Update app.py

app.py

@@ -57,8 +57,11 @@ def get_llm(model_id: str = MODEL_OPTIONS[0], max_tokens: int = 256, temperature
     Returns API URL, headers, and parameters for HuggingFace Inference API.
     """
     token = get_huggingface_token()
-    api_url = f"https://router.huggingface.co/models/{model_id}"
-    headers = {"Authorization": f"Bearer {token}"}
+    api_url = f"https://api-inference.huggingface.co/models/{model_id}"
+    headers = {
+        "Authorization": f"Bearer {token}",
+        "Content-Type": "application/json"
+    }
     
     return api_url, headers, max_tokens, temperature