We are using tritonserver:25.05 version and since it doesn't support model-analyzer, we would like to switch to performance analyzer.
The question is: what would be the most efficient way to test model performance (latency, throughput, GPU, CPU utilization) for different input shapes and batch sizes?
Also out of curiosity, why did you decide to stop supporting model analyzer? The framework was amazing.
We are using
tritonserver:25.05version and since it doesn't support model-analyzer, we would like to switch to performance analyzer.The question is: what would be the most efficient way to test model performance (latency, throughput, GPU, CPU utilization) for different input shapes and batch sizes?
Also out of curiosity, why did you decide to stop supporting model analyzer? The framework was amazing.