Curriculum Supervised Fine-Tuning
A supervised fine-tuning method that draws on the idea of curriculum learning, arranging training samples from easy to hard by difficulty or form so that the model transitions smoothly to the target distribution.
Timeline (last 90 days)
The LARC training pipeline is divided into two stages: curriculum supervised fine-tuning (curriculum SFT) and information-gain reinforcement learning (information-gain RL).
Curriculum SFT provides a smooth transition from discrete to continuous by gradually converting token-level reasoning into latent reasoning, avoiding training instability caused by the model jumping directly into latent space.
All Facts (2)
The LARC training pipeline is divided into two stages: curriculum supervised fine-tuning (curriculum SFT) and information-gain reinforcement learning (information-gain RL).
50%UnverifiedCurriculum SFT provides a smooth transition from discrete to continuous by gradually converting token-level reasoning into latent reasoning, avoiding training instability caused by the model jumping directly into latent space.
50%