meta-pytorch/data GitHub Actions scorecard

Public GitHub Actions data, last 30 days. Updated .

Data sourced from public GitHub. GitSpider is not affiliated with or endorsed by this repository's owners. Request removal.

5,293 min/mo
recoverable (~50% of CI time) · across 11 patterns · ≈$32/mo
Estimated from wall-clock time · public repos pay $0, $ = private-repo equivalent
See all 11 fixes, each with the exact YAML ↓
30.9%
failure rate, 30d
17h 37m
avg time to recover from a failure
Since : findings 10 → 11 · recoverable ~5,383 → ~5,293 min/mo (detector updated between scans, so part of this change is ours, not the repo's)
13 workflows · 94 runs (3.1/day) · 10,585 CI-min (wall-clock) · ≈$64 at private-repo rates (30d)

Where the minutes go (30d)

Cron job to validate TorchData Nightly Binaries~10,163 min · 30 runs
Build Linux Wheels~303 min · 30 runs
Push Nightly Release~114 min · 30 runs
or track on every push →

Waste detected

Biggest wins first, each with the exact config fix.

Workflow failing almost every run · Cron job to validate TorchData Nightly Binaries

~10162 min/mo · ≈$61/mo

Fails or times out on nearly all recent runs, burning minutes to produce only red, and an always-red workflow usually gets ignored. Fix it, or disable the trigger until it's ready.

Full guide: how to fix this →

No concurrency control · Cron job to validate TorchData Nightly Binaries

~1016 min/mo · ≈$6/mo

Add a `concurrency:` block keyed on branch to cancel superseded runs when devs push twice quickly.

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

Full guide: how to fix this →

No concurrency control · Build Linux Wheels

~30 min/mo

Add a `concurrency:` block keyed on branch to cancel superseded runs when devs push twice quickly.

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

Full guide: how to fix this →

No concurrency control · Push Nightly Release

~11 min/mo

Add a `concurrency:` block keyed on branch to cancel superseded runs when devs push twice quickly.

concurrency:
  group: ${{ github.workflow }}
  # no cancel-in-progress on deploys: queued deploys would be dropped

Full guide: how to fix this →

Also found: 7 more config fixes with negligible recoverable minutes: timeouts, retention, path filters

No job timeout · Push Nightly Release

~0 min/mo

No job sets `timeout-minutes`, so a hung step can run to GitHub's 6-hour default. Add `timeout-minutes` to each job.

jobs:
  build:
    runs-on: ubuntu-latest
    timeout-minutes: 15

Full guide: how to fix this →

Scheduled at the top of the hour · Push Nightly Release

~0 min/mo

The schedule fires at minute :00, GitHub's peak window, where scheduled runs get delayed or skipped. Shift to any other minute for the same cadence with less contention.

on:
  schedule:
    - cron: '6 11 * * *'  # was '00 11 * * *' — any non-:00 minute avoids the herd

Full guide: how to fix this →

No job timeout · Build Linux Wheels

~0 min/mo

No job sets `timeout-minutes`, so a hung step can run to GitHub's 6-hour default. Add `timeout-minutes` to each job.

jobs:
  build:
    runs-on: ubuntu-latest
    timeout-minutes: 15

Full guide: how to fix this →

No path filters on triggers · Build Linux Wheels

~0 min/mo

Runs on every push/PR with no `paths:` filter, so docs-only changes still trigger full CI. Add a `paths:` filter if that's common.

on:
  pull_request:
    paths:
      - 'src/**'
      - 'package.json'

Full guide: how to fix this →

Third-party actions pinned to a mutable ref · Build Linux Wheels

~0 min/mo

2 third-party actions pinned to a branch or other mutable ref, which can change under you (supply-chain risk, non-reproducible builds). Pin to a full commit SHA instead.

uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@<full-sha>  # was @main

No job timeout · Cron job to validate TorchData Nightly Binaries

~0 min/mo

No job sets `timeout-minutes`, so a hung step can run to GitHub's 6-hour default. Add `timeout-minutes` to each job.

jobs:
  build:
    runs-on: ubuntu-latest
    timeout-minutes: 15

Full guide: how to fix this →

Scheduled workflow runs in forks · Cron job to validate TorchData Nightly Binaries

~0 min/mo

The schedule has no repository guard, so forks that enable Actions inherit the cron and burn their own minutes. Gate the job with `if: github.repository == 'owner/repo'`.

jobs:
  nightly:
    if: github.repository == 'owner/repo'
    runs-on: ubuntu-latest

Full guide: how to fix this →

Copy every fix as one block (7 snippets, annotated per workflow)
# No concurrency control (applies to: Cron job to validate TorchData Nightly Binaries, Build Linux Wheels)
concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

# No concurrency control (applies to: Push Nightly Release)
concurrency:
  group: ${{ github.workflow }}
  # no cancel-in-progress on deploys: queued deploys would be dropped

# No job timeout (applies to: Push Nightly Release, Build Linux Wheels, Cron job to validate TorchData Nightly Binaries)
jobs:
  build:
    runs-on: ubuntu-latest
    timeout-minutes: 15

# Scheduled at the top of the hour (applies to: Push Nightly Release)
on:
  schedule:
    - cron: '6 11 * * *'  # was '00 11 * * *' — any non-:00 minute avoids the herd

# No path filters on triggers (applies to: Build Linux Wheels)
on:
  pull_request:
    paths:
      - 'src/**'
      - 'package.json'

# Third-party actions pinned to a mutable ref (applies to: Build Linux Wheels)
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@<full-sha>  # was @main

# Scheduled workflow runs in forks (applies to: Cron job to validate TorchData Nightly Binaries)
jobs:
  nightly:
    if: github.repository == 'owner/repo'
    runs-on: ubuntu-latest

Each snippet is representative. Merge into the named workflow files rather than pasting wholesale.

Add the badge to your README

Live CI-health badge → GitSpider badge

[![GitSpider](https://gitspider.com/badge/meta-pytorch/data.svg)](https://gitspider.com/scan/meta-pytorch/data)

Want this on every push?

This scorecard is a one-time snapshot. Install the free GitHub App to track this repo continuously: new regressions caught as they land, trends over time, on your public and private repos. Team adds the offending commit on the PR + Slack alerts.

Install & monitor this repo →

💬 On Team, this same analysis posts automatically to every PR: the regression, and the exact commit that caused it, right where your team already looks. See plans →

Not ready to install? Get this report by email. No spam, unsubscribe anytime.

Share this scorecard: https://gitspider.com/scan/meta-pytorch/data