Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

Описание: Paper: https://arxiv.org/abs/2502.05171
Code! https://github.com/seal-rg/recurrent-...

Notes: https://drive.google.com/file/d/10WQ_...
https://drive.google.com/file/d/1eglt...

00:00 Intro
01:42 Model
09:45 Loss and other model details
14:00 Results
18:20 How to actually use test-time compute and other cool results