Upload benchmark_headtohead.json with huggingface_hub
Browse files- benchmark_headtohead.json +242 -0
benchmark_headtohead.json
ADDED
|
@@ -0,0 +1,242 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": [
|
| 3 |
+
{
|
| 4 |
+
"name": "BitNet b1.58 2B-4T (I2_S)",
|
| 5 |
+
"model_path": "/home/cpinchington/MedusaBitNet/models/bitnet-b1.58-2B-4T/ggml-model-i2_s.gguf",
|
| 6 |
+
"model_size_mb": 1187.310112,
|
| 7 |
+
"n_runs": 8,
|
| 8 |
+
"avg_gen_tok_s": 72.73875,
|
| 9 |
+
"avg_prefill_tok_s": 434.95375,
|
| 10 |
+
"avg_ms_per_tok": 13.75375,
|
| 11 |
+
"runs": [
|
| 12 |
+
{
|
| 13 |
+
"gen_tok_s": 75.47,
|
| 14 |
+
"gen_ms_per_tok": 13.25,
|
| 15 |
+
"prefill_tok_s": 473.97,
|
| 16 |
+
"wall_time": 3.8744120597839355
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"gen_tok_s": 74.61,
|
| 20 |
+
"gen_ms_per_tok": 13.4,
|
| 21 |
+
"prefill_tok_s": 400.33,
|
| 22 |
+
"wall_time": 3.946519613265991
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"gen_tok_s": 73.39,
|
| 26 |
+
"gen_ms_per_tok": 13.63,
|
| 27 |
+
"prefill_tok_s": 468.56,
|
| 28 |
+
"wall_time": 3.981218099594116
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"gen_tok_s": 71.47,
|
| 32 |
+
"gen_ms_per_tok": 13.99,
|
| 33 |
+
"prefill_tok_s": 403.88,
|
| 34 |
+
"wall_time": 4.082836866378784
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"gen_tok_s": 71.75,
|
| 38 |
+
"gen_ms_per_tok": 13.94,
|
| 39 |
+
"prefill_tok_s": 425.03,
|
| 40 |
+
"wall_time": 4.0725319385528564
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"gen_tok_s": 71.9,
|
| 44 |
+
"gen_ms_per_tok": 13.91,
|
| 45 |
+
"prefill_tok_s": 399.68,
|
| 46 |
+
"wall_time": 4.0745580196380615
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"gen_tok_s": 71.23,
|
| 50 |
+
"gen_ms_per_tok": 14.04,
|
| 51 |
+
"prefill_tok_s": 466.73,
|
| 52 |
+
"wall_time": 4.097683429718018
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"gen_tok_s": 72.09,
|
| 56 |
+
"gen_ms_per_tok": 13.87,
|
| 57 |
+
"prefill_tok_s": 441.45,
|
| 58 |
+
"wall_time": 4.053175687789917
|
| 59 |
+
}
|
| 60 |
+
]
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"name": "Qwen2.5 1.5B (Q4_K_M)",
|
| 64 |
+
"model_path": "/home/cpinchington/MedusaBitNet/models/competing/qwen2.5-1.5b-instruct-q4_k_m.gguf",
|
| 65 |
+
"model_size_mb": 1117.320736,
|
| 66 |
+
"n_runs": 8,
|
| 67 |
+
"avg_gen_tok_s": 88.79125,
|
| 68 |
+
"avg_prefill_tok_s": 317.92375,
|
| 69 |
+
"avg_ms_per_tok": 11.2625,
|
| 70 |
+
"runs": [
|
| 71 |
+
{
|
| 72 |
+
"gen_tok_s": 88.83,
|
| 73 |
+
"gen_ms_per_tok": 11.26,
|
| 74 |
+
"prefill_tok_s": 277.6,
|
| 75 |
+
