How to fine tune a model using GPT-3 models?

API
1

Is there possible to fine tune a model using GPT-3 models?

We want to train and fine tune using one of the below models,

  1. text-davinci-003
  2. text-curie-001

We have tried to fine tune using this GPT-3 models, but it failed.

image
image959×152 5.56 KB

Kindly assist me to go forward on this.

Thanks in advance
Aravinth k

2

You can only use the following model names when you fine-tune:

  • ada,
  • babbage,
  • curie,
  • davinci, or
  • a fine-tuned model created by your organization

See this example:

irb(main):093:1* response = client.finetunes.create(
irb(main):094:2*         parameters: {
irb(main):095:2*         training_file: file_id,
irb(main):096:2*         model: "text-davinci-003"
irb(main):097:0>     })

=> 
{"error"=>                                       
  {"message"=>                                   
    "Invalid base model: text-davinci-003 (model must be one of ada, babbage, curie, davinci) or a fine-tuned model created by your organization: org-8NVL8Hblahblahblahblah7",           
   "type"=>"invalid_request_error",
   "param"=>nil,

So, if you parse / view the error message, you will see the message above, explaining to you the permitted model names. So in your case @aravinth.kumar , you need to change to (whatever code works for you, of course)

response = client.finetunes.create(
     parameters: {
          training_file: file_id,
          model: "davinci"
     })

I can confirm the model name “davinci” works, but in my case, it’s been in pending state for a long time:

"status"=>"pending",

Maybe you will have faster results @aravinth.kumar ? I’m on the cheap developer $18 work of credits plan :slight_smile:

HTH