diff --git a/sft/smoe_vcr/checkpoint-13312/latest b/sft/smoe_vcr/checkpoint-13312/latest new file mode 100644 index 0000000000000000000000000000000000000000..41750eb6d66126b6023d0560fd5c7875c0706774 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/latest @@ -0,0 +1 @@ +global_step13312 \ No newline at end of file diff --git a/sft/smoe_vcr/checkpoint-13312/model-00001-of-00002.safetensors b/sft/smoe_vcr/checkpoint-13312/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5ee6759e18ffe2324661938bdd61453415e2bfe2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1b08e9099f94c3d1a61dc176cb6ebace68cce03e5fb193dd2134237b6ec090d +size 4972489328 diff --git a/sft/smoe_vcr/checkpoint-13312/model-00002-of-00002.safetensors b/sft/smoe_vcr/checkpoint-13312/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..da06ad886c0ac72413c4c0b097c0eeb6db12dffc --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29d557b44abd6c334cc7711d303922022efcba465e24bd2c199779f8582b4485 +size 3759015720 diff --git a/sft/smoe_vcr/checkpoint-13312/rng_state_0.pth b/sft/smoe_vcr/checkpoint-13312/rng_state_0.pth new file mode 100644 index 0000000000000000000000000000000000000000..9231f69f5fd461899867106a669ce247e70c72c2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/rng_state_0.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f23d807f0e704f4ca79670a6631cbff43189cf7f8ff4e1fc0a4330e636a798 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-13312/rng_state_1.pth b/sft/smoe_vcr/checkpoint-13312/rng_state_1.pth new file mode 100644 index 0000000000000000000000000000000000000000..19fe2dcc766f192ea5de79cec4dcff17172a10f7 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/rng_state_1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d37f92f6aea5386e84d2d64a1a25d6ef96a10b3bbbfe63627981604c8934076 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-13312/rng_state_2.pth b/sft/smoe_vcr/checkpoint-13312/rng_state_2.pth new file mode 100644 index 0000000000000000000000000000000000000000..bfe492519c6b79b07a8d68b98c5f3d0c073667aa --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/rng_state_2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667ebf727735115f00a6bdbe090344e9846c726d11bb555cdc201c415f27ad85 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-13312/rng_state_3.pth b/sft/smoe_vcr/checkpoint-13312/rng_state_3.pth new file mode 100644 index 0000000000000000000000000000000000000000..838d42ad13e30851fdbd1d8801738a4106a9ce8b --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/rng_state_3.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d306f8c511cba8a225e3b723c5fa79d8a6ecc922f834da914ff0780c78b1fc +size 14960 diff --git a/sft/smoe_vcr/checkpoint-13312/tokenizer.model b/sft/smoe_vcr/checkpoint-13312/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/checkpoint-13312/training_args.bin b/sft/smoe_vcr/checkpoint-13312/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992 diff --git a/sft/smoe_vcr/checkpoint-13312/zero_to_fp32.py b/sft/smoe_vcr/checkpoint-13312/zero_to_fp32.py new file mode 100644 index 0000000000000000000000000000000000000000..24cc342e78d1a006c782b3a4cd68d9ce786d8fd8 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-13312/zero_to_fp32.py @@ -0,0 +1,604 @@ +#!/usr/bin/env python + +# Copyright (c) Microsoft Corporation. +# SPDX-License-Identifier: Apache-2.0 + +# DeepSpeed Team + +# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets +# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in +# the future. Once extracted, the weights don't require DeepSpeed and can be used in any +# application. +# +# example: python zero_to_fp32.py . pytorch_model.bin + +import argparse +import torch +import glob +import math +import os +import re +from collections import OrderedDict +from dataclasses import dataclass + +# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with +# DeepSpeed data structures it has to be available in the current python environment. +from deepspeed.utils import logger +from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS, + FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES, + FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS) + + +@dataclass +class zero_model_state: + buffers: dict() + param_shapes: dict() + shared_params: list + ds_version: int + frozen_param_shapes: dict() + frozen_param_fragments: dict() + + +debug = 0 + +# load to cpu +device = torch.device('cpu') + + +def atoi(text): + return int(text) if text.isdigit() else text + + +def natural_keys(text): + ''' + alist.sort(key=natural_keys) sorts in human order + http://nedbatchelder.com/blog/200712/human_sorting.html + (See Toothy's implementation in the comments) + ''' + return [atoi(c) for c in re.split(r'(\d+)', text)] + + +def get_model_state_file(checkpoint_dir, zero_stage): + if not os.path.isdir(checkpoint_dir): + raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist") + + # there should be only one file + if zero_stage <= 2: + file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt") + elif zero_stage == 3: + file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt") + + if not os.path.exists(file): + raise FileNotFoundError(f"can't find model states file at '{file}'") + + return file + + +def get_checkpoint_files(checkpoint_dir, glob_pattern): + # XXX: need to test that this simple glob rule works for multi-node setup too + ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys) + + if len(ckpt_files) == 0: + raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'") + + return ckpt_files + + +def get_optim_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt") + + +def get_model_state_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_model_states.pt") + + +def parse_model_states(files): + zero_model_states = [] + for file in files: + state_dict = torch.load(file, map_location=device) + + if BUFFER_NAMES not in state_dict: + raise ValueError(f"{file} is not a model state checkpoint") + buffer_names = state_dict[BUFFER_NAMES] + if debug: + print("Found buffers:", buffer_names) + + # recover just the buffers while restoring them to fp32 if they were saved in fp16 + buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names} + param_shapes = state_dict[PARAM_SHAPES] + + # collect parameters that are included in param_shapes + param_names = [] + for s in param_shapes: + for name in s.keys(): + param_names.append(name) + + # update with frozen parameters + frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None) + if frozen_param_shapes is not None: + if debug: + print(f"Found frozen_param_shapes: {frozen_param_shapes}") + param_names += list(frozen_param_shapes.keys()) + + # handle shared params + shared_params = [[k, v] for k, v in state_dict["shared_params"].items()] + + ds_version = state_dict.get(DS_VERSION, None) + + frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None) + + z_model_state = zero_model_state(buffers=buffers, + param_shapes=param_shapes, + shared_params=shared_params, + ds_version=ds_version, + frozen_param_shapes=frozen_param_shapes, + frozen_param_fragments=frozen_param_fragments) + zero_model_states.append(z_model_state) + + return zero_model_states + + +def parse_optim_states(files, ds_checkpoint_dir): + + total_files = len(files) + state_dicts = [] + for f in files: + state_dict = torch.load(f, map_location=device) + # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights + # and also handle the case where it was already removed by another helper script + state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None) + state_dicts.append(state_dict) + + if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]: + raise ValueError(f"{files[0]} is not a zero checkpoint") + zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE] + world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT] + + # For ZeRO-2 each param group can have different partition_count as data parallelism for expert + # parameters can be different from data parallelism for non-expert parameters. So we can just + # use the max of the partition_count to get the dp world_size. + + if type(world_size) is list: + world_size = max(world_size) + + if world_size != total_files: + raise ValueError( + f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. " + "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes." + ) + + # the groups are named differently in each stage + if zero_stage <= 2: + fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS + elif zero_stage == 3: + fp32_groups_key = FP32_FLAT_GROUPS + else: + raise ValueError(f"unknown zero stage {zero_stage}") + + if zero_stage <= 2: + fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))] + elif zero_stage == 3: + # if there is more than one param group, there will be multiple flattened tensors - one + # flattened tensor per group - for simplicity merge them into a single tensor + # + # XXX: could make the script more memory efficient for when there are multiple groups - it + # will require matching the sub-lists of param_shapes for each param group flattened tensor + + fp32_flat_groups = [ + torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts)) + ] + + return zero_stage, world_size, fp32_flat_groups + + +def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters): + """ + Returns fp32 state_dict reconstructed from ds checkpoint + + Args: + - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are) + + """ + print(f"Processing zero checkpoint '{ds_checkpoint_dir}'") + + optim_files = get_optim_files(ds_checkpoint_dir) + zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir) + print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}") + + model_files = get_model_state_files(ds_checkpoint_dir) + + zero_model_states = parse_model_states(model_files) + print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}') + + if zero_stage <= 2: + return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + elif zero_stage == 3: + return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + + +def _zero2_merge_frozen_params(state_dict, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + frozen_param_fragments = zero_model_states[0].frozen_param_fragments + + if debug: + num_elem = sum(s.numel() for s in frozen_param_shapes.values()) + print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in frozen_param_fragments.values()]) + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + state_dict[name] = frozen_param_fragments[name] + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _has_callable(obj, fn): + attr = getattr(obj, fn, None) + return callable(attr) + + +def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + + # Reconstruction protocol: + # + # XXX: document this + + if debug: + for i in range(world_size): + for j in range(len(fp32_flat_groups[0])): + print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}") + + # XXX: memory usage doubles here (zero2) + num_param_groups = len(fp32_flat_groups[0]) + merged_single_partition_of_fp32_groups = [] + for i in range(num_param_groups): + merged_partitions = [sd[i] for sd in fp32_flat_groups] + full_single_fp32_vector = torch.cat(merged_partitions, 0) + merged_single_partition_of_fp32_groups.append(full_single_fp32_vector) + avail_numel = sum( + [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups]) + + if debug: + wanted_params = sum([len(shapes) for shapes in param_shapes]) + wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes]) + # not asserting if there is a mismatch due to possible padding + print(f"Have {avail_numel} numels to process.") + print(f"Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + total_numel = 0 + total_params = 0 + for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups): + offset = 0 + avail_numel = full_single_fp32_vector.numel() + for name, shape in shapes.items(): + + unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape) + total_numel += unpartitioned_numel + total_params += 1 + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape) + offset += unpartitioned_numel + + # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and + # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex + # paddings performed in the code it's almost impossible to predict the exact numbers w/o the + # live optimizer object, so we are checking that the numbers are within the right range + align_to = 2 * world_size + + def zero2_align(x): + return align_to * math.ceil(x / align_to) + + if debug: + print(f"original offset={offset}, avail_numel={avail_numel}") + + offset = zero2_align(offset) + avail_numel = zero2_align(avail_numel) + + if debug: + print(f"aligned offset={offset}, avail_numel={avail_numel}") + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero2_merge_frozen_params(state_dict, zero_model_states) + + _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def zero3_partitioned_param_info(unpartitioned_numel, world_size): + remainder = unpartitioned_numel % world_size + padding_numel = (world_size - remainder) if remainder else 0 + partitioned_numel = math.ceil(unpartitioned_numel / world_size) + return partitioned_numel, padding_numel + + +def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + if debug: + for i in range(world_size): + num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values()) + print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in zero_model_states[0].frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states) + state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape) + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + avail_numel = fp32_flat_groups[0].numel() * world_size + # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each + # param, re-consolidating each param, while dealing with padding if any + + # merge list of dicts, preserving order + param_shapes = {k: v for d in param_shapes for k, v in d.items()} + + if debug: + for i in range(world_size): + print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}") + + wanted_params = len(param_shapes) + wanted_numel = sum(shape.numel() for shape in param_shapes.values()) + # not asserting if there is a mismatch due to possible padding + avail_numel = fp32_flat_groups[0].numel() * world_size + print(f"Trainable params: Have {avail_numel} numels to process.") + print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + offset = 0 + total_numel = 0 + total_params = 0 + for name, shape in param_shapes.items(): + + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + total_params += 1 + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + # XXX: memory usage doubles here + state_dict[name] = torch.cat( + tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)), + 0).narrow(0, 0, unpartitioned_numel).view(shape) + offset += partitioned_numel + + offset *= world_size + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero3_merge_frozen_params(state_dict, world_size, zero_model_states) + + _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with + ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example + via a model hub. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + + Returns: + - pytorch ``state_dict`` + + Note: this approach may not work if your application doesn't have sufficient free CPU memory and + you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with + the checkpoint. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint + # do the training and checkpoint saving + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu + model = model.cpu() # move to cpu + model.load_state_dict(state_dict) + # submit to model hub or save the model to share with others + + In this example the ``model`` will no longer be usable in the deepspeed context of the same + application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead. + + """ + if tag is None: + latest_path = os.path.join(checkpoint_dir, 'latest') + if os.path.isfile(latest_path): + with open(latest_path, 'r') as fd: + tag = fd.read().strip() + else: + raise ValueError(f"Unable to find 'latest' file at {latest_path}") + + ds_checkpoint_dir = os.path.join(checkpoint_dir, tag) + + if not os.path.isdir(ds_checkpoint_dir): + raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist") + + return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters) + + +def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be + loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + """ + + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters) + print(f"Saving fp32 state dict to {output_file}") + torch.save(state_dict, output_file) + + +def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None): + """ + 1. Put the provided model to cpu + 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` + 3. Load it into the provided model + + Args: + - ``model``: the model object to update + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + + Returns: + - ``model`: modified model + + Make sure you have plenty of CPU memory available before you call this function. If you don't + have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it + conveniently placed for you in the checkpoint folder. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint + model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) + # submit to model hub or save the model to share with others + + Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context + of the same application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + """ + logger.info(f"Extracting fp32 weights") + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag) + + logger.info(f"Overwriting model with fp32 weights") + model = model.cpu() + model.load_state_dict(state_dict, strict=False) + + return model + + +if __name__ == "__main__": + + parser = argparse.ArgumentParser() + parser.add_argument("checkpoint_dir", + type=str, + help="path to the desired checkpoint folder, e.g., path/checkpoint-12") + parser.add_argument( + "output_file", + type=str, + help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)") + parser.add_argument("-t", + "--tag", + type=str, + default=None, + help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1") + parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters") + parser.add_argument("-d", "--debug", action='store_true', help="enable debug") + args = parser.parse_args() + + debug = args.debug + + convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, + args.output_file, + tag=args.tag, + exclude_frozen_parameters=args.exclude_frozen_parameters) diff --git a/sft/smoe_vcr/checkpoint-16632/latest b/sft/smoe_vcr/checkpoint-16632/latest new file mode 100644 index 0000000000000000000000000000000000000000..e4087b037c4d90a88f08b57160ddc65e74a0c271 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/latest @@ -0,0 +1 @@ +global_step16632 \ No newline at end of file diff --git a/sft/smoe_vcr/checkpoint-16632/model-00001-of-00002.safetensors b/sft/smoe_vcr/checkpoint-16632/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a2e039e1186da0f5a2cb583c8aef3d2973e65d01 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b980095539e1791498907a3250d8c638d9f7efb6c942efaa1efaab69e6631208 +size 4972489328 diff --git a/sft/smoe_vcr/checkpoint-16632/model-00002-of-00002.safetensors b/sft/smoe_vcr/checkpoint-16632/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..332e627d0fccd3f23ca5f6d0e149534c99da3bee --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da33b82157a6d9d3f145750aa1ffe39fa400f5186ca3e89728d628e59816cda0 +size 3759015720 diff --git a/sft/smoe_vcr/checkpoint-16632/rng_state_0.pth b/sft/smoe_vcr/checkpoint-16632/rng_state_0.pth new file mode 100644 index 0000000000000000000000000000000000000000..9231f69f5fd461899867106a669ce247e70c72c2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/rng_state_0.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f23d807f0e704f4ca79670a6631cbff43189cf7f8ff4e1fc0a4330e636a798 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-16632/rng_state_1.pth b/sft/smoe_vcr/checkpoint-16632/rng_state_1.pth new file mode 100644 index 0000000000000000000000000000000000000000..19fe2dcc766f192ea5de79cec4dcff17172a10f7 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/rng_state_1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d37f92f6aea5386e84d2d64a1a25d6ef96a10b3bbbfe63627981604c8934076 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-16632/rng_state_2.pth b/sft/smoe_vcr/checkpoint-16632/rng_state_2.pth new file mode 100644 index 0000000000000000000000000000000000000000..bfe492519c6b79b07a8d68b98c5f3d0c073667aa --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/rng_state_2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667ebf727735115f00a6bdbe090344e9846c726d11bb555cdc201c415f27ad85 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-16632/rng_state_3.pth b/sft/smoe_vcr/checkpoint-16632/rng_state_3.pth new file mode 100644 index 0000000000000000000000000000000000000000..838d42ad13e30851fdbd1d8801738a4106a9ce8b --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/rng_state_3.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d306f8c511cba8a225e3b723c5fa79d8a6ecc922f834da914ff0780c78b1fc +size 14960 diff --git a/sft/smoe_vcr/checkpoint-16632/tokenizer.model b/sft/smoe_vcr/checkpoint-16632/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/checkpoint-16632/training_args.bin b/sft/smoe_vcr/checkpoint-16632/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992 diff --git a/sft/smoe_vcr/checkpoint-16632/zero_to_fp32.py b/sft/smoe_vcr/checkpoint-16632/zero_to_fp32.py new file mode 100644 index 0000000000000000000000000000000000000000..24cc342e78d1a006c782b3a4cd68d9ce786d8fd8 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-16632/zero_to_fp32.py @@ -0,0 +1,604 @@ +#!/usr/bin/env python + +# Copyright (c) Microsoft Corporation. +# SPDX-License-Identifier: Apache-2.0 + +# DeepSpeed Team + +# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets +# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in +# the future. Once extracted, the weights don't require DeepSpeed and can be used in any +# application. +# +# example: python zero_to_fp32.py . pytorch_model.bin + +import argparse +import torch +import glob +import math +import os +import re +from collections import OrderedDict +from dataclasses import dataclass + +# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with +# DeepSpeed data structures it has to be available in the current python environment. +from deepspeed.utils import logger +from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS, + FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES, + FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS) + + +@dataclass +class zero_model_state: + buffers: dict() + param_shapes: dict() + shared_params: list + ds_version: int + frozen_param_shapes: dict() + frozen_param_fragments: dict() + + +debug = 0 + +# load to cpu +device = torch.device('cpu') + + +def atoi(text): + return int(text) if text.isdigit() else text + + +def natural_keys(text): + ''' + alist.sort(key=natural_keys) sorts in human order + http://nedbatchelder.com/blog/200712/human_sorting.html + (See Toothy's implementation in the comments) + ''' + return [atoi(c) for c in re.split(r'(\d+)', text)] + + +def get_model_state_file(checkpoint_dir, zero_stage): + if not os.path.isdir(checkpoint_dir): + raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist") + + # there should be only one file + if zero_stage <= 2: + file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt") + elif zero_stage == 3: + file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt") + + if not os.path.exists(file): + raise FileNotFoundError(f"can't find model states file at '{file}'") + + return file + + +def get_checkpoint_files(checkpoint_dir, glob_pattern): + # XXX: need to test that this simple glob rule works for multi-node setup too + ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys) + + if len(ckpt_files) == 0: + raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'") + + return ckpt_files + + +def get_optim_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt") + + +def get_model_state_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_model_states.pt") + + +def parse_model_states(files): + zero_model_states = [] + for file in files: + state_dict = torch.load(file, map_location=device) + + if BUFFER_NAMES not in state_dict: + raise ValueError(f"{file} is not a model state checkpoint") + buffer_names = state_dict[BUFFER_NAMES] + if debug: + print("Found buffers:", buffer_names) + + # recover just the buffers while restoring them to fp32 if they were saved in fp16 + buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names} + param_shapes = state_dict[PARAM_SHAPES] + + # collect parameters that are included in param_shapes + param_names = [] + for s in param_shapes: + for name in s.keys(): + param_names.append(name) + + # update with frozen parameters + frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None) + if frozen_param_shapes is not None: + if debug: + print(f"Found frozen_param_shapes: {frozen_param_shapes}") + param_names += list(frozen_param_shapes.keys()) + + # handle shared params + shared_params = [[k, v] for k, v in state_dict["shared_params"].items()] + + ds_version = state_dict.get(DS_VERSION, None) + + frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None) + + z_model_state = zero_model_state(buffers=buffers, + param_shapes=param_shapes, + shared_params=shared_params, + ds_version=ds_version, + frozen_param_shapes=frozen_param_shapes, + frozen_param_fragments=frozen_param_fragments) + zero_model_states.append(z_model_state) + + return zero_model_states + + +def parse_optim_states(files, ds_checkpoint_dir): + + total_files = len(files) + state_dicts = [] + for f in files: + state_dict = torch.load(f, map_location=device) + # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights + # and also handle the case where it was already removed by another helper script + state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None) + state_dicts.append(state_dict) + + if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]: + raise ValueError(f"{files[0]} is not a zero checkpoint") + zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE] + world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT] + + # For ZeRO-2 each param group can have different partition_count as data parallelism for expert + # parameters can be different from data parallelism for non-expert parameters. So we can just + # use the max of the partition_count to get the dp world_size. + + if type(world_size) is list: + world_size = max(world_size) + + if world_size != total_files: + raise ValueError( + f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. " + "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes." + ) + + # the groups are named differently in each stage + if zero_stage <= 2: + fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS + elif zero_stage == 3: + fp32_groups_key = FP32_FLAT_GROUPS + else: + raise ValueError(f"unknown zero stage {zero_stage}") + + if zero_stage <= 2: + fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))] + elif zero_stage == 3: + # if there is more than one param group, there will be multiple flattened tensors - one + # flattened tensor per group - for simplicity merge them into a single tensor + # + # XXX: could make the script more memory efficient for when there are multiple groups - it + # will require matching the sub-lists of param_shapes for each param group flattened tensor + + fp32_flat_groups = [ + torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts)) + ] + + return zero_stage, world_size, fp32_flat_groups + + +def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters): + """ + Returns fp32 state_dict reconstructed from ds checkpoint + + Args: + - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are) + + """ + print(f"Processing zero checkpoint '{ds_checkpoint_dir}'") + + optim_files = get_optim_files(ds_checkpoint_dir) + zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir) + print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}") + + model_files = get_model_state_files(ds_checkpoint_dir) + + zero_model_states = parse_model_states(model_files) + print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}') + + if zero_stage <= 2: + return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + elif zero_stage == 3: + return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + + +def _zero2_merge_frozen_params(state_dict, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + frozen_param_fragments = zero_model_states[0].frozen_param_fragments + + if debug: + num_elem = sum(s.numel() for s in frozen_param_shapes.values()) + print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in frozen_param_fragments.values()]) + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + state_dict[name] = frozen_param_fragments[name] + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _has_callable(obj, fn): + attr = getattr(obj, fn, None) + return callable(attr) + + +def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + + # Reconstruction protocol: + # + # XXX: document this + + if debug: + for i in range(world_size): + for j in range(len(fp32_flat_groups[0])): + print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}") + + # XXX: memory usage doubles here (zero2) + num_param_groups = len(fp32_flat_groups[0]) + merged_single_partition_of_fp32_groups = [] + for i in range(num_param_groups): + merged_partitions = [sd[i] for sd in fp32_flat_groups] + full_single_fp32_vector = torch.cat(merged_partitions, 0) + merged_single_partition_of_fp32_groups.append(full_single_fp32_vector) + avail_numel = sum( + [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups]) + + if debug: + wanted_params = sum([len(shapes) for shapes in param_shapes]) + wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes]) + # not asserting if there is a mismatch due to possible padding + print(f"Have {avail_numel} numels to process.") + print(f"Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + total_numel = 0 + total_params = 0 + for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups): + offset = 0 + avail_numel = full_single_fp32_vector.numel() + for name, shape in shapes.items(): + + unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape) + total_numel += unpartitioned_numel + total_params += 1 + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape) + offset += unpartitioned_numel + + # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and + # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex + # paddings performed in the code it's almost impossible to predict the exact numbers w/o the + # live optimizer object, so we are checking that the numbers are within the right range + align_to = 2 * world_size + + def zero2_align(x): + return align_to * math.ceil(x / align_to) + + if debug: + print(f"original offset={offset}, avail_numel={avail_numel}") + + offset = zero2_align(offset) + avail_numel = zero2_align(avail_numel) + + if debug: + print(f"aligned offset={offset}, avail_numel={avail_numel}") + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero2_merge_frozen_params(state_dict, zero_model_states) + + _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def zero3_partitioned_param_info(unpartitioned_numel, world_size): + remainder = unpartitioned_numel % world_size + padding_numel = (world_size - remainder) if remainder else 0 + partitioned_numel = math.ceil(unpartitioned_numel / world_size) + return partitioned_numel, padding_numel + + +def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + if debug: + for i in range(world_size): + num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values()) + print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in zero_model_states[0].frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states) + state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape) + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + avail_numel = fp32_flat_groups[0].numel() * world_size + # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each + # param, re-consolidating each param, while dealing with padding if any + + # merge list of dicts, preserving order + param_shapes = {k: v for d in param_shapes for k, v in d.items()} + + if debug: + for i in range(world_size): + print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}") + + wanted_params = len(param_shapes) + wanted_numel = sum(shape.numel() for shape in param_shapes.values()) + # not asserting if there is a mismatch due to possible padding + avail_numel = fp32_flat_groups[0].numel() * world_size + print(f"Trainable params: Have {avail_numel} numels to process.") + print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + offset = 0 + total_numel = 0 + total_params = 0 + for name, shape in param_shapes.items(): + + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + total_params += 1 + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + # XXX: memory usage doubles here + state_dict[name] = torch.cat( + tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)), + 0).narrow(0, 0, unpartitioned_numel).view(shape) + offset += partitioned_numel + + offset *= world_size + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero3_merge_frozen_params(state_dict, world_size, zero_model_states) + + _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with + ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example + via a model hub. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + + Returns: + - pytorch ``state_dict`` + + Note: this approach may not work if your application doesn't have sufficient free CPU memory and + you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with + the checkpoint. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint + # do the training and checkpoint saving + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu + model = model.cpu() # move to cpu + model.load_state_dict(state_dict) + # submit to model hub or save the model to share with others + + In this example the ``model`` will no longer be usable in the deepspeed context of the same + application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead. + + """ + if tag is None: + latest_path = os.path.join(checkpoint_dir, 'latest') + if os.path.isfile(latest_path): + with open(latest_path, 'r') as fd: + tag = fd.read().strip() + else: + raise ValueError(f"Unable to find 'latest' file at {latest_path}") + + ds_checkpoint_dir = os.path.join(checkpoint_dir, tag) + + if not os.path.isdir(ds_checkpoint_dir): + raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist") + + return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters) + + +def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be + loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + """ + + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters) + print(f"Saving fp32 state dict to {output_file}") + torch.save(state_dict, output_file) + + +def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None): + """ + 1. Put the provided model to cpu + 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` + 3. Load it into the provided model + + Args: + - ``model``: the model object to update + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + + Returns: + - ``model`: modified model + + Make sure you have plenty of CPU memory available before you call this function. If you don't + have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it + conveniently placed for you in the checkpoint folder. