# TurboQuant **Source:** https://www.startuphub.ai/startups/turboquant **Profile type:** Startup > Google Research's AI compression algorithm that drastically reduces LLM memory requirements by compressing the KV cache. ## About TurboQuant is a novel AI compression algorithm developed by Google Research. It significantly reduces the memory requirements for large language models (LLMs) by employing advanced quantization techniques, such as PolarQuant and Quantized Johnson, to compress the key-value (KV) cache without compromising performance. ## Key facts - **Slug:** turboquant - **Website:** https://github.com/google-research/google-research/tree/master/turbo_quant - **Operating status:** Active - **Sectors / tags:** AI Foundation & Compute, Large Language Model, Generative AI, Transformer Architecture, AI Infrastructure, Vector Search, Data Compression, Foundation Model, Quantization, Machine Learning - **Social:** - Twitter: https://x.com/GoogleResearch - GitHub: https://github.com/google-research ## Frequently asked ### What does TurboQuant do? TurboQuant is a novel AI compression algorithm developed by Google Research. It significantly reduces the memory requirements for large language models (LLMs) by employing advanced quantization techniques, such as PolarQuant and Quantized Johnson, to compress the key-value (KV) cache without compromising performance. ### What industry does TurboQuant operate in? TurboQuant operates in AI Foundation & Compute, Large Language Model, Generative AI, Transformer Architecture, AI Infrastructure, Vector Search. ## JSON-LD Structured data aligned with this page's `