Skip to main content
Max PlatformDocsCompanyBlogModCon
Max PlatformDocsCompanyBlogModCon
Contact salesSign in

max benchmark

Benchmarks a model with MAX Engine.

Synopsis​

max benchmark [options] <model path>

Description​

Benchmarks a user-provided model with MAX Engine running under mlperf-loadgen. Inputs are randomly-generated by default but can be overridden with an input generation schema. For more detail,see the guide to benchmark a model.

Options​

Input data options​

--input-data-schema <PATH>​

Path to input data schema file.

Benchmarking options​

--mlperf-scenario <SCENARIO>​

MLPerf Inference scenario to test, out of:

  • single-stream
  • multiple-stream-2.0
  • server
  • offline

See https://mlcommons.org/en/inference-datacenter-30/#scenarios-and-metrics for definitions.

Compilation options​

--custom-ops-path <PATH>​

Path to custom ops.

Execution options​

--num-threads <N>​

Number of threads to use.

Common options​

--help, -h​

Displays help information.

--version​

Displays version information.