Code

Add HQQ support#605

Code

Merged

msaroufim merged 21 commits intopytorch:mainpytorch/ao:mainfrom mobiusml:mainCopy head branch name to clipboardAug 15, 2024

+428-8

Commits

Commits on Aug 6, 2024

• Add HQQ support

  mobicham

  committed
  f99a90c

  Copy full SHA for f99a90c

• use use_hqq flag in AffineQuantizedTensor.from_float + move hqq core to quantization api

  mobicham

  committed
  ab9ea3d

  Copy full SHA for ab9ea3d

Commits on Aug 7, 2024

• Merge branch 'pytorch:main' into main

  mobicham

  authored
  153fed1

  Copy full SHA for 153fed1

• move hqq quantization to quant_primitives

  mobicham

  committed
  e022654

  Copy full SHA for e022654

• update example with mutliple nbits

  mobicham

  committed
  f7a9e50

  Copy full SHA for f7a9e50

• clean-up imports in affine_quantized_tensor

  mobicham

  committed
  082dc58

  Copy full SHA for 082dc58

Commits on Aug 9, 2024

• add hqq to quant_api apply_int4_weight_only_quant

  mobicham

  committed
  77d498a

  Copy full SHA for 77d498a

• add random seed

  mobicham

  committed
  9a83eda

  Copy full SHA for 9a83eda

• add unittest

  mobicham

  committed
  c65e796

  Copy full SHA for c65e796

• Merge branch 'pytorch:main' into main

  mobicham

  authored
  b5abf39

  Copy full SHA for b5abf39

Commits on Aug 13, 2024

• Merge branch 'pytorch:main' into main

  mobicham

  authored
  c5834fe

  Copy full SHA for c5834fe

• replace from_float() with to_affine_quantized

  mobicham

  committed
  3303d95

  Copy full SHA for 3303d95

• add _ to private functions

  mobicham

  committed
  41b2fb0

  Copy full SHA for 41b2fb0

• add quantize_affine_hqq to __all__

  mobicham

  committed
  9382ec1

  Copy full SHA for 9382ec1

Commits on Aug 14, 2024

• Merge branch 'pytorch:main' into main

  mobicham

  authored
  d15accf

  Copy full SHA for d15accf

• separate xnbit tests + check device

  mobicham

  committed
  93ca471

  Copy full SHA for 93ca471

Commits on Aug 15, 2024

• Merge branch 'main' into main

  mobicham

  authored
  2146a5a

  Copy full SHA for 2146a5a

• add torch version for tensorcore dtype

  mobicham

  committed
  1e5eec8

  Copy full SHA for 1e5eec8

• fix torch 2.4 tensorcore dtype

  mobicham

  committed
  890a7be

  Copy full SHA for 890a7be

• fix core.py import

  mobicham

  committed
  9700d52

  Copy full SHA for 9700d52

• skip assertion error in test_dynamic_quant_per_channel_numerics_cuda

  mobicham

  committed
  0b511f5

  Copy full SHA for 0b511f5