{
 "schema": 2,
 "rows": [
  {
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "isl_target": 1024,
   "isl_short": "1K",
   "isl": 1042,
   "snow_decode": 27.7,
   "snow_prefill": 437.1,
   "snow_ttft": 2.38,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.378,
    "derived": false
   },
   "snow_arm": "MTP k=2",
   "snow_step_ms": 86.26,
   "llama_decode": 18.9,
   "ratio": 1.46,
   "llama_prefill": 185.6,
   "llama_ttft": 5.61,
   "ttft_ratio": 2.36,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.314,
    "derived": true
   },
   "llama_arm": "MTP k=2"
  },
  {
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "isl_target": 8192,
   "isl_short": "8K",
   "isl": 8211,
   "snow_decode": 25.8,
   "snow_prefill": 517.1,
   "snow_ttft": 15.88,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.284,
    "derived": false
   },
   "snow_arm": "MTP k=2",
   "snow_step_ms": 88.11,
   "llama_decode": 17.1,
   "ratio": 1.51,
   "llama_prefill": 208.6,
   "llama_ttft": 39.35,
   "ttft_ratio": 2.48,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.254,
    "derived": true
   },
   "llama_arm": "MTP k=2"
  },
  {
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "isl_target": 32768,
   "isl_short": "32K",
   "isl": 32787,
   "snow_decode": 26.2,
   "snow_prefill": 507.7,
   "snow_ttft": 64.58,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.315,
    "derived": false
   },
   "snow_arm": "MTP k=2",
   "snow_step_ms": 88.91,
   "llama_decode": 13.8,
   "ratio": 1.9,
   "llama_prefill": 126.5,
   "llama_ttft": 259.25,
   "ttft_ratio": 4.01,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.305,
    "derived": true
   },
   "llama_arm": "MTP k=2"
  },
  {
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "isl_target": 131072,
   "isl_short": "128K",
   "isl": 131091,
   "snow_decode": 26.9,
   "snow_prefill": 491.1,
   "snow_ttft": 266.95,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.487,
    "derived": false
   },
   "snow_arm": "MTP k=2",
   "snow_step_ms": 91.64,
   "llama_decode": 7.4,
   "ratio": 3.63,
   "llama_prefill": 48.6,
   "llama_ttft": 2694.98,
   "ttft_ratio": 10.1,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.329,
    "derived": true
   },
   "llama_arm": "MTP k=2"
  },
  {
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "isl_target": 225280,
   "isl_short": "220K",
   "isl": 225297,
   "snow_decode": 26.8,
   "snow_prefill": 480.7,
   "snow_ttft": 468.72,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.49,
    "derived": false
   },
   "snow_arm": "MTP k=2",
   "snow_step_ms": 93.79,
   "llama_decode": 6.2,
   "ratio": 4.34,
   "llama_prefill": 30.7,
   "llama_ttft": 7340.73,
   "ttft_ratio": 15.66,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.504,
    "derived": true
   },
   "llama_arm": "MTP k=2"
  },
  {
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "isl_target": 1024,
   "isl_short": "1K",
   "isl": 1041,
   "snow_decode": 86.4,
   "snow_prefill": 2131.9,
   "snow_ttft": 0.49,
   "snow_accept": {
    "unit": "AL",
    "mean": 3.231,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 37.7,
   "llama_decode": 57.4,
   "ratio": 1.5,
   "llama_prefill": 874.6,
   "llama_ttft": 1.19,
   "ttft_ratio": 2.44,
   "llama_accept": {
    "unit": "AL",
    "mean": 3.127,
    "derived": true
   },
   "llama_arm": "DFlash"
  },
  {
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "isl_target": 8192,
   "isl_short": "8K",
   "isl": 8209,
   "snow_decode": 78.2,
   "snow_prefill": 3715.2,
   "snow_ttft": 2.21,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.973,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 38.94,
   "llama_decode": 46.0,
   "ratio": 1.7,
   "llama_prefill": 1063.7,
   "llama_ttft": 7.72,
   "ttft_ratio": 3.49,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.804,
    "derived": true
   },
   "llama_arm": "DFlash"
  },
  {
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "isl_target": 32768,
   "isl_short": "32K",
   "isl": 32783,
   "snow_decode": 70.6,
   "snow_prefill": 3861.2,
   "snow_ttft": 8.49,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.982,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 41.94,
   "llama_decode": 41.8,
   "ratio": 1.69,
   "llama_prefill": 984.4,
   "llama_ttft": 33.3,
   "ttft_ratio": 3.92,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.647,
