360 Intelligent Brain launches Light-IF series models.

On August 12, the 360 Brain team announced the new Light-IF framework, with preview-self-testing reasoning and information entropy control as its core, to improve the model's compliance with complex instructions. The Light-IF framework includes five key steps: difficulty-aware instruction generation, Zero-RL reinforcement learning, reasoning pattern extraction and filtering, entropy-preserving supervised cold start, and entropy-adaptive regular reinforcement learning. The Light-IF-32B/14B/8B/4B/1.7B full series models will be gradually released on Hugging Face.