CHANGELOG.md 13 KB
Newer Older
1
2
3
4
5
6
# Changelog
All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
7
## NEXT - TBD
8
### Fixed
anj-s's avatar
anj-s committed
9
10
11
12
13
- FSDP: We don't attach post backward hooks for params that don't require grad. However in the hook
        triggered after the post backward hook, we assert on the POST_BACKWARD state which can only
        be set in the post backward hook. Modified the assert to account for the fact that the root
        FSDP module can have child modules with params that require grad and it can contain params
        that don't require grad and hence can fail the previous assert. [#761]
Min Xu's avatar
Min Xu committed
14
15

### Added
anj-s's avatar
anj-s committed
16
17
18
19
20
- FSDP: Added support for returning the original names of parameters when `named_parameters` is called on
        the module. To retrieve the orginal names of the parameters along with the params, you need to 
        call `named_parameters` under the `summon_full_params` context when using flattened params or original
        params. If you are using original params (i.e flatten_params=False), calling `named_parameters` outside
        of the `summon_full_params` context will still return the original param names along with the local shards. [#755]
Min Xu's avatar
Min Xu committed
21
22
23
24


## [0.4.0] - 2021-07-31
### Fixed
25
26
27
28
29
30
31
- FSDP: fixed final backward callback in certain activation checkpointed cases. Before this fix,
        if a model is activation checkpointed in a certain way, the final backward
        callback can fire incorrectly. That's due to autograd and reentrant backward
        graphs. With this fix, the final callback is always registered on the outer
        most root FSDP instance (i.e. the outer most backward graph), which result
        in reliably firing. This makes FSDP much more robust with respect to different
        models and activation checkpoints. [#753]
Min Xu's avatar
Min Xu committed
32
33

### Added
34
35
36
37
- FSDP: support gradient accumulation without the `no_sync` context. This is useful
        in training with smaller number of GPU with same overall batch size as large
        number of GPUs. Compared with the `no_sync` context, this mode consumes less
        GPU memory but uses more networking bandwidth. [#752]
Min Xu's avatar
Min Xu committed
38

Min Xu's avatar
Min Xu committed
39

Min Xu's avatar
Min Xu committed
40
41
42
43
44
## [0.3.9] - 2021-07-26
### Fixed
- FSDP: fixed metadata saving and shard consolidation for MoE cases. When a model has
        shared parameters or mixture of expert layers, the handling of state dict
        metadata was broken. This release fixes that. [#746]
45
- OSS: fixed the buckets which would stay in fp16 if `broadcast fp16` was required [#751]
anj-s's avatar
anj-s committed
46
47

### Added
Min Xu's avatar
Min Xu committed
48
49
- FSDP: better performance; use `_allgather_base` and `_reduce_scatter_base` when they are
        available from pytorch nightly version (will be in 1.10 releases) [#729]
50
- FSDP: prepared FSDP internals for supporting multiple groups of flatten parameters (to support more general optimization) [#746]
anj-s's avatar
anj-s committed
51
52
53
54
55
56
57
58
59
60

## [0.3.8] - 2021-07-12
### Fixed
- checkpointing: Use dummy tensor to ensure backward pass is called. [#701]
- checkpointing: Ensure internal fwd counter is not incremented in eval mode. [#709]
- checkpointing: Use non-blocking CPU transfer to improve perf. [#719]
- FSDP: Fixed bug where buffers returned in `state_dict()` could still be half precision when `mixed_precision` is set to `True`. [#705]
- FSDP: Ensure requires_grad of FlatParameter is consistent with requires_grad of the original parameters. [#721]
- doc: Thoroughly improved the doc for FSDP. [#711]
- cleanup: Remove examples/ doc from the repo. [#712]
61
- cleanup: Future proof storage size test. [#735]
anj-s's avatar
anj-s committed
62
63
- cleanup: Migrate away from legacy torchtext iterators. [#713]
- chore: Updated torch 1.9 to release version. [#717]
Min Xu's avatar
Min Xu committed
64
65

### Added
66
- FSDP: supporting multiple flatten parameter groups [#708] [#711]
anj-s's avatar
anj-s committed
67
- chore: Add the latest numpy version to requirements-test.txt to prevent mypy errors on certain PR commits [#732]
Min Xu's avatar
Min Xu committed
68
69
70
71
72
73
74

