From 284460c5954956c7c279abd1c5aac9ffdb5807da Mon Sep 17 00:00:00 2001 From: Faradawn Yang <73060648+faradawn@users.noreply.github.com> Date: Wed, 11 Mar 2026 09:32:08 -0700 Subject: [PATCH 1/2] Add FP8 B200 best config to MiniMax-M2.5 recipe Signed-off-by: Faradawn Yang <73060648+faradawn@users.noreply.github.com> --- MiniMax/MiniMax-M2.md | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) diff --git a/MiniMax/MiniMax-M2.md b/MiniMax/MiniMax-M2.md index 56f778967..16eacd48e 100644 --- a/MiniMax/MiniMax-M2.md +++ b/MiniMax/MiniMax-M2.md @@ -83,9 +83,24 @@ uv pip install vllm \ ## Launching M2.5/M2.1/M2 with vLLM -You can use 4x H200/H20 or 4x A100/A800 GPUs to launch this model. +You can use 2x B200, 4x H200/H20, or 4x A100/A800 GPUs to launch this model. -run tensor-parallel like this: +For 2x B200: FP8 serving on B200 GPUs with FlashInfer MoE FP8 and DeepGEMM MoE disabled: + +```bash +export VLLM_USE_FLASHINFER_MOE_FP8=0 +export VLLM_MOE_USE_DEEP_GEMM=0 + +vllm serve MiniMaxAI/MiniMax-M2.5 \ + --tensor-parallel-size 2 \ + --gpu-memory-utilization 0.95 \ + --block-size 32 \ + --tool-call-parser minimax_m2 \ + --reasoning-parser minimax_m2_append_think \ + --enable-auto-tool-choice +``` + +For 4x H200/H20 or 4x A100/A800, run like this ```bash vllm serve MiniMaxAI/MiniMax-M2.5 \ From ce967d3b64c51d150d28aa27c7db9b344058919e Mon Sep 17 00:00:00 2001 From: Faradawn Yang <73060648+faradawn@users.noreply.github.com> Date: Fri, 13 Mar 2026 10:04:10 -0700 Subject: [PATCH 2/2] Update MiniMax/MiniMax-M2.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Signed-off-by: Faradawn Yang <73060648+faradawn@users.noreply.github.com> --- MiniMax/MiniMax-M2.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/MiniMax/MiniMax-M2.md b/MiniMax/MiniMax-M2.md index f8983df5c..9e8eecbd6 100644 --- a/MiniMax/MiniMax-M2.md +++ b/MiniMax/MiniMax-M2.md @@ -120,7 +120,7 @@ vllm serve MiniMaxAI/MiniMax-M2.5 \ --enable-auto-tool-choice ``` -For 4x H200/H20 or 4x A100/A800, run like this +For 4x H200/H20 or 4x A100/A800: Standard serving configuration ```bash vllm serve MiniMaxAI/MiniMax-M2.5 \