Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 19, 2024

• fix local float8 tests on H100 (pytorch#1438)

  Show description for 38c79d4
  vkuzo

  authored
  38c79d4

  Copy full SHA for 38c79d4

Commits on Dec 4, 2024

• [float8] Allow specifying arbitrary dtype for each tensor (pytorch#1378)

  lw

  authored
  4f8021f

  Copy full SHA for 4f8021f

Commits on Dec 2, 2024

• check scale.ndim before applying t/transpose (pytorch#1339)

  Show description for 65b885f
  crcrpar

  authored
  65b885f

  Copy full SHA for 65b885f

Commits on Dec 1, 2024

• Update hardware check conditions (pytorch#1356)

  jainapurva

  authored
  22bec74

  Copy full SHA for 22bec74

Commits on Nov 28, 2024

• Add support for quantize_() with Float8Linear module (pytorch#1344)

  jainapurva

  authored
  c45d975

  Copy full SHA for c45d975

Commits on Nov 25, 2024

• float8 training: move module attribute setting to sync function (pytorch#1341)

  Show description for b65e513
  vkuzo

  authored
  b65e513

  Copy full SHA for b65e513

Commits on Nov 22, 2024

• fix bug with float8 training + FSDP2 + TP (pytorch#1327)

  Show description for 8f73e84
  vkuzo

  authored
  8f73e84

  Copy full SHA for 8f73e84

Commits on Nov 19, 2024

• float8 with delayed scaling: fix autocast handling (pytorch#1306)

  Show description for b714026
  vkuzo

  authored
  b714026

  Copy full SHA for b714026

Commits on Nov 18, 2024

• Lint fixes for test/float8 (pytorch#1303)

  jainapurva

  authored
  6234116

  Copy full SHA for 6234116

Commits on Nov 15, 2024

• float8 delayed scaling: remove need to use workaround for AC (pytorch#1291)

  Show description for 56bf2e8
  vkuzo

  authored
  56bf2e8

  Copy full SHA for 56bf2e8

Commits on Nov 14, 2024

• make float 8 error message test BC (pytorch#1284)

  Show description for 5effac5
  HDCharles

  authored
  5effac5

  Copy full SHA for 5effac5

Commits on Nov 12, 2024

• Update error message to unbreak main (pytorch#1274)

  Show description for f96e5ec
  HDCharles

  authored
  f96e5ec

  Copy full SHA for f96e5ec

Commits on Oct 31, 2024

• Add an option to use fp8-all-gather only without fp8 computation.

  Show description for 2761917
  y-sq

  authored
  2761917

  Copy full SHA for 2761917

Commits on Oct 24, 2024

• Move float8_aten_api to float8_ops (pytorch#1155)

  jainapurva

  authored
  79f364f

  Copy full SHA for 79f364f

Commits on Oct 23, 2024

• [reland][ROCm] use dataclass for fnuz type setting (pytorch#1150)

  jeffdaily

  authored
  75d0693

  Copy full SHA for 75d0693

• Revert "[ROCm] use dataclass for fnuz type setting" (pytorch#1148)

  Show description for d252612
  msaroufim

  authored
  d252612

  Copy full SHA for d252612

• [ROCm] use dataclass for fnuz type setting (pytorch#1142)

  Show description for eb1fb3a
  

  jeffdaily

  and

  msaroufim

  authored
  eb1fb3a

  Copy full SHA for eb1fb3a

Commits on Oct 7, 2024

• float8 training axiswise scaling support with per-gemm-argument configuration (pytorch#940)

  Show description for dec0313
  vkuzo

  authored
  dec0313

  Copy full SHA for dec0313

• add axiswise scaling to Float8Linear (pytorch#920)

  Show description for e76db70
  vkuzo

  authored
  e76db70

  Copy full SHA for e76db70

• add axiswise granularity to Float8Tensor (pytorch#919)

  Show description for 52d27a1
  vkuzo

  authored
  52d27a1

  Copy full SHA for 52d27a1

Commits on Oct 1, 2024

• [float8] improve eager numerics for dynamic scales and gets on par with torch.compile (pytorch#904)

  Show description for 71be315
  

  14 peopleauthored
  71be315

  Copy full SHA for 71be315

Commits on Sep 26, 2024

• [float8] all-reduce amax on dp mesh instead of global pg (pytorch#933)

  Show description for da0bbe3
  weifengpy

  authored
  da0bbe3

  Copy full SHA for da0bbe3

Commits on Sep 25, 2024

• Fix fp8-all-gather buck errors

  Show description for b521c9b
  y-sq

  authored
  b521c9b

  Copy full SHA for b521c9b

Commits on Sep 24, 2024

• clean up device checks in float8 unit test files (pytorch#923)

  Show description for 26e790d
  vkuzo

  authored
  26e790d

  Copy full SHA for 26e790d

Commits on Sep 23, 2024

• [float8] fix typo in bitwise_identical unit test (pytorch#918)

  Show description for 1d6f8e2
  weifengpy

  authored
  1d6f8e2

  Copy full SHA for 1d6f8e2

Commits on Sep 4, 2024

• [Float8] Make Inference and Training code independent (pytorch#808)

  drisspg

  authored
  848e123

  Copy full SHA for 848e123

Commits on Aug 28, 2024

• float8 training: add static scaling (pytorch#760)

  Show description for 983f565
  vkuzo

  authored
  983f565

  Copy full SHA for 983f565

Commits on Aug 22, 2024

• cast local_scale_tensor to fp32 for precompute of fp8 dynamic scaling (pytorch#713)

  Show description for cdb25a4
  

  crcrpar

  and

  msaroufim

  authored
  cdb25a4

  Copy full SHA for cdb25a4

Commits on Aug 15, 2024

• Add PyTorch 2.4 tests in CI (pytorch#654)

  msaroufim

  authored
  ffa88a4

  Copy full SHA for ffa88a4

Commits on Aug 14, 2024

• retry version guard fix (pytorch#679)

  Show description for 1acd710
  msaroufim

  authored
  1acd710

  Copy full SHA for 1acd710

Commits on Aug 12, 2024

• fix bug with float8 + inference_mode (pytorch#659)

  Show description for d3b8d43
  vkuzo

  authored
  d3b8d43

  Copy full SHA for d3b8d43

Commits on Aug 8, 2024

• Skip failing inference test which requires change to PT core (pytorch#640)

  drisspg

  authored
  5f35645

  Copy full SHA for 5f35645

Commits on Aug 7, 2024

• Enable float8 CI on sm89 (pytorch#587)

  jainapurva

  authored
  1cfe69e

  Copy full SHA for 1cfe69e

Commits on Aug 6, 2024

• Fix inference tests and add export test (pytorch#613)

  drisspg

  authored
  ac03072

  Copy full SHA for ac03072

• testing ghstack direct, part 2 (pytorch#608)

  Show description for 2d33197
  vkuzo

  authored
  2d33197

  Copy full SHA for 2d33197

Pagination

Previous

Next