{
  "generated_at": "2026-09-15T12:30:31.580Z",
  "total": 2,
  "exported": 2,
  "truncated": false,
  "filters": {
    "hardwareId": "unknown:arc-a770",
    "modelId": null,
    "quantId": null,
    "frameworkId": "exllama",
    "platform": null,
    "minEvidence": "L0",
    "range": "all",
    "sort": "recent",
    "page": 1,
    "perPage": 50,
    "keyword": null,
    "scope": "all"
  },
  "data_status": "sample",
  "runs": [
    {
      "model": "Llama-3.1-8B-Instruct · 未归一化",
      "quant": "Q5_K_M",
      "hardware": "Arc A770 · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 65.7,
      "prefill_tps": 2266,
      "ttft_cold_ms": 3113,
      "vram_peak_gb": 8,
      "sample_size": 9,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "GitHub",
      "source_url": "https://github.com/araoai-demo/benchmarks/issues/1021",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen2.5-7B-Instruct · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "Arc A770 · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 87.6,
      "prefill_tps": 3021.4,
      "ttft_cold_ms": 2530,
      "vram_peak_gb": 6.4,
      "sample_size": 15,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "GitHub",
      "source_url": "https://github.com/araoai-demo/benchmarks/issues/1042",
      "fetched_at": "2026-09-14T00:00:00Z"
    }
  ]
}