## [0.3.7] - 2021-05-17
### Fixed
- setup.py: hide CUDA extensions behind `BUILD_CUDA_EXTENSIONS` envvar [#634]
- checkpointing: rename and move the `checkpoint_activations` wrapper [#654]
- FSDP: fix `local_state_dict` potentially called child class's `state_dict` [#574]
- FSDP: fix extra process groups being created by default. Old behavior can cause excessive GPU memory usage [#678] [#681]
75
76
77
- FSDP: fix forward pass not overlapping compute and allgather [#671]
- FSDP: improved frozen weight support [#657]
- FSDP: workaround AMP autocast cache issue with `clear_autocast_cache` flag [#650]
Min Xu's avatar
Min Xu committed
78
- FSDP: Rename API arg `cpu_offload` to `move_params_to_cpu` to better reflect functionality. We will deprecate `cpu_offload` in an upcoming release [#676]
79
80
- MoE: several fixes [#666] [#667] [#668]
- SDP: re-expose the module property [#647]
Min Xu's avatar
Min Xu committed
81
- wrap: support wrapping based on `wrapper_config` [#685]
82

83
### Added
84
85
- FSDP: added `force_input_to_fp32` flag for SyncBatchNorm [#659]
- FSDP: better memory usage for reduce bucket [#633]
86
87
- FSDP: added `local_metadata_dict` to save sharding relating information [#683]
- FSDP: added `consolidate_shard_weights` to reconstruct the consolidated (non-sharded) model weights from saved sharded weights and metadata on the disk [#683]
Min Xu's avatar
Min Xu committed
88
- Experimental SyncBatchNorm [#662] [#680]
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
89

Min Xu's avatar
Min Xu committed
90
91
## [0.3.6] - 2021-04-26
### Added
92
- FSDP: Consolidate cpu\_adam optimizer state dict ([#607](https://github.com/facebookresearch/fairscale/pull/607))
Min Xu's avatar
Min Xu committed
93
94
95
96
97
98
99
100
101

### Fixed
- FSDP: handle model with multiple forward pass and checkpoint ([#621](https://github.com/facebookresearch/fairscale/pull/621))
- FSDP & SDP: check before calling `_specify_ddp_gpu_num` ([#626](https://github.com/facebookresearch/fairscale/pull/626))
- FSDP: relax checking root condition ([#620](https://github.com/facebookresearch/fairscale/pull/620))
- SDP: removing an assert which does not seem always accurate ([#625](https://github.com/facebookresearch/fairscale/pull/625))
- FSDP: changing FSDP init to by pass pg validation ([#619](https://github.com/facebookresearch/fairscale/pull/619))
- OSS: to 100% coverage ([#618](https://github.com/facebookresearch/fairscale/pull/618))

Min Xu's avatar
Min Xu committed
102
103
104
105
106
107
108
109
110
111
## [0.3.5] - 2021-04-19
### Added
- [offload] Add API, tutorial and smaller doc string changes. ([#576](https://github.com/facebookresearch/fairscale/pull/576))

### Fixed
- FSDP: fixing training with freezing weights ([#614](https://github.com/facebookresearch/fairscale/pull/614))
- SDP: privatizing all the things ([#611](https://github.com/facebookresearch/fairscale/pull/611))
- FSDP: Make `_get_default_cuda_device` more robust to modules without params ([#606](https://github.com/facebookresearch/fairscale/pull/606))
- OffloadModel: Add prev codepath of using OffloadModel without activation checkpointing ([#608](https://github.com/facebookresearch/fairscale/pull/608))

Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
112
## [0.3.4] - 2021-04-13
Min Xu's avatar
Min Xu committed
113
### Added
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
114
115
- FSDP: Add no broadcast optim state option ([#560](https://github.com/facebookresearch/fairscale/pull/560))

Min Xu's avatar
Min Xu committed
116
### Fixed
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
117
118
119
120
121
- ShardedDDP: Properly handle .eval() mode ([#587](https://github.com/facebookresearch/fairscale/pull/587))
- ShardedDDP: Handle model being moved back to CPU prior to state consolidation ([#573](https://github.com/facebookresearch/fairscale/pull/573))
- FSDP: much faster state consolidation ([#595](https://github.com/facebookresearch/fairscale/pull/595))
- FSDP: Add gradient pre-dedivide to prevent overflow with large world sizes ([#565](https://github.com/facebookresearch/fairscale/pull/565))
- Offload: (experimental) Fix activation offloading to CPU ([#588]((https://github.com/facebookresearch/fairscale/pull/588) )
Min Xu's avatar
Min Xu committed
122

