{"slug": "bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture", "title": "BDH Dragon Hatchling:attention-based state space sequence learning architecture", "summary": "Pathway researchers released BDH (Dragon Hatchling), a biologically inspired large language model architecture that matches GPT-2-scale Transformers on language and translation tasks at 10M–1B parameters while achieving 97.4% accuracy on Sudoku Extreme puzzles without chain-of-thought, backtracking, or external tools. The architecture, described in the paper 'The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain' (arXiv:2509.26507), uses a scale-free, locally interacting network of neurons with Hebbian working memory to provide interpretable, sparse activations and Transformer-like scaling laws.", "body_md": "**BDH (Dragon Hatchling)** is a biologically inspired large language model architecture that connects principles of deep learning with the foundations of neuroscience. Developed by researchers at [Pathway](https://pathway.com), BDH provides a theoretical and practical framework for understanding the emergence of reasoning and generalization in artificial systems.\n\nThis repository contains the official implementation from the paper:\n\nA. Kosowski, P. Uznański, J. Chorowski, Z. Stamirowska, M. Bartoszkiewicz.[, arXiv (2025).]The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain\n\nBDH represents a **scale-free, locally interacting network of neurons** capable of intrinsic reasoning dynamics. BDH scales like a Transformer on performance benchmarks—yet retains full interpretability and theoretical grounding in the fine-grained dynamics of neuron interactions.\n\n**Key properties:**\n\n**Scale-free network topology** mimicking biological connectivity**Locally interacting neuron particles** with excitatory/inhibitory dynamics**Hebbian working memory** based on synaptic plasticity, displaying monosemanticity**GPU-friendly state-space formulation** for efficient implementation**Interpretable activations** that are sparse and positive\n\nBDH formalizes a bridge between **neural computation and machine-based language understanding**. It shows how **macro reasoning behavior** in large AI models emerges from **micro-level neuron dynamics**, guided by principles of graph theory and local computation.\n\nEmpirically, BDH matches **GPT-2–scale Transformers** across language and translation tasks at equivalent parameter scales (10M–1B).\n\nBDH and the Transformer share attention-inspired computation; however, BDH’s graph-based architecture makes its attention **emerge naturally from neuron-level interactions**, reflecting attention as seen in biological systems.\n\nBDH follows **Transformer-like scaling laws**, maintaining parameter efficiency while achieving interpretability at any scale.\n\nNote: The Sudoku Extreme result refers to Pathway’s internal BDH implementation, not to the current open-source repository. This repository contains the implementation of the baseline variant as described in our [public paper](https://arxiv.org/abs/2509.26507) and does not reproduce the 97.4% benchmark result out of the box. See the dedicated Extreme Sudoku research blog post for additional benchmark context and the reported results.\n\nOn Sudoku Extreme, BDH reaches 97.4% accuracy across roughly 250,000 difficult puzzles, without chain-of-thought, solution backtracking, or external tool use, while leading LLMs struggle to perform on the benchmark at all.\n\nLanguage is not enough for intelligence. Transformers process information token by token with limited internal state, which makes search-heavy, non-linguistic reasoning tasks like Sudoku awkward. BDH uses a larger latent reasoning space with intrinsic memory that supports learning and adaptation during use.\n\nWe believe that the future of AI will belong to systems that can reason natively across domains, that can hold multiple possibilities in a rich latent space, and that can converge on solutions without needing to verbalize every step. BDH is our answer to that challenge. It is designed to be a universal reasoning system that can speak our language without being trapped inside it. And yes, it solves Sudoku.\n\nRead more: [Post-transformers: Sudoku Bench](https://pathway.com/research/beyond-transformers-sudoku-bench)\n\n| Model | Sudoku Extreme Accuracy | Relative Cost |\n|---|---|---|\n| Pathway BDH | 97.4% | 10× lower, No chain-of-thought |\n| Leading LLMs (O3-mini, DeepSeek R1, Claude 3.7 8K) | ~0% | High (chain-of-thought) |\n\n*Table 1: Performance comparison on extreme Sudoku benchmarks (~250,000 difficult puzzles).*\n\n*Source: Pathway internal data and https://arxiv.org/pdf/2506.21734 for the Leading LLMs’ accuracy score. Pathway’s approach reflects top-1 accuracy and does not rely on chain-of-thought nor solution backtracking.*\n\n```\n# install dependencies\npip install -r requirements.txt\n\n# train BDH on a toy dataset\npython train.py\n```\n\n-\nWatch the\n\n*SuperDataScience podcast*(72 min.) featuring Adrian Kosowski in conversation with Jon Krohn, unpacking BDH’s neuron-level architecture and sparse reasoning dynamics.▶️ *Dragon Hatchling: The Missing Link Between Transformers and the Brain* -\nRead about BDH in\n\n,*Forbes*,*Semafor*,*The Turing Post*,*Quantum Zeitgeist*, and elsewhere in the media.*Golem* -\nDiscuss and share the BDH paper on:\n\n,*Hugging Face Papers*, and*Alphaxiv*.*EmergentMind*\n\n[adamskrodzki/bdh](https://github.com/adamskrodzki/bdh): dynamic vocabulary, stateful attention[mosure/burn_dragon_hatchling](https://github.com/mosure/burn_dragon_hatchling): Burn port[severian42/bdh](https://github.com/severian42/bdh): MLX port[Git-Faisal/bdh](https://github.com/Git-Faisal/bdh)[GrahLnn/bdh](https://github.com/GrahLnn/bdh)\n\nWe thank Andrej Karpathy for the [nanoGPT](https://github.com/karpathy/nanoGPT/) code and the tiny Shapespeare dataset used in this demonstration.\n\nBDH research stands at the intersection of **AI architecture**, **biological learning models**, and **theoretical computer science**—an effort to map the *equations of reasoning* between artificial and biological intelligence.", "url": "https://wpnews.pro/news/bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture", "canonical_source": "https://github.com/pathwaycom/bdh", "published_at": "2026-08-14 01:43:16+00:00", "updated_at": "2026-08-14 02:11:24.360173+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "neural-networks"], "entities": ["Pathway", "BDH", "Dragon Hatchling", "A. Kosowski", "P. Uznański", "J. Chorowski", "Z. Stamirowska", "M. Bartoszkiewicz"], "alternates": {"html": "https://wpnews.pro/news/bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture", "markdown": "https://wpnews.pro/news/bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture.md", "text": "https://wpnews.pro/news/bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture.txt", "jsonld": "https://wpnews.pro/news/bdh-dragon-hatchling-attention-based-state-space-sequence-learning-architecture.jsonld"}}