Verifiers - Beam Search | Reasoning LLMs from Scratch
Автор: Vizuara
Загружено: Дата премьеры: 13 апр. 2025 г.
Просмотров: 1 173 просмотра
In this lecture, we will continue with exploring Inference-Time Compute Scaling for building reasoning models.
We will look at the method of “Search against Verifiers” and discuss about Outcome Reward Models and Process Reward Models. Then, we will explore the logic behind the following types of verifiers:
(1) Majority Voting
(2) Best-of-N Sampling
(3) Beam Search
Here is the Colab File shown in the lecture to build a beam search based verifier with a reward model:
https://colab.research.google.com/dri...

Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: