id
int64
0
190k
prompt
stringlengths
21
13.4M
docstring
stringlengths
1
12k
8,861
import argparse import sys from typing import List, Tuple import torch from fbgemm_gpu.split_table_batched_embeddings_ops_training import EmbeddingLocation from torchrec.github.benchmarks import ebc_benchmarks_utils from torchrec.modules.embedding_configs import EmbeddingBagConfig from torchrec.modules.embedding_modules import EmbeddingBagCollection from torchrec.modules.fused_embedding_modules import FusedEmbeddingBagCollection def parse_args(argv: List[str]) -> argparse.Namespace: parser = argparse.ArgumentParser(description="TorchRec ebc benchmarks") parser.add_argument( "--cpu_only", action="store_true", default=False, help="specify whether to use cpu", ) parser.add_argument( "--mode", type=str, default="ebc_comparison_dlrm", help="specify 'ebc_comparison_dlrm', 'ebc_comparison_scaling' or 'fused_ebc_uvm'", ) return parser.parse_args(argv)
null
8,862
import argparse import sys from typing import List, Tuple import torch from fbgemm_gpu.split_table_batched_embeddings_ops_training import EmbeddingLocation from torchrec.github.benchmarks import ebc_benchmarks_utils from torchrec.modules.embedding_configs import EmbeddingBagConfig from torchrec.modules.embedding_modules import EmbeddingBagCollection from torchrec.modules.fused_embedding_modules import FusedEmbeddingBagCollection def main(argv: List[str]) -> None: args = parse_args(argv) if not args.cpu_only and torch.cuda.is_available(): device = torch.device("cuda") else: device = torch.device("cpu") if args.mode == "ebc_comparison_dlrm": print("Running EBC vs. FusedEBC on DLRM EMB") for reduction_degree in [128, 64, 32]: embedding_bag_configs: List[EmbeddingBagConfig] = [ EmbeddingBagConfig( name=f"ebc_{idx}", embedding_dim=128, num_embeddings=num_embeddings, feature_names=[f"ebc_{idx}_feat_1"], ) for idx, num_embeddings in enumerate( get_shrunk_dlrm_num_embeddings(reduction_degree) ) ] ( ebc_time_avg, ebc_time_std, fused_ebc_time_avg, fused_ebc_time_std, speedup, ) = get_ebc_comparison(embedding_bag_configs, device) print(f"when DLRM EMB is reduced by {reduction_degree} times:") print(f"ebc_time = {ebc_time_avg} +/- {ebc_time_std} sec") print(f"fused_ebc_time = {fused_ebc_time_avg} +/- {fused_ebc_time_std} sec") print(f"speedup = {speedup}") elif args.mode == "fused_ebc_uvm": print("Running DLRM EMB on FusedEBC with UVM/UVM-caching") embedding_bag_configs: List[EmbeddingBagConfig] = [ EmbeddingBagConfig( name=f"ebc_{idx}", embedding_dim=128, num_embeddings=num_embeddings, feature_names=[f"ebc_{idx}_feat_1"], ) for idx, num_embeddings in enumerate(get_shrunk_dlrm_num_embeddings(2)) ] fused_ebc_time_avg, fused_ebc_time_std = get_fused_ebc_uvm_time( embedding_bag_configs, device, EmbeddingLocation.MANAGED_CACHING ) print( f"FusedEBC with UVM caching on DLRM: {fused_ebc_time_avg} +/- {fused_ebc_time_std} sec" ) embedding_bag_configs: List[EmbeddingBagConfig] = [ EmbeddingBagConfig( name=f"ebc_{idx}", embedding_dim=128, num_embeddings=num_embeddings, feature_names=[f"ebc_{idx}_feat_1"], ) for idx, num_embeddings in enumerate(DLRM_NUM_EMBEDDINGS_PER_FEATURE) ] fused_ebc_time_avg, fused_ebc_time_std = get_fused_ebc_uvm_time( embedding_bag_configs, device, EmbeddingLocation.MANAGED ) print( f"FusedEBC with UVM management on DLRM: {fused_ebc_time_avg} plus/minus {fused_ebc_time_std} sec" ) elif args.mode == "ebc_comparison_scaling": print("Running EBC vs. FusedEBC scaling experiment") num_tables_list = [10, 100, 1000] embedding_dim_list = [4, 8, 16, 32, 64, 128] num_embeddings_list = [4, 8, 16, 32, 64, 128, 256, 1024, 2048, 4096, 8192] for num_tables in num_tables_list: for num_embeddings in num_embeddings_list: for embedding_dim in embedding_dim_list: embedding_bag_configs: List[EmbeddingBagConfig] = [ EmbeddingBagConfig( name=f"ebc_{idx}", embedding_dim=embedding_dim, num_embeddings=num_embeddings, feature_names=[f"ebc_{idx}_feat_1"], ) for idx in range(num_tables) ] ebc_time, _, fused_ebc_time, _, speedup = get_ebc_comparison( embedding_bag_configs, device, epochs=3 ) print( f"EBC num_tables = {num_tables}, num_embeddings = {num_embeddings}, embedding_dim = {embedding_dim}:" ) print( f"ebc_time = {ebc_time} sec, fused_ebc_time = {fused_ebc_time} sec, speedup = {speedup}" ) def invoke_main() -> None: main(sys.argv[1:])
null
8,863
from typing import Dict, Optional, Tuple, Union import torch import torch.nn as nn from torchrec.modules.embedding_modules import ( EmbeddingBagCollection, EmbeddingCollection, ) from torchrec.modules.mc_modules import ManagedCollisionCollection from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor class EmbeddingBagCollection(EmbeddingBagCollectionInterface): """ EmbeddingBagCollection represents a collection of pooled embeddings (`EmbeddingBags`). It processes sparse data in the form of `KeyedJaggedTensor` with values of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (jagged) and outputs a `KeyedTensor` with values of the form [B * (F * D)] where: * F: features (keys) * D: each feature's (key's) embedding dimension * B: batch size Args: tables (List[EmbeddingBagConfig]): list of embedding tables. is_weighted (bool): whether input `KeyedJaggedTensor` is weighted. device (Optional[torch.device]): default compute device. Example:: table_0 = EmbeddingBagConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) table_1 = EmbeddingBagConfig( name="t2", embedding_dim=4, num_embeddings=10, feature_names=["f2"] ) ebc = EmbeddingBagCollection(tables=[table_0, table_1]) # 0 1 2 <-- batch # "f1" [0,1] None [2] # "f2" [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) pooled_embeddings = ebc(features) print(pooled_embeddings.values()) tensor([[-0.8899, -0.1342, -1.9060, -0.0905, -0.2814, -0.9369, -0.7783], [ 0.0000, 0.0000, 0.0000, 0.1598, 0.0695, 1.3265, -0.1011], [-0.4256, -1.1846, -2.1648, -1.0893, 0.3590, -1.9784, -0.7681]], grad_fn=<CatBackward0>) print(pooled_embeddings.keys()) ['f1', 'f2'] print(pooled_embeddings.offset_per_key()) tensor([0, 3, 7]) """ def __init__( self, tables: List[EmbeddingBagConfig], is_weighted: bool = False, device: Optional[torch.device] = None, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self._is_weighted = is_weighted self.embedding_bags: nn.ModuleDict = nn.ModuleDict() self._embedding_bag_configs = tables self._lengths_per_embedding: List[int] = [] self._dtypes: List[int] = [] table_names = set() for embedding_config in tables: if embedding_config.name in table_names: raise ValueError(f"Duplicate table name {embedding_config.name}") table_names.add(embedding_config.name) dtype = ( torch.float32 if embedding_config.data_type == DataType.FP32 else torch.float16 ) self.embedding_bags[embedding_config.name] = nn.EmbeddingBag( num_embeddings=embedding_config.num_embeddings, embedding_dim=embedding_config.embedding_dim, mode=pooling_type_to_str(embedding_config.pooling), device=device, include_last_offset=True, dtype=dtype, ) if device is None: device = self.embedding_bags[embedding_config.name].weight.device self._dtypes.append(embedding_config.data_type.value) if not embedding_config.feature_names: embedding_config.feature_names = [embedding_config.name] self._lengths_per_embedding.extend( len(embedding_config.feature_names) * [embedding_config.embedding_dim] ) self._device: torch.device = device or torch.device("cpu") self._embedding_names: List[str] = [ embedding for embeddings in get_embedding_names_by_table(tables) for embedding in embeddings ] self._feature_names: List[List[str]] = [table.feature_names for table in tables] self.reset_parameters() def forward(self, features: KeyedJaggedTensor) -> KeyedTensor: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: KeyedTensor """ flat_feature_names: List[str] = [] for names in self._feature_names: flat_feature_names.extend(names) inverse_indices = reorder_inverse_indices( inverse_indices=features.inverse_indices_or_none(), feature_names=flat_feature_names, ) pooled_embeddings: List[torch.Tensor] = [] feature_dict = features.to_dict() for i, embedding_bag in enumerate(self.embedding_bags.values()): for feature_name in self._feature_names[i]: f = feature_dict[feature_name] per_sample_weights: Optional[torch.Tensor] = None if self._is_weighted: per_sample_weights = ( f.weights().half() if self._dtypes[i] == DataType.FP16.value else f.weights() ) res = embedding_bag( input=f.values(), offsets=f.offsets(), per_sample_weights=( per_sample_weights if self._is_weighted else None ), ).float() pooled_embeddings.append(res) return KeyedTensor( keys=self._embedding_names, values=process_pooled_embeddings( pooled_embeddings=pooled_embeddings, inverse_indices=inverse_indices, ), length_per_key=self._lengths_per_embedding, ) def is_weighted(self) -> bool: return self._is_weighted def embedding_bag_configs(self) -> List[EmbeddingBagConfig]: return self._embedding_bag_configs def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_bag_configs: assert table_config.init_fn is not None param = self.embedding_bags[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) class EmbeddingCollection(EmbeddingCollectionInterface): """ EmbeddingCollection represents a collection of non-pooled embeddings. It processes sparse data in the form of `KeyedJaggedTensor` of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (variable) and outputs `Dict[feature (key), JaggedTensor]`. Each `JaggedTensor` contains values of the form (B * L) X D where: * B: batch size * L: length of sparse features (jagged) * D: each feature's (key's) embedding dimension and lengths are of the form L Args: tables (List[EmbeddingConfig]): list of embedding tables. device (Optional[torch.device]): default compute device. need_indices (bool): if we need to pass indices to the final lookup dict. Example:: e1_config = EmbeddingConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) e2_config = EmbeddingConfig( name="t2", embedding_dim=3, num_embeddings=10, feature_names=["f2"] ) ec = EmbeddingCollection(tables=[e1_config, e2_config]) # 0 1 2 <-- batch # 0 [0,1] None [2] # 1 [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor.from_offsets_sync( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) feature_embeddings = ec(features) print(feature_embeddings['f2'].values()) tensor([[-0.2050, 0.5478, 0.6054], [ 0.7352, 0.3210, -3.0399], [ 0.1279, -0.1756, -0.4130], [ 0.7519, -0.4341, -0.0499], [ 0.9329, -1.0697, -0.8095]], grad_fn=<EmbeddingBackward>) """ def __init__( # noqa C901 self, tables: List[EmbeddingConfig], device: Optional[torch.device] = None, need_indices: bool = False, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self.embeddings: nn.ModuleDict = nn.ModuleDict() self._embedding_configs = tables self._embedding_dim: int = -1 self._need_indices: bool = need_indices self._device: torch.device = ( device if device is not None else torch.device("cpu") ) table_names = set() for config in tables: if config.name in table_names: raise ValueError(f"Duplicate table name {config.name}") table_names.add(config.name) self._embedding_dim = ( config.embedding_dim if self._embedding_dim < 0 else self._embedding_dim ) if self._embedding_dim != config.embedding_dim: raise ValueError( "All tables in a EmbeddingCollection are required to have same embedding dimension." + f" Violating case: {config.name}'s embedding_dim {config.embedding_dim} !=" + f" {self._embedding_dim}" ) dtype = ( torch.float32 if config.data_type == DataType.FP32 else torch.float16 ) self.embeddings[config.name] = nn.Embedding( num_embeddings=config.num_embeddings, embedding_dim=config.embedding_dim, device=device, dtype=dtype, ) if config.init_fn is not None: config.init_fn(self.embeddings[config.name].weight) if not config.feature_names: config.feature_names = [config.name] self._embedding_names_by_table: List[List[str]] = get_embedding_names_by_table( tables ) self._feature_names: List[List[str]] = [table.feature_names for table in tables] def forward( self, features: KeyedJaggedTensor, ) -> Dict[str, JaggedTensor]: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: Dict[str, JaggedTensor] """ feature_embeddings: Dict[str, JaggedTensor] = {} jt_dict: Dict[str, JaggedTensor] = features.to_dict() for i, emb_module in enumerate(self.embeddings.values()): feature_names = self._feature_names[i] embedding_names = self._embedding_names_by_table[i] for j, embedding_name in enumerate(embedding_names): feature_name = feature_names[j] f = jt_dict[feature_name] lookup = emb_module( input=f.values(), ).float() feature_embeddings[embedding_name] = JaggedTensor( values=lookup, lengths=f.lengths(), weights=f.values() if self._need_indices else None, ) return feature_embeddings def need_indices(self) -> bool: return self._need_indices def embedding_dim(self) -> int: return self._embedding_dim def embedding_configs(self) -> List[EmbeddingConfig]: return self._embedding_configs def embedding_names_by_table(self) -> List[List[str]]: return self._embedding_names_by_table def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_configs: assert table_config.init_fn is not None param = self.embeddings[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) def evict( evictions: Dict[str, Optional[torch.Tensor]], ebc: Union[EmbeddingBagCollection, EmbeddingCollection], ) -> None: # TODO: write function return
null
8,864
import abc from typing import Dict, Optional import torch from torch import nn from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor # pyre-ignore class KeyedJaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """Represents an (optionally weighted) keyed jagged tensor. A `KeyedJaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. Keyed on first dimension and jagged on the last dimension. Implementation is torch.jit.script-able. Args: keys (List[str]): keys to the jagged Tensor. values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if the values have weights. Tensor with the same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. stride (Optional[int]): number of examples per batch. stride_per_key_per_rank (Optional[List[List[int]]]): batch size (number of examples) per key per rank, with the outer list representing the keys and the inner list representing the values. Each value in the inner list represents the number of examples in the batch from the rank of its index in a distributed context. length_per_key (Optional[List[int]]): start length for each key. offset_per_key (Optional[List[int]]): start offset for each key and final offset. index_per_key (Optional[Dict[str, int]]): index for each key. jt_dict (Optional[Dict[str, JaggedTensor]]): inverse_indices (Optional[Tuple[List[str], torch.Tensor]]): inverse indices to expand deduplicated embedding output for variable stride per key. Example:: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 dim_0: keyed dimension (ie. `Feature0`, `Feature1`) dim_1: optional second dimension (ie. batch size) dim_2: The jagged dimension which has slice lengths between 0-3 in the above example # We represent this data with following inputs: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset """ # This is the subset of fields on KJT which are required (all other fields # can be derived from these fields, and are only cached) _fields = [ "_values", "_weights", "_lengths", "_offsets", ] def __init__( self, keys: List[str], values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, # Below exposed to ensure torch.script-able length_per_key: Optional[List[int]] = None, offset_per_key: Optional[List[int]] = None, index_per_key: Optional[Dict[str, int]] = None, jt_dict: Optional[Dict[str, JaggedTensor]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> None: self._keys: List[str] = keys self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets self._stride_per_key_per_rank: List[List[int]] = [] self._stride_per_key: List[int] = [] self._variable_stride_per_key: bool = False self._stride: int = -1 if stride_per_key_per_rank is not None: if stride is not None: raise ValueError( "Cannot initialize KJT with both `stride` and `stride_per_key_per_rank`" ) self._stride_per_key_per_rank = stride_per_key_per_rank self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] self._variable_stride_per_key = True if not stride_per_key_per_rank: self._stride = 0 elif all(s == self.stride_per_key()[0] for s in self.stride_per_key()): self._stride = self.stride_per_key()[0] else: if torch.jit.is_tracing(): stride = _maybe_compute_stride_kjt_scripted( keys, stride, lengths, offsets )[0] else: stride = _maybe_compute_stride_kjt(keys, stride, lengths, offsets) self._stride = stride self._stride_per_key_per_rank = [[stride]] * len(self._keys) self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] # lazy fields self._length_per_key: Optional[List[int]] = length_per_key self._offset_per_key: Optional[List[int]] = offset_per_key self._index_per_key: Optional[Dict[str, int]] = index_per_key self._jt_dict: Optional[Dict[str, JaggedTensor]] = jt_dict self._inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = ( inverse_indices ) self._lengths_offset_per_key: List[int] = [] def from_offsets_sync( keys: List[str], values: torch.Tensor, offsets: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, offsets=offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def from_lengths_sync( keys: List[str], values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def concat( kjt_list: List["KeyedJaggedTensor"], ) -> "KeyedJaggedTensor": if len(kjt_list) == 0: raise ValueError("Can't concat empty KJT list") is_weighted: bool = kjt_list[0].weights_or_none() is not None has_length_per_key: bool = True length_per_key: List[int] = [] keys: List[str] = [] value_list: List[torch.Tensor] = [] weight_list: List[torch.Tensor] = [] length_list: List[torch.Tensor] = [] stride_per_key_per_rank: List[List[int]] = [] stride: Optional[int] = None variable_stride_per_key_list = [ kjt.variable_stride_per_key() for kjt in kjt_list ] assert all(variable_stride_per_key_list) or not any( variable_stride_per_key_list ), "variable stride per key must be consistent for all KJTs" variable_stride_per_key = all(variable_stride_per_key_list) for kjt in kjt_list: curr_is_weighted: bool = kjt.weights_or_none() is not None if is_weighted != curr_is_weighted: raise ValueError("Can't merge weighted KJT with unweighted KJT") _length_per_key: Optional[List[int]] = None if kjt._length_per_key is None: has_length_per_key = False else: _length_per_key = kjt._length_per_key if has_length_per_key and _length_per_key is not None: length_per_key += _length_per_key keys += kjt.keys() value_list.append(kjt.values()) if is_weighted: weight_list.append(kjt.weights()) length_list.append(kjt.lengths()) if variable_stride_per_key: stride_per_key_per_rank += kjt.stride_per_key_per_rank() elif stride is None: stride = kjt.stride() else: assert stride == kjt.stride(), "strides must be consistent for all KJTs" return KeyedJaggedTensor( keys=keys, values=torch.cat(value_list, dim=0), weights=torch.cat(weight_list, dim=0) if is_weighted else None, lengths=torch.cat(length_list, dim=0), stride=stride, stride_per_key_per_rank=( stride_per_key_per_rank if variable_stride_per_key else None ), length_per_key=length_per_key if has_length_per_key else None, ) def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "KeyedJaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return KeyedJaggedTensor( keys=torch.jit.annotate(List[str], []), values=torch.empty(0, dtype=values_dtype, device=device), weights=weights, lengths=torch.empty(0, dtype=lengths_dtype, device=device), stride=0, ) def empty_like(kjt: "KeyedJaggedTensor") -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, kjt.stride_per_key_per_rank()) if kjt.variable_stride_per_key() else (kjt.stride(), None) ) return KeyedJaggedTensor( keys=[], values=torch.empty(0, device=kjt.device(), dtype=kjt.values().dtype), weights=( None if kjt.weights_or_none() is None else torch.empty(0, device=kjt.device(), dtype=kjt.weights().dtype) ), lengths=torch.empty(0, device=kjt.device(), dtype=kjt.lengths().dtype), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, ) def from_jt_dict(jt_dict: Dict[str, JaggedTensor]) -> "KeyedJaggedTensor": """ Constructs a KeyedJaggedTensor from a Dict[str, JaggedTensor], but this function will ONLY work if the JaggedTensors all have the same "implicit" batch_size dimension. Basically, we can visualize JaggedTensors as 2-D tensors of the format of [batch_size x variable_feature_dim]. In case, we have some batch without a feature value, the input JaggedTensor could just not include any values. But KeyedJaggedTensor (by default) typically pad "None" so that all the JaggedTensors stored in the KeyedJaggedTensor have the same batch_size dimension. That is, in the case, the JaggedTensor input didn't automatically pad for the empty batches, this function would error / not work. Consider the visualization of the following KeyedJaggedTensor: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 Notice that the inputs for this KeyedJaggedTensor would have looked like: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset Now if the input jt_dict = { # "Feature0" [V0,V1] [V2] # "Feature1" [V3] [V4] [V5,V6,V7] } and the "None" is left out from each JaggedTensor, then this function would fail as we would not correctly be able to pad "None" as it does not technically know the correct batch / place to pad within the JaggedTensor. Essentially, the lengths Tensor inferred by this function would be [2, 1, 1, 1, 3] indicating variable batch_size dim_1 violates the existing assumption / precondition that KeyedJaggedTensor's should have fixed batch_size dimension. """ kjt_keys = list(jt_dict.keys()) kjt_vals_list: List[torch.Tensor] = [] kjt_lens_list: List[torch.Tensor] = [] kjt_weights_list: List[torch.Tensor] = [] stride_per_key: List[int] = [] for jt in jt_dict.values(): stride_per_key.append(len(jt.lengths())) kjt_vals_list.append(jt.values()) kjt_lens_list.append(jt.lengths()) weight = jt.weights_or_none() if weight is not None: kjt_weights_list.append(weight) kjt_vals = torch.concat(kjt_vals_list) kjt_lens = torch.concat(kjt_lens_list) kjt_weights = ( torch.concat(kjt_weights_list) if len(kjt_weights_list) > 0 else None ) kjt_stride, kjt_stride_per_key_per_rank = ( (stride_per_key[0], None) if all(s == stride_per_key[0] for s in stride_per_key) else (None, [[stride] for stride in stride_per_key]) ) kjt = KeyedJaggedTensor( keys=kjt_keys, values=kjt_vals, weights=kjt_weights, lengths=kjt_lens, stride=kjt_stride, stride_per_key_per_rank=kjt_stride_per_key_per_rank, ).sync() return kjt def sync(self) -> "KeyedJaggedTensor": self.length_per_key() self.offset_per_key() return self def unsync(self) -> "KeyedJaggedTensor": self._length_per_key = None self._offset_per_key = None return self def device(self) -> torch.device: return self._values.device def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def keys(self) -> List[str]: return self._keys def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def stride(self) -> int: return self._stride def stride_per_key(self) -> List[int]: return self._stride_per_key def stride_per_key_per_rank(self) -> List[List[int]]: return self._stride_per_key_per_rank def variable_stride_per_key(self) -> bool: return self._variable_stride_per_key def inverse_indices(self) -> Tuple[List[str], torch.Tensor]: return _get_inverse_indices_or_throw(self._inverse_indices) def inverse_indices_or_none(self) -> Optional[Tuple[List[str], torch.Tensor]]: return self._inverse_indices def _key_indices(self) -> Dict[str, int]: _index_per_key: Dict[str, int] = _maybe_compute_index_per_key( self._keys, self._index_per_key, ) self._index_per_key = _index_per_key return _index_per_key def length_per_key(self) -> List[int]: _length_per_key = _maybe_compute_length_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key return _length_per_key def length_per_key_or_none(self) -> Optional[List[int]]: return self._length_per_key def offset_per_key(self) -> List[int]: _length_per_key, _offset_per_key = _maybe_compute_offset_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key self._offset_per_key = _offset_per_key return _offset_per_key def offset_per_key_or_none(self) -> Optional[List[int]]: return self._offset_per_key def lengths_offset_per_key(self) -> List[int]: if not self._lengths_offset_per_key: self._lengths_offset_per_key = _cumsum(self.stride_per_key()) return self._lengths_offset_per_key def split(self, segments: List[int]) -> List["KeyedJaggedTensor"]: split_list: List[KeyedJaggedTensor] = [] start = 0 start_offset = 0 _length_per_key = self.length_per_key() _offset_per_key = self.offset_per_key() for segment in segments: end = start + segment end_offset = _offset_per_key[end] keys: List[str] = self._keys[start:end] stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()[start:end]) if self.variable_stride_per_key() else (self._stride, None) ) if segment == len(self._keys): # no torch slicing required split_list.append( KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self.weights_or_none(), lengths=self._lengths, offsets=self._offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=self._jt_dict, inverse_indices=None, ) ) elif segment == 0: empty_int_list: List[int] = torch.jit.annotate(List[int], []) split_list.append( KeyedJaggedTensor( keys=keys, values=torch.tensor( empty_int_list, device=self.device(), dtype=self._values.dtype, ), weights=( None if self.weights_or_none() is None else torch.tensor( empty_int_list, device=self.device(), dtype=self.weights().dtype, ) ), lengths=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), offsets=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) else: split_length_per_key = _length_per_key[start:end] if not torch.jit.is_scripting() and is_torchdynamo_compiling(): # Checks for dynamo dynamic shapes tracing torch._check_is_size(start_offset) torch._check_is_size(end_offset) torch._check_is_size(end_offset - start_offset) torch._check(start_offset <= self._values.size(0)) torch._check(end_offset <= self._values.size(0)) torch._check(end_offset >= start_offset) split_list.append( KeyedJaggedTensor( keys=keys, values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[ start ] : self.lengths_offset_per_key()[end] ], offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=split_length_per_key, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) start = end start_offset = end_offset return split_list def permute( self, indices: List[int], indices_tensor: Optional[torch.Tensor] = None, include_inverse_indices: bool = False, ) -> "KeyedJaggedTensor": if indices_tensor is None: indices_tensor = torch.tensor( indices, dtype=torch.int, device=self.device() ) length_per_key = self.length_per_key() permuted_keys: List[str] = [] permuted_stride_per_key_per_rank: List[List[int]] = [] permuted_length_per_key: List[int] = [] permuted_lengths_sum = 0 for index in indices: key = self.keys()[index] permuted_keys.append(key) permuted_stride_per_key_per_rank.append( self.stride_per_key_per_rank()[index] ) permuted_length_per_key.append(length_per_key[index]) permuted_lengths_sum += length_per_key[index] if self.variable_stride_per_key(): length_per_key_tensor = _pin_and_move( torch.tensor(self.length_per_key()), self.device() ) stride_per_key_tensor = _pin_and_move( torch.tensor(self.stride_per_key()), self.device() ) permuted_lengths, _ = _permute_tensor_by_segments( self.lengths(), stride_per_key_tensor, indices_tensor, None, ) permuted_values, permuted_weights = _permute_tensor_by_segments( self.values(), length_per_key_tensor, indices_tensor, self.weights_or_none(), ) else: ( permuted_lengths, permuted_values, permuted_weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( indices_tensor, self.lengths().view(len(self._keys), -1), self.values(), self.weights_or_none(), permuted_lengths_sum, ) stride, optional_permuted_stride_per_key_per_rank = ( (None, permuted_stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) kjt = KeyedJaggedTensor( keys=permuted_keys, values=permuted_values, weights=permuted_weights, lengths=permuted_lengths.view(-1), offsets=None, stride=stride, stride_per_key_per_rank=optional_permuted_stride_per_key_per_rank, length_per_key=permuted_length_per_key if len(permuted_keys) > 0 else None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=( self.inverse_indices_or_none() if include_inverse_indices else None ), ) return kjt def flatten_lengths(self) -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()) if self.variable_stride_per_key() else (self._stride, None) ) return KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self._weights, lengths=self.lengths().view(-1), offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self.length_per_key(), offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) def __getitem__(self, key: str) -> JaggedTensor: offset_per_key = self.offset_per_key() index = self._key_indices()[key] start_offset = offset_per_key[index] end_offset = ( offset_per_key[index + 1] if index + 1 < len(offset_per_key) else start_offset ) return JaggedTensor( values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[index] : self.lengths_offset_per_key()[ index + 1 ] ], offsets=None, ) def to_dict(self) -> Dict[str, JaggedTensor]: _jt_dict = _maybe_compute_kjt_to_jt_dict( stride=self.stride(), stride_per_key=self.stride_per_key(), keys=self.keys(), length_per_key=self.length_per_key(), lengths=self.lengths(), values=self.values(), variable_stride_per_key=self.variable_stride_per_key(), weights=self.weights_or_none(), jt_dict=self._jt_dict, ) self._jt_dict = _jt_dict return _jt_dict def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def to( self, device: torch.device, non_blocking: bool = False, dtype: Optional[torch.dtype] = None, ) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) length_per_key = self._length_per_key offset_per_key = self._offset_per_key index_per_key = self._index_per_key jt_dict = self._jt_dict inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = ( inverse_indices[0], inverse_indices[1].to(device, non_blocking=non_blocking), ) if weights is not None: if dtype is not None: weights = weights.to( dtype=dtype, device=device, non_blocking=non_blocking ) else: weights = weights.to(device=device, non_blocking=non_blocking) return KeyedJaggedTensor( keys=self._keys, values=self._values.to(device, non_blocking=non_blocking), weights=weights, lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=length_per_key, offset_per_key=offset_per_key, index_per_key=index_per_key, jt_dict=jt_dict, inverse_indices=inverse_indices, ) def __str__(self) -> str: if len(self._keys) == 0 or self._offsets is None and self._lengths is None: return "KeyedJaggedTensor()\n" offsets = self.offsets() return ( "KeyedJaggedTensor({\n" + ",\n".join( [ " " + _jagged_tensor_string( self._keys[index], self._values, self._weights, offsets, sum(self.stride_per_key()[:index]), sum(self.stride_per_key()[: index + 1]), ) for index in range(len(self._keys)) ] ) + "\n})\n" ) def pin_memory(self) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = (inverse_indices[0], inverse_indices[1].pin_memory()) return KeyedJaggedTensor( keys=self._keys, values=self._values.pin_memory(), weights=weights.pin_memory() if weights is not None else None, lengths=lengths.pin_memory() if lengths is not None else None, offsets=offsets.pin_memory() if offsets is not None else None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=None, inverse_indices=inverse_indices, ) def dist_labels(self) -> List[str]: labels = ["lengths", "values"] if self.variable_stride_per_key(): labels.append("strides") if self.weights_or_none() is not None: labels.append("weights") return labels def dist_splits(self, key_splits: List[int]) -> List[List[int]]: batch_size_per_split = _sum_by_splits(self.stride_per_key(), key_splits) length_per_split = _sum_by_splits(self.length_per_key(), key_splits) splits = [batch_size_per_split, length_per_split] if self.variable_stride_per_key(): splits.append(key_splits) if self.weights_or_none() is not None: splits.append(length_per_split) return splits def dist_tensors(self) -> List[torch.Tensor]: tensors = [self.lengths(), self.values()] if self.variable_stride_per_key(): strides = _pin_and_move(torch.tensor(self.stride_per_key()), self.device()) tensors.append(strides) if self.weights_or_none() is not None: tensors.append(self.weights()) return tensors def dist_init( keys: List[str], tensors: List[torch.Tensor], variable_stride_per_key: bool, num_workers: int, recat: Optional[torch.Tensor], stride_per_rank: Optional[List[int]], stagger: int = 1, ) -> "KeyedJaggedTensor": assert len(tensors) in [2, 3, 4] lengths = tensors[0] values = tensors[1] stride_per_rank_per_key = tensors[2] if variable_stride_per_key else None weights = ( tensors[-1] if (variable_stride_per_key and len(tensors) == 4) or (not variable_stride_per_key and len(tensors) == 3) else None ) if variable_stride_per_key: assert stride_per_rank_per_key is not None stride_per_key_per_rank: List[List[int]] = stride_per_rank_per_key.view( num_workers, len(keys) ).T.tolist() strides_cumsum: List[int] = torch.ops.fbgemm.asynchronous_complete_cumsum( stride_per_rank_per_key ).tolist() cumsum_lengths = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) length_per_key = ( cumsum_lengths[strides_cumsum[1:]] - cumsum_lengths[strides_cumsum[:-1]] ) with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: lengths, _ = _permute_tensor_by_segments( lengths, stride_per_rank_per_key, recat, None, ) values, weights = _permute_tensor_by_segments( values, length_per_key, recat, weights, ) if not stride_per_key_per_rank: stride_per_key_per_rank = [[0]] * len(keys) if stagger > 1: stride_per_key_per_rank_stagger: List[List[int]] = [] local_world_size = num_workers // stagger for i in range(len(keys)): stride_per_rank_stagger: List[int] = [] for j in range(local_world_size): stride_per_rank_stagger.extend( stride_per_key_per_rank[i][j::local_world_size] ) stride_per_key_per_rank_stagger.append(stride_per_rank_stagger) stride_per_key_per_rank = stride_per_key_per_rank_stagger kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride_per_key_per_rank=stride_per_key_per_rank, ) return kjt.sync() else: assert stride_per_rank is not None with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: stride = stride_per_rank[0] if all(s == stride for s in stride_per_rank): ( lengths, values, weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( recat, lengths.view(-1, stride), values, weights, values.numel(), ) lengths = lengths.view(-1) else: # variable batch size per rank ( lengths, values, weights, ) = torch.ops.fbgemm.permute_1D_sparse_data( recat, lengths.view(-1), values, weights, values.numel(), ) kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=sum(stride_per_rank), ) return kjt.sync() def get_weights_list( cat_seq: torch.Tensor, features: KeyedJaggedTensor, position_weights: Dict[str, nn.Parameter], ) -> Optional[torch.Tensor]: weights_list = [] seqs = torch.split(cat_seq, features.length_per_key()) for key, seq in zip(features.keys(), seqs): if key in position_weights.keys(): weights_list.append(torch.gather(position_weights[key], dim=0, index=seq)) else: weights_list.append( torch.ones(seq.shape[0], device=features.values().device) ) return torch.cat(weights_list) if weights_list else features.weights_or_none()
null
8,865
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def extract_module_or_tensor_callable( module_or_callable: Union[ Callable[[], torch.nn.Module], torch.nn.Module, Callable[[torch.Tensor], torch.Tensor], ] ) -> Union[torch.nn.Module, Callable[[torch.Tensor], torch.Tensor]]: try: # pyre-ignore[20]: PositionalOnly call expects argument in position 0 module = module_or_callable() if isinstance(module, torch.nn.Module): return module else: raise ValueError( "Expected callable that takes no input to return " "a torch.nn.Module, but got: {}".format(type(module)) ) except TypeError as e: if "required positional argument" in str(e): # pyre-ignore[7]: Expected `Union[typing.Callable[[torch.Tensor], torch.Tensor], torch.nn.Module]` return module_or_callable raise
null
8,866
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def get_module_output_dimension( module: Union[Callable[[torch.Tensor], torch.Tensor], torch.nn.Module], in_features: int, ) -> int: input = torch.zeros(1, in_features) output = module(input) return output.size(-1) The provided code snippet includes necessary dependencies for implementing the `check_module_output_dimension` function. Write a Python function `def check_module_output_dimension( module: Union[Iterable[torch.nn.Module], torch.nn.Module], in_features: int, out_features: int, ) -> bool` to solve the following problem: Verify that the out_features of a given module or a list of modules matches the specified number. If a list of modules or a ModuleList is given, recursively check all the submodules. Here is the function: def check_module_output_dimension( module: Union[Iterable[torch.nn.Module], torch.nn.Module], in_features: int, out_features: int, ) -> bool: """ Verify that the out_features of a given module or a list of modules matches the specified number. If a list of modules or a ModuleList is given, recursively check all the submodules. """ if isinstance(module, list) or isinstance(module, torch.nn.ModuleList): return all( check_module_output_dimension(submodule, in_features, out_features) for submodule in module ) else: # pyre-fixme[6]: Expected `Union[typing.Callable[[torch.Tensor], # torch.Tensor], torch.nn.Module]` for 1st param but got # `Union[Iterable[torch.nn.Module], torch.nn.Module]`. return get_module_output_dimension(module, in_features) == out_features
Verify that the out_features of a given module or a list of modules matches the specified number. If a list of modules or a ModuleList is given, recursively check all the submodules.
8,867
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def init_mlp_weights_xavier_uniform(m: torch.nn.Module) -> None: if isinstance(m, torch.nn.Linear): torch.nn.init.xavier_uniform_(m.weight) m.bias.data.fill_(0.0) The provided code snippet includes necessary dependencies for implementing the `construct_modulelist_from_single_module` function. Write a Python function `def construct_modulelist_from_single_module( module: torch.nn.Module, sizes: Tuple[int, ...] ) -> torch.nn.Module` to solve the following problem: Given a single module, construct a (nested) ModuleList of size of sizes by making copies of the provided module and reinitializing the Linear layers. Here is the function: def construct_modulelist_from_single_module( module: torch.nn.Module, sizes: Tuple[int, ...] ) -> torch.nn.Module: """ Given a single module, construct a (nested) ModuleList of size of sizes by making copies of the provided module and reinitializing the Linear layers. """ if len(sizes) == 1: return torch.nn.ModuleList( [ copy.deepcopy(module).apply(init_mlp_weights_xavier_uniform) for _ in range(sizes[0]) ] ) else: # recursively create nested ModuleList return torch.nn.ModuleList( [ construct_modulelist_from_single_module(module, sizes[1:]) for _ in range(sizes[0]) ] )
Given a single module, construct a (nested) ModuleList of size of sizes by making copies of the provided module and reinitializing the Linear layers.
8,868
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def convert_list_of_modules_to_modulelist( modules: Iterable[torch.nn.Module], sizes: Tuple[int, ...] ) -> torch.nn.Module: assert ( # pyre-fixme[6]: Expected `Sized` for 1st param but got # `Iterable[torch.nn.Module]`. len(modules) == sizes[0] ), f"the counts of modules ({len(modules)}) do not match with the required counts {sizes}" if len(sizes) == 1: return torch.nn.ModuleList(modules) else: # recursively create nested list return torch.nn.ModuleList( convert_list_of_modules_to_modulelist(m, sizes[1:]) for m in modules )
null
8,869
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def _permute_indices(indices: List[int], permute: List[int]) -> List[int]: permuted_indices = [0] * len(indices) for i, permuted_index in enumerate(permute): permuted_indices[i] = indices[permuted_index] return permuted_indices # pyre-ignore class JaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """ Represents an (optionally weighted) jagged tensor. A `JaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. See `KeyedJaggedTensor` for full example. Implementation is torch.jit.script-able. NOTE: We will NOT do input validation as it's expensive, you should always pass in the valid lengths, offsets, etc. Args: values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if values have weights. Tensor with same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. """ _fields = ["_values", "_weights", "_lengths", "_offsets"] def __init__( self, values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, ) -> None: self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights _assert_offsets_or_lengths_is_provided(offsets, lengths) if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "JaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return JaggedTensor( values=torch.empty(0, dtype=values_dtype, device=device), offsets=torch.empty(0, dtype=lengths_dtype, device=device), lengths=torch.empty(0, dtype=lengths_dtype, device=device), weights=weights, ) def from_dense_lengths( values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` is still of shape (B,). """ mask2d = ( _arange(end=values.size(1), device=values.device).expand(values.size(0), -1) ) < lengths.unsqueeze(-1) return JaggedTensor( values=values[mask2d], weights=_optional_mask(weights, mask2d), lengths=lengths, ) def from_dense( values: List[torch.Tensor], weights: Optional[List[torch.Tensor]] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` and `offsets` are still of shape (B,). Args: values (List[torch.Tensor]): a list of tensors for dense representation weights (Optional[List[torch.Tensor]]): if values have weights, tensor with the same shape as values. Returns: JaggedTensor: JaggedTensor created from 2D dense tensor. Example:: values = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] weights = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] j1 = JaggedTensor.from_dense( values=values, weights=weights, ) # j1 = [[1.0], [], [7.0], [8.0], [10.0, 11.0, 12.0]] """ values_tensor = torch.cat(values, dim=0) lengths = torch.tensor( [value.size(0) for value in values], dtype=torch.int32, device=values_tensor.device, ) weights_tensor = torch.cat(weights, dim=0) if weights is not None else None return JaggedTensor( values=values_tensor, weights=weights_tensor, lengths=lengths, ) def to_dense(self) -> List[torch.Tensor]: """ Constructs a dense-representation of the JT's values. Returns: List[torch.Tensor]: list of tensors. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) values_list = jt.to_dense() # values_list = [ # torch.tensor([1.0, 2.0]), # torch.tensor([]), # torch.tensor([3.0]), # torch.tensor([4.0]), # torch.tensor([5.0]), # torch.tensor([6.0, 7.0, 8.0]), # ] """ tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.values()[offset:next_offset]) return tensor_list def to_dense_weights(self) -> Optional[List[torch.Tensor]]: """ Constructs a dense-representation of the JT's weights. Returns: Optional[List[torch.Tensor]]: list of tensors, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) weights_list = jt.to_dense_weights() # weights_list = [ # torch.tensor([0.1, 0.2]), # torch.tensor([]), # torch.tensor([0.3]), # torch.tensor([0.4]), # torch.tensor([0.5]), # torch.tensor([0.6, 0.7, 0.8]), # ] """ if self.weights_or_none() is None: return None tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.weights()[offset:next_offset]) return tensor_list def to_padded_dense( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> torch.Tensor: """ Constructs a 2D dense tensor from the JT's values of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: torch.Tensor: 2d dense tensor. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) dt = jt.to_padded_dense( desired_length=2, padding_value=10.0, ) # dt = [ # [1.0, 2.0], # [10.0, 10.0], # [3.0, 10.0], # [4.0, 10.0], # [5.0, 10.0], # [6.0, 7.0], # ] """ if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.values(), [self.offsets()], [N], padding_value ) def to_padded_dense_weights( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> Optional[torch.Tensor]: """ Constructs a 2D dense tensor from the JT's weights of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: Optional[torch.Tensor]: 2d dense tensor, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) d_wt = jt.to_padded_dense_weights( desired_length=2, padding_value=1.0, ) # d_wt = [ # [0.1, 0.2], # [1.0, 1.0], # [0.3, 1.0], # [0.4, 1.0], # [0.5, 1.0], # [0.6, 0.7], # ] """ if self.weights_or_none() is None: return None if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.weights(), [self.offsets()], [N], padding_value ) def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def to(self, device: torch.device, non_blocking: bool = False) -> "JaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets return JaggedTensor( values=self._values.to(device, non_blocking=non_blocking), weights=( weights.to(device, non_blocking=non_blocking) if weights is not None else None ), lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def __str__(self) -> str: offsets = self.offsets() if self._weights is None: return ( "JaggedTensor({\n " + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + "\n})\n" ) return ( "JaggedTensor({\n" + ' "values": ' + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + ',\n "weights": ' + _jagged_values_string( _get_weights_or_throw(self._weights), offsets, 0, len(offsets) - 1 ) + "\n})\n" ) class KeyedJaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """Represents an (optionally weighted) keyed jagged tensor. A `KeyedJaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. Keyed on first dimension and jagged on the last dimension. Implementation is torch.jit.script-able. Args: keys (List[str]): keys to the jagged Tensor. values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if the values have weights. Tensor with the same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. stride (Optional[int]): number of examples per batch. stride_per_key_per_rank (Optional[List[List[int]]]): batch size (number of examples) per key per rank, with the outer list representing the keys and the inner list representing the values. Each value in the inner list represents the number of examples in the batch from the rank of its index in a distributed context. length_per_key (Optional[List[int]]): start length for each key. offset_per_key (Optional[List[int]]): start offset for each key and final offset. index_per_key (Optional[Dict[str, int]]): index for each key. jt_dict (Optional[Dict[str, JaggedTensor]]): inverse_indices (Optional[Tuple[List[str], torch.Tensor]]): inverse indices to expand deduplicated embedding output for variable stride per key. Example:: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 dim_0: keyed dimension (ie. `Feature0`, `Feature1`) dim_1: optional second dimension (ie. batch size) dim_2: The jagged dimension which has slice lengths between 0-3 in the above example # We represent this data with following inputs: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset """ # This is the subset of fields on KJT which are required (all other fields # can be derived from these fields, and are only cached) _fields = [ "_values", "_weights", "_lengths", "_offsets", ] def __init__( self, keys: List[str], values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, # Below exposed to ensure torch.script-able length_per_key: Optional[List[int]] = None, offset_per_key: Optional[List[int]] = None, index_per_key: Optional[Dict[str, int]] = None, jt_dict: Optional[Dict[str, JaggedTensor]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> None: self._keys: List[str] = keys self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets self._stride_per_key_per_rank: List[List[int]] = [] self._stride_per_key: List[int] = [] self._variable_stride_per_key: bool = False self._stride: int = -1 if stride_per_key_per_rank is not None: if stride is not None: raise ValueError( "Cannot initialize KJT with both `stride` and `stride_per_key_per_rank`" ) self._stride_per_key_per_rank = stride_per_key_per_rank self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] self._variable_stride_per_key = True if not stride_per_key_per_rank: self._stride = 0 elif all(s == self.stride_per_key()[0] for s in self.stride_per_key()): self._stride = self.stride_per_key()[0] else: if torch.jit.is_tracing(): stride = _maybe_compute_stride_kjt_scripted( keys, stride, lengths, offsets )[0] else: stride = _maybe_compute_stride_kjt(keys, stride, lengths, offsets) self._stride = stride self._stride_per_key_per_rank = [[stride]] * len(self._keys) self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] # lazy fields self._length_per_key: Optional[List[int]] = length_per_key self._offset_per_key: Optional[List[int]] = offset_per_key self._index_per_key: Optional[Dict[str, int]] = index_per_key self._jt_dict: Optional[Dict[str, JaggedTensor]] = jt_dict self._inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = ( inverse_indices ) self._lengths_offset_per_key: List[int] = [] def from_offsets_sync( keys: List[str], values: torch.Tensor, offsets: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, offsets=offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def from_lengths_sync( keys: List[str], values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def concat( kjt_list: List["KeyedJaggedTensor"], ) -> "KeyedJaggedTensor": if len(kjt_list) == 0: raise ValueError("Can't concat empty KJT list") is_weighted: bool = kjt_list[0].weights_or_none() is not None has_length_per_key: bool = True length_per_key: List[int] = [] keys: List[str] = [] value_list: List[torch.Tensor] = [] weight_list: List[torch.Tensor] = [] length_list: List[torch.Tensor] = [] stride_per_key_per_rank: List[List[int]] = [] stride: Optional[int] = None variable_stride_per_key_list = [ kjt.variable_stride_per_key() for kjt in kjt_list ] assert all(variable_stride_per_key_list) or not any( variable_stride_per_key_list ), "variable stride per key must be consistent for all KJTs" variable_stride_per_key = all(variable_stride_per_key_list) for kjt in kjt_list: curr_is_weighted: bool = kjt.weights_or_none() is not None if is_weighted != curr_is_weighted: raise ValueError("Can't merge weighted KJT with unweighted KJT") _length_per_key: Optional[List[int]] = None if kjt._length_per_key is None: has_length_per_key = False else: _length_per_key = kjt._length_per_key if has_length_per_key and _length_per_key is not None: length_per_key += _length_per_key keys += kjt.keys() value_list.append(kjt.values()) if is_weighted: weight_list.append(kjt.weights()) length_list.append(kjt.lengths()) if variable_stride_per_key: stride_per_key_per_rank += kjt.stride_per_key_per_rank() elif stride is None: stride = kjt.stride() else: assert stride == kjt.stride(), "strides must be consistent for all KJTs" return KeyedJaggedTensor( keys=keys, values=torch.cat(value_list, dim=0), weights=torch.cat(weight_list, dim=0) if is_weighted else None, lengths=torch.cat(length_list, dim=0), stride=stride, stride_per_key_per_rank=( stride_per_key_per_rank if variable_stride_per_key else None ), length_per_key=length_per_key if has_length_per_key else None, ) def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "KeyedJaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return KeyedJaggedTensor( keys=torch.jit.annotate(List[str], []), values=torch.empty(0, dtype=values_dtype, device=device), weights=weights, lengths=torch.empty(0, dtype=lengths_dtype, device=device), stride=0, ) def empty_like(kjt: "KeyedJaggedTensor") -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, kjt.stride_per_key_per_rank()) if kjt.variable_stride_per_key() else (kjt.stride(), None) ) return KeyedJaggedTensor( keys=[], values=torch.empty(0, device=kjt.device(), dtype=kjt.values().dtype), weights=( None if kjt.weights_or_none() is None else torch.empty(0, device=kjt.device(), dtype=kjt.weights().dtype) ), lengths=torch.empty(0, device=kjt.device(), dtype=kjt.lengths().dtype), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, ) def from_jt_dict(jt_dict: Dict[str, JaggedTensor]) -> "KeyedJaggedTensor": """ Constructs a KeyedJaggedTensor from a Dict[str, JaggedTensor], but this function will ONLY work if the JaggedTensors all have the same "implicit" batch_size dimension. Basically, we can visualize JaggedTensors as 2-D tensors of the format of [batch_size x variable_feature_dim]. In case, we have some batch without a feature value, the input JaggedTensor could just not include any values. But KeyedJaggedTensor (by default) typically pad "None" so that all the JaggedTensors stored in the KeyedJaggedTensor have the same batch_size dimension. That is, in the case, the JaggedTensor input didn't automatically pad for the empty batches, this function would error / not work. Consider the visualization of the following KeyedJaggedTensor: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 Notice that the inputs for this KeyedJaggedTensor would have looked like: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset Now if the input jt_dict = { # "Feature0" [V0,V1] [V2] # "Feature1" [V3] [V4] [V5,V6,V7] } and the "None" is left out from each JaggedTensor, then this function would fail as we would not correctly be able to pad "None" as it does not technically know the correct batch / place to pad within the JaggedTensor. Essentially, the lengths Tensor inferred by this function would be [2, 1, 1, 1, 3] indicating variable batch_size dim_1 violates the existing assumption / precondition that KeyedJaggedTensor's should have fixed batch_size dimension. """ kjt_keys = list(jt_dict.keys()) kjt_vals_list: List[torch.Tensor] = [] kjt_lens_list: List[torch.Tensor] = [] kjt_weights_list: List[torch.Tensor] = [] stride_per_key: List[int] = [] for jt in jt_dict.values(): stride_per_key.append(len(jt.lengths())) kjt_vals_list.append(jt.values()) kjt_lens_list.append(jt.lengths()) weight = jt.weights_or_none() if weight is not None: kjt_weights_list.append(weight) kjt_vals = torch.concat(kjt_vals_list) kjt_lens = torch.concat(kjt_lens_list) kjt_weights = ( torch.concat(kjt_weights_list) if len(kjt_weights_list) > 0 else None ) kjt_stride, kjt_stride_per_key_per_rank = ( (stride_per_key[0], None) if all(s == stride_per_key[0] for s in stride_per_key) else (None, [[stride] for stride in stride_per_key]) ) kjt = KeyedJaggedTensor( keys=kjt_keys, values=kjt_vals, weights=kjt_weights, lengths=kjt_lens, stride=kjt_stride, stride_per_key_per_rank=kjt_stride_per_key_per_rank, ).sync() return kjt def sync(self) -> "KeyedJaggedTensor": self.length_per_key() self.offset_per_key() return self def unsync(self) -> "KeyedJaggedTensor": self._length_per_key = None self._offset_per_key = None return self def device(self) -> torch.device: return self._values.device def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def keys(self) -> List[str]: return self._keys def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def stride(self) -> int: return self._stride def stride_per_key(self) -> List[int]: return self._stride_per_key def stride_per_key_per_rank(self) -> List[List[int]]: return self._stride_per_key_per_rank def variable_stride_per_key(self) -> bool: return self._variable_stride_per_key def inverse_indices(self) -> Tuple[List[str], torch.Tensor]: return _get_inverse_indices_or_throw(self._inverse_indices) def inverse_indices_or_none(self) -> Optional[Tuple[List[str], torch.Tensor]]: return self._inverse_indices def _key_indices(self) -> Dict[str, int]: _index_per_key: Dict[str, int] = _maybe_compute_index_per_key( self._keys, self._index_per_key, ) self._index_per_key = _index_per_key return _index_per_key def length_per_key(self) -> List[int]: _length_per_key = _maybe_compute_length_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key return _length_per_key def length_per_key_or_none(self) -> Optional[List[int]]: return self._length_per_key def offset_per_key(self) -> List[int]: _length_per_key, _offset_per_key = _maybe_compute_offset_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key self._offset_per_key = _offset_per_key return _offset_per_key def offset_per_key_or_none(self) -> Optional[List[int]]: return self._offset_per_key def lengths_offset_per_key(self) -> List[int]: if not self._lengths_offset_per_key: self._lengths_offset_per_key = _cumsum(self.stride_per_key()) return self._lengths_offset_per_key def split(self, segments: List[int]) -> List["KeyedJaggedTensor"]: split_list: List[KeyedJaggedTensor] = [] start = 0 start_offset = 0 _length_per_key = self.length_per_key() _offset_per_key = self.offset_per_key() for segment in segments: end = start + segment end_offset = _offset_per_key[end] keys: List[str] = self._keys[start:end] stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()[start:end]) if self.variable_stride_per_key() else (self._stride, None) ) if segment == len(self._keys): # no torch slicing required split_list.append( KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self.weights_or_none(), lengths=self._lengths, offsets=self._offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=self._jt_dict, inverse_indices=None, ) ) elif segment == 0: empty_int_list: List[int] = torch.jit.annotate(List[int], []) split_list.append( KeyedJaggedTensor( keys=keys, values=torch.tensor( empty_int_list, device=self.device(), dtype=self._values.dtype, ), weights=( None if self.weights_or_none() is None else torch.tensor( empty_int_list, device=self.device(), dtype=self.weights().dtype, ) ), lengths=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), offsets=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) else: split_length_per_key = _length_per_key[start:end] if not torch.jit.is_scripting() and is_torchdynamo_compiling(): # Checks for dynamo dynamic shapes tracing torch._check_is_size(start_offset) torch._check_is_size(end_offset) torch._check_is_size(end_offset - start_offset) torch._check(start_offset <= self._values.size(0)) torch._check(end_offset <= self._values.size(0)) torch._check(end_offset >= start_offset) split_list.append( KeyedJaggedTensor( keys=keys, values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[ start ] : self.lengths_offset_per_key()[end] ], offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=split_length_per_key, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) start = end start_offset = end_offset return split_list def permute( self, indices: List[int], indices_tensor: Optional[torch.Tensor] = None, include_inverse_indices: bool = False, ) -> "KeyedJaggedTensor": if indices_tensor is None: indices_tensor = torch.tensor( indices, dtype=torch.int, device=self.device() ) length_per_key = self.length_per_key() permuted_keys: List[str] = [] permuted_stride_per_key_per_rank: List[List[int]] = [] permuted_length_per_key: List[int] = [] permuted_lengths_sum = 0 for index in indices: key = self.keys()[index] permuted_keys.append(key) permuted_stride_per_key_per_rank.append( self.stride_per_key_per_rank()[index] ) permuted_length_per_key.append(length_per_key[index]) permuted_lengths_sum += length_per_key[index] if self.variable_stride_per_key(): length_per_key_tensor = _pin_and_move( torch.tensor(self.length_per_key()), self.device() ) stride_per_key_tensor = _pin_and_move( torch.tensor(self.stride_per_key()), self.device() ) permuted_lengths, _ = _permute_tensor_by_segments( self.lengths(), stride_per_key_tensor, indices_tensor, None, ) permuted_values, permuted_weights = _permute_tensor_by_segments( self.values(), length_per_key_tensor, indices_tensor, self.weights_or_none(), ) else: ( permuted_lengths, permuted_values, permuted_weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( indices_tensor, self.lengths().view(len(self._keys), -1), self.values(), self.weights_or_none(), permuted_lengths_sum, ) stride, optional_permuted_stride_per_key_per_rank = ( (None, permuted_stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) kjt = KeyedJaggedTensor( keys=permuted_keys, values=permuted_values, weights=permuted_weights, lengths=permuted_lengths.view(-1), offsets=None, stride=stride, stride_per_key_per_rank=optional_permuted_stride_per_key_per_rank, length_per_key=permuted_length_per_key if len(permuted_keys) > 0 else None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=( self.inverse_indices_or_none() if include_inverse_indices else None ), ) return kjt def flatten_lengths(self) -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()) if self.variable_stride_per_key() else (self._stride, None) ) return KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self._weights, lengths=self.lengths().view(-1), offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self.length_per_key(), offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) def __getitem__(self, key: str) -> JaggedTensor: offset_per_key = self.offset_per_key() index = self._key_indices()[key] start_offset = offset_per_key[index] end_offset = ( offset_per_key[index + 1] if index + 1 < len(offset_per_key) else start_offset ) return JaggedTensor( values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[index] : self.lengths_offset_per_key()[ index + 1 ] ], offsets=None, ) def to_dict(self) -> Dict[str, JaggedTensor]: _jt_dict = _maybe_compute_kjt_to_jt_dict( stride=self.stride(), stride_per_key=self.stride_per_key(), keys=self.keys(), length_per_key=self.length_per_key(), lengths=self.lengths(), values=self.values(), variable_stride_per_key=self.variable_stride_per_key(), weights=self.weights_or_none(), jt_dict=self._jt_dict, ) self._jt_dict = _jt_dict return _jt_dict def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def to( self, device: torch.device, non_blocking: bool = False, dtype: Optional[torch.dtype] = None, ) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) length_per_key = self._length_per_key offset_per_key = self._offset_per_key index_per_key = self._index_per_key jt_dict = self._jt_dict inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = ( inverse_indices[0], inverse_indices[1].to(device, non_blocking=non_blocking), ) if weights is not None: if dtype is not None: weights = weights.to( dtype=dtype, device=device, non_blocking=non_blocking ) else: weights = weights.to(device=device, non_blocking=non_blocking) return KeyedJaggedTensor( keys=self._keys, values=self._values.to(device, non_blocking=non_blocking), weights=weights, lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=length_per_key, offset_per_key=offset_per_key, index_per_key=index_per_key, jt_dict=jt_dict, inverse_indices=inverse_indices, ) def __str__(self) -> str: if len(self._keys) == 0 or self._offsets is None and self._lengths is None: return "KeyedJaggedTensor()\n" offsets = self.offsets() return ( "KeyedJaggedTensor({\n" + ",\n".join( [ " " + _jagged_tensor_string( self._keys[index], self._values, self._weights, offsets, sum(self.stride_per_key()[:index]), sum(self.stride_per_key()[: index + 1]), ) for index in range(len(self._keys)) ] ) + "\n})\n" ) def pin_memory(self) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = (inverse_indices[0], inverse_indices[1].pin_memory()) return KeyedJaggedTensor( keys=self._keys, values=self._values.pin_memory(), weights=weights.pin_memory() if weights is not None else None, lengths=lengths.pin_memory() if lengths is not None else None, offsets=offsets.pin_memory() if offsets is not None else None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=None, inverse_indices=inverse_indices, ) def dist_labels(self) -> List[str]: labels = ["lengths", "values"] if self.variable_stride_per_key(): labels.append("strides") if self.weights_or_none() is not None: labels.append("weights") return labels def dist_splits(self, key_splits: List[int]) -> List[List[int]]: batch_size_per_split = _sum_by_splits(self.stride_per_key(), key_splits) length_per_split = _sum_by_splits(self.length_per_key(), key_splits) splits = [batch_size_per_split, length_per_split] if self.variable_stride_per_key(): splits.append(key_splits) if self.weights_or_none() is not None: splits.append(length_per_split) return splits def dist_tensors(self) -> List[torch.Tensor]: tensors = [self.lengths(), self.values()] if self.variable_stride_per_key(): strides = _pin_and_move(torch.tensor(self.stride_per_key()), self.device()) tensors.append(strides) if self.weights_or_none() is not None: tensors.append(self.weights()) return tensors def dist_init( keys: List[str], tensors: List[torch.Tensor], variable_stride_per_key: bool, num_workers: int, recat: Optional[torch.Tensor], stride_per_rank: Optional[List[int]], stagger: int = 1, ) -> "KeyedJaggedTensor": assert len(tensors) in [2, 3, 4] lengths = tensors[0] values = tensors[1] stride_per_rank_per_key = tensors[2] if variable_stride_per_key else None weights = ( tensors[-1] if (variable_stride_per_key and len(tensors) == 4) or (not variable_stride_per_key and len(tensors) == 3) else None ) if variable_stride_per_key: assert stride_per_rank_per_key is not None stride_per_key_per_rank: List[List[int]] = stride_per_rank_per_key.view( num_workers, len(keys) ).T.tolist() strides_cumsum: List[int] = torch.ops.fbgemm.asynchronous_complete_cumsum( stride_per_rank_per_key ).tolist() cumsum_lengths = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) length_per_key = ( cumsum_lengths[strides_cumsum[1:]] - cumsum_lengths[strides_cumsum[:-1]] ) with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: lengths, _ = _permute_tensor_by_segments( lengths, stride_per_rank_per_key, recat, None, ) values, weights = _permute_tensor_by_segments( values, length_per_key, recat, weights, ) if not stride_per_key_per_rank: stride_per_key_per_rank = [[0]] * len(keys) if stagger > 1: stride_per_key_per_rank_stagger: List[List[int]] = [] local_world_size = num_workers // stagger for i in range(len(keys)): stride_per_rank_stagger: List[int] = [] for j in range(local_world_size): stride_per_rank_stagger.extend( stride_per_key_per_rank[i][j::local_world_size] ) stride_per_key_per_rank_stagger.append(stride_per_rank_stagger) stride_per_key_per_rank = stride_per_key_per_rank_stagger kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride_per_key_per_rank=stride_per_key_per_rank, ) return kjt.sync() else: assert stride_per_rank is not None with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: stride = stride_per_rank[0] if all(s == stride for s in stride_per_rank): ( lengths, values, weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( recat, lengths.view(-1, stride), values, weights, values.numel(), ) lengths = lengths.view(-1) else: # variable batch size per rank ( lengths, values, weights, ) = torch.ops.fbgemm.permute_1D_sparse_data( recat, lengths.view(-1), values, weights, values.numel(), ) kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=sum(stride_per_rank), ) return kjt.sync() def construct_jagged_tensors( embeddings: torch.Tensor, features: KeyedJaggedTensor, embedding_names: List[str], need_indices: bool = False, features_to_permute_indices: Optional[Dict[str, List[int]]] = None, original_features: Optional[KeyedJaggedTensor] = None, reverse_indices: Optional[torch.Tensor] = None, ) -> Dict[str, JaggedTensor]: with record_function("## construct_jagged_tensors ##"): if original_features is not None: features = original_features if reverse_indices is not None: embeddings = torch.index_select( embeddings, 0, reverse_indices.to(torch.int32) ) ret: Dict[str, JaggedTensor] = {} stride = features.stride() length_per_key = features.length_per_key() values = features.values() lengths = features.lengths().view(-1, stride) lengths_tuple = torch.unbind(lengths.view(-1, stride), dim=0) embeddings_list = torch.split(embeddings, length_per_key, dim=0) values_list = torch.split(values, length_per_key) if need_indices else None key_indices = defaultdict(list) for i, key in enumerate(embedding_names): key_indices[key].append(i) for key, indices in key_indices.items(): # combines outputs in correct order for CW sharding indices = ( _permute_indices(indices, features_to_permute_indices[key]) if features_to_permute_indices and key in features_to_permute_indices else indices ) ret[key] = JaggedTensor( lengths=lengths_tuple[indices[0]], values=( embeddings_list[indices[0]] if len(indices) == 1 else torch.cat([embeddings_list[i] for i in indices], dim=1) ), # pyre-ignore weights=values_list[indices[0]] if need_indices else None, ) return ret
null
8,870
import copy from collections import defaultdict from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union import torch from torch.profiler import record_function from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def _fx_to_list(tensor: torch.Tensor) -> List[int]: return tensor.long().tolist() def _permute_indices(indices: List[int], permute: List[int]) -> List[int]: permuted_indices = [0] * len(indices) for i, permuted_index in enumerate(permute): permuted_indices[i] = indices[permuted_index] return permuted_indices # pyre-ignore class JaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """ Represents an (optionally weighted) jagged tensor. A `JaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. See `KeyedJaggedTensor` for full example. Implementation is torch.jit.script-able. NOTE: We will NOT do input validation as it's expensive, you should always pass in the valid lengths, offsets, etc. Args: values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if values have weights. Tensor with same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. """ _fields = ["_values", "_weights", "_lengths", "_offsets"] def __init__( self, values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, ) -> None: self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights _assert_offsets_or_lengths_is_provided(offsets, lengths) if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "JaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return JaggedTensor( values=torch.empty(0, dtype=values_dtype, device=device), offsets=torch.empty(0, dtype=lengths_dtype, device=device), lengths=torch.empty(0, dtype=lengths_dtype, device=device), weights=weights, ) def from_dense_lengths( values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` is still of shape (B,). """ mask2d = ( _arange(end=values.size(1), device=values.device).expand(values.size(0), -1) ) < lengths.unsqueeze(-1) return JaggedTensor( values=values[mask2d], weights=_optional_mask(weights, mask2d), lengths=lengths, ) def from_dense( values: List[torch.Tensor], weights: Optional[List[torch.Tensor]] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` and `offsets` are still of shape (B,). Args: values (List[torch.Tensor]): a list of tensors for dense representation weights (Optional[List[torch.Tensor]]): if values have weights, tensor with the same shape as values. Returns: JaggedTensor: JaggedTensor created from 2D dense tensor. Example:: values = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] weights = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] j1 = JaggedTensor.from_dense( values=values, weights=weights, ) # j1 = [[1.0], [], [7.0], [8.0], [10.0, 11.0, 12.0]] """ values_tensor = torch.cat(values, dim=0) lengths = torch.tensor( [value.size(0) for value in values], dtype=torch.int32, device=values_tensor.device, ) weights_tensor = torch.cat(weights, dim=0) if weights is not None else None return JaggedTensor( values=values_tensor, weights=weights_tensor, lengths=lengths, ) def to_dense(self) -> List[torch.Tensor]: """ Constructs a dense-representation of the JT's values. Returns: List[torch.Tensor]: list of tensors. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) values_list = jt.to_dense() # values_list = [ # torch.tensor([1.0, 2.0]), # torch.tensor([]), # torch.tensor([3.0]), # torch.tensor([4.0]), # torch.tensor([5.0]), # torch.tensor([6.0, 7.0, 8.0]), # ] """ tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.values()[offset:next_offset]) return tensor_list def to_dense_weights(self) -> Optional[List[torch.Tensor]]: """ Constructs a dense-representation of the JT's weights. Returns: Optional[List[torch.Tensor]]: list of tensors, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) weights_list = jt.to_dense_weights() # weights_list = [ # torch.tensor([0.1, 0.2]), # torch.tensor([]), # torch.tensor([0.3]), # torch.tensor([0.4]), # torch.tensor([0.5]), # torch.tensor([0.6, 0.7, 0.8]), # ] """ if self.weights_or_none() is None: return None tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.weights()[offset:next_offset]) return tensor_list def to_padded_dense( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> torch.Tensor: """ Constructs a 2D dense tensor from the JT's values of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: torch.Tensor: 2d dense tensor. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) dt = jt.to_padded_dense( desired_length=2, padding_value=10.0, ) # dt = [ # [1.0, 2.0], # [10.0, 10.0], # [3.0, 10.0], # [4.0, 10.0], # [5.0, 10.0], # [6.0, 7.0], # ] """ if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.values(), [self.offsets()], [N], padding_value ) def to_padded_dense_weights( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> Optional[torch.Tensor]: """ Constructs a 2D dense tensor from the JT's weights of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: Optional[torch.Tensor]: 2d dense tensor, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) d_wt = jt.to_padded_dense_weights( desired_length=2, padding_value=1.0, ) # d_wt = [ # [0.1, 0.2], # [1.0, 1.0], # [0.3, 1.0], # [0.4, 1.0], # [0.5, 1.0], # [0.6, 0.7], # ] """ if self.weights_or_none() is None: return None if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.weights(), [self.offsets()], [N], padding_value ) def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def to(self, device: torch.device, non_blocking: bool = False) -> "JaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets return JaggedTensor( values=self._values.to(device, non_blocking=non_blocking), weights=( weights.to(device, non_blocking=non_blocking) if weights is not None else None ), lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def __str__(self) -> str: offsets = self.offsets() if self._weights is None: return ( "JaggedTensor({\n " + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + "\n})\n" ) return ( "JaggedTensor({\n" + ' "values": ' + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + ',\n "weights": ' + _jagged_values_string( _get_weights_or_throw(self._weights), offsets, 0, len(offsets) - 1 ) + "\n})\n" ) def construct_jagged_tensors_inference( embeddings: torch.Tensor, lengths: torch.Tensor, values: torch.Tensor, embedding_names: List[str], need_indices: bool = False, features_to_permute_indices: Optional[Dict[str, List[int]]] = None, reverse_indices: Optional[torch.Tensor] = None, ) -> Dict[str, JaggedTensor]: with record_function("## construct_jagged_tensors_inference ##"): if reverse_indices is not None: embeddings = torch.index_select( embeddings, 0, reverse_indices.to(torch.int32) ) ret: Dict[str, JaggedTensor] = {} length_per_key: List[int] = _fx_to_list( torch.sum(lengths.view(len(embedding_names), -1), dim=1) ) lengths = lengths.view(len(embedding_names), -1) lengths_tuple = torch.unbind(lengths, dim=0) embeddings_list = torch.split(embeddings, length_per_key, dim=0) values_list = torch.split(values, length_per_key) if need_indices else None key_indices = defaultdict(list) for i, key in enumerate(embedding_names): key_indices[key].append(i) for key, indices in key_indices.items(): # combines outputs in correct order for CW sharding indices = ( _permute_indices(indices, features_to_permute_indices[key]) if features_to_permute_indices and key in features_to_permute_indices else indices ) ret[key] = JaggedTensor( lengths=lengths_tuple[indices[0]], values=( embeddings_list[indices[0]] if len(indices) == 1 else torch.cat([embeddings_list[i] for i in indices], dim=1) ), # pyre-ignore weights=values_list[indices[0]] if need_indices else None, ) return ret
null
8,871
from typing import List, Optional, Tuple import torch import torch.nn as nn from torchrec.modules.embedding_modules import ( EmbeddingBagCollection, EmbeddingCollection, ) from torchrec.sparse.jagged_tensor import KeyedJaggedTensor class EmbeddingBagCollection(EmbeddingBagCollectionInterface): """ EmbeddingBagCollection represents a collection of pooled embeddings (`EmbeddingBags`). It processes sparse data in the form of `KeyedJaggedTensor` with values of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (jagged) and outputs a `KeyedTensor` with values of the form [B * (F * D)] where: * F: features (keys) * D: each feature's (key's) embedding dimension * B: batch size Args: tables (List[EmbeddingBagConfig]): list of embedding tables. is_weighted (bool): whether input `KeyedJaggedTensor` is weighted. device (Optional[torch.device]): default compute device. Example:: table_0 = EmbeddingBagConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) table_1 = EmbeddingBagConfig( name="t2", embedding_dim=4, num_embeddings=10, feature_names=["f2"] ) ebc = EmbeddingBagCollection(tables=[table_0, table_1]) # 0 1 2 <-- batch # "f1" [0,1] None [2] # "f2" [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) pooled_embeddings = ebc(features) print(pooled_embeddings.values()) tensor([[-0.8899, -0.1342, -1.9060, -0.0905, -0.2814, -0.9369, -0.7783], [ 0.0000, 0.0000, 0.0000, 0.1598, 0.0695, 1.3265, -0.1011], [-0.4256, -1.1846, -2.1648, -1.0893, 0.3590, -1.9784, -0.7681]], grad_fn=<CatBackward0>) print(pooled_embeddings.keys()) ['f1', 'f2'] print(pooled_embeddings.offset_per_key()) tensor([0, 3, 7]) """ def __init__( self, tables: List[EmbeddingBagConfig], is_weighted: bool = False, device: Optional[torch.device] = None, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self._is_weighted = is_weighted self.embedding_bags: nn.ModuleDict = nn.ModuleDict() self._embedding_bag_configs = tables self._lengths_per_embedding: List[int] = [] self._dtypes: List[int] = [] table_names = set() for embedding_config in tables: if embedding_config.name in table_names: raise ValueError(f"Duplicate table name {embedding_config.name}") table_names.add(embedding_config.name) dtype = ( torch.float32 if embedding_config.data_type == DataType.FP32 else torch.float16 ) self.embedding_bags[embedding_config.name] = nn.EmbeddingBag( num_embeddings=embedding_config.num_embeddings, embedding_dim=embedding_config.embedding_dim, mode=pooling_type_to_str(embedding_config.pooling), device=device, include_last_offset=True, dtype=dtype, ) if device is None: device = self.embedding_bags[embedding_config.name].weight.device self._dtypes.append(embedding_config.data_type.value) if not embedding_config.feature_names: embedding_config.feature_names = [embedding_config.name] self._lengths_per_embedding.extend( len(embedding_config.feature_names) * [embedding_config.embedding_dim] ) self._device: torch.device = device or torch.device("cpu") self._embedding_names: List[str] = [ embedding for embeddings in get_embedding_names_by_table(tables) for embedding in embeddings ] self._feature_names: List[List[str]] = [table.feature_names for table in tables] self.reset_parameters() def forward(self, features: KeyedJaggedTensor) -> KeyedTensor: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: KeyedTensor """ flat_feature_names: List[str] = [] for names in self._feature_names: flat_feature_names.extend(names) inverse_indices = reorder_inverse_indices( inverse_indices=features.inverse_indices_or_none(), feature_names=flat_feature_names, ) pooled_embeddings: List[torch.Tensor] = [] feature_dict = features.to_dict() for i, embedding_bag in enumerate(self.embedding_bags.values()): for feature_name in self._feature_names[i]: f = feature_dict[feature_name] per_sample_weights: Optional[torch.Tensor] = None if self._is_weighted: per_sample_weights = ( f.weights().half() if self._dtypes[i] == DataType.FP16.value else f.weights() ) res = embedding_bag( input=f.values(), offsets=f.offsets(), per_sample_weights=( per_sample_weights if self._is_weighted else None ), ).float() pooled_embeddings.append(res) return KeyedTensor( keys=self._embedding_names, values=process_pooled_embeddings( pooled_embeddings=pooled_embeddings, inverse_indices=inverse_indices, ), length_per_key=self._lengths_per_embedding, ) def is_weighted(self) -> bool: return self._is_weighted def embedding_bag_configs(self) -> List[EmbeddingBagConfig]: return self._embedding_bag_configs def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_bag_configs: assert table_config.init_fn is not None param = self.embedding_bags[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) class EmbeddingCollection(EmbeddingCollectionInterface): """ EmbeddingCollection represents a collection of non-pooled embeddings. It processes sparse data in the form of `KeyedJaggedTensor` of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (variable) and outputs `Dict[feature (key), JaggedTensor]`. Each `JaggedTensor` contains values of the form (B * L) X D where: * B: batch size * L: length of sparse features (jagged) * D: each feature's (key's) embedding dimension and lengths are of the form L Args: tables (List[EmbeddingConfig]): list of embedding tables. device (Optional[torch.device]): default compute device. need_indices (bool): if we need to pass indices to the final lookup dict. Example:: e1_config = EmbeddingConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) e2_config = EmbeddingConfig( name="t2", embedding_dim=3, num_embeddings=10, feature_names=["f2"] ) ec = EmbeddingCollection(tables=[e1_config, e2_config]) # 0 1 2 <-- batch # 0 [0,1] None [2] # 1 [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor.from_offsets_sync( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) feature_embeddings = ec(features) print(feature_embeddings['f2'].values()) tensor([[-0.2050, 0.5478, 0.6054], [ 0.7352, 0.3210, -3.0399], [ 0.1279, -0.1756, -0.4130], [ 0.7519, -0.4341, -0.0499], [ 0.9329, -1.0697, -0.8095]], grad_fn=<EmbeddingBackward>) """ def __init__( # noqa C901 self, tables: List[EmbeddingConfig], device: Optional[torch.device] = None, need_indices: bool = False, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self.embeddings: nn.ModuleDict = nn.ModuleDict() self._embedding_configs = tables self._embedding_dim: int = -1 self._need_indices: bool = need_indices self._device: torch.device = ( device if device is not None else torch.device("cpu") ) table_names = set() for config in tables: if config.name in table_names: raise ValueError(f"Duplicate table name {config.name}") table_names.add(config.name) self._embedding_dim = ( config.embedding_dim if self._embedding_dim < 0 else self._embedding_dim ) if self._embedding_dim != config.embedding_dim: raise ValueError( "All tables in a EmbeddingCollection are required to have same embedding dimension." + f" Violating case: {config.name}'s embedding_dim {config.embedding_dim} !=" + f" {self._embedding_dim}" ) dtype = ( torch.float32 if config.data_type == DataType.FP32 else torch.float16 ) self.embeddings[config.name] = nn.Embedding( num_embeddings=config.num_embeddings, embedding_dim=config.embedding_dim, device=device, dtype=dtype, ) if config.init_fn is not None: config.init_fn(self.embeddings[config.name].weight) if not config.feature_names: config.feature_names = [config.name] self._embedding_names_by_table: List[List[str]] = get_embedding_names_by_table( tables ) self._feature_names: List[List[str]] = [table.feature_names for table in tables] def forward( self, features: KeyedJaggedTensor, ) -> Dict[str, JaggedTensor]: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: Dict[str, JaggedTensor] """ feature_embeddings: Dict[str, JaggedTensor] = {} jt_dict: Dict[str, JaggedTensor] = features.to_dict() for i, emb_module in enumerate(self.embeddings.values()): feature_names = self._feature_names[i] embedding_names = self._embedding_names_by_table[i] for j, embedding_name in enumerate(embedding_names): feature_name = feature_names[j] f = jt_dict[feature_name] lookup = emb_module( input=f.values(), ).float() feature_embeddings[embedding_name] = JaggedTensor( values=lookup, lengths=f.lengths(), weights=f.values() if self._need_indices else None, ) return feature_embeddings def need_indices(self) -> bool: return self._need_indices def embedding_dim(self) -> int: return self._embedding_dim def embedding_configs(self) -> List[EmbeddingConfig]: return self._embedding_configs def embedding_names_by_table(self) -> List[List[str]]: return self._embedding_names_by_table def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_configs: assert table_config.init_fn is not None param = self.embeddings[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) The provided code snippet includes necessary dependencies for implementing the `tower_input_params` function. Write a Python function `def tower_input_params(module: nn.Module) -> Tuple[bool, bool]` to solve the following problem: Utilty to compute the mapping of tower KJT args to pass to the embedding modules. Args: module (nn.Module): Returns: Tuple[bool, bool]: tuple of 2 booleans representing if KJT and weighted KJT are required, respectively. Here is the function: def tower_input_params(module: nn.Module) -> Tuple[bool, bool]: """ Utilty to compute the mapping of tower KJT args to pass to the embedding modules. Args: module (nn.Module): Returns: Tuple[bool, bool]: tuple of 2 booleans representing if KJT and weighted KJT are required, respectively. """ if isinstance(module, EmbeddingCollection): return True, False elif isinstance(module, EmbeddingBagCollection): return not module.is_weighted(), module.is_weighted() # default to assuming both kjt and weight_kjt required return True, True
Utilty to compute the mapping of tower KJT args to pass to the embedding modules. Args: module (nn.Module): Returns: Tuple[bool, bool]: tuple of 2 booleans representing if KJT and weighted KJT are required, respectively.
