Skip to content

Benchmarking script output field deduplication #34

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Benchmarking script output field deduplication #34

wants to merge 4 commits into from

Conversation

Bslabe123
Copy link
Collaborator

@Bslabe123 Bslabe123 commented Apr 10, 2025 •
edited
Loading

We record duplicate values for summary stats both under the metrics and summary_stats keys, this change removes the duplicates under the metrics key

Before:

{
  "metrics": {
    "num_prompts_attempted": 100,
    "num_prompts_succeeded": 100,
    "request_rate": 10,
    "server_metrics": {
      "vllm:cpu_cache_usage_perc": {},
      "vllm:gpu_cache_usage_perc": {},
      "vllm:num_requests_waiting": {},
      "vllm:num_requests_running": {},
      "vllm:num_requests_swapped": {},
      "vllm:time_to_first_token_seconds": {},
      "vllm:time_per_output_token_seconds": {},
      "vllm:e2e_request_latency_seconds": {},
      "vllm:request_prefill_time_seconds": {},
      "vllm:request_queue_time_seconds": {},
      "vllm:request_decode_time_seconds": {},
      "vllm:request_inference_time_seconds": {},
      "vllm:time_in_queue_requests": {},
      "vllm:request_prompt_tokens": {},
      "vllm:request_generation_tokens": {},
      "vllm:iteration_tokens_total": {},
      "vllm:prompt_tokens_total": {},
      "vllm:generation_tokens_total": {},
      "vllm:request_success_total": {},
      "vllm:num_preemptions_total": {},
      "vllm:cpu_prefix_cache_hit_rate": {},
      "vllm:gpu_prefix_cache_hit_rate": {},
      "vllm:avg_generation_throughput_toks_per_s": {},
      "vllm:avg_prompt_throughput_toks_per_s": {}
    },
    "benchmark_time": 87.15983057022095,
    "throughput_rps": 1.1473175125028987,
    "throughput": 283.3071133623408,
    "total_output_token": 24693,
    "total_input_tokens": 21631,
    "input_tokens_per_sec": 248.17625112950202,
    "total_tokens": 46324,
    "tokens_per_sec": 531.4833644918427,
    "avg_per_token_latency_ms": 65.62879936484664,
    "median_per_token_latency_ms": 79.78041821641335,
    "sd_per_token_latency_ms": 58.12467562553413,
    "min_per_token_latency_ms": 1.6009688112202982,
    "max_per_token_latency_ms": 482.1956672668457,
    "p90_per_token_latency_ms": 104.27726848783006,
    "p99_per_token_latency_ms": 149.28262528806292,
    "avg_TTFT_ms": 681.1766710996628,
    "median_TTFT_ms": 523.3574693202972,
    "sd_TTFT_ms": 678.7279214216226,
    "min_TTFT_ms": 78.12633419036865,
    "max_TTFT_ms": 3456.4309854507446,
    "p90_TTFT_ms": 1301.4110309600842,
    "p99_TTFT_ms": 3312.3639695787438,
    "avg_ITL_ms": 100.06214696606354,
    "median_ITL_ms": 87.2403130531311,
    "sd_ITL_ms": 278.0551728819678,
    "min_ITL_ms": 0.009039878845214844,
    "max_ITL_ms": 23677.23402738571,
    "p90_ITL_ms": 90.99695301055908,
    "p99_ITL_ms": 352.11324405670166,
    "avg_latency_ms": 25318.23270263672,
    "median_latency_ms": 18174.828491210938,
    "sd_latency_ms": 22968.93149507456,
    "min_latency_ms": 510.90283203125,
    "max_latency_ms": 81436.54663085938,
    "p90_latency_ms": 61525.34230957033,
    "p99_latency_ms": 75749.16485351566,
    "avg_normalized_time_per_output_token_ms": 135.875691689547,
    "median_normalized_time_per_output_token_ms": 110.48952206123582,
    "sd_normalized_time_per_output_token_ms": 96.0256469921684,
    "min_normalized_time_per_output_token_ms": 63.86285400390625,
    "max_normalized_time_per_output_token_ms": 795.3367020456415,
    "p90_normalized_time_per_output_token_ms": 174.08503954541553,
    "p99_normalized_time_per_output_token_ms": 718.4630758117904,
    "avg_input_len": 216.31,
    "median_input_len": 73,
    "sd_input_len": 268.59451576679675,
    "min_input_len": 4,
    "max_input_len": 975,
    "p90_input_len": 601.9000000000007,
    "p99_input_len": 974.01,
    "avg_output_len": 246.93,
    "median_output_len": 146,
    "sd_output_len": 255.60411792457492,
    "min_output_len": 4,
    "max_output_len": 995,
    "p90_output_len": 667.6,
    "p99_output_len": 884.1200000000006,
    "ClientConnectorError": 0,
    "TimeoutError": 0,
    "ContentTypeError": 0,
    "ClientOSError": 0,
    "ServerDisconnectedError": 0,
    "unknown_error": 0
  },
  "dimensions": {
    "date": "20250410-161611",
    "backend": "vllm",
    "model_id": "weighted",
    "tokenizer_id": "meta-llama/Llama-2-7b-hf"
  },
  "config": {
    "model": "weighted",
    "num_models": 1,
    "model_server": "vllm",
    "start_time": {
      "seconds": 1744301771,
      "nanos": 385791000
    }
  },
  "summary_stats": {
    "stats": [
      {
        "request_rate": 10,
        "request_latency": {
          "mean": 25318.23270263672,
          "median": 18174.828491210938,
          "sd": 22968.93149507456,
          "min": 510.90283203125,
          "max": 81436.54663085938,
          "p90": 61525.34230957033,
          "p99": 75749.16485351566
        },
        "throughput": {
          "mean": 283.3071133623408
        },
        "input_length": {
          "mean": 216.31,
          "median": 73,
          "sd": 268.59451576679675,
          "min": 4,
          "max": 975,
          "p90": 601.9000000000007,
          "p99": 974.01
        },
        "output_length": {
          "mean": 246.93,
          "median": 146,
          "sd": 255.60411792457492,
          "min": 4,
          "max": 995,
          "p90": 667.6,
          "p99": 884.1200000000006
        },
        "tpot": {
          "mean": 135.875691689547,
          "median": 110.48952206123582,
          "sd": 96.0256469921684,
          "min": 63.86285400390625,
          "max": 795.3367020456415,
          "p90": 174.08503954541553,
          "p99": 718.4630758117904
        },
     ...

