return to table of content
Training LLMs to Reason in a Continuous Latent Space