Update app.py

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
-from huggingface_hub.inference_api import InferenceApiException
+# Import the correct exception class
+from huggingface_hub.utils import HfHubHTTPError
 import os
 
 """
@@ -13,17 +14,14 @@ Alternatively, pass your token directly: InferenceClient(token="hf_YOUR_TOKEN")
 # Initialize the Inference Client
 # It will try to use HUGGING_FACE_HUB_TOKEN environment variable or cached login
 try:
+    # You might need to provide a token if you haven't logged in via CLI
+    # token = os.getenv("HUGGING_FACE_HUB_TOKEN")
+    # client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=token)
     client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 except Exception as e:
     print(f"Error initializing InferenceClient: {e}")
-    # Optionally, provide a default token if needed and available
-    # token = os.getenv("HUGGING_FACE_HUB_TOKEN")
-    # if token:
-    #     client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=token)
-    # else:
-    #     raise ValueError("Could not initialize InferenceClient. Ensure you are logged in or provide a token.") from e
-    # For now, let's just raise it if initialization fails fundamentally
-    raise
+    raise ValueError("Could not initialize InferenceClient. Ensure you are logged in or provide a token.") from e
+
 
 def respond(
     message: str,
@@ -76,9 +74,20 @@ def respond(
                 response += token
                 yield response # Yield the accumulated response so far
 
-    except InferenceApiException as e:
-        print(f"Inference API Error: {e}")
-        yield f"Sorry, I encountered an error: {e}"
+    # Catch HTTP errors from the Hugging Face Hub API
+    except HfHubHTTPError as e:
+        error_message = f"Inference API Error: {e}"
+        # Try to get more details from the response if available
+        if e.response:
+            try:
+                details = e.response.json()
+                error_message += f"\nDetails: {details.get('error', 'N/A')}"
+            except Exception: # Catch potential JSON decoding errors
+                pass # Keep the original error message
+        print(error_message)
+        yield f"Sorry, I encountered an error communicating with the model service: {e}" # Display a user-friendly message
+
+    # Catch other potential errors
     except Exception as e:
         print(f"An unexpected error occurred: {e}")
         yield f"Sorry, an unexpected error occurred: {e}"
@@ -120,4 +129,6 @@ demo = gr.ChatInterface(
 
 
 if __name__ == "__main__":
+    # Ensure huggingface_hub library is up-to-date: pip install --upgrade huggingface_hub
+    print("Launching Gradio Interface...")
     demo.launch()