I tested the same file that has been on the server for months and ran successfully before in testing what models were completely denying images. Copying the training file ID right out of a previous success. A new fine tuning based on gpt-4o to see this issue.
The job failed due to an unsafe training file. Error processing images. Please re-upload the file and try again.
That took three hours.
The file has 10 of the same API calls with one image for vision.
This is a broken machine. Plus not even a “delete” method to get rid of failed jobs, which is the majority in the job listing confirming other’s issues.