medgemma 27b but now with image support

#1143
by jacek2024 - opened

Wow amazing. Let's see how @drwlf 's DrMedra 4B finetune using the medra-thinking-768 dataset turns out and we might finally have a base model justifying the 1000 to 2000 hours of training required for a DrMedra finetune of Gemma 3 27B

It's queued! :D
Hopefully mmproj extraction will be performed as it wasn't flagged as vision model due to beeing gated but I manualy queued it to nico1 and specified a token with access to the model.

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#medgemma-27b-it-GGUF for quants to appear.

Edit: Nice it did do the mmproj extraction.

(the vision detection at queuing time only affects the worker selection)

Sign up or log in to comment