To reproduce, on the older openai==0.25.0
openai api fine_tunes.results -i <job_id_here>
You'll get the output of the actual CSV file
E.g.
step,elapsed_tokens,elapsed_examples,training_loss,training_sequence_accuracy,training_token_accuracy
1,22048,32,0.011019098009307718,0.0,0.0
1,32288,32,0.007613447834811513,0.0,0.0
On the latest openai==0.26.2
You get instead
{
"object": "file",
"id": "file-66NnpqAXzw3In27fROPIU03P",
"purpose": "fine-tune-results",
"filename": "compiled_results.csv",
"bytes": 386756,
"created_at": 1674504271,
"status": "processed",
"status_details": null
}
This is probably due to a change in behavior in download of /api_resources/file.py. It probably affects everyone who depends on that download method, so i'm guessing wandb users who are syncing their runs too?
See https://github.com/openai/openai-python/pull/146/files#r1084884786
Probably related to #185, #183
To reproduce, on the older
openai==0.25.0openai api fine_tunes.results -i <job_id_here>You'll get the output of the actual CSV file
E.g.
On the latest
openai==0.26.2You get instead
This is probably due to a change in behavior in
downloadof /api_resources/file.py. It probably affects everyone who depends on that download method, so i'm guessing wandb users who are syncing their runs too?See https://github.com/openai/openai-python/pull/146/files#r1084884786
Probably related to #185, #183