8,872
import abc from collections import OrderedDict from typing import Any, Dict, Iterator, List, Optional, Tuple import torch import torch.nn as nn from torchrec.fx.tracer import is_fx_tracing from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor # pyre-ignore class JaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """ Represents an (optionally weighted) jagged tensor. A `JaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. See `KeyedJaggedTensor` for full example. Implementation is torch.jit.script-able. NOTE: We will NOT do input validation as it's expensive, you should always pass in the valid lengths, offsets, etc. Args: values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if values have weights. Tensor with same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. """ _fields = ["_values", "_weights", "_lengths", "_offsets"] def __init__( self, values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, ) -> None: self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights _assert_offsets_or_lengths_is_provided(offsets, lengths) if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "JaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return JaggedTensor( values=torch.empty(0, dtype=values_dtype, device=device), offsets=torch.empty(0, dtype=lengths_dtype, device=device), lengths=torch.empty(0, dtype=lengths_dtype, device=device), weights=weights, ) def from_dense_lengths( values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` is still of shape (B,). """ mask2d = ( _arange(end=values.size(1), device=values.device).expand(values.size(0), -1) ) < lengths.unsqueeze(-1) return JaggedTensor( values=values[mask2d], weights=_optional_mask(weights, mask2d), lengths=lengths, ) def from_dense( values: List[torch.Tensor], weights: Optional[List[torch.Tensor]] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` and `offsets` are still of shape (B,). Args: values (List[torch.Tensor]): a list of tensors for dense representation weights (Optional[List[torch.Tensor]]): if values have weights, tensor with the same shape as values. Returns: JaggedTensor: JaggedTensor created from 2D dense tensor. Example:: values = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] weights = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] j1 = JaggedTensor.from_dense( values=values, weights=weights, ) # j1 = [[1.0], [], [7.0], [8.0], [10.0, 11.0, 12.0]] """ values_tensor = torch.cat(values, dim=0) lengths = torch.tensor( [value.size(0) for value in values], dtype=torch.int32, device=values_tensor.device, ) weights_tensor = torch.cat(weights, dim=0) if weights is not None else None return JaggedTensor( values=values_tensor, weights=weights_tensor, lengths=lengths, ) def to_dense(self) -> List[torch.Tensor]: """ Constructs a dense-representation of the JT's values. Returns: List[torch.Tensor]: list of tensors. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) values_list = jt.to_dense() # values_list = [ # torch.tensor([1.0, 2.0]), # torch.tensor([]), # torch.tensor([3.0]), # torch.tensor([4.0]), # torch.tensor([5.0]), # torch.tensor([6.0, 7.0, 8.0]), # ] """ tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.values()[offset:next_offset]) return tensor_list def to_dense_weights(self) -> Optional[List[torch.Tensor]]: """ Constructs a dense-representation of the JT's weights. Returns: Optional[List[torch.Tensor]]: list of tensors, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) weights_list = jt.to_dense_weights() # weights_list = [ # torch.tensor([0.1, 0.2]), # torch.tensor([]), # torch.tensor([0.3]), # torch.tensor([0.4]), # torch.tensor([0.5]), # torch.tensor([0.6, 0.7, 0.8]), # ] """ if self.weights_or_none() is None: return None tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.weights()[offset:next_offset]) return tensor_list def to_padded_dense( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> torch.Tensor: """ Constructs a 2D dense tensor from the JT's values of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: torch.Tensor: 2d dense tensor. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) dt = jt.to_padded_dense( desired_length=2, padding_value=10.0, ) # dt = [ # [1.0, 2.0], # [10.0, 10.0], # [3.0, 10.0], # [4.0, 10.0], # [5.0, 10.0], # [6.0, 7.0], # ] """ if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.values(), [self.offsets()], [N], padding_value ) def to_padded_dense_weights( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> Optional[torch.Tensor]: """ Constructs a 2D dense tensor from the JT's weights of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: Optional[torch.Tensor]: 2d dense tensor, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) d_wt = jt.to_padded_dense_weights( desired_length=2, padding_value=1.0, ) # d_wt = [ # [0.1, 0.2], # [1.0, 1.0], # [0.3, 1.0], # [0.4, 1.0], # [0.5, 1.0], # [0.6, 0.7], # ] """ if self.weights_or_none() is None: return None if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.weights(), [self.offsets()], [N], padding_value ) def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def to(self, device: torch.device, non_blocking: bool = False) -> "JaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets return JaggedTensor( values=self._values.to(device, non_blocking=non_blocking), weights=( weights.to(device, non_blocking=non_blocking) if weights is not None else None ), lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def __str__(self) -> str: offsets = self.offsets() if self._weights is None: return ( "JaggedTensor({\n " + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + "\n})\n" ) return ( "JaggedTensor({\n" + ' "values": ' + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + ',\n "weights": ' + _jagged_values_string( _get_weights_or_throw(self._weights), offsets, 0, len(offsets) - 1 ) + "\n})\n" ) def position_weighted_module_update_features( features: Dict[str, JaggedTensor], weighted_features: Dict[str, JaggedTensor], ) -> Dict[str, JaggedTensor]: features.update(weighted_features) return features
null
8,873
import abc from collections import OrderedDict from typing import Any, Dict, Iterator, List, Optional, Tuple import torch import torch.nn as nn from torchrec.fx.tracer import is_fx_tracing from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor def offsets_to_range_traceble( offsets: torch.Tensor, values: torch.Tensor ) -> torch.Tensor: return torch.ops.fbgemm.offsets_range(offsets.long(), torch.numel(values))
null
8,874
import abc from collections import defaultdict from typing import Callable, Dict, List, NamedTuple, Optional, Tuple, Union import torch from torch import nn from torchrec.modules.embedding_configs import BaseEmbeddingConfig from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor # pyre-ignore class JaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """ Represents an (optionally weighted) jagged tensor. A `JaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. See `KeyedJaggedTensor` for full example. Implementation is torch.jit.script-able. NOTE: We will NOT do input validation as it's expensive, you should always pass in the valid lengths, offsets, etc. Args: values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if values have weights. Tensor with same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. """ _fields = ["_values", "_weights", "_lengths", "_offsets"] def __init__( self, values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, ) -> None: self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights _assert_offsets_or_lengths_is_provided(offsets, lengths) if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "JaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return JaggedTensor( values=torch.empty(0, dtype=values_dtype, device=device), offsets=torch.empty(0, dtype=lengths_dtype, device=device), lengths=torch.empty(0, dtype=lengths_dtype, device=device), weights=weights, ) def from_dense_lengths( values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` is still of shape (B,). """ mask2d = ( _arange(end=values.size(1), device=values.device).expand(values.size(0), -1) ) < lengths.unsqueeze(-1) return JaggedTensor( values=values[mask2d], weights=_optional_mask(weights, mask2d), lengths=lengths, ) def from_dense( values: List[torch.Tensor], weights: Optional[List[torch.Tensor]] = None, ) -> "JaggedTensor": """ Constructs `JaggedTensor` from dense values/weights of shape (B, N,). Note that `lengths` and `offsets` are still of shape (B,). Args: values (List[torch.Tensor]): a list of tensors for dense representation weights (Optional[List[torch.Tensor]]): if values have weights, tensor with the same shape as values. Returns: JaggedTensor: JaggedTensor created from 2D dense tensor. Example:: values = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] weights = [ torch.Tensor([1.0]), torch.Tensor(), torch.Tensor([7.0, 8.0]), torch.Tensor([10.0, 11.0, 12.0]), ] j1 = JaggedTensor.from_dense( values=values, weights=weights, ) # j1 = [[1.0], [], [7.0], [8.0], [10.0, 11.0, 12.0]] """ values_tensor = torch.cat(values, dim=0) lengths = torch.tensor( [value.size(0) for value in values], dtype=torch.int32, device=values_tensor.device, ) weights_tensor = torch.cat(weights, dim=0) if weights is not None else None return JaggedTensor( values=values_tensor, weights=weights_tensor, lengths=lengths, ) def to_dense(self) -> List[torch.Tensor]: """ Constructs a dense-representation of the JT's values. Returns: List[torch.Tensor]: list of tensors. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) values_list = jt.to_dense() # values_list = [ # torch.tensor([1.0, 2.0]), # torch.tensor([]), # torch.tensor([3.0]), # torch.tensor([4.0]), # torch.tensor([5.0]), # torch.tensor([6.0, 7.0, 8.0]), # ] """ tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.values()[offset:next_offset]) return tensor_list def to_dense_weights(self) -> Optional[List[torch.Tensor]]: """ Constructs a dense-representation of the JT's weights. Returns: Optional[List[torch.Tensor]]: list of tensors, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) weights_list = jt.to_dense_weights() # weights_list = [ # torch.tensor([0.1, 0.2]), # torch.tensor([]), # torch.tensor([0.3]), # torch.tensor([0.4]), # torch.tensor([0.5]), # torch.tensor([0.6, 0.7, 0.8]), # ] """ if self.weights_or_none() is None: return None tensor_list = [] for index in range(self.offsets().size(0) - 1): offset = self.offsets()[index].item() next_offset = self.offsets()[index + 1].item() tensor_list.append(self.weights()[offset:next_offset]) return tensor_list def to_padded_dense( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> torch.Tensor: """ Constructs a 2D dense tensor from the JT's values of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: torch.Tensor: 2d dense tensor. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, offsets=offsets) dt = jt.to_padded_dense( desired_length=2, padding_value=10.0, ) # dt = [ # [1.0, 2.0], # [10.0, 10.0], # [3.0, 10.0], # [4.0, 10.0], # [5.0, 10.0], # [6.0, 7.0], # ] """ if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.values(), [self.offsets()], [N], padding_value ) def to_padded_dense_weights( self, desired_length: Optional[int] = None, padding_value: float = 0.0, ) -> Optional[torch.Tensor]: """ Constructs a 2D dense tensor from the JT's weights of shape (B, N,). Note that `B` is the length of self.lengths() and `N` is the longest feature length or `desired_length`. If `desired_length` > `length` we will pad with `padding_value`, otherwise we will select the last value at `desired_length`. Args: desired_length (int): the length of the tensor. padding_value (float): padding value if we need to pad. Returns: Optional[torch.Tensor]: 2d dense tensor, `None` if no weights. Example:: values = torch.Tensor([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]) weights = torch.Tensor([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]) offsets = torch.IntTensor([0, 2, 2, 3, 4, 5, 8]) jt = JaggedTensor(values=values, weights=weights, offsets=offsets) d_wt = jt.to_padded_dense_weights( desired_length=2, padding_value=1.0, ) # d_wt = [ # [0.1, 0.2], # [1.0, 1.0], # [0.3, 1.0], # [0.4, 1.0], # [0.5, 1.0], # [0.6, 0.7], # ] """ if self.weights_or_none() is None: return None if desired_length is None: N = int(torch.max(self.lengths()).item()) else: N = desired_length return torch.ops.fbgemm.jagged_to_padded_dense( self.weights(), [self.offsets()], [N], padding_value ) def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def to(self, device: torch.device, non_blocking: bool = False) -> "JaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets return JaggedTensor( values=self._values.to(device, non_blocking=non_blocking), weights=( weights.to(device, non_blocking=non_blocking) if weights is not None else None ), lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def __str__(self) -> str: offsets = self.offsets() if self._weights is None: return ( "JaggedTensor({\n " + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + "\n})\n" ) return ( "JaggedTensor({\n" + ' "values": ' + _jagged_values_string(self._values, offsets, 0, len(offsets) - 1) + ',\n "weights": ' + _jagged_values_string( _get_weights_or_throw(self._weights), offsets, 0, len(offsets) - 1 ) + "\n})\n" ) The provided code snippet includes necessary dependencies for implementing the `apply_mc_method_to_jt_dict` function. Write a Python function `def apply_mc_method_to_jt_dict( method: str, features_dict: Dict[str, JaggedTensor], table_to_features: Dict[str, List[str]], managed_collisions: nn.ModuleDict, ) -> Dict[str, JaggedTensor]` to solve the following problem: Applies an MC method to a dictionary of JaggedTensors, returning the updated dictionary with same ordering Here is the function: def apply_mc_method_to_jt_dict( method: str, features_dict: Dict[str, JaggedTensor], table_to_features: Dict[str, List[str]], managed_collisions: nn.ModuleDict, ) -> Dict[str, JaggedTensor]: """ Applies an MC method to a dictionary of JaggedTensors, returning the updated dictionary with same ordering """ mc_output: Dict[str, JaggedTensor] = features_dict.copy() for table, features in table_to_features.items(): mc_input: Dict[str, JaggedTensor] = {} for feature in features: mc_input[feature] = features_dict[feature] mc_module = managed_collisions[table] attr = getattr(mc_module, method) mc_output.update(attr(mc_input)) return mc_output
Applies an MC method to a dictionary of JaggedTensors, returning the updated dictionary with same ordering
8,875
import abc from collections import defaultdict from typing import Callable, Dict, List, NamedTuple, Optional, Tuple, Union import torch from torch import nn from torchrec.modules.embedding_configs import BaseEmbeddingConfig from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor The provided code snippet includes necessary dependencies for implementing the `dynamic_threshold_filter` function. Write a Python function `def dynamic_threshold_filter( id_counts: torch.Tensor, threshold_skew_multiplier: float = 10.0, ) -> Tuple[torch.Tensor, torch.Tensor]` to solve the following problem: Threshold is total_count / num_ids * threshold_skew_multiplier. An id is added if its count is strictly greater than the threshold. Here is the function: def dynamic_threshold_filter( id_counts: torch.Tensor, threshold_skew_multiplier: float = 10.0, ) -> Tuple[torch.Tensor, torch.Tensor]: """ Threshold is total_count / num_ids * threshold_skew_multiplier. An id is added if its count is strictly greater than the threshold. """ num_ids = id_counts.numel() total_count = id_counts.sum() BASE_THRESHOLD = 1 / num_ids threshold_mass = BASE_THRESHOLD * threshold_skew_multiplier threshold = threshold_mass * total_count threshold_mask = id_counts > threshold return threshold_mask, threshold
Threshold is total_count / num_ids * threshold_skew_multiplier. An id is added if its count is strictly greater than the threshold.
8,876
import abc from collections import defaultdict from typing import Callable, Dict, List, NamedTuple, Optional, Tuple, Union import torch from torch import nn from torchrec.modules.embedding_configs import BaseEmbeddingConfig from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor The provided code snippet includes necessary dependencies for implementing the `average_threshold_filter` function. Write a Python function `def average_threshold_filter( id_counts: torch.Tensor, ) -> Tuple[torch.Tensor, torch.Tensor]` to solve the following problem: Threshold is average of id_counts. An id is added if its count is strictly greater than the mean. Here is the function: def average_threshold_filter( id_counts: torch.Tensor, ) -> Tuple[torch.Tensor, torch.Tensor]: """ Threshold is average of id_counts. An id is added if its count is strictly greater than the mean. """ if id_counts.dtype != torch.float: id_counts = id_counts.float() threshold = id_counts.mean() threshold_mask = id_counts > threshold return threshold_mask, threshold
Threshold is average of id_counts. An id is added if its count is strictly greater than the mean.
8,877
import abc from collections import defaultdict from typing import Callable, Dict, List, NamedTuple, Optional, Tuple, Union import torch from torch import nn from torchrec.modules.embedding_configs import BaseEmbeddingConfig from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor The provided code snippet includes necessary dependencies for implementing the `probabilistic_threshold_filter` function. Write a Python function `def probabilistic_threshold_filter( id_counts: torch.Tensor, per_id_probability: float = 0.01, ) -> Tuple[torch.Tensor, torch.Tensor]` to solve the following problem: Each id has probability per_id_probability of being added. For example, if per_id_probability is 0.01 and an id appears 100 times, then it has a 60% of being added. More precisely, the id score is 1 - (1 - per_id_probability) ^ id_count, and for a randomly generated threshold, the id score is the chance of it being added. Here is the function: def probabilistic_threshold_filter( id_counts: torch.Tensor, per_id_probability: float = 0.01, ) -> Tuple[torch.Tensor, torch.Tensor]: """ Each id has probability per_id_probability of being added. For example, if per_id_probability is 0.01 and an id appears 100 times, then it has a 60% of being added. More precisely, the id score is 1 - (1 - per_id_probability) ^ id_count, and for a randomly generated threshold, the id score is the chance of it being added. """ probability = torch.full_like(id_counts, 1 - per_id_probability, dtype=torch.float) id_scores = 1 - torch.pow(probability, id_counts) threshold: torch.Tensor = torch.rand(id_counts.size(), device=id_counts.device) threshold_mask = id_scores > threshold return threshold_mask, threshold
Each id has probability per_id_probability of being added. For example, if per_id_probability is 0.01 and an id appears 100 times, then it has a 60% of being added. More precisely, the id score is 1 - (1 - per_id_probability) ^ id_count, and for a randomly generated threshold, the id score is the chance of it being added.
8,878
from typing import List import torch from torch import nn from torch.fx import wrap def _get_flatten_input(inputs: List[torch.Tensor]) -> torch.Tensor: return torch.cat( [input.flatten(1) for input in inputs], dim=1, )
null
8,879
import copy import itertools from collections import defaultdict from typing import Any, cast, Dict, Iterator, List, Optional, Set, Tuple, Type import torch import torch.nn as nn import torchrec.optim as trec_optim from fbgemm_gpu.split_embedding_configs import EmbOptimType from fbgemm_gpu.split_table_batched_embeddings_ops_training import ( ComputeDevice, EmbeddingLocation, SplitTableBatchedEmbeddingBagsCodegen, ) from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DataType, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.optim.fused import FusedOptimizer, FusedOptimizerModule from torchrec.optim.keyed import CombinedOptimizer, KeyedOptimizer from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor def convert_optimizer_type_and_kwargs( optimizer_type: Type[torch.optim.Optimizer], optimizer_kwargs: Dict[str, Any], ) -> Optional[Tuple[EmbOptimType, Dict[str, Any]]]: optimizer_kwargs = copy.deepcopy(optimizer_kwargs) if "lr" in optimizer_kwargs: optimizer_kwargs["learning_rate"] = optimizer_kwargs["lr"] optimizer_kwargs.pop("lr") if optimizer_type == torch.optim.SGD: return ( EmbOptimType.EXACT_SGD, optimizer_kwargs, ) elif optimizer_type == torch.optim.Adagrad: return (EmbOptimType.EXACT_ADAGRAD, optimizer_kwargs) elif optimizer_type == trec_optim.RowWiseAdagrad: return (EmbOptimType.EXACT_ROWWISE_ADAGRAD, optimizer_kwargs) elif optimizer_type == torch.optim.Adam: return (EmbOptimType.ADAM, optimizer_kwargs) return None
null
8,880
import copy import itertools from collections import defaultdict from typing import Any, cast, Dict, Iterator, List, Optional, Set, Tuple, Type import torch import torch.nn as nn import torchrec.optim as trec_optim from fbgemm_gpu.split_embedding_configs import EmbOptimType from fbgemm_gpu.split_table_batched_embeddings_ops_training import ( ComputeDevice, EmbeddingLocation, SplitTableBatchedEmbeddingBagsCodegen, ) from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DataType, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.optim.fused import FusedOptimizer, FusedOptimizerModule from torchrec.optim.keyed import CombinedOptimizer, KeyedOptimizer from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor class FusedEmbeddingBagCollection( EmbeddingBagCollectionInterface, FusedOptimizerModule ): """ FusedEmbeddingBagCollection represents a collection of pooled embeddings (`EmbeddingBags`). It utilizes a technique called Optimizer fusion (register the optimizer with model). The semantics of this is that during the backwards pass, the registered optimizer will be called. It processes sparse data in the form of `KeyedJaggedTensor` with values of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (jagged) and outputs a `KeyedTensor` with values of the form [B x F x D] where: * F: features (keys) * D: each feature's (key's) embedding dimension * B: batch size Args: tables (List[EmbeddingBagConfig]): list of embedding tables. is_weighted (bool): whether input `KeyedJaggedTensor` is weighted. optimizer (Type[torch.optim.Optimizer]): fusion optimizer type optimizer_kwargs: Dict[str, Any]: fusion optimizer kwargs device (Optional[torch.device]): compute device. Example:: table_0 = EmbeddingBagConfig( name="t1", embedding_dim=4, num_embeddings=10, feature_names=["f1"] ) table_1 = EmbeddingBagConfig( name="t2", embedding_dim=8, num_embeddings=10, feature_names=["f2"] ) ebc = FusedEmbeddingBagCollection(tables=[table_0, table_1], optimizer_type=torch.optim.SGD, optimizer_kwargs={"lr": .01}) # 0 1 2 <-- batch # "f1" [0,1] None [2] # "f2" [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) pooled_embeddings = ebc(features) print(pooled_embeddings.values()) tensor([[ 0.2093, 0.1395, 0.1571, 0.3583, 0.0421, 0.0037, -0.0692, 0.0663, 0.2166, -0.3150, -0.2771, -0.0301], [ 0.0000, 0.0000, 0.0000, 0.0000, 0.0165, -0.1225, 0.2483, 0.0624, -0.1168, -0.0509, -0.1309, 0.3059], [ 0.0811, -0.1779, -0.1443, 0.1097, -0.4410, -0.4036, 0.4458, -0.2735, -0.3080, -0.2102, -0.0564, 0.5583]], grad_fn=<CatBackward0>) print(pooled_embeddings.keys()) ['f1', 'f2'] print(pooled_embeddings.offset_per_key()) [0, 4, 12] """ def __init__( self, tables: List[EmbeddingBagConfig], optimizer_type: Type[torch.optim.Optimizer], optimizer_kwargs: Dict[str, Any], is_weighted: bool = False, device: Optional[torch.device] = None, location: Optional[EmbeddingLocation] = None, ) -> None: super().__init__() self._optimizer_type = optimizer_type self._optimizer_kwargs = optimizer_kwargs self._device: torch.device = ( device if device is not None else torch.device("cpu") ) emb_optim_and_kwargs = convert_optimizer_type_and_kwargs( optimizer_type, optimizer_kwargs ) if emb_optim_and_kwargs is None: raise ValueError( f"Cannot fuse optimizer_type={optimizer_type} with kwargs {optimizer_kwargs}" ) (emb_optim_type, emb_opt_kwargs) = emb_optim_and_kwargs if location in [ EmbeddingLocation.DEVICE, EmbeddingLocation.MANAGED, EmbeddingLocation.MANAGED_CACHING, ]: assert device is not None and device.type in [ "cuda", "meta", ], f"Using location={location} requires device=cuda or meta" if device is None: device = torch.device("cpu") if location is None: if device.type in ["cpu", "meta"]: location = EmbeddingLocation.HOST elif device.type == "cuda": location = EmbeddingLocation.DEVICE else: raise ValueError("EmbeddingLocation could not be set") self._is_weighted = is_weighted self._embedding_bag_configs = tables # Registering in a List instead of ModuleList because we want don't want them to be auto-registered. # Their states will be modified via self.embedding_bags self._emb_modules: List[nn.Module] = [] self._key_to_tables: Dict[ Tuple[PoolingType, DataType], List[EmbeddingBagConfig] ] = defaultdict(list) self._length_per_key: List[int] = [] for table in tables: self._length_per_key.extend( [table.embedding_dim] * len(table.feature_names) ) key = (table.pooling, table.data_type) self._key_to_tables[key].append(table) optims = [] for key, tables in self._key_to_tables.items(): (pooling, data_type) = key emb_module = _BatchedFusedEmbeddingLookups( cast(List[BaseEmbeddingConfig], tables), data_type=data_type, pooling=pooling, optimizer_type=emb_optim_type, optimizer_kwargs=emb_opt_kwargs, device=device, embedding_location=location, ) self._emb_modules.append(emb_module) params: Dict[str, torch.Tensor] = {} for param_key, weight in emb_module.fused_optimizer().params.items(): params[f"embedding_bags.{param_key}"] = weight optims.append(("", emb_module.fused_optimizer())) self._optim: CombinedOptimizer = CombinedOptimizer(optims) self._embedding_names: List[str] = list( itertools.chain(*get_embedding_names_by_table(self._embedding_bag_configs)) ) # We map over the parameters from FBGEMM backed kernels to the canonical nn.EmbeddingBag # representation. This provides consistency between this class and the EmbeddingBagCollection's # nn.Module API calls (state_dict, named_modules, etc) self.embedding_bags: nn.ModuleDict = nn.ModuleDict() for (_key, tables), emb_module in zip( self._key_to_tables.items(), self._emb_modules ): for embedding_config, weight in zip( tables, emb_module.split_embedding_weights(), # torch._tensor.Tensor]` is not a function. ): self.embedding_bags[embedding_config.name] = torch.nn.Module() self.embedding_bags[embedding_config.name].register_parameter( "weight", torch.nn.Parameter(weight) ) def forward(self, features: KeyedJaggedTensor) -> KeyedTensor: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: KeyedTensor """ assert features is not None feature_dict = features.to_dict() embeddings = [] for emb_op, (_key, tables) in zip( self._emb_modules, self._key_to_tables.items() ): indicies = [] lengths = [] offsets = [] weights = [] for table in tables: for feature in table.feature_names: f = feature_dict[feature] indicies.append(f.values()) lengths.append(f.lengths()) if self._is_weighted: weights.append(f.weights()) indicies = torch.cat(indicies) lengths = torch.cat(lengths) offsets = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) if self._is_weighted: weights = torch.cat(weights) embeddings.append( emb_op( indicies.int(), offsets.int(), weights if self._is_weighted else None, ) ) embeddings = torch.cat(embeddings, dim=1) return KeyedTensor( keys=self._embedding_names, values=embeddings, length_per_key=self._length_per_key, ) def _get_name(self) -> str: return "FusedEmbeddingBagCollection" def device(self) -> torch.device: return self._device def embedding_bag_configs(self) -> List[EmbeddingBagConfig]: return self._embedding_bag_configs def is_weighted(self) -> bool: return self._is_weighted def optimizer_type(self) -> Type[torch.optim.Optimizer]: return self._optimizer_type def optimizer_kwargs(self) -> Dict[str, Any]: return self._optimizer_kwargs def fused_optimizer(self) -> KeyedOptimizer: return self._optim class FusedEmbeddingCollection(EmbeddingCollectionInterface, FusedOptimizerModule): """ EmbeddingCollection represents a unsharded collection of non-pooled embeddings. The semantics of this module is that during the backwards pass, the registered optimizer will be called. It processes sparse data in the form of `KeyedJaggedTensor` of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (variable) and outputs `Dict[feature (key), JaggedTensor]`. Each `JaggedTensor` contains values of the form (B * L) X D where: * B: batch size * L: length of sparse features (jagged) * D: each feature's (key's) embedding dimension and lengths are of the form L Args: tables (List[EmbeddingConfig]): list of embedding tables. device (Optional[torch.device]): default compute device. need_indices (bool): if we need to pass indices to the final lookup dict. Example:: e1_config = EmbeddingConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) e2_config = EmbeddingConfig( name="t2", embedding_dim=3, num_embeddings=10, feature_names=["f2"] ) ec = EmbeddingCollection(tables=[e1_config, e2_config]) # 0 1 2 <-- batch # 0 [0,1] None [2] # 1 [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor.from_offsets_sync( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) feature_embeddings = ec(features) print(feature_embeddings['f2'].values()) tensor([[-0.2050, 0.5478, 0.6054], [ 0.7352, 0.3210, -3.0399], [ 0.1279, -0.1756, -0.4130], [ 0.7519, -0.4341, -0.0499], [ 0.9329, -1.0697, -0.8095]], grad_fn=<EmbeddingBackward>) """ # noqa lint def __init__( self, tables: List[EmbeddingConfig], optimizer_type: Type[torch.optim.Optimizer], optimizer_kwargs: Dict[str, Any], device: Optional[torch.device] = None, need_indices: bool = False, location: Optional[EmbeddingLocation] = None, ) -> None: super().__init__() self._optimizer_type = optimizer_type self._optimizer_kwargs = optimizer_kwargs emb_optim_and_kwargs = convert_optimizer_type_and_kwargs( optimizer_type, optimizer_kwargs ) if emb_optim_and_kwargs is None: raise ValueError( f"Cannot fuse optimizer_type={optimizer_type} with kwargs {optimizer_kwargs}" ) (emb_optim_type, emb_opt_kwargs) = emb_optim_and_kwargs if location in [ EmbeddingLocation.DEVICE, EmbeddingLocation.MANAGED, EmbeddingLocation.MANAGED_CACHING, ]: assert device is not None and device.type in [ "cuda", "meta", ], f"Using location={location} requires device=cuda or meta" if device is None: device = torch.device("cpu") assert device.type in [ "cuda", "meta", ], "FusedEmbeddingCollection is only supported for device in [CUDA, meta] currently. There are plans to support device=CPU." if location is None: if device.type in ["cpu", "meta"]: location = EmbeddingLocation.HOST elif device.type == "cuda": location = EmbeddingLocation.DEVICE else: raise ValueError("EmbeddingLocation could not be set") self._embedding_configs = tables self._need_indices: bool = need_indices self._embedding_dim: int = -1 # Registering in a List instead of ModuleList because we want don't want them to be auto-registered. # Their states will be modified via self.embedding_bags self._emb_modules: List[nn.Module] = [] self._key_to_tables: Dict[DataType, List[EmbeddingConfig]] = defaultdict(list) seen_features = set() self._shared_features: Set[str] = set() for table in tables: key = table.data_type self._key_to_tables[key].append(table) if self._embedding_dim == -1: self._embedding_dim = table.embedding_dim elif self._embedding_dim != table.embedding_dim: raise ValueError( "All tables in a EmbeddingCollection are required to have same embedding dimension." + f" Violating case: {table}'s embedding_dim {table.embedding_dim} !=" + f" {self._embedding_dim}" ) for feature in table.feature_names: if feature in seen_features: self._shared_features.add(feature) else: seen_features.add(feature) optims = [] for key, tables in self._key_to_tables.items(): data_type = key emb_module = _BatchedFusedEmbeddingLookups( cast(List[BaseEmbeddingConfig], tables), data_type=data_type, pooling=PoolingType.NONE, optimizer_type=emb_optim_type, optimizer_kwargs=emb_opt_kwargs, device=device, embedding_location=location, ) self._emb_modules.append(emb_module) params: Dict[str, torch.Tensor] = {} for param_key, weight in emb_module.fused_optimizer().params.items(): params[f"embeddings.{param_key}"] = weight optims.append(("", emb_module.fused_optimizer())) self._optim: CombinedOptimizer = CombinedOptimizer(optims) self._embedding_names: List[str] = list( itertools.chain(*get_embedding_names_by_table(self._embedding_configs)) ) self._embedding_names_by_table: List[List[str]] = get_embedding_names_by_table( self._embedding_configs, ) # We map over the parameters from FBGEMM backed kernels to the canonical nn.EmbeddingBag # representation. This provides consistency between this class and the EmbeddingBagCollection's # nn.Module API calls (state_dict, named_modules, etc) self.embeddings: nn.ModuleDict = nn.ModuleDict() for (_key, tables), emb_module in zip( self._key_to_tables.items(), self._emb_modules ): for embedding_config, weight in zip( tables, emb_module.split_embedding_weights(), # torch._tensor.Tensor]` is not a function. ): self.embeddings[embedding_config.name] = torch.nn.Module() self.embeddings[embedding_config.name].register_parameter( "weight", torch.nn.Parameter(weight) ) def forward(self, features: KeyedJaggedTensor) -> Dict[str, JaggedTensor]: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: Dict[str, JaggedTensor] """ assert features is not None feature_dict = features.to_dict() feature_embeddings: Dict[str, JaggedTensor] = {} for emb_op, (_key, tables) in zip( self._emb_modules, self._key_to_tables.items() ): indicies = [] lengths = [] offsets = [] feature_names = [] feature_lengths = [] feature_values = [] splits = [] for table in tables: for feature in table.feature_names: f = feature_dict[feature] indicies.append(f.values()) lengths.append(f.lengths()) if feature in self._shared_features: feature = f"{feature}@{table.name}" feature_names.append(feature) feature_values.append(f.values()) feature_lengths.append(f.lengths()) splits.append(torch.sum(feature_lengths[-1])) indicies = torch.cat(indicies) lengths = torch.cat(lengths) offsets = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) lookups = emb_op(indicies.int(), offsets.int(), weights=None) lookups = torch.split(lookups, split_size_or_sections=splits) for feature, lookup, feature_length, values in zip( feature_names, lookups, feature_lengths, feature_values ): feature_embeddings[feature] = JaggedTensor( values=lookup, lengths=feature_length, # hack to return kJT positional indicies in return type. weights=values if self.need_indices() else None, ) return feature_embeddings def _get_name(self) -> str: return "FusedEmbeddingCollection" def embedding_configs(self) -> List[EmbeddingConfig]: return self._embedding_configs def embedding_names_by_table(self) -> List[List[str]]: return self._embedding_names_by_table def embedding_dim(self) -> int: return self._embedding_dim def optimizer_type(self) -> Type[torch.optim.Optimizer]: return self._optimizer_type def optimizer_kwargs(self) -> Dict[str, Any]: return self._optimizer_kwargs def fused_optimizer(self) -> KeyedOptimizer: return self._optim def need_indices(self) -> bool: return self._need_indices class EmbeddingBagCollection(EmbeddingBagCollectionInterface): """ EmbeddingBagCollection represents a collection of pooled embeddings (`EmbeddingBags`). It processes sparse data in the form of `KeyedJaggedTensor` with values of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (jagged) and outputs a `KeyedTensor` with values of the form [B * (F * D)] where: * F: features (keys) * D: each feature's (key's) embedding dimension * B: batch size Args: tables (List[EmbeddingBagConfig]): list of embedding tables. is_weighted (bool): whether input `KeyedJaggedTensor` is weighted. device (Optional[torch.device]): default compute device. Example:: table_0 = EmbeddingBagConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) table_1 = EmbeddingBagConfig( name="t2", embedding_dim=4, num_embeddings=10, feature_names=["f2"] ) ebc = EmbeddingBagCollection(tables=[table_0, table_1]) # 0 1 2 <-- batch # "f1" [0,1] None [2] # "f2" [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) pooled_embeddings = ebc(features) print(pooled_embeddings.values()) tensor([[-0.8899, -0.1342, -1.9060, -0.0905, -0.2814, -0.9369, -0.7783], [ 0.0000, 0.0000, 0.0000, 0.1598, 0.0695, 1.3265, -0.1011], [-0.4256, -1.1846, -2.1648, -1.0893, 0.3590, -1.9784, -0.7681]], grad_fn=<CatBackward0>) print(pooled_embeddings.keys()) ['f1', 'f2'] print(pooled_embeddings.offset_per_key()) tensor([0, 3, 7]) """ def __init__( self, tables: List[EmbeddingBagConfig], is_weighted: bool = False, device: Optional[torch.device] = None, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self._is_weighted = is_weighted self.embedding_bags: nn.ModuleDict = nn.ModuleDict() self._embedding_bag_configs = tables self._lengths_per_embedding: List[int] = [] self._dtypes: List[int] = [] table_names = set() for embedding_config in tables: if embedding_config.name in table_names: raise ValueError(f"Duplicate table name {embedding_config.name}") table_names.add(embedding_config.name) dtype = ( torch.float32 if embedding_config.data_type == DataType.FP32 else torch.float16 ) self.embedding_bags[embedding_config.name] = nn.EmbeddingBag( num_embeddings=embedding_config.num_embeddings, embedding_dim=embedding_config.embedding_dim, mode=pooling_type_to_str(embedding_config.pooling), device=device, include_last_offset=True, dtype=dtype, ) if device is None: device = self.embedding_bags[embedding_config.name].weight.device self._dtypes.append(embedding_config.data_type.value) if not embedding_config.feature_names: embedding_config.feature_names = [embedding_config.name] self._lengths_per_embedding.extend( len(embedding_config.feature_names) * [embedding_config.embedding_dim] ) self._device: torch.device = device or torch.device("cpu") self._embedding_names: List[str] = [ embedding for embeddings in get_embedding_names_by_table(tables) for embedding in embeddings ] self._feature_names: List[List[str]] = [table.feature_names for table in tables] self.reset_parameters() def forward(self, features: KeyedJaggedTensor) -> KeyedTensor: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: KeyedTensor """ flat_feature_names: List[str] = [] for names in self._feature_names: flat_feature_names.extend(names) inverse_indices = reorder_inverse_indices( inverse_indices=features.inverse_indices_or_none(), feature_names=flat_feature_names, ) pooled_embeddings: List[torch.Tensor] = [] feature_dict = features.to_dict() for i, embedding_bag in enumerate(self.embedding_bags.values()): for feature_name in self._feature_names[i]: f = feature_dict[feature_name] per_sample_weights: Optional[torch.Tensor] = None if self._is_weighted: per_sample_weights = ( f.weights().half() if self._dtypes[i] == DataType.FP16.value else f.weights() ) res = embedding_bag( input=f.values(), offsets=f.offsets(), per_sample_weights=( per_sample_weights if self._is_weighted else None ), ).float() pooled_embeddings.append(res) return KeyedTensor( keys=self._embedding_names, values=process_pooled_embeddings( pooled_embeddings=pooled_embeddings, inverse_indices=inverse_indices, ), length_per_key=self._lengths_per_embedding, ) def is_weighted(self) -> bool: return self._is_weighted def embedding_bag_configs(self) -> List[EmbeddingBagConfig]: return self._embedding_bag_configs def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_bag_configs: assert table_config.init_fn is not None param = self.embedding_bags[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) class EmbeddingCollection(EmbeddingCollectionInterface): """ EmbeddingCollection represents a collection of non-pooled embeddings. It processes sparse data in the form of `KeyedJaggedTensor` of the form [F X B X L] where: * F: features (keys) * B: batch size * L: length of sparse features (variable) and outputs `Dict[feature (key), JaggedTensor]`. Each `JaggedTensor` contains values of the form (B * L) X D where: * B: batch size * L: length of sparse features (jagged) * D: each feature's (key's) embedding dimension and lengths are of the form L Args: tables (List[EmbeddingConfig]): list of embedding tables. device (Optional[torch.device]): default compute device. need_indices (bool): if we need to pass indices to the final lookup dict. Example:: e1_config = EmbeddingConfig( name="t1", embedding_dim=3, num_embeddings=10, feature_names=["f1"] ) e2_config = EmbeddingConfig( name="t2", embedding_dim=3, num_embeddings=10, feature_names=["f2"] ) ec = EmbeddingCollection(tables=[e1_config, e2_config]) # 0 1 2 <-- batch # 0 [0,1] None [2] # 1 [3] [4] [5,6,7] # ^ # feature features = KeyedJaggedTensor.from_offsets_sync( keys=["f1", "f2"], values=torch.tensor([0, 1, 2, 3, 4, 5, 6, 7]), offsets=torch.tensor([0, 2, 2, 3, 4, 5, 8]), ) feature_embeddings = ec(features) print(feature_embeddings['f2'].values()) tensor([[-0.2050, 0.5478, 0.6054], [ 0.7352, 0.3210, -3.0399], [ 0.1279, -0.1756, -0.4130], [ 0.7519, -0.4341, -0.0499], [ 0.9329, -1.0697, -0.8095]], grad_fn=<EmbeddingBackward>) """ def __init__( # noqa C901 self, tables: List[EmbeddingConfig], device: Optional[torch.device] = None, need_indices: bool = False, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.modules.