huggingface/transformers

Pull Requests Commits

woops, wromg PR

molbap committed 57 days ago

1b1d5ad2

rf_detr changes

molbap committed 57 days ago

00bd1ba4

re-re-revert logic on backbone output tracing 🫪🫪🫪

molbap committed 57 days ago

0edbe929

remove manual setting

molbap committed 57 days ago

4273d633

molbap committed 57 days ago

6858e7b2

remove redundant dec

molbap committed 57 days ago

d76a2716

molbap committed 57 days ago

56d4f34d

fall back to usual hidden states routing for now

molbap committed 57 days ago

53ffa669

molbap committed 57 days ago

1a3f651b

registers conversion

molbap committed 57 days ago

a83a5521

molbap committed 57 days ago

f2cfd291

simplifications

molbap committed 57 days ago

f999712a

update decorators again, keep Encoders

molbap committed 57 days ago

3a20f3ba

molbap committed 57 days ago

a6790707

molbap committed 57 days ago

effd700e

propagate changes, fix tests

molbap committed 57 days ago

175ebd2f

First draft + todo

molbap committed 57 days ago

fab653f6

fix(afmoe): reduce tokens in test_compile_static_cache to avoid flaky bfloat16 drift (#46521)

ydshieh committed 57 days ago

Verified d2094607

[CB] Add a "max_requests_per_batch" parameter (#46434)

remi-or committed 58 days ago

Verified 64f94504

Extended & simplified n-to-1 kernel fusion via KernelConfig (#46339)

michaelbenayoun committed 58 days ago

Verified 5047f086

Speed up modular conversion by reusing process pool (#46438)

guarin committed 58 days ago

Verified cc041637

revamp cv docs and fix rf-detr (#46219)

merveenoyan committed 58 days ago

Verified 1b7bc25c

Update hub metadata (#46379)

zucchini-nlp committed 58 days ago

Verified 1d185d43

extend DeepseekV4FlashIntegrationTest to non-cuda device (#46517)

sywangyi committed 58 days ago

Verified e3144397

[docs] deepgemm (#46361)

stevhliu committed 58 days ago

Verified 55115549

Fix `{image/video/audio}_token_ids` in `ProcessorMixin` (#46500)

hmellor committed 58 days ago

Verified d6a82ba8

[fix] regression introduced by #45534 (#46456)

eustlb committed 58 days ago

Verified b6bad752

[docs] tp for continuous batching (#46019)

stevhliu committed 58 days ago

Verified 153d6c16

Use torchvision's native LANCZOS interpolation instead of PIL fallback (#46496)

NicolasHug committed 58 days ago

Verified ff331709

skip quant cache tests for deepseek v4 (#46362)

kaixuanliu committed 58 days ago

Verified e569f888

Older