Finetuning not working, pending status

API
1

Hello, I’ve been for a while finetuning a model and I’m guessing why is it not queuing and keeping the finetuning jobs as “pending”

Is there any outage?
I’ve tried with the 4 models.

{
“created_at”: 1676471215,
“events”: [
{
“created_at”: 1676471215,
“level”: “info”,
“message”: “Created fine-tune: ft-AnsQWvCK9hwAxGuDaHs5ITU6”,
“object”: “fine-tune-event”
}
],
“fine_tuned_model”: null,
“hyperparams”: {
“batch_size”: null,
“learning_rate_multiplier”: null,
“n_epochs”: 4,
“prompt_loss_weight”: 0.01
},
“id”: “ft-AnsQWvCK9hwAxGuDaHs5ITU6”,
“model”: “ada”,
“object”: “fine-tune”,
“organization_id”: “org-W35aFp2vTFBWfmzUv1UMJh3R”,
“result_files”: ,
“status”: “pending”,
“training_files”: [
{
“bytes”: 28674,
“created_at”: 1676471215,
“filename”: “./completions_prepared.jsonl”,
“id”: “file-2X4FWKIIA1tBZDQnhgMb22rf”,
“object”: “file”,
“purpose”: “fine-tune”,
“status”: “processed”,
“status_details”: null
}
],
“updated_at”: 1676471215,
“validation_files”:
}

2

I’m not sure sharing your fine tune model IDs is wise. I’m assuming these are just tests and you won’t use them for serious work, but just in case, I suggest you don’t provide them moving forward.

1 Like
3

This is normal, fine-tuning can take many hours, and the status will remain “pending” until it runs, often in the “pending” state for a very long time.

Actually, while its not important to share fine tune ids, I don’t think there is much of a cybersecurity risk in doing so.

Even if someone has it, it’s unlikely they can run the model unless they break into your account, and if they break into your account, it’s easy to get the fine tune model ids anyway.

Then, if someone was a kind of super hacker who could figure out a way to jailbreak the user authentication and run your model, models are ran by model name not id and the pending status models do not have a model name yet.

Then, even if someone could get your model name and hack the system to run your model, so what?

The terms and service of OpenAI is to not fine-tune with confidential or private data, and so your models should not contain any private data according to OpenAI policy.

So, while I understand your recommendation @humphree, as a cybersecurity person with decades of experience in this field, I fail to see any “practical risk” for a user in posting ids of pending fine-tune models, to be honest.

Don’t get me wrong, I am not encouraging people to post the ids and names of their fine tuned models; I’m simply pointing out that on the “risk curve”, I don’t see that @fcasals has done anything a reasonable cybersecurity professional would consider to have “significant risk”.

1 Like
4

Check you are on a paid plan. This is a possible (non obvious) reason why a file won’t train

1 Like
5

Thank you for the security breach pointing @humphree, I will bear that in mind for future cases!
ruby_coder also grateful for the information

@raymonddavey I will set the paid plan in order to prevent further waitings. I’m just introducing myself into this field and aiming to create multiple models to test them out, so it will be better!!

Tank you all :))

6

Hi everyone,

I’ve got 2 fine tune jobs that have not budged from pending status for almost 24 hours.

I’m happy to wait, as long as they’ll get done eventually, but I’m worrying something is wrong.

The jobs have not changed to “queued” status. And I’m not seeing any indication from my usage dashboard that it’s hit my account yet. Granted, this may not show unless the job is actually underway:

image
image821×238 10.2 KB

This is time sensitive (I need something by end of month), so I’m nervous just waiting around. @fcasals Have your jobs been completed yet? How long did it take?

@ruby_coder Do you still suspect this is ‘normal’ waiting time?

1 Like
7

Mi @mike13. Were you able to solve it? I’m in the same situation

8

mike13? Last seen March 31.

I would follow the prerequisites mentioned: you have an API payment method entered, you’ve budgeted enough hard limit for the job to complete, etc.