huggingface/transformers

Pull Requests Commits

Cast the unquantized weights of a dequantized FP8 checkpoint to the model's dtype

qgallouedec committed 1 day ago

4ee62cbd

Fix RecurrentGemma padded generation by declaring its `layer_types` (#50577)

ydshieh committed 1 day ago

Verified 07cc8e8c

fixed typo in docs - changed "a encoder-decoder" to "an encoder-decoder" (#50591)

Peter-Bergman committed 1 day ago

Verified e5ce72af

Revert "Add `transcription` input arg to `apply_transcription_request ` (#49869)" (#50587)

ebezzam committed 1 day ago

Verified f41b35a2

Bump doc-builder (#50550)

stevhliu committed 1 day ago

Verified 2ac1093a

[docs] Output capturing (#50529)

stevhliu committed 1 day ago

Verified 7ebb1328

Tests: skip audio model tests when torchaudio or librosa is missing (#50423)

IlyasMoutawwakil committed 1 day ago

Verified 2a540d5c

CI: install OpenVINO and NNCF in the daily GPU image (#50422)

IlyasMoutawwakil committed 1 day ago

Verified 9ae35c18

Clarify that causal LM labels are shifted internally (#44642)

joshuaswanson committed 1 day ago

Verified 6488cabb

Avoid device -> host syncs in load_balancing_loss_func (#50463)

ntenenz committed 1 day ago

Verified fee0f2b9

Fix grouped RMSNorm in Mamba2's non-fused path (#49744)

taking-lying-flat committed 1 day ago

Verified c4f1520d

Fix flaky ONNX export test for wav2vec2-style `ForPreTraining` models (argmax near-tie) (#50575)

ydshieh committed 1 day ago

Verified 04935b4f

[Model] LFM3 Family: LFM3-VL (#50392)

paulpak58 committed 1 day ago

Verified f1787be5

Draft updates for contribution/agent rules (#49783)

Rocketknight1 committed 1 day ago

Verified a8d43b20

Speed up `bad_words_ids` and `sequence_bias` by matching multi-token sequences in one pass (#50532)

deadstrobe5 committed 1 day ago

Verified ed0ecab6

Gemma4: fix static cache attending to empty cache slots in KV-shared layers at prefill (#50528)

ydshieh committed 1 day ago

Verified ca774efd

Fix Falcon-H1 gated RMSNorm shape during batched decoding (#50507)

taking-lying-flat committed 1 day ago

Verified fab2fd4e

QA / Add a mic runner in Make (#50407)

tarekziade committed 1 day ago

Verified 863d2620

[serge] Fix 8 integration tests for model `grounding_dino` failing with `output_mismatch` (tensor values differ (6), oth (#49886)

sergereview[bot] committed 1 day ago

Verified 9e7f5cd1

Fix flaky `test_causal_lm_loss_honors_shift_labels` (#50533)

ydshieh committed 1 day ago

Verified 78a4dafe

[Fix] Prevent duplicate weight renaming on save_pretrained for ALMs (#50265)

mrutkows committed 1 day ago

Verified a4f36274

Rename FP8 `modules_to_not_convert` entries like weight names (#50354)

albertvillanova committed 1 day ago

Verified c49879e7

[docs] Agent-specific content (#50276)

stevhliu committed 1 day ago

Verified fc6cdba1

Fix MiniMax-M3-VL packed vision attention isolation (#50346)

taking-lying-flat committed 1 day ago

Verified e20e0787

Tester always passes `per_layer_config` (#50486)

zucchini-nlp committed 1 day ago

Verified 800db143

Fix Qwen VL positions when a text cache is reused with a new image (#50427)

Arnav66692 committed 1 day ago

Verified 9f8c7c94

Drop None-valued keys from messages in `apply_chat_template` (#45422)

qgallouedec committed 2 days ago

Verified 86fdf159

[Models] Fix torch.compile fullgraph training across models when layerdrop=0.0 (#50475)

aarushdubey committed 2 days ago

Verified 147e66b3

[masking] Fix packed sequences for abitrary mask forms (#50513)

Cyrilvallez committed 2 days ago

Verified fa5127c5

🚨 Move processing defaults to class (#49089)

guarin committed 2 days ago

Verified 3c3ff8c1

Older