https://github.com/thiagosilvahyper/bihe-quantization
BIHE Protocol - Next-generation vector quantization combining E8 lattice geometry with Lloyd algorithm optimization. Achieves 16× compression ratio while maintaining 88.5% recall on real-world datasets. Validated on Stanford SQuAD v2.0 (5,000 texts) and Microsoft MS MARCO (6,000 texts). Features NSM below Shannon theoretical limits.
https://github.com/thiagosilvahyper/bihe-quantization
compression e8-lattice embeddings lattice-quantization machine-learning neural-networks python pytorch quantization vector-quantization
Last synced: 16 days ago
JSON representation
BIHE Protocol - Next-generation vector quantization combining E8 lattice geometry with Lloyd algorithm optimization. Achieves 16× compression ratio while maintaining 88.5% recall on real-world datasets. Validated on Stanford SQuAD v2.0 (5,000 texts) and Microsoft MS MARCO (6,000 texts). Features NSM below Shannon theoretical limits.
- Host: GitHub
- URL: https://github.com/thiagosilvahyper/bihe-quantization
- Owner: thiagosilvahyper
- License: mit
- Created: 2025-10-27T15:24:55.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2025-10-27T16:33:34.000Z (7 months ago)
- Last Synced: 2026-04-30T12:33:46.346Z (16 days ago)
- Topics: compression, e8-lattice, embeddings, lattice-quantization, machine-learning, neural-networks, python, pytorch, quantization, vector-quantization
- Language: Python
- Homepage:
- Size: 67.4 KB
- Stars: 4
- Watchers: 0
- Forks: 0
- Open Issues: 0