PR #3096 Save checkpoint with TP

Save checkpoint with TP #3096

BenjaminBossan merged 24 commits into huggingface:main from michaelbenayoun:save_tp

michaelbenayoun force pushed from 37b7fbcc to 045b9b14 28 days ago

michaelbenayoun force pushed from 045b9b14 to e8aa0414 23 days ago

feat: add hook for lora.Embedding

720c7d6f

fix: typos, deepcopy, minor

2c96acf0

feat: add saving support

f036b7c9

fix: remove embedding_colwise support for now

cf9fe8bc

refactor: cleanup implementation

c31f7dad

fix: broken conflict resolution during rebase

1c003287

fix: missing import and tests

dec1e01e

fix: replace logger.warning by warnings.warn

96e7d5cf

michaelbenayoun force pushed from e8aa0414 to 96e7d5cf 22 days ago

fix: make tp work with low level api

d5cb7389

style: ruff

dfcf687c

style: docstring

762c3c30

michaelbenayoun requested a review from

BenjaminBossan 21 days ago

BenjaminBossan requested changes on 2026-03-26

fix: restore unrelated files

70de544b

refactor: move TpInfo

4afada5f

fix: remove unnecessary check

70dd58de

refactor: simplify iteration on tp_plan

227baf3b

refactor: use fixture

76353c29

refactor: import safetensors at module level

4bff6d5c

feat: check if tp plan exists before looping over modules

e31d715f

feat: load_adapter support

49b460dc

test: use monitored barrier

a7a52e9f

test: rename tests

5a2a4f7c

fix: test name

c6b67168

test: skip if not latest transformers release

852a5049

feat: update function signature for TP hooks

a4417a1f

BenjaminBossan requested changes on 2026-04-10

BenjaminBossan approved these changes on 2026-04-10

BenjaminBossan merged 07a1db6f into main 5 days ago

michaelbenayoun deleted the save_tp branch 5 days ago

Reviewers

BenjaminBossan

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

peft Save checkpoint with TP #3096 Merged

Save checkpoint with TP #3096

peft
Save checkpoint with TP
#3096

Merged