跳到正文
cccskills
免费在 GitHub 公开

vllm-metax-upgrade

Orchestrate a repository-wide vLLM-MetaX compatibility upgrade or coordinated multi-component adaptation through a support matrix, dependency graph and scoped specialists. Use a domain skill for standalone component work.

中文简介正在准备中。当前显示原文说明。

实测评分 未实测

查看安装方法

包含的文件(3)

  • SKILL.md3.2 KB
  • agents/openai.yaml348 B
  • references/orchestration.md3.4 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

vLLM-MetaX Upgrade

Read and apply the complete shared upgrade workflow before this coordinator workflow. It owns environment confirmation, trust boundaries, upstream-first adaptation, validation and durable progress.

Own the whole requested support matrix, dependency graph, domain assignments and acceptance obligations. These roles may run sequentially; parallel agents are optional.

Plan the full upgrade

Use the existing run in full mode. After the common gate passes, define the agreed model/configuration matrix and evidence obligations, and persist its machine-readable rows before claiming support. Supported rows record their exact configuration, linked tasks and acceptance kinds; unsupported rows give the rejection reason. Create disposition tasks covering every baseline production/build/requirements inventory entry, including startup and build infrastructure outside the four specialist domains.

Read orchestration.md when constructing or expanding the graph. Build typed import, startup, side-effect, call, data and conditional edges. Plan around changed contracts and reachable dependencies; collapse strongly connected components into coordinated adaptation units rather than requiring cyclic members to import successfully before source adaptation.

Dispatch and schedule

  • Assign model, attention, registry and patch units to their corresponding skills. Their standalone scope boundaries permit coordinated dispatch here; shared rules remain in common and do not need to be copied into each assignment.
  • Cover platform selection, distributed communication, workers, allocator/cache, foundational ops, build/extension ABI and entry points through explicit contract review when no specialist exists.
  • Keep one owner per shared unit and one authoritative run. Workers return newly discovered edges, contract changes, dispositions and evidence to the coordinator.
  • When a prerequisite becomes ready, schedule its blocked consumer obligations. Prioritize an early minimal normal-entry vertical path, then expand to the agreed model, dtype/quantization, phase, parallel, MTP, graph and optional-feature matrix.

Full completion and handoff

Check every supported matrix row for actual production reachability and current acceptance evidence. Full completion requires a disposition for every baseline production/build/requirements file and a required passing normal-entry e2e obligation for every supported matrix row, together with each row's other declared evidence kinds. At least one row must be supported. Required blocked obligations keep the run incomplete; deliberately unsupported rows need their intended capability rejection, not a claimed pass.

At handoff, relate the remaining units and blocked closures to the support matrix and the common workflow's persisted next actions. Do not substitute a completed component or a narrower test result for the full-upgrade acceptance obligations.

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

Review and adapt MetaX attention backends, MLA, sparse indexers, cache layouts, and kernel wrappers against target vLLM and installed MetaX component contracts. Use for attention work outside vllm_metax/patch/.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

Trim large MetaX model directories for dummy smoke tests or real-checkpoint loading on limited GPUs. Preserve target execution paths while reducing depth; use the real-weight workflow when a loadable checkpoint is requested. Not for accuracy evaluation or model compatibility upgrades.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

Review and upgrade MetaX model support against target vLLM and installed MACA components, including model-dependent attention, weight loading and kernels. Use for model adaptations and shared-upstream-bug attribution, not standalone patch or registry audits.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

Audit and adapt monkey patches in vllm_metax/patch/ against a target upstream revision. Decide whether to retain, update, migrate, or remove each patch; maintain patch headers and explanations; validate changes and record evidence. Use only for compatibility work on this directory, not standalone attention backend, model, kernel, or other adaptations elsewhere in vllm_metax.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

Review and adapt vllm_metax/registry registrations, quantization configurations, CustomOps and kernel dispatch against a target vLLM revision and installed MetaX APIs. Verify registration, inherited state, weight layouts and actual backend selection. Use for registry compatibility work, not standalone attention algorithms or monkey patches.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

Shared workflow for vLLM-MetaX compatibility adaptations and reviews, including confirmed environments, upstream-first code, dependency contracts, validation and durable progress. Required by standalone domain skills and the full-upgrade coordinator; collecting environment facts may precede confirmation.

中文简介正在准备中。当前显示原文说明。

MetaX-MACA/vLLM-metax1842026年10月23日 更新

查看 MetaX-MACA 的全部技能

报告此技能的问题