{self.__class__.__name__}") self.embeddings: nn.ModuleDict = nn.ModuleDict() self._embedding_configs = tables self._embedding_dim: int = -1 self._need_indices: bool = need_indices self._device: torch.device = ( device if device is not None else torch.device("cpu") ) table_names = set() for config in tables: if config.name in table_names: raise ValueError(f"Duplicate table name {config.name}") table_names.add(config.name) self._embedding_dim = ( config.embedding_dim if self._embedding_dim < 0 else self._embedding_dim ) if self._embedding_dim != config.embedding_dim: raise ValueError( "All tables in a EmbeddingCollection are required to have same embedding dimension." + f" Violating case: {config.name}'s embedding_dim {config.embedding_dim} !=" + f" {self._embedding_dim}" ) dtype = ( torch.float32 if config.data_type == DataType.FP32 else torch.float16 ) self.embeddings[config.name] = nn.Embedding( num_embeddings=config.num_embeddings, embedding_dim=config.embedding_dim, device=device, dtype=dtype, ) if config.init_fn is not None: config.init_fn(self.embeddings[config.name].weight) if not config.feature_names: config.feature_names = [config.name] self._embedding_names_by_table: List[List[str]] = get_embedding_names_by_table( tables ) self._feature_names: List[List[str]] = [table.feature_names for table in tables] def forward( self, features: KeyedJaggedTensor, ) -> Dict[str, JaggedTensor]: """ Args: features (KeyedJaggedTensor): KJT of form [F X B X L]. Returns: Dict[str, JaggedTensor] """ feature_embeddings: Dict[str, JaggedTensor] = {} jt_dict: Dict[str, JaggedTensor] = features.to_dict() for i, emb_module in enumerate(self.embeddings.values()): feature_names = self._feature_names[i] embedding_names = self._embedding_names_by_table[i] for j, embedding_name in enumerate(embedding_names): feature_name = feature_names[j] f = jt_dict[feature_name] lookup = emb_module( input=f.values(), ).float() feature_embeddings[embedding_name] = JaggedTensor( values=lookup, lengths=f.lengths(), weights=f.values() if self._need_indices else None, ) return feature_embeddings def need_indices(self) -> bool: return self._need_indices def embedding_dim(self) -> int: return self._embedding_dim def embedding_configs(self) -> List[EmbeddingConfig]: return self._embedding_configs def embedding_names_by_table(self) -> List[List[str]]: return self._embedding_names_by_table def device(self) -> torch.device: return self._device def reset_parameters(self) -> None: if (isinstance(self.device, torch.device) and self.device.type == "meta") or ( isinstance(self.device, str) and self.device == "meta" ): return # Initialize embedding bags weights with init_fn for table_config in self._embedding_configs: assert table_config.init_fn is not None param = self.embeddings[f"{table_config.name}"].weight # pyre-ignore table_config.init_fn(param) The provided code snippet includes necessary dependencies for implementing the `fuse_embedding_optimizer` function. Write a Python function `def fuse_embedding_optimizer( model: nn.Module, optimizer_type: Type[torch.optim.Optimizer], optimizer_kwargs: Dict[str, Any], device: torch.device, location: Optional[EmbeddingLocation] = None, ) -> nn.Module` to solve the following problem: Recursively replaces EmbeddingBagCollection and EmbeddingCollection with FusedEmbeddingBagCollection and FusedEmbeddingCollection in a model subtree. The fused modules will be initialized using the passed in optimizer parameters, and model location. Args: model: (nn.Module): optimizer_type: (Type[torch.optim.Optimizer]): optimizer_kwargs: (Dict[Str, Any]): device (Optional[torch.device]): location: (Optional[EmbeddingLocation]): GPU location placement Returns nn.Module: input nn.Module with Fused Embedding Modules Example:: ebc = EmbeddingBagCollection() my_model = ExampleModel(ebc) my_model = fused_embedding_optimizer(my_model, optimizer_type=torch.optim.SGD, optimizer_kwargs={"lr": .01}) kjt = KeyedJaggedTensor() output = my_model(kjt) Here is the function: def fuse_embedding_optimizer( model: nn.Module, optimizer_type: Type[torch.optim.Optimizer], optimizer_kwargs: Dict[str, Any], device: torch.device, location: Optional[EmbeddingLocation] = None, ) -> nn.Module: """ Recursively replaces EmbeddingBagCollection and EmbeddingCollection with FusedEmbeddingBagCollection and FusedEmbeddingCollection in a model subtree. The fused modules will be initialized using the passed in optimizer parameters, and model location. Args: model: (nn.Module): optimizer_type: (Type[torch.optim.Optimizer]): optimizer_kwargs: (Dict[Str, Any]): device (Optional[torch.device]): location: (Optional[EmbeddingLocation]): GPU location placement Returns nn.Module: input nn.Module with Fused Embedding Modules Example:: ebc = EmbeddingBagCollection() my_model = ExampleModel(ebc) my_model = fused_embedding_optimizer(my_model, optimizer_type=torch.optim.SGD, optimizer_kwargs={"lr": .01}) kjt = KeyedJaggedTensor() output = my_model(kjt) """ # Replace all EBCs and ECs in a with a corresponding FusedEmbeddingModule. # check if top-level module is EBC/EC if isinstance(model, EmbeddingBagCollection): return FusedEmbeddingBagCollection( model.embedding_bag_configs(), optimizer_type=optimizer_type, optimizer_kwargs=optimizer_kwargs, device=device, location=location, ) if isinstance(model, EmbeddingCollection): return FusedEmbeddingCollection( model.embedding_configs(), optimizer_type=optimizer_type, optimizer_kwargs=optimizer_kwargs, device=device, location=location, ) def replace(_model: nn.Module) -> None: for child_name, child in _model.named_children(): if isinstance(child, EmbeddingBagCollection): setattr( _model, child_name, FusedEmbeddingBagCollection( tables=child.embedding_bag_configs(), optimizer_type=optimizer_type, optimizer_kwargs=optimizer_kwargs, device=device, location=location, ), ) elif isinstance(child, EmbeddingCollection): setattr( _model, child_name, FusedEmbeddingCollection( tables=child.embedding_configs(), optimizer_type=optimizer_type, optimizer_kwargs=optimizer_kwargs, device=device, location=location, ), ) else: replace(child) replace(model) return model
Recursively replaces EmbeddingBagCollection and EmbeddingCollection with FusedEmbeddingBagCollection and FusedEmbeddingCollection in a model subtree. The fused modules will be initialized using the passed in optimizer parameters, and model location. Args: model: (nn.Module): optimizer_type: (Type[torch.optim.Optimizer]): optimizer_kwargs: (Dict[Str, Any]): device (Optional[torch.device]): location: (Optional[EmbeddingLocation]): GPU location placement Returns nn.Module: input nn.Module with Fused Embedding Modules Example:: ebc = EmbeddingBagCollection() my_model = ExampleModel(ebc) my_model = fused_embedding_optimizer(my_model, optimizer_type=torch.optim.SGD, optimizer_kwargs={"lr": .01}) kjt = KeyedJaggedTensor() output = my_model(kjt)
8,881
from dataclasses import dataclass, field from enum import Enum, unique from functools import partial from math import sqrt from typing import Callable, Dict, List, NamedTuple, Optional import torch from fbgemm_gpu.split_embedding_configs import SparseType from fbgemm_gpu.split_table_batched_embeddings_ops_training import PoolingMode from torchrec.types import DataType class PoolingType(Enum): SUM = "SUM" MEAN = "MEAN" NONE = "NONE" def pooling_type_to_pooling_mode(pooling_type: PoolingType) -> PoolingMode: if pooling_type.value == PoolingType.SUM.value: return PoolingMode.SUM elif pooling_type.value == PoolingType.MEAN.value: return PoolingMode.MEAN elif pooling_type.value == PoolingType.NONE.value: return PoolingMode.NONE else: raise Exception(f"Invalid pooling type {pooling_type}")
null
8,882
from dataclasses import dataclass, field from enum import Enum, unique from functools import partial from math import sqrt from typing import Callable, Dict, List, NamedTuple, Optional import torch from fbgemm_gpu.split_embedding_configs import SparseType from fbgemm_gpu.split_table_batched_embeddings_ops_training import PoolingMode from torchrec.types import DataType class PoolingType(Enum): SUM = "SUM" MEAN = "MEAN" NONE = "NONE" def pooling_type_to_str(pooling_type: PoolingType) -> str: if pooling_type.value == PoolingType.SUM.value: return "sum" elif pooling_type.value == PoolingType.MEAN.value: return "mean" else: raise ValueError(f"Unsupported pooling type {pooling_type}")
null
8,883
from dataclasses import dataclass, field from enum import Enum, unique from functools import partial from math import sqrt from typing import Callable, Dict, List, NamedTuple, Optional import torch from fbgemm_gpu.split_embedding_configs import SparseType from fbgemm_gpu.split_table_batched_embeddings_ops_training import PoolingMode from torchrec.types import DataType class DataType(Enum): def __str__(self) -> str: def data_type_to_dtype(data_type: DataType) -> torch.dtype: if data_type.value == DataType.FP32.value: return torch.float32 elif data_type.value == DataType.FP16.value: return torch.float16 elif data_type.value == DataType.BF16.value: return torch.bfloat16 elif data_type.value == DataType.INT64.value: return torch.int64 elif data_type.value == DataType.INT32.value: return torch.int32 elif data_type.value == DataType.INT8.value: return torch.int8 elif data_type.value == DataType.UINT8.value: return torch.uint8 elif data_type.value == DataType.INT4.value: return torch.quint4x2 elif data_type.value == DataType.INT2.value: return torch.quint2x4 else: raise ValueError(f"DataType {data_type} cannot be converted to dtype")
null
8,884
from typing import Dict, List, Set, Union import torch import torch.nn as nn from torchrec.modules.embedding_modules import EmbeddingBagCollection from torchrec.modules.feature_processor_ import ( FeatureProcessor, FeatureProcessorsCollection, ) from torchrec.sparse.jagged_tensor import KeyedJaggedTensor, KeyedTensor # pyre-ignore class KeyedJaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """Represents an (optionally weighted) keyed jagged tensor. A `KeyedJaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. Keyed on first dimension and jagged on the last dimension. Implementation is torch.jit.script-able. Args: keys (List[str]): keys to the jagged Tensor. values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if the values have weights. Tensor with the same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. stride (Optional[int]): number of examples per batch. stride_per_key_per_rank (Optional[List[List[int]]]): batch size (number of examples) per key per rank, with the outer list representing the keys and the inner list representing the values. Each value in the inner list represents the number of examples in the batch from the rank of its index in a distributed context. length_per_key (Optional[List[int]]): start length for each key. offset_per_key (Optional[List[int]]): start offset for each key and final offset. index_per_key (Optional[Dict[str, int]]): index for each key. jt_dict (Optional[Dict[str, JaggedTensor]]): inverse_indices (Optional[Tuple[List[str], torch.Tensor]]): inverse indices to expand deduplicated embedding output for variable stride per key. Example:: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 dim_0: keyed dimension (ie. `Feature0`, `Feature1`) dim_1: optional second dimension (ie. batch size) dim_2: The jagged dimension which has slice lengths between 0-3 in the above example # We represent this data with following inputs: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset """ # This is the subset of fields on KJT which are required (all other fields # can be derived from these fields, and are only cached) _fields = [ "_values", "_weights", "_lengths", "_offsets", ] def __init__( self, keys: List[str], values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, # Below exposed to ensure torch.script-able length_per_key: Optional[List[int]] = None, offset_per_key: Optional[List[int]] = None, index_per_key: Optional[Dict[str, int]] = None, jt_dict: Optional[Dict[str, JaggedTensor]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> None: self._keys: List[str] = keys self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets self._stride_per_key_per_rank: List[List[int]] = [] self._stride_per_key: List[int] = [] self._variable_stride_per_key: bool = False self._stride: int = -1 if stride_per_key_per_rank is not None: if stride is not None: raise ValueError( "Cannot initialize KJT with both `stride` and `stride_per_key_per_rank`" ) self._stride_per_key_per_rank = stride_per_key_per_rank self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] self._variable_stride_per_key = True if not stride_per_key_per_rank: self._stride = 0 elif all(s == self.stride_per_key()[0] for s in self.stride_per_key()): self._stride = self.stride_per_key()[0] else: if torch.jit.is_tracing(): stride = _maybe_compute_stride_kjt_scripted( keys, stride, lengths, offsets )[0] else: stride = _maybe_compute_stride_kjt(keys, stride, lengths, offsets) self._stride = stride self._stride_per_key_per_rank = [[stride]] * len(self._keys) self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] # lazy fields self._length_per_key: Optional[List[int]] = length_per_key self._offset_per_key: Optional[List[int]] = offset_per_key self._index_per_key: Optional[Dict[str, int]] = index_per_key self._jt_dict: Optional[Dict[str, JaggedTensor]] = jt_dict self._inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = ( inverse_indices ) self._lengths_offset_per_key: List[int] = [] def from_offsets_sync( keys: List[str], values: torch.Tensor, offsets: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, offsets=offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def from_lengths_sync( keys: List[str], values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def concat( kjt_list: List["KeyedJaggedTensor"], ) -> "KeyedJaggedTensor": if len(kjt_list) == 0: raise ValueError("Can't concat empty KJT list") is_weighted: bool = kjt_list[0].weights_or_none() is not None has_length_per_key: bool = True length_per_key: List[int] = [] keys: List[str] = [] value_list: List[torch.Tensor] = [] weight_list: List[torch.Tensor] = [] length_list: List[torch.Tensor] = [] stride_per_key_per_rank: List[List[int]] = [] stride: Optional[int] = None variable_stride_per_key_list = [ kjt.variable_stride_per_key() for kjt in kjt_list ] assert all(variable_stride_per_key_list) or not any( variable_stride_per_key_list ), "variable stride per key must be consistent for all KJTs" variable_stride_per_key = all(variable_stride_per_key_list) for kjt in kjt_list: curr_is_weighted: bool = kjt.weights_or_none() is not None if is_weighted != curr_is_weighted: raise ValueError("Can't merge weighted KJT with unweighted KJT") _length_per_key: Optional[List[int]] = None if kjt._length_per_key is None: has_length_per_key = False else: _length_per_key = kjt._length_per_key if has_length_per_key and _length_per_key is not None: length_per_key += _length_per_key keys += kjt.keys() value_list.append(kjt.values()) if is_weighted: weight_list.append(kjt.weights()) length_list.append(kjt.lengths()) if variable_stride_per_key: stride_per_key_per_rank += kjt.stride_per_key_per_rank() elif stride is None: stride = kjt.stride() else: assert stride == kjt.stride(), "strides must be consistent for all KJTs" return KeyedJaggedTensor( keys=keys, values=torch.cat(value_list, dim=0), weights=torch.cat(weight_list, dim=0) if is_weighted else None, lengths=torch.cat(length_list, dim=0), stride=stride, stride_per_key_per_rank=( stride_per_key_per_rank if variable_stride_per_key else None ), length_per_key=length_per_key if has_length_per_key else None, ) def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "KeyedJaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return KeyedJaggedTensor( keys=torch.jit.annotate(List[str], []), values=torch.empty(0, dtype=values_dtype, device=device), weights=weights, lengths=torch.empty(0, dtype=lengths_dtype, device=device), stride=0, ) def empty_like(kjt: "KeyedJaggedTensor") -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, kjt.stride_per_key_per_rank()) if kjt.variable_stride_per_key() else (kjt.stride(), None) ) return KeyedJaggedTensor( keys=[], values=torch.empty(0, device=kjt.device(), dtype=kjt.values().dtype), weights=( None if kjt.weights_or_none() is None else torch.empty(0, device=kjt.device(), dtype=kjt.weights().dtype) ), lengths=torch.empty(0, device=kjt.device(), dtype=kjt.lengths().dtype), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, ) def from_jt_dict(jt_dict: Dict[str, JaggedTensor]) -> "KeyedJaggedTensor": """ Constructs a KeyedJaggedTensor from a Dict[str, JaggedTensor], but this function will ONLY work if the JaggedTensors all have the same "implicit" batch_size dimension. Basically, we can visualize JaggedTensors as 2-D tensors of the format of [batch_size x variable_feature_dim]. In case, we have some batch without a feature value, the input JaggedTensor could just not include any values. But KeyedJaggedTensor (by default) typically pad "None" so that all the JaggedTensors stored in the KeyedJaggedTensor have the same batch_size dimension. That is, in the case, the JaggedTensor input didn't automatically pad for the empty batches, this function would error / not work. Consider the visualization of the following KeyedJaggedTensor: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 Notice that the inputs for this KeyedJaggedTensor would have looked like: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset Now if the input jt_dict = { # "Feature0" [V0,V1] [V2] # "Feature1" [V3] [V4] [V5,V6,V7] } and the "None" is left out from each JaggedTensor, then this function would fail as we would not correctly be able to pad "None" as it does not technically know the correct batch / place to pad within the JaggedTensor. Essentially, the lengths Tensor inferred by this function would be [2, 1, 1, 1, 3] indicating variable batch_size dim_1 violates the existing assumption / precondition that KeyedJaggedTensor's should have fixed batch_size dimension. """ kjt_keys = list(jt_dict.keys()) kjt_vals_list: List[torch.Tensor] = [] kjt_lens_list: List[torch.Tensor] = [] kjt_weights_list: List[torch.Tensor] = [] stride_per_key: List[int] = [] for jt in jt_dict.values(): stride_per_key.append(len(jt.lengths())) kjt_vals_list.append(jt.values()) kjt_lens_list.append(jt.lengths()) weight = jt.weights_or_none() if weight is not None: kjt_weights_list.append(weight) kjt_vals = torch.concat(kjt_vals_list) kjt_lens = torch.concat(kjt_lens_list) kjt_weights = ( torch.concat(kjt_weights_list) if len(kjt_weights_list) > 0 else None ) kjt_stride, kjt_stride_per_key_per_rank = ( (stride_per_key[0], None) if all(s == stride_per_key[0] for s in stride_per_key) else (None, [[stride] for stride in stride_per_key]) ) kjt = KeyedJaggedTensor( keys=kjt_keys, values=kjt_vals, weights=kjt_weights, lengths=kjt_lens, stride=kjt_stride, stride_per_key_per_rank=kjt_stride_per_key_per_rank, ).sync() return kjt def sync(self) -> "KeyedJaggedTensor": self.length_per_key() self.offset_per_key() return self def unsync(self) -> "KeyedJaggedTensor": self._length_per_key = None self._offset_per_key = None return self def device(self) -> torch.device: return self._values.device def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def keys(self) -> List[str]: return self._keys def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def stride(self) -> int: return self._stride def stride_per_key(self) -> List[int]: return self._stride_per_key def stride_per_key_per_rank(self) -> List[List[int]]: return self._stride_per_key_per_rank def variable_stride_per_key(self) -> bool: return self._variable_stride_per_key def inverse_indices(self) -> Tuple[List[str], torch.Tensor]: return _get_inverse_indices_or_throw(self._inverse_indices) def inverse_indices_or_none(self) -> Optional[Tuple[List[str], torch.Tensor]]: return self._inverse_indices def _key_indices(self) -> Dict[str, int]: _index_per_key: Dict[str, int] = _maybe_compute_index_per_key( self._keys, self._index_per_key, ) self._index_per_key = _index_per_key return _index_per_key def length_per_key(self) -> List[int]: _length_per_key = _maybe_compute_length_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key return _length_per_key def length_per_key_or_none(self) -> Optional[List[int]]: return self._length_per_key def offset_per_key(self) -> List[int]: _length_per_key, _offset_per_key = _maybe_compute_offset_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key self._offset_per_key = _offset_per_key return _offset_per_key def offset_per_key_or_none(self) -> Optional[List[int]]: return self._offset_per_key def lengths_offset_per_key(self) -> List[int]: if not self._lengths_offset_per_key: self._lengths_offset_per_key = _cumsum(self.stride_per_key()) return self._lengths_offset_per_key def split(self, segments: List[int]) -> List["KeyedJaggedTensor"]: split_list: List[KeyedJaggedTensor] = [] start = 0 start_offset = 0 _length_per_key = self.length_per_key() _offset_per_key = self.offset_per_key() for segment in segments: end = start + segment end_offset = _offset_per_key[end] keys: List[str] = self._keys[start:end] stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()[start:end]) if self.variable_stride_per_key() else (self._stride, None) ) if segment == len(self._keys): # no torch slicing required split_list.append( KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self.weights_or_none(), lengths=self._lengths, offsets=self._offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=self._jt_dict, inverse_indices=None, ) ) elif segment == 0: empty_int_list: List[int] = torch.jit.annotate(List[int], []) split_list.append( KeyedJaggedTensor( keys=keys, values=torch.tensor( empty_int_list, device=self.device(), dtype=self._values.dtype, ), weights=( None if self.weights_or_none() is None else torch.tensor( empty_int_list, device=self.device(), dtype=self.weights().dtype, ) ), lengths=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), offsets=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) else: split_length_per_key = _length_per_key[start:end] if not torch.jit.is_scripting() and is_torchdynamo_compiling(): # Checks for dynamo dynamic shapes tracing torch._check_is_size(start_offset) torch._check_is_size(end_offset) torch._check_is_size(end_offset - start_offset) torch._check(start_offset <= self._values.size(0)) torch._check(end_offset <= self._values.size(0)) torch._check(end_offset >= start_offset) split_list.append( KeyedJaggedTensor( keys=keys, values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[ start ] : self.lengths_offset_per_key()[end] ], offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=split_length_per_key, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) start = end start_offset = end_offset return split_list def permute( self, indices: List[int], indices_tensor: Optional[torch.Tensor] = None, include_inverse_indices: bool = False, ) -> "KeyedJaggedTensor": if indices_tensor is None: indices_tensor = torch.tensor( indices, dtype=torch.int, device=self.device() ) length_per_key = self.length_per_key() permuted_keys: List[str] = [] permuted_stride_per_key_per_rank: List[List[int]] = [] permuted_length_per_key: List[int] = [] permuted_lengths_sum = 0 for index in indices: key = self.keys()[index] permuted_keys.append(key) permuted_stride_per_key_per_rank.append( self.stride_per_key_per_rank()[index] ) permuted_length_per_key.append(length_per_key[index]) permuted_lengths_sum += length_per_key[index] if self.variable_stride_per_key(): length_per_key_tensor = _pin_and_move( torch.tensor(self.length_per_key()), self.device() ) stride_per_key_tensor = _pin_and_move( torch.tensor(self.stride_per_key()), self.device() ) permuted_lengths, _ = _permute_tensor_by_segments( self.lengths(), stride_per_key_tensor, indices_tensor, None, ) permuted_values, permuted_weights = _permute_tensor_by_segments( self.values(), length_per_key_tensor, indices_tensor, self.weights_or_none(), ) else: ( permuted_lengths, permuted_values, permuted_weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( indices_tensor, self.lengths().view(len(self._keys), -1), self.values(), self.weights_or_none(), permuted_lengths_sum, ) stride, optional_permuted_stride_per_key_per_rank = ( (None, permuted_stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) kjt = KeyedJaggedTensor( keys=permuted_keys, values=permuted_values, weights=permuted_weights, lengths=permuted_lengths.view(-1), offsets=None, stride=stride, stride_per_key_per_rank=optional_permuted_stride_per_key_per_rank, length_per_key=permuted_length_per_key if len(permuted_keys) > 0 else None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=( self.inverse_indices_or_none() if include_inverse_indices else None ), ) return kjt def flatten_lengths(self) -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()) if self.variable_stride_per_key() else (self._stride, None) ) return KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self._weights, lengths=self.lengths().view(-1), offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self.length_per_key(), offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) def __getitem__(self, key: str) -> JaggedTensor: offset_per_key = self.offset_per_key() index = self._key_indices()[key] start_offset = offset_per_key[index] end_offset = ( offset_per_key[index + 1] if index + 1 < len(offset_per_key) else start_offset ) return JaggedTensor( values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[index] : self.lengths_offset_per_key()[ index + 1 ] ], offsets=None, ) def to_dict(self) -> Dict[str, JaggedTensor]: _jt_dict = _maybe_compute_kjt_to_jt_dict( stride=self.stride(), stride_per_key=self.stride_per_key(), keys=self.keys(), length_per_key=self.length_per_key(), lengths=self.lengths(), values=self.values(), variable_stride_per_key=self.variable_stride_per_key(), weights=self.weights_or_none(), jt_dict=self._jt_dict, ) self._jt_dict = _jt_dict return _jt_dict def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def to( self, device: torch.device, non_blocking: bool = False, dtype: Optional[torch.dtype] = None, ) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) length_per_key = self._length_per_key offset_per_key = self._offset_per_key index_per_key = self._index_per_key jt_dict = self._jt_dict inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = ( inverse_indices[0], inverse_indices[1].to(device, non_blocking=non_blocking), ) if weights is not None: if dtype is not None: weights = weights.to( dtype=dtype, device=device, non_blocking=non_blocking ) else: weights = weights.to(device=device, non_blocking=non_blocking) return KeyedJaggedTensor( keys=self._keys, values=self._values.to(device, non_blocking=non_blocking), weights=weights, lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=length_per_key, offset_per_key=offset_per_key, index_per_key=index_per_key, jt_dict=jt_dict, inverse_indices=inverse_indices, ) def __str__(self) -> str: if len(self._keys) == 0 or self._offsets is None and self._lengths is None: return "KeyedJaggedTensor()\n" offsets = self.offsets() return ( "KeyedJaggedTensor({\n" + ",\n".join( [ " " + _jagged_tensor_string( self._keys[index], self._values, self._weights, offsets, sum(self.stride_per_key()[:index]), sum(self.stride_per_key()[: index + 1]), ) for index in range(len(self._keys)) ] ) + "\n})\n" ) def pin_memory(self) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = (inverse_indices[0], inverse_indices[1].pin_memory()) return KeyedJaggedTensor( keys=self._keys, values=self._values.pin_memory(), weights=weights.pin_memory() if weights is not None else None, lengths=lengths.pin_memory() if lengths is not None else None, offsets=offsets.pin_memory() if offsets is not None else None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=None, inverse_indices=inverse_indices, ) def dist_labels(self) -> List[str]: labels = ["lengths", "values"] if self.variable_stride_per_key(): labels.append("strides") if self.weights_or_none() is not None: labels.append("weights") return labels def dist_splits(self, key_splits: List[int]) -> List[List[int]]: batch_size_per_split = _sum_by_splits(self.stride_per_key(), key_splits) length_per_split = _sum_by_splits(self.length_per_key(), key_splits) splits = [batch_size_per_split, length_per_split] if self.variable_stride_per_key(): splits.append(key_splits) if self.weights_or_none() is not None: splits.append(length_per_split) return splits def dist_tensors(self) -> List[torch.Tensor]: tensors = [self.lengths(), self.values()] if self.variable_stride_per_key(): strides = _pin_and_move(torch.tensor(self.stride_per_key()), self.device()) tensors.append(strides) if self.weights_or_none() is not None: tensors.append(self.weights()) return tensors def dist_init( keys: List[str], tensors: List[torch.Tensor], variable_stride_per_key: bool, num_workers: int, recat: Optional[torch.Tensor], stride_per_rank: Optional[List[int]], stagger: int = 1, ) -> "KeyedJaggedTensor": assert len(tensors) in [2, 3, 4] lengths = tensors[0] values = tensors[1] stride_per_rank_per_key = tensors[2] if variable_stride_per_key else None weights = ( tensors[-1] if (variable_stride_per_key and len(tensors) == 4) or (not variable_stride_per_key and len(tensors) == 3) else None ) if variable_stride_per_key: assert stride_per_rank_per_key is not None stride_per_key_per_rank: List[List[int]] = stride_per_rank_per_key.view( num_workers, len(keys) ).T.tolist() strides_cumsum: List[int] = torch.ops.fbgemm.asynchronous_complete_cumsum( stride_per_rank_per_key ).tolist() cumsum_lengths = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) length_per_key = ( cumsum_lengths[strides_cumsum[1:]] - cumsum_lengths[strides_cumsum[:-1]] ) with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: lengths, _ = _permute_tensor_by_segments( lengths, stride_per_rank_per_key, recat, None, ) values, weights = _permute_tensor_by_segments( values, length_per_key, recat, weights, ) if not stride_per_key_per_rank: stride_per_key_per_rank = [[0]] * len(keys) if stagger > 1: stride_per_key_per_rank_stagger: List[List[int]] = [] local_world_size = num_workers // stagger for i in range(len(keys)): stride_per_rank_stagger: List[int] = [] for j in range(local_world_size): stride_per_rank_stagger.extend( stride_per_key_per_rank[i][j::local_world_size] ) stride_per_key_per_rank_stagger.append(stride_per_rank_stagger) stride_per_key_per_rank = stride_per_key_per_rank_stagger kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride_per_key_per_rank=stride_per_key_per_rank, ) return kjt.sync() else: assert stride_per_rank is not None with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: stride = stride_per_rank[0] if all(s == stride for s in stride_per_rank): ( lengths, values, weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( recat, lengths.view(-1, stride), values, weights, values.numel(), ) lengths = lengths.view(-1) else: # variable batch size per rank ( lengths, values, weights, ) = torch.ops.fbgemm.permute_1D_sparse_data( recat, lengths.view(-1), values, weights, values.numel(), ) kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=sum(stride_per_rank), ) return kjt.sync() def apply_feature_processors_to_kjt( features: KeyedJaggedTensor, feature_processors: Dict[str, nn.Module], ) -> KeyedJaggedTensor: processed_weights = [] features_dict = features.to_dict() for key in features.keys(): jt = features_dict[key] if key in feature_processors: fp_jt = feature_processors[key](jt) processed_weights.append(fp_jt.weights()) else: processed_weights.append( torch.ones(jt.values().shape[0], device=jt.values().device), ) return KeyedJaggedTensor( keys=features.keys(), values=features.values(), weights=( torch.cat(processed_weights) if processed_weights else features.weights_or_none() ), lengths=features.lengths(), offsets=features._offsets, stride=features._stride, length_per_key=features._length_per_key, offset_per_key=features._offset_per_key, index_per_key=features._index_per_key, )
null
8,885
import functools import inspect from typing import Any, Callable import torch import torch.utils.hooks as hooks from torch.nn.modules.lazy import _LazyProtocol, LazyModuleMixin from torch.nn.modules.module import ( _global_backward_hooks, _global_forward_hooks, _global_forward_pre_hooks, ) def _apply_functions_after_first_forward( module: torch.nn.Module, # pyre-ignore[2] input: Any, # pyre-ignore[2] output: Any, ) -> None: _functions_to_lazy_apply = getattr(module, "_functions_to_lazy_apply", None) if _functions_to_lazy_apply is not None: for fn in _functions_to_lazy_apply: module.apply(fn) delattr(module, "_functions_to_lazy_apply") module._lazy_apply_hook.remove() delattr(module, "_lazy_apply_hook") The provided code snippet includes necessary dependencies for implementing the `lazy_apply` function. Write a Python function `def lazy_apply( module: torch.nn.Module, fn: Callable[[torch.nn.Module], None] ) -> torch.nn.Module` to solve the following problem: Attaches a function to a module, which will be applied recursively to every submodule (as returned by `.children()`) of the module as well as the module itself right after the first forward pass (i.e. after all submodules and parameters have been initialized). Typical use includes initializing the numerical value of the parameters of a lazy module (i.e. modules inherited from `LazyModuleMixin`). NOTE: `lazy_apply()` can be used on both lazy and non-lazy modules. Args: module (torch.nn.Module): module to recursively apply `fn` on. fn (Callable[[torch.nn.Module], None]): function to be attached to `module` and later be applied to each submodule of `module` and the `module` itself. Returns: torch.nn.Module: `module` with `fn` attached. Example:: @torch.no_grad() def init_weights(m): print(m) if type(m) == torch.nn.LazyLinear: m.weight.fill_(1.0) print(m.weight) linear = torch.nn.LazyLinear(2) lazy_apply(linear, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) linear(input) # runs `init_weights` only once, right after first forward pass seq = torch.nn.Sequential(torch.nn.LazyLinear(2), torch.nn.LazyLinear(2)) lazy_apply(seq, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) seq(input) # runs `init_weights` only once, right after first forward pass Here is the function: def lazy_apply( module: torch.nn.Module, fn: Callable[[torch.nn.Module], None] ) -> torch.nn.Module: """Attaches a function to a module, which will be applied recursively to every submodule (as returned by `.children()`) of the module as well as the module itself right after the first forward pass (i.e. after all submodules and parameters have been initialized). Typical use includes initializing the numerical value of the parameters of a lazy module (i.e. modules inherited from `LazyModuleMixin`). NOTE: `lazy_apply()` can be used on both lazy and non-lazy modules. Args: module (torch.nn.Module): module to recursively apply `fn` on. fn (Callable[[torch.nn.Module], None]): function to be attached to `module` and later be applied to each submodule of `module` and the `module` itself. Returns: torch.nn.Module: `module` with `fn` attached. Example:: @torch.no_grad() def init_weights(m): print(m) if type(m) == torch.nn.LazyLinear: m.weight.fill_(1.0) print(m.weight) linear = torch.nn.LazyLinear(2) lazy_apply(linear, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) linear(input) # runs `init_weights` only once, right after first forward pass seq = torch.nn.Sequential(torch.nn.LazyLinear(2), torch.nn.LazyLinear(2)) lazy_apply(seq, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) seq(input) # runs `init_weights` only once, right after first forward pass """ if not hasattr(module, "_functions_to_lazy_apply"): module._functions_to_lazy_apply = [] if not hasattr(module, "_lazy_apply_hook"): module._lazy_apply_hook = module.register_forward_hook( _apply_functions_after_first_forward ) module._functions_to_lazy_apply.append(fn) return module
Attaches a function to a module, which will be applied recursively to every submodule (as returned by `.children()`) of the module as well as the module itself right after the first forward pass (i.e. after all submodules and parameters have been initialized). Typical use includes initializing the numerical value of the parameters of a lazy module (i.e. modules inherited from `LazyModuleMixin`). NOTE: `lazy_apply()` can be used on both lazy and non-lazy modules. Args: module (torch.nn.Module): module to recursively apply `fn` on. fn (Callable[[torch.nn.Module], None]): function to be attached to `module` and later be applied to each submodule of `module` and the `module` itself. Returns: torch.nn.Module: `module` with `fn` attached. Example:: @torch.no_grad() def init_weights(m): print(m) if type(m) == torch.nn.LazyLinear: m.weight.fill_(1.0) print(m.weight) linear = torch.nn.LazyLinear(2) lazy_apply(linear, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) linear(input) # runs `init_weights` only once, right after first forward pass seq = torch.nn.Sequential(torch.nn.LazyLinear(2), torch.nn.LazyLinear(2)) lazy_apply(seq, init_weights) # doesn't run `init_weights` immediately input = torch.randn(2, 10) seq(input) # runs `init_weights` only once, right after first forward pass
8,886
import abc from typing import Dict, List, Optional, Tuple, Union import torch import torch.nn as nn from torchrec.modules.embedding_configs import ( DataType, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_str, ) from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor def reorder_inverse_indices( inverse_indices: Optional[Tuple[List[str], torch.Tensor]], feature_names: List[str], ) -> torch.Tensor: if inverse_indices is None: return torch.empty(0) index_per_name = {name: i for i, name in enumerate(inverse_indices[0])} index = torch.tensor( [index_per_name[name.split("@")[0]] for name in feature_names], device=inverse_indices[1].device, ) return torch.index_select(inverse_indices[1], 0, index)
null
8,887
import abc from typing import Dict, List, Optional, Tuple, Union import torch import torch.nn as nn from torchrec.modules.embedding_configs import ( DataType, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_str, ) from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor def process_pooled_embeddings( pooled_embeddings: List[torch.Tensor], inverse_indices: torch.Tensor, ) -> torch.Tensor: if inverse_indices.numel() > 0: pooled_embeddings = torch.ops.fbgemm.group_index_select_dim0( pooled_embeddings, list(torch.unbind(inverse_indices)) ) return torch.cat(pooled_embeddings, dim=1)
null
8,888
import abc from typing import Dict, List, Optional, Tuple, Union import torch import torch.nn as nn from torchrec.modules.embedding_configs import ( DataType, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_str, ) from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor class EmbeddingBagConfig(BaseEmbeddingConfig): pooling: PoolingType = PoolingType.SUM class EmbeddingConfig(BaseEmbeddingConfig): pass def get_embedding_names_by_table( tables: Union[List[EmbeddingBagConfig], List[EmbeddingConfig]], ) -> List[List[str]]: shared_feature: Dict[str, bool] = {} for embedding_config in tables: for feature_name in embedding_config.feature_names: if feature_name not in shared_feature: shared_feature[feature_name] = False else: shared_feature[feature_name] = True embedding_names_by_table: List[List[str]] = [] for embedding_config in tables: embedding_names: List[str] = [] for feature_name in embedding_config.feature_names: if shared_feature[feature_name]: embedding_names.append(feature_name + "@" + embedding_config.name) else: embedding_names.append(feature_name) embedding_names_by_table.append(embedding_names) return embedding_names_by_table
null
8,889
import abc import json from dataclasses import asdict, dataclass from typing import Any, Dict, List, Optional, Tuple, Type import torch import torch.nn as nn import torch.quantization as quant import torchrec as trec import torchrec.quant as trec_quant from torchrec.modules.embedding_configs import QuantConfig from torchrec.modules.embedding_modules import ( EmbeddingBagCollectionInterface, EmbeddingCollectionInterface, ) def trim_torch_package_prefix_from_typename(typename: str) -> str: if typename.startswith("<torch_package_"): # Trim off <torch_package_x> prefix. typename = ".".join(typename.split(".")[1:]) return typename
null
8,890
import abc import json from dataclasses import asdict, dataclass from typing import Any, Dict, List, Optional, Tuple, Type import torch import torch.nn as nn import torch.quantization as quant import torchrec as trec import torchrec.quant as trec_quant from torchrec.modules.embedding_configs import QuantConfig from torchrec.modules.embedding_modules import ( EmbeddingBagCollectionInterface, EmbeddingCollectionInterface, ) class QuantConfig(NamedTuple): def quantize_embeddings( module: nn.Module, dtype: torch.dtype, inplace: bool, additional_qconfig_spec_keys: Optional[List[Type[nn.Module]]] = None, additional_mapping: Optional[Dict[Type[nn.Module], Type[nn.Module]]] = None, output_dtype: torch.dtype = torch.float, per_table_weight_dtype: Optional[Dict[str, torch.dtype]] = None, ) -> nn.Module: qconfig = QuantConfig( activation=quant.PlaceholderObserver.with_args(dtype=output_dtype), weight=quant.PlaceholderObserver.with_args(dtype=dtype), per_table_weight_dtype=per_table_weight_dtype, ) qconfig_spec: Dict[Type[nn.Module], QuantConfig] = { trec.EmbeddingBagCollection: qconfig, } mapping: Dict[Type[nn.Module], Type[nn.Module]] = { trec.EmbeddingBagCollection: trec_quant.EmbeddingBagCollection, } if additional_qconfig_spec_keys is not None: for t in additional_qconfig_spec_keys: qconfig_spec[t] = qconfig if additional_mapping is not None: mapping.update(additional_mapping) return quant.quantize_dynamic( module, qconfig_spec=qconfig_spec, mapping=mapping, inplace=inplace, )
null
8,891
import abc import json from dataclasses import asdict, dataclass from typing import Any, Dict, List, Optional, Tuple, Type import torch import torch.nn as nn import torch.quantization as quant import torchrec as trec import torchrec.quant as trec_quant from torchrec.modules.embedding_configs import QuantConfig from torchrec.modules.embedding_modules import ( EmbeddingBagCollectionInterface, EmbeddingCollectionInterface, ) def quantize_feature( module: torch.nn.Module, inputs: Tuple[torch.Tensor, ...] ) -> Tuple[torch.Tensor, ...]: class PredictModule(nn.Module): def __init__( self, module: nn.Module, ) -> None: def predict_module( self, ) -> nn.Module: def predict_forward(self, batch: Dict[str, torch.Tensor]) -> Any: def forward(self, batch: Dict[str, torch.Tensor]) -> Any: def state_dict( self, destination: Optional[Dict[str, Any]] = None, prefix: str = "", keep_vars: bool = False, ) -> Dict[str, Any]: class EmbeddingBagCollectionInterface(abc.ABC, nn.Module): def forward( self, features: KeyedJaggedTensor, ) -> KeyedTensor: def embedding_bag_configs( self, ) -> List[EmbeddingBagConfig]: def is_weighted(self) -> bool: class EmbeddingCollectionInterface(abc.ABC, nn.Module): def forward( self, features: KeyedJaggedTensor, ) -> Dict[str, JaggedTensor]: def embedding_configs( self, ) -> List[EmbeddingConfig]: def need_indices(self) -> bool: def embedding_dim(self) -> int: def embedding_names_by_table(self) -> List[List[str]]: def quantize_dense( predict_module: PredictModule, dtype: torch.dtype, additional_embedding_module_type: List[Type[nn.Module]] = [], ) -> nn.Module: module = predict_module.predict_module reassign = {} for name, mod in module.named_children(): # both fused modules and observed custom modules are # swapped as one unit if not ( isinstance(mod, EmbeddingBagCollectionInterface) or isinstance(mod, EmbeddingCollectionInterface) or any([type(mod) is clazz for clazz in additional_embedding_module_type]) ): if dtype == torch.half: new_mod = mod.half() new_mod.register_forward_pre_hook(quantize_feature) reassign[name] = new_mod else: raise NotImplementedError( "only fp16 is supported for non-embedding module lowering" ) for key, value in reassign.items(): module._modules[key] = value return predict_module
null
8,892
import abc from pathlib import Path from typing import Any, BinaryIO, Dict, List, Type, TypeVar, Union import torch from torch.package import PackageExporter from torchrec.inference.modules import PredictFactory try: # pyre-fixme[21]: Could not find module `torch_package_importer`. import torch_package_importer # @manual except ImportError: pass def load_config_text(name: str) -> str: return torch_package_importer.load_text("__configs", name)
null
8,893
import abc from pathlib import Path from typing import Any, BinaryIO, Dict, List, Type, TypeVar, Union import torch from torch.package import PackageExporter from torchrec.inference.modules import PredictFactory T = TypeVar("T") try: # pyre-fixme[21]: Could not find module `torch_package_importer`. import torch_package_importer # @manual except ImportError: pass def load_pickle_config(name: str, clazz: Type[T]) -> T: loaded_obj = torch_package_importer.load_pickle("__configs", name) assert isinstance( loaded_obj, clazz ), f"The loaded config {type(loaded_obj)} is not of type {clazz}" return loaded_obj
null
8,894
import argparse import logging import grpc import torch from gen.torchrec.inference import predictor_pb2, predictor_pb2_grpc from torch.utils.data import DataLoader from torchrec.datasets.criteo import DEFAULT_CAT_NAMES, DEFAULT_INT_NAMES from torchrec.datasets.random import RandomRecDataset from torchrec.datasets.utils import Batch DEFAULT_INT_NAMES: List[str] = [f"int_{idx}" for idx in range(INT_FEATURE_COUNT)] DEFAULT_CAT_NAMES: List[str] = [f"cat_{idx}" for idx in range(CAT_FEATURE_COUNT)] class RandomRecDataset(IterableDataset[Batch]): """ Random iterable dataset used to generate batches for recommender systems (RecSys). Currently produces unweighted sparse features only. TODO: Add weighted sparse features. Args: keys (List[str]): List of feature names for sparse features. batch_size (int): batch size. hash_size (Optional[int]): Max sparse id value. All sparse IDs will be taken modulo this value. hash_sizes (Optional[List[int]]): Max sparse id value per feature in keys. Each sparse ID will be taken modulo the corresponding value from this argument. Note, if this is used, hash_size will be ignored. ids_per_feature (int): Number of IDs per sparse feature. ids_per_features (int): Number of IDs per sparse feature in each key. Note, if this is used, ids_per_feature will be ignored. num_dense (int): Number of dense features. manual_seed (int): Seed for deterministic behavior. num_batches: (Optional[int]): Num batches to generate before raising StopIteration num_generated_batches int: Num batches to cache. If num_batches > num_generated batches, then we will cycle to the first generated batch. If this value is negative, batches will be generated on the fly. min_ids_per_feature (int): Minimum number of IDs per features. Example:: dataset = RandomRecDataset( keys=["feat1", "feat2"], batch_size=16, hash_size=100_000, ids_per_feature=1, num_dense=13, ), example = next(iter(dataset)) """ def __init__( self, keys: List[str], batch_size: int, hash_size: Optional[int] = 100, hash_sizes: Optional[List[int]] = None, ids_per_feature: Optional[int] = 2, ids_per_features: Optional[List[int]] = None, num_dense: int = 50, manual_seed: Optional[int] = None, num_batches: Optional[int] = None, num_generated_batches: int = 10, min_ids_per_feature: Optional[int] = None, min_ids_per_features: Optional[List[int]] = None, ) -> None: super().__init__() if hash_sizes is None: hash_size = hash_size or 100 hash_sizes = [hash_size] * len(keys) assert hash_sizes is not None assert len(hash_sizes) == len( keys ), "length of hash_sizes must be equal to the number of keys" if ids_per_features is None: ids_per_feature = ids_per_feature or 2 ids_per_features = [ids_per_feature] * len(keys) assert ids_per_features is not None if min_ids_per_features is None: min_ids_per_feature = ( min_ids_per_feature if min_ids_per_feature is not None else ids_per_feature ) assert min_ids_per_feature is not None min_ids_per_features = [min_ids_per_feature] * len(keys) assert len(ids_per_features) == len( keys ), "length of ids_per_features must be equal to the number of keys" self.batch_generator = _RandomRecBatch( keys=keys, batch_size=batch_size, hash_sizes=hash_sizes, ids_per_features=ids_per_features, num_dense=num_dense, manual_seed=manual_seed, num_batches=None, num_generated_batches=num_generated_batches, min_ids_per_features=min_ids_per_features, ) self.num_batches: int = cast(int, num_batches if not None else sys.maxsize) def __iter__(self) -> Iterator[Batch]: return itertools.islice(iter(self.batch_generator), self.num_batches) def __len__(self) -> int: return self.num_batches class Batch(Pipelineable): dense_features: torch.Tensor sparse_features: KeyedJaggedTensor labels: torch.Tensor def to(self, device: torch.device, non_blocking: bool = False) -> "Batch": return Batch( dense_features=self.dense_features.to( device=device, non_blocking=non_blocking ), sparse_features=self.sparse_features.to( device=device, non_blocking=non_blocking ), labels=self.labels.to(device=device, non_blocking=non_blocking), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self.dense_features.record_stream(stream) self.sparse_features.record_stream(stream) self.labels.record_stream(stream) def pin_memory(self) -> "Batch": return Batch( dense_features=self.dense_features.pin_memory(), sparse_features=self.sparse_features.pin_memory(), labels=self.labels.pin_memory(), ) def create_training_batch(args: argparse.Namespace) -> Batch: return next( iter( DataLoader( RandomRecDataset( keys=DEFAULT_CAT_NAMES, batch_size=args.batch_size, hash_size=args.num_embedding_features, ids_per_feature=1, num_dense=len(DEFAULT_INT_NAMES), ), batch_sampler=None, pin_memory=False, num_workers=0, ) ) )
null
8,895
import argparse import logging import grpc import torch from gen.torchrec.inference import predictor_pb2, predictor_pb2_grpc from torch.utils.data import DataLoader from torchrec.datasets.criteo import DEFAULT_CAT_NAMES, DEFAULT_INT_NAMES from torchrec.datasets.random import RandomRecDataset from torchrec.datasets.utils import Batch class Batch(Pipelineable): dense_features: torch.Tensor sparse_features: KeyedJaggedTensor labels: torch.Tensor def to(self, device: torch.device, non_blocking: bool = False) -> "Batch": return Batch( dense_features=self.dense_features.to( device=device, non_blocking=non_blocking ), sparse_features=self.sparse_features.to( device=device, non_blocking=non_blocking ), labels=self.labels.to(device=device, non_blocking=non_blocking), ) def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self.dense_features.record_stream(stream) self.sparse_features.record_stream(stream) self.labels.record_stream(stream) def pin_memory(self) -> "Batch": return Batch( dense_features=self.dense_features.pin_memory(), sparse_features=self.sparse_features.pin_memory(), labels=self.labels.pin_memory(), ) def create_request( batch: Batch, args: argparse.Namespace ) -> predictor_pb2.PredictionRequest: def to_bytes(tensor: torch.Tensor) -> bytes: return tensor.cpu().numpy().tobytes() float_features = predictor_pb2.FloatFeatures( num_features=args.num_float_features, values=to_bytes(batch.dense_features), ) id_list_features = predictor_pb2.SparseFeatures( num_features=args.num_id_list_features, values=to_bytes(batch.sparse_features.values()), lengths=to_bytes(batch.sparse_features.lengths()), ) id_score_list_features = predictor_pb2.SparseFeatures(num_features=0) embedding_features = predictor_pb2.FloatFeatures(num_features=0) unary_features = predictor_pb2.SparseFeatures(num_features=0) return predictor_pb2.PredictionRequest( batch_size=args.batch_size, float_features=float_features, id_list_features=id_list_features, id_score_list_features=id_score_list_features, embedding_features=embedding_features, unary_features=unary_features, )
null
8,896
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_calibration( calibration_num: torch.Tensor, calibration_denom: torch.Tensor ) -> torch.Tensor: return torch.where( calibration_denom <= 0.0, 0.0, calibration_num / calibration_denom ).double()
null
8,897
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) CALIBRATION_NUM = "calibration_num" CALIBRATION_DENOM = "calibration_denom" def get_calibration_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> Dict[str, torch.Tensor]: return { CALIBRATION_NUM: torch.sum(predictions * weights, dim=-1), CALIBRATION_DENOM: torch.sum(labels * weights, dim=-1), }
null
8,898
from typing import Any, cast, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _validate_model_outputs( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, session_ids: torch.Tensor, ) -> None: assert predictions.shape == labels.shape == weights.shape == session_ids.shape assert ( predictions.dim() == 2 and predictions.shape[0] > 0 and predictions.shape[1] > 0 )
null
8,899
from typing import Any, cast, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) SUM_NDCG = "sum_ndcg" NUM_SESSIONS = "num_sessions" def session_ids_to_lengths( session_ids: torch.Tensor, device: torch.device ) -> torch.Tensor: """ Convert session_ids to lengths tensor. It is used in all session-wise loss computations. Args: session_ids: a tensor of session_ids, e.g., ["1", "2", "2"] device: the device to put session_ids into Returns: session_lengths(torch.Tensor): a tensor of session lengths, e.g., tensor([1, 2]) """ session_lengths: List[int] = [] if len(session_ids) == 0: return torch.zeros(1) length = 1 for i in range(len(session_ids) - 1): if session_ids[i] == session_ids[i + 1]: length += 1 else: session_lengths.append(length) length = 1 session_lengths.append(length) return torch.tensor(session_lengths, dtype=torch.int, device=device) def compute_lambda_ndcg( prediction: torch.Tensor, label: torch.Tensor, weight: torch.Tensor, session_lengths: torch.Tensor, use_exp_gain: bool, ) -> torch.Tensor: """ Compute the sum lambda NDCG loss from a group of sessions. Args: prediction(torch.Tensor): a tensor of predicted scores label(torch.Tensor): a tensor of labels weight(torch.Tensor): a tensor of weights session_lengths(torch.Tensor): a tensor of session lengths converted from session_ids use_exp_gain(bool): whether to use exponential gain or not Returns: sum_loss(torch.Tensor): a tensor of the sum of the ndcg loss """ loss = torch.zeros_like(session_lengths, dtype=torch.double) cur_index = int(0) for i, session_length in enumerate(session_lengths): data_indexes = torch.arange( cur_index, cur_index + int(session_length), dtype=torch.long, device=prediction.device, ) session_loss = compute_lambda_ndcg_by_session( prediction=torch.take(prediction, data_indexes), label=torch.take(label, data_indexes), use_exp_gain=use_exp_gain, ) loss[i] = session_loss * torch.max(torch.take(weight, data_indexes)) cur_index += session_length return torch.sum(loss) def get_ndcg_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, session_ids: torch.Tensor, exponential_gain: bool, ) -> Dict[str, torch.Tensor]: n_tasks = labels.shape[0] sum_ndcg = torch.zeros(n_tasks, dtype=torch.double).to(labels.device) for i in range(n_tasks): session_lengths = session_ids_to_lengths( session_ids=session_ids[i], device=predictions.device ) sum_ndcg[i] = compute_lambda_ndcg( prediction=predictions[i], label=labels[i], weight=weights[i], session_lengths=session_lengths, use_exp_gain=exponential_gain, ) return {SUM_NDCG: sum_ndcg, NUM_SESSIONS: session_lengths.shape[-1]}
null
8,900
from typing import Any, cast, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_ndcg(sum_ndcg: torch.Tensor, num_sessions: torch.Tensor) -> torch.Tensor: return sum_ndcg / num_sessions
null
8,901
from enum import Enum from typing import Optional class MetricNameBase(StrValueMixin, Enum): pass class MetricNamespaceBase(StrValueMixin, Enum): pass class MetricPrefix(StrValueMixin, Enum): DEFAULT = "" LIFETIME = "lifetime_" WINDOW = "window_" The provided code snippet includes necessary dependencies for implementing the `task_wildcard_metrics_pattern` function. Write a Python function `def task_wildcard_metrics_pattern( namespace: MetricNamespaceBase, metric_name: MetricNameBase, metric_prefix: MetricPrefix = MetricPrefix.DEFAULT, ) -> str` to solve the following problem: r"""Get the re (regular expression) pattern to find a set of metrics regardless task names. The motivation to have this API is from the past bugs which tools hard-code the patterns but the naming change, causing some testing issues. Here is the function: def task_wildcard_metrics_pattern( namespace: MetricNamespaceBase, metric_name: MetricNameBase, metric_prefix: MetricPrefix = MetricPrefix.DEFAULT, ) -> str: r"""Get the re (regular expression) pattern to find a set of metrics regardless task names. The motivation to have this API is from the past bugs which tools hard-code the patterns but the naming change, causing some testing issues. """ return rf"{namespace}-.+\|{metric_prefix}{metric_name}"
r"""Get the re (regular expression) pattern to find a set of metrics regardless task names. The motivation to have this API is from the past bugs which tools hard-code the patterns but the naming change, causing some testing issues.
8,902
from enum import Enum from typing import Optional class MetricNameBase(StrValueMixin, Enum): pass class MetricNamespaceBase(StrValueMixin, Enum): pass class MetricPrefix(StrValueMixin, Enum): DEFAULT = "" LIFETIME = "lifetime_" WINDOW = "window_" def compose_metric_namespace( namespace: MetricNamespaceBase, task_name: str, ) -> str: r"""Get the full namespace of a metric based on the input parameters""" return f"{namespace}-{task_name}" def compose_customized_metric_key( namespace: str, metric_name: str, description: Optional[str] = None, ) -> str: r"""Get the metric key. The input are unrestricted (string) namespace and metric_name. This API should only be used by compose_metric_key() and state metrics as the keys of state metrics are unknown. """ return f"{namespace}|{metric_name}{description or ''}" The provided code snippet includes necessary dependencies for implementing the `compose_metric_key` function. Write a Python function `def compose_metric_key( namespace: MetricNamespaceBase, task_name: str, metric_name: MetricNameBase, metric_prefix: MetricPrefix = MetricPrefix.DEFAULT, description: Optional[str] = None, ) -> str` to solve the following problem: r"""Get the metric key based on the input parameters Here is the function: def compose_metric_key( namespace: MetricNamespaceBase, task_name: str, metric_name: MetricNameBase, metric_prefix: MetricPrefix = MetricPrefix.DEFAULT, description: Optional[str] = None, ) -> str: r"""Get the metric key based on the input parameters""" return compose_customized_metric_key( compose_metric_namespace(namespace, task_name), f"{metric_prefix}{metric_name}", description, )
r"""Get the metric key based on the input parameters
8,903
from typing import Any, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _compute_cross_entropy_norm( mean_label: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: mean_label = mean_label.double() mean_label.clamp_(min=eta, max=1 - eta) return -pos_labels * torch.log2(mean_label) - neg_labels * torch.log2( 1.0 - mean_label ) def compute_ne_helper( ce_sum: torch.Tensor, weighted_num_samples: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: mean_label = pos_labels / weighted_num_samples ce_norm = _compute_cross_entropy_norm(mean_label, pos_labels, neg_labels, eta) return ce_sum / ce_norm
null
8,904
from typing import Any, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_logloss( ce_sum: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: # we utilize tensor broadcasting for operations labels_sum = pos_labels + neg_labels labels_sum.clamp_(min=eta) return ce_sum / labels_sum
null
8,905
from typing import Any, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _compute_cross_entropy_norm( mean_label: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: def compute_ne( ce_sum: torch.Tensor, weighted_num_samples: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, num_groups: int, eta: float, ) -> torch.Tensor: # size should be (num_groups) result_ne = torch.zeros(num_groups) for group in range(num_groups): mean_label = pos_labels[group] / weighted_num_samples[group] ce_norm = _compute_cross_entropy_norm( mean_label, pos_labels[group], neg_labels[group], eta ) ne = ce_sum[group] / ce_norm result_ne[group] = ne # ne indexed by group - tensor size (num_groups) return result_ne
null
8,906
from typing import Any, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_cross_entropy( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, eta: float, ) -> torch.Tensor: predictions = predictions.double() predictions.clamp_(min=eta, max=1 - eta) cross_entropy = -weights * labels * torch.log2(predictions) - weights * ( 1.0 - labels ) * torch.log2(1.0 - predictions) return cross_entropy def get_segemented_ne_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, grouping_keys: torch.Tensor, eta: float, num_groups: int, ) -> Dict[str, torch.Tensor]: groups = torch.unique(grouping_keys) cross_entropy, weighted_num_samples, pos_labels, neg_labels = ( torch.zeros(num_groups).to(labels.device), torch.zeros(num_groups).to(labels.device), torch.zeros(num_groups).to(labels.device), torch.zeros(num_groups).to(labels.device), ) for group in groups: group_mask = grouping_keys == group group_labels = labels[group_mask] group_predictions = predictions[group_mask] group_weights = weights[group_mask] ce_sum_group = torch.sum( compute_cross_entropy( labels=group_labels, predictions=group_predictions, weights=group_weights, eta=eta, ), dim=-1, ) weighted_num_samples_group = torch.sum(group_weights, dim=-1) pos_labels_group = torch.sum(group_weights * group_labels, dim=-1) neg_labels_group = torch.sum(group_weights * (1.0 - group_labels), dim=-1) cross_entropy[group] = ce_sum_group.item() weighted_num_samples[group] = weighted_num_samples_group.item() pos_labels[group] = pos_labels_group.item() neg_labels[group] = neg_labels_group.item() # tensor size for each value is (num_groups) return { "cross_entropy_sum": cross_entropy, "weighted_num_samples": weighted_num_samples, "pos_labels": pos_labels, "neg_labels": neg_labels, }
null
8,907
from typing import Any, Dict, List, Optional, Type import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _state_reduction_sum(state: torch.Tensor) -> torch.Tensor: return state.sum(dim=0)
null
8,908
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_multiclass_recall_at_k( tp_at_k: torch.Tensor, total_weights: torch.Tensor, ) -> torch.Tensor: return tp_at_k / torch.unsqueeze(total_weights, dim=-1)
null
8,909
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_true_positives_at_k( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, n_classes: int, ) -> torch.Tensor: """ Compute and return a list of weighted true positives (true predictions) at k. When k = 0, tp is counted when the 1st predicted class matches the label. When k = 1, tp is counted when either the 1st or 2nd predicted class matches the label. Args: predictions (Tensor): Tensor of label predictions with shape of (n_sample, n_class) or (n_task, n_sample, n_class). labels (Tensor): Tensor of ground truth labels with shape of (n_sample, ) or (n_task, n_sample). weights (Tensor): Tensor of weight on each sample, with shape of (n_sample, ) or (n_task, n_sample). n_classes (int): Number of classes. Output: true_positives_list (Tensor): Tensor of true positives with shape of (n_class, ) or (n_task, n_class). Examples: >>> predictions = torch.tensor([[0.9, 0.1, 0, 0, 0], [0.1, 0.2, 0.25, 0.15, 0.3], [0, 1.0, 0, 0, 0], [0, 0, 0.2, 0.7, 0.1]]) >>> labels = torch.tensor([0, 3, 1, 2]) >>> weights = torch.tensor([1, 0.25, 0.5, 0.25]) >>> n_classes = 5 >>> true_positives_list = compute_multiclass_k_sum(predictions, labels, n_classes) >>> true_positives_list tensor([1.5000, 1.7500, 1.7500, 2.0000, 2.0000]) """ ranks = torch.argsort(predictions, dim=-1, descending=True) true_positives = ( torch.zeros(1, device=predictions.device) if predictions.ndim == 2 else torch.zeros(predictions.shape[0], 1, device=predictions.device) ) true_positives_list = torch.tensor([], device=predictions.device) for k in range(n_classes): mask = torch.unsqueeze(labels, dim=-1) == ranks[..., k : k + 1] mask = mask * torch.unsqueeze(weights, dim=-1) true_positives += mask.sum(dim=-2) true_positives_list = torch.cat((true_positives_list, true_positives), dim=-1) return true_positives_list def get_multiclass_recall_states( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, n_classes: int, ) -> Dict[str, torch.Tensor]: true_positives_at_k_sum = compute_true_positives_at_k( predictions, labels, weights, n_classes ) return { "tp_at_k": true_positives_at_k_sum, "total_weights": torch.sum(weights, dim=-1), }
null
8,910
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_accuracy( accuracy_sum: torch.Tensor, weighted_num_samples: torch.Tensor ) -> torch.Tensor: return torch.where( weighted_num_samples == 0.0, 0.0, accuracy_sum / weighted_num_samples ).double()
null
8,911
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_accuracy_sum( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, threshold: float = 0.5, ) -> torch.Tensor: predictions = predictions.double() return torch.sum(weights * ((predictions >= threshold) == labels), dim=-1) def get_accuracy_states( labels: torch.Tensor, predictions: torch.Tensor, weights: Optional[torch.Tensor], threshold: float = 0.5, ) -> Dict[str, torch.Tensor]: if weights is None: weights = torch.ones_like(predictions) return { "accuracy_sum": compute_accuracy_sum(labels, predictions, weights, threshold), "weighted_num_samples": torch.sum(weights, dim=-1), }
null
8,912
import abc import logging import time from typing import Any, Dict, List, Optional, Type, Union import torch import torch.distributed as dist import torch.nn as nn from torch.profiler import record_function from torchrec.metrics.accuracy import AccuracyMetric from torchrec.metrics.auc import AUCMetric from torchrec.metrics.auprc import AUPRCMetric from torchrec.metrics.calibration import CalibrationMetric from torchrec.metrics.ctr import CTRMetric from torchrec.metrics.mae import MAEMetric from torchrec.metrics.metrics_config import ( MetricsConfig, RecMetricEnum, RecMetricEnumBase, RecTaskInfo, StateMetricEnum, ) from torchrec.metrics.metrics_namespace import ( compose_customized_metric_key, compose_metric_namespace, MetricNamespace, ) from torchrec.metrics.model_utils import parse_task_model_outputs from torchrec.metrics.mse import MSEMetric from torchrec.metrics.multiclass_recall import MulticlassRecallMetric from torchrec.metrics.ndcg import NDCGMetric from torchrec.metrics.ne import NEMetric from torchrec.metrics.rauc import RAUCMetric from torchrec.metrics.rec_metric import RecMetric, RecMetricList from torchrec.metrics.recall_session import RecallSessionMetric from torchrec.metrics.scalar import ScalarMetric from torchrec.metrics.segmented_ne import SegmentedNEMetric from torchrec.metrics.throughput import ThroughputMetric from torchrec.metrics.tower_qps import TowerQPSMetric from torchrec.metrics.weighted_avg import WeightedAvgMetric from torchrec.metrics.xauc import XAUCMetric class StateMetric(abc.ABC): """ The interface of state metrics for a component (e.g., optimizer, qat). """ def get_metrics(self) -> Dict[str, MetricValue]: pass class RecMetricModule(nn.Module): r""" For the current recommendation models, we assume there will be three types of metrics, 1.) RecMetric, 2.) Throughput, 3.) StateMetric. RecMetric is a metric that is computed from the model outputs (labels, predictions, weights). Throughput is being a standalone type as its unique characteristic, time-based. StateMetric is a metric that is computed based on a model componenet (e.g., Optimizer) internal logic. Args: batch_size (int): batch size used by this trainer. world_size (int): the number of trainers. rec_tasks (Optional[List[RecTaskInfo]]): the information of the model tasks. rec_metrics (Optional[RecMetricList]): the list of the RecMetrics. throughput_metric (Optional[ThroughputMetric]): the ThroughputMetric. state_metrics (Optional[Dict[str, StateMetric]]): the dict of StateMetrics. compute_interval_steps (int): the intervals between two compute calls in the unit of batch number memory_usage_limit_mb (float): the memory usage limit for OOM check Call Args: Not supported. Returns: Not supported. Example: >>> config = dataclasses.replace( >>> DefaultMetricsConfig, state_metrics=[StateMetricEnum.OPTIMIZERS] >>> ) >>> >>> metricModule = generate_metric_module( >>> metric_class=RecMetricModule, >>> metrics_config=config, >>> batch_size=128, >>> world_size=64, >>> my_rank=0, >>> state_metrics_mapping={StateMetricEnum.OPTIMIZERS: mock_optimizer}, >>> device=torch.device("cpu"), >>> pg=dist.new_group([0]), >>> ) """ batch_size: int world_size: int rec_tasks: List[RecTaskInfo] rec_metrics: RecMetricList throughput_metric: Optional[ThroughputMetric] state_metrics: Dict[str, StateMetric] memory_usage_limit_mb: float memory_usage_mb_avg: float oom_count: int compute_count: int last_compute_time: float # TODO(chienchin): Reorganize the argument to directly accept a MetricsConfig. def __init__( self, batch_size: int, world_size: int, rec_tasks: Optional[List[RecTaskInfo]] = None, rec_metrics: Optional[RecMetricList] = None, throughput_metric: Optional[ThroughputMetric] = None, state_metrics: Optional[Dict[str, StateMetric]] = None, compute_interval_steps: int = 100, min_compute_interval: float = 0.0, max_compute_interval: float = float("inf"), memory_usage_limit_mb: float = 512, ) -> None: super().__init__() self.rec_tasks = rec_tasks if rec_tasks else [] self.rec_metrics = rec_metrics if rec_metrics else RecMetricList([]) self.throughput_metric = throughput_metric self.state_metrics = state_metrics if state_metrics else {} self.trained_batches: int = 0 self.batch_size = batch_size self.world_size = world_size self.memory_usage_limit_mb = memory_usage_limit_mb self.memory_usage_mb_avg = 0.0 self.oom_count = 0 self.compute_count = 0 self.compute_interval_steps = compute_interval_steps self.min_compute_interval = min_compute_interval self.max_compute_interval = max_compute_interval if self.min_compute_interval == 0.0 and self.max_compute_interval == float( "inf" ): self.min_compute_interval = -1.0 self.max_compute_interval = -1.0 else: if self.max_compute_interval <= 0.0: raise ValueError("Max compute interval should not be smaller than 0.0.") if self.min_compute_interval < 0.0: raise ValueError("Min compute interval should not be smaller than 0.0.") self.register_buffer( "_compute_interval_steps", torch.zeros(1, dtype=torch.int32), persistent=False, ) self.last_compute_time = -1.0 def get_memory_usage(self) -> int: r"""Total memory of unique RecMetric tensors in bytes""" total = {} for metric in self.rec_metrics.rec_metrics: total.update(metric.get_memory_usage()) return sum(total.values()) def check_memory_usage(self, compute_count: int) -> None: memory_usage_mb = self.get_memory_usage() / (10**6) if memory_usage_mb > self.memory_usage_limit_mb: self.oom_count += 1 logger.warning( f"MetricModule is using {memory_usage_mb}MB. " f"This is larger than the limit{self.memory_usage_limit_mb}MB. " f"This is the f{self.oom_count}th OOM." ) if ( compute_count > MEMORY_AVG_WARNING_WARMUP and memory_usage_mb > self.memory_usage_mb_avg * ((100 + MEMORY_AVG_WARNING_PERCENTAGE) / 100) ): logger.warning( f"MetricsModule is using more than {MEMORY_AVG_WARNING_PERCENTAGE}% of " f"the average memory usage. Current usage: {memory_usage_mb}MB." ) self.memory_usage_mb_avg = ( self.memory_usage_mb_avg * (compute_count - 1) + memory_usage_mb ) / compute_count def _update_rec_metrics( self, model_out: Dict[str, torch.Tensor], **kwargs: Any ) -> None: r"""the internal update function to parse the model output. Override this function if the implementation cannot support the model output format. """ if self.rec_metrics and self.rec_tasks: labels, predictions, weights, required_inputs = parse_task_model_outputs( self.rec_tasks, model_out, self.get_required_inputs() ) if required_inputs: kwargs["required_inputs"] = required_inputs self.rec_metrics.update( predictions=predictions, labels=labels, weights=weights, **kwargs, ) def update(self, model_out: Dict[str, torch.Tensor], **kwargs: Any) -> None: r"""update() is called per batch, usually right after forward() to update the local states of metrics based on the model_output. Throughput.update() is also called due to the implementation sliding window throughput. """ with record_function("## RecMetricModule:update ##"): self._update_rec_metrics(model_out, **kwargs) if self.throughput_metric: self.throughput_metric.update() self.trained_batches += 1 def _adjust_compute_interval(self) -> None: """ Adjust the compute interval (in batches) based on the first two time elapsed between the first two compute(). """ if self.last_compute_time > 0 and self.min_compute_interval >= 0: now = time.time() interval = now - self.last_compute_time if not (self.max_compute_interval >= interval >= self.min_compute_interval): per_step_time = interval / self.compute_interval_steps assert ( self.max_compute_interval != float("inf") or self.min_compute_interval != 0.0 ), ( "The compute time interval is " f"[{self.max_compute_interval}, {self.min_compute_interval}]. " "Something is not correct of this range. __init__() should have " "captured this earlier." ) if self.max_compute_interval == float("inf"): # The `per_step_time` is not perfectly measured -- each # step training time can vary. Since max_compute_interval # is set to infinite, adding 1.0 to the `min_compute_interval` # increase the chance that the final compute interval is # indeed larger than `min_compute_interval`. self._compute_interval_steps[0] = int( (self.min_compute_interval + 1.0) / per_step_time ) elif self.min_compute_interval == 0.0: # Similar to the above if, subtracting 1.0 from # `max_compute_interval` to compute `_compute_interval_steps` # can increase the chance that the final compute interval # is indeed smaller than `max_compute_interval` offset = 0.0 if self.max_compute_interval <= 1.0 else 1.0 self._compute_interval_steps[0] = int( (self.max_compute_interval - offset) / per_step_time ) else: self._compute_interval_steps[0] = int( (self.max_compute_interval + self.min_compute_interval) / 2 / per_step_time ) dist.all_reduce(self._compute_interval_steps, op=dist.ReduceOp.MAX) self.compute_interval_steps = int(self._compute_interval_steps.item()) self.min_compute_interval = -1.0 self.max_compute_interval = -1.0 self.last_compute_time = time.time() def should_compute(self) -> bool: return self.trained_batches % self.compute_interval_steps == 0 def compute(self) -> Dict[str, MetricValue]: r"""compute() is called when the global metrics are required, usually right before logging the metrics results to the data sink. """ self.compute_count += 1 self.check_memory_usage(self.compute_count) with record_function("## RecMetricModule:compute ##"): ret: Dict[str, MetricValue] = {} if self.rec_metrics: self._adjust_compute_interval() ret.update(self.rec_metrics.compute()) if self.throughput_metric: ret.update(self.throughput_metric.compute()) if self.state_metrics: for namespace, component in self.state_metrics.items(): ret.update( { f"{compose_customized_metric_key(namespace, metric_name)}": metric_value for metric_name, metric_value in component.get_metrics().items() } ) return ret def local_compute(self) -> Dict[str, MetricValue]: r"""local_compute() is called when per-trainer metrics are required. It's can be used for debugging. Currently only rec_metrics is supported. """ ret: Dict[str, MetricValue] = {} if self.rec_metrics: ret.update(self.rec_metrics.local_compute()) return ret def sync(self) -> None: self.rec_metrics.sync() def unsync(self) -> None: self.rec_metrics.unsync() def reset(self) -> None: self.rec_metrics.reset() def get_required_inputs(self) -> Optional[List[str]]: return self.rec_metrics.get_required_inputs() def _generate_rec_metrics( metrics_config: MetricsConfig, world_size: int, my_rank: int, batch_size: int, process_group: Optional[dist.ProcessGroup] = None, ) -> RecMetricList: rec_metrics = [] for metric_enum, metric_def in metrics_config.rec_metrics.items(): kwargs: Dict[str, Any] = {} if metric_def and metric_def.arguments is not None: kwargs = metric_def.arguments rec_tasks: List[RecTaskInfo] = [] if metric_def.rec_tasks and metric_def.rec_task_indices: raise ValueError( "Only one of RecMetricDef.rec_tasks and RecMetricDef.rec_task_indices " "should be specified." ) if metric_def.rec_tasks: rec_tasks = metric_def.rec_tasks elif metric_def.rec_task_indices: rec_tasks = [ metrics_config.rec_tasks[idx] for idx in metric_def.rec_task_indices ] else: raise ValueError( "One of RecMetricDef.rec_tasks and RecMetricDef.rec_task_indices " "should be a non-empty list" ) rec_metrics.append( REC_METRICS_MAPPING[metric_enum]( world_size=world_size, my_rank=my_rank, batch_size=batch_size, tasks=rec_tasks, compute_mode=metrics_config.rec_compute_mode, window_size=metric_def.window_size, fused_update_limit=metrics_config.fused_update_limit, compute_on_all_ranks=metrics_config.compute_on_all_ranks, should_validate_update=metrics_config.should_validate_update, process_group=process_group, **kwargs, ) ) return RecMetricList(rec_metrics) def _generate_state_metrics( metrics_config: MetricsConfig, state_metrics_mapping: Dict[StateMetricEnum, StateMetric], ) -> Dict[str, StateMetric]: state_metrics: Dict[str, StateMetric] = {} for metric_enum in metrics_config.state_metrics: metric_namespace: Optional[MetricNamespace] = ( STATE_METRICS_NAMESPACE_MAPPING.get(metric_enum, None) ) if metric_namespace is None: raise ValueError(f"Unknown StateMetrics {metric_enum}") full_namespace = compose_metric_namespace( metric_namespace, str(metric_namespace) ) state_metrics[full_namespace] = state_metrics_mapping[metric_enum] return state_metrics class StateMetricEnum(StrValueMixin, Enum): OPTIMIZERS = "optimizers" MODEL_CONFIGURATOR = "model_configurator" class MetricsConfig: """The dataclass that lists all the configurations to be used by the MetricModule. Args: rec_tasks (List[RecTaskInfo]): the list of RecTasks that will be shared by all the metrics. rec_metrics (Dict[RecMetricEnum, RecMetricDef]): the confiurations of the RecMetric objects. throughput_metric: (Optional[ThroughputDef]): the configurations of the ThroughputMetric object. rec_compute_mode (RecComputeMode): the computation mode for the RecMetric objects. This will be applied to all the RecMetric objects defined by ``rec_metrics``. fused_update_limit (int): the maximum updates that can be fused. The default is 0 which means no fusion. Setting this field to 1 is logically identical to 0. If this field ii larger than 1, RecMetrics will perform the actual update every ``update()`` calls. state_metrics (List[StateMetricEnum]): indicates what state_metrics will be enabled. compute_interval_steps(int): computing metrics every step can be expsensive. This field is used to specify the computation interval in batch count. `should_compute()` return True if the current trained batch count match the setting. min_compute_interval(float): minimum compute interval in seconds. If this value is set (should be larger than 0), MetricModule will adjust `compute_interval_steps` after the second compute() is called. max_compute_interval(float): maximum compute interval in seconds. If this value is set (should be larger than 0), MetricModule will adjust `compute_interval_steps` after the second compute() is called. compute_on_all_ranks (bool): whether to compute rec metrics on all ranks. If False, only compute on rank 0. should_validate_update (bool): whether to check the inputs of update() and skip update if the inputs are invalid. Invalid inputs include the case where all examples have 0 weights for a batch. """ rec_tasks: List[RecTaskInfo] = field(default_factory=list) rec_metrics: Dict[RecMetricEnum, RecMetricDef] = field(default_factory=dict) throughput_metric: Optional[ThroughputDef] = None rec_compute_mode: RecComputeMode = RecComputeMode.UNFUSED_TASKS_COMPUTATION fused_update_limit: int = 0 state_metrics: List[StateMetricEnum] = field(default_factory=list) compute_interval_steps: int = 100 min_compute_interval: float = 0.0 max_compute_interval: float = float("inf") compute_on_all_ranks: bool = False should_validate_update: bool = False class ThroughputMetric(nn.Module): """ The module to calculate throughput. Throughput is defined as the trained examples across all ranks per second. For example, if the batch size on each rank is 512 and there are 32 ranks, throughput is 512 * 32 / time_to_train_one_step. Args: batch_size (int): batch size for the trainer world_size (int): the number of trainers window_seconds (int): Throughput use time-based window for window_throughput. This argument specify the window size in seconds. warmup_steps (int): the number of warmup batches. No Throughput will be calculated before the warmup batches count reached. Call Args: Not supported. Returns: Not supported. Example:: throughput = ThroughputMetric( batch_size=128, world_size=4, window_seconds=100, warmup_steps=100 ) """ _namespace: MetricNamespace = MetricNamespace.THROUGHPUT _metric_name: MetricName = MetricName.THROUGHPUT _batch_examples: int _window_seconds: int _warmup_steps: int _window_time_lapse_buffer: Deque[float] _window_time_lapse: float _previous_ts: float _lifetime_throughput_key: str _window_throughput_key: str _total_examples_key: str _steps: int def __init__( self, *, batch_size: int, world_size: int, window_seconds: int, warmup_steps: int = 100, ) -> None: super().__init__() if window_seconds < 1: raise ValueError( "window_seconds must be at least 1 to give window throughput " "the minimum time window" ) if warmup_steps < 1: raise ValueError( "warmup_steps must be at least 1 to give throughput a " "reasonable begin time." ) if window_seconds > MAX_WINDOW_TS: logger.warn( f"window_seconds is greater than {MAX_WINDOW_TS}, capping to {MAX_WINDOW_TS} to make sure window_qps is not staled" ) window_seconds = MAX_WINDOW_TS self._batch_examples = batch_size * world_size self._window_seconds = window_seconds self._warmup_steps = warmup_steps self.register_buffer("total_examples", torch.tensor(0, dtype=torch.long)) self.register_buffer("warmup_examples", torch.tensor(0, dtype=torch.long)) self.register_buffer( "time_lapse_after_warmup", torch.tensor(0, dtype=torch.double) ) self._window_time_lapse_buffer = deque(maxlen=MAX_WINDOW_TS) self._window_time_lapse = 0 self._previous_ts = 0 self._lifetime_throughput_key = compose_metric_key( self._namespace, str(self._namespace), self._metric_name, MetricPrefix.LIFETIME, ) self._window_throughput_key = compose_metric_key( self._namespace, str(self._namespace), self._metric_name, MetricPrefix.WINDOW, ) self._total_examples_key = compose_metric_key( self._namespace, str(self._namespace), MetricName.TOTAL_EXAMPLES, ) self._steps = 0 def _check_window(self) -> None: while self._window_time_lapse > self._window_seconds: self._window_time_lapse -= self._window_time_lapse_buffer.popleft() def update(self) -> None: ts = time.monotonic() self._steps += 1 self.total_examples += self._batch_examples if self._steps <= self._warmup_steps: self.warmup_examples += self._batch_examples if self._steps == self._warmup_steps: self._previous_ts = ts else: time_lapse = ts - self._previous_ts self.time_lapse_after_warmup += time_lapse self._window_time_lapse += time_lapse self._window_time_lapse_buffer.append(time_lapse) self._check_window() self._previous_ts = ts def compute(self) -> Dict[str, torch.Tensor]: ret = {self._total_examples_key: self.total_examples} if self._steps > self._warmup_steps and ( not math.isclose(self.time_lapse_after_warmup.item(), 0) ): lifetime_throughput = ( self.total_examples - self.warmup_examples ) / self.time_lapse_after_warmup if not math.isclose(self._window_time_lapse, 0): window_throughput = ( len(self._window_time_lapse_buffer) * self._batch_examples / self._window_time_lapse ) else: window_throughput = 0.0 if not math.isclose(lifetime_throughput.item(), 0): ret.update( { self._lifetime_throughput_key: torch.tensor( lifetime_throughput, dtype=torch.double ), self._window_throughput_key: torch.tensor( window_throughput, dtype=torch.double ), } ) return ret def generate_metric_module( metric_class: Type[RecMetricModule], metrics_config: MetricsConfig, batch_size: int, world_size: int, my_rank: int, state_metrics_mapping: Dict[StateMetricEnum, StateMetric], device: torch.device, process_group: Optional[dist.ProcessGroup] = None, ) -> RecMetricModule: rec_metrics = _generate_rec_metrics( metrics_config, world_size, my_rank, batch_size, process_group ) if metrics_config.throughput_metric: throughput_metric = ThroughputMetric( batch_size=batch_size, world_size=world_size, window_seconds=metrics_config.throughput_metric.window_size, ) else: throughput_metric = None state_metrics = _generate_state_metrics(metrics_config, state_metrics_mapping) metrics = metric_class( batch_size=batch_size, world_size=world_size, rec_tasks=metrics_config.rec_tasks, rec_metrics=rec_metrics, throughput_metric=throughput_metric, state_metrics=state_metrics, compute_interval_steps=metrics_config.compute_interval_steps, min_compute_interval=metrics_config.min_compute_interval, max_compute_interval=metrics_config.max_compute_interval, ) metrics.to(device) return metrics
null
8,913
from functools import partial from typing import Any, cast, Dict, List, Optional, Type import torch import torch.distributed as dist import torch.nn.functional as F from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _compute_auprc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor: sorted_indices = torch.argsort(predictions, descending=True, dim=-1) threshold = torch.index_select(predictions, dim=0, index=sorted_indices) sorted_labels = torch.index_select(labels, dim=0, index=sorted_indices) sorted_weights = torch.index_select(weights, dim=0, index=sorted_indices) mask = F.pad(threshold.diff(dim=0) != 0, [0, 1], value=1.0) num_tp = torch.cumsum(sorted_weights * sorted_labels, dim=0)[mask] num_fp = torch.cumsum(sorted_weights * (1.0 - sorted_labels), dim=0)[mask] precision = (num_tp / (num_tp + num_fp)).flip(0) recall = (num_tp / num_tp[-1]).flip(0) # The last precision and recall values are 1.0 and 0.0 without a corresponding threshold. # This ensures that the graph starts on the y-axis. precision = torch.cat([precision, precision.new_ones(1)]) recall = torch.cat([recall, recall.new_zeros(1)]) # If recalls are NaNs, set NaNs to 1.0s. if torch.isnan(recall[0]): recall = torch.nan_to_num(recall, 1.0) auprc = _riemann_integral(recall, precision) return auprc The provided code snippet includes necessary dependencies for implementing the `compute_auprc` function. Write a Python function `def compute_auprc( n_tasks: int, predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor` to solve the following problem: Computes AUPRC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (torch.Tensor): tensor of size (n_tasks, n_examples). labels (torch.Tensor): tensor of size (n_tasks, n_examples). weights (torch.Tensor): tensor of size (n_tasks, n_examples). Here is the function: def compute_auprc( n_tasks: int, predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor: """ Computes AUPRC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (torch.Tensor): tensor of size (n_tasks, n_examples). labels (torch.Tensor): tensor of size (n_tasks, n_examples). weights (torch.Tensor): tensor of size (n_tasks, n_examples). """ auprcs = [] for predictions_i, labels_i, weights_i in zip(predictions, labels, weights): auprc = _compute_auprc_helper(predictions_i, labels_i, weights_i) auprcs.append(auprc.view(1)) return torch.cat(auprcs)
Computes AUPRC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (torch.Tensor): tensor of size (n_tasks, n_examples). labels (torch.Tensor): tensor of size (n_tasks, n_examples). weights (torch.Tensor): tensor of size (n_tasks, n_examples).
8,914
from functools import partial from typing import Any, cast, Dict, List, Optional, Type import torch import torch.distributed as dist import torch.nn.functional as F from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _compute_auprc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor: sorted_indices = torch.argsort(predictions, descending=True, dim=-1) threshold = torch.index_select(predictions, dim=0, index=sorted_indices) sorted_labels = torch.index_select(labels, dim=0, index=sorted_indices) sorted_weights = torch.index_select(weights, dim=0, index=sorted_indices) mask = F.pad(threshold.diff(dim=0) != 0, [0, 1], value=1.0) num_tp = torch.cumsum(sorted_weights * sorted_labels, dim=0)[mask] num_fp = torch.cumsum(sorted_weights * (1.0 - sorted_labels), dim=0)[mask] precision = (num_tp / (num_tp + num_fp)).flip(0) recall = (num_tp / num_tp[-1]).flip(0) # The last precision and recall values are 1.0 and 0.0 without a corresponding threshold. # This ensures that the graph starts on the y-axis. precision = torch.cat([precision, precision.new_ones(1)]) recall = torch.cat([recall, recall.new_zeros(1)]) # If recalls are NaNs, set NaNs to 1.0s. if torch.isnan(recall[0]): recall = torch.nan_to_num(recall, 1.0) auprc = _riemann_integral(recall, precision) return auprc The provided code snippet includes necessary dependencies for implementing the `compute_auprc_per_group` function. Write a Python function `def compute_auprc_per_group( n_tasks: int, predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, grouping_keys: torch.Tensor, ) -> torch.Tensor` to solve the following problem: Computes AUPRC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (torch.Tensor): tensor of size (n_tasks, n_examples) labels (torch.Tensor): tensor of size (n_tasks, n_examples) weights (torch.Tensor): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUPRCs per group. Here is the function: def compute_auprc_per_group( n_tasks: int, predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, grouping_keys: torch.Tensor, ) -> torch.Tensor: """ Computes AUPRC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (torch.Tensor): tensor of size (n_tasks, n_examples) labels (torch.Tensor): tensor of size (n_tasks, n_examples) weights (torch.Tensor): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUPRCs per group. """ auprcs = [] if grouping_keys.numel() != 0 and grouping_keys[0] == -1: # we added padding as the first elements during init to avoid floating point exception in sync() # removing the paddings to avoid numerical errors. grouping_keys = grouping_keys[1:] predictions = predictions[:, 1:] labels = labels[:, 1:] weights = weights[:, 1:] # get unique group indices group_indices = torch.unique(grouping_keys) for predictions_i, labels_i, weights_i in zip(predictions, labels, weights): # Loop over each group auprc_groups_sum = torch.tensor([0], dtype=torch.float32) for group_idx in group_indices: # get predictions, labels, and weights for this group group_mask = grouping_keys == group_idx grouped_predictions = predictions_i[group_mask] grouped_labels = labels_i[group_mask] grouped_weights = weights_i[group_mask] auprc = _compute_auprc_helper( grouped_predictions, grouped_labels, grouped_weights ) auprc_groups_sum = auprc_groups_sum.to(auprc.device) auprc_groups_sum += auprc.view(1) avg_auprc = ( auprc_groups_sum / len(group_indices) if len(group_indices) > 0 else torch.tensor([0.5], dtype=torch.float32) ) auprcs.append(avg_auprc) return torch.cat(auprcs)
Computes AUPRC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (torch.Tensor): tensor of size (n_tasks, n_examples) labels (torch.Tensor): tensor of size (n_tasks, n_examples) weights (torch.Tensor): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUPRCs per group.
8,915
from functools import partial from typing import Any, cast, Dict, List, Optional, Type import torch import torch.distributed as dist import torch.nn.functional as F from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _state_reduction(state: List[torch.Tensor], dim: int = 1) -> List[torch.Tensor]: return [torch.cat(state, dim=dim)]
null
8,916
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_ctr(ctr_num: torch.Tensor, ctr_denom: torch.Tensor) -> torch.Tensor: return torch.where(ctr_denom == 0.0, 0.0, ctr_num / ctr_denom).double()
null
8,917
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) CTR_NUM = "ctr_num" CTR_DENOM = "ctr_denom" def get_ctr_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> Dict[str, torch.Tensor]: return { CTR_NUM: torch.sum(labels * weights, dim=-1), CTR_DENOM: torch.sum(weights, dim=-1), }
null
8,918
import logging from typing import Any, cast, Dict, List, Optional, Set, Type, Union import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecTaskInfo, SessionMetricDef from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecComputeMode, RecMetric, RecMetricComputation, RecMetricException, ) def _validate_model_outputs( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, sessions: torch.Tensor, ) -> None: # check if tensors are of the same shape assert labels.dim() == 2 assert labels.shape == predictions.shape assert labels.shape == weights.shape assert labels.shape == sessions.shape
null
8,919
import logging from typing import Any, cast, Dict, List, Optional, Set, Type, Union import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecTaskInfo, SessionMetricDef from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecComputeMode, RecMetric, RecMetricComputation, RecMetricException, ) def ranking_within_session( predictions: torch.Tensor, session: torch.Tensor, ) -> torch.Tensor: # rank predictions that belong to the same session # Example: # predictions = [1.0, 0.0, 0.51, 0.8, 1.0, 0.0, 0.51, 0.8, 1.0, 0.0, 0.51, 0.8] # sessions = [1, 1, 1, 1, 1, 1, 1, -1, -1, -1, -1, -1] # return = [0, 5, 3, 2, 1, 6, 4, 1, 0, 4, 3, 2] n_tasks = predictions.size(0) matching_session_id = session.view(-1, n_tasks) == session.view(n_tasks, -1) predictions_relation = predictions.view(-1, n_tasks) >= predictions.view( n_tasks, -1 ) relation_within_session = matching_session_id & predictions_relation rank_within_session = torch.sum(matching_session_id, dim=-1) - torch.sum( relation_within_session, dim=-1 ) return rank_within_session
null
8,920
import logging from typing import Any, cast, Dict, List, Optional, Set, Type, Union import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecTaskInfo, SessionMetricDef from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecComputeMode, RecMetric, RecMetricComputation, RecMetricException, ) def _calc_num_true_pos( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> torch.Tensor: # predictions are expected to be 0 or 1 integers. num_true_pos = torch.sum(weights * labels * (predictions == 1).double(), dim=-1) return num_true_pos
null
8,921
import logging from typing import Any, cast, Dict, List, Optional, Set, Type, Union import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecTaskInfo, SessionMetricDef from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecComputeMode, RecMetric, RecMetricComputation, RecMetricException, ) def _calc_num_false_neg( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> torch.Tensor: # predictions are expected to be 0 or 1 integers. num_false_neg = torch.sum(weights * labels * (predictions == 0).double(), dim=-1) return num_false_neg
null
8,922
import logging from typing import Any, cast, Dict, List, Optional, Set, Type, Union import torch from torch import distributed as dist from torchrec.metrics.metrics_config import RecTaskInfo, SessionMetricDef from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecComputeMode, RecMetric, RecMetricComputation, RecMetricException, ) logger: logging.Logger = logging.getLogger(__name__) def _calc_recall( num_true_pos: torch.Tensor, num_false_neg: torch.Tensor ) -> torch.Tensor: # if num_true_pos + num_false_neg == 0 then we set recall = NaN by default. recall = torch.tensor([float("nan")]) if (num_true_pos + num_false_neg).item() != 0: recall = num_true_pos / (num_true_pos + num_false_neg) else: logger.warning( "Recall = NaN. Likely, it means that there were no positive examples passed to the metric yet." " Please, debug if you expect every batch to include positive examples." ) return recall
null
8,923
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_mae( error_sum: torch.Tensor, weighted_num_samples: torch.Tensor ) -> torch.Tensor: return torch.where( weighted_num_samples == 0.0, 0.0, error_sum / weighted_num_samples ).double()
null
8,924
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_error_sum( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> torch.Tensor: predictions = predictions.double() return torch.sum(weights * torch.abs(labels - predictions), dim=-1) def get_mae_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> Dict[str, torch.Tensor]: return { "error_sum": compute_error_sum(labels, predictions, weights), "weighted_num_samples": torch.sum(weights, dim=-1), }
null
8,925
from typing import Dict, List, Optional, Tuple import torch from torchrec.metrics.rec_metric import RecTaskInfo def is_empty_signals( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, ) -> bool: return ( torch.numel(labels) <= 0 and torch.numel(predictions) <= 0 and torch.numel(weights) <= 0 ) def parse_model_outputs( label_name: str, prediction_name: str, weight_name: str, model_out: Dict[str, torch.Tensor], ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[torch.Tensor]]: labels = model_out[label_name].squeeze() if not prediction_name: assert not weight_name, "weight name must be empty if prediction name is empty" return (labels, None, None) assert isinstance(labels, torch.Tensor) predictions = model_out[prediction_name].squeeze() assert isinstance(predictions, torch.Tensor) weights = model_out[weight_name].squeeze() assert isinstance(weights, torch.Tensor) if not is_empty_signals(labels, predictions, weights): if labels.dim() == predictions.dim(): assert (torch.numel(labels) == torch.numel(predictions)) and ( torch.numel(labels) == torch.numel(weights) ), ( "Expect the same number of elements in labels, predictions, and weights. " f"Instead got {torch.numel(labels)}, {torch.numel(predictions)}, " f"{torch.numel(weights)}" ) else: # For multiclass models, labels.size() = (batch_size), and predictions.size() = (batch_size, number_of_classes) assert torch.numel(labels) == torch.numel(predictions) / predictions.size()[ -1 ] and torch.numel(labels) == torch.numel(weights) # non-empty tensors need to have rank 1 if len(labels.size()) == 0: labels = labels.unsqueeze(0) predictions = predictions.unsqueeze(0) weights = weights.unsqueeze(0) return labels, predictions, weights def parse_required_inputs( model_out: Dict[str, torch.Tensor], required_inputs_list: List[str], ndcg_transform_input: bool = False, ) -> Dict[str, torch.Tensor]: required_inputs: Dict[str, torch.Tensor] = {} for feature in required_inputs_list: # convert feature defined from config only if ndcg_transform_input: model_out[feature] = ( # pyre-ignore[6] session_ids_to_tensor(model_out[feature]) if isinstance(model_out[feature], list) else model_out[feature] ) required_inputs[feature] = model_out[feature].squeeze() assert isinstance(required_inputs[feature], torch.Tensor) return required_inputs def parse_task_model_outputs( tasks: List[RecTaskInfo], model_out: Dict[str, torch.Tensor], required_inputs_list: Optional[List[str]] = None, ) -> Tuple[ Dict[str, torch.Tensor], Dict[str, torch.Tensor], Dict[str, torch.Tensor], Dict[str, torch.Tensor], ]: all_labels: Dict[str, torch.Tensor] = {} all_predictions: Dict[str, torch.Tensor] = {} all_weights: Dict[str, torch.Tensor] = {} all_required_inputs: Dict[str, torch.Tensor] = {} # Convert session_ids to tensor if NDCG metric ndcg_transform_input = False for task in tasks: labels, predictions, weights = parse_model_outputs( task.label_name, task.prediction_name, task.weight_name, model_out ) if predictions is not None and weights is not None: if not is_empty_signals(labels, predictions, weights): all_labels[task.name] = labels all_predictions[task.name] = predictions all_weights[task.name] = weights else: if torch.numel(labels) > 0: all_labels[task.name] = labels if task.name and task.name.startswith("ndcg"): ndcg_transform_input = True if required_inputs_list is not None: all_required_inputs = parse_required_inputs( model_out, required_inputs_list, ndcg_transform_input ) return all_labels, all_predictions, all_weights, all_required_inputs
null
8,926
import time from typing import Any, cast, Dict, List, Optional, Type import torch import torch.distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, RecModelOutput, ) def _compute_tower_qps( num_examples: torch.Tensor, time_lapse: torch.Tensor ) -> torch.Tensor: return torch.where(time_lapse <= 0.0, 0.0, num_examples / time_lapse).double()
null
8,927
import time from typing import Any, cast, Dict, List, Optional, Type import torch import torch.distributed as dist from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, RecModelOutput, ) def _max_reduction(state: torch.Tensor) -> torch.Tensor: return torch.max(state, dim=0).values
null
8,928
from functools import partial from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type import torch import torch.distributed as dist from torchmetrics.utilities.distributed import gather_all_tensors from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _concat_if_needed( predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: """ This check exists because of how the state is organized due to quirks in RecMetrics. Since we do not do tensor concatenatation in the compute or update call, there are cases (in non-distributed settings) where the tensors from updates are not concatted into a single tensor. Which is determined by the length of the list. """ preds_t, labels_t, weights_t = None, None, None if len(predictions) > 1: preds_t = torch.cat(predictions, dim=-1) labels_t = torch.cat(labels, dim=-1) weights_t = torch.cat(weights, dim=-1) else: preds_t = predictions[0] labels_t = labels[0] weights_t = weights[0] return preds_t, labels_t, weights_t def _compute_auc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, apply_bin: bool = False, ) -> torch.Tensor: sorted_indices = torch.argsort(predictions, descending=True, dim=-1) sorted_labels = torch.index_select(labels, dim=0, index=sorted_indices) if apply_bin: # TODO - [add flag to set bining dyamically] for use with soft labels, >=0.039 --> 1, <0.039 --> 0 sorted_labels = torch.ge(sorted_labels, 0.039).to(dtype=sorted_labels.dtype) sorted_weights = torch.index_select(weights, dim=0, index=sorted_indices) cum_fp = torch.cumsum(sorted_weights * (1.0 - sorted_labels), dim=0) cum_tp = torch.cumsum(sorted_weights * sorted_labels, dim=0) auc = torch.where( cum_fp[-1] * cum_tp[-1] == 0, 0.5, # 0.5 is the no-signal default value for auc. torch.trapz(cum_tp, cum_fp) / cum_fp[-1] / cum_tp[-1], ) return auc The provided code snippet includes necessary dependencies for implementing the `compute_auc` function. Write a Python function `def compute_auc( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], apply_bin: bool = False, ) -> torch.Tensor` to solve the following problem: Computes AUC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples). Here is the function: def compute_auc( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], apply_bin: bool = False, ) -> torch.Tensor: """ Computes AUC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples). """ preds_t, labels_t, weights_t = _concat_if_needed(predictions, labels, weights) aucs = [] for predictions_i, labels_i, weights_i in zip(preds_t, labels_t, weights_t): auc = _compute_auc_helper(predictions_i, labels_i, weights_i, apply_bin) aucs.append(auc.view(1)) return torch.cat(aucs)
Computes AUC (Area Under the Curve) for binary classification. Args: n_tasks (int): number of tasks. predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples).
8,929
from functools import partial from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type import torch import torch.distributed as dist from torchmetrics.utilities.distributed import gather_all_tensors from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _concat_if_needed( predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: """ This check exists because of how the state is organized due to quirks in RecMetrics. Since we do not do tensor concatenatation in the compute or update call, there are cases (in non-distributed settings) where the tensors from updates are not concatted into a single tensor. Which is determined by the length of the list. """ preds_t, labels_t, weights_t = None, None, None if len(predictions) > 1: preds_t = torch.cat(predictions, dim=-1) labels_t = torch.cat(labels, dim=-1) weights_t = torch.cat(weights, dim=-1) else: preds_t = predictions[0] labels_t = labels[0] weights_t = weights[0] return preds_t, labels_t, weights_t def _compute_auc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, apply_bin: bool = False, ) -> torch.Tensor: sorted_indices = torch.argsort(predictions, descending=True, dim=-1) sorted_labels = torch.index_select(labels, dim=0, index=sorted_indices) if apply_bin: # TODO - [add flag to set bining dyamically] for use with soft labels, >=0.039 --> 1, <0.039 --> 0 sorted_labels = torch.ge(sorted_labels, 0.039).to(dtype=sorted_labels.dtype) sorted_weights = torch.index_select(weights, dim=0, index=sorted_indices) cum_fp = torch.cumsum(sorted_weights * (1.0 - sorted_labels), dim=0) cum_tp = torch.cumsum(sorted_weights * sorted_labels, dim=0) auc = torch.where( cum_fp[-1] * cum_tp[-1] == 0, 0.5, # 0.5 is the no-signal default value for auc. torch.trapz(cum_tp, cum_fp) / cum_fp[-1] / cum_tp[-1], ) return auc The provided code snippet includes necessary dependencies for implementing the `compute_auc_per_group` function. Write a Python function `def compute_auc_per_group( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], grouping_keys: torch.Tensor, ) -> torch.Tensor` to solve the following problem: Computes AUC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUCs per group. Here is the function: def compute_auc_per_group( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], grouping_keys: torch.Tensor, ) -> torch.Tensor: """ Computes AUC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUCs per group. """ preds_t, labels_t, weights_t = _concat_if_needed(predictions, labels, weights) aucs = [] if grouping_keys.numel() != 0 and grouping_keys[0] == -1: # we added padding as the first elements during init to avoid floating point exception in sync() # removing the paddings to avoid numerical errors. grouping_keys = grouping_keys[1:] # get unique group indices group_indices = torch.unique(grouping_keys) for predictions_i, labels_i, weights_i in zip(preds_t, labels_t, weights_t): # Loop over each group auc_groups_sum = torch.tensor([0], dtype=torch.float32) for group_idx in group_indices: # get predictions, labels, and weights for this group group_mask = grouping_keys == group_idx grouped_predictions = predictions_i[group_mask] grouped_labels = labels_i[group_mask] grouped_weights = weights_i[group_mask] auc = _compute_auc_helper( grouped_predictions, grouped_labels, grouped_weights ) auc_groups_sum = auc_groups_sum.to(auc.device) auc_groups_sum += auc.view(1) avg_auc = ( auc_groups_sum / len(group_indices) if len(group_indices) > 0 else torch.tensor([0.5], dtype=torch.float32) ) aucs.append(avg_auc) return torch.cat(aucs)
Computes AUC (Area Under the Curve) for binary classification for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of AUCs per group.
8,930
from functools import partial from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type import torch import torch.distributed as dist from torchmetrics.utilities.distributed import gather_all_tensors from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _state_reduction(state: List[torch.Tensor], dim: int = 1) -> List[torch.Tensor]: return [torch.cat(state, dim=dim)]
null
8,931
from functools import partial from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type import torch import torch.distributed as dist from torchmetrics.utilities.distributed import gather_all_tensors from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _concat_if_needed( predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: """ This check exists because of how the state is organized due to quirks in RecMetrics. Since we do not do tensor concatenatation in the compute or update call, there are cases (in non-distributed settings) where the tensors from updates are not concatted into a single tensor. Which is determined by the length of the list. """ preds_t, labels_t, weights_t = None, None, None if len(predictions) > 1: preds_t = torch.cat(predictions, dim=-1) labels_t = torch.cat(labels, dim=-1) weights_t = torch.cat(weights, dim=-1) else: preds_t = predictions[0] labels_t = labels[0] weights_t = weights[0] return preds_t, labels_t, weights_t def _compute_rauc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor: array = [ x for x, _ in sorted( zip(labels.tolist(), predictions.tolist()), key=lambda x: (x[1], x[0]) ) ] return torch.tensor(1 - count_reverse_pairs_divide_and_conquer(array)) The provided code snippet includes necessary dependencies for implementing the `compute_rauc` function. Write a Python function `def compute_rauc( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> torch.Tensor` to solve the following problem: Computes RAUC (Regression AUC) for regression tasks. Args: predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples). Here is the function: def compute_rauc( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> torch.Tensor: """ Computes RAUC (Regression AUC) for regression tasks. Args: predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples). """ preds_t, labels_t, weights_t = _concat_if_needed(predictions, labels, weights) raucs = [] for predictions_i, labels_i, weights_i in zip(preds_t, labels_t, weights_t): rauc = _compute_rauc_helper(predictions_i, labels_i, weights_i) raucs.append(rauc.view(1)) return torch.cat(raucs)
Computes RAUC (Regression AUC) for regression tasks. Args: predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples). labels (List[torch.Tensor]): tensor of size (n_tasks, n_examples). weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples).
8,932
from functools import partial from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type import torch import torch.distributed as dist from torchmetrics.utilities.distributed import gather_all_tensors from torchrec.metrics.metrics_config import RecComputeMode, RecTaskInfo from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _concat_if_needed( predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: """ This check exists because of how the state is organized due to quirks in RecMetrics. Since we do not do tensor concatenatation in the compute or update call, there are cases (in non-distributed settings) where the tensors from updates are not concatted into a single tensor. Which is determined by the length of the list. """ preds_t, labels_t, weights_t = None, None, None if len(predictions) > 1: preds_t = torch.cat(predictions, dim=-1) labels_t = torch.cat(labels, dim=-1) weights_t = torch.cat(weights, dim=-1) else: preds_t = predictions[0] labels_t = labels[0] weights_t = weights[0] return preds_t, labels_t, weights_t def _compute_rauc_helper( predictions: torch.Tensor, labels: torch.Tensor, weights: torch.Tensor, ) -> torch.Tensor: array = [ x for x, _ in sorted( zip(labels.tolist(), predictions.tolist()), key=lambda x: (x[1], x[0]) ) ] return torch.tensor(1 - count_reverse_pairs_divide_and_conquer(array)) The provided code snippet includes necessary dependencies for implementing the `compute_rauc_per_group` function. Write a Python function `def compute_rauc_per_group( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], grouping_keys: torch.Tensor, ) -> torch.Tensor` to solve the following problem: Computes RAUC (Regression AUC) for regression tasks for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of RAUCs per group. Here is the function: def compute_rauc_per_group( n_tasks: int, predictions: List[torch.Tensor], labels: List[torch.Tensor], weights: List[torch.Tensor], grouping_keys: torch.Tensor, ) -> torch.Tensor: """ Computes RAUC (Regression AUC) for regression tasks for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of RAUCs per group. """ preds_t, labels_t, weights_t = _concat_if_needed(predictions, labels, weights) raucs = [] if grouping_keys.numel() != 0 and grouping_keys[0] == -1: # we added padding as the first elements during init to avoid floating point exception in sync() # removing the paddings to avoid numerical errors. grouping_keys = grouping_keys[1:] # get unique group indices group_indices = torch.unique(grouping_keys) for predictions_i, labels_i, weights_i in zip(preds_t, labels_t, weights_t): # Loop over each group rauc_groups_sum = torch.tensor([0], dtype=torch.float32) for group_idx in group_indices: # get predictions, labels, and weights for this group group_mask = grouping_keys == group_idx grouped_predictions = predictions_i[group_mask] grouped_labels = labels_i[group_mask] grouped_weights = weights_i[group_mask] rauc = _compute_rauc_helper( grouped_predictions, grouped_labels, grouped_weights ) rauc_groups_sum = rauc_groups_sum.to(rauc.device) rauc_groups_sum += rauc.view(1) avg_rauc = ( rauc_groups_sum / len(group_indices) if len(group_indices) > 0 else torch.tensor([0.5], dtype=torch.float32) ) raucs.append(avg_rauc) return torch.cat(raucs)
Computes RAUC (Regression AUC) for regression tasks for groups of predictions/labels. Args: n_tasks (int): number of tasks predictions (List[torch.Tensor]): tensor of size (n_tasks, n_examples) labels (List[torch.Tensor]: tensor of size (n_tasks, n_examples) weights (List[torch.Tensor]): tensor of size (n_tasks, n_examples) grouping_keys (torch.Tensor): tensor of size (n_examples,) Returns: torch.Tensor: tensor of size (n_tasks,), average of RAUCs per group.
