazy_property + def scale_tril(self) -> Tensor: + return self._unbroadcasted_scale_tril.expand( + self._batch_shape + self._event_shape + ) + + @lazy_property + def covariance_matrix(self) -> Tensor: + return ( + self._unbroadcasted_scale_tril + @ self._unbroadcasted_scale_tril.transpose(-2, -1) + ).expand(self._batch_shape + self._event_shape) + + @lazy_property + def precision_matrix(self) -> Tensor: + identity = torch.eye( + self._event_shape[-1], + device=self._unbroadcasted_scale_tril.device, + dtype=self._unbroadcasted_scale_tril.dtype, + ) + return torch.cholesky_solve(identity, self._unbroadcasted_scale_tril).expand( + self._batch_shape + self._event_shape + ) + + @property + def mean(self) -> Tensor: + return self.df.view(self._batch_shape + (1, 1)) * self.covariance_matrix + + @property + def mode(self) -> Tensor: + factor = self.df - self.covariance_matrix.shape[-1] - 1 + factor[factor <= 0] = nan + return factor.view(self._batch_shape + (1, 1)) * self.covariance_matrix + + @property + def variance(self) -> Tensor: + V = self.covariance_matrix # has shape (batch_shape x event_shape) + diag_V = V.diagonal(dim1=-2, dim2=-1) + return self.df.view(self._batch_shape + (1, 1)) * ( + V.pow(2) + torch.einsum("...i,...j->...ij", diag_V, diag_V) + ) + + def _bartlett_sampling(self, sample_shape=torch.Size()): + p = self._event_shape[-1] # has singleton shape + + # Implemented Sampling using Bartlett decomposition + noise = _clamp_above_eps( + self._dist_chi2.rsample(sample_shape).sqrt() + ).diag_embed(dim1=-2, dim2=-1) + + i, j = torch.tril_indices(p, p, offset=-1) + noise[..., i, j] = torch.randn( + torch.Size(sample_shape) + self._batch_shape + (int(p * (p - 1) / 2),), + dtype=noise.dtype, + device=noise.device, + ) + chol = self._unbroadcasted_scale_tril @ noise + return chol @ chol.transpose(-2, -1) + + def rsample( + self, sample_shape: _size = torch.Size(), max_try_correction=None + ) -> Tensor: + r""" + .. warning:: + In some cases, sampling algorithm based on Bartlett decomposition may return singular matrix samples. + Several tries to correct singular samples are performed by default, but it may end up returning + singular matrix samples. Singular samples may return `-inf` values in `.log_prob()`. + In those cases, the user should validate the samples and either fix the value of `df` + or adjust `max_try_correction` value for argument in `.rsample` accordingly. + """ + + if max_try_correction is None: + max_try_correction = 3 if torch._C._get_tracing_state() else 10 + + sample_shape = torch.Size(sample_shape) + sample = self._bartlett_sampling(sample_shape) + + # Below part is to improve numerical stability temporarily and should be removed in the future + is_singular = self.support.check(sample) + if self._batch_shape: + is_singular = is_singular.amax(self._batch_dims) + + if torch._C._get_tracing_state(): + # Less optimized version for JIT + for _ in range(max_try_correction): + sample_new = self._bartlett_sampling(sample_shape) + sample = torch.where(is_singular, sample_new, sample) + + is_singular = ~self.support.check(sample) + if self._batch_shape: + is_singular = is_singular.amax(self._batch_dims) + + else: + # More optimized version with data-dependent control flow. + if is_singular.any(): + warnings.warn("Singular sample detected.", stacklevel=2) + + for _ in range(max_try_correction): + sample_new = self._bartlett_sampling(is_singular[is_singular].shape) + sample[is_singular] = sample_new + + is_singular_new = ~self.support.check(sample_new) + if self._batch_shape: + is_singular_new = is_singular_new.amax(self._batch_dims) + is_singular[is_singular.clone()] = is_singular_new + + if not is_singular.any(): + break + + return sample + + def log_prob(self, value): + if self._validate_args: + self._validate_sample(value) + nu = self.df # has shape (batch_shape) + p = self._event_shape[-1] # has singleton shape + return ( + -nu + * ( + p * _log_2 / 2 + + self._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1) + .log() + .sum(-1) + ) + - torch.mvlgamma(nu / 2, p=p) + + (nu - p - 1) / 2 * torch.linalg.slogdet(value).logabsdet + - torch.cholesky_solve(value, self._unbroadcasted_scale_tril) + .diagonal(dim1=-2, dim2=-1) + .sum(dim=-1) + / 2 + ) + + def entropy(self): + nu = self.df # has shape (batch_shape) + p = self._event_shape[-1] # has singleton shape + return ( + (p + 1) + * ( + p * _log_2 / 2 + + self._unbroadcasted_scale_tril.diagonal(dim1=-2, dim2=-1) + .log() + .sum(-1) + ) + + torch.mvlgamma(nu / 2, p=p) + - (nu - p - 1) / 2 * _mvdigamma(nu / 2, p=p) + + nu * p / 2 + ) + + @property + def _natural_params(self) -> tuple[Tensor, Tensor]: + nu = self.df # has shape (batch_shape) + p = self._event_shape[-1] # has singleton shape + return -self.precision_matrix / 2, (nu - p - 1) / 2 + + # pyrefly: ignore [bad-override] + def _log_normalizer(self, x, y): + p = self._event_shape[-1] + return (y + (p + 1) / 2) * ( + -torch.linalg.slogdet(-2 * x).logabsdet + _log_2 * p + ) + torch.mvlgamma(y + (p + 1) / 2, p=p) diff --git a/.venv/lib/python3.12/site-packages/torch/export/__init__.py b/.venv/lib/python3.12/site-packages/torch/export/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..ff6f87b59543ec30c557ba56d145272424c801ca --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/export/__init__.py @@ -0,0 +1,573 @@ +import logging +import os +import warnings +import zipfile +from collections.abc import Callable, Mapping +from typing import Any +from typing_extensions import deprecated + +import torch +import torch.utils._pytree as pytree +from torch.fx.passes.infra.pass_base import PassResult +from torch.types import FileLike + + +__all__ = [ + "AdditionalInputs", + "Constraint", + "CustomDecompTable", + "default_decompositions", + "Dim", + "dims", + "draft_export", + "export", + "ExportBackwardSignature", + "ExportedProgram", + "ExportGraphSignature", + "FlatArgsAdapter", + "load", + "ModuleCallEntry", + "ModuleCallSignature", + "register_dataclass", + "save", + "ShapesCollection", + "unflatten", + "UnflattenedModule", +] + +# To make sure export specific custom ops are loaded +import torch.export.custom_ops + +from ._state_dict_utils import _restore_state_dict +from .decomp_utils import CustomDecompTable +from .dynamic_shapes import AdditionalInputs, Constraint, Dim, dims, ShapesCollection +from .exported_program import ( + default_decompositions, + ExportedProgram, + ModuleCallEntry, + ModuleCallSignature, +) +from .graph_signature import ExportBackwardSignature, ExportGraphSignature +from .unflatten import FlatArgsAdapter, unflatten, UnflattenedModule + + +PassType = Callable[[torch.fx.GraphModule], PassResult | None] + +log: logging.Logger = logging.getLogger(__name__) + + +def export( + mod: torch.nn.Module, + args: tuple[Any, ...], + kwargs: Mapping[str, Any] | None = None, + *, + # dynamic_shapes: _DynamicShapesInput | AdditionalInputs | ShapesCollection + dynamic_shapes: Any = None, + strict: bool = False, + preserve_module_call_signature: tuple[str, ...] = (), + prefer_deferred_runtime_asserts_over_guards: bool = False, +) -> ExportedProgram: + """ + :func:`export` takes any nn.Module along with example inputs, and produces a traced graph representing + only the Tensor computation of the function in an Ahead-of-Time (AOT) fashion, + which can subsequently be executed with different inputs or serialized. The + traced graph (1) produces normalized operators in the functional ATen operator set + (as well as any user-specified custom operators), (2) has eliminated all Python control + flow and data structures (with certain exceptions), and (3) records the set of + shape constraints needed to show that this normalization and control-flow elimination + is sound for future inputs. + + **Soundness Guarantee** + + While tracing, :func:`export()` takes note of shape-related assumptions + made by the user program and the underlying PyTorch operator kernels. + The output :class:`ExportedProgram` is considered valid only when these + assumptions hold true. + + Tracing makes assumptions on the shapes (not values) of input tensors. + Such assumptions must be validated at graph capture time for :func:`export` + to succeed. Specifically: + + - Assumptions on static shapes of input tensors are automatically validated without additional effort. + - Assumptions on dynamic shape of input tensors require explicit specification + by using the :func:`Dim` API to construct dynamic dimensions and by associating + them with example inputs through the ``dynamic_shapes`` argument. + + If any assumption can not be validated, a fatal error will be raised. When that happens, + the error message will include suggested fixes to the specification that are needed + to validate the assumptions. For example :func:`export` might suggest the + following fix to the definition of a dynamic dimension ``dim0_x``, say appearing in the + shape associated with input ``x``, that was previously defined as ``Dim("dim0_x")``:: + + dim = Dim("dim0_x", max=5) + + This example means the generated code requires dimension 0 of input ``x`` to be less + than or equal to 5 to be valid. You can inspect the suggested fixes to dynamic dimension + definitions and then copy them verbatim into your code without needing to change the + ``dynamic_shapes`` argument to your :func:`export` call. + + Args: + mod: We will trace the forward method of this module. + + args: Example positional inputs. + + kwargs: Optional example keyword inputs. + + dynamic_shapes: + An optional argument where the type should either be: + 1) a dict from argument names of ``f`` to their dynamic shape specifications. + For a variadic ``**kwargs`` argument, the dict may use the actual keyword names + passed to ``f``, unless a keyword name collides with another input name. + 2) a tuple that specifies dynamic shape specifications for each input in original order. + If you are specifying dynamism on keyword args, you will need to pass them in the order that + is defined in the original function signature. + + The dynamic shape of a tensor argument can be specified as either + (1) a dict from dynamic dimension indices to :func:`Dim` types, where it is + not required to include static dimension indices in this dict, but when they are, + they should be mapped to None; or (2) a tuple / list of :func:`Dim` types or None, + where the :func:`Dim` types correspond to dynamic dimensions, and static dimensions + are denoted by None. Arguments that are dicts or tuples / lists of tensors are + recursively specified by using mappings or sequences of contained specifications. + + **ShapesSpec API.** ``dynamic_shapes`` may also be a + :class:`torch.fx.experimental.dynamic_spec.ShapesSpec` (or its + shorthand :class:`torch.fx.experimental.dynamic_spec.ParamsSpec`). + This is a newer unbacked unified API across compile, pre-compile, + export, etc., and is the recommended way to specify dynamic + shapes for export going forward. It is the same spec API exposed + via ``dynamic_shapes=`` in :func:`torch.compile`. + + The keys of ``ParamsSpec`` are **parameter names of the callable + being traced** (for an ``nn.Module``, the parameters of + ``forward``); keyword and ``**kwargs`` arguments are addressed by + name too. + + Key properties (see :mod:`torch.fx.experimental.dynamic_spec` for + full details): + + * **Unbacked-only.** Dims / scalars marked dynamic become unbacked + SymInts (``u`` symbols) and are never specialized (including no + 0/1 specialization). + * **Assumptions and derived expressions.** A dim can be an + expression over the spec's symbols (e.g. ``TensorSpec([B * 2, + ...])``), and you can pass relational ``assumptions`` between + symbols (e.g. ``[B % 2 == 0]``) that export validates. + * **No silent specialization.** The exported graph is guaranteed valid + for every assumption provided, otherwise export fails. (By + contrast, ``Dim.DYNAMIC`` / ``Dim.AUTO`` may silently specialize a + dynamic dim, yielding a graph that is not valid for all the inputs). + + Example:: + + batch = ShapeVar("batch", min=2, max=128) + ep = torch.export.export( + mod, + (torch.randn(8, 3), torch.randn(16, 3)), + dynamic_shapes=ShapesSpec( + params=ParamsSpec( + { + "x": TensorSpec([batch, 3]), + "y": TensorSpec([batch * 2, 3]), # derived expression + } + ), + assumptions=[batch % 2 == 0], + ), + strict=True, + ) + + strict: When disabled (default), the export function will trace the program through + Python runtime, which by itself will not validate some of the implicit assumptions + baked into the graph. It will still validate most critical assumptions like shape + safety. When enabled (by setting ``strict=True``), the export function will trace + the program through TorchDynamo which will ensure the soundness of the