Min Xu's avatar
Min Xu committed
123
124
## [0.3.3] - 2021-04-1
### Added
125
- FSDP: changed `auto_wrap_bn` utility function so that single FSDP group is optional ([#556](https://github.com/facebookresearch/fairscale/pull/556))
Min Xu's avatar
Min Xu committed
126
127
128
129
130
131
132
133
- FSDP: optimizer state load/save ([#537](https://github.com/facebookresearch/fairscale/pull/537))
- FSDP: fix weight init when using apply() ([#543](https://github.com/facebookresearch/fairscale/pull/543))
- Multiprocess Pipe: retired old implementation
- Experimental: xpipe

### Fixed
- ShardedDDP deferred init ([#558](https://github.com/facebookresearch/fairscale/pull/558))

Min Xu's avatar
Min Xu committed
134
## [0.3.2] - 2021-03-18
Min Xu's avatar
Min Xu committed
135
### Added
136
- Experimental: Add spectrain support ([#372](https://github.com/facebookresearch/fairscale/issues/372))
Min Xu's avatar
Min Xu committed
137
- FSDP: enabled pytorch SyncBN (no asserting) ([#527](https://github.com/facebookresearch/fairscale/issues/527))
138
- FSDP: added `auto_wrap_bn` utility function ([#531](https://github.com/facebookresearch/fairscale/pull/531))
Min Xu's avatar
Min Xu committed
139
140

### Fixed
141
- OSS: fix a compatibily problem with lightning wrt optimizer state dict ([#510](https://github.com/facebookresearch/fairscale/issues/510))
142
- FSDP: fixed a bug when part of autograd graph is traversed multiple times in mixed precision mode ([#513](https://github.com/facebookresearch/fairscale/pull/513))
Min Xu's avatar
Min Xu committed
143
144
145
146

## [0.3.1] - 2021-03-09
### Added
- FSDP docs ([#455](https://github.com/facebookresearch/fairscale/issues/455))
147
- `enable_wrap` and `auto_wrap` APIs ([#446](https://github.com/facebookresearch/fairscale/issues/446))
Min Xu's avatar
Min Xu committed
148
149
- Added experimental.nn.OffloadModel API for training large models on a single GPU.([#432](https://github.com/facebookresearch/fairscale/issues/432))

150
### Fixed
Min Xu's avatar
Min Xu committed
151
152
153
154
- OSS: fix a broken state dict when using non contiguous param groups
- Several SDP fixes around performance and corner cases
- Many FSDP fixes
- AdaScale & SDP/FSDP test added but not officially supported
155

Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
156
## [0.3.0] - 2021-02-22
Min Xu's avatar
Min Xu committed
157
158
### Added
- FullyShardedDataParallel (FSDP) ([#413](https://github.com/facebookresearch/fairscale/issues/413))
159
- ShardedDDP fp16 grad reduction option ([#402](https://github.com/facebookresearch/fairscale/issues/402))
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
160
- Expose experimental algorithms within the pip package ([#410](https://github.com/facebookresearch/fairscale/pull/410))
Min Xu's avatar
Min Xu committed
161

162
### Fixed
Min Xu's avatar
Min Xu committed
163
- Catch corner case when the model is too small with respect to the world size, and shards are empty ([#406](https://github.com/facebookresearch/fairscale/pull/406))
164
- Memory leak in `checkpoint_wrapper` ([#412](https://github.com/facebookresearch/fairscale/pull/412))
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
165
166

## [0.1.7] - 2021-02-19
167
168
### Fixed
- ShardedDDP and OSS handle model trainability changes during training ([#369](https://github.com/facebookresearch/fairscale/issues/369))
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
169
170
171
- ShardedDDP state dict load/save bug ([#386](https://github.com/facebookresearch/fairscale/issues/386))
- ShardedDDP handle train/eval modes ([#393](https://github.com/facebookresearch/fairscale/issues/393))
- AdaScale handling custom scaling factors ([#401](https://github.com/facebookresearch/fairscale/issues/401))
172
173

