Text Generation
Transformers
Safetensors
English
mixtral
conversational
Inference Endpoints
text-generation-inference

How is it?

#1
by jackboot - opened

Does it follow instructions as good as the mixtral-instruct but without the instability?

Might be a bit broken. I can't seem to get it to reply fully using the built-in prompt template. When it responds, it appears to stop early, after a newline, so maybe an issue with the training and stop token config?

Edit: though, to be fair, getting the right template and responses is a bit fragile with base Mixtral as well. Need more time to test and make sure it's not user error on my side.

I am getting some replies with alpaca. Was unable to generate SD pics in the official format. Would just gen empty and ST froze.

Going through my characters it's ok.. but the magic sauce it doesn't have. Follows instructions above average. I guess it's onto noromaid which is only GGUF format.

Will anyone be able to replicate and top instruct?

edit: it can't handle a 2 character card. One character falls away.

Sign up or log in to comment