"wall_time": 3.4317729473114014
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"gen_tok_s": 88.81,
|
| 79 |
+
"gen_ms_per_tok": 11.26,
|
| 80 |
+
"prefill_tok_s": 349.13,
|
| 81 |
+
"wall_time": 3.419360637664795
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"gen_tok_s": 87.75,
|
| 85 |
+
"gen_ms_per_tok": 11.4,
|
| 86 |
+
"prefill_tok_s": 362.62,
|
| 87 |
+
"wall_time": 3.448280096054077
|
| 88 |
+
},
|
| 89 |
+
{
|
| 90 |
+
"gen_tok_s": 88.93,
|
| 91 |
+
"gen_ms_per_tok": 11.24,
|
| 92 |
+
"prefill_tok_s": 371.15,
|
| 93 |
+
"wall_time": 2.9640591144561768
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"gen_tok_s": 88.83,
|
| 97 |
+
"gen_ms_per_tok": 11.26,
|
| 98 |
+
"prefill_tok_s": 278.95,
|
| 99 |
+
"wall_time": 3.4554200172424316
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"gen_tok_s": 89.49,
|
| 103 |
+
"gen_ms_per_tok": 11.17,
|
| 104 |
+
"prefill_tok_s": 271.77,
|
| 105 |
+
"wall_time": 2.8726541996002197
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"gen_tok_s": 88.94,
|
| 109 |
+
"gen_ms_per_tok": 11.24,
|
| 110 |
+
"prefill_tok_s": 278.1,
|
| 111 |
+
"wall_time": 3.42142915725708
|
| 112 |
+
},
|
| 113 |
+
{
|
| 114 |
+
"gen_tok_s": 88.75,
|
| 115 |
+
"gen_ms_per_tok": 11.27,
|
| 116 |
+
"prefill_tok_s": 354.07,
|
| 117 |
+
"wall_time": 3.4249227046966553
|
| 118 |
+
}
|
| 119 |
+
]
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"name": "Llama 3.2 1B (Q4_K_M)",
|
| 123 |
+
"model_path": "/home/cpinchington/MedusaBitNet/models/competing/Llama-3.2-1B-Instruct-Q4_K_M.gguf",
|
| 124 |
+
"model_size_mb": 807.694464,
|
| 125 |
+
"n_runs": 8,
|
| 126 |
+
"avg_gen_tok_s": 115.94624999999999,
|
| 127 |
+
"avg_prefill_tok_s": 440.73375,
|
| 128 |
+
"avg_ms_per_tok": 8.62625,
|
| 129 |
+
"runs": [
|
| 130 |
+
{
|
| 131 |
+
"gen_tok_s": 115.23,
|
| 132 |
+
"gen_ms_per_tok": 8.68,
|
| 133 |
+
"prefill_tok_s": 373.11,
|
| 134 |
+
"wall_time": 3.3805642127990723
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"gen_tok_s": 115.66,
|
| 138 |
+
"gen_ms_per_tok": 8.65,
|
| 139 |
+
"prefill_tok_s": 499.09,
|
| 140 |
+
"wall_time": 3.3459088802337646
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"gen_tok_s": 115.25,
|
| 144 |
+
"gen_ms_per_tok": 8.68,
|
| 145 |
+
"prefill_tok_s": 505.72,
|
| 146 |
+
"wall_time": 3.353907346725464
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"gen_tok_s": 116.84,
|
| 150 |
+
"gen_ms_per_tok": 8.56,
|
| 151 |
+
"prefill_tok_s": 324.38,
|
| 152 |
+
"wall_time": 3.3358867168426514
|
| 153 |
+
},
|
| 154 |
+
{
|
| 155 |
+
"gen_tok_s": 115.81,
|
| 156 |
+
"gen_ms_per_tok": 8.63,
|
| 157 |
+
"prefill_tok_s": 531.45,
