ConicCat/Gemma4-GarnetV3-31B

GarnetV3 is a finetune of Gemma 4 focused on improving roleplay and writing performance using DPO with an emphasis on prose quality and humanlike characters. The dataset consists of approximately 1/3rds writing and 2/3ds roleplay.

How to Get Started with the Model

I recommend grabbing the Q4_K_M gguf and using koboldcpp.

Training Details

Training was conducted on 1xA100 80GB for 9 hours.

Datasets

• ConicCat/Lamp_P_Preference: Human revised vs AI writing to improve prose.
• ConicCat/Charcards_Delta_Qwen3_5V2: Qwen3.5 27B vs Qwen3.5 2B using the AI2 delta tuning recipe for data generation.
• ConicCat/Charcards_Context_Distill_Gemma4_26BV2: Gemma 4 26B with full context as chosen and missing context as rejected.