{"slug": "cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in", "title": "CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models", "summary": "Researchers have introduced CosmicFish-HRM, a compact language model that uses a Hierarchical Reasoning Module (HRM) to dynamically allocate computational effort based on input complexity rather than applying fixed computation to every input. The model iterates through high-level and low-level reasoning cycles, learning when to halt, and demonstrates non-uniform reasoning behavior across different tasks. This adaptive reasoning depth approach offers a potential alternative to relying solely on massive parameter counts for achieving strong reasoning capabilities in language models.", "body_md": "arXiv:2605.28919v1 Announce Type: new\nAbstract: Large language models have achieved strong reasoning capabilities, though often at the cost of massive parameter counts and expensive inference. In this work, we explore a different direction: adaptive reasoning depth in compact language models. We present CosmicFish-HRM, a compact language model built around a Hierarchical Reasoning Module (HRM) that dynamically allocates computational effort during inference. Instead of applying fixed computation to every input, the model iterates through high-level and low-level reasoning cycles and learns when to halt based on input complexity. CosmicFish-HRM combines this adaptive reasoning core with modern transformer components including Grouped Query Attention, RoPE, and SwiGLU activations. While the additional reasoning infrastructure introduces overhead at small scale, we hypothesize that this tradeoff becomes increasingly favorable as model size grows and the relative cost of the HRM core diminishes. Our results show that the model learns non-uniform reasoning behavior, allocating different numbers of reasoning steps across tasks and inputs. These findings suggest that adaptive reasoning depth may offer a promising alternative to relying solely on parameter scale for reasoning capability.", "url": "https://wpnews.pro/news/cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in", "canonical_source": "https://arxiv.org/abs/2605.28919", "published_at": "2026-05-29 04:00:00+00:00", "updated_at": "2026-05-29 04:18:53.622792+00:00", "lang": "en", "topics": ["large-language-models", "machine-learning", "artificial-intelligence", "neural-networks", "ai-research"], "entities": ["CosmicFish-HRM", "Hierarchical Reasoning Module", "Grouped Query Attention", "RoPE", "SwiGLU"], "alternates": {"html": "https://wpnews.pro/news/cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in", "markdown": "https://wpnews.pro/news/cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in.md", "text": "https://wpnews.pro/news/cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in.txt", "jsonld": "https://wpnews.pro/news/cosmicfish-hrm-adaptive-reasoning-via-hierarchical-recurrent-mechanisms-in.jsonld"}}