maybe this will fix the stuck on restart?

app.py

@@ -114,16 +114,17 @@ class GaiaAgent:
         # Initialize the LLM
         print("Loading LLM...")
         llm = HuggingFacePipeline.from_model_id(
-            model_id="HuggingFaceH4/zephyr-7b-beta",
+            model_id="microsoft/Phi-3-mini-4k-instruct",
             task="text-generation",
             pipeline_kwargs={
                 "max_new_tokens": 512,
                 "top_k": 50,
                 "temperature": 0.1,
                 "do_sample": False,
-                "torch_dtype": torch.bfloat16,
+                "torch_dtype": "auto",  # Let transformers figure out the best dtype
                 "device_map": "auto",
             },
+            trust_remote_code=True,  # Phi-3 requires this
         )
         print("LLM loaded.")