Repository profile
scrya-com/rotorquant
KV cache compression via block-diagonal rotation. Beats TurboQuant: better PPL (6.91 vs 7.07), 28% faster decode, 5.3x faster prefill, 44x fewer params. Drop-in llama.cpp integration.
Why this page exists
Use this profile to move from awareness into adoption-oriented inspection.
Best next step
Check the summary, then compare it against similar projects before touching production.
Research posture
Momentum helps discovery. Fit, maintenance quality, and reversibility decide adoption.
Editorial summary
RotorQuant is a cutting-edge KV cache compression technology designed for large language models (LLMs). By utilizing block-diagonal rotations, RotorQuant bypasses the conventional butterfly network, resulting in superior performance metrics compared to Google's TurboQuant. It achieves better perplexity (PPL), faster decoding, and significantly reduced parameter counts, making it an attractive option for developers looking to optimize LLM performance without compromising quality. The drop-in integration with llama.cpp further simplifies its adoption, allowing users to enhance their projects with minimal friction.
Adoption analysis
Best-fit use case
scrya-com/rotorquant is most useful to evaluate when your team is researching Python ecosystem tooling. Compare its documented workflow with your runtime, deployment model, and maintenance capacity before adopting it.
Momentum signal
Recent tracked star growth is modest, so maintenance quality and fit may matter more than momentum. Daily and three-day changes are discovery signals, while total stars show accumulated awareness.
Adoption caution
Before adding it to production, review license terms, dependency footprint, security guidance, open issue quality, and whether there is a clear path to migrate away later.
What to inspect next
- 1Run the quick install in a disposable project before touching production code.
- 2Check whether the README clearly states the project scope and non-goals.
- 3Identify at least two alternatives so the decision is not based on one ranking page.
- 4Read recent issues and releases to understand maintenance rhythm, breaking changes, and common failure modes.