Exécutez n'importe quel Skill dans Manus
en un clic

Exécutez n'importe quel Skill dans Manus en un clic

model-pr-diff-dossier

Étoiles662

Forks58

Mis à jour16 mai 2026 à 16:11

Use when creating or revising model PR optimization history documents for SGLang, vLLM, or another serving framework that cite GitHub PRs. Requires manual, per-PR source-diff review and documentation of motivation, key implementation approach, most important code excerpts, reviewed files, and validation implications instead of generated or one-line summaries.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Exécuter dans Manus

Source

BBuf

BBuf/AI-Infra-Auto-Driven-SKILLS

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Exécuter dans Manus

Métiers associésSOC

Basé sur la classification professionnelle SOC

Développeurs de logicielsProfessions informatiques et mathématiques·SOC 15-1252

Explorateur de fichiers

2 fichiers

SKILL.md

readonly

name	model-pr-diff-dossier
description	Use when creating or revising model PR optimization history documents for SGLang, vLLM, or another serving framework that cite GitHub PRs. Requires manual, per-PR source-diff review and documentation of motivation, key implementation approach, most important code excerpts, reviewed files, and validation implications instead of generated or one-line summaries.

Model PR Diff Dossier

Use this skill before publishing any model PR optimization history document that cites framework PRs.

Non-Negotiable Standard

Do not summarize a PR with only a title-level sentence. Do not use a script or bulk generator to fill motivation, implementation notes, or code excerpts.

For every PR cited as model optimization evidence, the document must include or link to a diff-reviewed PR card with:

PR link, title, state, merge time when available, additions/deletions, and changed-file count.
Motivation: why the PR existed, inferred from PR body, title, issue context, docs changes, tests, and code diff.
Key implementation idea: what runtime path changed and how the patch implements the change.
Key code excerpts: short, relevant snippets from the actual diff, not invented pseudocode.
Reviewed files: important files from the full diff, grouped by runtime/docs/tests where possible.
Validation implications: tests, benchmark paths, launch flags, or regression lanes implied by the diff.
Diff coverage note: state that the full diff was fetched/read and include diff line count.

Workflow

Collect exact PR links from the target model history files. Use GitHub PR URLs, not bare #123 text.
Open each PR diff directly with GitHub, gh pr diff, or the local framework repository commit. Read the changed source files, not just the PR title.
For merged PRs, cross-check the final mainline code in the relevant framework checkout when the diff is ambiguous.
Write the PR card manually in the matching model history document. Use references/card-schema.md when you need the exact card shape. The card must name concrete files/functions/classes and include a short real code excerpt.
For docs-only or config-only PRs, quote the exact command/config line that changed and explain why it matters for serving or validation.
After each model family, review the cards for repeated shallow words such as "follow-up", "bugfix", or "optimization"; replace them with concrete implementation detail.
Run repository tests and formatting before publishing.

Review Gate

A model PR history is not ready if any PR card says only "follow-up", "bugfix", "docs", or "optimization" without:

named files/functions/classes touched by the diff,
a concrete motivation,
a concrete implementation summary,
and at least one code excerpt or an explicit reason why the PR is docs-only.

Plus depuis ce dépôt

même dépôt

model-pr-history-knowledge

BBuf/AI-Infra-Auto-Driven-SKILLS

Use when an SGLang, vLLM, TensorRT-LLM, or TokenSpeed serving/model optimization task needs prior model-family PR evidence. Query and read the PR-driven history docs under model-pr-optimization-history before choosing source paths, fast paths, kernel/fusion ideas, regression risks, or validation lanes.

2026-06-27662

llm-serving-auto-benchmark

BBuf/AI-Infra-Auto-Driven-SKILLS

Framework-independent LLM serving benchmark skill for comparing SGLang, vLLM, TensorRT-LLM, TokenSpeed, or another serving framework. Use when a user wants to find the best deployment command for one model across multiple serving frameworks under the same workload, GPU budget, and latency SLA.

