Latent-GRPO: Reinforcement Learning in Continuous Thought Space

Automated excerpt

By replacing discrete vocabulary tokens with continuous recurrent thought vectors in embedding space, models can reason internally across continuous manifold dimensions. Latent thought step: A recurrent model cycle where continuous vectors loop internally without emitting vocabulary logits. GRPO gradients only optimize the subsequent discrete action tokens.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.