huggingface/transformers

Pull Requests Commits

Add nCompass triton-moe experts integration + sonic-moe dep nudges

IlyasMoutawwakil committed 127 days ago

6c84a0f2

Fix accelerate hooks for all models using linear attention (#46978)

Cyrilvallez committed 134 days ago

Verified 36778030

Add support for RADIO models (#46425)

meatybobby committed 134 days ago

Verified 17518d53

only in the original repo (#46982)

tarekziade committed 135 days ago

Verified 18693fc6

🚨 Modeling changes for export, compile, and hybrid-attention standardization (#46738)

IlyasMoutawwakil committed 135 days ago

Verified 181beb3b

Fix typos in Gemma 4 Assistant documentation (#46975)

RaunaqDavidNath committed 135 days ago

Verified 45b004d7

🚨 [gemma 3/4] Fix bidirectional attention masking crossing sliding window boundaries (#46850)

douglas-reid committed 135 days ago

Verified 1f3294c8

the CI status should be a comment (#46976)

tarekziade committed 135 days ago

Verified 9f7719f3

QwenVL model conversion (#46881)

zucchini-nlp committed 135 days ago

Verified 03197493

Remove default dtype in FusedRMSNormGated modules (#46953)

Cyrilvallez committed 135 days ago

Verified 6e1b2546

[docs] EP plan validation (#46900)

stevhliu committed 136 days ago

Verified 957e6032

Fix Xcodec2 attention to be non-causal. (#46963)

ebezzam committed 136 days ago

Verified 7f5d5d1a

Fix Mamba2 chunked-prefill / speculative decoding for Zamba2, Nemotron-H, Bamba, FalconH1 and GraniteMoeHybrid (#46741)

Sunt-ing committed 136 days ago

Verified fdb6d318

FIX PEFT test changed error type (#46959)

BenjaminBossan committed 136 days ago

Verified b779eb6c

Fix path traversal via vocab-file arguments in tokenizer_config.json (#46279)

LinZiyuu committed 136 days ago

Verified 61fa365b

docs(conditional_detr): fix num_queries default in docstring (100 -> 300) (#46939)

Kropiunig committed 136 days ago

Verified 2e68d89e

Use common floats_list method for feature extractor tests. (#46956)

ebezzam committed 136 days ago

Verified fab209b2

Fix RT-DETR indexing error when num_feature_levels exceeds backbone o… (#46833)

c1prk committed 136 days ago

Verified d9c4bf39

Fix Florence2 training-loss double-shift (same pattern as Moonshine #… (#46898)

sharmax-vikas committed 136 days ago

Verified 2370ba6e

Remove some unnecessary generate warnings (#46955)

Cyrilvallez committed 136 days ago

Verified 8f8edc24

[Olmo3] different RoPE per layer type (#46911)

zucchini-nlp committed 136 days ago

Verified 326683dc

Use inspect.getsource instead of open() for source-reading in _can_set_*_implementation (#46207)

rasmi committed 136 days ago

Verified d9e7791b

Reject assisted generation for LFM2 and LFM2-MoE (set _is_stateful) (#46937)

Sunt-ing committed 136 days ago

Verified 2934a0fe

Don't pin the gated delta net norm to `cuda:0` with a hardcoded device (#46817)

Sunt-ing committed 136 days ago

Verified 38d8dd32

Fix auto-mappings registration for remote code & fixes a few custom code issues (#46876)

Cyrilvallez committed 136 days ago

Verified e5dbd129

Fix broken internal documentation links (#46945)

sezer-muhammed committed 137 days ago

Verified 39f89b91

Insert a Grafana badge in the PR (#46774)

tarekziade committed 137 days ago

Verified ea5c6cf3

Add Nemotron 3.5 ASR Streaming (#46565)

eustlb committed 138 days ago

Verified 4fd7f1a0

[NemotronAsrStreaming] fix pipeline (#46870)

eustlb committed 138 days ago

Verified dd192bad

[NemotronAsrStreaming] processor without modular (#46865)

eustlb committed 138 days ago

Verified cf293b69

Older