PR #3817 Deepspeed Ulysses/ALST integration

Deepspeed Ulysses/ALST integration #3817

kashif merged 53 commits into huggingface:main from stas00:alst-integration

Feat: initial impl

439fa7ea

improve

e238f7bb

SunMarc commented on 2025-10-22

s/flavour/backend/

876fa2db

style + ver

b016317c

better check

a63d0948

check

73d3dbce

docs + example

8b4a4f41

add tests

76288a8d

add tests

209eab73

cleanup

c013677f

cleanup

685453c3

SunMarc approved these changes on 2025-10-23

Apply suggestions from code review

a3969042

add experimental notice

fb80e025

style

21a4a2d0

Merge branch 'alst-integration' of https://github.com/stas00/accelera…

05b6ac12

new deepspeed version

60f74930

additional checks + tests

453fb55f

more docs

8677f23a

more docs

8ee9b03c

working now

2e577d3d

style

5c51897d

update docs

83172414

more robust config parsing

94f558bb

fix

a2388cdb

SunMarc approved these changes on 2025-11-04

Apply suggestions from code review

e6e243f3

check backend, integrate ulysses API improvement

2330dcdc

style

9dbcf919

fix default to match the doc

e79034f1

SunMarc approved these changes on 2025-11-05

Apply suggestions from code review

61873c66

fix

756bd9f6

deepspeed=0.18.2 is out

56df6213

kashif commented on 2025-11-09

kashif commented on 2025-11-10

Apply suggestions from code review

380747ca

kashif approved these changes on 2025-11-13

s/cp/sp

38c84fab

fixes

5c2f34e8

SunMarc commented on 2025-11-14

Apply suggestions from code review

190494be

Update src/accelerate/parallelism_config.py

285e24fd

suggestion

a7d2e5d2

SunMarc commented on 2025-11-17

Update docs/source/concept_guides/sequence_parallelism.md

d4ee1566

Update sequence_parallelism.md

99b321a4

fix

b769fc83

fix

04b4dc3e

fix

a4005e72