8,934
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_mse( error_sum: torch.Tensor, weighted_num_samples: torch.Tensor ) -> torch.Tensor: return torch.where( weighted_num_samples == 0.0, 0.0, error_sum / weighted_num_samples ).double()
null
8,935
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_rmse( error_sum: torch.Tensor, weighted_num_samples: torch.Tensor ) -> torch.Tensor: return torch.where( weighted_num_samples == 0.0, 0.0, torch.sqrt(error_sum / weighted_num_samples) ).double()
null
8,936
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_error_sum( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> torch.Tensor: predictions = predictions.double() return torch.sum(weights * torch.square(labels - predictions), dim=-1) def get_mse_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> Dict[str, torch.Tensor]: return { "error_sum": compute_error_sum(labels, predictions, weights), "weighted_num_samples": torch.sum(weights, dim=-1), }
null
8,937
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def _compute_cross_entropy_norm( mean_label: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: mean_label = mean_label.double() mean_label.clamp_(min=eta, max=1 - eta) return -pos_labels * torch.log2(mean_label) - neg_labels * torch.log2( 1.0 - mean_label ) def compute_ne( ce_sum: torch.Tensor, weighted_num_samples: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, allow_missing_label_with_zero_weight: bool = False, ) -> torch.Tensor: if allow_missing_label_with_zero_weight and not weighted_num_samples.all(): # If nan were to occur, return a dummy value instead of nan if # allow_missing_label_with_zero_weight is True return torch.tensor([eta]) # Goes into this block if all elements in weighted_num_samples > 0 mean_label = pos_labels / weighted_num_samples ce_norm = _compute_cross_entropy_norm(mean_label, pos_labels, neg_labels, eta) return ce_sum / ce_norm
null
8,938
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_logloss( ce_sum: torch.Tensor, pos_labels: torch.Tensor, neg_labels: torch.Tensor, eta: float, ) -> torch.Tensor: labels_sum = pos_labels + neg_labels labels_sum.clamp_(min=eta) return ce_sum / labels_sum
null
8,939
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_cross_entropy( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, eta: float, ) -> torch.Tensor: predictions = predictions.double() predictions.clamp_(min=eta, max=1 - eta) cross_entropy = -weights * labels * torch.log2(predictions) - weights * ( 1.0 - labels ) * torch.log2(1.0 - predictions) return cross_entropy def get_ne_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor, eta: float ) -> Dict[str, torch.Tensor]: cross_entropy = compute_cross_entropy( labels, predictions, weights, eta, ) return { "cross_entropy_sum": torch.sum(cross_entropy, dim=-1), "weighted_num_samples": torch.sum(weights, dim=-1), "pos_labels": torch.sum(weights * labels, dim=-1), "neg_labels": torch.sum(weights * (1.0 - labels), dim=-1), }
null
8,940
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_xauc( error_sum: torch.Tensor, weighted_num_pairs: torch.Tensor ) -> torch.Tensor: return torch.where( weighted_num_pairs == 0.0, 0.0, error_sum / weighted_num_pairs ).double()
null
8,941
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, RecMetricException, ) def compute_error_sum( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> torch.Tensor: predictions = predictions.double() errors = [] for predictions_i, labels_i, weights_i in zip(predictions, labels, weights): preds_x, preds_y = torch.meshgrid(predictions_i, predictions_i) labels_x, labels_y = torch.meshgrid(labels_i, labels_i) weights_x, weights_y = torch.meshgrid(weights_i, weights_i) weights_flag = weights_x * weights_y match = torch.logical_or( torch.logical_and(preds_x > preds_y, labels_x > labels_y), torch.logical_and(preds_x < preds_y, labels_x < labels_y), ) match = ( weights_flag * torch.logical_or( match, torch.logical_and(preds_x == preds_y, labels_x == labels_y) ).double() ) errors.append(torch.sum(torch.triu(match, diagonal=1)).view(1)) return torch.cat(errors) def compute_weighted_num_pairs(weights: torch.Tensor) -> torch.Tensor: num_pairs = [] for weight_i in weights: weights_x, weights_y = torch.meshgrid(weight_i, weight_i) weights_flag = weights_x * weights_y num_pairs.append(torch.sum(torch.triu(weights_flag, diagonal=1)).view(1)) return torch.cat(num_pairs) def get_xauc_states( labels: torch.Tensor, predictions: torch.Tensor, weights: torch.Tensor ) -> Dict[str, torch.Tensor]: return { "error_sum": compute_error_sum(labels, predictions, weights), "weighted_num_pairs": compute_weighted_num_pairs(weights), }
null
8,942
from typing import Any, cast, Dict, List, Optional, Type import torch from torchrec.metrics.metrics_namespace import MetricName, MetricNamespace, MetricPrefix from torchrec.metrics.rec_metric import ( MetricComputationReport, RecMetric, RecMetricComputation, ) def get_mean(value_sum: torch.Tensor, num_samples: torch.Tensor) -> torch.Tensor: return value_sum / num_samples
null
8,943
from typing import Optional, Union import torch from torch import nn from torchrec.distributed.model_parallel import DistributedModelParallel from torchrec.distributed.quant_embeddingbag import ShardedQuantEmbeddingBagCollection from torchrec.quant.embedding_modules import ( EmbeddingBagCollection as QuantEmbeddingBagCollection, EmbeddingCollection as QuantEmbeddingCollection, ) def populate_fx_names( quant_ebc: Union[QuantEmbeddingBagCollection, ShardedQuantEmbeddingBagCollection] ) -> None: """ Assigns fx path to non registered lookup modules. This allows the Torchrec tracer to fallback to emb_module._fx_path for table batched embeddings. """ if isinstance(quant_ebc, QuantEmbeddingBagCollection): for emb_configs, emb_module in zip( quant_ebc._key_to_tables, quant_ebc._emb_modules ): table_names = [] for config in emb_configs: table_names.append(config.name) joined_table_names = ",".join(table_names) emb_module._fx_path = f"emb_module.{joined_table_names}" elif isinstance(quant_ebc, ShardedQuantEmbeddingBagCollection): for i, (emb_module, emb_dist_module) in enumerate( zip(quant_ebc._lookups, quant_ebc._output_dists) ): embedding_fx_path = f"embedding_lookup.sharding_{i}" emb_module._fx_path = embedding_fx_path emb_dist_module._fx_path = f"embedding_dist.{i}" for rank, rank_module in enumerate(emb_module._embedding_lookups_per_rank): rank_fx_path = f"{embedding_fx_path}.rank_{rank}" rank_module._fx_path = rank_fx_path for group, group_module in enumerate(rank_module._emb_modules): group_module._fx_path = f"{rank_fx_path}.group_{group}" group_module._emb_module._fx_path = ( f"{rank_fx_path}.group_{group}.tbe" ) class ShardedQuantEmbeddingBagCollection( ShardedQuantEmbeddingModuleState[ ListOfKJTList, List[List[torch.Tensor]], KeyedTensor, NullShardedModuleContext, ], ): """ Sharded implementation of `EmbeddingBagCollection`. This is part of the public API to allow for manual data dist pipelining. """ def __init__( self, module: EmbeddingBagCollectionInterface, table_name_to_parameter_sharding: Dict[str, ParameterSharding], env: ShardingEnv, fused_params: Optional[Dict[str, Any]] = None, device: Optional[torch.device] = None, ) -> None: super().__init__() self._embedding_bag_configs: List[EmbeddingBagConfig] = ( module.embedding_bag_configs() ) self._sharding_type_to_sharding_infos: Dict[ str, List[EmbeddingShardingInfo] ] = create_sharding_infos_by_sharding( module, table_name_to_parameter_sharding, "embedding_bags.", fused_params ) self._sharding_type_to_sharding: Dict[ str, EmbeddingSharding[ NullShardingContext, KJTList, List[torch.Tensor], torch.Tensor, ], ] = { sharding_type: create_infer_embedding_bag_sharding( sharding_type, embedding_confings, env ) for sharding_type, embedding_confings in self._sharding_type_to_sharding_infos.items() } self._device = device self._is_weighted: bool = module.is_weighted() self._input_dists: List[nn.Module] = [] self._lookups: List[nn.Module] = [] self._create_lookups(fused_params, device) # Ensure output dist is set for post processing from an inference runtime (ie. setting device from runtime). self._output_dists: torch.nn.ModuleList = torch.nn.ModuleList() self._embedding_names: List[str] = [] self._embedding_dims: List[int] = [] self._feature_splits: List[int] = [] self._features_order: List[int] = [] # forward pass flow control self._has_uninitialized_input_dist: bool = True self._has_uninitialized_output_dist: bool = True self._has_features_permute: bool = True tbes: Dict[IntNBitTableBatchedEmbeddingBagsCodegen, GroupedEmbeddingConfig] = ( get_tbes_to_register_from_iterable(self._lookups) ) self._tbes_configs: Dict[ IntNBitTableBatchedEmbeddingBagsCodegen, GroupedEmbeddingConfig ] = tbes # Optional registration of TBEs for model post processing utilities if is_fused_param_register_tbe(fused_params): self.tbes: torch.nn.ModuleList = torch.nn.ModuleList(tbes.keys()) quant_state_dict_split_scale_bias = ( is_fused_param_quant_state_dict_split_scale_bias(fused_params) ) if quant_state_dict_split_scale_bias: self._initialize_torch_state( tbes=tbes, table_name_to_parameter_sharding=table_name_to_parameter_sharding, tables_weights_prefix="embedding_bags", ) else: table_wise_sharded_only: bool = all( sharding_type == ShardingType.TABLE_WISE.value for sharding_type in self._sharding_type_to_sharding.keys() ) assert ( table_wise_sharded_only ), "ROW_WISE,COLUMN_WISE shardings can be used only in 'quant_state_dict_split_scale_bias' mode, specify fused_params[FUSED_PARAM_QUANT_STATE_DICT_SPLIT_SCALE_BIAS]=True to __init__ argument" self.embedding_bags: nn.ModuleDict = nn.ModuleDict() for table in self._embedding_bag_configs: self.embedding_bags[table.name] = torch.nn.Module() for _sharding_type, lookup in zip( self._sharding_type_to_sharding.keys(), self._lookups ): lookup_state_dict = lookup.state_dict() for key in lookup_state_dict: if key.endswith(".weight"): table_name = key[: -len(".weight")] self.embedding_bags[table_name].register_buffer( "weight", lookup_state_dict[key] ) def tbes_configs( self, ) -> Dict[IntNBitTableBatchedEmbeddingBagsCodegen, GroupedEmbeddingConfig]: return self._tbes_configs def sharding_type_to_sharding_infos(self) -> Dict[str, List[EmbeddingShardingInfo]]: return self._sharding_type_to_sharding_infos def embedding_bag_configs(self) -> List[EmbeddingBagConfig]: return self._embedding_bag_configs def _create_input_dist( self, input_feature_names: List[str], features_device: torch.device, input_dist_device: Optional[torch.device] = None, ) -> None: feature_names: List[str] = [] for sharding in self._sharding_type_to_sharding.values(): self._input_dists.append( sharding.create_input_dist(device=input_dist_device) ) feature_names.extend(sharding.feature_names()) self._feature_splits.append(len(sharding.feature_names())) if feature_names == input_feature_names: self._has_features_permute = False else: for f in feature_names: self._features_order.append(input_feature_names.index(f)) self.register_buffer( "_features_order_tensor", torch.tensor( self._features_order, device=features_device, dtype=torch.int32 ), persistent=False, ) def _create_lookups( self, fused_params: Optional[Dict[str, Any]], device: Optional[torch.device] = None, ) -> None: for sharding in self._sharding_type_to_sharding.values(): self._lookups.append( sharding.create_lookup( device=device, fused_params=fused_params, ) ) def _create_output_dist(self, device: Optional[torch.device] = None) -> None: for sharding in self._sharding_type_to_sharding.values(): self._output_dists.append(sharding.create_output_dist(device)) self._embedding_names.extend(sharding.embedding_names()) self._embedding_dims.extend(sharding.embedding_dims()) # pyre-ignore [14] # pyre-ignore def input_dist( self, ctx: NullShardedModuleContext, features: KeyedJaggedTensor ) -> ListOfKJTList: if self._has_uninitialized_input_dist: self._create_input_dist( features.keys(), features.device(), self._device, ) self._has_uninitialized_input_dist = False if self._has_uninitialized_output_dist: self._create_output_dist(features.device()) self._has_uninitialized_output_dist = False with torch.no_grad(): if self._has_features_permute: features = features.permute( self._features_order, self._features_order_tensor, ) else: features = flatten_feature_lengths(features) features_by_shards = ( [features] if len(self._feature_splits) == 1 else features.split(self._feature_splits) ) return ListOfKJTList( [ self._input_dists[i].forward(features_by_shards[i]) for i in range(len(self._input_dists)) ] ) def compute( self, ctx: NullShardedModuleContext, dist_input: ListOfKJTList, ) -> List[List[torch.Tensor]]: # syntax for torchscript return [lookup.forward(dist_input[i]) for i, lookup in enumerate(self._lookups)] # pyre-ignore def output_dist( self, ctx: NullShardedModuleContext, output: List[List[torch.Tensor]], ) -> KeyedTensor: return construct_output_kt( embeddings=[ dist.forward(output[i]) for i, dist in enumerate(self._output_dists) ], embedding_dims=self._embedding_dims, embedding_names=self._embedding_names, ) # pyre-ignore def compute_and_output_dist( self, ctx: NullShardedModuleContext, input: ListOfKJTList ) -> KeyedTensor: return self.output_dist(ctx, self.compute(ctx, input)) # pyre-ignore def forward(self, *input, **kwargs) -> KeyedTensor: ctx = self.create_context() dist_input = self.input_dist(ctx, *input, **kwargs) return self.compute_and_output_dist(ctx, dist_input) def copy(self, device: torch.device) -> nn.Module: if self._has_uninitialized_output_dist: self._create_output_dist(device) self._has_uninitialized_output_dist = False return super().copy(device) def shardings(self) -> Dict[str, FeatureShardingMixIn]: # pyre-ignore [7] return self._sharding_type_to_sharding def create_context(self) -> NullShardedModuleContext: if is_torchdynamo_compiling(): # Context creation is not supported by dynamo yet. # Context is not needed for TW sharding => # Unblocking dynamo TW with None. # pyre-ignore return None return NullShardedModuleContext() def recursive_populate_fx_names(module: nn.Module) -> None: if isinstance(module, QuantEmbeddingBagCollection) or isinstance( module, ShardedQuantEmbeddingBagCollection ): populate_fx_names(module) return for submodule in module.children(): recursive_populate_fx_names(submodule)
null
8,944
from typing import Optional, Union import torch from torch import nn from torchrec.distributed.model_parallel import DistributedModelParallel from torchrec.distributed.quant_embeddingbag import ShardedQuantEmbeddingBagCollection from torchrec.quant.embedding_modules import ( EmbeddingBagCollection as QuantEmbeddingBagCollection, EmbeddingCollection as QuantEmbeddingCollection, ) def _meta_to_cpu_placement( module: nn.Module, root_module: nn.Module, name: Optional[str] = None ) -> None: if ( name is not None and isinstance(module, QuantEmbeddingBagCollection) and module.device.type == "meta" ): qebc_cpu = QuantEmbeddingBagCollection( tables=module.embedding_bag_configs(), is_weighted=module.is_weighted(), device=torch.device("cpu"), output_dtype=module.output_dtype(), register_tbes=module.register_tbes, row_alignment=module.row_alignment, ) setattr(root_module, name, qebc_cpu) elif ( name is not None and isinstance(module, QuantEmbeddingCollection) and module.device.type == "meta" ): qec_cpu = QuantEmbeddingCollection( tables=module.embedding_configs(), device=torch.device("cpu"), need_indices=module.need_indices(), output_dtype=module.output_dtype(), register_tbes=module.register_tbes, row_alignment=module.row_alignment, ) setattr(root_module, name, qec_cpu) else: for name, submodule in module.named_children(): _meta_to_cpu_placement(submodule, module, name) class DistributedModelParallel(nn.Module, FusedOptimizerModule): """ Entry point to model parallelism. Args: module (nn.Module): module to wrap. env (Optional[ShardingEnv]): sharding environment that has the process group. device (Optional[torch.device]): compute device, defaults to cpu. plan (Optional[ShardingPlan]): plan to use when sharding, defaults to `EmbeddingShardingPlanner.collective_plan()`. sharders (Optional[List[ModuleSharder[nn.Module]]]): `ModuleSharders` available to shard with, defaults to `EmbeddingBagCollectionSharder()`. init_data_parallel (bool): data-parallel modules can be lazy, i.e. they delay parameter initialization until the first forward pass. Pass `True` to delay initialization of data parallel modules. Do first forward pass and then call DistributedModelParallel.init_data_parallel(). init_parameters (bool): initialize parameters for modules still on meta device. data_parallel_wrapper (Optional[DataParallelWrapper]): custom wrapper for data parallel modules. Example:: def init_weights(m): if isinstance(m, nn.Linear): m.weight.fill_(1.0) elif isinstance(m, EmbeddingBagCollection): for param in m.parameters(): init.kaiming_normal_(param) m = MyModel(device='meta') m = DistributedModelParallel(m) m.apply(init_weights) """ def __init__( self, module: nn.Module, env: Optional[ShardingEnv] = None, device: Optional[torch.device] = None, plan: Optional[ShardingPlan] = None, sharders: Optional[List[ModuleSharder[torch.nn.Module]]] = None, init_data_parallel: bool = True, init_parameters: bool = True, data_parallel_wrapper: Optional[DataParallelWrapper] = None, ) -> None: super().__init__() torch._C._log_api_usage_once(f"torchrec.distributed.{self.__class__.__name__}") self.init_parameters = init_parameters self._ddp_wrapped: bool = False if env is None: pg = dist.GroupMember.WORLD assert pg is not None, "Process group is not initialized" env = ShardingEnv.from_process_group(pg) self._env: ShardingEnv = env if device is None: device = torch.device("cpu") self.device: torch.device = device if sharders is None: sharders = get_default_sharders() self._sharder_map: Dict[Type[nn.Module], ModuleSharder[nn.Module]] = { sharder.module_type: sharder for sharder in sharders } if data_parallel_wrapper is None: data_parallel_wrapper = DefaultDataParallelWrapper() self._data_parallel_wrapper: DataParallelWrapper = data_parallel_wrapper if plan is None: planner = EmbeddingShardingPlanner( topology=Topology( local_world_size=get_local_size(self._env.world_size), world_size=self._env.world_size, compute_device=self.device.type, ) ) pg = self._env.process_group if pg is not None: plan = planner.collective_plan(module, sharders, pg) else: plan = planner.plan(module, sharders) self._plan: ShardingPlan = plan self._dmp_wrapped_module: nn.Module = self._init_dmp(module) self._optim: CombinedOptimizer = self._init_optim(self._dmp_wrapped_module) if init_parameters: self._init_parameters(self.module) if init_data_parallel: self.init_data_parallel() def module(self) -> nn.Module: """ Property to directly access sharded module, which will not be wrapped in DDP, FSDP, DMP, or any other parallelism wrappers. """ return get_unwrapped_module(self) def module(self, value: nn.Module) -> None: if isinstance(self.module, DistributedDataParallel) or isinstance( self.module, FullyShardedDataParallel ): raise RuntimeError( "module can't be set after calling init_data_parallel(...)" ) else: self._dmp_wrapped_module = value # pyre-ignore [2, 3] def forward(self, *args, **kwargs) -> Any: return self._dmp_wrapped_module(*args, **kwargs) def init_data_parallel(self) -> None: """ See init_data_parallel c-tor argument for usage. It's safe to call this method multiple times. """ if not self._ddp_wrapped: # Allocate any 'meta' tensors if self.init_parameters: self._init_parameters(self._dmp_wrapped_module) self._data_parallel_wrapper.wrap(self, self._env, self.device) self._ddp_wrapped = True def copy( self, device: torch.device, ) -> "DistributedModelParallel": """ Recursively copy submodules to new device by calling per-module customized copy process, since some modules needs to use the original references (like `ShardedModule` for inference). """ assert isinstance(device, torch.device) # dmp code deep copy with sharded_model_copy(device=None): copy_dmp = copy.deepcopy(self) # tensor resident module deep copy copy_dmp_wrapped_module = copy_to_device( self._dmp_wrapped_module, self.device, device ) copy_dmp._dmp_wrapped_module = copy_dmp_wrapped_module return copy_dmp def _init_dmp(self, module: nn.Module) -> nn.Module: return self._shard_modules_impl(module) def _init_optim(self, module: nn.Module) -> CombinedOptimizer: # pyre-ignore [6] return CombinedOptimizer(self._fused_optim_impl(module, [])) def _fused_optim_impl( self, module: nn.Module, fused_optims: List[Tuple[str, KeyedOptimizer]], path: str = "", ) -> List[Tuple[str, KeyedOptimizer]]: if isinstance(module, FusedOptimizerModule): fused_optims.append((path, module.fused_optimizer)) return fused_optims for name, child in module.named_children(): self._fused_optim_impl( child, fused_optims, path + "." + name if path else name, ) return fused_optims def _shard_modules_impl( self, module: nn.Module, path: str = "", ) -> nn.Module: # pre-sharded module if isinstance(module, ShardedModule): return module # shardable module module_sharding_plan = self._plan.get_plan_for_module(path) if module_sharding_plan: sharder_key = type(module) module = self._sharder_map[sharder_key].shard( module, module_sharding_plan, self._env, self.device, ) return module for name, child in module.named_children(): child = self._shard_modules_impl( child, path + "." + name if path else name, ) setattr(module, name, child) return module def _init_parameters(self, module: nn.Module) -> None: def init_parameters(module: nn.Module) -> None: # Allocate parameters and buffers if over 'meta' device. has_meta_param = False for name, param in module._parameters.items(): if isinstance(param, torch.Tensor) and param.device.type == "meta": module._parameters[name] = nn.Parameter( torch.empty_like(param, device=self.device), requires_grad=param.requires_grad, ) has_meta_param = True for name, buffer in module._buffers.items(): if isinstance(buffer, torch.Tensor) and buffer.device.type == "meta": module._buffers[name] = torch.zeros_like(buffer, device=self.device) # Init parameters if at least one parameter is over 'meta' device. if has_meta_param and hasattr(module, "reset_parameters"): module.reset_parameters() module.apply(init_parameters) def sparse_grad_parameter_names( self, destination: Optional[List[str]] = None, prefix: str = "" ) -> List[str]: destination = [] if destination is None else destination return self._sparse_grad_parameter_names(self.module, destination, prefix) def _sparse_grad_parameter_names( self, module: nn.Module, destination: List[str], prefix: str = "" ) -> List[str]: module = get_unwrapped_module(module) if isinstance(module, ShardedModule): pass elif isinstance(module, nn.Embedding): if module.sparse: destination.append(append_prefix(prefix, "weight")) elif isinstance(module, nn.EmbeddingBag): if module.sparse: destination.append(append_prefix(prefix, "weight")) else: for name, child in module.named_children(): self._sparse_grad_parameter_names( child, destination, append_prefix(prefix, name) ) return destination # pyre-ignore [14] def state_dict( self, destination: Optional[Dict[str, Any]] = None, prefix: str = "", keep_vars: bool = False, ) -> Dict[str, Any]: state_dict = get_module(self).state_dict( destination=destination, prefix=prefix, keep_vars=keep_vars ) torch.nn.modules.utils.consume_prefix_in_state_dict_if_present( state_dict, prefix + _DDP_STATE_DICT_PREFIX ) add_prefix_to_state_dict(state_dict, prefix) return state_dict # pyre-fixme[14]: `load_state_dict` overrides method defined in `Module` # inconsistently. def load_state_dict( self, state_dict: "OrderedDict[str, torch.Tensor]", prefix: str = "", strict: bool = True, ) -> _IncompatibleKeys: return self._load_state_dict(self, state_dict, prefix, strict) def _load_state_dict( self, module: nn.Module, state_dict: "OrderedDict[str, torch.Tensor]", prefix: str = "", strict: bool = True, ) -> _IncompatibleKeys: missing_keys = [] unexpected_keys = [] module = get_module(module) if isinstance(module, DistributedDataParallel): torch.nn.modules.utils.consume_prefix_in_state_dict_if_present( state_dict, prefix ) add_prefix_to_state_dict(state_dict, prefix + _DDP_STATE_DICT_PREFIX) if isinstance(module, ShardedModule): return module.load_state_dict(state_dict, strict=strict) else: module._load_from_state_dict( state_dict, prefix, {}, strict, missing_keys, unexpected_keys, [] ) for name, child in module.named_children(): m_keys, u_keys = self._load_state_dict( child, filter_state_dict(state_dict, prefix + name), "", strict, ) missing_keys.extend(m_keys) unexpected_keys.extend(u_keys) return _IncompatibleKeys( missing_keys=missing_keys, unexpected_keys=unexpected_keys ) def _named_parameters( self, module: nn.Module, prefix: str = "", recurse: bool = True, strip_ddp: bool = True, ) -> Iterator[Tuple[str, torch.nn.Parameter]]: if strip_ddp: module = get_unwrapped_module(module) if isinstance(module, ShardedModule): yield from module.named_parameters(prefix, recurse) else: yield from module.named_parameters(prefix, recurse=False) for name, child in module.named_children(): yield from self._named_parameters( child, append_prefix(prefix, name), recurse, strip_ddp, ) def named_parameters( self, prefix: str = "", recurse: bool = True, remove_duplicate: bool = True, ) -> Iterator[Tuple[str, torch.nn.Parameter]]: gen = self._named_parameters( self.module, prefix, recurse, ) memo = set() for key, param in gen: if param in memo: continue if remove_duplicate: memo.add(param) yield key, param def bare_named_parameters( self, prefix: str = "", recurse: bool = True, ) -> Iterator[Tuple[str, torch.nn.Parameter]]: gen = self._named_parameters( self.module, prefix, recurse, ) memo = set() for key, param in gen: if param in memo: continue memo.add(param) yield key, param def _sharded_parameter_names(module: nn.Module, prefix: str = "") -> Iterator[str]: module = get_unwrapped_module(module) if isinstance(module, ShardedModule): yield from module.sharded_parameter_names(prefix) else: for name, child in module.named_children(): yield from DistributedModelParallel._sharded_parameter_names( child, append_prefix(prefix, name) ) def _named_buffers( self, module: nn.Module, prefix: str = "", recurse: bool = True ) -> Iterator[Tuple[str, torch.Tensor]]: module = get_unwrapped_module(module) if isinstance(module, ShardedModule): yield from module.named_buffers(prefix, recurse) else: yield from module.named_buffers(prefix, recurse=False) for name, child in module.named_children(): yield from self._named_buffers( child, append_prefix(prefix, name), recurse ) def named_buffers( self, prefix: str = "", recurse: bool = True, remove_duplicate: bool = True ) -> Iterator[Tuple[str, torch.Tensor]]: gen = self._named_buffers(self.module, prefix, recurse) memo = set() for key, param in gen: if param in memo: continue if remove_duplicate: memo.add(param) yield key, param def fused_optimizer(self) -> KeyedOptimizer: return self._optim def plan(self) -> ShardingPlan: return self._plan def _reset_parameters(module: nn.Module) -> None: for _, m in module.named_modules(): if hasattr(m, "reset_parameters"): m.reset_parameters() def meta_to_cpu_placement(module: torch.nn.Module) -> None: if hasattr(module, "_dmp_wrapped_module"): # for placement update of dmp module, we need to fetch .module (read access) and write # to .dmp_wrapped_module (write access) assert type(module) == DistributedModelParallel _meta_to_cpu_placement(module.module, module, "_dmp_wrapped_module") else: # shard module case _meta_to_cpu_placement(module, module)
null
8,945
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin def _get_batching_hinted_output(lengths: Tensor, output: Tensor) -> Tensor: # this is a fx rule to help with batching hinting jagged sequence tensor coalescing. return output
null
8,946
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin # pyre-ignore class KeyedJaggedTensor(Pipelineable, metaclass=JaggedTensorMeta): """Represents an (optionally weighted) keyed jagged tensor. A `KeyedJaggedTensor` is a tensor with a *jagged dimension* which is dimension whose slices may be of different lengths. Keyed on first dimension and jagged on the last dimension. Implementation is torch.jit.script-able. Args: keys (List[str]): keys to the jagged Tensor. values (torch.Tensor): values tensor in dense representation. weights (Optional[torch.Tensor]): if the values have weights. Tensor with the same shape as values. lengths (Optional[torch.Tensor]): jagged slices, represented as lengths. offsets (Optional[torch.Tensor]): jagged slices, represented as cumulative offsets. stride (Optional[int]): number of examples per batch. stride_per_key_per_rank (Optional[List[List[int]]]): batch size (number of examples) per key per rank, with the outer list representing the keys and the inner list representing the values. Each value in the inner list represents the number of examples in the batch from the rank of its index in a distributed context. length_per_key (Optional[List[int]]): start length for each key. offset_per_key (Optional[List[int]]): start offset for each key and final offset. index_per_key (Optional[Dict[str, int]]): index for each key. jt_dict (Optional[Dict[str, JaggedTensor]]): inverse_indices (Optional[Tuple[List[str], torch.Tensor]]): inverse indices to expand deduplicated embedding output for variable stride per key. Example:: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 dim_0: keyed dimension (ie. `Feature0`, `Feature1`) dim_1: optional second dimension (ie. batch size) dim_2: The jagged dimension which has slice lengths between 0-3 in the above example # We represent this data with following inputs: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset """ # This is the subset of fields on KJT which are required (all other fields # can be derived from these fields, and are only cached) _fields = [ "_values", "_weights", "_lengths", "_offsets", ] def __init__( self, keys: List[str], values: torch.Tensor, weights: Optional[torch.Tensor] = None, lengths: Optional[torch.Tensor] = None, offsets: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, # Below exposed to ensure torch.script-able length_per_key: Optional[List[int]] = None, offset_per_key: Optional[List[int]] = None, index_per_key: Optional[Dict[str, int]] = None, jt_dict: Optional[Dict[str, JaggedTensor]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> None: self._keys: List[str] = keys self._values: torch.Tensor = values self._weights: Optional[torch.Tensor] = weights if offsets is not None: _assert_tensor_has_no_elements_or_has_integers(offsets, "offsets") if lengths is not None: _assert_tensor_has_no_elements_or_has_integers(lengths, "lengths") self._lengths: Optional[torch.Tensor] = lengths self._offsets: Optional[torch.Tensor] = offsets self._stride_per_key_per_rank: List[List[int]] = [] self._stride_per_key: List[int] = [] self._variable_stride_per_key: bool = False self._stride: int = -1 if stride_per_key_per_rank is not None: if stride is not None: raise ValueError( "Cannot initialize KJT with both `stride` and `stride_per_key_per_rank`" ) self._stride_per_key_per_rank = stride_per_key_per_rank self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] self._variable_stride_per_key = True if not stride_per_key_per_rank: self._stride = 0 elif all(s == self.stride_per_key()[0] for s in self.stride_per_key()): self._stride = self.stride_per_key()[0] else: if torch.jit.is_tracing(): stride = _maybe_compute_stride_kjt_scripted( keys, stride, lengths, offsets )[0] else: stride = _maybe_compute_stride_kjt(keys, stride, lengths, offsets) self._stride = stride self._stride_per_key_per_rank = [[stride]] * len(self._keys) self._stride_per_key = [sum(s) for s in self._stride_per_key_per_rank] # lazy fields self._length_per_key: Optional[List[int]] = length_per_key self._offset_per_key: Optional[List[int]] = offset_per_key self._index_per_key: Optional[Dict[str, int]] = index_per_key self._jt_dict: Optional[Dict[str, JaggedTensor]] = jt_dict self._inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = ( inverse_indices ) self._lengths_offset_per_key: List[int] = [] def from_offsets_sync( keys: List[str], values: torch.Tensor, offsets: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, offsets=offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def from_lengths_sync( keys: List[str], values: torch.Tensor, lengths: torch.Tensor, weights: Optional[torch.Tensor] = None, stride: Optional[int] = None, stride_per_key_per_rank: Optional[List[List[int]]] = None, inverse_indices: Optional[Tuple[List[str], torch.Tensor]] = None, ) -> "KeyedJaggedTensor": kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, inverse_indices=inverse_indices, ) return kjt.sync() def concat( kjt_list: List["KeyedJaggedTensor"], ) -> "KeyedJaggedTensor": if len(kjt_list) == 0: raise ValueError("Can't concat empty KJT list") is_weighted: bool = kjt_list[0].weights_or_none() is not None has_length_per_key: bool = True length_per_key: List[int] = [] keys: List[str] = [] value_list: List[torch.Tensor] = [] weight_list: List[torch.Tensor] = [] length_list: List[torch.Tensor] = [] stride_per_key_per_rank: List[List[int]] = [] stride: Optional[int] = None variable_stride_per_key_list = [ kjt.variable_stride_per_key() for kjt in kjt_list ] assert all(variable_stride_per_key_list) or not any( variable_stride_per_key_list ), "variable stride per key must be consistent for all KJTs" variable_stride_per_key = all(variable_stride_per_key_list) for kjt in kjt_list: curr_is_weighted: bool = kjt.weights_or_none() is not None if is_weighted != curr_is_weighted: raise ValueError("Can't merge weighted KJT with unweighted KJT") _length_per_key: Optional[List[int]] = None if kjt._length_per_key is None: has_length_per_key = False else: _length_per_key = kjt._length_per_key if has_length_per_key and _length_per_key is not None: length_per_key += _length_per_key keys += kjt.keys() value_list.append(kjt.values()) if is_weighted: weight_list.append(kjt.weights()) length_list.append(kjt.lengths()) if variable_stride_per_key: stride_per_key_per_rank += kjt.stride_per_key_per_rank() elif stride is None: stride = kjt.stride() else: assert stride == kjt.stride(), "strides must be consistent for all KJTs" return KeyedJaggedTensor( keys=keys, values=torch.cat(value_list, dim=0), weights=torch.cat(weight_list, dim=0) if is_weighted else None, lengths=torch.cat(length_list, dim=0), stride=stride, stride_per_key_per_rank=( stride_per_key_per_rank if variable_stride_per_key else None ), length_per_key=length_per_key if has_length_per_key else None, ) def empty( is_weighted: bool = False, device: Optional[torch.device] = None, values_dtype: Optional[torch.dtype] = None, weights_dtype: Optional[torch.dtype] = None, lengths_dtype: torch.dtype = torch.int32, ) -> "KeyedJaggedTensor": weights = ( torch.empty(0, dtype=weights_dtype, device=device) if is_weighted else None ) return KeyedJaggedTensor( keys=torch.jit.annotate(List[str], []), values=torch.empty(0, dtype=values_dtype, device=device), weights=weights, lengths=torch.empty(0, dtype=lengths_dtype, device=device), stride=0, ) def empty_like(kjt: "KeyedJaggedTensor") -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, kjt.stride_per_key_per_rank()) if kjt.variable_stride_per_key() else (kjt.stride(), None) ) return KeyedJaggedTensor( keys=[], values=torch.empty(0, device=kjt.device(), dtype=kjt.values().dtype), weights=( None if kjt.weights_or_none() is None else torch.empty(0, device=kjt.device(), dtype=kjt.weights().dtype) ), lengths=torch.empty(0, device=kjt.device(), dtype=kjt.lengths().dtype), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, ) def from_jt_dict(jt_dict: Dict[str, JaggedTensor]) -> "KeyedJaggedTensor": """ Constructs a KeyedJaggedTensor from a Dict[str, JaggedTensor], but this function will ONLY work if the JaggedTensors all have the same "implicit" batch_size dimension. Basically, we can visualize JaggedTensors as 2-D tensors of the format of [batch_size x variable_feature_dim]. In case, we have some batch without a feature value, the input JaggedTensor could just not include any values. But KeyedJaggedTensor (by default) typically pad "None" so that all the JaggedTensors stored in the KeyedJaggedTensor have the same batch_size dimension. That is, in the case, the JaggedTensor input didn't automatically pad for the empty batches, this function would error / not work. Consider the visualization of the following KeyedJaggedTensor: # 0 1 2 <-- dim_1 # "Feature0" [V0,V1] None [V2] # "Feature1" [V3] [V4] [V5,V6,V7] # ^ # dim_0 Notice that the inputs for this KeyedJaggedTensor would have looked like: values: torch.Tensor = [V0, V1, V2, V3, V4, V5, V6, V7] # V == any tensor datatype weights: torch.Tensor = [W0, W1, W2, W3, W4, W5, W6, W7] # W == any tensor datatype lengths: torch.Tensor = [2, 0, 1, 1, 1, 3] # representing the jagged slice offsets: torch.Tensor = [0, 2, 2, 3, 4, 5, 8] # offsets from 0 for each jagged slice keys: List[str] = ["Feature0", "Feature1"] # correspond to each value of dim_0 index_per_key: Dict[str, int] = {"Feature0": 0, "Feature1": 1} # index for each key offset_per_key: List[int] = [0, 3, 8] # start offset for each key and final offset Now if the input jt_dict = { # "Feature0" [V0,V1] [V2] # "Feature1" [V3] [V4] [V5,V6,V7] } and the "None" is left out from each JaggedTensor, then this function would fail as we would not correctly be able to pad "None" as it does not technically know the correct batch / place to pad within the JaggedTensor. Essentially, the lengths Tensor inferred by this function would be [2, 1, 1, 1, 3] indicating variable batch_size dim_1 violates the existing assumption / precondition that KeyedJaggedTensor's should have fixed batch_size dimension. """ kjt_keys = list(jt_dict.keys()) kjt_vals_list: List[torch.Tensor] = [] kjt_lens_list: List[torch.Tensor] = [] kjt_weights_list: List[torch.Tensor] = [] stride_per_key: List[int] = [] for jt in jt_dict.values(): stride_per_key.append(len(jt.lengths())) kjt_vals_list.append(jt.values()) kjt_lens_list.append(jt.lengths()) weight = jt.weights_or_none() if weight is not None: kjt_weights_list.append(weight) kjt_vals = torch.concat(kjt_vals_list) kjt_lens = torch.concat(kjt_lens_list) kjt_weights = ( torch.concat(kjt_weights_list) if len(kjt_weights_list) > 0 else None ) kjt_stride, kjt_stride_per_key_per_rank = ( (stride_per_key[0], None) if all(s == stride_per_key[0] for s in stride_per_key) else (None, [[stride] for stride in stride_per_key]) ) kjt = KeyedJaggedTensor( keys=kjt_keys, values=kjt_vals, weights=kjt_weights, lengths=kjt_lens, stride=kjt_stride, stride_per_key_per_rank=kjt_stride_per_key_per_rank, ).sync() return kjt def sync(self) -> "KeyedJaggedTensor": self.length_per_key() self.offset_per_key() return self def unsync(self) -> "KeyedJaggedTensor": self._length_per_key = None self._offset_per_key = None return self def device(self) -> torch.device: return self._values.device def lengths(self) -> torch.Tensor: _lengths = _maybe_compute_lengths(self._lengths, self._offsets) self._lengths = _lengths return _lengths def lengths_or_none(self) -> Optional[torch.Tensor]: return self._lengths def offsets(self) -> torch.Tensor: _offsets = _maybe_compute_offsets(self._lengths, self._offsets) self._offsets = _offsets return _offsets def offsets_or_none(self) -> Optional[torch.Tensor]: return self._offsets def keys(self) -> List[str]: return self._keys def values(self) -> torch.Tensor: return self._values def weights(self) -> torch.Tensor: return _get_weights_or_throw(self._weights) def weights_or_none(self) -> Optional[torch.Tensor]: return self._weights def stride(self) -> int: return self._stride def stride_per_key(self) -> List[int]: return self._stride_per_key def stride_per_key_per_rank(self) -> List[List[int]]: return self._stride_per_key_per_rank def variable_stride_per_key(self) -> bool: return self._variable_stride_per_key def inverse_indices(self) -> Tuple[List[str], torch.Tensor]: return _get_inverse_indices_or_throw(self._inverse_indices) def inverse_indices_or_none(self) -> Optional[Tuple[List[str], torch.Tensor]]: return self._inverse_indices def _key_indices(self) -> Dict[str, int]: _index_per_key: Dict[str, int] = _maybe_compute_index_per_key( self._keys, self._index_per_key, ) self._index_per_key = _index_per_key return _index_per_key def length_per_key(self) -> List[int]: _length_per_key = _maybe_compute_length_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key return _length_per_key def