Update app.py

app.py

@@ -32,17 +32,27 @@ def respond(
 
     response = ""
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    try:
+        for message in client.chat_completion(
+            messages,
+            max_tokens=max_tokens,
+            stream=True,
+            temperature=temperature,
+            top_p=top_p,
+        ):
+            payload = message.choices[0].delta.content
+            # Attempt to decode the payload, but handle JSON errors gracefully
+            try:
+                token = json.loads(payload.lstrip("data:").rstrip("/n"))["content"]
+                response += token
+                yield response
+            except json.JSONDecodeError as e:
+                print(f"JSON decoding error: {e} - payload: {payload}")
+                continue  # Skip malformed payloads and continue
+    except Exception as e:
+        print(f"Error in chat completion: {e}")
+        yield "Error processing response from the model."
 
-        response += token
-        yield response
 
 """
 For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface