[NeurIPS 2025 Spotlight] LLM post-training suite for long-CoT reasoning, PRM, and code generation — featuring ReasonFlux, ReasonFlux-PRM, and ReasonFlux-Coder.

realize the reinforcement learning training for gpt2 llama bloom and so on llm model