mistralai

/

Mistral-7B-Instruct-v0.2

like 2.36k

set sliding_window to 0 or back to 4096 in the config until they fix transformers code, it won't hurt the finetuning

Hi @asmittal can you try with the latest version of transformers? I think this has been fixed with the attention mask refactor pip install -U transformers

@asmittal how you do fine tuning of this model ? do you have some guide ? I tried axolotl but no success

@Nondzu I don't really have a guide, but I'm fine tuning this model like any other LLM, I think prompt is the only thing that makes any difference.

@ybelkada Thanks alot! It's working with the latest version of transformers.