From the source
- Pathway announced its brain-inspired BDH (Dragon Hatchling) architecture, a post-transformer model that performs reasoning in latent space using sparse, local neuron interactions and Hebbian learning.
- The BDH-CQ variant updates internal memory during inference without fine-tuning or retraining.
- The model integrates with PyTorch and uses
Amazon SageMaker HyperPod for scalable training.





