[contrib] Add MiniMax-M2 (229B / ~10B active MoE, TP=64, EP=64)#138

Open

whn09 wants to merge 16 commits intoaws-neuron:mainfrom

whn09:contrib/MiniMax-M2

Commits on Apr 22, 2026

[contrib] Add MiniMax-M2 NeuronX port (TP=64, EP=64 MoE)

whn09
and
claude
committed
Switch vllm-neuron patch to a runtime registration hook on release-0.5.0

whn09
and
claude
committed
vllm-neuron patch: move registration into _get_neuron_model_cls

whn09
and
claude
committed
vllm-neuron patch: also monkey-patch AutoConfig for trust_remote_code

whn09
and
claude
committed
bench: set use_torch_block_wise=true to avoid missing NKI kernel

whn09
and
claude
committed
bench: extend wait_for_server timeout to 2h for MoE first-compile

whn09
and
claude
committed
perf_test: add sanity_check.sh and run_bench_single.sh helpers

whn09
and
claude
committed

Commits on Apr 25, 2026