misc: update output file logic

[zhyncs]

Motivation

as titled cc @merrymercy @Ying1123 @hnyls2002

1. save output file when single benchmark and multi benchmarks
2. refine the output file name

Modification

as titled

Checklist

1. Ensure pre-commit pre-commit run --all-files or other linting tools are used to fix potential lint issues.
2. Confirm that modifications are covered by complete unit tests. If not, please add more unit tests for correctness.
3. Modify documentation as needed, such as docstrings or example tutorials.

[zhyncs]

example 1

python3 -m sglang.bench_serving --backend sglang --num-prompts 64

result

# sglang_0721_64_sharegpt.jsonl
{"backend": "sglang", "dataset_name": "sharegpt", "request_rate": Infinity, "total_input": 14688, "total_output": 12794, "median_ttft": 899.2171995341778, "median_itl": 20.11713619122408, "output_token_throughput": 1114.0553206688626, "sharegpt_output_len": null, "random_input_len": 1024, "random_output_len": 128, "random_range_ratio": 1.0}

example 2

python3 -m sglang.bench_serving --backend sglang --request-rate-range 1,2 --multi --num-prompts 64

result

# sglang_0721_64_sharegpt.jsonl
{"backend": "sglang", "dataset_name": "sharegpt", "request_rate": Infinity, "total_input": 14688, "total_output": 12794, "median_ttft": 899.2171995341778, "median_itl": 20.11713619122408, "output_token_throughput": 1114.0553206688626, "sharegpt_output_len": null, "random_input_len": 1024, "random_output_len": 128, "random_range_ratio": 1.0}
{"backend": "sglang", "dataset_name": "sharegpt", "request_rate": 1, "total_input": 14688, "total_output": 12785, "median_ttft": 34.721650183200836, "median_itl": 12.611075227676041, "output_token_throughput": 187.63401440175718, "sharegpt_output_len": null, "random_input_len": 1024, "random_output_len": 128, "random_range_ratio": 1.0}
{"backend": "sglang", "dataset_name": "sharegpt", "request_rate": 2, "total_input": 14688, "total_output": 12789, "median_ttft": 36.969633772969246, "median_itl": 13.274253509486124, "output_token_throughput": 344.3559774625713, "sharegpt_output_len": null, "random_input_len": 1024, "random_output_len": 128, "random_range_ratio": 1.0}

example 3

python3 -m sglang.bench_serving --backend sglang --dataset-name random --num-prompts 64 --random-input 32 --random-output 64 --random-range-ratio 0.5

result

# sglang_0721_64_32_64.jsonl
{"backend": "sglang", "dataset_name": "random", "request_rate": Infinity, "total_input": 1476, "total_output": 3006, "median_ttft": 420.9714438766241, "median_itl": 25.011468476574922, "output_token_throughput": 2131.062314178063, "sharegpt_output_len": null, "random_input_len": 32, "random_output_len": 64, "random_range_ratio": 0.5}

example 4

python3 -m sglang.bench_serving --backend sglang --dataset-name random --num-prompts 64 --random-input 32 --random-output 64 --random-range-ratio 0.5 --request-rate-range 1,2 --multi

result

# sglang_0721_64_32_64.jsonl
{"backend": "sglang", "dataset_name": "random", "request_rate": 1, "total_input": 1476, "total_output": 3006, "median_ttft": 24.624092504382133, "median_itl": 12.247147382089443, "output_token_throughput": 49.562472284844425, "sharegpt_output_len": null, "random_input_len": 32, "random_output_len": 64, "random_range_ratio": 0.5}
{"backend": "sglang", "dataset_name": "random", "request_rate": 2, "total_input": 1476, "total_output": 3006, "median_ttft": 24.88376945257187, "median_itl": 12.330721587577608, "output_token_throughput": 81.97581017691886, "sharegpt_output_len": null, "random_input_len": 32, "random_output_len": 64, "random_range_ratio": 0.5}