Models from successful gpt-3.5-turbo-1106 fine-tune jobs missing

API Bugs
, ,
1

Hello! I recently fine-tuned about 8 models successfully. Below are their job IDs, output models, and status from the OpenAI API:

ftjob-RhdOCPFf3OJ9CrK9p2rFq0bX ft:gpt-3.5-turbo-1106:cornell-university:m60:9tTDBTWY succeeded
ftjob-y7M78LIRybuTQvpD1yhTiA6q ft:gpt-3.5-turbo-1106:cornell-university:m60:9tRUL2DQ succeeded
ftjob-NgPkMuxpPTqGADkfzLSXsjeR ft:gpt-3.5-turbo-1106:cornell-university:m60-prompt2:9tPuDaM7 succeeded
ftjob-WxvG9VS7hoNqvzHIHgBK1JqS ft:gpt-3.5-turbo-1106:cornell-university:m60:9tOOpn2x succeeded
ftjob-IllTAUKk6Gy6fiMxcJYKKreQ ft:gpt-3.5-turbo-1106:cornell-university:m60:9tLyCdcE succeeded
ftjob-4HSJxJI64K5GoYwxI1yQGPDQ ft:gpt-3.5-turbo-1106:cornell-university:m120:9tLY0iBc succeeded
ftjob-yZzoUy9TwNYUT5VUohtmsNRo ft:gpt-3.5-turbo-1106:cornell-university:m90:9tLaSAqB succeeded
ftjob-zGFFtp8XmDb12TVWAC5m5W6g ft:gpt-3.5-turbo-1106:cornell-university:m120:9tKXfsqc succeeded
ftjob-guFFePs2WswH0rWOaVbByVCS ft:gpt-3.5-turbo-1106:cornell-university:m90:9tKr4J4l succeeded
ftjob-TQcFbJTO8zNGKm001pyjhp97 ft:gpt-3.5-turbo-1106:cornell-university::9tKWxjxc succeeded
ftjob-IQn2KYiDELDFvzF7mtICtpU6 ft:gpt-3.5-turbo-1106:cornell-university::9tHa3bwN succeeded

When I proceed to do inference, I can only use the first 2 models! The error I receive looks like the following:

NotFoundError: Error code: 404 - {'error': {'message': 'The model `ft:gpt-3.5-turbo-1106:cornell-university:m60:9tLyCdcE` does not exist or you do not have access to it.', 'type': 'invalid_request_error', 'param': None, 'code': 'model_not_found'}}

Has anyone faced the same issue and found a workaround? OpenAI support has been slow to get back to me, and I am hoping I didn’t waste a couple of hundred dollars on all of these fine-tunes.

Update: A workaround that seems to work is to create a new project under the same organization, a new API key, and submit a fine-tuning job there. The first fine-tuning job I submitted in the new project completed training and I am able to use it for inference. I’ll report back on how subsequent fine-tuned models fare.

Emaad

2

I have the same issue with fine tunes of GPT-4o mini-0718.

The job completed fine, no errors, no issues, but when I go to retrieve the model I get told it doesn’t exist.

I ran my dataset through the moderator API and yanked out anything that might be considered objectionable (mine is mostly a bunch of cooking sites and car forums I’m building a classifier for) and am running the job again. I will update this if that fixes it.

I would love any kind of clarity from OpenAI as to why this is happening as opposed to continually throwing training runs down the well here.

1 Like