{
  "created_at_unix": 1781285877.5537736,
  "base_url": "http://127.0.0.1:18080",
  "model": "qwen36-35b-a3b-fp8",
  "tokenizer": "/mnt/fast-ai/llm-cache/hf/models--nameistoken--Qwen3.6-35B-A3B-Quark-W8A8-INT8/snapshots/cced56592e8c8935f8220836b4baa04dfd389118",
  "server_model_record": {
    "id": "qwen36-35b-a3b-fp8",
    "object": "model",
    "created": 1781285855,
    "owned_by": "vllm",
    "root": "/mnt/fast-ai/llm-cache/hf/models--nameistoken--Qwen3.6-35B-A3B-Quark-W8A8-INT8/snapshots/cced56592e8c8935f8220836b4baa04dfd389118",
    "parent": null,
    "max_model_len": 32768,
    "permission": [
      {
        "id": "modelperm-af5cd5df4adc7617",
        "object": "model_permission",
        "created": 1781285855,
        "allow_create_engine": false,
        "allow_sampling": true,
        "allow_logprobs": true,
        "allow_search_indices": false,
        "allow_view": true,
        "allow_fine_tuning": false,
        "organization": "*",
        "group": null,
        "is_blocking": false
      }
    ]
  },
  "prompt_kind": "vllm-random",
  "prompt_preset": "repetitive",
  "prompt_file": null,
  "endpoint": "completions",
  "seed": 20260612,
  "random_prefix_len": 0,
  "prompt_tokens_requested": 512,
  "prompt_tokens_actual": 512,
  "output_tokens_requested": 256,
  "mode": "stream",
  "ignore_eos": true,
  "skip_vram": true,
  "repeats": 1,
  "measurement_notes": [
    "TTFT and e2e are measured both client-side and from vLLM Prometheus histogram deltas.",
    "tok_s_prefill_lower_bound_from_ttft is prompt_tokens / TTFT, so it is conservative and includes first-token scheduling/decode overhead.",
    "tok_s_out_client_after_first_chunk is generated-token-only throughput after the first streamed text chunk.",
    "tok_s_out_client_after_first_chunk_corrected subtracts the estimated first streamed chunk tokens from the numerator; this is more stable when --stream-interval > 1."
  ],
  "vram_mib_before_all": {},
  "peak_vram_mib_observed": {},
  "records": [
    {
      "repeat": 1,
      "request_id": "cmpl-bff0c4ae0f448a18",
      "request_started_at_unix": 1781285874.6864047,
      "request_finished_at_unix": 1781285877.5484524,
      "prompt_tokens_client": 512,
      "prompt_tokens_estimated_before_request": 512,
      "output_tokens_client": 256,
      "elapsed_s_client": 2.8619739760179073,
      "ttft_ms_client": 77.89753400720656,
      "tok_s_out_client_after_first_chunk": 91.95149821932083,
      "tok_s_out_client_after_first_chunk_corrected": 91.5923126794016,
      "tok_s_out_client_e2e": 89.44875185629508,
      "tok_s_total_client": 268.3462555688852,
      "tok_s_prefill_lower_bound_from_ttft": 6572.73694893388,
      "vllm_metric_deltas": {
        "prompt_tokens": 512.0,
        "generation_tokens": 256.0,
        "ttft_count": 1.0,
        "ttft_sum_s": 0.07656240463256836,
        "e2e_count": 1.0,
        "e2e_sum_s": 2.860894203186035
      },
      "vllm_histogram_deltas": {
        "vllm:request_queue_time_seconds": {
          "count": 1.0,
          "sum": 1.0430114343762398e-05,
          "mean": 1.0430114343762398e-05
        },
        "vllm:request_prefill_time_seconds": {
          "count": 1.0,
          "sum": 0.0707689500413835,
          "mean": 0.0707689500413835
        },
        "vllm:request_decode_time_seconds": {
          "count": 1.0,
          "sum": 2.784385201986879,
          "mean": 2.784385201986879
        },
        "vllm:request_inference_time_seconds": {
          "count": 1.0,
          "sum": 2.8551541520282626,
          "mean": 2.8551541520282626
        },
        "vllm:request_time_per_output_token_seconds": {
          "count": 1.0,
          "sum": 0.010919157654850499,
          "mean": 0.010919157654850499
        },
        "vllm:inter_token_latency_seconds": {
          "count": 255.0,
