# pathwaycom/bdh

BDH (Dragon Hatchling) – Architecture and Code

Repository: https://github.com/pathwaycom/bdh
Canonical: https://ross.abutalabs.com/products/bdh
Language: Python
License: MIT
License Family: permissive
Last push: 2026-05-16T02:53:54+00:00

## Health v2 (maintenance only)
Score: 54/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 82, release rhythm 35, longevity 24
- inputs: {"age_days": 337, "days_push": 109, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3519, forks 248 (observed 2026-08-28T04:08:07.907262+00:00)

## What it is
BDH (Dragon Hatchling) is a biologically inspired large language model architecture that bridges deep learning and neuroscience, implemented as an official PyTorch research codebase. It uses a scale-free network of locally interacting neurons with Hebbian working memory, matching GPT-2-scale Transformers on language and translation benchmarks while remaining interpretable.

## Use cases
- train a biologically inspired language model from scratch
- experiment with alternatives to transformer architectures
- study interpretability and monosemanticity in neural networks
- reproduce research on neuron-level reasoning dynamics
- benchmark scaling laws for non-transformer LLM architectures
- explore state-space formulations of attention-like computation

## When to choose
- you want an interpretable, neuroscience-grounded alternative to Transformers for language modeling
- you are doing research on emergent reasoning and biological neural dynamics
- you need a GPU-friendly implementation of a novel LLM architecture at 10M-1B parameter scale

## When to avoid
- you need a production-ready LLM with ecosystem tooling like mainstream transformer libraries
- you expect out-of-the-box reproduction of internal benchmark results like the 97.4% Sudoku Extreme accuracy
- you need fine-tuning or inference of existing pretrained models rather than training new architectures

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training, llm-inference, transformers
- domain: artificial-intelligence, deep-learning, large-language-models
- platform: python, cross-platform
- tags: biologically-inspired-architecture, neuroscience, state-space-models, interpretability, scaling-laws, research-code, alternative-to-transformers, natural-language-processing, research, gpu, linux

## Member repositories
- pathwaycom/bdh (main) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:07.907262+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:35:33.170269+00:00, confidence not recorded.
  - readme: https://github.com/pathwaycom/bdh (fetched 2026-08-28T04:08:07.907262+00:00, sha c11f85e5a905)
- Data as of 2026-08-30T08:39:29.467469+00:00.