After:

{
  "metrics": {
    "num_prompts_attempted": 100,
    "num_prompts_succeeded": 100,
    "request_rate": 10,
    "server_metrics": {
      "vllm:cpu_cache_usage_perc": {},
      "vllm:gpu_cache_usage_perc": {},
      "vllm:num_requests_waiting": {},
      "vllm:num_requests_running": {},
      "vllm:num_requests_swapped": {},
      "vllm:time_to_first_token_seconds": {},
      "vllm:time_per_output_token_seconds": {},
      "vllm:e2e_request_latency_seconds": {},
      "vllm:request_prefill_time_seconds": {},
      "vllm:request_queue_time_seconds": {},
      "vllm:request_decode_time_seconds": {},
      "vllm:request_inference_time_seconds": {},
      "vllm:time_in_queue_requests": {},
      "vllm:request_prompt_tokens": {},
      "vllm:request_generation_tokens": {},
      "vllm:iteration_tokens_total": {},
      "vllm:prompt_tokens_total": {},
      "vllm:generation_tokens_total": {},
      "vllm:request_success_total": {},
      "vllm:num_preemptions_total": {},
      "vllm:cpu_prefix_cache_hit_rate": {},
      "vllm:gpu_prefix_cache_hit_rate": {},
      "vllm:avg_generation_throughput_toks_per_s": {},
      "vllm:avg_prompt_throughput_toks_per_s": {}
    },
    "benchmark_time": 89.79840230941772,
    "throughput_rps": 1.1136055589879061,
    "throughput": 252.20938699958097,
    "total_output_token": 22648,
    "total_input_tokens": 24724,
    "input_tokens_per_sec": 275.3278384041699,
    "total_tokens": 47372,
    "tokens_per_sec": 527.5372254037509,
    "ClientConnectorError": 0,
    "TimeoutError": 0,
    "ContentTypeError": 0,
    "ClientOSError": 0,
    "ServerDisconnectedError": 0,
    "unknown_error": 0
  },
  "dimensions": {
    "date": "20250410-164545",
    "backend": "vllm",
    "model_id": "weighted",
    "tokenizer_id": "meta-llama/Llama-2-7b-hf"
  },
  "config": {
    "model": "weighted",
    "num_models": 1,
    "model_server": "vllm",
    "start_time": {
      "seconds": 1744303545,
      "nanos": 960754000
    }
  },
  "summary_stats": {
    "stats": [
      {
        "request_rate": 10,
        "request_latency": {
          "mean": 31747127.46826172,
          "median": 31635978.759765625,
          "sd": 21039342.704095166,
          "min": 1089250.9765625,
          "max": 80900479.00390625,
          "p90": 62639594.55566406,
          "p99": 80862957.37548828
        },
        "throughput": {
          "mean": 252.20938699958097
        },
        "input_length": {
          "mean": 247.24,
          "median": 119.5,
          "sd": 283.9288333368064,
          "min": 5,
          "max": 1012,
          "p90": 674.0000000000006,
          "p99": 955.5700000000003
        },
        "output_length": {
          "mean": 226.48,
          "median": 176,
          "sd": 206.0769021506292,
          "min": 4,
          "max": 921,
          "p90": 518.8000000000006,
          "p99": 691.3200000000012
        },
        "ttft": {
          "mean": 9933.80773329258,
          "median": 583.4563143253326,
          "sd": 14529.419121424398,
          "min": 101.06280565261841,
          "max": 37110.926535606384,
          "p90": 36094.4530960083,
          "p99": 37066.00822801113
        },
        "inter_token_latency": {
          "mean": 96.5978099885208,
          "median": 87.64755535125732,
          "sd": 149.29015189934833,
          "min": 0.008959770202636719,
          "max": 15574.367049217224,
          "p90": 90.9701452255249,
          "p99": 323.8559623718262
        },
        "tpot": {
          "mean": 422597.06265315035,
          "median": 121694.35466963978,
          "sd": 1089551.8412275861,
          "min": 87839.82519425217,
          "max": 8983535.034179688,
          "p90": 670745.2299854641,
          "p99": 4052263.4875488533
        },
        "per_token_latencies": {
          "mean": 111.2268114998901,
          "median": 89.54365993742783,
          "sd": 226.35747867726514,
          "min": 1.331602660834352,
          "max": 2226.0436150045957,
          "p90": 175.7089749850049,
          "p99": 533.3310897145078
        },
      ...

@Bslabe123 Bslabe123 requested a review from achandrasekar April 10, 2025 20:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@achandrasekar achandrasekar Awaiting requested review from achandrasekar

At least 1 approving review is required to merge this pull request.

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@Bslabe123