Transformers Explained: Sampling LLM Output
Автор: McData
Загружено: 2026-01-08
Просмотров: 15
LLM output a probability distribution over a vocabulary. This video presents the different sampling techniques that sample from this distribution. The video also discuss how the probability distribution is generated using the output from the transformer block.
#ai #llm #transformers #mathematics #statistics
---------------CHAPTER-------------
00:00 Introduction
01:39 Why a probability distribution?
04:24 Why we need a linear layer?
07:21 Softmax
11:12 Greedy sampling
13:18 Multinomial sampling
15:14 top-k approach
17:03 top-p approach
19:46 Temperature scaling
22:00 Autoregression
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: