{
  "schema": "MAKNOW_FLR014_LAB_FIXTURES_v1",
  "cases": [
    {
      "id": "baseline_safe",
      "model_footprint_gib": 8.0,
      "request_tokens": 2048,
      "concurrency": 2,
      "fb_total_gib": 24.0,
      "fb_used_gib": 12.0,
      "gpu_util_pct": 62,
      "allocation_result": "SUCCESS",
      "queue_depth": 1,
      "latency_ms": 72,
      "latency_slo_ms": 120,
      "throughput_rps": 31,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 2048,
      "post_concurrency": 2,
      "post_fb_used_gib": 12.0,
      "post_gpu_util_pct": 62,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 1,
      "post_latency_ms": 72,
      "post_throughput_rps": 31
    },
    {
      "id": "high_util_negative_control",
      "model_footprint_gib": 8.0,
      "request_tokens": 2048,
      "concurrency": 4,
      "fb_total_gib": 24.0,
      "fb_used_gib": 15.0,
      "gpu_util_pct": 99,
      "allocation_result": "SUCCESS",
      "queue_depth": 2,
      "latency_ms": 95,
      "latency_slo_ms": 120,
      "throughput_rps": 28,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 2048,
      "post_concurrency": 4,
      "post_fb_used_gib": 15.0,
      "post_gpu_util_pct": 99,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 2,
      "post_latency_ms": 95,
      "post_throughput_rps": 28
    },
    {
      "id": "concurrency_oom_uncontained",
      "model_footprint_gib": 8.0,
      "request_tokens": 2048,
      "concurrency": 16,
      "fb_total_gib": 24.0,
      "fb_used_gib": 23.8,
      "gpu_util_pct": 98,
      "allocation_result": "OOM",
      "queue_depth": 18,
      "latency_ms": 420,
      "latency_slo_ms": 120,
      "throughput_rps": 10,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 2048,
      "post_concurrency": 12,
      "post_fb_used_gib": 23.6,
      "post_gpu_util_pct": 97,
      "post_allocation_result": "OOM",
      "post_queue_depth": 12,
      "post_latency_ms": 330,
      "post_throughput_rps": 14
    },
    {
      "id": "concurrency_oom_closed",
      "model_footprint_gib": 8.0,
      "request_tokens": 2048,
      "concurrency": 16,
      "fb_total_gib": 24.0,
      "fb_used_gib": 23.8,
      "gpu_util_pct": 98,
      "allocation_result": "OOM",
      "queue_depth": 18,
      "latency_ms": 420,
      "latency_slo_ms": 120,
      "throughput_rps": 10,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 2048,
      "post_concurrency": 4,
      "post_fb_used_gib": 16.5,
      "post_gpu_util_pct": 88,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 2,
      "post_latency_ms": 96,
      "post_throughput_rps": 27
    },
    {
      "id": "oversized_request_transfer",
      "model_footprint_gib": 8.0,
      "request_tokens": 20000,
      "concurrency": 1,
      "fb_total_gib": 24.0,
      "fb_used_gib": 23.9,
      "gpu_util_pct": 76,
      "allocation_result": "OOM",
      "queue_depth": 1,
      "latency_ms": 390,
      "latency_slo_ms": 120,
      "throughput_rps": 3,
      "minimum_throughput_rps": 10,
      "post_request_tokens": 6000,
      "post_concurrency": 1,
      "post_fb_used_gib": 17.0,
      "post_gpu_util_pct": 72,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 1,
      "post_latency_ms": 105,
      "post_throughput_rps": 12
    },
    {
      "id": "compute_saturation_no_oom",
      "model_footprint_gib": 8.0,
      "request_tokens": 4096,
      "concurrency": 6,
      "fb_total_gib": 24.0,
      "fb_used_gib": 18.0,
      "gpu_util_pct": 99,
      "allocation_result": "SUCCESS",
      "queue_depth": 14,
      "latency_ms": 245,
      "latency_slo_ms": 120,
      "throughput_rps": 14,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 4096,
      "post_concurrency": 4,
      "post_fb_used_gib": 17.0,
      "post_gpu_util_pct": 92,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 3,
      "post_latency_ms": 110,
      "post_throughput_rps": 23
    },
    {
      "id": "non_gpu_bottleneck",
      "model_footprint_gib": 8.0,
      "request_tokens": 2048,
      "concurrency": 4,
      "fb_total_gib": 24.0,
      "fb_used_gib": 13.0,
      "gpu_util_pct": 48,
      "allocation_result": "SUCCESS",
      "queue_depth": 22,
      "latency_ms": 300,
      "latency_slo_ms": 120,
      "throughput_rps": 8,
      "minimum_throughput_rps": 20,
      "post_request_tokens": 2048,
      "post_concurrency": 4,
      "post_fb_used_gib": 13.0,
      "post_gpu_util_pct": 50,
      "post_allocation_result": "SUCCESS",
      "post_queue_depth": 19,
      "post_latency_ms": 265,
      "post_throughput_rps": 9
    },
    {
      "id": "ambiguous_oom",
      "model_footprint_gib": 8.0,
      "request_tokens": 8000,
      "concurrency": 4,
      "fb_total_gib": 24.0,
      "fb_used_gib": 23.85,
      "gpu_util_pct": 91,
      "allocation_result": "OOM",
      "queue_depth": 7,
      "latency_ms": 310,
      "latency_slo_ms": 120,
      "throughput_rps": 7,
      "minimum_throughput_rps": 15,
      "post_request_tokens": 7000,
      "post_concurrency": 3,
      "post_fb_used_gib": 23.4,
      "post_gpu_util_pct": 89,
      "post_allocation_result": "OOM",
      "post_queue_depth": 5,
      "post_latency_ms": 250,
      "post_throughput_rps": 9
    }
  ]
}