How to delete a fine-tune model (via api)?

General API discussion
4

Delete endpoint is working fine but engines are still appearing on GET /v1/fine-tunes endpoint with status succeeded. At this point, it is impossible to understand which engines are available.

If you call delete endpoint again you will get a 404. I think delete endpoint is working but there is a bug at fine-tunes endpoint or I am missing something.

5

Hi!
I had the same question but no, its ok like it is. Fine tunes is different from engines. Think like this: a fine-tune is a process that runs and finishes, producing a new engine as a result. Even if you delete the engine, the fine-tune record will remain, which it should. To know which engines are available you should use the engines end-point, not the fine-tunes endpoint.

6

Hi Nunodonato, I’m look for this, do you mind telling me how to gain access to that page?

7

there’s no page, you can only access that via API

8

Hello, When I use “openai.Model.delete(model_name)”, I found the model was deleted successful, but when I use “openai.FuneTune.list()” to check, the model still there.
Is there a way to delete all resource of the model in the result of “openai.FuneTune.list()”?

9

I think it takes time for all the user data to sync.

How long has it been since you deleted ?

:slight_smile:

10

It is about 24 hours. There is still no change.

11

Please use the retrieve API method using the fine-tune ID and post the results back here.

:slight_smile:

12

you are confusing things. Fine-tune process is one thing, models are another. You are deleting a fine-tuned model, but the fine-tuned process is still on the records. Its kind of historical data of what was done, with which file, etc.

13

The result is:

Finetune retrieve status: failed
Checking other finetune jobs in the subscription.
[‘curie:ft-personal-2023-03-01-07-26-01’, ‘babbage:ft-personal:grg-model-001-2023-03-01-12-17-24’, None, None, None, None, None]

14

So I can not delete these historical data?

15

I don’t think so, at least the docs don’t list an endpoint for that.

16

I have no an idea.
And I found all my finetune jobs were always failed, confusing.

17

This often happens when the JSONL training file is not validated and incorrectly JSONL formatted.

HTH

:slight_smile:

18

Thanks, you are right, now I get success:
Status: pending
Status: pending
Status: pending
Status: pending
Status: pending
Status: running
Status: succeeded

1 Like
19

Welcome!

Send Bitcoin (Joking)

:slight_smile:

20

Thanks a lot.
One more question, is there a rule of making training data? can you show me a sample?

21

Sorry, do you mean validation rules?

Maybe check out this very detailed lab-style tutorial:

22

Thank you very much!
But I confused, why I still got fine-tune failure when I only use your sample(complete copy):
{“prompt”:“What is your favorite color? ++++”, “completion”:" My super favorite color is blue. ####"}
But these samples can work:
{“prompt”: “When I go to the store, I want an”, “completion”: “apple.”}
{“prompt”: “When I go to work, I want a”, “completion”: “coffee.”}
{“prompt”: “When I go home, I want a”, “completion”: “soda.”}

23

Are you going to publish this tool publicly?
Me want plz. :slight_smile: