targetLatencyinteger required
trainingCyclesinteger
Maximum number of training cycles
tuningWorkersinteger
Maximum number of parallel workers/jobs
optimizationRoundsinteger
Number of optimization rounds
trialsPerOptimizationRoundinteger
Number of trials per optimization round
tuningAlgorithm'random' | 'hyperband' | 'bayesian' | 'custom'
Tuning algorithm to use to search hyperparameter space
notificationOnCompletionboolean
importProjectMetricsboolean
Whether to import metrics for previous EON tuner runs in the same project to accelerate the hyperparameter search process
importResourceMetricsboolean
Whether to import resource usage (RAM/ROM/latency) metrics to accelerate the hyperparameter search process
numImportProjectMetricsnumber
Number of project trials to import
numImportResourceMetricsnumber
Number of resource usage trials to import
enableSEMboolean
Enable standard error of the mean (SEM)
accuracySEMnumber
Standard error of the trial accuracy mean
latencySEMnumber
Standard error of the trial latency mean
rawObjectivesstring
Hyperparameter optimization objectives + weights in string format
optimizationPrecision'float32' | 'int8'
Model variant to optimize for
earlyStoppingboolean
Enable trial level early stopping based on loss metrics during training
earlyStoppingWindowSizenumber
Stops the EON tuner if the feasible (mean) objective has not improved over the past “window_size” iterations
earlyStoppingImprovementBarnumber
Threshold (in [0,1]) for considering relative improvement over the best point.
MOMFboolean
Enable Multi-fidelity Multi-Objective optimization
disableConstraintsboolean
Disable search constraints
disableDeduplicateboolean
Disable trial deduplication
maxTotalTrainingTimenumber
Maximum total training time in seconds
tunerChildJobMemoryMiBinteger
Optional memory override for tuner child jobs (MiB), capped at 65536 MiB.