4
$\begingroup$

Would it be necessary to retrain the entire model if we were to perform fine-tuning?

Let's say we somehow got the GPT-3 model from OpenAI (I know GPT-3 is closed source).

Would anyone with access to a couple of RTX 3080 GPUs be able to fine tune it if they got the GPT-3 model weights?

Or would it need infrastructure like the big companies?

$\endgroup$

2 Answers 2

9
$\begingroup$

No, you don't need to retrain the entire model. Fine-tuning refers to taking the weights trained in the general model and then continuing training a bit using your specific data. Using this approach, typically the only things you need to fully train are the models performing the downstream task from the model creating the representation of the data, often just a handful of densely connected layers to perform e.g. classification, which are orders of magnitude less expensive to train than the representation model.

$\endgroup$
3
  • $\begingroup$ What kind of GPU would be needed to fine-tune GPT-3 models and can anyone do that? $\endgroup$
    – Exploring
    Nov 17, 2022 at 19:42
  • 2
    $\begingroup$ Generally with models of this size it's less about what kind of GPU and more about how many you can get your hands on. But yes if they have the model, training data, and compute, anyone can fine tune any model. $\endgroup$
    – Andy
    Nov 17, 2022 at 19:50
  • $\begingroup$ you are really cool. Thanks @Andy. $\endgroup$
    – Exploring
    Nov 17, 2022 at 19:51
5
$\begingroup$

Yes. If open-sourced, we will be able to customize the model to our requirements. This is one of the most important modelling techniques called Transfer Learning

A pre-trained model, such as GPT-3, essentially takes care of massive amounts of hard-work for the developers: It teaches the model to do basic understanding of the problem and provide solutions in generic format.

With transfer learning, given that the pre-trained models can generate basic solutions, we can transfer the learning to another context.

This is the reason why GPT-3 has applications everywhere: Building chatbots, Q&A models, context inference, etc.

Simple Example

GPT-3 can understand the paragraph and generate generic summary. A custom GPT-3 would be able to understand medical journals and generate a subject relevant summary.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.