Metadata-Version: 2.4
Name: hydra_attention
Version: 3.0.1
Summary: Hierarchical Hybrid Dynamic Recurrent Attention with Local-Window Anchoring for Efficient Long-Context Processing
Home-page: https://github.com/WiMProject/project-hydra
Author: Wildan Miladji
Project-URL: Homepage, https://github.com/WiMProject/project-hydra
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: torch>=2.0.0
Requires-Dist: transformers>=4.30.0
Dynamic: home-page
Dynamic: license-file
Dynamic: requires-python

# ProjectHydra (H-HYDRA V3.0)

**H-HYDRA V3.0** is a novel hybrid neural network architecture unifying linear-time State Space Models (SSM) with Sparse Anchor Attention via a gated dynamic compute router.

Developed by **Wildan Miladji**.

## Installation

```bash
pip install hydra-attention
