Why are the responses in the chat playground on Azure Open AI studio different from the responses in the deployed web app?

Asked Aug 21 '23 at 13:45

Active Aug 21 '23 at 14:30

Viewed 41 times

I am trying to customize Azure Open AI assistant to do specific tasks. I added system message and user/assistant examples to the chat playground of Azure Open AI studio. When I test the model on the chat session in the chat playground, the responses are perfect, but when I deploy it on a web app the responses are awful and different. I use the deployment option on the chat playground page on Azure itself, so the exact same model should be deployed on the web app. What am I missing?

I tried with different temperature values and other parameters. I restarted the web app a few times as well, but it's still not working.

My problem is that the responses on the testing session change due to the different parameters or different system messages, but the responses on the web app are so much different from the responses on the chat playground.

edited Aug 21 '23 at 14:30

asked Aug 21 '23 at 13:45

Paria

What's the value of `temperature` parameter? That could be one reason why the responses are different. – Gaurav Mantri Aug 21 '23 at 14:08
The temperature is 0.7. I tried to reduce it, but still, I have perfect responses on the chat playground and awful responses on the web app. They are so much different – Paria Aug 21 '23 at 14:11
Try with `0` in both the places. Also it would help if you edit your question and provide more details. – Gaurav Mantri Aug 21 '23 at 14:15
I tried 0, it's still not working the same – Paria Aug 21 '23 at 14:22
Like I said, please edit your question and provide more details. – Gaurav Mantri Aug 21 '23 at 14:23

Why are the responses in the chat playground on Azure Open AI studio different from the responses in the deployed web app?

0 Answers0