New model request

#2
by HR1777 - opened

Is it possible to train Mistral-7B-Instruct-v0.2 (https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) on this dataset (https://huggingface.co/datasets/cris177/Arguments)? This dataset, although relatively unknown, is one of the best available, based on my experience of testing more than 500 models.

If it's possible, I recommend using a learning rate of 5e-07 and running it for 3 epochs to create a well-trained model.

of course! I use Axololt to fine-tune. This dataset has one column, I wonder what should be the format of the dataset (completion?): https://github.com/OpenAccess-AI-Collective/axolotl?tab=readme-ov-file#dataset

I haven't had time to learn how to do custom dataset by maybe pre-processing this into prompt/generation. Once I figure this out I can launch the fine-tuning

Thank you so much. I'm not aware of that either. Whenever you have time and figure it out, please let me know about creating this new model. This dataset is truly excellent, and I'm quite confident that this model will perform exceptionally well.

Sign up or log in to comment