Files
vllm-omni/tests/dfx/perf/scripts/diffusion_result_template.json
T

90 lines
2.0 KiB
JSON

[
{
"test_name": null,
"endpoint": null,
"timestamp": null,
"server_params": {
"model": null,
"serve_args": {
"enable-diffusion-pipeline-profiler": false
}
},
"benchmark_params": {
"name": null,
"dataset": null,
"task": null,
"width": 0,
"height": 0,
"num-inference-steps": 0,
"num-prompts": 0,
"max-concurrency": 0,
"num-input-images": 0,
"enable-negative-prompt": false,
"baseline": {
"H100": {
"throughput_qps": 0,
"latency_mean": 0,
"peak_memory_mb_max": 0,
"peak_memory_mb_mean": 0
}
}
},
"result": {
"duration": 0,
"completed_requests": 0,
"failed_requests": 0,
"throughput_qps": 0,
"latency_mean": 0,
"latency_median": 0,
"latency_p99": 0,
"latency_p95": 0,
"latency_p50": 0,
"peak_memory_mb_max": 0,
"peak_memory_mb_mean": 0,
"peak_memory_mb_median": 0,
"stage_durations_mean": {},
"stage_durations_p50": {},
"stage_durations_p99": {},
"endpoint": null,
"model": null,
"dataset": null,
"task": null
},
"log_file": null,
"Model": null,
"Framework": null,
"API Endpoint": null,
"Hardware": null,
"Deployment": null,
"Task": null,
"Dataset": null,
"resolution": null,
"Parallelism": null,
"max_concurrency": 0,
"Cache": null,
"Quantization": null,
"offload": null,
"compile": null,
"Attn_backend": null,
"num_inference_steps": 0,
"completed": 0,
"failed": 0,
"throughput_qps": 0,
"latency_mean": 0,
"latency_median": 0,
"latency_p99": 0,
"latency_p95": 0,
"latency_p50": 0,
"peak_memory_mb_max": 0,
"peak_memory_mb_mean": 0,
"peak_memory_mb_median": 0,
"stage_durations_mean": {},
"stage_durations_p50": {},
"stage_durations_p99": {},
"commit_sha": null,
"build_id": null,
"build_url": null,
"source_file": null
}
]