Reporting an issue

When the steps in Diagnosis and log reference do not resolve a problem, report it. This page covers when to report and what to include, so that the report can be acted on promptly.

When to report

Report the issue if any of the following applies:

• The same symptom recurs after applying the first-line action
• One of the errors marked to report in Error codes occurs, for example NP-X-999 (unhandled exception), NP-SRN-500 (pipeline stage failure, repeated), or NP-SSM-503 (GPU unavailable)
• The result is outside the normal range described in the following section
• The cause cannot be identified, or the whole service is affected

Normal and abnormal boundary

Aspect	Acceptable range (self-service)	Report it
Accuracy	Within allowed tolerance vs. expected model	Sharp accuracy drop or meaningless output
Pipeline	Completes after a configuration fix	Still stops at the same stage after np run --retry
Service	A single command fails transiently, then recovers	Persistent API or service outage

These are general guidelines. Model-specific and device-specific tolerances follow the agreed delivery criteria.

What to include

Include all of the following information. Missing information delays root cause analysis and the response.

• Time of occurrence, including the time zone
• Impact: whether you are blocked, have a workaround, or have a question
• Environment: np CLI version, OS, and OS version
• Command run, or run.yaml (and run.detail.yaml if used)
• Full error message, including the error code, with the log file attached
• Reproducibility result (reproduced or not reproduced)
• Actions already attempted

Logs are written automatically to the experiment's logs/ folder. See Log reference for what to attach.

Report template

[Time]: YYYY-MM-DD HH:MM (time zone)
[Impact]: blocked / workaround available / question
[Environment]: np CLI version, OS / OS version
[Symptom summary]:
[Error code / logs]: NP-***-*** (log file attached)
[Reproducible]: Y / N
[Actions attempted]:

If the issue turns out to be on the server side, your operator adds the deployment details on their side. You do not need to collect them.

Where to report

Report through the support channel agreed for this delivery. If you are not sure which channel that is, ask your project contact.

Your report is triaged and routed to the right engineer, so you do not need to identify the team yourself. Support hours are 09:00 to 18:00 KST, Monday to Friday.

After the issue is resolved

• Normal operation is restored and verified
• What fixed it is recorded (workaround or fix applied) for next time
• Any version or configuration change provided by support is noted

If the same symptom returns, report it again with the new logs and reference the previous report.