Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Sep 21, 2022

• ignore version check

  yytdfc

  authored
  50f949a

  Copy full SHA for 50f949a

Commits on Sep 20, 2022

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

  Show description for 52c5128
  jpool-nv

  authored
  52c5128

  Copy full SHA for 52c5128

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

  Show description for 2b0e837
  timmoon10

  authored
  2b0e837

  Copy full SHA for 2b0e837

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

• Update test_fused_layer_norm.py (NVIDIA#1485)

  Show description for 136a13c
  Fuzzkatt

  authored
  136a13c

  Copy full SHA for 136a13c

• 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 30, 2022

• [fmha] fixes spurious illegal memaccess (NVIDIA#1469)

  Show description for 21e4154
  yjk21

  authored
  21e4154

  Copy full SHA for 21e4154

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

• Optimize distributed Adam kernels and implementation (NVIDIA#1463)

  Show description for 6f7d8ac
  timmoon10

  authored
  6f7d8ac

  Copy full SHA for 6f7d8ac

Commits on Aug 24, 2022

• Label smoothing in vocab parallel cross entropy (NVIDIA#1457)

  Show description for 8cea0b4
  MaximumEntropy

  authored
  8cea0b4

  Copy full SHA for 8cea0b4

Commits on Aug 22, 2022

• Permutation tests (NVIDIA#1410)

  Show description for 28a96cf
  jpool-nv

  authored
  28a96cf

  Copy full SHA for 28a96cf

Commits on Aug 18, 2022

• Merge pull request NVIDIA#1460 from NVIDIA/asp_fix_torchvision_ops_misc

  Show description for 9866160
  ChongyuNVIDIA

  authored
  9866160

  Copy full SHA for 9866160

• Fix the torchvision.ops.misc.Conv2d error as Torchvision remove deprecated APIs in 0.12.0

  ChongyuNVIDIA

  committed
  23f80a2

  Copy full SHA for 23f80a2

Commits on Aug 16, 2022

• Merge pull request NVIDIA#1459 from NVIDIA/faster_2way_halo_transfers

  Show description for 78be3a7
  thorjohnsen

  authored
  78be3a7

  Copy full SHA for 78be3a7

Commits on Aug 15, 2022

• Fixed peer halo exchange module test

  thorjohnsen

  committed
  07e3632

  Copy full SHA for 07e3632

• 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 10, 2022

• Store optimizer state_dict in uint8 tensor instead of bytearray (NVIDIA#1446)

  Show description for a0f5f3a
  timmoon10

  authored
  a0f5f3a

  Copy full SHA for a0f5f3a

Commits on Aug 9, 2022

• Update mlp_cuda test (NVIDIA#1425)

  Show description for 71e5871
  crcrpar

  authored
  71e5871

  Copy full SHA for 71e5871

• fix: apply dropout only in training mode (NVIDIA#1449)

  yjk21

  authored
  26ff4e4

  Copy full SHA for 26ff4e4

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

  Show description for 2f67d22
  crcrpar

  authored
  2f67d22

  Copy full SHA for 2f67d22

Commits on Aug 5, 2022

• Add a generic fused softmax (NVIDIA#1440)

  Show description for cd0a1f1
  yidong72

  authored
  cd0a1f1

  Copy full SHA for cd0a1f1

Commits on Aug 4, 2022

• support for full mask in fused softmax kernel + avoiding inplace operation in bwd pass (NVIDIA#1444)

  Show description for 31cbdd1
  kvareddy

  authored
  31cbdd1

  Copy full SHA for 31cbdd1

Commits on Aug 1, 2022

• add customized fused op index mulitiplication (NVIDIA#1438)

  Show description for a4f4258
  

  BaoHhhhhhan

  and

  Han Bao

  authored
  a4f4258

  Copy full SHA for a4f4258

Commits on Jul 28, 2022

• Sequence parallel perf updates (NVIDIA#1437)

  Show description for 3c19f10
  ericharper

  authored
  3c19f10

  Copy full SHA for 3c19f10

Commits on Jul 26, 2022

• Improvements in distributed Adam optimizer for Megatron (NVIDIA#1432)

  Show description for 2e025ab
  timmoon10

  authored
  2e025ab

  Copy full SHA for 2e025ab

• Fix bug when initializing model-parallel process groups for GPT-3 (NVIDIA#1435)

  Show description for fb21698
  timmoon10

  authored
  fb21698

  Copy full SHA for fb21698

Commits on Jul 25, 2022

• [transformer] update tests (NVIDIA#1428)

  Aidyn-A

  authored
  e57d9e7

  Copy full SHA for e57d9e7

Pagination

Previous

Next