MiniMax M2.1 229B

Aurora system unifies RL training and serving for faster LLM inference

Researchers have developed Aurora, a novel system that unifies the training and serving of speculative decoding for large language models. This approach addresses the delays and performance degradation associated with t…