numerical-determinism

仓库创建 2026年7月6日最近提交 22 天前SkillHot 收录 22 天前

⬈ GitHub 仓库 ⬈ SKILL.md 原文件 ⬇ Raw 原始文件

▸ 精选理由

对需比对位级或计时结果的实验尤为重要。

▸ 风险提示

可能需要采集或暴露底层硬件和环境信息。

这个 Skill 做什么

定义锁定线程、并行顺序与硬件信息以实现数值可复现性。

当同样代码和 seed 还会跑出不一样的数时用它：会把线程数、并行规约顺序、GPU kernel 和引擎级非确定性都固定住，帮你追求位级可复现或给出有意义的时间/硬件上下文。也会规定每次运行要记录的硬件元信息，方便跨机比对和复现问题。

▸ 展开 SKILL.md 英文原文

Use when the same code and same seed still produce different numbers — pinning thread counts, parallel reduction order, GPU kernels, and engine-level nondeterminism — and when a claim requires bit-reproducibility or a timing result requires hardware context. Also defines the hardware-context fields every run's meta file should record. Trigger phrases: "same seed different result", "pin the threads", "deterministic mode", "results differ across machines", "record the hardware", "timing needs the machine specs".

开发编程数值重现硬件上下文随机性控制通用

Stars

Forks

仓库内 Skill

7 日增星

安装 / 使用

给你的 Agent 一句话（通用）

帮我安装这个 skill：https://raw.githubusercontent.com/huguryildiz/ALETHEIA/main/skills/extended/numerical-determinism/SKILL.md

或 curl 直取 SKILL.md

curl -fsSL "https://raw.githubusercontent.com/huguryildiz/ALETHEIA/main/skills/extended/numerical-determinism/SKILL.md"

SKILL.MD 节选查看完整文件 ↗

# numerical-determinism

Seeds pin the *draws*; they do not pin the *arithmetic*. Parallel reductions reorder
floating-point sums, GPU atomics race, and optimization engines take time-dependent paths —
so "same seed, different number" is expected behavior until execution is pinned too. This
skill sets the pinning knobs, defines honest reproducibility tiers, and (folded in) the
hardware-context record every run should carry.

## When to use

- A seeded computation gives different results across runs or machines.
- A claim needs bit-reproducibility (regression pins, cross-checking two implementations).
- Any timing/performance number is about to be reported.
- Setting up the meta-file fields f

via SKILL·HOT · 数据来自 GitHub 公开信息 · 原文版权归作者所有