{
  "generated_at": "2026-09-15T10:08:47.709Z",
  "total": 22,
  "exported": 22,
  "truncated": false,
  "filters": {
    "hardwareId": null,
    "modelId": null,
    "quantId": "q4_k_m",
    "frameworkId": null,
    "platform": null,
    "minEvidence": "L0",
    "range": "30d",
    "sort": "recent",
    "page": 1,
    "perPage": 50,
    "keyword": null,
    "scope": "all"
  },
  "data_status": "sample",
  "runs": [
    {
      "model": "DeepSeek-R1-Distill-Qwen-14B · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RTX 3060 · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 28.2,
      "prefill_tps": 971.2,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 6,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "V2EX",
      "source_url": "https://www.v2ex.com/t/araoai-demo-3",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "phi-4 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RX 7900 GRE · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 45.1,
      "prefill_tps": 1553.8,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 10,
      "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/1007",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Mistral-Nemo-Instruct-2407 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "M2 Ultra · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 73,
      "prefill_tps": 2517.8,
      "ttft_cold_ms": 3780,
      "vram_peak_gb": 9.8,
      "sample_size": 14,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-11",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "phi-4 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "NVIDIA RTX 3090 · 24G",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 73.2,
      "prefill_tps": 2525,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 15,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "B站",
      "source_url": "https://www.bilibili.com/video/araoai-demo-12",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-32B",
      "quant": "Q4_K_M",
      "hardware": "Apple M3 Max · 128G 统一内存",
      "framework": "llama.cpp b6120",
      "decode_tps": 12.9,
      "prefill_tps": 439.2,
      "ttft_cold_ms": 8780,
      "vram_peak_gb": 23.5,
      "sample_size": 16,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "Arao 自产",
      "source_url": "https://araoai.com/demo/rig-run-13",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen2.5-7B-Instruct · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "AMD Radeon RX 7900 XTX · 24G",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 138.9,
      "prefill_tps": 4699.7,
      "ttft_cold_ms": 2530,
      "vram_peak_gb": 6.4,
      "sample_size": 17,
      "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/1014",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen2.5-7B-Instruct · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "M2 Ultra · 未归一化",
      "framework": "ollama 0.5.1",
      "decode_tps": 109.8,
      "prefill_tps": 3672.7,
      "ttft_cold_ms": 2530,
      "vram_peak_gb": 6.4,
      "sample_size": 6,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-18",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-14B",
      "quant": "Q4_K_M",
      "hardware": "RTX 4080 SUPER · 未归一化",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 53.2,
      "prefill_tps": 1801.6,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 10,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "Reddit",
      "source_url": "https://www.reddit.com/r/araoai_demo/comments/demo22_qwen3-14b-gguf/",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Llama-3.1-8B-Instruct · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RX 7800 XT · 未归一化",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 85.4,
      "prefill_tps": 2945.8,
      "ttft_cold_ms": 2780,
      "vram_peak_gb": 7.1,
      "sample_size": 16,
      "evidence": "L2",
      "repro_count": 3,
      "repro_chain": "一致 3 次，偏差 0 次，共 3 格",
      "status": "稳定",
      "source_platform": "GitHub",
      "source_url": "https://github.com/araoai-demo/benchmarks/issues/1028",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Llama-3.1-8B-Instruct · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RX 7800 XT · 未归一化",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 79,
      "prefill_tps": 2673,
      "ttft_cold_ms": 2780,
      "vram_peak_gb": 7.1,
      "sample_size": 17,
      "evidence": "L2",
      "repro_count": 2,
      "repro_chain": "一致 2 次，偏差 0 次，共 3 格",
      "status": "稳定",
      "source_platform": "Reddit",
      "source_url": "https://www.reddit.com/r/araoai_demo/comments/demo2900_llama-3-1-8b-instruct/",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "DeepSeek-R1-Distill-Qwen-14B · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "Arc A770 · 未归一化",
