← Back to the collection
Adaptive depth ·
Think-at-Hard
Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models
Allocates extra latent iterations to selected tokens, using a decider and duo-causal attention.
Inside the method
Token → Decider → Latent iterations
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- Selective token iteration
- KV / state strategy
- Duo-causal attention
Reading note
Selective additional iteration and ordinary layer skipping have different computational semantics.
Sources checked 2026-09-15. This catalog does not imply independent reproduction.
Cite this work
@misc{thinkathard2025,
title = {Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models},
author = {Tianyu Fu and Yichen You and Zekai Chen and Guohao Dai and Huazhong Yang and Yu Wang},
year = {2025},
eprint = {2511.08577},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2511.08577}
}