Qwen3.8-27B-Ridge-GGUF: Local Multimodal with Image Support via 12.6GB GGUF

This is a GGUF quantized version of the Qwen3.8-27B model, performing multimodal inference that simultaneously understands images and text. Distributed under the Apache-2.0 license, it allows for commercial use and supports English and Chinese.

It is compatible with various local inference frameworks such as llama.cpp, vLLM, and Ollama, enabling CPU-based execution without a GPU. In particular, it is possible to launch an OpenAI-compatible API server or perform direct inference in the terminal via llama.cpp.

By applying the Gated-DeltaNet architecture and MTP (Multi-Token Prediction) technology, it provides more efficient inference than previous methods. Three levels of reasoning effort—xhigh, medium, and low—can be configured to adjust the depth of thought according to task difficulty.

It is suitable for developers looking to build multimodal AI while maintaining privacy in a local environment. It provides a 12.6GB GGUF file along with an mmproj file, facilitating vision model configuration.