Fine-tune failed. Returning to queue.. perplexing?

I guess the Finetuning API is fairly new, so gave it whirl. I performed the following steps

  1. Prep’d a jsonl that was ~63MB. I’m following the suggestions to essentially do style transfer.

  2. openai tools fine_tunes.prepare_data -f <LOCAL_FILE>.
    This gave a few suggestions which I accepted, it also gave a reusable file ID, nice.

  3. openai api fine_tunes.create -t “jsonlmstar2021_prepared.jsonl” -m curie

This put me in the queue. A day later I made it through the queue but it said failed and put me in the queue again. Will it ever finish? Any stack trace so I can self-correct? Seems like another day’s wait for a try…

[2021-07-18 22:48:40] Fine-tune is in the queue. Queue number: 0
[2021-07-18 22:48:45] Fine-tune started
[2021-07-18 23:30:46] Fine-tune failed. Returning to queue
[2021-07-18 23:30:47] Fine-tune is in the queue. Queue number: 5


Alright i put the email in. Will report back. Sharing is caring.

EDIT 7/30. Got this update

Thanks for your patience! Our engineering team has confirmed that the issue is likely that your completions exceed 2,048 tokens, so you’ll want to pas in the --use_packing parameter when creating the fine-tune, which has the side effect of chunking the examples to fit in the max context length.

Hope this helps