After you install the Python agent for your large language model (LLM) application, ARMS begins monitoring it. The Performance Analysis page displays key metrics such as model invocations, average model invocation time, and model invocation errors.
Prerequisites
You have installed an agent for your LLM application. For more information, see Connect LLM applications or inference services to ARMS.
View performance analysis
Log on to the ARMS console. In the left-side navigation pane, choose .
On the Application List page, select the region and click the name of your application.
In the top navigation bar, click Performance Analysis.
Panel
Description
Model invocations
Total model invocations within the selected time period.
Average model invocation time
Average time per model invocation during the selected time period.
Model invocation errors
Total failed model invocations within the selected time period.
Model invocations/1m
The number of model invocations per minute.
Average model invocation time/1m
The average model invocation time per minute.
Model invocation errors/1m
The number of failed model invocations per minute.
Model invocation time quantile (P99)/1m
The P99 quantile for model invocation time per minute. This means 99% of invocations were faster than this value.
Average time to first packet/1m
The average time to receive the first data packet from a model invocation, per minute.
Time to first packet quantile (P99)/1m
The P99 quantile for the time to first packet (TTFP) from model invocations per minute.
Top 5 models by invocation count
The top 5 models with the most invocations, in descending order.
Top 5 models by average invocation time
The top 5 models with the longest average invocation time, in descending order.
Top 5 models by invocation errors
The top 5 models with the most invocation errors, in descending order.