Commits

Branch selector

 main

User selector

caogao

Datepicker

All time

Commit History

Commits on Feb 23, 2022

• Allow skipping dequantization after P2P. (pytorch#947)

  Show description for d791702
  

  caogao

  authored and

  facebook-github-bot

  committed
  d791702

Commits on Feb 15, 2022

• Add TBE support to benchmark. (pytorch#919)

  Show description for 5ae23f9
  

  caogao

  authored and

  facebook-github-bot

  committed
  5ae23f9

Commits on Feb 10, 2022

• Add INT8 and INT4 support to P2P benchmark. (pytorch#918)

  Show description for 501dfa7
  

  caogao

  authored and

  facebook-github-bot

  committed
  501dfa7

Commits on Jun 17, 2021

• Deprecate old quanzation / dequantization util functions. (pytorch#630)

  Show description for 1bad64b
  

  caogao

  authored and

  facebook-github-bot

  committed
  1bad64b

Commits on Jun 16, 2021

• Update call site for FBGemm quantization util functions. (pytorch#624)

  Show description for 8865079
  

  caogao

  authored and

  facebook-github-bot

  committed
  8865079

Commits on Jun 10, 2021

• Add quantize and dequantize functions between rowwise-quantized int8 + fp32 scale/bias and fp32 embeddings. (pytorch#623)

  Show description for 3182125
  

  caogao

  authored and

  facebook-github-bot

  committed
  3182125

Commits on Jun 5, 2021

• Add functions for dequantizing rowwise-quantized int2/int4/int8 + fp16 scale/bias to fp16 embeddings. (pytorch#619)

  Show description for 77a4792
  

  caogao

  authored and

  facebook-github-bot

  committed
  77a4792

Commits on Jun 3, 2021

• Remove usage of if constexpr from D28620537. (pytorch#618)

  Show description for 5885ee1
  

  caogao

  authored and

  facebook-github-bot

  committed
  5885ee1

Commits on Jun 2, 2021

• Add functions for quantizing fp16 embeddings to rowwise quantized int2/int4/int8 + fp16 scale/bias. (pytorch#616)

  Show description for 9cb33bc
  

  caogao

  authored and

  facebook-github-bot

  committed
  9cb33bc