Paper 2605.14386
Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning
- Published
- May 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- Not linked
01 In brief
Summary
The Darwin Family framework enables training-free evolutionary merging of large language models via gradient-free weight-space recombination, improving reasoning performance without additional training.
It introduces a 14-dimensional adaptive merge genome, MRI-Trust Fusion (balancing diagnostic layer-importance signals with evolutionary search via a learnable trust parameter τ), and an Architecture Mapper for cross-architecture breeding.
The flagship Darwin-27B-Opus achieves 86.9% on GPQA Diamond, ranking #6 among 1,252 models, outperforming its fully trained parent without gradient-based training.
Across scales (4B–35B), Darwin models consistently improve over parents, support recursive multi-generation evolution, and enable Transformer–Mamba hybrid merging.
Ablations show adaptive τ yields +2.5pp over no-MRI baseline on GPQA.
The framework generalizes across scales and architectures, with stable patterns of attention preservation and feed-forward recombination.
Limitations include dependence on parent capabilities and shared pretrained bases.
All models, code, and tooling are released under Apache 2.0, with community downloads exceeding 96,000.
02 From the paper
Abstract
We present Darwin Family, a framework for training-free evolutionary merging of large language models via gradient-free weight-space recombination. We ask whether frontier-level reasoning performance can be improved without additional training, by reorganizing latent capabilities already encoded in existing checkpoints. Darwin introduces three key ideas: (i) a 14-dimensional adaptive merge genome enabling fine-grained component- and block-level recombination; (ii) MRI-Trust Fusion, which adaptively balances diagnostic layer-importance signals with evolutionary search through a learnable trust parameter; and (iii) an Architecture Mapper that enables cross-architecture breeding between heterogeneous model families. Empirically, the flagship Darwin-27B-Opus achieves 86.9% on GPQA Diamond, ranking #6 among 1,252 evaluated models, and outperforming its fully trained foundation model without any gradient-based training. Across scales from 4B to 35B parameters, Darwin models consistently improve over their parents, support recursive multi-generation evolution, and enable a training-free evolutionary merge that combines Transformer- and Mamba-based components. Together, the Darwin Family demonstrates that diagnostic-guided evolutionary merging is a practical and reproducible alternative to costly post-training pipelines for reasoning-centric language models.