          "sum": 2.784385201986879,
          "mean": 0.010919157654850506
        },
        "vllm:iteration_tokens_total": {
          "count": 256.0,
          "sum": 768.0,
          "mean": 3.0
        }
      },
      "ttft_ms_vllm_metrics": 76.56240463256836,
      "e2e_ms_vllm_metrics": 2860.894203186035,
      "queue_ms_vllm_histogram": 0.010430114343762398,
      "prefill_ms_vllm_histogram": 70.7689500413835,
      "decode_ms_vllm_histogram": 2784.385201986879,
      "decode_ms_per_generation_token_vllm_histogram": 10.876504695261247,
      "inference_ms_vllm_histogram": 2855.1541520282626,
      "time_per_output_token_ms_vllm_histogram": 10.9191576548505,
      "inter_token_ms_vllm_histogram": 10.919157654850506,
      "iteration_tokens_per_step_vllm_histogram": 3.0,
      "vram_mib_before": {},
      "vram_mib_after": {},
      "first_chunk_tokens_client_estimate": 1,
      "streamed_text_chunks": 256,
      "text_preview": " 10000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"
    }
  ],
  "summary": {
    "tok_s_out_client_after_first_chunk": {
      "mean": 91.95149821932083,
      "median": 91.95149821932083,
      "min": 91.95149821932083,
      "max": 91.95149821932083
    },
    "tok_s_out_client_after_first_chunk_corrected": {
      "mean": 91.5923126794016,
      "median": 91.5923126794016,
      "min": 91.5923126794016,
      "max": 91.5923126794016
    },
    "tok_s_out_client_e2e": {
      "mean": 89.44875185629508,
      "median": 89.44875185629508,
      "min": 89.44875185629508,
      "max": 89.44875185629508
    },
    "tok_s_total_client": {
      "mean": 268.3462555688852,
      "median": 268.3462555688852,
      "min": 268.3462555688852,
      "max": 268.3462555688852
    },
    "ttft_ms_client": {
      "mean": 77.89753400720656,
      "median": 77.89753400720656,
      "min": 77.89753400720656,
      "max": 77.89753400720656
    },
    "ttft_ms_vllm_metrics": {
      "mean": 76.56240463256836,
      "median": 76.56240463256836,
      "min": 76.56240463256836,
      "max": 76.56240463256836
    },
    "tok_s_prefill_lower_bound_from_ttft": {
      "mean": 6572.73694893388,
      "median": 6572.73694893388,
      "min": 6572.73694893388,
      "max": 6572.73694893388
    },
    "e2e_ms_vllm_metrics": {
      "mean": 2860.894203186035,
      "median": 2860.894203186035,
      "min": 2860.894203186035,
      "max": 2860.894203186035
    },
    "queue_ms_vllm_histogram": {
      "mean": 0.010430114343762398,
      "median": 0.010430114343762398,
      "min": 0.010430114343762398,
      "max": 0.010430114343762398
    },
    "prefill_ms_vllm_histogram": {
      "mean": 70.7689500413835,
      "median": 70.7689500413835,
      "min": 70.7689500413835,
      "max": 70.7689500413835
    },
    "decode_ms_vllm_histogram": {
      "mean": 2784.385201986879,
      "median": 2784.385201986879,
      "min": 2784.385201986879,
      "max": 2784.385201986879
    },
    "decode_ms_per_generation_token_vllm_histogram": {
      "mean": 10.876504695261247,
      "median": 10.876504695261247,
      "min": 10.876504695261247,
      "max": 10.876504695261247
    },
    "inference_ms_vllm_histogram": {
      "mean": 2855.1541520282626,
      "median": 2855.1541520282626,
      "min": 2855.1541520282626,
      "max": 2855.1541520282626
    },
    "time_per_output_token_ms_vllm_histogram": {
      "mean": 10.9191576548505,
      "median": 10.9191576548505,
      "min": 10.9191576548505,
      "max": 10.9191576548505
    },
    "inter_token_ms_vllm_histogram": {
      "mean": 10.919157654850506,
      "median": 10.919157654850506,
      "min": 10.919157654850506,
      "max": 10.919157654850506
    },
    "iteration_tokens_per_step_vllm_histogram": {
      "mean": 3.0,
      "median": 3.0,
      "min": 3.0,
      "max": 3.0
    }
  }
}