    "derived": true
   },
   "llama_arm": "DFlash"
  },
  {
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "isl_target": 131072,
   "isl_short": "131K",
   "isl": 131086,
   "snow_decode": 55.0,
   "snow_prefill": 1835.3,
   "snow_ttft": 71.42,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.925,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 53.34,
   "llama_decode": 27.9,
   "ratio": 1.97,
   "llama_prefill": 485.8,
   "llama_ttft": 269.86,
   "ttft_ratio": 3.78,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.617,
    "derived": true
   },
   "llama_arm": "DFlash"
  },
  {
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "isl_target": 225280,
   "isl_short": "225K",
   "isl": 225297,
   "snow_decode": 30.8,
   "snow_prefill": 1311.6,
   "snow_ttft": 171.77,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.557,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 83.09,
   "llama_decode": 20.4,
   "ratio": 1.51,
   "llama_prefill": 297.7,
   "llama_ttft": 756.89,
   "ttft_ratio": 4.41,
   "llama_accept": {
    "unit": "AL",
    "mean": 2.431,
    "derived": true
   },
   "llama_arm": "DFlash"
  },
  {
   "model": "Qwen3.6-35B-A3B FP8",
   "isl_target": 1024,
   "isl_short": "1K",
   "isl": 1041,
   "snow_decode": 68.5,
   "snow_prefill": 2133.8,
   "snow_ttft": 0.49,
   "snow_accept": {
    "unit": "AL",
    "mean": 3.12,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 45.66
  },
  {
   "model": "Qwen3.6-35B-A3B FP8",
   "isl_target": 8192,
   "isl_short": "8K",
   "isl": 8209,
   "snow_decode": 62.5,
   "snow_prefill": 3605.5,
   "snow_ttft": 2.28,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.931,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 46.88
  },
  {
   "model": "Qwen3.6-35B-A3B FP8",
   "isl_target": 32768,
   "isl_short": "32K",
   "isl": 32783,
   "snow_decode": 58.1,
   "snow_prefill": 3019.6,
   "snow_ttft": 10.86,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.872,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 49.64
  },
  {
   "model": "Qwen3.6-35B-A3B FP8",
   "isl_target": 131072,
   "isl_short": "131K",
   "isl": 131086,
   "snow_decode": 42.2,
   "snow_prefill": 1811.4,
   "snow_ttft": 72.37,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.809,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 68.57
  },
  {
   "model": "Qwen3.6-35B-A3B FP8",
   "isl_target": 225280,
   "isl_short": "225K",
   "isl": 225297,
   "snow_decode": 25.6,
   "snow_prefill": 1303.4,
   "snow_ttft": 172.85,
   "snow_accept": {
    "unit": "AL",
    "mean": 2.476,
    "derived": false
   },
   "snow_arm": "DFlash",
   "snow_step_ms": 98.49
  }
 ],
 "series": [
  {
   "key": "deepseek_v4_flash_iq2_xxs_snowllm_128k",
   "engine": "SnowLLM",
   "model": "DeepSeek-V4-Flash IQ2_XXS",
   "arm": "DSpark, ctx 132096",
   "label": "SnowLLM \u00b7 DeepSeek-V4-Flash IQ2_XXS \u00b7 DSpark, ctx 132096",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1033,
     "decode": 23.09,
     "prefill": 360.6,
     "ttft": 2.865,
     "step_ms": 115.25,
     "al": 2.679,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8202,
     "decode": 21.9,
     "prefill": 629.8,
     "ttft": 13.024,
     "step_ms": 115.37,
     "al": 2.567,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32777,
     "decode": 20.78,
     "prefill": 632.1,
     "ttft": 51.853,
     "step_ms": 119.08,
     "al": 2.47,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131082,
     "decode": 19.93,
     "prefill": 681.4,
     "ttft": 192.38,
     "step_ms": 131.25,
     "al": 2.615,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "deepseek_v4_flash_iq2_xxs_snowllm",
   "engine": "SnowLLM",
   "model": "DeepSeek-V4-Flash IQ2_XXS",
   "arm": "DSpark, ctx 227328",
   "label": "SnowLLM \u00b7 DeepSeek-V4-Flash IQ2_XXS \u00b7 DSpark, ctx 227328",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1033,
     "decode": 23.06,
     "prefill": 360.7,
     "ttft": 2.864,
     "step_ms": 115.54,
     "al": 2.679,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8202,
     "decode": 21.9,
     "prefill": 629.1,
     "ttft": 13.037,
     "step_ms": 115.51,
     "al": 2.551,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32777,
     "decode": 20.73,
     "prefill": 632.5,
     "ttft": 51.823,
     "step_ms": 119.37,
