HRM (Hierarchical Reasoning Model) is a hierarchical reasoning model designed for complex reasoning tasks, whose core feature is to perform efficient reasoning by mimicking the human brain's multi-timescale information processing mechanism. The key features of the model are as follows:
- parameter scale: only 27 million parameters, more lightweight than traditional large models
- structural design: A two-module recurrent neural network architecture is used:
- High-level module (responsible for abstraction planning)
- Low-level module (dealing with specific calculations)
- Working mechanisms: Sequential inference can be accomplished by a single forward propagation without intermediate process supervision
The design is directly inspired by research in the field of cognitive science on how the human brain works, specifically:
- Hierarchical information processing mechanisms in the cerebral cortex
- Differences in the time scale of information processing in different brain regions
- A synergistic working model of abstract thinking and embodied computing
This answer comes from the articleHRM: Hierarchical Reasoning Model for Complex ReasoningThe
May not be reproduced without permission:AI productivity tools " What is the HRM model? What inspired its design?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil