[feat] Support more precision (e.g., nvfp4) for IPC weight loading - #32398

Open

xuantengh wants to merge 16 commits into

sgl-project:mainfrom

xuantengh:fast_recovery

Commits on Aug 7, 2026

add nvfp4
xuantengh
committed
77ac535
fix conflict with main
xuantengh
committed
a442253
remove unnecessary configs and utils
xuantengh
committed
6c5347d
remove some comments
xuantengh
committed
98802e2
restore the comments
xuantengh
committed
0569e5c
validate nvfp4 quant config
xuantengh
committed
a858475
limit diff change region
xuantengh
committed
1b1289e
map quant_method to rebind method
xuantengh
committed
07541ff
move the transfer logic into quant_methods
xuantengh
committed
c62b435
refactor
xuantengh
committed
122701b
refactor
xuantengh
committed
1390e7e
refactor minor
xuantengh
committed
f7eb60b
restore interface
xuantengh
committed
a096f38
split env and quant stamp
xuantengh
committed
292ffed
update type
xuantengh
committed
dd21915
dedicate each weight cache quant states class to init backend
xuantengh
committed
89a304f