←View full Fiddler AI status
Fiddler AI: Pipeline / Training Run Failed
Current Status: Operational
Last checked: 7m ago
What We're Seeing Right Now
No recent issues reported. If you're experiencing problems with Fiddler AI, report below to help the community.
What is this error?
Your ML pipeline or training run on Fiddler AI has failed unexpectedly. This can be caused by platform-side issues (compute unavailability, orchestration bugs) or configuration problems on your end. Distinguishing between the two is key to a fast resolution.
Error Signatures
Run failed with exit code 1OOMKilledOut of memoryPipeline execution errorStep timed outCompute not availableJob preemptedContainer failed to startCommon Causes
- Compute resources (GPU/CPU) temporarily unavailable on the platform
- Out-of-memory errors due to model size or batch size misconfiguration
- Dependency or environment conflicts in the runtime container
- Network timeout between pipeline steps or data sources
- Platform-side orchestration bugs causing premature job termination
✓ How to Fix It
- Check the run logs for the exact error message and step that failed
- Verify that the required compute tier is available on Fiddler AI's status page
- Reduce batch size or model size if you're hitting memory limits
- Check your data pipeline for connectivity issues to storage (S3, GCS, etc.)
- Try re-running with a smaller test dataset to isolate the issue
- Open a support ticket with the run ID if the error appears platform-side
Live Signals
Service Components
Fiddler AI Web
OperationalRecent Incidents
No incidents in the past 30 days
Frequently Asked Questions
How do I tell if the pipeline failure is my fault or Fiddler AI's?
Check the error logs carefully. OOM errors and code exceptions are usually your issue. 'Compute not available', 'job preempted', or infrastructure-level errors point to the platform. Check if other users are reporting similar failures on this page.
Will I be charged for a failed pipeline run?
Most MLOps platforms bill for compute time used, even on failed runs. If the failure is clearly platform-side, contact Fiddler AI support to request a credit.
How do I prevent pipeline failures from losing hours of training progress?
Enable checkpointing in your training code (most frameworks support this natively). Configure your pipeline to checkpoint every N steps so a failure only loses a fraction of progress.