how much more fix

Dockerfile

@@ -1,15 +1,16 @@
 FROM ollama/ollama:latest
 
-# Pull model during build
 RUN ollama serve & \
     sleep 10 && \
     ollama pull embeddinggemma && \
     pkill ollama
 
-COPY start.sh /start.sh
-
-RUN chmod +x /start.sh
+ENV OLLAMA_HOST=0.0.0.0:7860
+ENV OLLAMA_NUM_PARALLEL=1
+ENV OLLAMA_MAX_QUEUE=128
+ENV OLLAMA_KEEP_ALIVE=1h
 
 EXPOSE 7860
 
-ENTRYPOINT ["/start.sh"]
+ENTRYPOINT ["ollama"]
+CMD ["serve"]

start.sh

@@ -2,13 +2,13 @@
 
 export OLLAMA_HOST=0.0.0.0:7860
 
-# Keep model loaded forever
-export OLLAMA_KEEP_ALIVE=-1
+# One request at a time
+export OLLAMA_NUM_PARALLEL=1
 
-# Slightly increase concurrency
-export OLLAMA_NUM_PARALLEL=4
+# Smaller queue
+export OLLAMA_MAX_QUEUE=128
 
-# Larger queue
-export OLLAMA_MAX_QUEUE=1024
+# Keep model alive for 1 hour
+export OLLAMA_KEEP_ALIVE=1h
 
 exec ollama serve