All pull requests

Pull requests

• Use grouped_mm on TPU devices under torch.compile

  #49097·salkan0 opened Sep 24, 2026·

  ·
  1

  1

• [Zamba] Fix associative scan breaking ONNX export and OOM in integration test

  #49092·ydshieh opened Sep 24, 2026Collaborator·

  ·
  11

• [MPS] Remove cu_seqlens_k clone workaround for metal-flash-sdpa

  #49091·Isalia20 opened Sep 24, 2026Contributor·

  ·
  2

• [WIP] move processing defaults to class

  #49089·guarin opened Sep 24, 2026Member·

  ·
  6

• 🚨 [CB] 🚨 Fuse update for index and block table path

  #49088·remi-or opened Sep 24, 2026Collaborator·3/3·

  ·
  6

• [encodec] separate encoder and decoder

  #49087·eustlb opened Sep 24, 2026Collaborator·

  ·
  3

• [WIP] Modality tester auto-parameterize based on model class

  #49085·zucchini-nlp opened Sep 24, 2026Member·

  ·

• Remap the legacy Gemma 1 hidden_act in the config post-init

  #49084·PCfVW opened Sep 24, 2026·

  ·
  1

  22

• Keep the eos ids the config declares

  #49082·albertvillanova opened Sep 24, 2026Member·

  ·
  7

• Fix Gemma 1.0 legacy GELU activation mapping in config

  #49081·Talhax55z opened Sep 24, 2026·

  ·
  1

  3

• Composite lm head tp plan

  #49080·IlyasMoutawwakil opened Sep 24, 2026Member·

  ·
  10

• Composite transparent PIL images in shared RGB conversion

  #49079·Nikhi00718 opened Sep 24, 2026·

  ·
  4

• Scope GITHUB_TOKEN permissions per job

  #49078·hf-security-analysis[bot] opened Sep 24, 2026Bot·

  ·
  3

• Fix Qwen2Tokenizer ignoring pretokenize_regex (Qwen3.6 / Qwen3.8 wrong token IDs)

  #49077·tanvir-ux opened Sep 24, 2026·

  ·
  1

  2

• [serge] Fix 2 integration tests for model cvt failing with output_mismatch (tensor values differ (2))

  #49076·sergereview[bot] opened Sep 24, 2026Bot·

  ·
  3

• [Refactor] Make some flash-attention utils more readable

  #49071·remi-or opened Sep 24, 2026Collaborator·2/3·

  ·
  13

  

• 🚨 [CB] 🚨 Little fixes before removing "paged"

  #49069·remi-or opened Sep 24, 2026Collaborator·1/3·

  ·
  14

• [serge] Fix 2 integration tests for model hy_v3 failing with output_mismatch (tensor values differ (2))

  #49068·sergereview[bot] opened Sep 23, 2026Bot·

  ·
  3

• [serge] Fix 2 integration tests for model pvt_v2 failing with other (other (2))

  #49067·sergereview[bot] opened Sep 23, 2026Bot·

  ·
  3

• dont merge - i'm checking smth

  #49064·zucchini-nlp opened Sep 23, 2026Member·

  ·
  3

• [docs] TP for PEFT adapters

  #49060·stevhliu opened Sep 23, 2026Member·

  ·
  2

• Document running the example scripts on Hugging Face Jobs

  #49050·davanstrien opened Sep 23, 2026Member·

  ·
  2

• fix: treat empty protobuf bytes as absent in SentencePiece charsmap

  #49048·codewithyug06 opened Sep 23, 2026·

  ·
  1

  2

• [serge] Fix 2 integration tests for model jamba failing with other (other (2))

  #49044·sergereview[bot] opened Sep 23, 2026Bot·

  ·
  3

• fix(cache): roll back recurrent states in LinearAttentionCacheLayerMi…

  #49043·Varun-51 opened Sep 23, 2026·

  ·
  1

  1