MiniMax M3.1 Flash open weights: NeMo RL recipe with GRPO, CP2 + EP128 and vLLM
An NVIDIA-side guide shows RL fine-tuning MiniMax M3.1 Flash with NeMo RL: GRPO on the AutoModel backend, context plus expert parallelism (CP2 + EP128) and vLLM generation.
Space Bunny used to live behind an API. Today it lives on your GPUs. thank you @MiniMax_AI for the open weights. 🐰 MiniMax-M3.1-Flash is a seriously impressive model: frontier-level agentic coding, a big jump on the AA Intelligence Index, topping OpenRouter for 10 straight days. RL it today with NeMo RL: GRPO on the AutoModel backend with context + expert parallelism (CP2 + EP128) and vLLM generation: https://github.com/NVIDIA-NeMo/RL/blob/minimax-m3-1-flash/docs/guides/models/minimax/minimax-m3-1-flash.md
This post quotes another post: “MiniMax M3.1 Flash open weights: 428B total, 23B activated”