Create main.py

main.py

@@ -0,0 +1,28 @@
+from llama_cpp.server.app import create_app, Settings
+from fastapi.responses import HTMLResponse
+import os
+
+app = create_app(
+    Settings(
+        n_threads=2,  # set to number of cpu cores
+        model="https://huggingface.co/TheBloke/OpenHermes-2-Mistral-7B-GGUF/resolve/main/openhermes-2-mistral-7b.Q4_0.gguf",
+        embedding=False
+    )
+)
+
+# Read the content of index.html once and store it in memory
+with open("index.html", "r") as f:
+    content = f.read()
+
+
+@app.get("/", response_class=HTMLResponse)
+async def read_items():
+    return content
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app,
+                host="0.0.0.0",
+                port=int("2600")
+                )
+