Qwen3_4b abliterated with refusal rate of 16/100.KL divergence: 0.0000this can be used as Text encoder for models like Z-image/z-image-turbo/Flux2Klein 4B or as a chat assistant eg can be used inside of LLm loaders
v1.00