跳到正文
cccskills
免费在 GitHub 公开

organize-experiments

Collect experiment reports from issues and organize them in an existing Markdown index. It places entries in relevant sections, merges WandB experiment links into existing records, and checks for duplicate entries and formatting inconsistencies.

中文简介正在准备中。当前显示英语简介。

原文查看原文说明

Harvest experiment issue reports and curate docs/reports/index.md only when explicitly requested.

查看安装方法

适用场景

  • Adding new experiment reports to an index
  • Sorting uncategorized experiments
  • Merging WandB links into existing entries
  • Checking index duplicates and Markdown formatting

包含的文件(1)

  • SKILL.md2.5 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Skill: Organize Experiment Reports

Overview

Curate docs/reports/index.md after new experiment issues are harvested: fold fresh entries into the right sections, refresh links, and leave ## Uncategorized empty.

Prerequisites

  • Local checkout of the marin repository with write access.
  • Ability to run uv commands.
  • Familiarity with the existing experiment categories in docs/reports/index.md.

Guidelines for Humans

Standard Workflow

  1. Run uv run scripts/pm/itemize_experiment_issues.py to append new experiments.
  2. Open the diff for docs/reports/index.md and identify the additions in ## Uncategorized.
  3. For each experiment:
    • Match it to an existing section (e.g., Training and Performance, Data Experiments).
    • Merge new WandB links into the canonical entry in that section.
    • Remove the placeholder entry from ## Uncategorized.
  4. If an experiment truly does not fit, leave it under ## Uncategorized and add a note explaining why.
  5. Proofread for duplicate bullets, broken Markdown, and consistent title casing.
  6. Commit the curated report, open a PR, or, if an interactive agent, signal to the user to look.

Link Hygiene

  • Prefer direct https://wandb.ai/... links when available; keep legacy api.wandb.ai links only if the direct share link does not exist.
  • Do not add Data Browser links. The service was retired in #8624; the script collects only WandB URLs.
  • Preserve access tokens embedded in links.

Rules for Agents

  • NEVER delete existing sections or conclusions.
  • Keep badge styling consistent: [![#NNN](https://img.shields.io/...)] next to each experiment title.
  • When merging new content, update the canonical entry instead of duplicating it.
  • Leave ## Uncategorized empty whenever possible; a single placeholder sentence is fine.
  • Ask for guidance if an experiment does not map cleanly to known categories.

Validation

  • rg "^- " docs/reports/index.md to confirm bullets exist only under curated sections, not under ## Uncategorized.
  • Re-run uv run scripts/pm/itemize_experiment_issues.py if unsure that all experiments were captured.
  • Optional: markdownlint docs/reports/index.md to catch formatting drift.

See Also

  • scripts/pm/itemize_experiment_issues.py for source data generation.
  • .agents/skills/archive-experiments/SKILL.md for retiring legacy experiments.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Compare Zephyr changes using the same prepared dataset and matching execution settings. Analyze Finelog metrics for each processing stage to distinguish CPU efficiency, elapsed time, and peak memory changes, then report improvements, regressions, or tradeoffs.

中文简介正在准备中。当前显示英语简介。

  • Compare Zephyr performance before and after a PR
  • Assess speed gains against additional CPU usage
  • Check peak memory and out-of-memory regressions
marin-community/marin40022026年10月22日 更新

Register Hugging Face datasets for Marin by inspecting their structure without downloading the full dataset. Identify text fields, configurations, and data splits, then add a dataset module and validate the mapping with a sample.

中文简介正在准备中。当前显示英语简介。

  • Registering a Hugging Face dataset for Marin
  • Inspecting text fields without a full dataset download
  • Checking configurations and data splits
marin-community/marin40022026年10月22日 更新

add-grant

免费

Turn a specified GCP access request into a reviewable Pulumi change and pull request. Select the narrowest suitable role and encrypt personal email identities, covering resource permissions and Cloud Run viewer access through IAP, an access gateway.

中文简介正在准备中。当前显示英语简介。

  • Prepare a request for GCS bucket read access
  • Add a viewer to a Cloud Run service protected by IAP
  • Prepare service account impersonation permissions
marin-community/marin40022026年10月22日 更新

Add or update Pallas/Mosaic kernels, specialized computation routines for JAX. The skill covers reference implementations, output and gradient checks, public wrappers, fallback behavior, performance benchmarks, and hardware-specific tuning when requested.

中文简介正在准备中。当前显示英语简介。

  • Adding a Pallas kernel for a JAX computation
  • Checking outputs and gradients against a reference
  • Maintaining backend selection and fallback behavior
marin-community/marin40022026年10月22日 更新

Preserve old experiment scripts with a dated Git tag, a named snapshot of the code, and add comments linking related GitHub issues to the archived files and cleanup pull request.

中文简介正在准备中。当前显示英语简介。

  • Preserving a snapshot of old experiment scripts
  • Linking experiment issues to archived files
  • Updating archive notices after a tag changes
marin-community/marin40022026年10月22日 更新

Launch and continuously monitor a specified Zephyr pipeline on Iris when asked. It checks coordinator and worker health, tracks stage progress, identifies stalled or repeated failures, and requires explicit approval before stopping or restarting jobs.

中文简介正在准备中。当前显示英语简介。

  • Launch a specified Zephyr pipeline and check startup
  • Monitor stage progress and worker health over time
  • Investigate stalled stages or repeated worker failures
marin-community/marin40022026年10月22日 更新

查看 marin-community 的全部技能

报告此技能的问题