CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models

In three linesCosmicFish-HRM is a compact model with a Hierarchical Reasoning Module (HRM) that dynamically allocates computational effort during inference. The model learns when to halt based on input complexity, combining high-level and low-level reasoning cycles with Grouped Query Attention, RoPE, and SwiGLU. Results show non-uniform reasoning behavior adapted to tasks and inputs.

Summary generated by Claude — human-verified