Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Jan 12, 2023

• Enable bias decoding tests with beam_size=1 (#1046)

  guillaumekln

  authored
  92de578

  Copy full SHA for 92de578

• feat: support bias decoding for beam_size 1 (#1044)

  linxiao-zeng-zm

  authored
  e1ff860

  Copy full SHA for e1ff860

Commits on Jan 9, 2023

• Simplify how examples are read from a text stream (#1043)

  guillaumekln

  authored
  fafa687

  Copy full SHA for fafa687

Commits on Jan 5, 2023

• doc: Manually list the frameworks to avoid updating the navigation

  guillaumekln

  committed
  349942b

  Copy full SHA for 349942b

Commits on Jan 2, 2023

• Bump version to 3.3.0

  guillaumekln

  committed
  0d81ef2

  Copy full SHA for 0d81ef2

• Apply SmoothQuant for OPT models (#1038)

  guillaumekln

  authored
  bfdc886

  Copy full SHA for bfdc886

Commits on Dec 30, 2022

• Update oneMKL to 2023.0 (#1037)

  Show description for 91c85ad
  guillaumekln

  authored
  91c85ad

  Copy full SHA for 91c85ad

• Remove unnecessary CXX flag when compiling on macOS (#1036)

  guillaumekln

  authored
  90275f4

  Copy full SHA for 90275f4

• Update oneDNN to 3.0 (#1035)

  Show description for b40beec
  guillaumekln

  authored
  b40beec

  Copy full SHA for b40beec

Commits on Dec 29, 2022

• Remove attribute check in test

  guillaumekln

  committed
  7027b20

  Copy full SHA for 7027b20

• Reuse existing synchronization function in profiler

  guillaumekln

  committed
  c894953

  Copy full SHA for c894953

• Add Tanh operator (#1033)

  guillaumekln

  authored
  c8bb82b

  Copy full SHA for c8bb82b

• Reduce disk usage when running Python tests in the CI (#1034)

  guillaumekln

  authored
  0dddb43

  Copy full SHA for 0dddb43

• Support loading model files from memory (#1030)

  Show description for a6b9121
  guillaumekln

  authored
  a6b9121

  Copy full SHA for a6b9121

• Support T5 models (#1027)

  guillaumekln

  authored
  ca4fe74

  Copy full SHA for ca4fe74

Commits on Dec 28, 2022

• Read decoder_layers parameters for BART-like models (#1032)

  guillaumekln

  authored
  a7c0366

  Copy full SHA for a7c0366

Commits on Dec 21, 2022

• Do not register spec attributes that should not be set (#1029)

  guillaumekln

  authored
  d6de581

  Copy full SHA for d6de581

• Add CUDA 11.8 in PyPI classifiers

  guillaumekln

  committed
  b646145

  Copy full SHA for b646145

Commits on Dec 20, 2022

• Pin oneDNN version in the CI (#1028)

  guillaumekln

  authored
  d559922

  Copy full SHA for d559922

• Remove CMake warning when oneDNN is missing

  Show description for 095a581
  guillaumekln

  committed
  095a581

  Copy full SHA for 095a581

• Add RMSNorm op (#1026)

  Show description for e020385
  guillaumekln

  authored
  e020385

  Copy full SHA for e020385

Commits on Dec 19, 2022

• Fix compilation when no BLAS CPU backend is selected (#1025)

  guillaumekln

  authored
  2b05ed9

  Copy full SHA for 2b05ed9

Commits on Dec 14, 2022

• Remove out of date comment since CUDA 10 is no longer supported

  guillaumekln

  authored
  c797a92

  Copy full SHA for c797a92

Commits on Dec 12, 2022

• Unpin PyTorch version for macOS tests (#1022)

  guillaumekln

  authored
  84388ea

  Copy full SHA for 84388ea

• Bump version to 3.2.0

  guillaumekln

  committed
  9e3c389

  Copy full SHA for 9e3c389

• Update OpenNMT-tf to version 2.30 in tests (#1021)

  guillaumekln

  authored
  9afef4b

  Copy full SHA for 9afef4b

• Enable ffast-math locally (#1020)

  guillaumekln

  authored
  06762aa

  Copy full SHA for 06762aa

Commits on Dec 10, 2022

• Use the correct quantization value in the size table

  guillaumekln

  authored
  3fd7aba

  Copy full SHA for 3fd7aba

Commits on Dec 9, 2022

• Support the new generator structure in OpenNMT-py (#1019)

  guillaumekln

  authored
  b4f2448

  Copy full SHA for b4f2448

• Fix CMakeLists #1015 Windows OpenMP runtime (#1016)

  Show description for 410c908
  

  A2va

  and

  guillaumekln

  authored
  410c908

  Copy full SHA for 410c908

• Add option end_token to stop the decoding on a specific token (#1014)

  guillaumekln

  authored
  f9ca83b

  Copy full SHA for f9ca83b

• Improve batch generation with the Whisper model (#1012)

  Show description for 338aac9
  guillaumekln

  authored
  338aac9

  Copy full SHA for 338aac9

Commits on Dec 8, 2022

• Update Python test to not assume it is run with an Intel CPU (#1013)

  guillaumekln

  authored
  89c69c4

  Copy full SHA for 89c69c4

• Fallback to a basic Conv1D implementation when oneDNN is missing (#1010)

  Show description for 2dcdf57
  guillaumekln

  authored
  2dcdf57

  Copy full SHA for 2dcdf57

• Generalize Gather op to support all cases where axis == batch_dims (#1011)

  Show description for a8a9078
  guillaumekln

  authored
  a8a9078

  Copy full SHA for a8a9078