Any particular difference between this model and TheBloke/laser-dolphin-mixtral-2x7b-dpo-GGUF?

#3
by Kalemnor - opened

Since you did this after TheBloke, is there anything different between this model and the TheBloKe version?

Yes, TheBloke’s version is deprecated. I updated the model to correct template issues and I didn’t wanna bother him to redo everything.

Please use this version for the best results

What were the template issues, and what changes have you brought to the template, I'm going to use this with LM Studio (accept suggestions to make a preset for it).

The template issues were related to the merge. It caused EOS and BOS tokens to be put into the message and not be recognized.

I corrected the chat template to properly use ChatML by performing a new merge with a different model for one of the experts.

I don’t have a template preset for LM studio, but I was messing around with it earlier today using this model. I will post my config later today or tomorrow

Kalemnor changed discussion status to closed
Kalemnor changed discussion status to open

I'll wait for it. In the meanwhile I will stick with ChatML

Sign up or log in to comment