---
title: "Training Check - Air Market"
description: "A training run can break quietly and keep burning GPU hours doing nothing useful. This checks the metrics as it goes and flags the early signs - NaNs, loss going the wrong way, idle GPUs. Reach for i…"
canonical: "https://airmarket.ai/app/training-check"
---

[Air Market](/) / [Data & Analytics](/category/data) / Training Check

Skill by [Yang Ruofeng](/@wanshuiyin)

# Training Check

Notice the run went wrong while the GPUs are still worth saving

## What this does

A training run can break quietly and keep burning GPU hours doing nothing useful. This checks the metrics as it goes and flags the early signs - NaNs, loss going the wrong way, idle GPUs. Reach for it when you have a run going and do not want to keep checking it by hand.

### Top 3 things it does for you

- Watch a training run for problems while it goes
- Catch a diverging loss or a NaN early
- Find out when GPUs have gone idle

## Information

**Training Check at a glance**

|  |  |
| --- | --- |
| Kind | Skill |
| Publisher | [Yang Ruofeng](/@wanshuiyin) |
| Category | [Data & Analytics](/category/data) |
| Works with | Agent Skills - an open standard supported by 40+ agents |
| Needs | \[object Object\] |
| Size | 6 KB |
| Licence | MIT |
| Repository | [wanshuiyin/auto-claude-code-research-in-sleep](https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/tree/c5f3d5bfc694a812012729841e9697223e4f2130/skills/training-check) |
| GitHub stars | 15,082 (as of Nov 2026) |

## Security report

**Verified.** AIR scanned this exact version of Training Check and flagged nothing. AIR's verdict: **safe**.

**What Training Check's check was run against**

|  |  |
| --- | --- |
| Commit | c5f3d5bfc694a812012729841e9697223e4f2130 |
| Content hash | ad8170a8ba452d8f3479ae0506f004b6779a2e8f7feb3b82c93a36d7ee777be8 |
| Files read | [Browse the files at this exact commit](https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/c5f3d5bfc694a812012729841e9697223e4f2130/skills/training-check/SKILL.md) |

A scan is named by the exact content it read, not by a version number. A different version is a different artifact, and nothing here applies to a version AIR has not scanned.

[Read the full security report for Training Check](/security/training-check)

## About Training Check

The worst training failures are the quiet ones. The job keeps running, the GPUs keep billing, and the numbers went bad hours ago.

This checks WandB metrics periodically during training to catch problems early - NaNs, loss divergence, idle GPUs - so a broken run does not keep consuming GPU hours.

Use it while training is running and you want the health checks to happen on their own.

## What's inside

- SKILL.md - instructions

## Questions

### Is Training Check safe to install?

AIR scanned this exact version of Training Check and flagged nothing. AIR's verdict: safe. The verdict is named by the content that was read, at commit c5f3d5bfc694, so it cannot drift onto a version nobody scanned. The full report is at https://airmarket.ai/security/training-check.

### How do I install Training Check?

Open https://airmarket.ai/app/training-check on Air Market and use the install action: Air Market hands Claude a prompt naming this exact skill, and Claude sets it up. Source, pinned to the exact commit this listing describes: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/tree/c5f3d5bfc694a812012729841e9697223e4f2130/skills/training-check. Install from that link, not the branch. It needs \[object Object\] to be reachable before it can do anything.

## Related on Air Market

- [More Data & Analytics listings](/category/data)
- [Everything by Yang Ruofeng](/@wanshuiyin)

---

Canonical URL: <https://airmarket.ai/app/training-check>