resulting + graph. TorchDynamo has limited Python feature coverage, thus you may experience more + errors. Note that toggling this argument does not affect the resulting IR spec to be + different and the model will be serialized in the same way regardless of what value + is passed here. + + preserve_module_call_signature: A list of submodule paths for which the original + calling conventions are preserved as metadata. The metadata will be used when calling + torch.export.unflatten to preserve the original calling conventions of modules. + + Returns: + An :class:`ExportedProgram` containing the traced callable. + + **Acceptable input/output types** + + Acceptable types of inputs (for ``args`` and ``kwargs``) and outputs include: + + - Primitive types, i.e. ``torch.Tensor``, ``int``, ``float``, ``bool`` and ``str``. + - Dataclasses, but they must be registered by calling :func:`register_dataclass` first. + - (Nested) Data structures comprising of ``dict``, ``list``, ``tuple``, ``namedtuple`` and + ``OrderedDict`` containing all above types. + + """ + from ._trace import _export + + if not isinstance(mod, torch.nn.Module): + raise ValueError( + f"Expected `mod` to be an instance of `torch.nn.Module`, got {type(mod)}." + ) + if isinstance(mod, torch.jit.ScriptModule): + raise ValueError( + "Exporting a ScriptModule is not supported. " + "Maybe try converting your ScriptModule to an ExportedProgram " + "using `TS2EPConverter(mod, args, kwargs).convert()` instead." + ) + + # If ``mod.forward`` carries an ``@dynamic_spec(...)`` decorator, the + # attached ``ShapesSpec`` is used as ``dynamic_shapes``. Passing both raises. + from torch.fx.experimental.dynamic_spec import _resolve_dynamic_shapes + + dynamic_shapes = _resolve_dynamic_shapes(mod, dynamic_shapes) + + try: + return _export( + mod, + args, + kwargs, + dynamic_shapes, + strict=strict, + preserve_module_call_signature=preserve_module_call_signature, + pre_dispatch=True, + prefer_deferred_runtime_asserts_over_guards=prefer_deferred_runtime_asserts_over_guards, + ) + except Exception as e: + draft_export_msg = ( + "The error above occurred when calling torch.export.export. If you would " + "like to view some more information about this error, and get a list " + "of all other errors that may occur in your export call, you can " + "replace your `export()` call with `draft_export()`." + ) + + # For errors that we know can be caught by draft-export, add the message + # to ask users to try out draft-export + if isinstance( + e, + ( + torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode, + torch._subclasses.fake_tensor.UnsupportedOperatorException, + torch._dynamo.exc.UserError, + torch.fx.experimental.symbolic_shapes.ConstraintViolationError, + ), + ): + new_msg = str(e) + "\n\n" + draft_export_msg + e.args = (new_msg,) + elif isinstance(e, RuntimeError) and "no fake impl registered" in str(e): + new_msg = str(e) + "\n\n" + draft_export_msg + e.args = (new_msg,) + raise e + + +DEFAULT_PICKLE_PROTOCOL = 2 + + +def save( + ep: ExportedProgram, + f: FileLike, + *, + extra_files: dict[str, Any] | None = None, + opset_version: dict[str, int] | None = None, + pickle_protocol: int = DEFAULT_PICKLE_PROTOCOL, +) -> None: + """ + + .. warning:: + Under active development, saved files may not be usable in newer versions + of PyTorch. + + Saves an :class:`ExportedProgram` to a file-like object. It can then be + loaded using the Python API :func:`torch.export.load `. + + Args: + ep (ExportedProgram): The exported program to save. + + f (str | os.PathLike[str] | IO[bytes]) A file-like object (has to + implement write and flush) or a string containing a file name. + + extra_files (Optional[Dict[str, Any]]): Map from filename to contents + which will be stored as part of f. + + opset_version (Optional[Dict[str, int]]): A map of opset names + to the version of this opset + + pickle_protocol: can be specified to override the default protocol + + Example:: + + import torch + import io + + + class MyModule(torch.nn.Module): + def forward(self, x): + return x + 10 + + + ep = torch.export.export(MyModule(), (torch.randn(5),)) + + # Save to file + torch.export.save(ep, "exported_program.pt2") + + # Save to io.BytesIO buffer + buffer = io.BytesIO() + torch.export.save(ep, buffer) + + # Save with extra files + extra_files = {"foo.txt": b"bar".decode("utf-8")} + torch.export.save(ep, "exported_program.pt2", extra_files=extra_files) + + """ + if not isinstance(ep, ExportedProgram): + raise TypeError( + f"The 'ep' parameter must be an instance of 'ExportedProgram', got '{type(ep).