Code

Gptq mt refactor#914

Code

Merged

HDCharles merged 24 commits intopytorch:mainpytorch/ao:mainfrom danielpatrickhug:gptq_MT_refactordanielpatrickhug/ao:gptq_MT_refactorCopy head branch name to clipboardOct 1, 2024

+1,243-11

Commits

Commits on Oct 1, 2024

• add inital [WIP] of MultiTensor rewrite of GPTQ

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  c8d0eca

  Copy full SHA for c8d0eca

• Implement GPTQQuantizer and Int4WeightOnlyGPTQQuantizer

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  4c2a8ed

  Copy full SHA for 4c2a8ed

• added GPTQuantizer and Int4WeightOnlyGPTQQuantizer classes. Removed QuantizableModel class

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  66d913c

  Copy full SHA for 66d913c

• removed print statement

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  4b6e1ce

  Copy full SHA for 4b6e1ce

• fix control structure in torch_function. layer outputs not properly handled

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  c08385c

  Copy full SHA for c08385c

• add testing script for gptq Multitensor

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  3e6a456

  Copy full SHA for 3e6a456

• testing

  Show description for cbdaec5
  HDCharles

  committed
  cbdaec5

  Copy full SHA for cbdaec5

• more fixes

  Show description for 67bc6ac
  HDCharles

  committed
  67bc6ac

  Copy full SHA for 67bc6ac

• more fixes

  Show description for 8615084
  HDCharles

  committed
  8615084

  Copy full SHA for 8615084

• this is working now

  Show description for 5644d02
  HDCharles

  committed
  5644d02

  Copy full SHA for 5644d02

• fixes for in place ops

  Show description for 22c711b
  HDCharles

  committed
  22c711b

  Copy full SHA for 22c711b

• removing some debug code

  Show description for 2e75905
  HDCharles

  committed
  2e75905

  Copy full SHA for 2e75905

• AQT fixes

  Show description for 45f8cb5
  HDCharles

  committed
  45f8cb5

  Copy full SHA for 45f8cb5

• small speed improvement, force unpad unelss we designated op as in place

  Show description for 49526fc
  HDCharles

  committed
  49526fc

  Copy full SHA for 49526fc

• testing

  Show description for b84139d
  HDCharles

  committed
  b84139d

  Copy full SHA for b84139d

• final fixes

  Show description for 6e57d80
  HDCharles

  committed
  6e57d80

  Copy full SHA for 6e57d80

• refactor test_gptq_quantizer_int4_weight_only to use new MT class. add unit tests for MT. Refactor inplace ops classification for readablility.

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  8649eb4

  Copy full SHA for 8649eb4

• change path to model.pth

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  1f2498d

  Copy full SHA for 1f2498d

• add correct imports to tests

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  8819cc5

  Copy full SHA for 8819cc5

• fix imports and duplicate lines of code

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  2d36bb6

  Copy full SHA for 2d36bb6

• fix import of model.pth

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  200473b

  Copy full SHA for 200473b

• add decorator to skip test due to lm_eval not in CI

  

  danielpatrickhug

  authored and

  HDCharles

  committed
  826b830

  Copy full SHA for 826b830

• Merge branch 'pytorch:main' into gptq_MT_refactor

  danielpatrickhug

  authored
  2ed51c5

  Copy full SHA for 2ed51c5

• skip MT tests in test_quant_api

  danielpatrickhug

  committed
  83d8cdd

  Copy full SHA for 83d8cdd