     "al": 2.47,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131082,
     "decode": 18.43,
     "prefill": 680.4,
     "ttft": 192.643,
     "step_ms": 132.81,
     "al": 2.42,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225287,
     "decode": 16.66,
     "prefill": 513.4,
     "ttft": 438.848,
     "step_ms": 144.31,
     "al": 2.408,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "deepseek_v4_flash_iq2_xxs_nospec_snowllm_128k",
   "engine": "SnowLLM",
   "model": "DeepSeek-V4-Flash IQ2_XXS",
   "arm": "no spec, ctx 132096",
   "label": "SnowLLM \u00b7 DeepSeek-V4-Flash IQ2_XXS \u00b7 no spec, ctx 132096",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1033,
     "decode": 18.15,
     "prefill": 370.4,
     "ttft": 2.789,
     "step_ms": 55.08,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8202,
     "decode": 18.13,
     "prefill": 634.4,
     "ttft": 12.929,
     "step_ms": 55.15,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32777,
     "decode": 17.78,
     "prefill": 636.0,
     "ttft": 51.536,
     "step_ms": 56.24,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131082,
     "decode": 16.56,
     "prefill": 682.5,
     "ttft": 192.055,
     "step_ms": 60.39,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "deepseek_v4_flash_iq2_xxs_nospec_llama",
   "engine": "llama.cpp",
   "model": "DeepSeek-V4-Flash IQ2_XXS",
   "arm": "no spec, ctx 132096",
   "label": "llama.cpp \u00b7 DeepSeek-V4-Flash IQ2_XXS \u00b7 no spec, ctx 132096",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1033,
     "decode": 15.36,
     "prefill": 179.2,
     "ttft": 5.766,
     "step_ms": 65.09,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8202,
     "decode": 12.91,
     "prefill": 177.1,
     "ttft": 46.311,
     "step_ms": 77.47,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32777,
     "decode": 11.75,
     "prefill": 149.4,
     "ttft": 219.439,
     "step_ms": 85.12,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131082,
     "decode": 8.77,
     "prefill": 72.7,
     "ttft": 1802.114,
     "step_ms": 113.97,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "glm5_3_flash_iq2_xxs_mtp_snowllm",
   "engine": "SnowLLM",
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "arm": "MTP k=2",
   "label": "SnowLLM \u00b7 GLM-5.3-Flash UD-IQ2_XXS \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1042,
     "decode": 27.65,
     "prefill": 437.1,
     "ttft": 2.384,
     "step_ms": 86.26,
     "al": 2.378,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8211,
     "decode": 25.76,
     "prefill": 517.1,
     "ttft": 15.878,
     "step_ms": 88.11,
     "al": 2.284,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32787,
     "decode": 26.17,
     "prefill": 507.7,
     "ttft": 64.579,
     "step_ms": 88.91,
     "al": 2.315,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131091,
     "decode": 26.92,
     "prefill": 491.1,
     "ttft": 266.947,
     "step_ms": 91.64,
     "al": 2.487,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 26.79,
     "prefill": 480.7,
     "ttft": 468.715,
     "step_ms": 93.79,
     "al": 2.49,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "glm5_3_flash_iq2_xxs_mtp_llama",
   "engine": "llama.cpp",
   "model": "GLM-5.3-Flash UD-IQ2_XXS",
   "arm": "MTP k=2",
   "label": "llama.cpp \u00b7 GLM-5.3-Flash UD-IQ2_XXS \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1042,
     "decode": 18.91,
     "prefill": 185.6,
     "ttft": 5.615,
     "step_ms": 122.38,
     "al": 2.314,
     "accept_pct": 0.657,
     "derived": true
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8211,
     "decode": 17.07,
     "prefill": 208.6,
     "ttft": 39.354,
     "step_ms": 132.01,
     "al": 2.254,
     "accept_pct": 0.6269,
     "derived": true
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32787,
     "decode": 13.78,
     "prefill": 126.5,
     "ttft": 259.255,
     "step_ms": 167.34,
     "al": 2.305,
     "accept_pct": 0.6526,
     "derived": true
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131091,
     "decode": 7.42,
     "prefill": 48.6,
     "ttft": 2694.976,
     "step_ms": 313.83,
     "al": 2.329,
     "accept_pct": 0.6645,
     "derived": true
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 6.17,
     "prefill": 30.7,
     "ttft": 7340.729,
