Commits

Branch selector

 master

User selector

crcrpar

Datepicker

All time

Commit History

Commits on May 12, 2023

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

  Show description for 2d876b8
  crcrpar

  authored
  2d876b8

  Copy full SHA for 2d876b8

Commits on Apr 21, 2023

• Mention Ninja for faster build (NVIDIA#1607)

  Show description for b3e3bab
  crcrpar

  authored
  b3e3bab

  Copy full SHA for b3e3bab

Commits on Mar 22, 2023

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

  Show description for 4060aef
  crcrpar

  authored
  4060aef

  Copy full SHA for 4060aef

Commits on Feb 16, 2023

• use torch.tensor to create a tensor with initializer values (NVIDIA#1588)

  Show description for 6943fd2
  crcrpar

  authored
  6943fd2

  Copy full SHA for 6943fd2

Commits on Feb 9, 2023

• decrease warnings & remove append_nvcc_threads from setup.py (NVIDIA#1581)

  Show description for 0145d69
  crcrpar

  authored
  0145d69

  Copy full SHA for 0145d69

Commits on Jan 26, 2023

• bump cutlass to v2.11.0 for contrib/multihead_attn (NVIDIA#1570)

  Show description for 536e549
  crcrpar

  authored
  536e549

  Copy full SHA for 536e549

Commits on Dec 9, 2022

• change param of list of tensor to tensor for _make_nccl_premul_sum (NVIDIA#1542)

  Show description for 91fcaaf
  crcrpar

  authored
  91fcaaf

  Copy full SHA for 91fcaaf

Commits on Nov 21, 2022

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

  Show description for c216175
  

  crcrpar

  and

  mikolajblaz

  authored
  c216175

  Copy full SHA for c216175

Commits on Nov 14, 2022

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

  Show description for a9950b1
  crcrpar

  authored
  a9950b1

  Copy full SHA for a9950b1

Commits on Oct 27, 2022

• update exception message (NVIDIA#1524)

  Show description for 6a40a0a
  crcrpar

  authored
  6a40a0a

  Copy full SHA for 6a40a0a

• check is_ucc_available (NVIDIA#1523)

  Show description for e76ade0
  crcrpar

  authored
  e76ade0

  Copy full SHA for e76ade0

Commits on Oct 25, 2022

• Partial activation checkpoint (NVIDIA#1511)

  Show description for ee42e2d
  

  crcrpar

  and

  erhoo82

  authored
  ee42e2d

  Copy full SHA for ee42e2d

Commits on Oct 12, 2022

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

  Show description for 806f9b0
  

  3 peopleauthored
  806f9b0

  Copy full SHA for 806f9b0

Commits on Oct 10, 2022

• Deprecate outdated features (NVIDIA#1506)

  Show description for 48bf0df
  crcrpar

  authored
  48bf0df

  Copy full SHA for 48bf0df

Commits on Oct 6, 2022

• check CUDA_HOME before using it (NVIDIA#1499)

  Show description for b213772
  crcrpar

  authored
  b213772

  Copy full SHA for b213772

• Update for CUDA 11.8 (NVIDIA#1502)

  Show description for 5f1a73e
  

  3 peopleauthored
  5f1a73e

  Copy full SHA for 5f1a73e

Commits on Sep 16, 2022

• introducing APEX_RUN_WITH_SLOW_TESTS env var (NVIDIA#1489)

  Show description for 6b5405e
  crcrpar

  authored
  6b5405e

  Copy full SHA for 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

  Copy full SHA for a091a3c

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

  Show description for 5ba7af1
  crcrpar

  authored
  5ba7af1

  Copy full SHA for 5ba7af1

Commits on Sep 10, 2022

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

  Show description for ce9df7d
  crcrpar

  authored
  ce9df7d

  Copy full SHA for 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

  Copy full SHA for 70dbc20

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

  Show description for abf78c7
  crcrpar

  authored
  abf78c7

  Copy full SHA for abf78c7

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

  Show description for 81c9aca
  crcrpar

  authored
  81c9aca

  Copy full SHA for 81c9aca

Commits on Sep 8, 2022

• unittest-ish (NVIDIA#1478)

  Show description for 19ca350
  crcrpar

  authored
  19ca350

  Copy full SHA for 19ca350

• [contrib] cuDNN based GBN (NVIDIA#1476)

  Show description for 7a12ee6
  

  crcrpar

  and

  nvcastet

  authored
  7a12ee6

  Copy full SHA for 7a12ee6

Commits on Sep 7, 2022

• bump cudnn-frontend to v0.7.1, released on Aug. 26th (NVIDIA#1473)

  Show description for 12e5bfa
  crcrpar

  authored
  12e5bfa

  Copy full SHA for 12e5bfa

Commits on Aug 25, 2022

• [contrib] Guard for torch.distributed._make_nccl_premul_sum (NVIDIA#1465)

  Show description for 7e90f95
  crcrpar

  authored
  7e90f95

  Copy full SHA for 7e90f95

Commits on Aug 15, 2022

• Use xmlrunner.XMLTestRunner accordingly in tests/L0/run_test.py (NVIDIA#1451)

  Show description for e75cb25
  crcrpar

  authored
  e75cb25

  Copy full SHA for e75cb25

Commits on Aug 9, 2022

• Update mlp_cuda test (NVIDIA#1425)

  Show description for 71e5871
  crcrpar

  authored
  71e5871

  Copy full SHA for 71e5871

• [transformer] unittest: less mem consumption for generic softmax (NVIDIA#1448)

  Show description for 2f67d22
  crcrpar

  authored
  2f67d22

  Copy full SHA for 2f67d22

Commits on Jul 14, 2022

• [contrib] Fix the reference implementation of multihead_attn (NVIDIA#1423)

  Show description for 809043f
  crcrpar

  authored
  809043f

  Copy full SHA for 809043f

Commits on Jul 7, 2022

• Remove pyprof and reparameterization (NVIDIA#1404)

  Show description for 8a7a332
  crcrpar

  authored
  8a7a332

  Copy full SHA for 8a7a332

Commits on Jun 23, 2022

• [transformer] Port Sequence Parallelism (takeover of NVIDIA#1396) (NVIDIA#1400)

  Show description for 3ff1a10
  crcrpar

  authored
  3ff1a10

  Copy full SHA for 3ff1a10

Commits on Jun 22, 2022

• Temporary Solution to Let FusedAdam support BFloat16 (NVIDIA#1407)

  Show description for 81f8ba7
  crcrpar

  authored
  81f8ba7

  Copy full SHA for 81f8ba7

Commits on May 18, 2022

• [transformer] Allow for different backend for Pipeline Parallel ProcessGroups (NVIDIA#1380)

  Show description for 3490b9e
  

  crcrpar

  and

  Aidyn-A

  authored
  3490b9e

  Copy full SHA for 3490b9e

Pagination

Previous

Next