huggingface/transformers

Pull Requests Commits

ArthurZucker committed 76 days ago

21ae32d4

tp loading: minimal sync distribute_tensor, drop async machinery

ArthurZucker committed 79 days ago

c60156f8

tp loading: async scatter + broadcast via mesh_scatter(async_op=True)

ArthurZucker committed 79 days ago

cdb4a7d9

tp loading: switch to distribute_tensor, drop 200 lines of packing

ArthurZucker committed 79 days ago

be9278eb

tp loading: remove dead code, shared thread pool, allocator warmup

ArthurZucker committed 79 days ago

e186123d

tp loading: pipelined batch_isend_irecv with overlap

ArthurZucker committed 79 days ago

63d7486e

tp loading: single-batch packed scatter, document NCCL transfer floor

ArthurZucker committed 79 days ago

ffe7a339

tp loading: clean NCCL packed-scatter, remove CUDA IPC experiment

ArthurZucker committed 79 days ago

498a7269

tp loading: per-batch pinned-memory reads + pipeline overlap

ArthurZucker committed 80 days ago

5d9c61e3

ArthurZucker committed 80 days ago

8dc4d753

ArthurZucker committed 80 days ago

4e66348e

tp loading: packed scatter per source rank; drop coalescing_manager

ArthurZucker committed 80 days ago

b9d49ae9

tp loading: batch coalesce scatter/broadcast via _coalescing_manager

ArthurZucker committed 80 days ago

84ee0694

tp loading: 3-stage pipeline — async convert + async scatter + finalize

ArthurZucker committed 80 days ago

5301a9af

tp loading: materialize to GPU directly, drop redundant CPU→GPU copy

ArthurZucker committed 80 days ago

6a4dea69

tp loading: async pipelined scatter + eliminate per-mapping metadata broadcast

ArthurZucker committed 80 days ago

ffac18ae

tp loading: batched pre-slice via torch.chunk instead of per-rank loop

ArthurZucker committed 80 days ago

fa073e84

tp loading: use dist.scatter for sharded redistribute

ArthurZucker committed 81 days ago

aea39f09

tp loading: fix view-keeps-buffer-alive, skip warmup under TP

ArthurZucker committed 81 days ago

1fa1c78b

tp loading: CPU staging, per-rank scheduling, slice-first FP8, no_grad

ArthurZucker committed 81 days ago

b5dcdf21

ArthurZucker committed 81 days ago

1ee7fff9

ArthurZucker committed 81 days ago

5c4efd53

Add `step3_vl` to `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS` (#45449)

hmellor committed 81 days ago

Verified 331ea339

Update workflow references to new commit hash (#45442)

paulinebm committed 82 days ago

Verified 7028c30b

[Gemma4] Add docstrings for Per-Layer Embeddings (PLE) pipeline (#45207)

w4nderlust committed 82 days ago

Verified 155db714

[Doc] Correct checkpoint path in Dinov2 model_docs (#45430)

ambroiseodt committed 82 days ago

Verified 99ce466f

Fix ty for transformers cli (#45190)

SunMarc committed 82 days ago

Verified 5e3af512

fix(models): Resolve regressions in Wav2Vec2PhonemeCTCTokenizer (wav2vec2-lv-60-espeak-cv-ft) (#45199)

harshaljanjani committed 82 days ago

Verified 5b565a58

fix: prevent accelerate from splitting vision encoder by setting _no_… (#43047)

CodersAcademy006 committed 82 days ago

Verified 27fbb514

Fix Qwen2.5VL temporal grid positions (#45400)

zucchini-nlp committed 82 days ago

Verified fd63aa22

Older