Fail to deploy finetuned GPT-3 model

Joe Hu 6 Microsoft Employee

Hi there,

I successfully finetuned the GPT-3 models using Azure OpenAI, but it fails all the time when I try to deploy the finetuned model. And the deployment took about 24 hours before it fails.

Lauri Lehman 136 Reputation points

2023-03-16T12:56:08.2733333+00:00

According to the documentation, "davinci" models are not available for fine-tuning. Not sure if that is the issue in this case, because apparently you were able to train the fine-tuning model anyway.

I have the same problem with a "curie" fine-tuned model. The training is successful but the deployment fails every time. I have tried to deploy with both OpenAI Studio UI and the Python SDK, none of them are working.

It is impossible for developers to debug this issue, because there is absolutely no error log given for the deployment. Hopefully Microsoft will address this issue soon.

EDIT: I have contacted Microsoft support about this issue. Seems that it was just a temporary platform issue. I am now able to fine-tune models.

1 answer

Berry, Ivan 1 Reputation point

2023-01-08T16:59:09.187+00:00

Hi @Joe Hu

I have the same issue! Did you manage to resolve this problem?

Many Thanks,
Ivan
Please sign in to rate this answer.

0 comments No comments
Sign in to comment