Code

Flexible Pack DType#1158

Code

Merged

Qubitium merged 15 commits intomainModelCloud/GPTQModel:mainfrom flex-storage-dtypeModelCloud/GPTQModel:flex-storage-dtypeCopy head branch name to clipboardJan 26, 2025

+258-161

Commits

Commits on Jan 24, 2025

• init flex dtype

  Qubitium

  committed
  dc28d12

  Copy full SHA for dc28d12

• bits

  Qubitium

  committed
  6eca433

  Copy full SHA for 6eca433

• fix tensors_per_storage_dtype

  LRL-ModelCloud

  committed
  406653e

  Copy full SHA for 406653e

• add storage_np_dtype

  LRL-ModelCloud

  committed
  c29dcbb

  Copy full SHA for c29dcbb

• wrong np dtype

  Qubitium

  committed
  c1b5acb

  Copy full SHA for c1b5acb

• dtype err

  Qubitium

  committed
  ebfe746

  Copy full SHA for ebfe746

• cleanup

  Qubitium

  committed
  44d239c

  Copy full SHA for 44d239c

• add int64

  Qubitium

  committed
  9ca4124

  Copy full SHA for 9ca4124

Commits on Jan 25, 2025

• simplify

  Qubitium

  committed
  9a39ff8

  Copy full SHA for 9a39ff8

• simplify

  Qubitium

  committed
  fdda28e

  Copy full SHA for fdda28e

Commits on Jan 26, 2025

• move control to cfg.pack_dtype

  Qubitium

  committed
  53dd6a2

  Copy full SHA for 53dd6a2

• disable int64 (ppl failed)

  Qubitium

  committed
  b30f7c6

  Copy full SHA for b30f7c6

• add flex pack dtype to triton

  Qubitium

  committed
  9ea27f9

  Copy full SHA for 9ea27f9

• wrong qzero buffer dtype

  Qubitium

  committed
  e820660

  Copy full SHA for e820660

• marlin add prelim int32 refractor

  Qubitium

  committed
  1a6e9aa

  Copy full SHA for 1a6e9aa