Commits

Branch selector

 85af77c

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 8, 2022

• Refactor run_transformer bert minimal and gpt minimal tests (NVIDIA#1540)

  Show description for 85af77c
  Fuzzkatt

  authored
  85af77c

Commits on Nov 30, 2022

• Merge pull request NVIDIA#1543 from timmoon10/peer-halo-exchange-perf-optimization

  Show description for 190a446
  thorjohnsen

  authored
  190a446

Commits on Nov 22, 2022

• Update megatron fused softmax follow megatron-lm (NVIDIA#1539)

  Show description for abeca58
  

  yaoyu-33

  and

  Yu Yao

  authored
  abeca58

Commits on Nov 21, 2022

• Avoid index bookkeeping in halo exchange kernel if data is contiguous

  timmoon10

  committed
  fa34475

• [contrib] fix reduce_scatter_tensor fallouts in DistributedFusedLamb (NVIDIA#1536)

  Show description for c216175
  

  crcrpar

  and

  mikolajblaz

  authored
  c216175

Commits on Nov 18, 2022

• Optimize peer memory halo exchange kernel

  Show description for 0bd620e
  timmoon10

  committed
  0bd620e

Commits on Nov 17, 2022

• add setter for virtual world size (NVIDIA#1541)

  Show description for 3229126
  

  timmoon10

  and

  ericharper

  authored
  3229126

• Perform dist Adam communication in side stream (NVIDIA#1528)

  Show description for d79f69a
  timmoon10

  authored
  d79f69a

Commits on Nov 14, 2022

• [README] Add a list of install options (NVIDIA#1492)

  Show description for a9950b1
  crcrpar

  authored
  a9950b1

Commits on Oct 27, 2022

• update exception message (NVIDIA#1524)

  Show description for 6a40a0a
  crcrpar

  authored
  6a40a0a

• check is_ucc_available (NVIDIA#1523)

  Show description for e76ade0
  crcrpar

  authored
  e76ade0

• Remove weight gradients if model is frozen for linear layer (NVIDIA#1517)

  Show description for 36ffac0
  

  3 peopleauthored
  36ffac0

• update run_transformer tests to default to using pytorch native UCC instead of torch_ucc (NVIDIA#1495)

  Show description for 0d06a73
  Fuzzkatt

  authored
  0d06a73

• Support overlapped grad sync with Megatron interleaved pipeline parallelism (NVIDIA#1514)

  Show description for f683961
  timmoon10

  authored
  f683961

Commits on Oct 26, 2022

• Use all_gather_into_tensor and reduce_scatter_tensor (NVIDIA#1513)

  Show description for f680fab
  Aidyn-A

  authored
  f680fab

Commits on Oct 25, 2022

• Partial activation checkpoint (NVIDIA#1511)

  Show description for ee42e2d
  

  crcrpar

  and

  erhoo82

  authored
  ee42e2d

• Fix bug with grad clipping and distributed Adam (NVIDIA#1512)

  Show description for da9f5ae
  timmoon10

  authored
  da9f5ae

Commits on Oct 13, 2022

• add tearDown (NVIDIA#1508)

  Aidyn-A

  authored
  2c3b8f9

Commits on Oct 12, 2022

• [transformer] Allow for skipping stream sync (NVIDIA#1505)

  Show description for 806f9b0
  

  3 peopleauthored
  806f9b0

Commits on Oct 10, 2022

• Deprecate outdated features (NVIDIA#1506)

  Show description for 48bf0df
  crcrpar

  authored
  48bf0df

Commits on Oct 6, 2022

• check CUDA_HOME before using it (NVIDIA#1499)

  Show description for b213772
  crcrpar

  authored
  b213772

• Update for CUDA 11.8 (NVIDIA#1502)

  Show description for 5f1a73e
  

  3 peopleauthored
  5f1a73e

Commits on Oct 2, 2022

• Replace usage of internal with public APIs (NVIDIA#1496)

  Show description for 6816ef6
  kumpera

  authored
  6816ef6

Commits on Sep 22, 2022

• update fused_dense to use torch.autocast instead of apex amp (NVIDIA#1490)

  Show description for 1d77111
  Fuzzkatt

  authored
  1d77111

Commits on Sep 20, 2022

• [contrib][sparsity] Conv1 d support (NVIDIA#1466)

  Show description for 52c5128
  jpool-nv

  authored
  52c5128

• [contrib][DistributedFusedAdam] Support overlapped grad sync with Megatron pipeline parallelism (NVIDIA#1475)

  Show description for 2b0e837
  timmoon10

  authored
  2b0e837

Commits on Sep 16, 2022

• introducing APEX_RUN_WITH_SLOW_TESTS env var (NVIDIA#1489)

  Show description for 6b5405e
  crcrpar

  authored
  6b5405e

Commits on Sep 14, 2022

• [contrib] Skip peer_memory test if world_size is not a multiple of 2 (NVIDIA#1486)

  Show description for a091a3c
  crcrpar

  authored
  a091a3c

• Update test_fused_layer_norm.py (NVIDIA#1485)

  Show description for 136a13c
  Fuzzkatt

  authored
  136a13c

• Skip test if related contrib modules are not found (NVIDIA#1483)

  Show description for 5ba7af1
  crcrpar

  authored
  5ba7af1

Commits on Sep 10, 2022

• [contrib] xfail decorator to some of transducer tests (NVIDIA#1482)

  Show description for ce9df7d
  crcrpar

  authored
  ce9df7d

Commits on Sep 9, 2022

• [contrib][fast_layer_norm] test_ to _test_impl for pytest to not recognize (NVIDIA#1481)

  Show description for 70dbc20
  crcrpar

  authored
  70dbc20

• [contrib] bottleneck test refactoring (NVIDIA#1479)

  Show description for abf78c7
  crcrpar

  authored
  abf78c7

• Update apex.mlp to use fp16 in autocast (NVIDIA#1477)

  Show description for 81c9aca
  crcrpar

  authored
  81c9aca

Commits on Sep 8, 2022

• unittest-ish (NVIDIA#1478)

  Show description for 19ca350
  crcrpar

  authored
  19ca350

Pagination

PreviousNext