vLLM Adds Day-0 Support for MiniMax-M3.1-Flash
vLLM Blog reports day-0 support for MiniMax-M3.1-Flash on NVIDIA Blackwell GPUs, NVIDIA Rubin GPUs with dedicated kernels, and AMD MI350-series GPUs. Prebuilt NVIDIA and AMD images are available. The implementation uses MiniMax Sparse Attention, an NVFP4 KV cache, NVFP4-quantized MoE weights and a DSpark drafter. vLLM reports gains in kernel and serving performance, including optimized agentic serving on GB300, Rubin and MI355X hardware. It also reports day-0 end-to-end reinforcement-learning support with veRL, FSDP2 and LoRA, demonstrated on 16 GB300 GPUs.
The integration provides optimized inference paths for the model across NVIDIA Blackwell, NVIDIA Rubin and AMD MI350-series hardware. The reinforcement-learning support and results were demonstrated specifically on NVIDIA GB300 systems.
MiniMax-M3.1-Flash is the latest model in the MiniMax-M3 family, with improvements across reasoning, coding, agentic, and visual tasks. It is also built for inference efficiency, with a native NVFP4 KV cache, NVFP4-quantized MoE weights, and a DSpark draft model for speculative decoding.
vLLM supports MiniMax-M3.1-Flash on day 0 on NVIDIA Blackwell GPUs, on NVIDIA Rubin GPUs with dedicated Rubin kernels, and on AMD MI350-series GPUs, with strong performance on agentic workloads.
TL;DR
- MiniMax Sparse Attention (MSA) with an NVFP4 KV cache for both the indexer and the attention path, plus NVFP4 MoE weights and a DSpark drafter.
- Blackwell MSA kernels use the new CUDA 13.4
QMUL4instruction to dequantize NVFP4 into FP8 inside the kernel: up to 1.7× faster indexer QK decode and 1.3× faster sparse attention decode. - NVFP4 MegaMoE with MXFP8 shared-expert fusion and early TMEM release: 20% lower kernel latency in prefill and 7% higher prefill throughput.
- Indexer decode context parallelism (DCP): up to 3.3× faster indexer per sparse layer at 512K context on 4× GB300.
- QKV GEMM + norm + RoPE + KV-insert fusion: 1.1–1.6× faster prefill at 512+ tokens and about 19% lower P50 TTFT at TP4.
- FlashInfer PrimiTS backend for the DSpark drafter attention: 1.3–2× faster than FlashAttention 4, 5% lower total forward time.
- Rubin (SM107) MSA kernels are also integrated into vLLM with state-of-the-art performance.
- AMD MI350 series: the NVFP4-native MSA kernel, indexer decode context parallelism, dual-batch overlap, and the mono kernel deliver a 2–3× performance boost across the board.
- Agentic Serving: on NVIDIA GB300, up to 418.9 tok/s per user at the latency-sensitive end; on NVIDIA Rubin, up to 724.9 tok/s per user and 191.5K total tok/s per GPU; on AMD MI355X, up to 133.3M total tokens per dollar of TCO on the SemiAnalysis AgentX benchmark.
- Day-0 RL support: end-to-end RL with veRL, FSDP2, and LoRA on the NVFP4 checkpoint.
#MiniMax-M3.1-Flash#vLLM#MiniMax#NVIDIA Blackwell#NVIDIA Rubin#AMD MI350#MiniMax Sparse Attention#DSpark