Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF: Uncensored 27B Multimodal Model with MTP for Maximum Local Inference Speed

This repository distributes a 27B parameter model based on the Qwen3.8 architecture, converted to GGUF format. It features a multimodal structure capable of processing both text and image inputs, and operates in English, Chinese, and multilingual environments. Released under the Apache-2.0 license, it can be freely used in both commercial and non-commercial projects.

Unlike typical quantized models, it is provided in 'Uncensored' and 'Aggressive' variants with content filtering removed. It also applies MTP (Multi-Token Prediction) and FastMTP technologies to enhance inference speed. It is configured for immediate use in major local inference frameworks such as llama.cpp, vLLM, and Ollama.

Files with various quantization levels, from IQ2_M to Q8_K_P, are included, allowing memory usage to be adjusted according to hardware specifications. mmproj files with BF16 precision are also provided to maintain vision processing performance. FastMTP-specific files supporting a 32K context window are included, making it suitable for processing long documents or conversations.

It is suitable for developers who want to reduce cloud API dependency and achieve high-quality responses in local environments. It is particularly valuable for building agents and chatbots that require uncensored, free generation or MTP-based fast inference performance. Its stability has been verified through 304 likes and active community discussions.