What you would like to be added?
Context:
Phase 1 of OptimizationJob in KEP-3562 restricts Objectives to exactly 1 metric (MinItems=1, MaxItems=1).
Problem:
Real-world ML problems often require balancing trade-offs between competing objectives (e.g., maximizing accuracy while minimizing model latency or memory usage).
Future Goals:
-
Update CEL validation on OptimizationJobSpec.objectives to allow multiple metrics.
-
Extend Optuna gRPC adapter for multi-objective optimization.
-
Update OptimizationJobStatus to track multi-metric optimal results.
Relates to KEP-3562 Master Tracking Issue.
Why is this needed?
Multi-objective support allows users to optimize production trade-offs (e.g., model size vs. throughput) directly within a single OptimizationJob.
Love this feature?
Give it a 👍 We prioritize the features with most 👍
What you would like to be added?
Context:
Phase 1 of OptimizationJob in KEP-3562 restricts Objectives to exactly 1 metric (MinItems=1, MaxItems=1).
Problem:
Real-world ML problems often require balancing trade-offs between competing objectives (e.g., maximizing accuracy while minimizing model latency or memory usage).
Future Goals:
Update CEL validation on
OptimizationJobSpec.objectivesto allow multiple metrics.Extend
OptunagRPC adapter for multi-objective optimization.Update
OptimizationJobStatusto track multi-metric optimal results.Relates to KEP-3562 Master Tracking Issue.
Why is this needed?
Multi-objective support allows users to optimize production trade-offs (e.g., model size vs. throughput) directly within a single OptimizationJob.
Love this feature?
Give it a 👍 We prioritize the features with most 👍