Every premium and members-only resource for $9.99/month, as a no-commitment subscription. See the plans

PalateKit

Sign in
Menu

Request a resource

Describe the resource you are missing in one sentence. We will email you when it is published.

0 / 280

A page, a video or an image online.

Unlock PalateKit

Unlock everything

Plan

Paying means accepting our terms of sale.

Secure payment by Stripe. Receipts and invoices sent by email.

Welcome back

Sign in to install, download or save this resource. A free account is enough for free resources.

Diagnosing TPU multi-slice training hangs on GKE v1.0.0

Use this resource to diagnose a frozen TPU multi-slice training job on GKE

New

  • 316 downloads
  • 145 likes
  • 128 saves

Free

Install with the CLI

In a terminal, from the folder you want:

npx --yes palatekit@latest install gke-tpu-multislice-hang-diagnosis

With no option, the skill goes to ~/.agents/skills/; add --agent claude-code (or codex, cursor…) to target an agent.

First time: connect the CLI with

npx --yes palatekit@latest login

The agent correlates HANG_DETECTED logs and MXLA analyzer events with one-minute latency metrics and node topology. It routes the diagnosis to compiler divergence, a data input stall, or a chip, interconnect or network fault. You get the likely cause and the next checks.

Mostly picked by

Updated November 2, 2026

Card for the “Diagnosing TPU multi-slice training hangs on GKE” skill

Details

Works in
Install guides
Generates with
Your agent's model
Licence
Apache-2.0

Similar resources

View all

What you get

  • SKILL.md: context collection, hang queries and correlation with metrics
  • references/path-a-compiler-hlo-divergence.md: HLO compiler divergence
  • references/path-b-program-queueing-input-stall.md: program queueing and input stalls

Good fit for

  • A multi-slice TPU training job freezes without advancing its steps
  • You need to tell a compilation problem apart from a hardware fault

Not for

    Tags

    • tpu
    • gke
    • training
    • diagnostics

    About

    You say

    “My multi-slice TPU job has been frozen for an hour with HANG_DETECTED events: find the cause.”

    Card for the “Diagnosing TPU multi-slice training hangs on GKE” skill

    Skill published by google under the Apache-2.0 license, delivered in its original version.

    Show more

    How to use it

    1. Install the skill in your agent's skills folder
    2. Ask your agent to diagnose a frozen TPU multi-slice training job on GKE
    3. Review and adjust the result

    Content preview

    Files

    • LICENSE11.3 KB
    • NOTICE-PALATEKIT.md0.5 KB
    • references/attribution.md1.9 KB
    • references/path-a-compiler-hlo-divergence.md1 KB
    • references/path-b-program-queueing-input-stall.md1.4 KB
    • references/path-c-hardware-network-faults.md2.5 KB
    • references/path-d-no-detected-or-unknown.md1.5 KB
    • references/path-e-healthy-telemetry.md0.3 KB
    • SKILL.md15.2 KB

    Origin and licence

    Listed unmodified

    Author
    google
    Source
    github.com/google/skills/tree/81b0e8f8f40513c31b3388e600506e8b48a85bc2/skills/cloud/gke-ai-troubleshooting-tpu-mxla-hang
    Licence
    Apache-2.0

    Notice

    « gke-ai-troubleshooting-tpu-mxla-hang » (google, Apache-2.0) livré dans sa version d'origine : fiche en français et en anglais par PalateKit.

    User reviews

    –

    No ratings yet

    Loading reviews…

    Delete your review?

    Your rating and comment will be removed from the page.