fixes

app.py

@@ -31,12 +31,13 @@ def respond(
         cleaned_output = response.text.strip().rstrip('%')
         output = json.loads(cleaned_output)
 
-    if output['error']['code'] == 'concurrency_limit_exceeded':
-        return "Error: Concurrency limit exceeded. Try submit your request again in a few seconds."
-    elif 'error' in output:
-        return f"Error: {output['error']['message']}. Refresh the page and try again later, post the error on OLAT forum if your issue persists."
-    else:
+    if isinstance(output, str):
         return output
+    elif 'error' in output:
+        if output['error']['code'] == 'concurrency_limit_exceeded':
+            return "Error: Concurrency limit exceeded. Try submit your request again in a few seconds."
+        else:
+            return f"Error: {output['error']['message']}. Refresh the page and try again later, post the error on OLAT forum if your issue persists."
 
 
 demo = gr.Interface(