Feature request:
How can i get logits (probabilites of each next token), during generation, just like I can do it in Open AI API (logprobs)? This feature will be helpfull for apps, that use logprobs to measure model avareness and confidence.