trainer
https://github.com/kubeflow/trainer
Go
Distributed AI Model Training and LLM Fine-Tuning on Kubernetes
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Go not yet supported1 Subscribers
Add a CodeTriage badge to trainer
Help out
- Issues
- KEP-3562: [OptimizationJob] Validate Grid search against the search space and trial budget
- MPI runLauncherAsNode allocates an extra worker when numNodes is 1
- fix(framework): prevent initializer STORAGE_URI env from overriding storageUri
- fix(framework): make GPU count detection deterministic and unit-aware
- fix(framework): stop Coscheduling Build from mutating PodSet requests
- ML plugins size processes from the raw TrainJob resourcesPerNode, so a partial override hides the runtime's GPUs
- spec.trainer.resourcesPerNode is silently dropped when the runtime has no mlPolicy
- fix(runtimes): validate runtimePatches against the unpatched runtime template
- S3 initializers default the region to "auto", so s3:// URIs against AWS fail unless a region is set explicitly
- KEP-2899: [Phase 2] Implement TTLSecondsAfterFinished and RunPolicy runtime defaults
- Docs
- Go not yet supported