-
Initial release: TurboQuant practical guide for consumer hardware
AI Engineering Lab
- README.md: bilingual (EN/DE), benchmark results, quick start, 5 error fixes
- WHITEPAPER.de.md: full German white paper
- Dockerfile: reproducible CUDA build (feature/turboquant-kv-cache branch)
- scripts/: download-model.sh, run-baseline.sh, run-turbo.sh
- results/: RTX 3090 benchmark JSON (April 2026)
Results: 100K context on RTX 3090 with +1.8 GB VRAM and -8.5% TPS
Based on TurboQuant (ICLR 2026, arXiv:2504.19874)
