components compute_performance_metrics

Compute Performance Metrics

compute_performance_metrics

Overview

Performs performance metric post processing using data from a model inference run.

Version: 0.0.11

View in Studio: https://ml.azure.com/registries/azureml/components/compute_performance_metrics/version/0.0.11

Inputs

Name	Description	Type	Default	Optional	Enum
performance_data	Data outputted by model inferencing that contains performance data.	uri_folder		False
percentiles	Comma-separated list of percentiles of latency to be calculated.	string	50,90,99	True
batch_size_column_name	The name of the column that contains the batch size information. Ex. "batch_size"	string		False
start_time_column_name	The name of the column that contains the start timestamp in ISO 8601 format. Ex. "start_time_iso"	string		False
end_time_column_name	The name of the column that contains the end timestamp in ISO 8601 format. Ex. "end_time_iso"	string		False
input_token_count_column_name	The name of the column that contains the input token count information. Ex. "input_token_count"	string		True
output_token_count_column_name	The name of the column that contains the output token count information. Ex. "output_token_count"	string		True
input_char_count_column_name	The name of the column that contains the input character count information. Ex. "input_char_count"	string		True
output_char_count_column_name	The name of the column that contains the output character count information. Ex. "output_char_count"	string		True
is_batch_inference_result	If True, we will use the time between the first and last request to calculate the tokens per second and request per second. If False, we will use individual request time to calculate the tokens per second and request per second.	boolean	True	False

Outputs

Name	Description	Type
performance_result	Path to the file where the calculated performance metric results will be stored.	uri_file

Environment

azureml://registries/azureml/environments/model-evaluation/labels/latest