huggingface/transformers

Pull Requests Commits

refactor merging plans

3outeille committed 8 hours ago

036cd6df

[distributed] Document SP+EP and TP+EP flag combinations

3outeille committed 12 hours ago

8bbad67c

[distributed] Drop intra-expert moe_tp_* from MoE SP plans

3outeille committed 12 hours ago

4bb9d29a

[distributed] Use merged plan in tp_plan property and load path

3outeille committed 12 hours ago

0b346b82

[distributed] Wire resolve_parallel_plan into apply_tensor_parallel

3outeille committed 12 hours ago

24ab0dcc

[distributed] Add resolve_parallel_plan merge helper

3outeille committed 12 hours ago

5f4729be

[distributed] Rename MoE intra-expert TP styles to moe_tp_*

3outeille committed 12 hours ago

ec2212bd

[distributed] Document decomposed MoE TP/EP plans

3outeille committed 14 hours ago

6d73f841

[distributed] Migrate MoE configs to decomposed TP/SP expert plans

3outeille committed 14 hours ago

6a6dda75

[distributed] Add MoE TP/EP plan tests and a two-sided sharding assertion

3outeille committed 14 hours ago

963777a0

[distributed] Add param-level apply pass to apply_tensor_parallel

3outeille committed 15 hours ago

bcecab89

[distributed] Add param-level MoE TP/EP styles and ep_router

3outeille committed 15 hours ago

f20c14c1

handle dense + sparse mixing for mellum model in SP plan

3outeille committed 1 day ago

2d01eb3f

mellum now inherits tp and sp plan from qwen3_moe

3outeille committed 1 day ago

f914e301

not unifying but still better than before

3outeille committed 1 day ago

863e12ba

Revert "[`Revert`] FSDP+Dtensor refactor related changes (#46246)"

3outeille committed 1 day ago

18a712b9

Fix model parallel issue for altclip model and ChineseClip model (#45487)

kaixuanliu committed 1 day ago

Verified 08f13097

feat: Add support for JetBrains' `Mellum` v2 code generation model (#46112)

shadeMe committed 1 day ago

Verified 9cfa266f

Fix `StaticCache` building an empty layer list when `num_kv_shared_layers == 0` (#46235)

tengomucho committed 1 day ago

Verified 0ff380e0

Model parallel fix (#46230)

kaixuanliu committed 1 day ago

Verified e120f7ea

Fix TAPAS tokenizer crash on pandas 3.x string-dtype tables (#45916)

Rocketknight1 committed 1 day ago

Verified 1656d90b

Enable kernels-community/metal-flash-sdpa on MPS (#45974)

ArthurZucker committed 1 day ago

Verified bdca59aa

Fix inverted assertion in remove_handler (#46227)

SebTardif committed 1 day ago

Verified a8a01e53

[ShieldGemma2] Support attn_implementation dispatch (#46069)

YangKai0616 committed 1 day ago

Verified ec06284c

[Gemma4] Replace one-hot matmul with F.embedding in position embeddings (#46176)

Sriniketh24 committed 1 day ago

Verified bc8f70a9

fix: kosmos2.5: properly expand embeddings table (#45835)

nunq committed 1 day ago

Verified fd1d5b7d

find pytest launch error in torch 2.13.0.dev20260526 (#46252)

sywangyi committed 2 days ago

Verified f336a53d

Update a test to avoid writing to the default xet cache (#46250)

ydshieh committed 2 days ago

Verified 2d31b762

[Test][Kosmos2.5] Add XPU expectations for integration tests (#46135)

YangKai0616 committed 2 days ago

Verified 88070eff

[`Revert`] FSDP+Dtensor refactor related changes (#46246)

vasqu committed 2 days ago

Verified 295cee3e

Older