more optimization

.dockerignore

@@ -0,0 +1,7 @@
+.git
+.gitignore
+README.md
+__pycache__
+*.pyc
+.env
+test.py

.gitignore

@@ -0,0 +1 @@
+test.py

Dockerfile

@@ -1,4 +1,4 @@
-FROM ollama/ollama
+FROM ollama/ollama:latest
 
 # Pull model during build
 RUN ollama serve & \

start.sh

@@ -1,7 +1,14 @@
 #!/bin/sh
 
-# Listen on HF's expected port
 export OLLAMA_HOST=0.0.0.0:7860
 
-# Start Ollama in foreground
+# Keep model loaded forever
+export OLLAMA_KEEP_ALIVE=-1
+
+# Slightly increase concurrency
+export OLLAMA_NUM_PARALLEL=4
+
+# Larger queue
+export OLLAMA_MAX_QUEUE=1024
+
 exec ollama serve