Modelpublic

debais/G4-MeroMero-26B-A4B-StyleSwap-GGUF

sourceHugging Facegemmaupdated 3mo agoView on Hugging Face

1likes106downloads

Model Card

G4-MeroMero-26B-A4B — StyleSwap (Q8_0 GGUF)

This is `zerofata/G4-MeroMero-26B-A4B-gguf` with its output.weight (the lm_head) tensor grafted on from `mradermacher/Gemma-4-26B-A4B-StyleTune-GGUF`.

Everything else — all other tensors, metadata, tokenizer, and chat template — is zerofata's, unchanged. Only the head, which controls how hidden states map to vocabulary logits, comes from StyleTune. Load and run it like any other Gemma-4 GGUF.

How it was made

Made with stylegraft.py, from `andyoneal`. It works for any Gemma 4 26B or 31B model (source and target must share hidden size and vocab):

bash

stylegraft TARGET.gguf SOURCE.gguf OUTPUT.gguf

This release was built with:

bash

stylegraft G4-MeroMero-26B-A4B-Q8_0.gguf \
           Gemma-4-26B-A4B-StyleTune.Q8_0.gguf \
           G4-MeroMero-26B-A4B-StyleSwap-Q8_0.gguf

License

Model weights: Gemma Terms of Use.