BgGPT Gemma 3 27B IT V2 GGUF Q4_K_M

This repository contains a GGUF export of a partially abliterated variant of BgGPT Gemma 3 27B IT.

Contents

• BgGPT-Gemma-3-27B-IT-V2-Q4_K_M.gguf: text model quantized to Q4_K_M
• mmproj-out-bggpt27b-full-uncensor-v2-F16.gguf: multimodal projector for image support

Note

This model is partially abliterated. The intervention targets refusal behavior in the language model, so outputs can differ from the original instruct model, especially on sensitive topics.

Base model

• Base: INSAIT-Institute/BgGPT-Gemma-3-27B-IT
• Format: GGUF
• Quantization: Q4_K_M
• Multimodal: yes, requires the included mmproj file

Example with llama.cpp

./llama-mtmd-cli   -m BgGPT-Gemma-3-27B-IT-V2-Q4_K_M.gguf   --mmproj mmproj-out-bggpt27b-full-uncensor-v2-F16.gguf   --image your_image.jpg   -p "What is shown in this image?"   -dev CUDA0 -ngl all