Versi-StyleTune-31B

A text-only Gemma 4 31B model: Nimbz/Versipellis-31B as the base, with the output projection (lm_head) grafted from Gryphe/Gemma-4-31B-StyleTune.

All credits go to Nimbz and Gryphe for the original models, I only committed the merge.

Recipe

  • Base: Versipellis-31B, text layer extracted (vision tower removed).
  • Output head: StyleTune lm_head.weight, with tie_word_embeddings set to false.
  • No layer-level merging — Versipellis weights are otherwise unchanged.

Notes

  • Text-only (vision/audio removed).
  • Run with the Gemma 4 chat template; thinking off by default.
Downloads last month
5
Safetensors
Model size
32B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Iloqt/Versi-StyleTune-31B

Merge model
this model
Quantizations
3 models