{
  "cliff": 2560,
  "models": [
    {
      "below_cliff": true,
      "block_speedup": {
        "1": 1.0159533073929963,
        "16": 1.469910716534255,
        "8": 1.3035249819862567
      },
      "hidden": 896,
      "intermediate": 4864,
      "layers": 24,
      "model": "Qwen2.5-0.5B-Instruct-4bit",
      "prefill_down_speedup": 2.321689948699001,
      "prefill_up_speedup": 1.014739871221443
    },
    {
      "below_cliff": true,
      "block_speedup": {
        "1": 1.022573126641721,
        "16": 1.3213056003029673,
        "8": 1.2321990164360201
      },
      "hidden": 1536,
      "intermediate": 8960,
      "layers": 28,
      "model": "Qwen2.5-1.5B-Instruct-4bit",
      "prefill_down_speedup": 2.7943500774507637,
      "prefill_up_speedup": 1.0272895714639583
    },
    {
      "below_cliff": true,
      "block_speedup": {
        "1": 0.9787414011121801,
        "16": 1.811065020086297,
        "8": 1.2725086269998958
      },
      "hidden": 2048,
      "intermediate": 8192,
      "layers": 16,
      "model": "Llama-3.2-1B-Instruct-4bit",
      "prefill_down_speedup": 3.0859020719513452,
      "prefill_up_speedup": 1.056946805239965
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 0.974022266153366,
        "16": 1.724416536883332,
        "8": 1.2317307292085395
      },
      "hidden": 3072,
      "intermediate": 8192,
      "layers": 28,
      "model": "Llama-3.2-3B-Instruct-4bit",
      "prefill_down_speedup": 1.0645489258522691,
      "prefill_up_speedup": 1.07363158369368
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 0.9710773375238523,
        "16": 1.2851061764861016,
        "8": 1.2361424589025911
      },
      "hidden": 2560,
      "intermediate": 9216,
      "layers": 32,
      "model": "Qwen3.5-4B-4bit",
      "prefill_down_speedup": 1.2154574607092807,
      "prefill_up_speedup": 1.0894659285999502
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 1.0042250014020526,
        "16": 1.272089784203057,
        "8": 1.208501116726675
      },
      "hidden": 4096,
      "intermediate": 12288,
      "layers": 32,
      "model": "Qwen3.5-9B-4bit",
      "prefill_down_speedup": 1.1451193725706423,
      "prefill_up_speedup": 1.0005195995440563
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 0.9884400278107033,
        "16": 1.24229027695934,
        "8": 1.206608494224176
      },
      "hidden": 5120,
      "intermediate": 17408,
      "layers": 64,
      "model": "Qwen3.6-27B-4bit",
      "prefill_down_speedup": 1.005869821857639,
      "prefill_up_speedup": 1.0921425678412315
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 1.00382554364555,
        "16": 1.2683794810841196,
        "8": 1.2228912270883843
      },
      "hidden": 2560,
      "intermediate": 9728,
      "layers": 36,
      "model": "Qwen3-VL-4B-Instruct-4bit",
      "prefill_down_speedup": 1.2199208883635804,
      "prefill_up_speedup": 1.0959107412266742
    },
    {
      "below_cliff": false,
      "block_speedup": {
        "1": 1.0064916164535491,
        "16": 1.253771699205944,
        "8": 1.2030589690154248
      },
      "hidden": 3584,
      "intermediate": 18944,
      "layers": 28,
      "model": "Qwen2.5-VL-7B-Instruct-4bit",
      "prefill_down_speedup": 1.0006981471579988,
      "prefill_up_speedup": 1.0729846073571614
    }
  ],
  "precision_comparison": {
    "class": "same_representation",
    "same_weight_representation": true
  },
  "prompt_rows": 127,
  "rounds": 25,
  "scope": "model_shape_matrix"
}
