Once the endpoint and miner are setup and running, benchmark your setup:

python tests/benchmark_sdxl.py \\
--max_concurrent_requests <your-max-concurrent-requests> \\ # should equal to your miner.max_concurrent_requests
--n_times <number-of-iterations> \\ # n_times * max_concurrent_requests should be less than or equal to your miner.total_volume
--model_name <selected-model-name>

• This script will run the miner with the specified number of concurrent requests and measure the average latency and throughput.

• Console Print:

  • report (dict): A dictionary with keys are status_code. Values are list of latencies for each request. Example: {200: [0.1, 0.2, 0.3], 408: [12, 12, 12]}

• Plot latency histogram:

  • x-axis: latency in seconds
  • y-axis: number of requests

• Example Plot

  • [with ControlNet] RealitiesEdgeXL model with 3 concurrent requests and 100 iterations

  

  • [without ControlNet] RealitiesEdgeXL model with 3 concurrent requests and 100 iterations