2026-06-27662

llm-torch-profiler-analysis

BBuf/AI-Infra-Auto-Driven-SKILLS

Unified LLM torch-profiler triage skill for `sglang`, `vllm`, `TensorRT-LLM`, and `TokenSpeed`. Use it to inspect an existing `trace.json(.gz)` or profile directory, or to drive live profiling against a running server when supported and return one three-table report with kernel, overlap-opportunity, and fuse-pattern tables.

2026-06-27662

sglang-sota-humanize-loop

BBuf/AI-Infra-Auto-Driven-SKILLS

Run an autonomous Humanize-governed SGLang SOTA performance loop for one LLM model: first perform a fixed fair SGLang benchmark against the requested comparison framework set, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches SGLang code, optionally uses ncu-report-skill for kernel evidence, and revalidates until SGLang matches or beats the best observed requested framework under the same workload and SLA.

2026-06-27662

sglang-humanize-review

BBuf/AI-Infra-Auto-Driven-SKILLS

Perform SGLang code review in the style of human maintainers by consulting the full non-agent PR review episode corpus from project start through the latest refresh (June 2026), including inline review threads, top-level PR comments, review submissions, original multilingual text, and multi-round discussions. Use when reviewing SGLang PRs, diffs, patches, or local changes for correctness, tests, performance, GPU/runtime risks, API compatibility, and maintainability.

2026-06-13662

vllm-sota-humanize-loop

BBuf/AI-Infra-Auto-Driven-SKILLS

Run an autonomous Humanize-governed vLLM SOTA performance loop for one LLM model: first perform the fixed fair vLLM/SGLang/TensorRT-LLM deployment search and benchmark, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches vLLM code, optionally uses ncu-report-skill for kernel evidence, and revalidates until vLLM matches or beats the best observed framework under the same workload and SLA.

2026-05-26662

name	model-pr-diff-dossier
description	Use when creating or revising model PR optimization history documents for SGLang, vLLM, or another serving framework that cite GitHub PRs. Requires manual, per-PR source-diff review and documentation of motivation, key implementation approach, most important code excerpts, reviewed files, and validation implications instead of generated or one-line summaries.

Model PR Diff Dossier

Use this skill before publishing any model PR optimization history document that cites framework PRs.

Non-Negotiable Standard

Do not summarize a PR with only a title-level sentence. Do not use a script or bulk generator to fill motivation, implementation notes, or code excerpts.

For every PR cited as model optimization evidence, the document must include or link to a diff-reviewed PR card with:

PR link, title, state, merge time when available, additions/deletions, and changed-file count.
Motivation: why the PR existed, inferred from PR body, title, issue context, docs changes, tests, and code diff.
Key implementation idea: what runtime path changed and how the patch implements the change.
Key code excerpts: short, relevant snippets from the actual diff, not invented pseudocode.
Reviewed files: important files from the full diff, grouped by runtime/docs/tests where possible.
Validation implications: tests, benchmark paths, launch flags, or regression lanes implied by the diff.
Diff coverage note: state that the full diff was fetched/read and include diff line count.

Workflow

Collect exact PR links from the target model history files. Use GitHub PR URLs, not bare #123 text.
Open each PR diff directly with GitHub, gh pr diff, or the local framework repository commit. Read the changed source files, not just the PR title.
For merged PRs, cross-check the final mainline code in the relevant framework checkout when the diff is ambiguous.
Write the PR card manually in the matching model history document. Use references/card-schema.md when you need the exact card shape. The card must name concrete files/functions/classes and include a short real code excerpt.
For docs-only or config-only PRs, quote the exact command/config line that changed and explain why it matters for serving or validation.
After each model family, review the cards for repeated shallow words such as "follow-up", "bugfix", or "optimization"; replace them with concrete implementation detail.
Run repository tests and formatting before publishing.

Review Gate

A model PR history is not ready if any PR card says only "follow-up", "bugfix", "docs", or "optimization" without:

named files/functions/classes touched by the diff,
a concrete motivation,
a concrete implementation summary,
and at least one code excerpt or an explicit reason why the PR is docs-only.