     "step_ms": 405.87,
     "al": 2.504,
     "accept_pct": 0.752,
     "derived": true
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_fp8_dflash_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.6-35B-A3B FP8",
   "arm": "DFlash",
   "label": "SnowLLM \u00b7 Qwen3.6-35B-A3B FP8 \u00b7 DFlash",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 68.53,
     "prefill": 2133.8,
     "ttft": 0.488,
     "step_ms": 45.66,
     "al": 3.12,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 62.53,
     "prefill": 3605.5,
     "ttft": 2.277,
     "step_ms": 46.88,
     "al": 2.931,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 58.09,
     "prefill": 3019.6,
     "ttft": 10.857,
     "step_ms": 49.64,
     "al": 2.872,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 42.23,
     "prefill": 1811.4,
     "ttft": 72.368,
     "step_ms": 68.57,
     "al": 2.809,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 25.62,
     "prefill": 1303.4,
     "ttft": 172.852,
     "step_ms": 98.49,
     "al": 2.476,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_fp8_mtp_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.6-35B-A3B FP8",
   "arm": "MTP k=2",
   "label": "SnowLLM \u00b7 Qwen3.6-35B-A3B FP8 \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 67.39,
     "prefill": 2563.4,
     "ttft": 0.406,
     "step_ms": 36.95,
     "al": 2.504,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 65.1,
     "prefill": 3851.4,
     "ttft": 2.131,
     "step_ms": 37.85,
     "al": 2.481,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 61.9,
     "prefill": 3033.6,
     "ttft": 10.807,
     "step_ms": 40.67,
     "al": 2.507,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 49.59,
     "prefill": 1720.8,
     "ttft": 76.177,
     "step_ms": 52.55,
     "al": 2.573,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 39.38,
     "prefill": 1214.6,
     "ttft": 185.493,
     "step_ms": 63.28,
     "al": 2.477,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_q4_k_xl_dflash_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "arm": "DFlash",
   "label": "SnowLLM \u00b7 Qwen3.6-35B-A3B UD-Q4_K_XL \u00b7 DFlash",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 86.37,
     "prefill": 2131.9,
     "ttft": 0.488,
     "step_ms": 37.7,
     "al": 3.231,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 78.18,
     "prefill": 3715.2,
     "ttft": 2.21,
     "step_ms": 38.94,
     "al": 2.973,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 70.61,
     "prefill": 3861.2,
     "ttft": 8.49,
     "step_ms": 41.94,
     "al": 2.982,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 55.01,
     "prefill": 1835.3,
     "ttft": 71.425,
     "step_ms": 53.34,
     "al": 2.925,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 30.83,
     "prefill": 1311.6,
     "ttft": 171.775,
     "step_ms": 83.09,
     "al": 2.557,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_q4_k_xl_dflash_llama",
   "engine": "llama.cpp",
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "arm": "DFlash",
   "label": "llama.cpp \u00b7 Qwen3.6-35B-A3B UD-Q4_K_XL \u00b7 DFlash",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 57.4,
     "prefill": 874.6,
     "ttft": 1.19,
     "step_ms": 54.47,
     "al": 3.127,
     "accept_pct": 0.3039,
     "derived": true
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 45.96,
     "prefill": 1063.7,
     "ttft": 7.718,
     "step_ms": 61.01,
     "al": 2.804,
     "accept_pct": 0.2577,
     "derived": true
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 41.82,
     "prefill": 984.4,
     "ttft": 33.302,
     "step_ms": 63.3,
     "al": 2.647,
     "accept_pct": 0.2353,
     "derived": true
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 27.86,
     "prefill": 485.8,
     "ttft": 269.858,
     "step_ms": 93.92,
     "al": 2.617,
     "accept_pct": 0.231,
     "derived": true
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 20.42,
     "prefill": 297.7,
     "ttft": 756.894,
     "step_ms": 119.06,
     "al": 2.431,
     "accept_pct": 0.2044,
     "derived": true
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_q4_k_xl_mtp_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "arm": "MTP k=2",
   "label": "SnowLLM \u00b7 Qwen3.6-35B-A3B UD-Q4_K_XL \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 86.4,
