turboquant-pytorch
🎯Skillfrom aradotso/trending-skills
From-scratch PyTorch implementation of Google TurboQuant (ICLR 2026) for LLM KV-cache compression: Stage 1 random orthogonal rotation + Lloyd-Max scalar quantization, Stage 2 QJL 1-bit sign residual correction for unbiased inner products — achieving 5x compression at 3-bit (58MB vs 289MB FP16) with 99.5% attention fidelity on Qwen2.5-3B.
Same repository
aradotso/trending-skills(263 items)
Installation
npx vibeindex add aradotso/trending-skills --skill turboquant-pytorchnpx skills add aradotso/trending-skills --skill turboquant-pytorch~/.claude/skills/turboquant-pytorch/SKILL.mdSKILL.md
More from this repository10
Skill
InkOS is a multi-agent CLI (`@actalk/inkos`) that autonomously writes, audits, and revises novels through a Writer → Validator → Auditor → Reviser pipeline with configurable human review gates. Supports book creation, per-chapter writing/audit, and a continuous-production daemon, tracking subplots, emotional arcs, and character matrices.
Skill
Curated collection of 50+ specialized AI agent personalities (engineering, design, marketing, sales, paid media) for Claude Code, Cursor, Aider, Windsurf, and Copilot, each with distinct expertise, workflows, and deliverables. Ships an interactive installer that auto-detects target tools and writes agent files to the right paths (`~/.claude/agents/`, `.cursor/rules/`, `.windsurf/rules/`, etc.).
Skill
Skill
Skill
Skill
Skill
Skill