accelerate
(Part 1) fix: make TP training compatible with new transformers
#3457

Merged

(Part 1) fix: make TP training compatible with new transformers #3457

S1ro1 merged 7 commits into huggingface:main from kmehant:tp-compa

SunMarc commented on 2025-03-25

kmehant requested a review from

SunMarc 1 year ago

muellerzr approved these changes on 2025-03-25

kmehant changed the title ~~fix: make TP training compatible with new transformers~~ (Part 1) fix: make TP training compatible with new transformers 1 year ago

kmehant force pushed from 4bf02843 to 72d52c24 1 year ago

SunMarc commented on 2025-03-28

S1ro1 commented on 2025-04-07

S1ro1 commented on 2025-04-10

kmehant force pushed from 72d52c24 to ff804c11 1 year ago

S1ro1 approved these changes on 2025-04-10

kmehant requested a review from

SunMarc 1 year ago

SunMarc commented on 2025-04-10

kmehant force pushed from ae5fed7e to e3e833fa 1 year ago

kmehant force pushed from e3e833fa to b16eb905 1 year ago

kmehant force pushed from b16eb905 to b26fcb5f 1 year ago

kmehant force pushed from b26fcb5f to 862bffce 1 year ago

kmehant force pushed from 862bffce to 398afa38 1 year ago

SunMarc commented on 2025-04-11

SunMarc approved these changes on 2025-04-11

feat: support new tp refactor for training

6fb90891

fix: @S1ro1 review cmt

55abbb71

fix: @S1ro1 review cmt - tp_plan flag docstr

80828ef5

fix: @SunMarc review cmt on un used flag

552e9e95

fix: pick approach 3 as discussed in the PR

b129999f

fix: styling errors

02b98cd0

fix: bump up transformers for tp_size feature

0f7e998f

kmehant force pushed from e99f0c31 to 0f7e998f 1 year ago

S1ro1 merged 67adb473 into main 1 year ago

Reviewers

SunMarc

S1ro1

muellerzr

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

accelerate (Part 1) fix: make TP training compatible with new transformers #3457 Merged

(Part 1) fix: make TP training compatible with new transformers #3457

accelerate
(Part 1) fix: make TP training compatible with new transformers
#3457

Merged