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint + model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) + # submit to model hub or save the model to share with others + + Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context + of the same application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + """ + logger.info(f"Extracting fp32 weights") + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag) + + logger.info(f"Overwriting model with fp32 weights") + model = model.cpu() + model.load_state_dict(state_dict, strict=False) + + return model + + +if __name__ == "__main__": + + parser = argparse.ArgumentParser() + parser.add_argument("checkpoint_dir", + type=str, + help="path to the desired checkpoint folder, e.g., path/checkpoint-12") + parser.add_argument( + "output_file", + type=str, + help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)") + parser.add_argument("-t", + "--tag", + type=str, + default=None, + help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1") + parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters") + parser.add_argument("-d", "--debug", action='store_true', help="enable debug") + args = parser.parse_args() + + debug = args.debug + + convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, + args.output_file, + tag=args.tag, + exclude_frozen_parameters=args.exclude_frozen_parameters) diff --git a/sft/smoe_vcr/checkpoint-3328/latest b/sft/smoe_vcr/checkpoint-3328/latest new file mode 100644 index 0000000000000000000000000000000000000000..2c27d5aabecd1a20f5d8e01a05251ed2cf0a7fec --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/latest @@ -0,0 +1 @@ +global_step3328 \ No newline at end of file diff --git a/sft/smoe_vcr/checkpoint-3328/model-00001-of-00002.safetensors b/sft/smoe_vcr/checkpoint-3328/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d65469b9b173abc2e0219205d2769e34df1f9475 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cd3580f5ef232bb8b4dbb925ef4d70dabfd796ce8ee29219ef113aac8f8799a +size 4972489328 diff --git a/sft/smoe_vcr/checkpoint-3328/model-00002-of-00002.safetensors b/sft/smoe_vcr/checkpoint-3328/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..681d0e2b7da0715392acfb3b7e563729dbb26080 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ecdaaf1be9653d016d6ab7aabeb7615950b7c0f5c5f958421ee21255e0140483 +size 3759015720 diff --git a/sft/smoe_vcr/checkpoint-3328/rng_state_0.pth b/sft/smoe_vcr/checkpoint-3328/rng_state_0.pth new file mode 100644 index 0000000000000000000000000000000000000000..9231f69f5fd461899867106a669ce247e70c72c2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/rng_state_0.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f23d807f0e704f4ca79670a6631cbff43189cf7f8ff4e1fc0a4330e636a798 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-3328/rng_state_1.pth b/sft/smoe_vcr/checkpoint-3328/rng_state_1.pth new file mode 100644 index 0000000000000000000000000000000000000000..19fe2dcc766f192ea5de79cec4dcff17172a10f7 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/rng_state_1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d37f92f6aea5386e84d2d64a1a25d6ef96a10b3bbbfe63627981604c8934076 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-3328/rng_state_2.pth b/sft/smoe_vcr/checkpoint-3328/rng_state_2.pth new file mode 100644 index 0000000000000000000000000000000000000000..bfe492519c6b79b07a8d68b98c5f3d0c073667aa --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/rng_state_2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667ebf727735115f00a6bdbe090344e9846c726d11bb555cdc201c415f27ad85 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-3328/rng_state_3.pth b/sft/smoe_vcr/checkpoint-3328/rng_state_3.pth new file mode 100644 index 0000000000000000000000000000000000000000..838d42ad13e30851fdbd1d8801738a4106a9ce8b --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/rng_state_3.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d306f8c511cba8a225e3b723c5fa79d8a6ecc922f834da914ff0780c78b1fc +size 14960 diff --git a/sft/smoe_vcr/checkpoint-3328/tokenizer.model b/sft/smoe_vcr/checkpoint-3328/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/checkpoint-3328/training_args.bin b/sft/smoe_vcr/checkpoint-3328/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992 diff --git a/sft/smoe_vcr/checkpoint-3328/zero_to_fp32.py b/sft/smoe_vcr/checkpoint-3328/zero_to_fp32.py new file mode 100644 index 0000000000000000000000000000000000000000..24cc342e78d1a006c782b3a4cd68d9ce786d8fd8 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-3328/zero_to_fp32.py @@ -0,0 +1,604 @@ +#!/usr/bin/env python + +# Copyright (c) Microsoft Corporation. +# SPDX-License-Identifier: Apache-2.0 + +# DeepSpeed Team + +# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets +# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in +# the future. Once extracted, the weights don't require DeepSpeed and can be used in any +# application. +# +# example: python zero_to_fp32.py . pytorch_model.bin + +import argparse +import torch +import glob +import math +import os +import re +from collections import OrderedDict +from dataclasses import dataclass + +# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with +# DeepSpeed data structures it has to be available in the current python environment. +from deepspeed.utils import logger +from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS, + FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES, + FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS) + + +@dataclass +class zero_model_state: + buffers: dict() + param_shapes: dict() + shared_params: list + ds_version: int + frozen_param_shapes: dict() + frozen_param_fragments: dict() + + +debug = 0 + +# load to cpu +device = torch.device('cpu') + + +def atoi(text): + return int(text) if text.isdigit() else text + + +def natural_keys(text): + ''' + alist.sort(key=natural_keys) sorts in human order + http://nedbatchelder.com/blog/200712/human_sorting.html + (See Toothy's implementation in the comments) + ''' + return [atoi(c) for c in re.split(r'(\d+)', text)] + + +def get_model_state_file(checkpoint_dir, zero_stage): + if not os.path.isdir(checkpoint_dir): + raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist") + + # there should be only one file + if zero_stage <= 2: + file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt") + elif zero_stage == 3: + file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt") + + if not os.path.exists(file): + raise FileNotFoundError(f"can't find model states file at '{file}'") + + return file + + +def get_checkpoint_files(checkpoint_dir, glob_pattern): + # XXX: need to test that this simple glob rule works for multi-node setup too + ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys) + + if len(ckpt_files) == 0: + raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'") + + return ckpt_files + + +def get_optim_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt") + + +def get_model_state_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_model_states.pt") + + +def parse_model_states(files): + zero_model_states = [] + for file in files: + state_dict = torch.load(file, map_location=device) + + if BUFFER_NAMES not in state_dict: + raise ValueError(f"{file} is not a model state checkpoint") + buffer_names = state_dict[BUFFER_NAMES] + if debug: + print("Found buffers:", buffer_names) + + # recover just the buffers while restoring them to fp32 if they were saved in fp16 + buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names} + param_shapes = state_dict[PARAM_SHAPES] + + # collect parameters that are included in param_shapes + param_names = [] + for s in param_shapes: + for name in s.keys(): + param_names.append(name) + + # update with frozen parameters + frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None) + if frozen_param_shapes is not None: + if debug: + print(f"Found frozen_param_shapes: {frozen_param_shapes}") + param_names += list(frozen_param_shapes.keys()) + + # handle shared params + shared_params = [[k, v] for k, v in state_dict["shared_params"].items()] + + ds_version = state_dict.get(DS_VERSION, None) + + frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None) + + z_model_state = zero_model_state(buffers=buffers, + param_shapes=param_shapes, + shared_params=shared_params, + ds_version=ds_version, + frozen_param_shapes=frozen_param_shapes, + frozen_param_fragments=frozen_param_fragments) + zero_model_states.append(z_model_state) + + return zero_model_states + + +def parse_optim_states(files, ds_checkpoint_dir): + + total_files = len(files) + state_dicts = [] + for f in files: + state_dict = torch.load(f, map_location=device) + # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights + # and also handle the case where it was already removed by another helper script + state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None) + state_dicts.append(state_dict) + + if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]: + raise ValueError(f"{files[0]} is not a zero checkpoint") + zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE] + world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT] + + # For ZeRO-2 each param group can have different partition_count as data parallelism for expert + # parameters can be different from data parallelism for non-expert parameters. So we can just + # use the max of the partition_count to get the dp world_size. + + if type(world_size) is list: + world_size = max(world_size) + + if world_size != total_files: + raise ValueError( + f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. " + "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes." + ) + + # the groups are named differently in each stage + if zero_stage <= 2: + fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS + elif zero_stage == 3: + fp32_groups_key = FP32_FLAT_GROUPS + else: + raise ValueError(f"unknown zero stage {zero_stage}") + + if zero_stage <= 2: + fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))] + elif zero_stage == 3: + # if there is more than one param group, there will be multiple flattened tensors - one + # flattened tensor per group - for simplicity merge them into a single tensor + # + # XXX: could make the script more memory efficient for when there are multiple groups - it + # will require matching the sub-lists of param_shapes for each param group flattened tensor + + fp32_flat_groups = [ + torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts)) + ] + + return zero_stage, world_size, fp32_flat_groups + + +def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters): + """ + Returns fp32 state_dict reconstructed from ds checkpoint + + Args: + - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are) + + """ + print(f"Processing zero checkpoint '{ds_checkpoint_dir}'") + + optim_files = get_optim_files(ds_checkpoint_dir) + zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir) + print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}") + + model_files = get_model_state_files(ds_checkpoint_dir) + + zero_model_states = parse_model_states(model_files) + print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}') + + if zero_stage <= 2: + return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + elif zero_stage == 3: + return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + + +def _zero2_merge_frozen_params(state_dict, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + frozen_param_fragments = zero_model_states[0].frozen_param_fragments + + if debug: + num_elem = sum(s.numel() for s in frozen_param_shapes.values()) + print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in frozen_param_fragments.values()]) + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + state_dict[name] = frozen_param_fragments[name] + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _has_callable(obj, fn): + attr = getattr(obj, fn, None) + return callable(attr) + + +def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + + # Reconstruction protocol: + # + # XXX: document this + + if debug: + for i in range(world_size): + for j in range(len(fp32_flat_groups[0])): + print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}") + + # XXX: memory usage doubles here (zero2) + num_param_groups = len(fp32_flat_groups[0]) + merged_single_partition_of_fp32_groups = [] + for i in range(num_param_groups): + merged_partitions = [sd[i] for sd in fp32_flat_groups] + full_single_fp32_vector = torch.cat(merged_partitions, 0) + merged_single_partition_of_fp32_groups.append(full_single_fp32_vector) + avail_numel = sum( + [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups]) + + if debug: + wanted_params = sum([len(shapes) for shapes in param_shapes]) + wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes]) + # not asserting if there is a mismatch due to possible padding + print(f"Have {avail_numel} numels to process.") + print(f"Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + total_numel = 0 + total_params = 0 + for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups): + offset = 0 + avail_numel = full_single_fp32_vector.numel() + for name, shape in shapes.items(): + + unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape) + total_numel += unpartitioned_numel + total_params += 1 + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape) + offset += unpartitioned_numel + + # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and + # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex + # paddings performed in the code it's almost impossible to predict the exact numbers w/o the + # live optimizer object, so we are checking that the numbers are within the right range + align_to = 2 * world_size + + def zero2_align(x): + return align_to * math.ceil(x / align_to) + + if debug: + print(f"original offset={offset}, avail_numel={avail_numel}") + + offset = zero2_align(offset) + avail_numel = zero2_align(avail_numel) + + if debug: + print(f"aligned offset={offset}, avail_numel={avail_numel}") + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero2_merge_frozen_params(state_dict, zero_model_states) + + _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def zero3_partitioned_param_info(unpartitioned_numel, world_size): + remainder = unpartitioned_numel % world_size + padding_numel = (world_size - remainder) if remainder else 0 + partitioned_numel = math.ceil(unpartitioned_numel / world_size) + return partitioned_numel, padding_numel + + +def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + if debug: + for i in range(world_size): + num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values()) + print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in zero_model_states[0].frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states) + state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape) + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + avail_numel = fp32_flat_groups[0].numel() * world_size + # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each + # param, re-consolidating each param, while dealing with padding if any + + # merge list of dicts, preserving order + param_shapes = {k: v for d in param_shapes for k, v in d.items()} + + if debug: + for i in range(world_size): + print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}") + + wanted_params = len(param_shapes) + wanted_numel = sum(shape.numel() for shape in param_shapes.values()) + # not asserting if there is a mismatch due to possible padding + avail_numel = fp32_flat_groups[0].numel() * world_size + print(f"Trainable params: Have {avail_numel} numels to process.") + print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + offset = 0 + total_numel = 0 + total_params = 0 + for name, shape in param_shapes.items(): + + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + total_params += 1 + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + # XXX: memory usage doubles here + state_dict[name] = torch.cat( + tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)), + 0).narrow(0, 0, unpartitioned_numel).view(shape) + offset += partitioned_numel + + offset *= world_size + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero3_merge_frozen_params(state_dict, world_size, zero_model_states) + + _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with + ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example + via a model hub. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + + Returns: + - pytorch ``state_dict`` + + Note: this approach may not work if your application doesn't have sufficient free CPU memory and + you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with + the checkpoint. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint + # do the training and checkpoint saving + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu + model = model.cpu() # move to cpu + model.load_state_dict(state_dict) + # submit to model hub or save the model to share with others + + In this example the ``model`` will no longer be usable in the deepspeed context of the same + application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead. + + """ + if tag is None: + latest_path = os.path.join(checkpoint_dir, 'latest') + if os.path.isfile(latest_path): + with open(latest_path, 'r') as fd: + tag = fd.read().strip() + else: + raise ValueError(f"Unable to find 'latest' file at {latest_path}") + + ds_checkpoint_dir = os.path.join(checkpoint_dir, tag) + + if not os.path.isdir(ds_checkpoint_dir): + raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist") + + return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters) + + +def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be + loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + """ + + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters) + print(f"Saving fp32 state dict to {output_file}") + torch.save(state_dict, output_file) + + +def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None): + """ + 1. Put the provided model to cpu + 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` + 3. Load it into the provided model + + Args: + - ``model``: the model object to update + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + + Returns: + - ``model`: modified model + + Make sure you have plenty of CPU memory available before you call this function. If you don't + have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it + conveniently placed for you in the checkpoint folder. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint + model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) + # submit to model hub or save the model to share with others + + Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context + of the same application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + """ + logger.info(f"Extracting fp32 weights") + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag) + + logger.info(f"Overwriting model with fp32 weights") + model = model.cpu() + model.load_state_dict(state_dict, strict=False) + + return model + + +if __name__ == "__main__": + + parser = argparse.ArgumentParser() + parser.add_argument("checkpoint_dir", + type=str, + help="path to the desired checkpoint folder, e.g., path/checkpoint-12") + parser.add_argument( + "output_file", + type=str, + help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)") + parser.add_argument("-t", + "--tag", + type=str, + default=None, + help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1") + parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters") + parser.add_argument("-d", "--debug", action='store_true', help="enable debug") + args = parser.parse_args() + + debug = args.debug + + convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, + args.output_file, + tag=args.tag, + exclude_frozen_parameters=args.exclude_frozen_parameters) diff --git a/sft/smoe_vcr/checkpoint-6656/latest b/sft/smoe_vcr/checkpoint-6656/latest new file mode 100644 index 0000000000000000000000000000000000000000..24f37f789c4e6eb86270647db8ff45788e484aa2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/latest @@ -0,0 +1 @@ +global_step6656 \ No newline at end of file diff --git a/sft/smoe_vcr/checkpoint-6656/model-00001-of-00002.safetensors b/sft/smoe_vcr/checkpoint-6656/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7cfbba86c1e8058983bf1495bcc852e92b4bbaef --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d89657f3f9f5a138987ff2a2ad3b59b04956a660f4b56c50ff2af692c1d5a62d +size 4972489328 diff --git a/sft/smoe_vcr/checkpoint-6656/model-00002-of-00002.safetensors b/sft/smoe_vcr/checkpoint-6656/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f9bcc6d0ef875e46a01807424c7de1b8cb9e2c81 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b77dcb153143b1e8e406b442235536ae4600ce71ba52a5be3a61e751b21d547 +size 3759015720 diff --git a/sft/smoe_vcr/checkpoint-6656/rng_state_0.pth b/sft/smoe_vcr/checkpoint-6656/rng_state_0.pth new file mode 100644 index 0000000000000000000000000000000000000000..9231f69f5fd461899867106a669ce247e70c72c2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/rng_state_0.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f23d807f0e704f4ca79670a6631cbff43189cf7f8ff4e1fc0a4330e636a798 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-6656/rng_state_1.pth b/sft/smoe_vcr/checkpoint-6656/rng_state_1.pth new file mode 100644 index 0000000000000000000000000000000000000000..19fe2dcc766f192ea5de79cec4dcff17172a10f7 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/rng_state_1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d37f92f6aea5386e84d2d64a1a25d6ef96a10b3bbbfe63627981604c8934076 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-6656/rng_state_2.pth b/sft/smoe_vcr/checkpoint-6656/rng_state_2.pth new file mode 100644 index 0000000000000000000000000000000000000000..bfe492519c6b79b07a8d68b98c5f3d0c073667aa --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/rng_state_2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667ebf727735115f00a6bdbe090344e9846c726d11bb555cdc201c415f27ad85 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-6656/rng_state_3.pth b/sft/smoe_vcr/checkpoint-6656/rng_state_3.pth new file mode 100644 index 0000000000000000000000000000000000000000..838d42ad13e30851fdbd1d8801738a4106a9ce8b --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/rng_state_3.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d306f8c511cba8a225e3b723c5fa79d8a6ecc922f834da914ff0780c78b1fc +size 14960 diff --git a/sft/smoe_vcr/checkpoint-6656/tokenizer.model b/sft/smoe_vcr/checkpoint-6656/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/checkpoint-6656/training_args.bin b/sft/smoe_vcr/checkpoint-6656/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992 diff --git a/sft/smoe_vcr/checkpoint-6656/zero_to_fp32.py b/sft/smoe_vcr/checkpoint-6656/zero_to_fp32.py new file mode 100644 index 0000000000000000000000000000000000000000..24cc342e78d1a006c782b3a4cd68d9ce786d8fd8 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-6656/zero_to_fp32.py @@ -0,0 +1,604 @@ +#!