Hey everyone! In this video we’ll be fine-tuning OpenAI’s amazing GPT-3 model to do something really interesting: come up with new science fiction TV show ideas.
That’s right. We’ll be generating show ideas and synopses for one of my favorite shows of all time: Dr. Who. I’m going to take you from gathering the data to preprocessing it to training it and visualizing every step in Weights & Biases.
As a bonus, towards the end I’ll also show you how we can use our custom-trained model to generate new Doctor Who episode synopses on OpenAI’s Playground.
If you want to get more information about the project or recreate it or just learn how to fine tune GPT-3, check out the links below!
—
Links
Google Colab notebook to fine-tune GPT-3 with OpenAI and W&B: http://wandb.me/gpt-3-video-colab
Blogpost I wrote about fine-tuning GPT-3: http://wandb.me/dr-who-gpt-3
OpenAI fine-tuning docs: https://beta.openai.com/docs/guides/fine-tuning
OpenAI X W&B docs: https://docs.wandb.ai/guides/integrations/other/openai
—
Timestamps
00:00 Intro
01:17 Misconceptions about training GPT-3
04:19 Colab notebook for fine-tuning
07:16 Exploring GPT-3 training data with W&B Tables
11:19 Preprocessing training data with OpenAI CLI tool
13:07 Starting fine-tuning GPT-3
14:46 Syncing OpenAI fine-tunes to Weights & Biases
19:06 Visualizing GPT-3 model predictions on validation data as W&B Table
21:52 Using OpenAI Playground to generate new Doctor Who synopses
23:38 Outro
—
Follow Ivan:
Twitter: https://twitter.com/Ivangrov
YouTube: https://www.youtube.com/c/IvanGoncharovAI
Get started with W&B: http://wandb.me/intro
Follow us:
Twitter: http://twitter.com/weights_biases
Linkedin: https://www.linkedin.com/company/weights-biases
Thanks for watching! If you have any questions, please don’t hesitate to ask! If you have any suggestions, please don’t hesitate either! We love hearing from the community and look forward to seeing you join our community!
Doctor Who and all affiliated concepts are owned by the BBC. The Daleks are owned by Terry Nation.