length_per_key_or_none(self) -> Optional[List[int]]: return self._length_per_key def offset_per_key(self) -> List[int]: _length_per_key, _offset_per_key = _maybe_compute_offset_per_key( keys=self._keys, stride=self.stride(), stride_per_key=self.stride_per_key(), variable_stride_per_key=self.variable_stride_per_key(), length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, lengths=self._lengths, offsets=self._offsets, values=self._values, ) self._length_per_key = _length_per_key self._offset_per_key = _offset_per_key return _offset_per_key def offset_per_key_or_none(self) -> Optional[List[int]]: return self._offset_per_key def lengths_offset_per_key(self) -> List[int]: if not self._lengths_offset_per_key: self._lengths_offset_per_key = _cumsum(self.stride_per_key()) return self._lengths_offset_per_key def split(self, segments: List[int]) -> List["KeyedJaggedTensor"]: split_list: List[KeyedJaggedTensor] = [] start = 0 start_offset = 0 _length_per_key = self.length_per_key() _offset_per_key = self.offset_per_key() for segment in segments: end = start + segment end_offset = _offset_per_key[end] keys: List[str] = self._keys[start:end] stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()[start:end]) if self.variable_stride_per_key() else (self._stride, None) ) if segment == len(self._keys): # no torch slicing required split_list.append( KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self.weights_or_none(), lengths=self._lengths, offsets=self._offsets, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=self._jt_dict, inverse_indices=None, ) ) elif segment == 0: empty_int_list: List[int] = torch.jit.annotate(List[int], []) split_list.append( KeyedJaggedTensor( keys=keys, values=torch.tensor( empty_int_list, device=self.device(), dtype=self._values.dtype, ), weights=( None if self.weights_or_none() is None else torch.tensor( empty_int_list, device=self.device(), dtype=self.weights().dtype, ) ), lengths=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), offsets=torch.tensor( empty_int_list, device=self.device(), dtype=torch.int ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) else: split_length_per_key = _length_per_key[start:end] if not torch.jit.is_scripting() and is_torchdynamo_compiling(): # Checks for dynamo dynamic shapes tracing torch._check_is_size(start_offset) torch._check_is_size(end_offset) torch._check_is_size(end_offset - start_offset) torch._check(start_offset <= self._values.size(0)) torch._check(end_offset <= self._values.size(0)) torch._check(end_offset >= start_offset) split_list.append( KeyedJaggedTensor( keys=keys, values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[ start ] : self.lengths_offset_per_key()[end] ], offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=split_length_per_key, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) ) start = end start_offset = end_offset return split_list def permute( self, indices: List[int], indices_tensor: Optional[torch.Tensor] = None, include_inverse_indices: bool = False, ) -> "KeyedJaggedTensor": if indices_tensor is None: indices_tensor = torch.tensor( indices, dtype=torch.int, device=self.device() ) length_per_key = self.length_per_key() permuted_keys: List[str] = [] permuted_stride_per_key_per_rank: List[List[int]] = [] permuted_length_per_key: List[int] = [] permuted_lengths_sum = 0 for index in indices: key = self.keys()[index] permuted_keys.append(key) permuted_stride_per_key_per_rank.append( self.stride_per_key_per_rank()[index] ) permuted_length_per_key.append(length_per_key[index]) permuted_lengths_sum += length_per_key[index] if self.variable_stride_per_key(): length_per_key_tensor = _pin_and_move( torch.tensor(self.length_per_key()), self.device() ) stride_per_key_tensor = _pin_and_move( torch.tensor(self.stride_per_key()), self.device() ) permuted_lengths, _ = _permute_tensor_by_segments( self.lengths(), stride_per_key_tensor, indices_tensor, None, ) permuted_values, permuted_weights = _permute_tensor_by_segments( self.values(), length_per_key_tensor, indices_tensor, self.weights_or_none(), ) else: ( permuted_lengths, permuted_values, permuted_weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( indices_tensor, self.lengths().view(len(self._keys), -1), self.values(), self.weights_or_none(), permuted_lengths_sum, ) stride, optional_permuted_stride_per_key_per_rank = ( (None, permuted_stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) kjt = KeyedJaggedTensor( keys=permuted_keys, values=permuted_values, weights=permuted_weights, lengths=permuted_lengths.view(-1), offsets=None, stride=stride, stride_per_key_per_rank=optional_permuted_stride_per_key_per_rank, length_per_key=permuted_length_per_key if len(permuted_keys) > 0 else None, offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=( self.inverse_indices_or_none() if include_inverse_indices else None ), ) return kjt def flatten_lengths(self) -> "KeyedJaggedTensor": stride, stride_per_key_per_rank = ( (None, self.stride_per_key_per_rank()) if self.variable_stride_per_key() else (self._stride, None) ) return KeyedJaggedTensor( keys=self._keys, values=self._values, weights=self._weights, lengths=self.lengths().view(-1), offsets=None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self.length_per_key(), offset_per_key=None, index_per_key=None, jt_dict=None, inverse_indices=None, ) def __getitem__(self, key: str) -> JaggedTensor: offset_per_key = self.offset_per_key() index = self._key_indices()[key] start_offset = offset_per_key[index] end_offset = ( offset_per_key[index + 1] if index + 1 < len(offset_per_key) else start_offset ) return JaggedTensor( values=self._values[start_offset:end_offset], weights=( None if self.weights_or_none() is None else self.weights()[start_offset:end_offset] ), lengths=self.lengths()[ self.lengths_offset_per_key()[index] : self.lengths_offset_per_key()[ index + 1 ] ], offsets=None, ) def to_dict(self) -> Dict[str, JaggedTensor]: _jt_dict = _maybe_compute_kjt_to_jt_dict( stride=self.stride(), stride_per_key=self.stride_per_key(), keys=self.keys(), length_per_key=self.length_per_key(), lengths=self.lengths(), values=self.values(), variable_stride_per_key=self.variable_stride_per_key(), weights=self.weights_or_none(), jt_dict=self._jt_dict, ) self._jt_dict = _jt_dict return _jt_dict def record_stream(self, stream: torch.cuda.streams.Stream) -> None: self._values.record_stream(stream) weights = self._weights lengths = self._lengths offsets = self._offsets if weights is not None: weights.record_stream(stream) if lengths is not None: lengths.record_stream(stream) if offsets is not None: offsets.record_stream(stream) def to( self, device: torch.device, non_blocking: bool = False, dtype: Optional[torch.dtype] = None, ) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) length_per_key = self._length_per_key offset_per_key = self._offset_per_key index_per_key = self._index_per_key jt_dict = self._jt_dict inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = ( inverse_indices[0], inverse_indices[1].to(device, non_blocking=non_blocking), ) if weights is not None: if dtype is not None: weights = weights.to( dtype=dtype, device=device, non_blocking=non_blocking ) else: weights = weights.to(device=device, non_blocking=non_blocking) return KeyedJaggedTensor( keys=self._keys, values=self._values.to(device, non_blocking=non_blocking), weights=weights, lengths=( lengths.to(device, non_blocking=non_blocking) if lengths is not None else None ), offsets=( offsets.to(device, non_blocking=non_blocking) if offsets is not None else None ), stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=length_per_key, offset_per_key=offset_per_key, index_per_key=index_per_key, jt_dict=jt_dict, inverse_indices=inverse_indices, ) def __str__(self) -> str: if len(self._keys) == 0 or self._offsets is None and self._lengths is None: return "KeyedJaggedTensor()\n" offsets = self.offsets() return ( "KeyedJaggedTensor({\n" + ",\n".join( [ " " + _jagged_tensor_string( self._keys[index], self._values, self._weights, offsets, sum(self.stride_per_key()[:index]), sum(self.stride_per_key()[: index + 1]), ) for index in range(len(self._keys)) ] ) + "\n})\n" ) def pin_memory(self) -> "KeyedJaggedTensor": weights = self._weights lengths = self._lengths offsets = self._offsets stride, stride_per_key_per_rank = ( (None, self._stride_per_key_per_rank) if self.variable_stride_per_key() else (self._stride, None) ) inverse_indices = self._inverse_indices if inverse_indices is not None: inverse_indices = (inverse_indices[0], inverse_indices[1].pin_memory()) return KeyedJaggedTensor( keys=self._keys, values=self._values.pin_memory(), weights=weights.pin_memory() if weights is not None else None, lengths=lengths.pin_memory() if lengths is not None else None, offsets=offsets.pin_memory() if offsets is not None else None, stride=stride, stride_per_key_per_rank=stride_per_key_per_rank, length_per_key=self._length_per_key, offset_per_key=self._offset_per_key, index_per_key=self._index_per_key, jt_dict=None, inverse_indices=inverse_indices, ) def dist_labels(self) -> List[str]: labels = ["lengths", "values"] if self.variable_stride_per_key(): labels.append("strides") if self.weights_or_none() is not None: labels.append("weights") return labels def dist_splits(self, key_splits: List[int]) -> List[List[int]]: batch_size_per_split = _sum_by_splits(self.stride_per_key(), key_splits) length_per_split = _sum_by_splits(self.length_per_key(), key_splits) splits = [batch_size_per_split, length_per_split] if self.variable_stride_per_key(): splits.append(key_splits) if self.weights_or_none() is not None: splits.append(length_per_split) return splits def dist_tensors(self) -> List[torch.Tensor]: tensors = [self.lengths(), self.values()] if self.variable_stride_per_key(): strides = _pin_and_move(torch.tensor(self.stride_per_key()), self.device()) tensors.append(strides) if self.weights_or_none() is not None: tensors.append(self.weights()) return tensors def dist_init( keys: List[str], tensors: List[torch.Tensor], variable_stride_per_key: bool, num_workers: int, recat: Optional[torch.Tensor], stride_per_rank: Optional[List[int]], stagger: int = 1, ) -> "KeyedJaggedTensor": assert len(tensors) in [2, 3, 4] lengths = tensors[0] values = tensors[1] stride_per_rank_per_key = tensors[2] if variable_stride_per_key else None weights = ( tensors[-1] if (variable_stride_per_key and len(tensors) == 4) or (not variable_stride_per_key and len(tensors) == 3) else None ) if variable_stride_per_key: assert stride_per_rank_per_key is not None stride_per_key_per_rank: List[List[int]] = stride_per_rank_per_key.view( num_workers, len(keys) ).T.tolist() strides_cumsum: List[int] = torch.ops.fbgemm.asynchronous_complete_cumsum( stride_per_rank_per_key ).tolist() cumsum_lengths = torch.ops.fbgemm.asynchronous_complete_cumsum(lengths) length_per_key = ( cumsum_lengths[strides_cumsum[1:]] - cumsum_lengths[strides_cumsum[:-1]] ) with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: lengths, _ = _permute_tensor_by_segments( lengths, stride_per_rank_per_key, recat, None, ) values, weights = _permute_tensor_by_segments( values, length_per_key, recat, weights, ) if not stride_per_key_per_rank: stride_per_key_per_rank = [[0]] * len(keys) if stagger > 1: stride_per_key_per_rank_stagger: List[List[int]] = [] local_world_size = num_workers // stagger for i in range(len(keys)): stride_per_rank_stagger: List[int] = [] for j in range(local_world_size): stride_per_rank_stagger.extend( stride_per_key_per_rank[i][j::local_world_size] ) stride_per_key_per_rank_stagger.append(stride_per_rank_stagger) stride_per_key_per_rank = stride_per_key_per_rank_stagger kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride_per_key_per_rank=stride_per_key_per_rank, ) return kjt.sync() else: assert stride_per_rank is not None with record_function("## all2all_data:recat_values ##"): if recat is not None and recat.numel() > 0: stride = stride_per_rank[0] if all(s == stride for s in stride_per_rank): ( lengths, values, weights, ) = torch.ops.fbgemm.permute_2D_sparse_data( recat, lengths.view(-1, stride), values, weights, values.numel(), ) lengths = lengths.view(-1) else: # variable batch size per rank ( lengths, values, weights, ) = torch.ops.fbgemm.permute_1D_sparse_data( recat, lengths.view(-1), values, weights, values.numel(), ) kjt = KeyedJaggedTensor( keys=keys, values=values, weights=weights, lengths=lengths, stride=sum(stride_per_rank), ) return kjt.sync() def _get_feature_length(feature: KeyedJaggedTensor) -> Tensor: return feature.lengths()
null
8,947
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin MODULE_ATTR_QUANT_STATE_DICT_SPLIT_SCALE_BIAS: str = ( "__quant_state_dict_split_scale_bias" ) def quant_prep_enable_quant_state_dict_split_scale_bias(module: nn.Module) -> None: setattr(module, MODULE_ATTR_QUANT_STATE_DICT_SPLIT_SCALE_BIAS, True)
null
8,948
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin try: torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops") torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu") except OSError: pass MODULE_ATTR_QUANT_STATE_DICT_SPLIT_SCALE_BIAS: str = ( "__quant_state_dict_split_scale_bias" ) def for_each_module_of_type_do( module: nn.Module, module_types: List[Type[torch.nn.Module]], op: Callable[[torch.nn.Module], None], ) -> None: for m in module.modules(): if any([isinstance(m, t) for t in module_types]): op(m) def quant_prep_enable_quant_state_dict_split_scale_bias_for_types( module: nn.Module, module_types: List[Type[torch.nn.Module]] ) -> None: for_each_module_of_type_do( module, module_types, lambda m: setattr(m, MODULE_ATTR_QUANT_STATE_DICT_SPLIT_SCALE_BIAS, True), )
null
8,949
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin try: torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops") torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu") except OSError: pass MODULE_ATTR_REGISTER_TBES_BOOL: str = "__register_tbes_in_named_modules" def for_each_module_of_type_do( module: nn.Module, module_types: List[Type[torch.nn.Module]], op: Callable[[torch.nn.Module], None], ) -> None: for m in module.modules(): if any([isinstance(m, t) for t in module_types]): op(m) def quant_prep_enable_register_tbes( module: nn.Module, module_types: List[Type[torch.nn.Module]] ) -> None: for_each_module_of_type_do( module, module_types, lambda m: setattr(m, MODULE_ATTR_REGISTER_TBES_BOOL, True), )
null
8,950
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin try: torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops") torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu") except OSError: pass MODULE_ATTR_ROW_ALIGNMENT_INT: str = "__register_row_alignment_in_named_modules" def for_each_module_of_type_do( module: nn.Module, module_types: List[Type[torch.nn.Module]], op: Callable[[torch.nn.Module], None], ) -> None: for m in module.modules(): if any([isinstance(m, t) for t in module_types]): op(m) def quant_prep_customize_row_alignment( module: nn.Module, module_types: List[Type[torch.nn.Module]], row_alignment: int ) -> None: for_each_module_of_type_do( module, module_types, lambda m: setattr(m, MODULE_ATTR_ROW_ALIGNMENT_INT, row_alignment), )
null
8,951
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin try: torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops") torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu") except OSError: pass def pruned_num_embeddings(pruning_indices_mapping: Tensor) -> int: return int(torch.max(pruning_indices_mapping).item()) + 1 DATA_TYPE_NUM_BITS: Dict[DataType, int] = { DataType.FP32: 32, DataType.FP16: 16, DataType.BF16: 16, DataType.INT8: 8, DataType.UINT8: 8, DataType.INT4: 4, DataType.INT2: 2, } def quantize_state_dict( module: nn.Module, table_name_to_quantized_weights: Dict[str, Tuple[Tensor, Tensor]], table_name_to_data_type: Dict[str, DataType], table_name_to_pruning_indices_mapping: Optional[Dict[str, Tensor]] = None, ) -> torch.device: device = torch.device("cpu") if not table_name_to_pruning_indices_mapping: table_name_to_pruning_indices_mapping = {} for key, tensor in module.state_dict().items(): # Extract table name from state dict key. # e.g. ebc.embedding_bags.t1.weight splits = key.split(".") assert splits[-1] == "weight" table_name = splits[-2] data_type = table_name_to_data_type[table_name] num_rows = tensor.shape[0] pruning_indices_mapping: Optional[Tensor] = None if table_name in table_name_to_pruning_indices_mapping: pruning_indices_mapping = table_name_to_pruning_indices_mapping[table_name] if pruning_indices_mapping is not None: num_rows = pruned_num_embeddings(pruning_indices_mapping) device = tensor.device num_bits = DATA_TYPE_NUM_BITS[data_type] if tensor.is_meta: quant_weight = torch.empty( (num_rows, (tensor.shape[1] * num_bits) // 8), device="meta", dtype=torch.uint8, ) if ( data_type == DataType.INT8 or data_type == DataType.INT4 or data_type == DataType.INT2 ): scale_shift = torch.empty( (num_rows, 4), device="meta", dtype=torch.uint8, ) else: scale_shift = None else: if pruning_indices_mapping is not None: rows_mask = pruning_indices_mapping.gt(-1) tensor = tensor[rows_mask, :] if tensor.dtype == torch.float or tensor.dtype == torch.float16: if data_type == DataType.FP16: if tensor.dtype == torch.float: tensor = tensor.half() quant_res = tensor.view(torch.uint8) else: quant_res = ( torch.ops.fbgemm.FloatOrHalfToFusedNBitRowwiseQuantizedSBHalf( tensor, num_bits ) ) else: raise Exception("Unsupported dtype: {tensor.dtype}") if ( data_type == DataType.INT8 or data_type == DataType.INT4 or data_type == DataType.INT2 ): quant_weight, scale_shift = ( quant_res[:, :-4], quant_res[:, -4:], ) else: quant_weight, scale_shift = quant_res, None table_name_to_quantized_weights[table_name] = (quant_weight, scale_shift) return device
null
8,952
import copy import itertools from collections import defaultdict from typing import Callable, cast, Dict, List, Optional, Tuple, Type, Union import torch import torch.nn as nn from fbgemm_gpu.split_table_batched_embeddings_ops_inference import ( EmbeddingLocation, IntNBitTableBatchedEmbeddingBagsCodegen, PoolingMode, ) from torch import Tensor from torchrec.distributed.utils import none_throws from torchrec.modules.embedding_configs import ( BaseEmbeddingConfig, DATA_TYPE_NUM_BITS, data_type_to_sparse_type, DataType, dtype_to_data_type, EmbeddingBagConfig, EmbeddingConfig, pooling_type_to_pooling_mode, PoolingType, QuantConfig, ) from torchrec.modules.embedding_modules import ( EmbeddingBagCollection as OriginalEmbeddingBagCollection, EmbeddingBagCollectionInterface, EmbeddingCollection as OriginalEmbeddingCollection, EmbeddingCollectionInterface, get_embedding_names_by_table, ) from torchrec.modules.feature_processor_ import FeatureProcessorsCollection from torchrec.modules.fp_embedding_modules import ( FeatureProcessedEmbeddingBagCollection as OriginalFeatureProcessedEmbeddingBagCollection, ) from torchrec.modules.utils import construct_jagged_tensors_inference from torchrec.sparse.jagged_tensor import JaggedTensor, KeyedJaggedTensor, KeyedTensor from torchrec.tensor_types import UInt2Tensor, UInt4Tensor from torchrec.types import ModuleNoCopyMixin try: torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops") torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu") except OSError: pass def dtype_to_data_type(dtype: torch.dtype) -> DataType: if dtype == torch.float: return DataType.FP32 elif dtype == torch.float16 or dtype == torch.half: return DataType.FP16 elif dtype == torch.bfloat16: return DataType.BF16 elif dtype in {torch.int, torch.int32}: return DataType.INT32 elif dtype in {torch.long, torch.int64}: return DataType.INT64 elif dtype in {torch.quint8, torch.qint8, torch.int8}: return DataType.INT8 elif dtype == torch.uint8: return DataType.UINT8 elif dtype == torch.quint4x2: return DataType.INT4 elif dtype == torch.quint2x4: return DataType.INT2 else: raise Exception(f"Invalid data type {dtype}") class BaseEmbeddingConfig: num_embeddings: int embedding_dim: int name: str = "" data_type: DataType = DataType.FP32 feature_names: List[str] = field(default_factory=list) weight_init_max: Optional[float] = None weight_init_min: Optional[float] = None # if table is pruned: # num_embeddings is pruned number of embeddings, # pruning_indices_mapping is not None and contains remapping tensor of shape=[unpruned_num_embeddings], dtype=int32 pruning_indices_remapping: Optional[torch.Tensor] = None init_fn: Optional[Callable[[torch.Tensor], Optional[torch.Tensor]]] = None # when the position_weighted feature is in this table config, # enable this flag to support rw_sharding need_pos: bool = False def get_weight_init_max(self) -> float: if self.weight_init_max is None: return sqrt(1 / self.num_embeddings) else: return self.weight_init_max def get_weight_init_min(self) -> float: if self.weight_init_min is None: return -sqrt(1 / self.num_embeddings) else: return self.weight_init_min def num_features(self) -> int: return len(self.feature_names) def __post_init__(self) -> None: if self.init_fn is None: self.init_fn = partial( torch.nn.init.uniform_, a=self.get_weight_init_min(), b=self.get_weight_init_max(), ) class QuantConfig(NamedTuple): activation: torch.quantization.PlaceholderObserver weight: torch.quantization.PlaceholderObserver per_table_weight_dtype: Optional[Dict[str, torch.dtype]] = None def _update_embedding_configs( embedding_configs: List[BaseEmbeddingConfig], quant_config: Union[QuantConfig, torch.quantization.QConfig], ) -> None: per_table_weight_dtype = ( quant_config.per_table_weight_dtype if isinstance(quant_config, QuantConfig) and quant_config.per_table_weight_dtype else {} ) for config in embedding_configs: config.data_type = dtype_to_data_type( per_table_weight_dtype[config.name] if config.name in per_table_weight_dtype else quant_config.weight().dtype )
null
8,953
from typing import Any, Callable, Dict, List, Optional, Union import torch from torch.fx._compatibility import compatibility from torch.fx.graph import Graph from torch.fx.node import Argument from torchrec.distributed.types import LazyAwaitable, NoWait from torchrec.fx.utils import dmp_fx_trace_forward _is_fx_tracing_flag = False def is_fx_tracing() -> bool: return _is_fx_tracing_flag
null
8,954
from typing import Any, Callable, Dict, List, Optional, Union import torch from torch.fx._compatibility import compatibility from torch.fx.graph import Graph from torch.fx.node import Argument from torchrec.distributed.types import LazyAwaitable, NoWait from torchrec.fx.utils import dmp_fx_trace_forward class Tracer(torch.fx.Tracer): """ Custom FX tracer for torchrec See `Torch.FX documentation <https://pytorch.org/docs/stable/fx.html>`_ We create a custom FX tracer to trace torchrec based models. The custom tracer handles python generic types (i.e. NoWait[T], Awaitable[T]) and lower it to TorchScript if needed """ def __init__(self, leaf_modules: Optional[List[str]] = None) -> None: super().__init__() self._leaf_modules: List[str] = leaf_modules if leaf_modules is not None else [] def is_leaf_module(self, m: torch.nn.Module, module_qualified_name: str) -> bool: """ Override FX definition to include quantized embedding bags """ if type(m).__name__ in self._leaf_modules: return True return super().is_leaf_module(m, module_qualified_name) def trace( self, # pyre-ignore[2]: Missing parameter annotation [2]: Parameter `root` must have a type that does not contain `Any` root: Union[torch.nn.Module, Callable[..., Any]], concrete_args: Optional[Dict[str, Any]] = None, ) -> Graph: global _is_fx_tracing_flag old_is_fx_tracing_flag = _is_fx_tracing_flag _is_fx_tracing_flag = True try: # TODO(ivankobzarev): support DMP not only on the root level from torchrec.distributed.model_parallel import DistributedModelParallel if isinstance(root, DistributedModelParallel): # In the case where the module is wrapped in DMP, you need to replace DMP's forward # call with a new signature, one with explicit args, because fx can't handle variable args. # Furthermore, we need to provide the `fn_root` argument because when tracing a function, # fx uses an empty module as the root (unless one is explicitly provided), which leads to # issues with path_of_module and named_buffers. # TODO(shababayub): This can be removed if we either stop supporting dmp wrapping # for fx trace or strip dmp name in named_modules path (much like named_buffers). if isinstance(root, torch.nn.Module): for prefix, module in root.named_modules(): # TODO(T140754678): Remove this workaround to _fx_path module._fx_path = prefix dmp = root graph = super().trace( root=dmp_fx_trace_forward(dmp, self), concrete_args=concrete_args, ) self.root._dmp_wrapped_module = dmp._dmp_wrapped_module else: # Unwrapped dmp modules and composibility api will enter here. graph = super().trace( root=root, concrete_args=concrete_args, ) finally: _is_fx_tracing_flag = old_is_fx_tracing_flag return graph # pyre-ignore[2] def create_arg(self, a: Any) -> Argument: """ A method to specify the behavior of tracing when preparing values to be used as arguments to nodes in the ``Graph``. Adds support for the NoWait type in addition to the default tracer Args: a (Any): The value to be emitted as an ``Argument`` in the ``Graph``. Returns: Argument: The value ``a`` converted into the appropriate ``Argument`` """ if isinstance(a, NoWait): return self.create_node( "call_function", target=NoWait, args=self.create_arg((a._obj,)), kwargs={}, type_expr=NoWait, ) # jit script has explicit convertions to torch.device from str if isinstance(a, torch.device): return super().create_arg(f"{a.type}:{a.index}") # Not equivalent to when LazyAwaitable.wait() is called in eager. Here can be called earlier, as attr was not requested and this is not guranteed to be torch function # TODO(ivankobzarev): support equivalent timing of LazyAwaitable if isinstance(a, LazyAwaitable): if a._result is None: a._result = a.wait() return super().create_arg(a._result) return super().create_arg(a) def path_of_module(self, mod: torch.nn.Module) -> str: """ Allows trace-ability of non registered modules. This is typically used for Table Batched Embeddings made to look like nn.EmbeddingBags """ if hasattr(mod, "_fx_path"): return mod._fx_path else: return super().path_of_module(mod) The provided code snippet includes necessary dependencies for implementing the `symbolic_trace` function. Write a Python function `def symbolic_trace( # pyre-ignore[24] root: Union[torch.nn.Module, Callable], concrete_args: Optional[Dict[str, Any]] = None, leaf_modules: Optional[List[str]] = None, ) -> torch.fx.GraphModule` to solve the following problem: Symbolic tracing API Given an ``nn.Module`` or function instance ``root``, this function will return a ``GraphModule`` constructed by recording operations seen while tracing through ``root``. ``concrete_args`` allows you to partially specialize your function, whether it's to remove control flow or data structures. Args: root (Union[torch.nn.Module, Callable]): Module or function to be traced and converted into a Graph representation. concrete_args (Optional[Dict[str, any]]): Inputs to be partially specialized Returns: GraphModule: a Module created from the recorded operations from ``root``. Here is the function: def symbolic_trace( # pyre-ignore[24] root: Union[torch.nn.Module, Callable], concrete_args: Optional[Dict[str, Any]] = None, leaf_modules: Optional[List[str]] = None, ) -> torch.fx.GraphModule: """ Symbolic tracing API Given an ``nn.Module`` or function instance ``root``, this function will return a ``GraphModule`` constructed by recording operations seen while tracing through ``root``. ``concrete_args`` allows you to partially specialize your function, whether it's to remove control flow or data structures. Args: root (Union[torch.nn.Module, Callable]): Module or function to be traced and converted into a Graph representation. concrete_args (Optional[Dict[str, any]]): Inputs to be partially specialized Returns: GraphModule: a Module created from the recorded operations from ``root``. """ tracer = Tracer(leaf_modules) graph = tracer.trace(root, concrete_args) return torch.fx.GraphModule(root, graph)
Symbolic tracing API Given an ``nn.Module`` or function instance ``root``, this function will return a ``GraphModule`` constructed by recording operations seen while tracing through ``root``. ``concrete_args`` allows you to partially specialize your function, whether it's to remove control flow or data structures. Args: root (Union[torch.nn.Module, Callable]): Module or function to be traced and converted into a Graph representation. concrete_args (Optional[Dict[str, any]]): Inputs to be partially specialized Returns: GraphModule: a Module created from the recorded operations from ``root``.
8,955
import inspect from typing import Any, Dict, List, Set import torch from torch.fx._symbolic_trace import is_fx_tracing def fake_range(): # pyre-fixme[16]: Module `_C` has no attribute `_jit_tree_views`. return torch._C._jit_tree_views.SourceRangeFactory("", None, 0, 0).make_raw_range( 0, 1 ) def dmp_fx_trace_forward( # noqa: C901 # pyre-ignore dmp, tracer: torch.fx.Tracer, ): func = dmp._dmp_wrapped_module.forward sign: inspect.Signature = inspect.signature(func) module_to_type_str: Dict[str, Set[str]] = {} def add_if_missing(module: str, type_str: str) -> None: if module not in module_to_type_str: _set = set() _set.add(type_str) module_to_type_str[module] = _set else: s = module_to_type_str[module] if type_str not in s: s.add(type_str) def torch_no_import(t: torch.Type) -> bool: return isinstance( t, (torch.FloatType, torch.IntType, torch.ComplexType, torch.StringType) ) def torch_typing(t: torch.Type) -> bool: return isinstance( t, ( torch.TupleType, torch.ListType, torch.DictType, torch.OptionalType, torch.AnyType, ), ) exec_imports = [] args_call = ", ".join([f"{p.name}" for p in sign.parameters.values()]) types = [] try: args_decls: List[str] = [] for p in sign.parameters.values(): pann = p.annotation ptype = torch.jit.annotations.try_ann_to_type(pann, fake_range()) types.append(ptype) args_decls.append(f"{p.name}: {ptype}") while len(types) > 0: t = types.pop() if torch_no_import(t): continue t_base_name = f"{t}".split("[")[0] if torch_typing(t): add_if_missing("typing", t_base_name) else: if hasattr(t, "__module__") and not torch_no_import(t): m = t.__module__ add_if_missing(f"{m}", f"{t}".split("[")[0]) if hasattr(t, "containedTypes"): contained_types = getattr(t, "containedTypes", None)() for ctype in contained_types: types.append(ctype) if hasattr(t, "getElementType"): el_type = getattr(t, "getElementType", None)() args_decl = ", ".join(args_decls) for m, s in module_to_type_str.items(): ts = ", ".join(s) exec_imports.append(f"from {m} import {ts}") except Exception as e: print(f"Exception:{e}") # Catching here if source is not available to proceed hoping that jit will infer correct types without annotations. # Often it fails here when can not access to dataclass generated __init__ args_decl = args_call exec_def_fn_name = "__fx_forward" exec_dmp_wrapper_local_name = "_dmp_wrapped_module_local" _dmp_wrapped_module_local = dmp locals_dict = locals() exec_def = f"def {exec_def_fn_name}({args_decl}):\n return {exec_dmp_wrapper_local_name}({args_call})" exec_imports_str = "\n".join(exec_imports) pycode = f"{exec_imports_str}\n{exec_def}" exec(pycode, locals_dict) # noqa: P204 Allow use of exec wrapper = locals_dict[exec_def_fn_name] wrapper.__signature__ = sign return wrapper
null
8,956
import inspect from typing import Any, Dict, List, Set import torch from torch.fx._symbolic_trace import is_fx_tracing def _fx_marker(s: str, any_proxy_unused: Any) -> None: pass def fx_marker(s: str, any_proxy_unused: Any) -> None: if is_fx_tracing(): _fx_marker(s, any_proxy_unused)
null
8,957
import inspect from typing import Any, Dict, List, Set import torch from torch.fx._symbolic_trace import is_fx_tracing def _fx_marker(s: str, any_proxy_unused: Any) -> None: pass def is_marker_node(node: torch.fx.Node, marker_name: str) -> bool: # bool() syntax for pyre return bool( node.op == "call_function" and node.target == _fx_marker and isinstance(node.args[0], str) and node.args[0] == marker_name )
null
8,958
import ast import json from argparse import ArgumentParser, Namespace from typing import Any, Dict, List, Optional, Tuple def check_class_definition(python_path: str, node: ast.ClassDef) -> None: """ This function will run set of sanity checks against class definitions and their docstrings. Args: python_path: Path to the file that is getting checked node: AST node with the ClassDef that needs to be checked Returns: None """ assert ( type(node) == ast.ClassDef ), "Received invalid node type. Expected ClassDef, got: {}".format(type(node)) is_TorchRec_module = False is_test_file = "tests" in python_path for base in node.bases: # For now only names and attributes are supported if type(base) != ast.Name and type(base) != ast.Attribute: # pragma: nocover continue # We assume that TorchRec module has one of the following inheritance patterns: # 1. `class SomeTorchRecModule(LazyModuleExtensionMixin, torch.nn.Module)` # 2. `class SomeTorchRecModule(torch.nn.Module)` # pyre-ignore[16]: `_ast.expr` has no attribute `id`. if hasattr(base, "id") and base.id == "LazyModuleExtensionMixin": is_TorchRec_module = True break # pyre-ignore[16]: `_ast.expr` has no attribute `id`. elif hasattr(base, "attr") and base.attr == "Module": is_TorchRec_module = True break if not is_TorchRec_module or is_test_file: return docstring: Optional[str] = ast.get_docstring(node) if docstring is None: print_error_message( python_path, node, "No docstring found in a TorchRec module", "TorchRec modules are required to have a docstring describing how " "to use them. Given Module don't have a docstring, please fix this.", ) return # Check presence of the example: if "Example:" not in docstring: print_error_message( python_path, node, "No runnable example in a TorchRec module", "TorchRec modules are required to have runnable examples in " '"Example:" section. Please fix the docstring', ) # Check correctness of the Args for a class definition: required_keywords = ["Args:"] missing_keywords = [] for keyword in required_keywords: if keyword not in docstring: missing_keywords.append(keyword) if len(missing_keywords) > 0: print_error_message( python_path, node, "Missing required keywords from TorchRec module", "TorchRec modules are required to description of their args and " 'results in "Args:". ' "Missing keywords: {}.".format(missing_keywords), ) # Check actual args from the functions # pyre-ignore[33]: Explicit annotation for `functions` cannot contain `Any`. functions: Dict[str, Tuple[List[Any], List[Any]]] = {} function_sub_nodes = {} for sub_node in node.body: if type(sub_node) == ast.FunctionDef: assert isinstance(sub_node, ast.FunctionDef) functions[sub_node.name] = get_function_args(sub_node) function_sub_nodes[sub_node.name] = sub_node def check_function(function_name: str) -> None: if function_name not in functions: return if function_name == "__init__": # NOTE: -1 to not count the `self` argument. num_args = sum([len(args) for args in functions[function_name]]) - 1 if num_args > MAX_NUM_ARGS_IN_MODULE_CTOR: print_error_message( python_path, node, "TorchRec module has too many constructor arguments", "TorchRec module can have at most {} constructor arguments, but this module has {}.".format( MAX_NUM_ARGS_IN_MODULE_CTOR, num_args, ), ) if function_name in functions: missing_required_args = [] missing_optional_args = [] for arg in functions[function_name][0]: # Ignore checks for required self and net args if arg == "self" or arg == "net": continue assert docstring is not None if arg not in docstring: missing_required_args.append(arg) for arg in functions[function_name][1]: assert docstring is not None if arg not in docstring: missing_optional_args.append(arg) if len(missing_required_args) > 0 or len(missing_optional_args) > 0: print_error_message( python_path, node, "Missing docstring descriptions for {} function arguments.".format( function_name ), ( "Missing descriptions for {} function arguments. " "Missing required args: {}, missing optional args: {}" ).format( function_name, missing_required_args, missing_optional_args, ), ) # pyre-ignore[53] def check_function_docstring(function_name: str) -> None: if function_name not in functions: return function_docstring: Optional[str] = None function_docstring = ast.get_docstring(function_sub_nodes[function_name]) if function_docstring is None: print_error_message( python_path, node, "Missing docstring for {} function".format(function_name), "Missing docstring for {} function".format(function_name), ) return missing_required_args = [] missing_optional_args = [] for arg in functions[function_name][0]: # Ignore checks for required self and net args if arg == "self" or arg == "net": continue assert function_docstring is not None if arg not in function_docstring: missing_required_args.append(arg) for arg in functions[function_name][1]: assert function_docstring is not None if arg not in function_docstring: missing_optional_args.append(arg) if len(missing_required_args) > 0 or len(missing_optional_args) > 0: print_error_message( python_path, node, "Missing docstring descriptions for {} function arguments.".format( function_name ), ( "Missing descriptions for {} function arguments. " "Missing required args: {}, missing optional args: {}" ).format( function_name, missing_required_args, missing_optional_args, ), ) assert function_docstring is not None if "Returns:" not in function_docstring: print_error_message( python_path, node, "Missing docstring descriptions for {} function arguments.".format( function_name ), ( "Missing descriptions for {} function arguments. " "Missing Returns section" ).format( function_name, ), ) check_function("__init__") check_function_docstring("forward") def read_file(path: str) -> str: # pragma: nocover """ This function simply reads contents of the file. It's moved out to a function purely to simplify testing process. Args: path: File to read. Returns: content(str): Content of given file. """ return open(path).read() The provided code snippet includes necessary dependencies for implementing the `linter_one_file` function. Write a Python function `def linter_one_file(python_path: str) -> None` to solve the following problem: This function will check all Modules defined in the given file for a valid documentation based on the AST. Input args: python_path: Path to the file that need to be verified with the linter. Returns: None Here is the function: def linter_one_file(python_path: str) -> None: """ This function will check all Modules defined in the given file for a valid documentation based on the AST. Input args: python_path: Path to the file that need to be verified with the linter. Returns: None """ python_path = python_path.strip() try: for node in ast.parse(read_file(python_path)).body: if type(node) == ast.ClassDef: assert isinstance(node, ast.ClassDef) check_class_definition(python_path, node) except SyntaxError as e: # pragma: nocover # possible failing due to file parsing error lint_item = { "path": python_path, "line": e.lineno, "char": e.offset, "severity": "warning", "name": "syntax-error", "description": ( f"There is a linter parser error with message: {e.msg}. " "Please report the diff to torchrec oncall" ), "bypassChangedLineFiltering": True, } print(json.dumps(lint_item))
This function will check all Modules defined in the given file for a valid documentation based on the AST. Input args: python_path: Path to the file that need to be verified with the linter. Returns: None
8,959
import ast import json from argparse import ArgumentParser, Namespace from typing import Any, Dict, List, Optional, Tuple def _make_argparse() -> ArgumentParser: # pragma: nocover parser = ArgumentParser( description="TorchRec docstring linter", fromfile_prefix_chars="@" ) parser.add_argument("source_files", nargs="+", help="Path to python source files") return parser def _parse_args() -> Namespace: # pragma: nocover ap = _make_argparse() return ap.parse_args()
null
8,960
import itertools from typing import List, Tuple import torch import torch._prims_common as utils def down_size(N: int, size: torch.Size) -> Tuple[int, int]: assert size[-1] % N == 0, f"{size} last dim not divisible by {N}" return (*size[:-1], size[-1] // N)
null
8,961
import itertools from typing import List, Tuple import torch import torch._prims_common as utils def up_size(N: int, size: torch.Size) -> Tuple[int, int]: return (*size[:-1], size[-1] * N)
null