pytorch-runners-routing

Resolve PyTorch CI runner labels to final infrastructure and fleet.

17|14|Updated Jun 15, 2023
One-click install
npx skills add https://github.com/pytorch/ci-infra --skill pytorch-runners-routing
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pytorch-runners-routing
Source: https://github.com/pytorch/ci-infra/tree/main/osdc/.claude/skills/pytorch-runners-routing
Command: npx skills add https://github.com/pytorch/ci-infra --skill pytorch-runners-routing

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill resolves how PyTorch CI jobs map from workflow runner labels to their actual runner fleet and infrastructure, preventing misrouting between legacy EC2 autoscaling and current ARC on EKS.

Core Features & Use Cases

  • Runner Resolution: Trace runs-on values, test matrices, fleet prefixes, and reusable workflow outputs to determine the runner a job will use.
  • Infrastructure Identification: Distinguish Meta OSDC ARC, Linux Foundation OSDC ARC, legacy ALI EC2, and identity-passthrough partner runners.
  • Routing Diagnostics: Analyze experiment rollouts, EC2-to-ARC translation, Meta-only H100 and B200 overrides, build-test decoupling, and missing mapping failures.
  • Use Case: When a PyTorch GPU job is queued unexpectedly, use this Skill to determine its final label, owning infrastructure, experiment settings, and likely capacity or mapping issue.

Quick Start

Use the PyTorch runners routing skill to trace the workflow and test matrix for the specified job, then report its final runner label, infrastructure, fleet owner, and any routing risks.

Frequently Asked Questions about pytorch-runners-routing

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I trace a PyTorch CI job to its actual runner infrastructure?

Distinguish Meta OSDC ARC from Linux Foundation OSDC ARC and legacy ALI EC2 by applying strict arc.yaml mappings and prefix semantics to the job's runs-on values, identifying the owning fleet and any Meta-only hardware overrides.

Why does my PyTorch GPU job route to an unexpected runner fleet?

Diagnose routing risks by analyzing determinator experiments, build-test decoupling, and identity-passthrough partner runners to detect missing mappings or capacity issues affecting the final runner placement.

When do I need to analyze runner labels for PyTorch CI infrastructure diagnosis?

You need runner label analysis to resolve test matrices and reusable build workflows, verifying that prefix semantics and strict arc.yaml mappings align with the correct runner fleet and experiment settings.

Does PyTorch CI runner routing support Meta-only hardware overrides for H100 and B200?

The routing model supports Meta-only hardware overrides by translating runs-on values through arc.yaml mappings, ensuring H100 and B200 jobs bypass legacy EC2 autoscaling and route to the appropriate ARC fleet.

What causes missing mapping failures in PyTorch CI runner routing?

These failures happen when prefix semantics or experiment opt-outs prevent proper EC2-to-ARC translation, leaving the job without a valid runner fleet assignment or fallback behavior.