{
  "kind": "native_mps_benchmark_results",
  "date_pdt": "2026-10-29",
  "gpu": "NVIDIA A100-SXM4-40GB",
  "model": "Qwen/Qwen3-1.7B-FP8",
  "configuration": {
    "tenants": 2,
    "engine_gib_each": 18,
    "vllm_target_gib_each": 16.2,
    "mps_compute_percent_each": 100,
    "max_num_batched_tokens": 8192,
    "max_num_seqs": 64,
    "offered_window_s": 120,
    "output_tokens_per_request": 256
  },
  "results": {
    "12.0_req_s": {
      "offered": 1440,
      "completed": 1440,
      "within_latency_targets_percent": 86.96,
      "completed_output_tok_s": 2627.58,
      "slo_qualified_output_tok_s": 2627.58,
      "latency": {
        "ttft_p50_s": 0.0741,
        "ttft_p95_s": 0.08,
        "ttft_p99_s": 0.0812,
        "tpot_p99_s": 0.01551
      },
      "gpu_activity_percent": {
        "kernel_busy_mean": 86.26,
        "memory_busy_mean": 64.09,
        "sm_active_mean": 78.78,
        "tensor_active_mean": 23.57,
        "dram_active_mean": 49.48
      }
    },
    "16.0_req_s": {
      "offered": 1920,
      "completed": 1920,
      "within_latency_targets_percent": 31.7,
      "completed_output_tok_s": 3325.05,
      "slo_qualified_output_tok_s": 1212.26,
      "latency": {
        "ttft_p50_s": 1.898,
        "ttft_p95_s": 5.814,
        "ttft_p99_s": 6.104,
        "tpot_p99_s": 0.03889
      },
      "gpu_activity_percent": {
        "kernel_busy_mean": 86.26,
        "memory_busy_mean": 69.04,
        "sm_active_mean": 82.78,
        "tensor_active_mean": 30.09,
        "dram_active_mean": 53.74
      }
    }
  }
}