GitLab

Make ShardedGradScaler __init__ mirror GradScaler so super can forward parameters. Without this one cannot configure a ShardedGradScaler object like one can with the PyTorch native GradScaler object.
Updated with black linter.
Added stub for GradScaler __init__ which solves mypy issues and removed
ignore comment.

* rename codecov yaml file

* remove status checks

* [chore] 0.3.7 release

* fixed changelog
Co-authored-by: Min Xu <min.xu.public@gmail.com>

* [fix] auto_wrap: support wrapping based on wrapper_config

- user can use this to avoid assert if auto_wrap is used multiple times on a module
- user can traverse the modules multiple times and assign a wrapper_config
  to the module and then use auto_wrap once to wrap them

fix #649
fix #585

* added changelog

* fix tests

* fix a test

* added an optional assert for collision based on discussions with Quentin

* added config_auto_wrap_policy

* lint
Co-authored-by: Min Xu <min.xu.public@gmail.com>

* Save FSDP metadata for offline unflattening

* Complete the meta-data saving method with all the information needed to reconstruct a checkpoint offline, and implement the method that reconstruct a consolidated checkpoint from a sharded checkpoint

* Complete the meta-data saving method with all the information needed to reconstruct a checkpoint offline, and implement the method that reconstruct a consolidated checkpoint from a sharded checkpoint

* Add a unit test to show how to use the function

* Code review + improvement of the unit tests

* Code review: extract clean_path

* Make meta data and consolidation of checkpoint work for flatten_parameter=False

* Add new unit test file in CI

* Complete changelog and fix mypy issues

* Add support for module buffers in the consolidation of sharded checkpoints

* Better support for module buffers: save them in the meta data

* Refactoring: use a data-format for the meta data that is simpler to understand (move from object of array to array of object format)

* Renaming to make code clearer

* Code review: in_temporary_directory rework and typo correction

* Renaming
Co-authored-by: Sam Shleifer <sshleifer@gmail.com>
Co-authored-by: QuentinDuval <QuentinDuval@users.noreply.github.com>

* api changes

* fix list

* modify changelog

* modify changelog

* modify changelog

* move function

* [minor] use dist.group.WORLD for default process group

- this is slightly more efficient than the previous commit
  for get_process_group_cached.

* fix

* better fix

* fixed for pytorch 1.6 and 1.7

* Update fairscale/utils/parallel.py
Co-authored-by: Min Xu <min.xu@acm.org>
Co-authored-by: Min Xu <min.xu.public@gmail.com>

* fix saving and loading checkpoints with use_sharded_state=True

* mypy fix

* better fix of the infinite recursion

- we need to specifically call FSDP.state_dict from its local state_dict
- added unit test that fails without the fix and works with the fix
- fixed mypy for the overloaded functions

* make cpu-only fsdp work for state_dict at least
Co-authored-by: Min Xu <min.xu@acm.org>
Co-authored-by: Min Xu <min.xu.public@gmail.com>
Co-authored-by: Min Xu <m1n@fb.com>

* [fix] add and use get_process_group_cached

- This commit makes FSDP avoid making too many process groups by default
- Extra process group is bad for GPU memory and init time

* add changelog

* lint

* note on speed

* add better assert output

test seems to be flaky:
https://app.circleci.com/pipelines/github/facebookresearch/fairscale/2957/workflows/383c9f9f-f1a5-461c-8c41-e2e28ece037b/jobs/26783/steps



* update test reference memory values

- With cached process groups, the memory is reduced as reported by
pytorch as well (due to bucket buffer memory for the reduction buffer)
- The effect on memory is actually more on the SMI memory, which is not
reported by pytorch and checked by this test.

* Update fairscale/nn/data_parallel/fully_sharded_data_parallel.py

* Update fairscale/nn/data_parallel/fully_sharded_data_parallel.py

* Update CHANGELOG.md

* Update fairscale/utils/parallel.py

* Update fairscale/utils/parallel.py

* Update fairscale/utils/parallel.py

* Update fairscale/utils/parallel.py

* improved changelog

* better handling of underscores in the md file
Co-authored-by: Min Xu <min.xu@acm.org>

* rename files

* add newly renamed file

* rename and move checkpoint activations related files

* add test files to ci list

* fix lint errors

* modify docs

* add changelog

* retain old path for now

* fix lint errors

* add another import test case

* fix merge conflict

* add missing test file

* add pragma

* add mypy ignore comments

* fix comment

* add more no cover comments

* add comments

* [fix] FSDP forward pass overlap between compute and all-gather

- much thanks for @cyanguwa for report and @QuentinDuval for debugging it
- a new unit test is added to check for this and ensure we detect
  issue with overlapping and cpu/gpu blocking wait calls

* fix

* fix

* fix

* better assertion outputs

* fix format and tune all_gather mb for CI

* more tuning with non_flatten

* undo an accidental change

* tuning all gather mb and del model

* Update + fix overlapping test to use patched all_gather w/ delay (#672)

* fixing get_cycles_per_ms

* add get_smi_memory

* update the docstring
Co-authored-by: Min Xu <min.xu@acm.org>
Co-authored-by: Myle Ott <myleott@fb.com>

