How LLMs Actually Work

Рет қаралды 1,540

6 ай бұрын

In this video, I explain how language models generate text, why most of the process is actually deterministic (not random), and how you can shape the probability when selecting a next token from LLMs using parameters like temperature and top p.
I cover temperature in-depth and demonstrate with a spreadsheet how different values change the probabilities.
Topics:
00:10 Tokens & Why They Matter
03:27 Special Tokens
04:35 The Inference Loop
07:26 Random or Not?
08:11 Deep Dive into Temperature
14:19 Tips for Setting Temperature
16:11 Top P
If you'd like to play with the temperature calculator spreadsheet, you can make a copy of it here (read-only):
docs.google.com/spreadsheets/...
To learn more about Entry Point AI, visit our website at www.entrypointai.com
Like this video? Hit that subscribe button ⭐️
PS. PyTorch, TensorFlow, and underlying GPU libraries can introduce randomness that is tricky to pin down - these are implementation details that will change and presumably get easier over time.It doesn't change the fundamental nature of LLMs.

Пікірлер: 10

@RoryWilliamson 2 ай бұрын

This is one of the best explanations i've seen, thanks

@Titan-axle 6 ай бұрын

Thank you for this great explanation. It's given me a better clarity to temperature, top-p and their involvement in the loop. Really looking forward to the next one 😃

@ai-automation-agents 5 ай бұрын

Helpful and interesting - nicely done.🎓

@AnjelaPetkova 6 ай бұрын

Thanks 🙏 super helpful video

@pratik9882 6 ай бұрын

Informative 👍🏽

@Ak_Seeker 2 ай бұрын

Thank you

@NY1man1NY 26 күн бұрын

Watching the first 10 minutes of this video, explains how AI works. This should be showed in schools.

@DrJaneLuciferian 6 ай бұрын

Thank you. Very well explained :^) Do you think you'll touch on non-LLM topics in the future: LSTM, GRUs, SNNs, etc?

@EntryPointAI 6 ай бұрын

I expect to get to diffusion image models and multi-modal at some point, but likely won't backtrack to cover any architectures that aren't achieving new state-of-the-art.