The paper title is "Training Large Language Models to Reason in a Continuous Latent Space". It's true it says training in the title, but the goal (reasoning in

The paper title is "Training Large Language Models to Reason in a Continuous Latent Space". It's true it says training in the title, but the goal (reasoning in continuous space) happens at inference time.