return to table of content

Training LLMs to Reason in a Continuous Latent Space