search

thunlp / UltraChat

Large-scale, Informative, and Diverse Multi-round Chat Data (and Models)
MIT License
2.22k stars 114 forks source link

Support Fine-tune on PeFT adaptation #4

Open Ejafa opened 1 year ago

Ejafa commented 1 year ago

This project is great and the dataset is unique. To provide help to the community, it will be a great idea to support PeFT training on this dataset. Also, there's a chance to increase the training to RLHF if PEFT is supported (https://huggingface.co/CarperAI/vicuna-13b-fine-tuned-rlhf). We can compare the finetuned model with Vicuna as their dataset is also a synthetic dataset (ShareGPT)

ningding97 commented 1 year ago

This project is great and the dataset is unique. To provide help to the community, it will be a great idea to support PeFT training on this dataset. Also, there's a chance to increase the training to RLHF if PEFT is supported (https://huggingface.co/CarperAI/vicuna-13b-fine-tuned-rlhf).

We can compare the finetuned model with Vicuna as their dataset is also a synthetic dataset (ShareGPT)

Thanks, it is a great suggestion. We are working on a full model and will then train peft modules.