Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Jul 7, 2023

• [Transformer][UCC] Fix async p2p ops (#1691)

  Show description for 7b2e71b
  Aidyn-A

  authored
  7b2e71b

  Copy full SHA for 7b2e71b

Commits on Jul 2, 2023

• Tkurth/sgbn fixes (#1685)

  Show description for 8ffc901
  azrael417

  authored
  8ffc901

  Copy full SHA for 8ffc901

Commits on Jun 22, 2023

• Tkurth/mplamb fixed (#1684)

  azrael417

  authored
  30a7ad3

  Copy full SHA for 30a7ad3

Commits on Jun 12, 2023

• Add `pyproject.toml` (#1669)

  Show description for 2d8302a
  crcrpar

  authored
  2d8302a

  Copy full SHA for 2d8302a

Commits on Jun 9, 2023

• do not use batch_isend_irecv for UCC (#1675)

  Aidyn-A

  authored
  82ee367

  Copy full SHA for 82ee367

Commits on Jun 2, 2023

• [Test][Transformer] Pre-parse container version (#1673)

  Show description for 05091d4
  Aidyn-A

  authored
  05091d4

  Copy full SHA for 05091d4

Commits on Jun 1, 2023

• tensor.type() is deprecated, we should use scalar_type() (#1672)

  Show description for ecb5614
  

  ycsos

  and

  yuchuang

  authored
  ecb5614

  Copy full SHA for ecb5614

Commits on May 13, 2023

• Update distributed optimizer with new coalescing manager API in PyTorch (#1663)

  Show description for 8b7a1ff
  

  timmoon10

  and

  crcrpar

  authored
  8b7a1ff

  Copy full SHA for 8b7a1ff

Commits on May 12, 2023

• allow for custom directory of xml reports (#1661)

  Show description for 2d876b8
  crcrpar

  authored
  2d876b8

  Copy full SHA for 2d876b8

Commits on Apr 27, 2023

• Using regular layernorm for ONNX and TS export (#1634)

  Show description for 85e9edd
  

  borisfom

  and

  crcrpar

  authored
  85e9edd

  Copy full SHA for 85e9edd

• [Transformer] Update p2p communication routine (#1650)

  Aidyn-A

  authored
  f2d6f29

  Copy full SHA for f2d6f29

Commits on Apr 26, 2023

• [Transformer][test/L0] Increase max world_size for BERT and GPT (#1651)

  Aidyn-A

  authored
  5b5d410

  Copy full SHA for 5b5d410

Commits on Apr 21, 2023

• Mention `Ninja` for faster build (#1607)

  Show description for b3e3bab
  crcrpar

  authored
  b3e3bab

  Copy full SHA for b3e3bab

Commits on Apr 19, 2023

• [Transformer] Manipulate NCCL_NET for hybrid IB/Socket setups (#1620)

  Show description for 817e818
  Aidyn-A

  authored
  817e818

  Copy full SHA for 817e818

Commits on Apr 18, 2023

• use nvcr 23.03 (#1644)

  ZhiyuanChen

  authored
  eae1ddf

  Copy full SHA for eae1ddf

• add bf16 support for FusedDense (#1627)

  Agoniii

  authored
  1ba89ee

  Copy full SHA for 1ba89ee

• include stdint.h & stdio.h in fast_layer_norm/ln.h (#1637)

  Show description for 4e1ae43
  ZhiyuanChen

  authored
  4e1ae43

  Copy full SHA for 4e1ae43

Commits on Apr 14, 2023

• Add support for FP32 master weights in FusedAdam optimizer (#1625)

  Show description for 6bd01c4
  

  4 peopleauthored
  6bd01c4

  Copy full SHA for 6bd01c4

Commits on Mar 31, 2023

• [apex.contrib.sparsity] Grouped conv permutations (#1628)

  Show description for 89cc215
  jpool-nv

  authored
  89cc215

  Copy full SHA for 89cc215

Commits on Mar 29, 2023

• create mpi proc group: (#1626)

  erhoo82

  authored
  57057e2

  Copy full SHA for 57057e2

Commits on Mar 23, 2023

• Overlap first param sync with optimizer step in distributed Adam (#1621)

  Show description for 7150e20
  timmoon10

  authored
  7150e20

  Copy full SHA for 7150e20

• check for empty params in group (#1596)

  Show description for 0420735
  

  3 peopleauthored
  0420735

  Copy full SHA for 0420735

Commits on Mar 22, 2023

• Revert "Add support for FP32 master weights in FusedAdam optimizer (#1623)" (#1624)

  Show description for 4060aef
  crcrpar

  authored
  4060aef

  Copy full SHA for 4060aef

• Add support for FP32 master weights in FusedAdam optimizer (#1623)

  Show description for 6952004
  

  3 peopleauthored
  6952004

  Copy full SHA for 6952004

Commits on Mar 20, 2023

• p2p communication overlap at interleaved pipelining (#1616)

  Show description for d8643ef
  erhoo82

  authored
  d8643ef

  Copy full SHA for d8643ef

Commits on Mar 17, 2023

• Align distopt communication in interleaved pipeline parallelism (#1611)

  Show description for 27de66c
  

  timmoon10

  and

  crcrpar

  authored
  27de66c

  Copy full SHA for 27de66c

• [fmha] Use 8bit comparison to generate dropout mask (#1614)

  vasunvidia

  authored
  07b2935

  Copy full SHA for 07b2935

Commits on Mar 13, 2023

• ConvFrozenScaleBiasReLU fusion (#1608)

  Show description for 623315a
  minitu

  authored
  623315a

  Copy full SHA for 623315a

Commits on Mar 7, 2023

• [FMHA] Remove zero fill for softmax output (#1565)

  Show description for b1c7600
  vasunvidia

  authored
  b1c7600

  Copy full SHA for b1c7600

• s/maximimize/maximize/ (#1604)

  Show description for be819ea
  EIFY

  authored
  be819ea

  Copy full SHA for be819ea

Commits on Mar 6, 2023

• gracefully skipping permutations when torch.fx fails tracing (#1595)

  jpool-nv

  authored
  e29fb45

  Copy full SHA for e29fb45

Commits on Feb 28, 2023

• Update layers.py to ensure grad_output is contiguous (#1601)

  Show description for 95cea8e
  markelsanz14

  authored
  95cea8e

  Copy full SHA for 95cea8e

Commits on Feb 25, 2023

• use a single communicator for amax reduction in both TP and DP domains (#1597)

  erhoo82

  authored
  eec7250

  Copy full SHA for eec7250

• Decouple optimizer state and grad dtypes in distributed Adam optimizer (#1575)

  Show description for 03c9d80
  timmoon10

  authored
  03c9d80

  Copy full SHA for 03c9d80

Commits on Feb 16, 2023

• Use a separate communicator for AMAX reduction (#1585)

  Show description for 0c8400a
  erhoo82

  authored
  0c8400a

  Copy full SHA for 0c8400a