Exceeded allowed number of retries. Fine-tune failed

API
1

openai api fine_tunes.follow -i ft-xx
[2023-06-16 17:34:16] Created fine-tune: ft-xx
[2023-06-16 17:35:52] Fine-tune costs $0.04
[2023-06-16 17:35:52] Fine-tune enqueued. Queue number: 0
[2023-06-16 17:45:54] Fine-tune started
[2023-06-16 17:45:55] Server error. Returning to queue for retry
[2023-06-16 17:46:06] Fine-tune started
[2023-06-16 17:46:06] Server error. Returning to queue for retry
[2023-06-16 17:46:17] Exceeded allowed number of retries. Fine-tune failed. For help, please contact OpenAI and include your fine-tune ID: ft-xx

I have tried train GPT a few times and often get the above error, does anyone know what the problem is
In the data set, I used many questions with the same meaning and the same answer, is that the problem?

1 Like
2

I have the same issue. has anyone figured it out?

3

UPDATING AS NOW FIXED, IN CASE IT HELPS OTHERS, AS THERE ARE MANY THREADS ON THIS ISSUE. SUGGESTION TO OPENAI: if possible, update docs and cookbook with this.

This finally resolved itself but only after ~20 attempts. I didnā€™t do anything other than try running the line of code below that was mentioned in another thread.

openai.FineTune.retrieve(id=ā€œā€)

After I ran it, I tried to continue the ā€˜createā€™ step again and it worked:
openai api fine_tunes.follow -i

Original Message below -----------
(Note that I ā€˜aggregatedā€™ my rows to remove duplicate prompts as mentioned below, and given each prompt multiple categories in the ā€˜completionā€™ field instead - seems thatā€™s a better way to train it, but I still got a ā€˜failedā€™ error message for hours. )

I am also getting this same error and the response time from OpenAI is given as ā€˜1 weekā€™.

Maybe Boris can help (Boris, I followed your recipeā€™s approach - https://github.com/openai/openai-cookbook/blob/main/examples/Fine-tuned_classification.ipynb ).

Interesting that @phantbinh99 says:

"In the data set, I used many questions with the same meaning and the same answer, is that the problem?"

I am also using the same prompt in several places but with a different completion each time. This is because I am trying to assign multiple categories to an article, so there could be 3 categories that are relevant e.g.

Prompt
In the marketing process, companies work to understand consumers, create customer value, and build strong customer relationships. In the final step, companies reap the rewards of creating superior customer value in the form of sales, profits, and long-term customer equity.

Completion
marketing process

Prompt
In the marketing process, companies work to understand consumers, create customer value, and build strong customer relationships. In the final step, companies reap the rewards of creating superior customer value in the form of sales, profits, and long-term customer equity.

Completion
value creation

maybe I cannot have the same prompt a few times in the dataset, even if the completion is different each time. Should I create a single row with the categories aggregated for that prompt? Will this fix the error we are all getting?

Thanks very much
Mim

1 Like