     "prefill": 2145.4,
     "ttft": 0.485,
     "step_ms": 28.63,
     "al": 2.489,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 83.84,
     "prefill": 3806.5,
     "ttft": 2.157,
     "step_ms": 29.52,
     "al": 2.468,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 75.55,
     "prefill": 3864.4,
     "ttft": 8.483,
     "step_ms": 32.43,
     "al": 2.452,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 54.4,
     "prefill": 1765.3,
     "ttft": 74.258,
     "step_ms": 46.73,
     "al": 2.557,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 33.59,
     "prefill": 1244.0,
     "ttft": 181.111,
     "step_ms": 72.33,
     "al": 2.465,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_6_35b_a3b_q4_k_xl_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.6-35B-A3B UD-Q4_K_XL",
   "arm": "no spec",
   "label": "SnowLLM \u00b7 Qwen3.6-35B-A3B UD-Q4_K_XL \u00b7 no spec",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1041,
     "decode": 62.02,
     "prefill": 2245.2,
     "ttft": 0.464,
     "step_ms": 16.12,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8209,
     "decode": 59.06,
     "prefill": 3936.0,
     "ttft": 2.086,
     "step_ms": 16.93,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32783,
     "decode": 52.45,
     "prefill": 4052.9,
     "ttft": 8.089,
     "step_ms": 19.06,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131086,
     "decode": 35.14,
     "prefill": 1764.3,
     "ttft": 74.3,
     "step_ms": 28.46,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225297,
     "decode": 20.56,
     "prefill": 1267.8,
     "ttft": 177.711,
     "step_ms": 48.64,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_8_27b_q4_k_xl_dflash2_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.8-27B UD-Q4_K_XL",
   "arm": "DFlash2",
   "label": "SnowLLM \u00b7 Qwen3.8-27B UD-Q4_K_XL \u00b7 DFlash2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1083,
     "decode": 28.9,
     "prefill": 511.9,
     "ttft": 2.116,
     "step_ms": 115.45,
     "al": 3.417,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8251,
     "decode": 24.61,
     "prefill": 602.7,
     "ttft": 13.69,
     "step_ms": 117.98,
     "al": 2.902,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32825,
     "decode": 18.66,
     "prefill": 502.3,
     "ttft": 65.355,
     "step_ms": 165.81,
     "al": 3.173,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131128,
     "decode": 13.84,
     "prefill": 350.5,
     "ttft": 374.126,
     "step_ms": 221.68,
     "al": 3.154,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225339,
     "decode": 11.46,
     "prefill": 272.3,
     "ttft": 827.468,
     "step_ms": 261.16,
     "al": 3.058,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_8_27b_q4_k_xl_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.8-27B UD-Q4_K_XL",
   "arm": "MTP k=2",
   "label": "SnowLLM \u00b7 Qwen3.8-27B UD-Q4_K_XL \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1083,
     "decode": 20.28,
     "prefill": 503.8,
     "ttft": 2.15,
     "step_ms": 112.29,
     "al": 2.284,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8251,
     "decode": 19.26,
     "prefill": 602.9,
     "ttft": 13.685,
     "step_ms": 115.08,
     "al": 2.218,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32825,
     "decode": 15.29,
     "prefill": 499.6,
     "ttft": 65.702,
     "step_ms": 142.59,
     "al": 2.189,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131128,
     "decode": 10.31,
     "prefill": 343.9,
     "ttft": 381.3,
     "step_ms": 220.89,
     "al": 2.279,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225339,
     "decode": 9.23,
     "prefill": 264.9,
     "ttft": 850.649,
     "step_ms": 267.74,
     "al": 2.464,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_8_27b_q4_k_xl_llama",
   "engine": "llama.cpp",
   "model": "Qwen3.8-27B UD-Q4_K_XL",
   "arm": "MTP k=2",
   "label": "llama.cpp \u00b7 Qwen3.8-27B UD-Q4_K_XL \u00b7 MTP k=2",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1083,
     "decode": 17.65,
     "prefill": 301.0,
     "ttft": 3.598,
     "step_ms": 125.03,
     "al": 2.207,
     "accept_pct": 0.6035,
     "derived": true
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8251,
     "decode": 23.54,
     "prefill": 300.6,
     "ttft": 27.446,