__name__}' instead." + ) + + from torch.export.pt2_archive._package import package_pt2 + + package_pt2( + f, + exported_programs={"model": ep}, + extra_files=extra_files, + pickle_protocol=pickle_protocol, + opset_version=opset_version, + ) + + +def load( + f: FileLike, + *, + extra_files: dict[str, Any] | None = None, + expected_opset_version: dict[str, int] | None = None, +) -> ExportedProgram: + """ + + .. warning:: + Under active development, saved files may not be usable in newer versions + of PyTorch. + + .. warning:: + :func:`torch.export.load()` uses pickle under the hood to load models. **Never load data from an untrusted source.** + + Loads an :class:`ExportedProgram` previously saved with + :func:`torch.export.save `. + + Args: + f (str | os.PathLike[str] | IO[bytes]): A file-like object (has to + implement write and flush) or a string containing a file name. + + extra_files (Optional[Dict[str, Any]]): The extra filenames given in + this map would be loaded and their content would be stored in the + provided map. + + expected_opset_version (Optional[Dict[str, int]]): A map of opset names + to expected opset versions + + Returns: + An :class:`ExportedProgram` object + + Example:: + + import torch + import io + + # Load ExportedProgram from file + ep = torch.export.load("exported_program.pt2") + + # Load ExportedProgram from io.BytesIO object + with open("exported_program.pt2", "rb") as f: + buffer = io.BytesIO(f.read()) + buffer.seek(0) + ep = torch.export.load(buffer) + + # Load with extra files. + extra_files = {"foo.txt": ""} # values will be replaced with data + ep = torch.export.load("exported_program.pt2", extra_files=extra_files) + print(extra_files["foo.txt"]) + print(ep(torch.randn(5))) + """ + if isinstance(f, (str, os.PathLike)): + f = os.fspath(f) + + extra_files = extra_files or {} + + from torch.export.pt2_archive._package import load_pt2, PT2ArchiveContents + + try: + pt2_contents = load_pt2( + f, + expected_opset_version=expected_opset_version, + ) + except RuntimeError: + log.warning("Ran into the following error when deserializing", exc_info=True) + pt2_contents = PT2ArchiveContents({}, {}, {}) + + if len(pt2_contents.exported_programs) > 0 or len(pt2_contents.extra_files) > 0: + for k, v in pt2_contents.extra_files.items(): + extra_files[k] = v + + return pt2_contents.exported_programs["model"] + + # TODO: For backward compatibility, we support loading a zip file from 2.7. Delete this path in 2.9(?) + with zipfile.ZipFile(f, "r") as zipf: + if "version" not in zipf.namelist(): + raise RuntimeError( + "We ran into an error when deserializing the saved file. " + "Please check the warnings above for possible errors. " + ) + + log.warning( + "Trying to deserialize for the older format. This version of file is " + "deprecated. Please generate a new pt2 saved file." + ) + + # Check the version + version = zipf.read("version").decode().split(".") + from torch._export.serde.schema import ( + SCHEMA_VERSION, # todo change archive version to schema version + ) + + if len(version) != len(SCHEMA_VERSION): + raise AssertionError( + "Version in the saved file has incorrect length, double check if the file is generated by torch.export.save()" + ) + if version[0] != str(SCHEMA_VERSION[0]): + raise RuntimeError( + f"Serialized version {version} does not match our current " + f"schema version {SCHEMA_VERSION}." + ) + + from torch._export.serde.serialize import deserialize, SerializedArtifact + + # Load serialized_ep and serialized_state_dict from the zip file + + serialized_exported_program: bytes | None = None + serialized_state_dict: bytes | None = None + serialized_constants: bytes | None = None + serialized_example_inputs: bytes | None = None + + for file_info in zipf.infolist(): + file_content = zipf.read(file_info.filename) + + if file_info.filename == "serialized_exported_program.json": + serialized_exported_program = file_content + elif