Compile on startup

Dockerfile

@@ -20,15 +20,9 @@ WORKDIR $HOME
 RUN mkdir app
 WORKDIR $HOME/app
 
-# Build llama.cpp (CPU, only `quantize`)
-# Should we pin a known commit?
 RUN git clone https://github.com/ggerganov/llama.cpp.git
-WORKDIR $HOME/app/llama.cpp
-RUN make quantize
-WORKDIR $HOME/app
 
 # App and Python reqs
-# Should we pin versions here?
 COPY --chown=user ./ $HOME/app
 RUN pip install --no-cache-dir -r requirements.txt
 RUN pip install --no-cache-dir -r llama.cpp/requirements.txt
@@ -42,4 +36,4 @@ ENV GRADIO_THEME huggingface
 ENV SYSTEM spaces
 ENV HF_HUB_ENABLE_HF_TRANSFER 1
 
-#CMD ["python", "app.py"]
+ENTRYPOINT /bin/sh start.sh

start.sh

@@ -0,0 +1,4 @@
+cd llama.cpp
+make quantize
+cd ..
+python app.py