      "framework": "llama.cpp b5980",
      "decode_tps": 39.7,
      "prefill_tps": 1336.4,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 3,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "HuggingFace",
      "source_url": "https://huggingface.co/araoai-demo/deepseek-r1-distill-qwen-14b-bench/discussions/30",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Mistral-Nemo-Instruct-2407 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "NVIDIA RTX 4090 · 24G",
      "framework": "ollama 0.5.1",
      "decode_tps": 80.7,
      "prefill_tps": 2699.4,
      "ttft_cold_ms": 3780,
      "vram_peak_gb": 9.8,
      "sample_size": 5,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-32",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "DeepSeek-R1-Distill-Qwen-14B · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "NVIDIA RTX 3090 · 24G",
      "framework": "llama.cpp b5980",
      "decode_tps": 66.3,
      "prefill_tps": 2233.7,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 7,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "Arao 自产",
      "source_url": "https://araoai.com/demo/rig-run-34",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-14B",
      "quant": "Q4_K_M",
      "hardware": "NVIDIA RTX 3090 · 24G",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 67.7,
      "prefill_tps": 2291.1,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 11,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "V2EX",
      "source_url": "https://www.v2ex.com/t/araoai-demo-38",
      "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"
    },
    {
      "model": "phi-4 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "Apple M3 Max · 128G 统一内存",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 28.9,
      "prefill_tps": 979.1,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 16,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "Reddit",
      "source_url": "https://www.reddit.com/r/araoai_demo/comments/demo43_phi-4-gguf/",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Mistral-Nemo-Instruct-2407 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RTX 3060 · 未归一化",
      "framework": "llama.cpp b6120",
      "decode_tps": 31,
      "prefill_tps": 1054,
      "ttft_cold_ms": 3780,
      "vram_peak_gb": 9.8,
      "sample_size": 4,
      "evidence": "L1",
      "repro_count": 1,
      "repro_chain": "一致 1 次，偏差 0 次，共 1 格",
      "status": "待判定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-46",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-14B",
      "quant": "Q4_K_M",
      "hardware": "RTX 3060 · 未归一化",
      "framework": "vLLM 0.6.2",
      "decode_tps": 29.8,
      "prefill_tps": 1040.4,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 6,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "Arao 自产",
      "source_url": "https://araoai.com/demo/rig-run-48",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-14B",
      "quant": "Q4_K_M",
      "hardware": "RX 7900 GRE · 未归一化",
      "framework": "mlx-lm 0.18.2",
      "decode_tps": 41.7,
      "prefill_tps": 1409.9,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 7,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "GitHub",
      "source_url": "https://github.com/araoai-demo/benchmarks/issues/1049",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "phi-4 · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "AMD Radeon RX 7900 XTX · 24G",
      "framework": "ExLlamaV2 0.2.5",
      "decode_tps": 75.1,
      "prefill_tps": 2589.7,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 12,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "B站",
      "source_url": "https://www.bilibili.com/video/araoai-demo-54",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "DeepSeek-R1-Distill-Qwen-14B · 未归一化",
      "quant": "Q4_K_M",
      "hardware": "RTX 4080 SUPER · 未归一化",
      "framework": "vLLM 0.6.2",
      "decode_tps": 60.8,
      "prefill_tps": 2127.1,
      "ttft_cold_ms": 4280,
      "vram_peak_gb": 11.2,
      "sample_size": 3,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "待判定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-60",
      "fetched_at": "2026-09-14T00:00:00Z"
    },
    {
      "model": "Qwen3-32B",
      "quant": "Q4_K_M",
      "hardware": "M2 Ultra · 未归一化",
      "framework": "llama.cpp b6120",
      "decode_tps": 25.8,
      "prefill_tps": 878.3,
      "ttft_cold_ms": 8780,
      "vram_peak_gb": 23.5,
      "sample_size": 10,
      "evidence": "L0",
      "repro_count": 0,
      "repro_chain": "待复现",
      "status": "稳定",
      "source_platform": "知乎",
      "source_url": "https://zhuanlan.zhihu.com/p/araoai-demo-67",
      "fetched_at": "2026-09-14T00:00:00Z"
    }
  ]
}