Update app.py

app.py

@@ -149,8 +149,34 @@ def safe_calculator(expression: str) -> str:
         print(f"Error during calculation for '{expression}': {e}")
         return f"Error calculating '{expression}': Invalid expression or calculation error ({e})."
 
-# --- Basic Agent Definition ---
-# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
+# --- Custom Agent to Force Correct Behavior ---
+class MyCustomHfAgent(HfAgent):
+    """
+    A custom agent that inherits from HfAgent to override the text generation method,
+    forcing it to use the local pipeline instead of attempting a faulty web request.
+    """
+    def generate_one(self, prompt: str, stop: list):
+        print("--- INSIDE CUSTOM HfAgent's generate_one method ---")
+
+        # This is the crucial check. We're logging what the agent thinks its state is.
+        is_pipeline = self.llm.is_hf_pipeline if hasattr(self.llm, "is_hf_pipeline") else "LLM has no is_hf_pipeline attr"
+        print(f"--> self.llm.is_hf_pipeline is: {is_pipeline}")
+
+        # Regardless of what the agent thinks, we KNOW we gave it a pipeline.
+        # So, we will force it to execute the code path for local pipelines.
+        print("--> Forcing execution of the local pipeline path...")
+
+        try:
+            # This is the code from the 'if self.llm.is_hf_pipeline:' block in the original Agent.generate_one method
+            processed_prompt = self.llm.processor.process_prompt(prompt, **self.tokenizer_kwargs)
+            model_outputs = self.llm.pipeline(processed_prompt, stop_sequence=stop, **self.generate_kwargs)
+            return self.llm.processor.process_outputs(model_outputs, stop_sequence=stop)
+        except Exception as e:
+            print(f"--- ERROR during forced pipeline execution: {e} ---")
+            traceback.print_exc()
+            # If this fails, we return an error string.
+            return f"Error during custom pipeline execution: {e}"
+
 # --- Agent Definition using HfAgent ---
 class HfAgentWrapper:
     def __init__(self):