How to fine tune a model with already fine tuned model name?

API
1

We had uploaded our dataset file to OpenAI and fine tuned a model with davinci base model name. We got the fine_tuned_model name as davinci:ft-organization-name-2022-12-22-11-29-14.

Now we are planning to train and fine tune another dataset file. We want to fine tune the model with already fine tuned model (davinci:ft-organization-name-2022-12-22-11-29-14).

If I try to fine tune with already fine tuned model, I’m facing the below issue. And my uploaded file size is 14.8MB.

image
image972×220 8.19 KB

Could you please assist me how to achieve this in OpenAI?

Thanks in advance!y

2

Welcome to the community!

From the error it looks like you have a hard billing limit and fine-tuning the file will take you over that limit, so it’s failing …

Did you check your billing settings?

Good luck!

1 Like
3

Thanks for your response Paul!

Initially I have fine tuned with davinci model with a small data file (10KB) and it gets fine tuned with succeeded status. Also, I got fine_tuned_model name as davinci:ft-org-name-2022-12-22-11-29-14.

Response:

{
    "data": [
        {
            "created_at": 1671705275,
            "fine_tuned_model": "davinci:ft-org-name-2022-12-22-11-29-14",
            "hyperparams": {
                "batch_size": 1,
                "learning_rate_multiplier": 0.1,
                "n_epochs": 4,
                "prompt_loss_weight": 0.01
            },
            "id": "ft-e0IrBwFvFDMMZJPl2FCxBTBW",
            "model": "davinci",
            "object": "fine-tune",
            "organization_id": "org-etoq30QMSdIOT2QfJeLPIXrE",
            "result_files": [
                {
                    "bytes": 6424,
                    "created_at": 1671708555,
                    "filename": "compiled_results.csv",
                    "id": "file-JitfsAzLazbG7txiyJI5KgD0",
                    "object": "file",
                    "purpose": "fine-tune-results",
                    "status": "processed",
                    "status_details": null
                }
            ],
            "status": "succeeded",
            "training_files": [
                {
                    "bytes": 46150,
                    "created_at": 1671693460,
                    "filename": "blazor-speeddial.jsonl",
                    "id": "file-MSjf7HkykF19erFRSQMAZOaV",
                    "object": "file",
                    "purpose": "fine-tune",
                    "status": "processed",
                    "status_details": null
                }
            ],
            "updated_at": 1671708555,
            "validation_files": []
        }
    ],
    "object": "list"
}

Then, I tried to fine tune with another data file (size 9KB) with previously fine tuned model name (i.e.,) davinci:ft-org-name-2022-12-22-11-29-14 and i got the below response as failed status.

{
    "data": [
        {
            "created_at": 1672213412,
            "fine_tuned_model": null,
            "hyperparams": {
                "batch_size": null,
                "learning_rate_multiplier": null,
                "n_epochs": 4,
                "prompt_loss_weight": 0.01
            },
            "id": "ft-daUgFK1lmXgb2LeLWQncI4se",
            "model": "davinci:ft-org-name-2022-12-22-11-29-14",
            "object": "fine-tune",
            "organization_id": "org-etoq30QMSdIOT2QfJeLPIXrE",
            "result_files": [],
            "status": "failed",
            "training_files": [
                {
                    "bytes": 15228579,
                    "created_at": 1672213270,
                    "filename": "finetuning_prepared.jsonl",
                    "id": "file-2oviTsUgI4VYd5iR5QP2J0FE",
                    "object": "file",
                    "purpose": "fine-tune",
                    "status": "processed",
                    "status_details": null
                }
            ],
            "updated_at": 1672213501,
            "validation_files": []
        }
    ],
    "object": "list"
}

How could I fine tune with davinci:ft-org-name-2022-12-22-11-29-14 model again and again?

4

I’m interested in this question if anyone can help.

5

Hey, you can indeed fine-tune a fine-tuned model, see here: OpenAI API and please let me know if this doesn’t resolve the issue.

2 Likes
6

Can u pls provide the model name which u r using while retraining ur model?

7

Can any one plz guide me I am fine tuning the model with custom dataset manually via coding.
Is there any option on the official site of openai to re fine tune the already tunned model
note: I have first time fine tunned the model on their offical site but it is not giving accurate results now i want to refine tune the model so that i give better results, how i can?
https://platform.openai.com/finetune
plz anyone guide me Thankyou
@logankilpatrick

8

you can do it manually using job id and fine tune model id

Assuming you already have training_file_id and validation_file_id defined

Step 1: Start the fine-tuning job

try:
response = openai.FineTuningJob.create(
training_file=training_file_id,
validation_file=validation_file_id,
model=“ft:gpt-3.5-turbo-0125:smarty::97zV0KRm”, # Your model ID here
)

job_id = response["id"]
print("Fine-tuning job ID:", job_id)
print("Fine-tuning job response:", response)

except Exception as e:
print(f"An error occurred while starting fine-tuning: {e}")

Step 2: Check if the fine-tuning job was successful and get the new model ID

existing_model_id = response.get(“model”)

if existing_model_id:

print("Existing model ID:", existing_model_id)

try:
    # Start fine-tuning with the existing model
    response = openai.FineTuningJob.create(
        training_file=training_file_id,
        validation_file=validation_file_id,
        model=existing_model_id, # Use the existing model ID
    )

    new_model_id = response["id"]
    print(f"Fine-tuning completed! New model ID: {new_model_id}")
except Exception as e:
    print(f"An error occurred while fine-tuning with existing model: {e}")

else:
print(“No existing model ID found.”)

===========================================================================