Code

Hqq support#21

Code

Draft

ElizaWszola wants to merge 17 commits intomainneuralmagic/vllm:mainfrom hqq-supportneuralmagic/vllm:hqq-supportCopy head branch name to clipboard

+607-89

Commits

Commits on Oct 7, 2024

• pre-merge

  ElizaWszola

  committed
  6cdd1c7

  Copy full SHA for 6cdd1c7

• Merge branch 'main' into hqq-support

  ElizaWszola

  committed
  467ccd8

  Copy full SHA for 467ccd8

• Merge branch 'main' into hqq-support

  ElizaWszola

  committed
  117eff6

  Copy full SHA for 117eff6

Commits on Oct 14, 2024

• try different shapes

  ElizaWszola

  committed
  ee54bca

  Copy full SHA for ee54bca

Commits on Oct 15, 2024

• works with a hack

  ElizaWszola

  committed
  d3b5c12

  Copy full SHA for d3b5c12

• cleanup

  ElizaWszola

  committed
  45a993b

  Copy full SHA for 45a993b

• more cleanup

  ElizaWszola

  committed
  3324b2e

  Copy full SHA for 3324b2e

Commits on Oct 22, 2024

• it works

  ElizaWszola

  committed
  271ff95

  Copy full SHA for 271ff95

• cleanup

  ElizaWszola

  committed
  de8c5f0

  Copy full SHA for de8c5f0

• further cleanup

  ElizaWszola

  committed
  1521370

  Copy full SHA for 1521370

Commits on Oct 23, 2024

• Merge branch 'main' into hqq-support

  ElizaWszola

  committed
  05f1fce

  Copy full SHA for 05f1fce

Commits on Oct 24, 2024

• Adapt to model format prepared with transformers

  ElizaWszola

  committed
  9972c88

  Copy full SHA for 9972c88

• small cleanup

  ElizaWszola

  committed
  7717b55

  Copy full SHA for 7717b55

• reshape cleanup

  ElizaWszola

  committed
  b9106bd

  Copy full SHA for b9106bd

Commits on Oct 25, 2024

• hqq unit tests

  ElizaWszola

  committed
  5ef4b80

  Copy full SHA for 5ef4b80

• remove hardcoded type

  ElizaWszola

  committed
  5340ce8

  Copy full SHA for 5340ce8

• force fp16 type in kernel to reduce wheel size

  ElizaWszola

  committed
  1da2d97

  Copy full SHA for 1da2d97