Update app.py

app.py

@@ -20,14 +20,11 @@ class ChatRequest(BaseModel):
 
 @app.on_event("startup")
 def load_model():
-    global llm
-    model_path = "phi-2.Q4_K_M.gguf"
+    with open("/tmp/model_path.txt", "r") as f:
+        model_path = f.read().strip()
+
     if not os.path.exists(model_path):
         raise RuntimeError(f"Model not found: {model_path}")
-    llm = Llama(
-        model_path=model_path,
-        n_ctx=2048,
-        n_threads=2
     )
 
 @app.post("/v1/chat/completions")