* [chore] Updating PR template

Add N/A (Not Applicable) options to some of the questions in the PR template

* Update PULL_REQUEST_TEMPLATE.md

* Update PULL_REQUEST_TEMPLATE.md

- we do have a use case of empty params inside a FSDP -- for the
overlapping fsdp unit test, we use it to measure timing of compute
when no params is needed for all_gather
- therefore, I updated the comment to be more correct there.
- fixes #661

Co-authored-by: @myleott

Co-authored-by: @myleott

* add license file headers for all files

* fix lint

* rename and move optim/utils.py

* attach the new file

Workaround for https://github.com/pytorch/pytorch/issues/55579

Co-authored-by: @shruti-bh, @myleott

* [test]: add a more general test case

- also rebalance the tests a bit

* added missing arg

* balance

* better checking

* balance

* make test smaller and faster

* make ddp results cached and enable sync_bn

* clean up

* fix tests

* changelog

* blance

* fix

* addressing comments
Co-authored-by: Min Xu <min.xu@acm.org>

* [feat] experimental.nn.SyncBatchNorm: initial commit

Fast/simple re-implementation of SyncBatchNorm.

When profiling SSL Vision, I was seeing a majority of cycles spent in
SyncBatchNorm. With this change, I see a 10% to 20% speedup on the
model I was profiling.

When running benchmarks/experimental/sync_batchnorm.py on 8 x V100,
I get a 6x speedup:

<class 'torch.nn.modules.batchnorm.BatchNorm2d'>
Elapsed time is  0.08709120750427246
Elapsed time is  0.12632274627685547
Elapsed time is  0.14095258712768555
Elapsed time is  0.16529417037963867
Elapsed time is  0.1419970989227295
Elapsed time is  0.15166854858398438
Elapsed time is  0.12000870704650879
Elapsed time is  0.17534875869750977
<class 'torch.nn.modules.batchnorm.SyncBatchNorm'>
Elapsed time is  2.5087168216705322
Elapsed time is  2.497001886367798
Elapsed time is  2.5204885005950928
Elapsed time is  2.526789903640747
Elapsed time is  2.5080230236053467
Elapsed time is  2.524489641189575
Elapsed time is  2.513214588165283
Elapsed time is  2.5359973907470703
<class 'fairscale.experimental.nn.sync_batchnorm.SyncBatchNorm'>
Elapsed time is  0.4126114845275879
Elapsed time is  0.39051294326782227
Elapsed time is  0.40685415267944336
Elapsed time is  0.4159870147705078
Elapsed time is  0.42383885383605957
Elapsed time is  0.4080159664154053
Elapsed time is  0.41202712059020996
Elapsed time is  0.42400121688842773

* [fix] better assert and better test for frozen weights

- the precise condition should have been check m.parameters(), not
  m.params.
- fixes #643

* add changelog

* use enum is so much better
Co-authored-by: Min Xu <min.xu@acm.org>

* fix infinite loop in docstring

* fix docstring

* increasing the code coverage, good practice and raising bugs.  hopefully getting to 100%
* small bugfix

* rename files

* add newly renamed file

Co-authored-by: Anjali Sridhar <anj@devfair0443.h2.fair>

* [fix] add clear_autocast_cache flag

- when training in AMP model with weight dtype32, FSDP may need to
  optionally clear the autocast cache to avoid GPU OOM
- this flag is default false, automatically doing it is a future TODO
- also added a verbose flag to make print(fsdp_model) a bit shorter
- updated the memory test to cover those new code
- added a couple of useful functions in parallel.py and testing.py

* minor

* address comments

* format

* improve the test
Co-authored-by: Min Xu <min.xu@acm.org>

* dynamic loss scaler

* isort

* black

* flake8

* comments

* added the test to ci file, added a line to catch the overflow error, fixed some formatting errors

* adding type annotation

* added todo for adding more test cases for handling Nan gradients

* fix some doc string and comments, add more tods

* fix two doc strings

* fix + unit test
* changelog update

* [minor] not creating a temp file on import

* address review

* Revert "address review"

This reverts commit f65eb9bc7f7ea8829b1ac0a369ef9a3e6b56420a.
Co-authored-by: Min Xu <min.xu@acm.org>

* Improving test coverage on SDP
* using pytest exception catcher

* [test] improve BN test coverage

- Added sync_bn on/off cases
- Added conv and linear bias on/off cases
- clarified when sync_bn is off, when is BN wrapping needed with the test

* adding a comment
Co-authored-by: Min Xu <min.xu@acm.org>

* adding auto graph generation for distributed pipeline

* ignore trace.py for my for now, since it needs pytorch 1.8

* fixing tests

* simplifying graph api

* remove unused debug utilities

* use inspect to find argument lists

* use sharded linear layer

* flkae8

* comment

* polishing

* polishing

* [feat] save memory by using bucket buffer only in backward

- this fixes bug #627
- added documentation to clarify the buffer's cost and speed/memory
  tradeoff
- added setup/teardown calls so that the buffer is only allocated
  during the backward pass, saving more memory for forward and stepping
  so that they can be used for things like activations.
- added a unit test that assert the memory is in range.

Comparing with DDP:

  1. buffer size scales with # of FSDP not model size
  2. buffer is only allocated during backward
  3. buffer is used for small tensors only to reduce overhead
  4. overlapping of compute-reduction is very different

* add PR number to changelog

* filled in with memory number on 1.9

* addressed comments

* update comments

* fix for 1.6

* add a todo
Co-authored-by: Min Xu <min.xu@acm.org>