Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Apr 15, 2025

• Add @requires_multicast_support to test_multimem_all_gather (#151227)

  Show description for 7e5f6dc
  

  pragupta

  authored and

  pytorchmergebot

  committed
  7e5f6dc

  Copy full SHA for 7e5f6dc

Commits on Apr 9, 2025

• add reduce_scatter to symm mem ops (#150813)

  Show description for d04a6ec
  

  ngimel

  authored and

  pytorchmergebot

  committed
  d04a6ec

  Copy full SHA for d04a6ec

Commits on Apr 1, 2025

• Add one_shot_all_reduce_copy to allow non-symm-mem allocated tensors to be reduced (#150129)

  Show description for 1700599
  

  ngimel

  authored and

  pytorchmergebot

  committed
  1700599

  Copy full SHA for 1700599

• enable out variant of 2-shot reduction (#150153)

  Show description for 414b9ae
  

  ngimel

  authored and

  pytorchmergebot

  committed
  414b9ae

  Copy full SHA for 414b9ae

Commits on Mar 31, 2025

• Revert "enable out variant of 2-shot reduction (#150153)"

  Show description for 57fa99c
  pytorchmergebot

  committed
  57fa99c

  Copy full SHA for 57fa99c

• Revert "Add one_shot_all_reduce_copy to allow non-symm-mem allocated tensors to be reduced (#150129)"

  Show description for e57fa18
  pytorchmergebot

  committed
  e57fa18

  Copy full SHA for e57fa18

Commits on Mar 28, 2025

• enable out variant of 2-shot reduction (#150153)

  Show description for cdeb32d
  

  ngimel

  authored and

  pytorchmergebot

  committed
  cdeb32d

  Copy full SHA for cdeb32d

• Add one_shot_all_reduce_copy to allow non-symm-mem allocated tensors to be reduced (#150129)

  Show description for 8a87226
  

  ngimel

  authored and

  pytorchmergebot

  committed
  8a87226

  Copy full SHA for 8a87226

Commits on Jan 28, 2025

• [SymmetricMemory] fix an issue where rendezvous is performed with wrong device context when torch.cuda.set_device() is not callled (#144886)

  Show description for db33d23
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  db33d23

  Copy full SHA for db33d23

Commits on Jan 2, 2025

• ROCm: Enable 4 gpu tests for distributed config (#140319)

  Show description for 8f3eb84
  

  jagadish-amd

  authored and

  pytorchmergebot

  committed
  8f3eb84

  Copy full SHA for 8f3eb84

Commits on Dec 17, 2024

• [fused_all_gather_matmul] use _multimem_all_gather_matmul for small global Ms (#143160)

  Show description for af19047
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  af19047

  Copy full SHA for af19047

• [SymmetricMemory] introduce multimem_all_gather (#142810)

  Show description for 6fae60a
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  6fae60a

  Copy full SHA for 6fae60a

Commits on Dec 13, 2024

• Fix unused Python variables in test/[a-d]* (#134665)

  Show description for d25e6e6
  

  rec

  authored and

  pytorchmergebot

  committed
  d25e6e6

  Copy full SHA for d25e6e6

• Enable cutlass-based all-gather matmul when TORCH_SYMM_MEM_ENABLE_NATIVE_ASYNC_TP is set (#142283)

  Show description for 810808d
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  810808d

  Copy full SHA for 810808d

Commits on Dec 5, 2024

• [SymmetricMemory] use the python version of empty() and rendezvous() for tests and library ops (#142154)

  Show description for 5513e2e
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  5513e2e

  Copy full SHA for 5513e2e

Commits on Nov 26, 2024

• [torch/distributed] Make _SymmetricMemory.has_multicast_support() ret… (#141598)

  Show description for 5b0b16c
  

  Kiuk Chung

  authored and

  pytorchmergebot

  committed
  5b0b16c

  Copy full SHA for 5b0b16c

Commits on Nov 21, 2024

• Implements user buffer registration using MemPool (#133603)

  Show description for e0482fd
  

  syed-ahmed

  authored and

  pytorchmergebot

  committed
  e0482fd

  Copy full SHA for e0482fd

Commits on Nov 19, 2024

• Revert "Implements user buffer registration using MemPool (#133603)"

  Show description for 496c1e7
  pytorchmergebot

  committed
  496c1e7

  Copy full SHA for 496c1e7

Commits on Nov 17, 2024

• [SymmetricMemory] introduce user-facing APIs empty() and rendezvous() (#139677)

  Show description for 5a7e147
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  5a7e147

  Copy full SHA for 5a7e147

• [SymmetricMemory] support specifying group_name at rendezvous time (#139529)

  Show description for ab5c885
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  ab5c885

  Copy full SHA for ab5c885

Commits on Nov 15, 2024

• Implements user buffer registration using MemPool (#133603)

  Show description for 25d9be3
  

  syed-ahmed

  authored and

  pytorchmergebot

  committed
  25d9be3

  Copy full SHA for 25d9be3

Commits on Nov 14, 2024

• [SymmetricMemory] fix a bug in symm_mem::memset32_ where the ops fails when offset=0 (#140129)

  Show description for 02d0c43
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  02d0c43

  Copy full SHA for 02d0c43

• [SymmetricMemory] fix a bug where get_signal_pad() returns a tensor backed by a buffer ptr instead of a signal_pad ptr (#140128)

  Show description for 684db9b
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  684db9b

  Copy full SHA for 684db9b

• [SymmetricMemory] allow overlapping devices for testing (#140127)

  Show description for c3d61bd
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  c3d61bd

  Copy full SHA for c3d61bd

Commits on Nov 11, 2024

• [SymmetricMemory] improve the API for stream_write_value32 (#139934)

  Show description for 0a0915f
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  0a0915f

  Copy full SHA for 0a0915f

• Revert "[SymmetricMemory] improve the API for stream_write_value32 (#139934)"

  Show description for 5f4a21d
  pytorchmergebot

  committed
  5f4a21d

  Copy full SHA for 5f4a21d

• [SymmetricMemory] improve the API for stream_write_value32 (#139934)

  Show description for 2f3a5a1
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  2f3a5a1

  Copy full SHA for 2f3a5a1

Commits on Nov 8, 2024

• [experimental] async-tp impl with cutlass-based, progress aware kernel (#139227)

  Show description for 1659e24
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  1659e24

  Copy full SHA for 1659e24

Commits on Nov 7, 2024

• Revert "[experimental] async-tp impl with cutlass-based, progress aware kernel (#139227)"

  Show description for 36e0f11
  pytorchmergebot

  committed
  36e0f11

  Copy full SHA for 36e0f11

• [experimental] async-tp impl with cutlass-based, progress aware kernel (#139227)

  Show description for 5203138
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  5203138

  Copy full SHA for 5203138

Commits on Nov 5, 2024

• [SymmetricMemory] introduce a binding for cuMemset32Async (#138755)

  Show description for ee42a99
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  ee42a99

  Copy full SHA for ee42a99

Commits on Nov 4, 2024

• Revert "[SymmetricMemory] introduce a binding for cuMemset32Async (#138755)"

  Show description for 3ca7947
  pytorchmergebot

  committed
  3ca7947

  Copy full SHA for 3ca7947

Commits on Nov 3, 2024

• [SymmetricMemory] introduce a binding for cuMemset32Async (#138755)

  Show description for 924e726
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  924e726

  Copy full SHA for 924e726

Commits on Nov 1, 2024

• [SymmetricMemory] expose signal_pads as tensors in Python (#138754)

  Show description for 0dbc284
  

  Yifu Wang

  authored and

  pytorchmergebot

  committed
  0dbc284

  Copy full SHA for 0dbc284

• [ROCM][CUDA][NCCL] Disable `test_lowering_one_shot_all_reduce` on ROCM (#139414)

  Show description for 6fc63b4
  

  eqy

  authored and

  pytorchmergebot

  committed
  6fc63b4

  Copy full SHA for 6fc63b4

Pagination

Previous

Next