agentmods.dev

lightcompress-batch-quantize

lightcompress-batch-quantize is a skill for Claude Code, Codex from HorizonRobotics/OE-Skills. It costs 23 tokens per session (721 once invoked), scanned A, original, Apache-2.0.

Instructions for running many machine-learning model quantization experiments and comparing their results. Quantization reduces the numerical precision used by a model, which can make it smaller or faster but may affect accuracy.

In plain words
What is it for?
Use it to test combinations of models, quantization methods, and settings; skip experiments already cached; run the remaining tests; and compare perplexity results.
Why use it?
It removes the need to start each model-and-method combination by hand. It also reuses saved results and gathers completed experiments into one comparison table.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

not rated 27repo 4mo ago A scan 23 tokens original Apache-2.0

Good fit Use it to test combinations of models, quantization methods, and settings; skip experiments already cached; run the remaining tests; and compare perplexity results.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/horizonrobotics/oe-skills/lightcompress-batch-quantize
View source ↗ HorizonRobotics/OE-Skills
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add HorizonRobotics/OE-Skills --skill lightcompress-batch-quantize
Clone the repo
git clone --depth 1 https://github.com/HorizonRobotics/OE-Skills

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for lightcompress-batch-quantize

README.md
[![agentmods](https://agentmods.dev/badge/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize/github.svg)](https://agentmods.dev/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize)
Your own site
<a href="https://agentmods.dev/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize"><img src="https://agentmods.dev/badge/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for lightcompress-batch-quantize

Your own site · 80×15
<a href="https://agentmods.dev/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize"><img src="https://agentmods.dev/badge/skills/horizonrobotics/oe-skills/lightcompress-batch-quantize.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 23 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 721 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00023 $0.00721
Opus 5.5 $0.00009 $0.00288
Sonnet 5.5 $0.00005 $0.00144
Haiku 4.5 $0.00002 $0.00072

Measured 2mo ago against content hash 85f4fdc34ba0, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

lightcompress-batch-quantize scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

horizon/skills/llm/lightcompress/lightcompress-batch-quantize/SKILL.md · 73 lines

What it actually says

LightCompress 批量量化实验

概述

批量执行 = 循环调用 lightcompress-quant-explore skill。本 skill 只负责编排，不重复实现单实验逻辑。

Phase 1: 解析用户输入，展开实验列表

将用户输入展开为实验的笛卡尔积：models x methods x configs。

示例输入：

对 Qwen3-0.6B、Qwen2.5-7B 跑 RTN W8A8 和 GPTQ W4A8 量化

展开为 4 个实验：

  1. Qwen3-0.6B + RTN W8A8
  2. Qwen3-0.6B + GPTQ W4A8
  3. Qwen2.5-7B + RTN W8A8
  4. Qwen2.5-7B + GPTQ W4A8

Phase 2: 检查精度缓存

调用 quant-accuracy-cache skill 查询每个实验是否已有结果。

匹配规则：model.name + quant.method + quant.w_bit + quant.a_bit。

  • 命中缓存：记录结果，从待执行列表中移除
  • 未命中缓存：保留在待执行列表

Phase 3: 逐个调用 lightcompress-quant-explore

对待执行列表中的每个实验：

  1. 调用 lightcompress-quant-explore skill 启动实验
    • 不指定 GPU，让子 skill 自动选择空闲 GPU
  2. 等待实验真正启动（通过 nvidia-smi 确认显存占用上升）
  3. 启动下一个实验

自动回答子 skill 的交互问题：

子 skill 问题 预设答案
开始实验 / 取消 开始实验
保存到缓存 / 跳过 保存到缓存
缓存冲突处理 保留两者

Phase 4: 汇总结果，生成对比表

所有实验完成后，汇总为对比表：

| 模型 | 方法 | 配置 | Pretrain PPL | Fake Quant PPL | PPL 变化 | 来源 |
|------|------|------|--------------|----------------|----------|------|
| Qwen3-0.6B | RTN | W8A8 | 21.30 | 22.15 | +0.85 | 实验 |
| Qwen2.5-7B | GPTQ | W4A8 | 7.00 | 7.45 | +0.45 | 缓存 |

调用 quant-accuracy-cache 保存新产生的实验结果。

与其他 Skill 的协作

Skill 关系 说明
lightcompress-quant-explore 依赖 单实验执行引擎，负责 GPU 选择、配置生成、实验运行
quant-accuracy-cache 协作 实验前查询缓存跳过重复，实验后保存结果
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2mo ago First seen · 73 lines · 23 tokens per session scan A 85f4fdc34ba0

Subscribe to this mod's changes

lightcompress-batch-quantize is a skill published in the GitHub repository HorizonRobotics/OE-Skills (27 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 23 tokens to every session and 721 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Data and AIScientific
Related

lightcompress-batch-quantize compared with skills from other repositories

lightcompress-batch-quantize next to 6 skills from other repositories
ModTypeStarsLast pushLicenceScanTokensWorks with
lightcompress-batch-quantizeHorizonRobotics/OE-SkillsSkill274mo agoApache-2.0A23Claude Code, Codex
arboretoK-Dense-AI/scientific-agent-skillsSkill56k4d agoMITA55Claude Code, Codex
deepspot-mK-Dense-AI/scientific-agent-skillsSkill56k4d agoMITA70Claude Code
torchdrugK-Dense-AI/scientific-agent-skillsSkill56k4d agoMITA64Claude Code
pyhealthK-Dense-AI/scientific-agent-skillsSkill56k4d agoMITA84Claude Code, Codex
nemo-mbridge-perf-expert-parallel-overlapNVIDIA/skillsSkill3.9kyesterdayApache-2.0A56Claude Code
ml-training-recipesOrchestra-Research/AI-Research-SKILLsSkill14k26d agoMITA88Claude Code