Dockerfile

FROM ghcr.io/ggml-org/llama.cpp:full

RUN apt update && apt install wget -y

RUN wget "https://huggingface.co/mradermacher/coconut-qwen2.5-7b-GGUF/resolve/main/coconut-qwen2.5-7b.Q4_K_S.gguf" -O /coconut-qwen2.5-7b.Q4_K_S.gguf


CMD ["--server", "-m", "coconut-qwen2.5-7b.Q4_K_S.gguf", "--port", "7860", "--host", "0.0.0.0", "-n", "512", "--no-warmup"]