{
  "schemaVersion": 1,
  "recordedAt": "2026-09-20T20:03:54.765554Z",
  "model": {
    "servedName": "qwen3.8-flash-next-nvfp4",
    "checkpoint": "RadixArk/Qwen3.8-Flash-Next-NVFP4"
  },
  "runtime": {
    "name": "FreeToken",
    "commit": "cc1f5c2c91855f2cc7787ad6b909f7e46a5d5825",
    "attentionBackend": "qsa_sparse",
    "moeStrategy": "offload",
    "expertKernel": "nvfp4 via triton",
    "pleBackend": "disk"
  },
  "worker": {
    "location": "New Jersey, US",
    "gpu": "NVIDIA GeForce RTX 5090",
    "gpuMemoryMiB": 32607,
    "cpu": "AMD EPYC 9534 64-Core Processor",
    "pcieAdvertisedGBps": 54.1,
    "disk": "WD_BLACK SN850P for PS5 8000GB"
  },
  "bandwidth": {
    "cpuStreamReadGBps": 149.58,
    "pcieLinearH2DGBps": 56.81,
    "pcieLinearD2HGBps": 56.57,
    "nvfp4PcieGatherGBps": 51.92,
    "nvfp4CpuMoeGBps": 59.89,
    "recommendedStrategy": "offload"
  },
  "singleStream": [
    {
      "maxOutputTokens": 128,
      "trials": 3,
      "medianTTFTSeconds": 1.233,
      "medianSteadyDecodeTokensPerSecond": 75.057,
      "medianEndToEndTokensPerSecond": 43.616
    },
    {
      "maxOutputTokens": 1024,
      "trials": 2,
      "medianTTFTSeconds": 1.226,
      "medianSteadyDecodeTokensPerSecond": 80.918,
      "medianEndToEndTokensPerSecond": 73.832
    }
  ],
  "concurrency": [
    {"level": 1, "activeRequestLimit": 8, "promptTokens": 66, "completionTokens": 88, "batchElapsedSeconds": 2.518, "medianTTFTSeconds": 1.234, "aggregateCompletionTokensPerSecond": 34.942, "medianSteadyDecodeTokensPerSecond": 67.823, "successful": true},
    {"level": 2, "activeRequestLimit": 8, "promptTokens": 132, "completionTokens": 185, "batchElapsedSeconds": 4.002, "medianTTFTSeconds": 2.252, "aggregateCompletionTokensPerSecond": 46.223, "medianSteadyDecodeTokensPerSecond": 54.324, "successful": true},
    {"level": 4, "activeRequestLimit": 8, "promptTokens": 264, "completionTokens": 342, "batchElapsedSeconds": 4.306, "medianTTFTSeconds": 2.447, "aggregateCompletionTokensPerSecond": 79.430, "medianSteadyDecodeTokensPerSecond": 46.214, "successful": true},
    {"level": 8, "activeRequestLimit": 8, "promptTokens": 528, "completionTokens": 740, "batchElapsedSeconds": 6.421, "medianTTFTSeconds": 3.061, "effectiveAggregatePrefillTokensPerSecond": 172.5, "aggregateCompletionTokensPerSecond": 115.249, "medianSteadyDecodeTokensPerSecond": 32.428, "successful": true},
    {"level": 16, "activeRequestLimit": 16, "promptTokens": 1079, "completionTokens": 1543, "batchElapsedSeconds": 12.537, "medianTTFTSeconds": 3.370, "effectiveAggregatePrefillTokensPerSecond": 320.2, "aggregateCompletionTokensPerSecond": 123.079, "medianSteadyDecodeTokensPerSecond": 11.614, "successful": true}
  ],
  "interpretation": {
    "recommendedConcurrency": 8,
    "reason": "Concurrency 16 increased aggregate output throughput by 6.8% while reducing median per-request steady decode throughput from 32.428 to 11.614 tokens per second."
  }
}