According to Foresight News, decentralized AI protocol Prime Intellect is set to open source its distributed reinforcement learning model, INTELLECT-2. This model supports decentralized training research and introduces the PRIME-RL framework, TOPLOC verification, and SHARDCAST weight distribution to enhance training stability. It is based on the QwQ-32B model, improving mathematical and coding performance. Future plans include increasing inference computation ratio, tool invocation, multi-round reinforcement learning, and model integration.

#AI