CTranslate2 icon indicating copy to clipboard operation
CTranslate2 copied to clipboard

Published 20 hours ago • verified publisher icon OpenNMT

Trouble converting Mistral-Nemo despite architecture indicating MistralForCausalLM

Open BBC-Esq opened this issue 7 months ago • 6 comments

Has anyone had luck converting the model located here:

https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407

I haven't. I've even tried renaming the consolidated.safetensors file to model.safetensors just to be safe, but no dice. I'm wondering if despite the architecture being Mistral, there's some kind of nuance that Ctranslate2 didn't take account of. I noticed that the HF Repo indicates that only the development version of Transformers supports it, not the latest PyPi release so...this lends credibility to my hypothesis but I'm no expert.

Thanks!

BBC-Esq avatar Jul 20 '24 15:07 BBC-Esq