     "step_ms": 126.31,
     "al": 2.974,
     "accept_pct": 0.9869,
     "derived": true
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32825,
     "decode": 17.27,
     "prefill": 275.7,
     "ttft": 119.075,
     "step_ms": 154.77,
     "al": 2.673,
     "accept_pct": 0.8364,
     "derived": true
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131128,
     "decode": 14.35,
     "prefill": 117.4,
     "ttft": 1116.969,
     "step_ms": 187.05,
     "al": 2.684,
     "accept_pct": 0.8419,
     "derived": true
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225339,
     "decode": 9.56,
     "prefill": 66.4,
     "ttft": 3392.35,
     "step_ms": 252.95,
     "al": 2.419,
     "accept_pct": 0.7095,
     "derived": true
    }
   ]
  },
  {
   "key": "qwen3_8_flash_next_q3_k_xl_mtp_engine_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.8-Flash-Next UD-Q3_K_XL",
   "arm": "MTP k=2, Engine",
   "label": "SnowLLM \u00b7 Qwen3.8-Flash-Next UD-Q3_K_XL \u00b7 MTP k=2, Engine",
   "featured": true,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1024,
     "decode": 43.29,
     "prefill": 1324.3,
     "ttft": 0.773,
     "step_ms": 55.01,
     "al": 2.391,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8192,
     "decode": 42.93,
     "prefill": 1812.2,
     "ttft": 4.521,
     "step_ms": 56.03,
     "al": 2.429,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32768,
     "decode": 41.81,
     "prefill": 1842.0,
     "ttft": 17.789,
     "step_ms": 56.47,
     "al": 2.361,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131072,
     "decode": 39.84,
     "prefill": 1775.1,
     "ttft": 73.84,
     "step_ms": 57.93,
     "al": 2.304,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225280,
     "decode": 38.56,
     "prefill": 1723.8,
     "ttft": 130.686,
     "step_ms": 59.12,
     "al": 2.284,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_8_flash_next_q3_k_xl_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.8-Flash-Next UD-Q3_K_XL",
   "arm": "MTP k=2, HTTP",
   "label": "SnowLLM \u00b7 Qwen3.8-Flash-Next UD-Q3_K_XL \u00b7 MTP k=2, HTTP",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1083,
     "decode": 39.13,
     "prefill": 1152.0,
     "ttft": 0.94,
     "step_ms": 56.33,
     "al": 2.208,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8251,
     "decode": 39.09,
     "prefill": 1668.7,
     "ttft": 4.944,
     "step_ms": 57.49,
     "al": 2.239,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32825,
     "decode": 36.13,
     "prefill": 1703.3,
     "ttft": 19.272,
     "step_ms": 57.87,
     "al": 2.118,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131128,
     "decode": 35.57,
     "prefill": 1672.7,
     "ttft": 78.392,
     "step_ms": 59.74,
     "al": 2.171,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225339,
     "decode": 35.61,
     "prefill": 1622.5,
     "ttft": 138.881,
     "step_ms": 60.92,
     "al": 2.191,
     "accept_pct": null,
     "derived": false
    }
   ]
  },
  {
   "key": "qwen3_8_flash_next_q3_k_xl_nospec_engine_snowllm",
   "engine": "SnowLLM",
   "model": "Qwen3.8-Flash-Next UD-Q3_K_XL",
   "arm": "no spec, Engine",
   "label": "SnowLLM \u00b7 Qwen3.8-Flash-Next UD-Q3_K_XL \u00b7 no spec, Engine",
   "featured": false,
   "points": [
    {
     "isl_target": 1024,
     "isl_short": "1K",
     "isl": 1024,
     "decode": 28.3,
     "prefill": 1380.0,
     "ttft": 0.742,
     "step_ms": 35.69,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 8192,
     "isl_short": "8K",
     "isl": 8192,
     "decode": 28.0,
     "prefill": 1874.5,
     "ttft": 4.37,
     "step_ms": 35.75,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 32768,
     "isl_short": "32K",
     "isl": 32768,
     "decode": 27.86,
     "prefill": 1890.8,
     "ttft": 17.331,
     "step_ms": 35.85,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 131072,
     "isl_short": "128K",
     "isl": 131072,
     "decode": 27.25,
     "prefill": 1811.9,
     "ttft": 72.338,
     "step_ms": 36.67,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    },
    {
     "isl_target": 225280,
     "isl_short": "220K",
     "isl": 225280,
     "decode": 26.46,
     "prefill": 1761.8,
     "ttft": 127.87,
     "step_ms": 37.76,
     "al": 1.0,
     "accept_pct": null,
     "derived": false
    }
   ]
  }
 ]
}