Ruby helpers and a curated link map for TurboQuant, PolarQuant, and Quantized Johnson-Lindenstrauss (QJL) as described by Google Research.
This release focuses on research metadata and stable URL constants. It does not ship a native quantisation implementation; see the papers below for algorithms.
Add to your Gemfile:
gem 'turboquant'Or install the gem:
gem install turboquantrequire 'turboquant'
TurboQuant::VERSION
# => "0.1.0"
TurboQuant::Research::BLOG_POST
TurboQuant::Research::TURBOQUANT_ARXIV_ABS
TurboQuant::Research.urls # => { blog_post: "...", ... }| Topic | URL |
|---|---|
| Google blog post | https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/ |
| TurboQuant (arXiv) | https://arxiv.org/abs/2504.19874 |
| PolarQuant (arXiv) | https://arxiv.org/abs/2502.02617 |
| Quantized JL (ACM) | https://dl.acm.org/doi/10.1609/aaai.v39i24.34773 |
| JL background (PDF) | https://arxiv.org/pdf/2103.00564 |
| KV caching explainer | https://huggingface.co/blog/not-lain/kv-caching |
| Vector quantisation | https://en.wikipedia.org/wiki/Vector_quantization |
Further benchmarks and baselines (LongBench, ZeroSCROLLS, needle-in-a-haystack, KIVI, GloVe, etc.) are available as constants on TurboQuant::Research.
MIT. See LICENSE.txt.