GPT-3 Alternative - OPT-175B Hugging Face Language Model Tutorial

Meta AI’s recently shared Open Pretrained Transformer (OPT-175B), a language model with 175 billion parameters trained on publicly available data sets.

For the first time for a language technology system of this size, the release includes both the pretrained models and the code needed to train and use them.

This video contains three parts:

00:00 Quick Intro about OPT-175B
03:51 OPT-175B Live Demo with Alpa
08:07 OPT1.3B Text Generation Hands-on Coding using Hugging Face Transformers

[mystwalker]

Being Meta, I'm scared Meta might become the new Google someday, it may not be but the possibility is there.

[knowledgelover]

Davinci has a 4000 token limit. Does OPT state what their token limit is?

[mvrdara]

can you try china's trillion paramaeter language model?

[ScriptureFirst]

Hmm 🤔 We should have a contest for the most efficient Wikipedia “fact check” of an output using the model.

[neilsal]

All in all can we say, there aren't good results with this technique, right ?

[alvarochuiso]

it throws me an error: NameError: name 'generator' is not defined

[kutupbear]

title says 175b but you using 1.3b model.