12b variant?

#1
by Moleculo - opened

Hello, thanks for you work, do you plan 12b variant?

This quants are based of https://huggingface.co/jbeslt/Gemma-3-27B-Heretic from jbeslt. I assume with 12B variant you mean https://huggingface.co/p-e-w/gemma-3-12b-it-heretic from p-e-w?

It's queued! :D
Unfortunately it will be slightly delayed as we are currently computing the importance matrix for Dolphin-X1-Llama-3.1-405B on nico1 which will probably take at least another 12 hours. nico1 is the only worker capable of quantizing vision models.

You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#gemma-3-12b-it-heretic-GGUF for quants to appear.

Heretic is such a cool technique to remove censorship of models. I really should start using it myself: https://github.com/p-e-w/heretic

There was a small delay as the model missed processor_config.json and preprocessor_config.json which made it fail MMPROJ vision extraction. I now fixed this issue by copying the required files form google/gemma-3-12b-it. The model is now finally getting processed and should be completed in an hour.

Oh no even more missing files. That is crazy. Luckely tokenizer.model can be copied from google/gemma-3-12b-it as well.

-2000   25 si gemma-3-12b-it-heretic                       error/1 missing tokenizer.model

MMPROJ vision extraction and GGUF conversion finally completed successfully. and the importance matrix is currently being computed so it should be completed in an hour.

Sign up or log in to comment