https://github.com/hummbl-dev/governed-compression
Governed vector and KV-cache compression research — CPU reference, reproducible benchmarks, tuple-based experiment logging
https://github.com/hummbl-dev/governed-compression
Last synced: 30 days ago
JSON representation
Governed vector and KV-cache compression research — CPU reference, reproducible benchmarks, tuple-based experiment logging
- Host: GitHub
- URL: https://github.com/hummbl-dev/governed-compression
- Owner: hummbl-dev
- License: apache-2.0
- Created: 2026-04-14T12:52:48.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-05-13T21:02:28.000Z (about 2 months ago)
- Last Synced: 2026-05-13T23:08:30.874Z (about 2 months ago)
- Language: Python
- Size: 12.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Security: SECURITY.md
- Roadmap: docs/ROADMAP.md
Awesome Lists containing this project
README
# governed-compression
Private research implementation surface for governed vector and KV-cache compression.
## Purpose
This repo exists to provide a better implementation surface than the current fragmented TurboQuant-adjacent landscape.
Core goals:
- CPU reference implementation first
- reproducible benchmarks
- Windows via WSL2 first
- tuple-based experiment logging from day one
- method comparison across TurboQuant-style, QJL, and simple baselines
## Initial Scope
- vector encode / decode
- approximate dot product
- distortion metrics
- simple benchmark harness
- experiment logging
This repo is not yet a full inference-runtime integration project.
## Layout
```text
governed_compression/
core/
bench/
logging/
tests/
examples/
docs/
```
## Quick Start
```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
python -m governed_compression.cli
```
## Current Status
Stage 1 scaffold only.