file_info.filename == "serialized_state_dict.json": + warnings.warn("This version of file is deprecated", stacklevel=2) + serialized_state_dict = file_content + elif file_info.filename == "serialized_constants.json": + warnings.warn("This version of file is deprecated", stacklevel=2) + serialized_constants = file_content + elif file_info.filename == "serialized_state_dict.pt": + serialized_state_dict = file_content + elif file_info.filename == "serialized_constants.pt": + serialized_constants = file_content + elif file_info.filename == "serialized_example_inputs.pt": + serialized_example_inputs = file_content + elif file_info.filename.startswith("extra_files"): + filename = file_info.filename.split("/", 1)[1] + extra_files[filename] = file_content.decode("utf-8") + + if serialized_exported_program is None: + raise AssertionError("serialized_exported_program is None") + if serialized_state_dict is None: + raise AssertionError("serialized_state_dict is None") + if serialized_constants is None: + raise AssertionError("serialized_constants is None") + if serialized_example_inputs is None: + raise AssertionError("serialized_example_inputs is None") + artifact: SerializedArtifact = SerializedArtifact( + serialized_exported_program, + serialized_state_dict, + serialized_constants, + serialized_example_inputs, + ) + + # Deserialize ExportedProgram + ep = deserialize(artifact, expected_opset_version) + + return ep + + +def draft_export( + mod: torch.nn.Module, + args: tuple[Any, ...], + kwargs: Mapping[str, Any] | None = None, + *, + dynamic_shapes: dict[str, Any] | tuple[Any, ...] | list[Any] | None = None, + preserve_module_call_signature: tuple[str, ...] = (), + strict: bool = False, + prefer_deferred_runtime_asserts_over_guards: bool = False, +) -> ExportedProgram: + """ + A version of torch.export.export which is designed to consistently produce + an ExportedProgram, even if there are potential soundness issues, and to + generate a report listing the issues found. + """ + from torch.fx.experimental.dynamic_spec import ParamsSpec, ShapesSpec + + from ._draft_export import draft_export + + if isinstance(dynamic_shapes, (ShapesSpec, ParamsSpec)): + raise NotImplementedError( + f"draft_export does not support the new dynamic shapes API " + f"({type(dynamic_shapes).__name__}); use torch.export.export instead." + ) + + return draft_export( + mod=mod, + args=args, + kwargs=kwargs, + dynamic_shapes=dynamic_shapes, + preserve_module_call_signature=preserve_module_call_signature, + strict=strict, + prefer_deferred_runtime_asserts_over_guards=prefer_deferred_runtime_asserts_over_guards, + ) + + +def register_dataclass( + cls: type[Any], + *, + serialized_type_name: str | None = None, +) -> None: + """ + Registers a dataclass as a valid input/output type for :func:`torch.export.export`. + + Args: + cls: the dataclass type to register + serialized_type_name: The serialized name for the dataclass. This is + required if you want to serialize the pytree TreeSpec containing this + dataclass. + + Example:: + + import torch + from dataclasses import dataclass + + + @dataclass + class InputDataClass: + feature: torch.Tensor + bias: int + + + @dataclass + class OutputDataClass: + res: torch.Tensor + + + torch.export.register_dataclass(InputDataClass) + torch.export.register_dataclass(OutputDataClass) + + + class Mod(torch.nn.Module): + def forward(self, x: InputDataClass) -> OutputDataClass: + res = x.feature + x.bias + return OutputDataClass(res=res) + + + ep = torch.export.export(Mod(), (InputDataClass(torch.ones(2, 2), 1),)) + print(ep) + + """ + pytree.register_dataclass(cls, serialized_type_name=serialized_type_name) diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/__init__.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..4bec56f36f6e938a37115ae1b1f5f1321b20919e Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/__init__.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_draft_export.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_draft_export.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e705778574381fd9e7efb469fc1053d5b1f406da Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_draft_export.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_leakage_detection_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_leakage_detection_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..819b293c90a89def8dd7c2ca02ef19f61f4835d2 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_leakage_detection_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_patches.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_patches.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5fdb6e4e0e79025b1dddad3d8c82e259b3f545d5 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_patches.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_auto_functionalized_pass.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_auto_functionalized_pass.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b7a46016736ed8cbc64a3ef94792d64076def9b9 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_auto_functionalized_pass.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_effect_tokens_pass.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_effect_tokens_pass.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cf569276d95e807b21a847f3f6e534593933d263 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_remove_effect_tokens_pass.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_safeguard.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_safeguard.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..015d8ceeeed149a6b9fc728349e59315c2fef100 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_safeguard.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_state_dict_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_state_dict_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..29d3f4c110664cb2d4fbf3c633c94478a143e2b8 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_state_dict_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_swap.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_swap.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..954dc32468d32945a2ab30cc73f97928c0e87c70 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_swap.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_trace.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_trace.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..7fbc8a42e5f3cb9192fafd74b89ec4327d82db3e --- /dev/null +++ b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_trace.cpython-312.pyc @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3dfed0876ab1f07660ecd5268b2e6fffd5a4d402d2fed04cd73b8bac6bd0ca24 +size 110333 diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_tree_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_tree_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6f908472c650b29dab8e69865f1a679f9e6717bd Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_tree_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_unlift.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_unlift.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6606fc94b955a3ca757cbdda510f85a1d1bf5b58 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_unlift.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_wrapper_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_wrapper_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..c8431ae66ab8e4eed4fe87c159b09c177c119e46 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/_wrapper_utils.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_obj.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_obj.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ccec096d8e28b878df373ebcd62827681755bd70 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_obj.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_ops.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_ops.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..880f0f0c097e0602c676c3cf1003cb76d419bc19 Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/custom_ops.cpython-312.pyc differ diff --git a/.venv/lib/python3.12/site-packages/torch/export/__pycache__/decomp_utils.cpython-312.pyc b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/decomp_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..44e815876c75435fae686231cf0b32f792f86ebc Binary files /dev/null and b/.venv/lib/python3.12/site-packages/torch/export/__pycache__/decomp_utils.cpython-312.pyc differ