gghfez

/

gemma-3-12b-novision

like 2

Mostly for tool compatibility. If the multimodal model is working for you, this probably doesn't offer you anything now.

I made these myself so I could:

• Train control-vectors for Gemma3

• Create / run EXL2 quants in a modified exllamav2. I managed to get text working but vision was beyond me.
  (no longer needed as exllamav2 supports the gemma3 properly on dev branch)

• Abliteration tools weren't working with vision
  (these are probably updated by now as I see abliterated multimodal models on HF)

• I was having issues finetuning the multimodal models just after the model release.
  (No longer needed as unsloth supports multimodal now)