### Added
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
174
- ShardedDDP manual reduce option for checkpointing ([#389](https://github.com/facebookresearch/fairscale/issues/389))
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
175

Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
176
177
178
179
180
181
182
183
184
185
## [0.1.6] - 2021-02-10
### Added
- Checkpointing model wrapper (#376)
- Faster OSS, flatbuffers (#371)
- Small speedup in OSS clipgradnorm (#363)

### Fixed
- Bug in ShardedDDP with 0.1.5 depending the init (KeyError / OSS)
- Much refactoring in Pipe (#357, #358, #360, #362, #370, #373)
- Better pip integration / resident pytorch (#375)
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
186
187

## [0.1.5] - 2021-02-03
188
### Added
189
190
191
- Pytorch compatibility for OSS checkpoints (#310)
- Elastic checkpoints for OSS, world size can vary in between save and loads (#310)
- Tensor views for OSS bucketing, reduced CPU use (#300)
192
- Bucket calls in ShardedDDP, for faster inter node communications (#327)
Benjamin Lefaudeux's avatar
Benjamin Lefaudeux committed
193
194
195
196
197
198
199
- FlattenParamWrapper, which flattens module parameters into a single tensor seamlessly (#317)
- AMPnet experimental support (#304)

### Fixed
- ShardedDDP properly handles device changes via `.to()` (#353)
- Add a new interface for AdaScale, AdaScaleWrapper, which makes it compatible with OSS (#347)

200

201
202
203
204
205
## [0.1.4] - 2021-01-07
### Fixed
- Missing cu files in the pip package


206
207
208
## [0.1.3] - 2021-01-04
### Fixed
- Release numbering within python and from pypi
209

210
## [0.1.2] - 2021-01-04
211
### Added
212
213
- AdaScale:
  . Added gradient accumulation feature (#202)
214
215
216
  . Added support of `torch.lr_scheduler` (#229)
  . Added support for `add_param_groups` (#266)
  . Added support for `scale != world_size` (#266)
217
218

### Fixed
Min Xu's avatar
Min Xu committed
219
220
- AdaScale: smoothing factor value fixed when using gradient accumulation (#235)
- Pipe: documentation on balancing functions (#243)
221
222
223
- ShardedDDP: handle typical NLP models
- ShardedDDP: better partitioning when finetuning

224

225
226
227
228
## [0.1.1] - 2020-12-01
### Fixed
- make sure pip package includes header files (#221)

msbaines's avatar
msbaines committed
229
230
231
232
233
234
## [0.1.0] - 2020-12-01
### Added
- ShardedDataParallel with autoreduce (#157)
- cpu support for Pipe (#188)
- ShardedOptim: Distributed Grad Scaler (for torch AMP)  (#182)
- OSS-aware clip grads, bridge sharded states (#167)
235
- oss: add `rank_local_state_dict` staticmethod (#174)
msbaines's avatar
msbaines committed
236
237
238
239
240
241
- support for PyTorch 1.7.0 (#171)
- Add implementation of AdaScale (#139)

### Fixed
- pip package install (#196, #200)

msbaines's avatar
msbaines committed
242
243
244
245
246
247
248
249
## [0.0.3] - 2020-10-14
### Added
- multi-process pipe

### Fixed
- multiple OSS fixes
- MegaTron+OSS DDP fix

msbaines's avatar
msbaines committed
250
251
## [0.0.2] - 2020-08-28
### Added
252
- add ddp that works with oss with `reduce()` not `all_reduce()` (#19)
msbaines's avatar
msbaines committed
253
254
255
256
257
258
259
260
261
- support for PyTorch v1.6
- add mixed precision Adam (#40)
- Adam optimizer state scaling (#44)

### Fixed
- properly restore a sharded optim state (#39)
- OSS restore state to proper device (#46)
- optim/oss: support optimizers with additional step kwargs (#53)
- optim/oss: fix state cast (#56)
262
- fix eval for `oss_ddp` (#55)
msbaines's avatar
msbaines committed
263
264
265
- optim/oss: work correctly with LRScheduler (#58)

## [0.0.1] - 2020-07-31
266
- Initial release.