GPT-3 for custom dataset classification with custom labels

API
1

Hello - good morning.

I am a data scientist at a lab and recently designed a classifier using bert model. The model uses text to predict business labels for data. I want to understand if we can use GPT-3 to predict our business labels?
My current dataset is not very large, its about 40k text records with 450-500 words per record with 85 unique labels, and it is in CSV format. I am using python for this work. Also can we deploy a GPT-3 model on AWS? I think we can, but if someone can assist me - I would greatly appreciate it.

Thanks

2

That’s way larger than you need for GPT3. Since you have 85 classes, I would suspect you’d need no more than 5 to 10 samples per class, so maybe 850 samples to finetune GPT3.

Sure, if you’ve got $7M - otherwise just use the OpenAI API :stuck_out_tongue:

Just start here, it’s super easy https://beta.openai.com/docs/introduction