H all,
I want to connect my Azure Data (from a Cognitive Search resource) using PVA Conversation Booster and I want to ask if anyone has experienced using the Model Data configuration. Any help or advice is much appreciated.
Thank you!
Hi @jthacker,
As it turns out to be more an Azure OpenAI question than a PVA one, can you raise this on the Microsoft Q&A?
Ask a question - Microsoft Q&A
Henry
Thank you for the response. I really dont think that they are account numbers or other codes as Im familiar with the documents. I tried it a few times, entering random letters and numbers but response from bot are either a copy of the previous response or a more detailed explanation of the previous prompt. It's really weird.
Agree with Henry on this last point; if a data source comes back with a valid answer, we have no other signal to indicate to us that the input was gibberish. It's even possible for some data sources that that may have been an account number or some other code that was intentional to hit a specific doc.
Thats interesting @HenryJammes . I tried to copy your P control but it doesnt fix the issue. Interestingly after a few tries, it happens more in PVA.
Do you think it has something to do with the way I loaded the data?
How did you load yours?
Additionally, may I also ask what is the meaning of this error? I'm getting this often
Thanks!
If it happens in the Azure OpenAI studio, I'm guessing it's fair to expect the same to happen in PVA.
How have you setup the temperature or top P that control how deterministic or random the response is?
I get the model to respond that they can't find an answer:
Yes @HenryJammes the same thing happens. If it helps, I added my data from azure search service.
I tried adjusting the parameters but performance did not improve.
@jthacker do you also experience the same in the Azure OpenAI studio?
Thank you @jthacker - I see. Interesting. Let me flag @Jameslew_MSFT
Sorry about that.
Im stress testing the bot from azure openai and deployed in PVA. And after a response from a bot I tried prompting it with random letters and/or texts see photo below. How can we fix this issue?
Thank you @HenryJammes
I'm not sure I follow, do you have a screenshot?
Unrecognized user utterances trigger the UnrecognizedIntent trigger.
It first goes through the Conversational Boosting system topic.
If no answer is returned by Generative Answers, it continues to the Fallback topic.
Henry