/usr/bin/env python + +# Copyright (c) Microsoft Corporation. +# SPDX-License-Identifier: Apache-2.0 + +# DeepSpeed Team + +# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets +# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in +# the future. Once extracted, the weights don't require DeepSpeed and can be used in any +# application. +# +# example: python zero_to_fp32.py . pytorch_model.bin + +import argparse +import torch +import glob +import math +import os +import re +from collections import OrderedDict +from dataclasses import dataclass + +# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with +# DeepSpeed data structures it has to be available in the current python environment. +from deepspeed.utils import logger +from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS, + FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES, + FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS) + + +@dataclass +class zero_model_state: + buffers: dict() + param_shapes: dict() + shared_params: list + ds_version: int + frozen_param_shapes: dict() + frozen_param_fragments: dict() + + +debug = 0 + +# load to cpu +device = torch.device('cpu') + + +def atoi(text): + return int(text) if text.isdigit() else text + + +def natural_keys(text): + ''' + alist.sort(key=natural_keys) sorts in human order + http://nedbatchelder.com/blog/200712/human_sorting.html + (See Toothy's implementation in the comments) + ''' + return [atoi(c) for c in re.split(r'(\d+)', text)] + + +def get_model_state_file(checkpoint_dir, zero_stage): + if not os.path.isdir(checkpoint_dir): + raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist") + + # there should be only one file + if zero_stage <= 2: + file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt") + elif zero_stage == 3: + file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt") + + if not os.path.exists(file): + raise FileNotFoundError(f"can't find model states file at '{file}'") + + return file + + +def get_checkpoint_files(checkpoint_dir, glob_pattern): + # XXX: need to test that this simple glob rule works for multi-node setup too + ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys) + + if len(ckpt_files) == 0: + raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'") + + return ckpt_files + + +def get_optim_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt") + + +def get_model_state_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_model_states.pt") + + +def parse_model_states(files): + zero_model_states = [] + for file in files: + state_dict = torch.load(file, map_location=device) + + if BUFFER_NAMES not in state_dict: + raise ValueError(f"{file} is not a model state checkpoint") + buffer_names = state_dict[BUFFER_NAMES] + if debug: + print("Found buffers:", buffer_names) + + # recover just the buffers while restoring them to fp32 if they were saved in fp16 + buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names} + param_shapes = state_dict[PARAM_SHAPES] + + # collect parameters that are included in param_shapes + param_names = [] + for s in param_shapes: + for name in s.keys(): + param_names.append(name) + + # update with frozen parameters + frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None) + if frozen_param_shapes is not None: + if debug: + print(f"Found frozen_param_shapes: {frozen_param_shapes}") + param_names += list(frozen_param_shapes.keys()) + + # handle shared params + shared_params = [[k, v] for k, v in state_dict["shared_params"].items()] + + ds_version = state_dict.get(DS_VERSION, None) + + frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None) + + z_model_state = zero_model_state(buffers=buffers, + param_shapes=param_shapes, + shared_params=shared_params, + ds_version=ds_version, + frozen_param_shapes=frozen_param_shapes, + frozen_param_fragments=frozen_param_fragments) + zero_model_states.append(z_model_state) + + return zero_model_states + + +def parse_optim_states(files, ds_checkpoint_dir): + + total_files = len(files) + state_dicts = [] + for f in files: + state_dict = torch.load(f, map_location=device) + # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights + # and also handle the case where it was already removed by another helper script + state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None) + state_dicts.append(state_dict) + + if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]: + raise ValueError(f"{files[0]} is not a zero checkpoint") + zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE] + world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT] + + # For ZeRO-2 each param group can have different partition_count as data parallelism for expert + # parameters can be different from data parallelism for non-expert parameters. So we can just + # use the max of the partition_count to get the dp world_size. + + if type(world_size) is list: + world_size = max(world_size) + + if world_size != total_files: + raise ValueError( + f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. " + "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes." + ) + + # the groups are named differently in each stage + if zero_stage <= 2: + fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS + elif zero_stage == 3: + fp32_groups_key = FP32_FLAT_GROUPS + else: + raise ValueError(f"unknown zero stage {zero_stage}") + + if zero_stage <= 2: + fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))] + elif zero_stage == 3: + # if there is more than one param group, there will be multiple flattened tensors - one + # flattened tensor per group - for simplicity merge them into a single tensor + # + # XXX: could make the script more memory efficient for when there are multiple groups - it + # will require matching the sub-lists of param_shapes for each param group flattened tensor + + fp32_flat_groups = [ + torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts)) + ] + + return zero_stage, world_size, fp32_flat_groups + + +def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters): + """ + Returns fp32 state_dict reconstructed from ds checkpoint + + Args: + - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are) + + """ + print(f"Processing zero checkpoint '{ds_checkpoint_dir}'") + + optim_files = get_optim_files(ds_checkpoint_dir) + zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir) + print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}") + + model_files = get_model_state_files(ds_checkpoint_dir) + + zero_model_states = parse_model_states(model_files) + print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}') + + if zero_stage <= 2: + return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + elif zero_stage == 3: + return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + + +def _zero2_merge_frozen_params(state_dict, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + frozen_param_fragments = zero_model_states[0].frozen_param_fragments + + if debug: + num_elem = sum(s.numel() for s in frozen_param_shapes.values()) + print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in frozen_param_fragments.values()]) + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + state_dict[name] = frozen_param_fragments[name] + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _has_callable(obj, fn): + attr = getattr(obj, fn, None) + return callable(attr) + + +def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + + # Reconstruction protocol: + # + # XXX: document this + + if debug: + for i in range(world_size): + for j in range(len(fp32_flat_groups[0])): + print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}") + + # XXX: memory usage doubles here (zero2) + num_param_groups = len(fp32_flat_groups[0]) + merged_single_partition_of_fp32_groups = [] + for i in range(num_param_groups): + merged_partitions = [sd[i] for sd in fp32_flat_groups] + full_single_fp32_vector = torch.cat(merged_partitions, 0) + merged_single_partition_of_fp32_groups.append(full_single_fp32_vector) + avail_numel = sum( + [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups]) + + if debug: + wanted_params = sum([len(shapes) for shapes in param_shapes]) + wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes]) + # not asserting if there is a mismatch due to possible padding + print(f"Have {avail_numel} numels to process.") + print(f"Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + total_numel = 0 + total_params = 0 + for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups): + offset = 0 + avail_numel = full_single_fp32_vector.numel() + for name, shape in shapes.items(): + + unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape) + total_numel += unpartitioned_numel + total_params += 1 + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape) + offset += unpartitioned_numel + + # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and + # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex + # paddings performed in the code it's almost impossible to predict the exact numbers w/o the + # live optimizer object, so we are checking that the numbers are within the right range + align_to = 2 * world_size + + def zero2_align(x): + return align_to * math.ceil(x / align_to) + + if debug: + print(f"original offset={offset}, avail_numel={avail_numel}") + + offset = zero2_align(offset) + avail_numel = zero2_align(avail_numel) + + if debug: + print(f"aligned offset={offset}, avail_numel={avail_numel}") + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero2_merge_frozen_params(state_dict, zero_model_states) + + _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def zero3_partitioned_param_info(unpartitioned_numel, world_size): + remainder = unpartitioned_numel % world_size + padding_numel = (world_size - remainder) if remainder else 0 + partitioned_numel = math.ceil(unpartitioned_numel / world_size) + return partitioned_numel, padding_numel + + +def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + if debug: + for i in range(world_size): + num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values()) + print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in zero_model_states[0].frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states) + state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape) + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + avail_numel = fp32_flat_groups[0].numel() * world_size + # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each + # param, re-consolidating each param, while dealing with padding if any + + # merge list of dicts, preserving order + param_shapes = {k: v for d in param_shapes for k, v in d.items()} + + if debug: + for i in range(world_size): + print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}") + + wanted_params = len(param_shapes) + wanted_numel = sum(shape.numel() for shape in param_shapes.values()) + # not asserting if there is a mismatch due to possible padding + avail_numel = fp32_flat_groups[0].numel() * world_size + print(f"Trainable params: Have {avail_numel} numels to process.") + print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + offset = 0 + total_numel = 0 + total_params = 0 + for name, shape in param_shapes.items(): + + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + total_params += 1 + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + # XXX: memory usage doubles here + state_dict[name] = torch.cat( + tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)), + 0).narrow(0, 0, unpartitioned_numel).view(shape) + offset += partitioned_numel + + offset *= world_size + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero3_merge_frozen_params(state_dict, world_size, zero_model_states) + + _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with + ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example + via a model hub. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + + Returns: + - pytorch ``state_dict`` + + Note: this approach may not work if your application doesn't have sufficient free CPU memory and + you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with + the checkpoint. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint + # do the training and checkpoint saving + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu + model = model.cpu() # move to cpu + model.load_state_dict(state_dict) + # submit to model hub or save the model to share with others + + In this example the ``model`` will no longer be usable in the deepspeed context of the same + application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead. + + """ + if tag is None: + latest_path = os.path.join(checkpoint_dir, 'latest') + if os.path.isfile(latest_path): + with open(latest_path, 'r') as fd: + tag = fd.read().strip() + else: + raise ValueError(f"Unable to find 'latest' file at {latest_path}") + + ds_checkpoint_dir = os.path.join(checkpoint_dir, tag) + + if not os.path.isdir(ds_checkpoint_dir): + raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist") + + return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters) + + +def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be + loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + """ + + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters) + print(f"Saving fp32 state dict to {output_file}") + torch.save(state_dict, output_file) + + +def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None): + """ + 1. Put the provided model to cpu + 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` + 3. Load it into the provided model + + Args: + - ``model``: the model object to update + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + + Returns: + - ``model`: modified model + + Make sure you have plenty of CPU memory available before you call this function. If you don't + have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it + conveniently placed for you in the checkpoint folder. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint + model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) + # submit to model hub or save the model to share with others + + Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context + of the same application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + """ + logger.info(f"Extracting fp32 weights") + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag) + + logger.info(f"Overwriting model with fp32 weights") + model = model.cpu() + model.load_state_dict(state_dict, strict=False) + + return model + + +if __name__ == "__main__": + + parser = argparse.ArgumentParser() + parser.add_argument("checkpoint_dir", + type=str, + help="path to the desired checkpoint folder, e.g., path/checkpoint-12") + parser.add_argument( + "output_file", + type=str, + help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)") + parser.add_argument("-t", + "--tag", + type=str, + default=None, + help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1") + parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters") + parser.add_argument("-d", "--debug", action='store_true', help="enable debug") + args = parser.parse_args() + + debug = args.debug + + convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, + args.output_file, + tag=args.tag, + exclude_frozen_parameters=args.exclude_frozen_parameters) diff --git a/sft/smoe_vcr/checkpoint-9984/latest b/sft/smoe_vcr/checkpoint-9984/latest new file mode 100644 index 0000000000000000000000000000000000000000..6c2bd85bc7c6d33e172c9c565d8517bade9572ca --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/latest @@ -0,0 +1 @@ +global_step9984 \ No newline at end of file diff --git a/sft/smoe_vcr/checkpoint-9984/model-00001-of-00002.safetensors b/sft/smoe_vcr/checkpoint-9984/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b07c2ae523340ce4cb7072e1fb9b706974fba8cd --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c2d8d9707620fd53145a82126507abbf6cad7994d252b0d44078afa1c2d09f3 +size 4972489328 diff --git a/sft/smoe_vcr/checkpoint-9984/model-00002-of-00002.safetensors b/sft/smoe_vcr/checkpoint-9984/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2b21b54eba05ac04c35bc185da22adcbb63b4172 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e36b4b833ff0dbd2e5e9972d3eea8b6d3426b3957a3d220d4339993610aa1ac6 +size 3759015720 diff --git a/sft/smoe_vcr/checkpoint-9984/rng_state_0.pth b/sft/smoe_vcr/checkpoint-9984/rng_state_0.pth new file mode 100644 index 0000000000000000000000000000000000000000..9231f69f5fd461899867106a669ce247e70c72c2 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/rng_state_0.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9f23d807f0e704f4ca79670a6631cbff43189cf7f8ff4e1fc0a4330e636a798 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-9984/rng_state_1.pth b/sft/smoe_vcr/checkpoint-9984/rng_state_1.pth new file mode 100644 index 0000000000000000000000000000000000000000..19fe2dcc766f192ea5de79cec4dcff17172a10f7 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/rng_state_1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d37f92f6aea5386e84d2d64a1a25d6ef96a10b3bbbfe63627981604c8934076 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-9984/rng_state_2.pth b/sft/smoe_vcr/checkpoint-9984/rng_state_2.pth new file mode 100644 index 0000000000000000000000000000000000000000..bfe492519c6b79b07a8d68b98c5f3d0c073667aa --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/rng_state_2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:667ebf727735115f00a6bdbe090344e9846c726d11bb555cdc201c415f27ad85 +size 14960 diff --git a/sft/smoe_vcr/checkpoint-9984/rng_state_3.pth b/sft/smoe_vcr/checkpoint-9984/rng_state_3.pth new file mode 100644 index 0000000000000000000000000000000000000000..838d42ad13e30851fdbd1d8801738a4106a9ce8b --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/rng_state_3.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d306f8c511cba8a225e3b723c5fa79d8a6ecc922f834da914ff0780c78b1fc +size 14960 diff --git a/sft/smoe_vcr/checkpoint-9984/tokenizer.model b/sft/smoe_vcr/checkpoint-9984/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/checkpoint-9984/training_args.bin b/sft/smoe_vcr/checkpoint-9984/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992 diff --git a/sft/smoe_vcr/checkpoint-9984/zero_to_fp32.py b/sft/smoe_vcr/checkpoint-9984/zero_to_fp32.py new file mode 100644 index 0000000000000000000000000000000000000000..24cc342e78d1a006c782b3a4cd68d9ce786d8fd8 --- /dev/null +++ b/sft/smoe_vcr/checkpoint-9984/zero_to_fp32.py @@ -0,0 +1,604 @@ +#!/usr/bin/env python + +# Copyright (c) Microsoft Corporation. +# SPDX-License-Identifier: Apache-2.0 + +# DeepSpeed Team + +# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets +# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in +# the future. Once extracted, the weights don't require DeepSpeed and can be used in any +# application. +# +# example: python zero_to_fp32.py . pytorch_model.bin + +import argparse +import torch +import glob +import math +import os +import re +from collections import OrderedDict +from dataclasses import dataclass + +# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with +# DeepSpeed data structures it has to be available in the current python environment. +from deepspeed.utils import logger +from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS, + FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES, + FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS) + + +@dataclass +class zero_model_state: + buffers: dict() + param_shapes: dict() + shared_params: list + ds_version: int + frozen_param_shapes: dict() + frozen_param_fragments: dict() + + +debug = 0 + +# load to cpu +device = torch.device('cpu') + + +def atoi(text): + return int(text) if text.isdigit() else text + + +def natural_keys(text): + ''' + alist.sort(key=natural_keys) sorts in human order + http://nedbatchelder.com/blog/200712/human_sorting.html + (See Toothy's implementation in the comments) + ''' + return [atoi(c) for c in re.split(r'(\d+)', text)] + + +def get_model_state_file(checkpoint_dir, zero_stage): + if not os.path.isdir(checkpoint_dir): + raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist") + + # there should be only one file + if zero_stage <= 2: + file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt") + elif zero_stage == 3: + file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt") + + if not os.path.exists(file): + raise FileNotFoundError(f"can't find model states file at '{file}'") + + return file + + +def get_checkpoint_files(checkpoint_dir, glob_pattern): + # XXX: need to test that this simple glob rule works for multi-node setup too + ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys) + + if len(ckpt_files) == 0: + raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'") + + return ckpt_files + + +def get_optim_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt") + + +def get_model_state_files(checkpoint_dir): + return get_checkpoint_files(checkpoint_dir, "*_model_states.pt") + + +def parse_model_states(files): + zero_model_states = [] + for file in files: + state_dict = torch.load(file, map_location=device) + + if BUFFER_NAMES not in state_dict: + raise ValueError(f"{file} is not a model state checkpoint") + buffer_names = state_dict[BUFFER_NAMES] + if debug: + print("Found buffers:", buffer_names) + + # recover just the buffers while restoring them to fp32 if they were saved in fp16 + buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names} + param_shapes = state_dict[PARAM_SHAPES] + + # collect parameters that are included in param_shapes + param_names = [] + for s in param_shapes: + for name in s.keys(): + param_names.append(name) + + # update with frozen parameters + frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None) + if frozen_param_shapes is not None: + if debug: + print(f"Found frozen_param_shapes: {frozen_param_shapes}") + param_names += list(frozen_param_shapes.keys()) + + # handle shared params + shared_params = [[k, v] for k, v in state_dict["shared_params"].items()] + + ds_version = state_dict.get(DS_VERSION, None) + + frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None) + + z_model_state = zero_model_state(buffers=buffers, + param_shapes=param_shapes, + shared_params=shared_params, + ds_version=ds_version, + frozen_param_shapes=frozen_param_shapes, + frozen_param_fragments=frozen_param_fragments) + zero_model_states.append(z_model_state) + + return zero_model_states + + +def parse_optim_states(files, ds_checkpoint_dir): + + total_files = len(files) + state_dicts = [] + for f in files: + state_dict = torch.load(f, map_location=device) + # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights + # and also handle the case where it was already removed by another helper script + state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None) + state_dicts.append(state_dict) + + if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]: + raise ValueError(f"{files[0]} is not a zero checkpoint") + zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE] + world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT] + + # For ZeRO-2 each param group can have different partition_count as data parallelism for expert + # parameters can be different from data parallelism for non-expert parameters. So we can just + # use the max of the partition_count to get the dp world_size. + + if type(world_size) is list: + world_size = max(world_size) + + if world_size != total_files: + raise ValueError( + f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. " + "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes." + ) + + # the groups are named differently in each stage + if zero_stage <= 2: + fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS + elif zero_stage == 3: + fp32_groups_key = FP32_FLAT_GROUPS + else: + raise ValueError(f"unknown zero stage {zero_stage}") + + if zero_stage <= 2: + fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))] + elif zero_stage == 3: + # if there is more than one param group, there will be multiple flattened tensors - one + # flattened tensor per group - for simplicity merge them into a single tensor + # + # XXX: could make the script more memory efficient for when there are multiple groups - it + # will require matching the sub-lists of param_shapes for each param group flattened tensor + + fp32_flat_groups = [ + torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts)) + ] + + return zero_stage, world_size, fp32_flat_groups + + +def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters): + """ + Returns fp32 state_dict reconstructed from ds checkpoint + + Args: + - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are) + + """ + print(f"Processing zero checkpoint '{ds_checkpoint_dir}'") + + optim_files = get_optim_files(ds_checkpoint_dir) + zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir) + print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}") + + model_files = get_model_state_files(ds_checkpoint_dir) + + zero_model_states = parse_model_states(model_files) + print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}') + + if zero_stage <= 2: + return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + elif zero_stage == 3: + return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters) + + +def _zero2_merge_frozen_params(state_dict, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + frozen_param_fragments = zero_model_states[0].frozen_param_fragments + + if debug: + num_elem = sum(s.numel() for s in frozen_param_shapes.values()) + print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in frozen_param_fragments.values()]) + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + state_dict[name] = frozen_param_fragments[name] + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _has_callable(obj, fn): + attr = getattr(obj, fn, None) + return callable(attr) + + +def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + + # Reconstruction protocol: + # + # XXX: document this + + if debug: + for i in range(world_size): + for j in range(len(fp32_flat_groups[0])): + print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}") + + # XXX: memory usage doubles here (zero2) + num_param_groups = len(fp32_flat_groups[0]) + merged_single_partition_of_fp32_groups = [] + for i in range(num_param_groups): + merged_partitions = [sd[i] for sd in fp32_flat_groups] + full_single_fp32_vector = torch.cat(merged_partitions, 0) + merged_single_partition_of_fp32_groups.append(full_single_fp32_vector) + avail_numel = sum( + [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups]) + + if debug: + wanted_params = sum([len(shapes) for shapes in param_shapes]) + wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes]) + # not asserting if there is a mismatch due to possible padding + print(f"Have {avail_numel} numels to process.") + print(f"Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + total_numel = 0 + total_params = 0 + for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups): + offset = 0 + avail_numel = full_single_fp32_vector.numel() + for name, shape in shapes.items(): + + unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape) + total_numel += unpartitioned_numel + total_params += 1 + + if debug: + print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ") + state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape) + offset += unpartitioned_numel + + # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and + # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex + # paddings performed in the code it's almost impossible to predict the exact numbers w/o the + # live optimizer object, so we are checking that the numbers are within the right range + align_to = 2 * world_size + + def zero2_align(x): + return align_to * math.ceil(x / align_to) + + if debug: + print(f"original offset={offset}, avail_numel={avail_numel}") + + offset = zero2_align(offset) + avail_numel = zero2_align(avail_numel) + + if debug: + print(f"aligned offset={offset}, avail_numel={avail_numel}") + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero2_merge_frozen_params(state_dict, zero_model_states) + + _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def zero3_partitioned_param_info(unpartitioned_numel, world_size): + remainder = unpartitioned_numel % world_size + padding_numel = (world_size - remainder) if remainder else 0 + partitioned_numel = math.ceil(unpartitioned_numel / world_size) + return partitioned_numel, padding_numel + + +def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states): + if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0: + return + + if debug: + for i in range(world_size): + num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values()) + print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}') + + frozen_param_shapes = zero_model_states[0].frozen_param_shapes + wanted_params = len(frozen_param_shapes) + wanted_numel = sum(s.numel() for s in frozen_param_shapes.values()) + avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size + print(f'Frozen params: Have {avail_numel} numels to process.') + print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params') + + total_params = 0 + total_numel = 0 + for name, shape in zero_model_states[0].frozen_param_shapes.items(): + total_params += 1 + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + + param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states) + state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape) + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements") + + +def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states): + param_shapes = zero_model_states[0].param_shapes + avail_numel = fp32_flat_groups[0].numel() * world_size + # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each + # param, re-consolidating each param, while dealing with padding if any + + # merge list of dicts, preserving order + param_shapes = {k: v for d in param_shapes for k, v in d.items()} + + if debug: + for i in range(world_size): + print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}") + + wanted_params = len(param_shapes) + wanted_numel = sum(shape.numel() for shape in param_shapes.values()) + # not asserting if there is a mismatch due to possible padding + avail_numel = fp32_flat_groups[0].numel() * world_size + print(f"Trainable params: Have {avail_numel} numels to process.") + print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.") + + # params + # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support + # out-of-core computing solution + offset = 0 + total_numel = 0 + total_params = 0 + for name, shape in param_shapes.items(): + + unpartitioned_numel = shape.numel() + total_numel += unpartitioned_numel + total_params += 1 + + partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size) + + if debug: + print( + f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}" + ) + + # XXX: memory usage doubles here + state_dict[name] = torch.cat( + tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)), + 0).narrow(0, 0, unpartitioned_numel).view(shape) + offset += partitioned_numel + + offset *= world_size + + # Sanity check + if offset != avail_numel: + raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong") + + print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements") + + +def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states, + exclude_frozen_parameters): + state_dict = OrderedDict() + + # buffers + buffers = zero_model_states[0].buffers + state_dict.update(buffers) + if debug: + print(f"added {len(buffers)} buffers") + + if not exclude_frozen_parameters: + _zero3_merge_frozen_params(state_dict, world_size, zero_model_states) + + _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states) + + # recover shared parameters + for pair in zero_model_states[0].shared_params: + if pair[1] in state_dict: + state_dict[pair[0]] = state_dict[pair[1]] + + return state_dict + + +def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with + ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example + via a model hub. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + + Returns: + - pytorch ``state_dict`` + + Note: this approach may not work if your application doesn't have sufficient free CPU memory and + you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with + the checkpoint. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint + # do the training and checkpoint saving + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu + model = model.cpu() # move to cpu + model.load_state_dict(state_dict) + # submit to model hub or save the model to share with others + + In this example the ``model`` will no longer be usable in the deepspeed context of the same + application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead. + + """ + if tag is None: + latest_path = os.path.join(checkpoint_dir, 'latest') + if os.path.isfile(latest_path): + with open(latest_path, 'r') as fd: + tag = fd.read().strip() + else: + raise ValueError(f"Unable to find 'latest' file at {latest_path}") + + ds_checkpoint_dir = os.path.join(checkpoint_dir, tag) + + if not os.path.isdir(ds_checkpoint_dir): + raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist") + + return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters) + + +def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False): + """ + Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be + loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed. + + Args: + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + - ``exclude_frozen_parameters``: exclude frozen parameters + """ + + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters) + print(f"Saving fp32 state dict to {output_file}") + torch.save(state_dict, output_file) + + +def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None): + """ + 1. Put the provided model to cpu + 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` + 3. Load it into the provided model + + Args: + - ``model``: the model object to update + - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``) + - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14`` + + Returns: + - ``model`: modified model + + Make sure you have plenty of CPU memory available before you call this function. If you don't + have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it + conveniently placed for you in the checkpoint folder. + + A typical usage might be :: + + from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint + model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) + # submit to model hub or save the model to share with others + + Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context + of the same application. i.e. you will need to re-initialize the deepspeed engine, since + ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it. + + """ + logger.info(f"Extracting fp32 weights") + state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag) + + logger.info(f"Overwriting model with fp32 weights") + model = model.cpu() + model.load_state_dict(state_dict, strict=False) + + return model + + +if __name__ == "__main__": + + parser = argparse.ArgumentParser() + parser.add_argument("checkpoint_dir", + type=str, + help="path to the desired checkpoint folder, e.g., path/checkpoint-12") + parser.add_argument( + "output_file", + type=str, + help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)") + parser.add_argument("-t", + "--tag", + type=str, + default=None, + help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1") + parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters") + parser.add_argument("-d", "--debug", action='store_true', help="enable debug") + args = parser.parse_args() + + debug = args.debug + + convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, + args.output_file, + tag=args.tag, + exclude_frozen_parameters=args.exclude_frozen_parameters) diff --git a/sft/smoe_vcr/model-00001-of-00002.safetensors b/sft/smoe_vcr/model-00001-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a2e039e1186da0f5a2cb583c8aef3d2973e65d01 --- /dev/null +++ b/sft/smoe_vcr/model-00001-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b980095539e1791498907a3250d8c638d9f7efb6c942efaa1efaab69e6631208 +size 4972489328 diff --git a/sft/smoe_vcr/model-00002-of-00002.safetensors b/sft/smoe_vcr/model-00002-of-00002.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..332e627d0fccd3f23ca5f6d0e149534c99da3bee --- /dev/null +++ b/sft/smoe_vcr/model-00002-of-00002.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da33b82157a6d9d3f145750aa1ffe39fa400f5186ca3e89728d628e59816cda0 +size 3759015720 diff --git a/sft/smoe_vcr/tokenizer.model b/sft/smoe_vcr/tokenizer.model new file mode 100644 index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899 --- /dev/null +++ b/sft/smoe_vcr/tokenizer.model @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347 +size 499723 diff --git a/sft/smoe_vcr/training_args.bin b/sft/smoe_vcr/training_args.bin new file mode 100644 index 0000000000000000000000000000000000000000..a6bf5dd8c2124ae4390fcff1f839a55a05673fb5 --- /dev/null +++ b/sft/smoe_vcr/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e04cbeb93b4abbbbbea1e73f341a6942676570aa79d4df252055868d2df91fe +size 7992