genai_perf_python_api_llm_endpoint_profiling_quickstart.py

1import genai_perf.main as genai_perf
2from genai_perf.wrapper import Wrapper
3
4def main():
5    # 1. Configure the GenAi-Perf arguments
6    # This mimics the command line arguments used in the CLI version
7    args = [
8        "genai-perf",
9        "-m", "ensemble",              # Model name
10        "--service-kind", "triton",     # Service type (triton, openai, etc.)
11        "--endpoint", "v1/completions", # API endpoint
12        "--streaming",                  # Enable streaming mode
13        "--concurrency", "1",           # Number of concurrent requests
14        "--measurement-interval", "10000" # Time in ms to gather samples
15    ]
16
17    # 2. Run GenAi-Perf
18    # This will generate the profile data and output files (e.g., profile_export_genai_perf.csv)
19    try:
20        genai_perf.run(args)
21        print("GenAi-Perf run completed successfully.")
22    except Exception as e:
23        print(f"GenAi-Perf run failed: {e}")
24
25if __name__ == "__main__":
26    main()