Diego Fiori comments

Results 58 comments of


                                            Diego Fiori

Input/completion in reward training vs. RL training

Hi @menandro, thank you for reporting the mismatch! I think it's just a typo that we need to fix as soon as possible. Would you mind opening a PR to...

[Chatllama]: MultiGPU support for training

Hi @TejaGollapudi, thank you very much for reaching out. We are currently working on supporting the Accelerate library. You can check directly the updates on the PR #233.

Refactor of models and trainers with base class for common methods

@PierpaoloSorbellini please add a description of what this PR is adding in terms of features and which bugs it is fixing.

[ChatLLaMA] RLHF Training: Prompt too long

Hi @swang99, thank you for reaching out! What model are you currently using?

[Chatllama] Merge the datasets to create more insightful training data

Hello @mohsinmahmood12, thanks a lot for the interest in ChatLLaMA. I assigned the issue to you! Let us know if you face any difficulties with the task 😄

[Chatllama]why use entropy as a part of loss

Hi @HuangLK, good question! In the paper they maximize entropy, while in the code we minimize `-entropy` which is mathematically equivalent to their approach. Clearly it is possible that we...

[Chatllama] Training Japanese language for finetuning or training from scratch

Hi @TakafumiYano, thank you for reaching out. Personally, I would suggest to mix your actual training data with some datasets available open-source as SHP and Anthropic HH RLHF, which are...

[Chatllama] Supervised Finetune on llama-7B

Hello @TonyZhanghm, thank you very much for reaching out. I'll investigate the errors you are getting. Could you please share with us the `config.yaml` file you are currently using?