Skip to content

GitLab

Switch branch/tag
  1. 02 Nov, 2021 1 commit
  3. 01 Nov, 2021 2 commits
    • Min Xu's avatar
      [feat] [FSDP]: add experimental support to shared weights (#836) · f2af4c66
      Min Xu authored 
      

* added a new test, passing without shared weights

* tested weight sharing

* added the test to test list file

* extended to world_size = 2

* fixed test

* [feat]: add limited and experimental support for shared parameter

* fixed tests

* simplify to work with layer with at least 1 non-shared params and add code to pick up linked_param field for sharding the shared param

* fixed the case where linked param is not in separate FSDP

* changelog and remove old code
Co-authored-by: default avatarMin Xu <min.xu.public@gmail.com>
      f2af4c66
    • anj-s's avatar
      [feature] Add the low level SSD APIs (#829) · a9fcaa28
      anj-s authored 
      * add doc strings

* add lower level SSD APIs and tests

* add the test to the list to be run

* remove unused imports

* more doc string changes

* fix lint errors
      a9fcaa28
  5. 28 Oct, 2021 1 commit
  7. 27 Oct, 2021 6 commits
  9. 24 Oct, 2021 1 commit
  11. 22 Oct, 2021 2 commits
    • anj-s's avatar
      modify golden data (#825) · 35f327f3
      anj-s authored
      35f327f3
    • Eugen Hotaj's avatar
      Extend auto shard capabilities to work around torch.fx edge cases. (#817) · 7bdf50a3
      Eugen Hotaj authored 
      auto_shard.py currently uses torch.fx to create a symbolic DAG of
operations and linearizes that DAG into an nn.Sequential so it can later
be used for model offloading. This works in most cases but runs into
issues for certain eager mode features, such as dynamic conditionals,
shape-dependent computation, etc.

This PR extends auto_shard.py to first run a preprocessing step which wraps
any nn.Module which cannot be traced through. It adds a test for dynamic
conditionals and updates existing failing test code.

There are some immediate extensions to this approach which are marked as
TODO in the code.
      7bdf50a3
  13. 21 Oct, 2021 2 commits
  15. 20 Oct, 2021 3 commits
  17. 19 Oct, 2021 1 commit
  19. 28 Sep, 2021 1 commit
  21. 24 Sep, 2021 1 commit
  23. 22 Sep, 2021 1 commit
    • tmarkstrum's avatar
      Switch default branch from master to main (#807) · b09ddb2d
      tmarkstrum authored 
      * update master branch to main

* added FAQ about updating the branch from master to main

* fixed some false positive correction

* added what is new section

* fixed the quoted code area

* added release what is new section

* added a step in release.md

* fixed a word
      b09ddb2d
  25. 21 Sep, 2021 1 commit
  27. 20 Sep, 2021 1 commit
  29. 17 Sep, 2021 1 commit
  31. 13 Sep, 2021 1 commit
  33. 12 Sep, 2021 2 commits
    • Min Xu's avatar
      [fix] minor fixes for master branch (#792) · 31e36453
      Min Xu authored 
      

* add changelog for previous commit

* add changelog for previous commit

* add changelog for previous commit

* fix a merge induced error
Co-authored-by: default avatarMin Xu <min.xu.public@gmail.com>
      31e36453
    • Darryl Barnhart's avatar
      [fix] FSDP intra-backwards gradient accumulation. (#784) · 4fa2ab9b
      Darryl Barnhart authored 
      * [fix] FSDP intra-backwards gradient accumulation.

Ensure gradient reduction accumulates into the unsharded gradient tensor
within a backwards pass. This matters when an FSDP module is called
multiple times within a forward pass, and reduction is _not_ deferred
using activation checkpoint forward counters, bucketing or some other
mechanism.

Closes #780

* [refactor] Remove forward counters. Comments.

Removed forward counters from the activation checkpointing utility, now
that FSDP does not require them for correct operation. Add more detailed
comment about memory usage behaviour with gradient reduction.

* [refactor] Delete deprecated forward counter usage.

* [refactor] Add state assertion as end of pre-backward hook.
      4fa2ab9b
  35. 11 Sep, 2021 1 commit
    • Alex Xiao's avatar
      [feat] set requires_grad of output tensors of checkpointed modules properly (#787) · 482944d9
      Alex Xiao authored 
      

Before this commit, output tensors of checkpointed modules always
require grad, even if they shouldn't. This commit makes it so that
the outputs of checkpointed modules only require grad if either
the input requires grad or if the parameters require grad.

To achieve this, this commit also adds a new _unflattened_param_views
attribute to modules being flattened. This allows the checkpointing
to still access the parameters and check if gradients need to be
computed.
Co-authored-by: default avatarAlex Xiao <axiao@fb.com>
      482944d9
  37. 10 Sep, 2021 2 commits
  39. 07 Sep, 2021 1 commit
  41. 06 Sep, 2021 2 commits
  43. 05 Sep, 2021 1 commit
  45. 18 Aug, 2021 1 commit
  47. 12 Aug, 2021 4 commits