triton-inference-server / model_analyzer

Triton Model Analyzer is a CLI tool to help with better understanding of the compute and memory requirements of the Triton Inference Server models.
Apache License 2.0
423 stars 74 forks source link

Optuna Search Mode (Alpha) Release #896

Closed nv-braf closed 3 months ago

nv-braf commented 3 months ago

This is the Alpha release of Optuna Search Mode.

This mode has the following limitations:

This mode uses a hyperparameter optimization framework to search the configuration space, looking for the maximal objective value within the specified constraints. Please see the Optuna website if you are interested in specific details on how the algorithm functions.

Optuna allows you to search for every parameter that can be specified in the model configuration. Parameters can be specified with a min/max range (using the run-config-search options) or a list of parameters to test against can be set in the parameters/model_config_parameters field.

After Optuna search has found the best config(s), it will then sweep the top-N configurations found (specified by --num-configs-per-model) over the default concurrency range before generation of the summary reports.