Fix/multi lora ep memory efficient init - #265

Draft

meichangsu1 wants to merge 13 commits into

modelscope:mainfrom

meichangsu1:fix/multi-lora-ep-memory-efficient-init

Commits on Aug 12, 2026

fix(transformers): support memory-efficient multi-LoRA EP init
meichangsu1
committed
c5ac8c6
fix(transformers): align multi-LoRA dtype before FSDP
meichangsu1
committed
58edad2
wip
meichangsu1
committed
50eedab
wip
meichangsu1
committed
5a1eade

Commits on Aug 13, 2026

wip
meichangsu1
committed
a7bff39

Commits on Aug 14, 2026

wip
meichangsu1
committed
b06ae3e
wip
meichangsu1
committed
d2b54ff
multinode script
meichangsu1
committed
d62c4d5
wip
meichangsu1
committed
485733f

Commits on Aug 16, 2026

wip
meichangsu1
committed
a1f1a16

Commits on Aug 18, 2026

wip
meichangsu1
committed
ab0b8bc

Commits on Aug 20, 2026

wip
meichangsu1
committed
620aff4

Commits on Aug 21, 2026

fix statedict
meichangsu1
committed
21f74da