all publishers

Ryuketsukami

@ryuketsukami source repo

2 published skills

  1. ▌
    Turboquant · ryuketsukami bundle
    Implement, use, or explain TurboQuant — Google's data-oblivious vector quantization algorithm for LLM KV cache compression (ICLR 2026). Use this skill when the user asks about KV cache compression, TurboQuant, PolarQuant, QJL (Quantized Johnson-Lindenstrauss), Lloyd-Max quantization for high-dimensional vectors, reducing LLM memory usage, compressing attention keys/values, or implementing any component of the TurboQuant pipeline. Also trigger when the user mentions vector quantization for inference optimization, 3-4 bit KV cache quantization, or inner product preserving compression.
    5 installs
  2. ▌
    Simple Explanation · ryuketsukami bundle
    Explain anything simply — lead with the answer, layer the depth, use plain words, keep every fact. Use when asked to explain, describe, clarify, simplify, summarize, teach, or break down a concept, system, codebase, architecture, decision, error, or document; when writing a TL;DR, overview, onboarding note, tutorial intro, or the answer to a how/why/what question; when asked for ELI5, layman's terms, beginner-friendly, non-technical, or plain-English explanations; or when an existing explanation buries its point, front-loads background, or overwhelms the reader. Do not apply to code or identifiers.
    6 installs