Commits

Branch selector

 master

User selector

fegin

Datepicker

All time

Commit History

Commits on Oct 3, 2022

• [FSDP] Delay moving tensor to CPU until necessary for optim_state_dict() (#85761)

  Show description for 2067b76
  

  fegin

  authored and

  pytorchmergebot

  committed
  2067b76

  Copy full SHA for 2067b76

Commits on Oct 1, 2022

• [FSDP] Ignore buffers that are non-persistent. (#85740)

  Show description for be29ca9
  

  fegin

  authored and

  pytorchmergebot

  committed
  be29ca9

  Copy full SHA for be29ca9

Commits on Sep 28, 2022

• [FSDP] Handle the state_dict on CPU cases (#85640)

  Show description for 1c1f3a9
  

  fegin

  authored and

  pytorchmergebot

  committed
  1c1f3a9

  Copy full SHA for 1c1f3a9

Commits on Sep 9, 2022

• [FSDP] Optimizer states may be on CPU, copy them to GPU before gathering (#84708)

  Show description for 28c830a
  

  fegin

  authored and

  pytorchmergebot

  committed
  28c830a

  Copy full SHA for 28c830a

• [FSDP] Ensure that all ranks use the same order to iterate through optimizer states (#84654)

  Show description for 1840f24
  

  fegin

  authored and

  pytorchmergebot

  committed
  1840f24

  Copy full SHA for 1840f24

Commits on Aug 31, 2022

• [FSDP] Fix the FQN not found issue for load sharded_state_dict when using activation checkpoint (#84253)

  Show description for 305c6a6
  

  fegin

  authored and

  pytorchmergebot

  committed
  305c6a6

  Copy full SHA for 305c6a6

Commits on Aug 18, 2022

• [FSDP] Implement sharded_optim_state_dict and flatten_sharded_optim_state_dict. (#77628)

  Show description for 3e1fc85
  

  fegin

  authored and

  pytorchmergebot

  committed
  3e1fc85

  Copy full SHA for 3e1fc85

Commits on Aug 17, 2022

• [FSDP] Use _init_from_local_tensor to create ShardedTensor to avoid communication overhead (#82911)

  Show description for 2446902
  

  fegin

  authored and

  pytorchmergebot

  committed
  2446902

  Copy full SHA for 2446902

Commits on Aug 3, 2022

• [FSDP] Move the sharded_state_dict logic to the post hook to avoid OOM (#82613)

  Show description for b750c10
  

  fegin

  authored and

  pytorchmergebot

  committed
  b750c10

  Copy full SHA for b750c10

Commits on Aug 2, 2022

• [FSDP] Implement _param_fqns() to return all parameter FQNs for the FSDP module (#82595)

  Show description for c05c395
  

  fegin

  authored and

  pytorchmergebot

  committed
  c05c395

  Copy full SHA for c05c395

Commits on Jul 15, 2022

• [FSDP] Construct FQN in _full_post_state_dict_hook (#81253)

  Show description for 3ea1b9b
  

  fegin

  authored and

  pytorchmergebot

  committed
  3ea1b9b

  Copy full SHA for 3ea1b9b

Commits on Jul 5, 2022

• Consolidate the naming of named_parameter and state_dict for CheckpointWrapper (#80089)

  Show description for e0eeb06
  

  fegin

  authored and

  pytorchmergebot

  committed
  e0eeb06

  Copy full SHA for e0eeb06

Commits on May 16, 2022

• [FSDP] Add unittests to test distributed checkpointing

  Show description for a0cc38e
  

  fegin

  authored and

  pytorchmergebot

  committed
  a0cc38e

  Copy full SHA for a0cc38e

Commits on May 15, 2022

• [FSDP] Implement sharded_state_dict and load_sharded_state_dict

  Show description for 58c9d52
  

  fegin

  authored and

  pytorchmergebot

  committed
  58c9d52

  Copy full SHA for 58c9d52

Commits on May 11, 2022

• [FSDP] Fix local_state_dict and state_dict_type bugs

  Show description for 4f4ebc5
  

  fegin

  authored and

  pytorchmergebot

  committed
  4f4ebc5

  Copy full SHA for 4f4ebc5

Commits on Apr 12, 2022

• [FSDP] Implement reshard_flatten_tensor

  Show description for 577c9ff
  

  fegin

  authored and

  pytorchmergebot

  committed
  577c9ff

  Copy full SHA for 577c9ff

• [FSDP] Avoid using the legacy _sharded_tensor package

  Show description for c2d4e20
  

  fegin

  authored and

  pytorchmergebot

  committed
  c2d4e20

  Copy full SHA for c2d4e20

Commits on Mar 31, 2022

• [FSDP] Register state_dict hooks for FlatParamsWrapper even if params_list is empty (#74860)

  Show description for fd4ad5d
  

  fegin

  authored and

  pytorchmergebot

  committed
  fd4ad5d

  Copy full SHA for fd4ad5d

Commits on Mar 8, 2022

• [FSDP] Provide a utility API to allow users easily to set state_dict_type (#73716)

  Show description for be535e2
  

  fegin

  authored and

  pytorchmergebot

  committed
  be535e2

  Copy full SHA for be535e2

• [FSDP] Generalize fsdp_modules() (#73553)

  Show description for 5372dcd
  

  fegin

  authored and

  pytorchmergebot

  committed
  5372dcd

  Copy full SHA for 5372dcd

Commits on Mar 3, 2022

• [FSDP] Add the unittests for the _replace_by_prefix (#73530)

  Show description for e7a786f
  

  fegin

  authored and

  pytorchmergebot

  committed
  e7a786f

  Copy full SHA for e7a786f

Commits on Mar 1, 2022

• [FSDP] Make summon_full_params a public method (#73116)

  Show description for 6396547
  

  fegin

  authored and

  pytorchmergebot

  committed
  6396547

  Copy full SHA for 6396547

Commits on Feb 25, 2022

• [FSDP][Reland] Implement local_state_dict and load_local_state_dict (#73300)

  Show description for dc1bd97
  

  fegin

  authored and

  pytorchmergebot

  committed
  dc1bd97

  Copy full SHA for dc1bd97

Commits on Feb 19, 2022

• [FSDP] Implement local_state_dict and load_local_state_dict (#72469)

  Show description for d50643a
  

  fegin

  authored and

  pytorchmergebot

  committed
  d50643a

  Copy full SHA for d50643a

Commits on Feb 12, 2022

• [FSDP] Use unflatten_parameter in _summon_full_parameters (#72467)

  Show description for c73cc92
  

  fegin

  authored and

  pytorchmergebot

  committed
  c73cc92

  Copy full SHA for c73cc92

Commits on Feb 7, 2022

• [FSDP] Add FlatParameter to track the information of a flat parameter (#69241)

  Show description for 224093d
  

  fegin

  authored and

  pytorchmergebot

  committed
  224093d

  Copy full SHA for 224093d