DDIM does: deterministic sampling for faster generation with fewer steps

Ever feel like your text generator keeps repeating boring phrases?

Imagine writing a story and your AI keeps suggesting the same tired ending. It's frustrating and stifles creativity.

Top-p sampling helps by choosing words that are more likely to lead to interesting outcomes, instead of just picking the most obvious ones.

Example

Instead of always choosing the ending "and they lived happily ever after," top-p sampling might pick "and they embarked on a thrilling adventure," because it's a less predictable choice.

Remember this

Top-p sampling ensures your story stays fresh and exciting by picking less obvious, more interesting words.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

Large language model

LLMs can generate, summarize, translate, and analyze text in many contexts

Shannon's source coding theorem: you can't compress below entropy

Can you squeeze endless text into fewer bits without losing anything?

ring attention does: distributes long sequences across multiple devices

How can a machine understand and generate human language?

rejection sampling does: samples from target by accepting/rejecting proposals

How do you randomly pick a number following a strange rule?

Stable Diffusion

Stable Diffusion generates images from text descriptions

gradient accumulation simulates larger batch sizes without more memory

Can you train a machine like you do with a computer?

Swipe through 100 ML concepts daily

Open Pocket Polymath