TurboQuant KV cache compression (Hadamard, Lloyd-Max, QJL)
Parent: Mac local LLMs: KV cache sizing and quantization · Published reference · snapshot 2026-10-05
↓ Facts as markdownall context files
The concept is already covered by turboquant-and-rotation-based-kv-quantization-on-metal.md, llama-cpp-attention-rotation-hadamard-for-quanti.md and qjl-residual-correction-needed-or-harmful.md; sources checked in this batch (arXiv 2504.19874, llama.cpp discussion 20969, the Hannecke five-integra...
These notes link each claim to its source. A source may be a research report hosted on this site rather than the primary document. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
Facts
- The concept is already covered by turboquant-and-rotation-based-kv-quantization-on-metal.md, llama-cpp-attention-rotation-hadamard-for-quanti.md and qjl-residual-correction-needed-or-harmful.md; sources checked in this batch (arXiv 2504.19874, llama.cpp discussion 20969, the Hannecke five-integration-paths article) are already cited there and add nothing new. [source]
Children
- No children recorded.