Create start.sh

start.sh

@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+MODEL_NAME="${MODEL_NAME:-qwen2.5:7b-instruct-q4_K_M}"
+OLLAMA_URL="${OLLAMA_URL:-http://127.0.0.1:11434}"
+
+# Start Ollama server in background
+ollama serve > /tmp/ollama.log 2>&1 &
+OLLAMA_PID=$!
+
+# Wait until Ollama is ready
+python3 - << 'PY'
+import time, urllib.request
+url = "http://127.0.0.1:11434/api/tags"
+for _ in range(120):
+    try:
+        urllib.request.urlopen(url, timeout=1).read()
+        print("Ollama is up")
+        raise SystemExit(0)
+    except Exception:
+        time.sleep(1)
+raise SystemExit("Ollama did not start")
+PY
+
+# Pull model (first run will download it)
+ollama pull "${MODEL_NAME}"
+
+# Start Gradio app
+exec python3 /app/app.py