[atom-vllm] enable minimax_m3 for atom vllm#1408

Open

lirui927 wants to merge 16 commits into

mainfrom

lirui/m3_vllm_0630

Commits on Jun 30, 2026

[MiniMax-M3] support sparse MHA serving in vLLM

authored and
XiaobingSuper
committed
[MiniMax-M3] optimize sparse MHA vLLM output path

authored and
XiaobingSuper
committed
[MiniMax-M3] initialize sparse MHA vLLM cache state

XiaobingSuper
and
cursoragent
committed
[MiniMax-M3] register ATOM MXFP8 quant config

XiaobingSuper
and
cursoragent
committed
[MiniMax-M3] fix sparse MHA fp8 metadata alignment
XiaobingSuper
committed
[MiniMax-M3] add vLLM attention correctness baseline
XiaobingSuper
committed
[MiniMax-M3] route vLLM attention through ATOM adapters

lirui927
authored and
XiaobingSuper
committed
[MiniMax-M3] clean up vLLM sparse attention adapter

lirui927
authored and
XiaobingSuper
committed
[MiniMax-M3] stabilize vLLM graph attention path

lirui927
authored and
XiaobingSuper
committed
[MiniMax-M3] restore core files and stabilize sparse capture
lirui927
committed
[MiniMax-M3] simplify vLLM attention adapter
lirui927
committed
[MiniMax-M3] fix sparse fp8 KV cache scales
lirui927
committed
[MiniMax-M3] add vLLM recipe
lirui927
committed

Commits on Jul 1, 2026