Use GPU

Dockerfile

@@ -1,4 +1,4 @@
-FROM ghcr.io/ggerganov/llama.cpp:server
+FROM ghcr.io/ggerganov/llama.cpp:server-cuda
 
 RUN apt update && apt install -y curl
 
@@ -8,4 +8,4 @@ RUN curl -L https://huggingface.co/TheBloke/deepseek-coder-6.7B-instruct-GGUF/re
 
 ENTRYPOINT [ "/server" ]
 
-CMD [ "--host", "0.0.0.0", "--port", "7860", "--model", "/models/deepseek-coder-6.7b-instruct.Q6_K.gguf", "-c", "4096", "--chat-template", "deepseek" ]
+CMD [ "-ngl", "32" "--host", "0.0.0.0", "--port", "7860", "--model", "/models/deepseek-coder-6.7b-instruct.Q6_K.gguf", "-c", "4096", "--chat-template", "deepseek" ]