Code

Transformers 4.48#2158

Code

Merged

IlyasMoutawwakil merged 26 commits intomainhuggingface/optimum:mainfrom transformers-4.48huggingface/optimum:transformers-4.48Copy head branch name to clipboardJan 29, 2025

+272-115

Commits

Commits on Jan 16, 2025

• test

  IlyasMoutawwakil

  committed
  5190280

  Copy full SHA for 5190280

Commits on Jan 20, 2025

• testing tensor cache x)

  IlyasMoutawwakil

  committed
  6a03d76

  Copy full SHA for 6a03d76

• fix logger

  IlyasMoutawwakil

  committed
  7207215

  Copy full SHA for 7207215

• condition cache class usage

  IlyasMoutawwakil

  committed
  6261094

  Copy full SHA for 6261094

• update opset for beit and data2vec vision and skip flattened/fused pkv (e.g. gpt bigcode)

  IlyasMoutawwakil

  committed
  822066d

  Copy full SHA for 822066d

• style

  IlyasMoutawwakil

  committed
  3ab38fd

  Copy full SHA for 3ab38fd

• fix args patcher

  IlyasMoutawwakil

  committed
  d713e5a

  Copy full SHA for d713e5a

• fix modernbert testing

  IlyasMoutawwakil

  committed
  bf4d1f3

  Copy full SHA for bf4d1f3

• adaot to new whisper returned generation length

  IlyasMoutawwakil

  committed
  230c3a0

  Copy full SHA for 230c3a0

• fix is_causal in transformers

  IlyasMoutawwakil

  committed
  3d5d9c9

  Copy full SHA for 3d5d9c9

• fix modernbert failures

  IlyasMoutawwakil

  committed
  96e2714

  Copy full SHA for 96e2714

• style

  IlyasMoutawwakil

  committed
  78a2dba

  Copy full SHA for 78a2dba

• traceable cache

  IlyasMoutawwakil

  committed
  967c6e2

  Copy full SHA for 967c6e2

Commits on Jan 24, 2025

• use pkv index

  IlyasMoutawwakil

  committed
  1d74388

  Copy full SHA for 1d74388

• add version gard and clean up other model patcher version gards

  IlyasMoutawwakil

  committed
  d452c46

  Copy full SHA for d452c46

• patch sdpa attention in optimum for now

  IlyasMoutawwakil

  committed
  5dcab7f

  Copy full SHA for 5dcab7f

• remove modernbert condition

  IlyasMoutawwakil

  committed
  656941a

  Copy full SHA for 656941a

• style

  IlyasMoutawwakil

  committed
  1bcb38f

  Copy full SHA for 1bcb38f

• fix MistralModelPatcher

  IlyasMoutawwakil

  committed
  23fa20e

  Copy full SHA for 23fa20e

• correctly patch gpt2 in vision encoder decoder

  IlyasMoutawwakil

  committed
  24c8f4b

  Copy full SHA for 24c8f4b

Commits on Jan 26, 2025

• patch sdpa attention forward everywhere

  IlyasMoutawwakil

  committed
  3694ea4

  Copy full SHA for 3694ea4

• fix gpt2 cross attention in seq2seq as well

  IlyasMoutawwakil

  committed
  3d7d586

  Copy full SHA for 3d7d586

Commits on Jan 29, 2025

• moved traceable cache to a file for simplicity of model patcher

  IlyasMoutawwakil

  committed
  10833d8

  Copy full SHA for 10833d8

• Apply suggestions from code review

  IlyasMoutawwakil

  authored
  9491d17

  Copy full SHA for 9491d17

• style

  IlyasMoutawwakil

  committed
  2b73129

  Copy full SHA for 2b73129

• fix

  IlyasMoutawwakil

  committed
  dea98a0

  Copy full SHA for dea98a0