No description provided.

This PR copies the fixed mistral-common v3 tokenizer from this PR, minus the chat template.

Unofficial Mistral Community org

Hi @Rocketknight1
Have you tested this change yourself? Will it work without any issue?

Hi @MaziyarPanahi , it was tested against the Mistral v3 tokenizer for chat completions and the string outputs and token sequences were identical. I'm not sure if the Mistral tokenizer even allows standard string tokenizations though - is there a way to check that?

Unofficial Mistral Community org

Thanks @Rocketknight1
I will merge and we can test it. It is simpler this way.

Thanks again for your contribution

MaziyarPanahi changed pull request status to merged

Sign up or log in to comment