|
| 158 |
+
"wall_time": 3.3553287982940674
|
| 159 |
+
},
|
| 160 |
+
{
|
| 161 |
+
"gen_tok_s": 116.16,
|
| 162 |
+
"gen_ms_per_tok": 8.61,
|
| 163 |
+
"prefill_tok_s": 524.34,
|
| 164 |
+
"wall_time": 3.340409278869629
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"gen_tok_s": 117.13,
|
| 168 |
+
"gen_ms_per_tok": 8.54,
|
| 169 |
+
"prefill_tok_s": 341.84,
|
| 170 |
+
"wall_time": 3.348862648010254
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"gen_tok_s": 115.49,
|
| 174 |
+
"gen_ms_per_tok": 8.66,
|
| 175 |
+
"prefill_tok_s": 425.94,
|
| 176 |
+
"wall_time": 3.349743366241455
|
| 177 |
+
}
|
| 178 |
+
]
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"name": "Gemma 2 2B (Q4_K_M)",
|
| 182 |
+
"model_path": "/home/cpinchington/MedusaBitNet/models/competing/gemma-2-2b-it-Q4_K_M.gguf",
|
| 183 |
+
"model_size_mb": 1708.582752,
|
| 184 |
+
"n_runs": 8,
|
| 185 |
+
"avg_gen_tok_s": 50.53125,
|
| 186 |
+
"avg_prefill_tok_s": 200.96,
|
| 187 |
+
"avg_ms_per_tok": 19.7875,
|
| 188 |
+
"runs": [
|
| 189 |
+
{
|
| 190 |
+
"gen_tok_s": 50.47,
|
| 191 |
+
"gen_ms_per_tok": 19.81,
|
| 192 |
+
"prefill_tok_s": 184.03,
|
| 193 |
+
"wall_time": 5.728861331939697
|
| 194 |
+
},
|
| 195 |
+
{
|
| 196 |
+
"gen_tok_s": 50.67,
|
| 197 |
+
"gen_ms_per_tok": 19.73,
|
| 198 |
+
"prefill_tok_s": 229.99,
|
| 199 |
+
"wall_time": 5.6800384521484375
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"gen_tok_s": 50.37,
|
| 203 |
+
"gen_ms_per_tok": 19.85,
|
| 204 |
+
"prefill_tok_s": 165.91,
|
| 205 |
+
"wall_time": 5.7390992641448975
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"gen_tok_s": 50.7,
|
| 209 |
+
"gen_ms_per_tok": 19.72,
|
| 210 |
+
"prefill_tok_s": 174.06,
|
| 211 |
+
"wall_time": 5.694071292877197
|
| 212 |
+
},
|
| 213 |
+
{
|
| 214 |
+
"gen_tok_s": 50.63,
|
| 215 |
+
"gen_ms_per_tok": 19.75,
|
| 216 |
+
"prefill_tok_s": 190.64,
|
| 217 |
+
"wall_time": 5.7339208126068115
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"gen_tok_s": 50.33,
|
| 221 |
+
"gen_ms_per_tok": 19.87,
|
| 222 |
+
"prefill_tok_s": 253.86,
|
| 223 |
+
"wall_time": 5.723286867141724
|
| 224 |
+
},
|
| 225 |
+
{
|
| 226 |
+
"gen_tok_s": 50.52,
|
| 227 |
+
"gen_ms_per_tok": 19.79,
|
| 228 |
+
"prefill_tok_s": 233.19,
|
| 229 |
+
"wall_time": 5.69863224029541
|
| 230 |
+
},
|
| 231 |
+
{
|
| 232 |
+
"gen_tok_s": 50.56,
|
| 233 |
+
"gen_ms_per_tok": 19.78,
|
| 234 |
+
"prefill_tok_s": 176.0,
|
| 235 |
+
"wall_time": 5.730913162231445
|
| 236 |
+
}
|
| 237 |
+
]
|
| 238 |
+
}
|
| 239 |
+
],
|
| 240 |
+
"hardware": "AMD Ryzen AI MAX+ 395 (Strix Halo)",
|
| 241 |
+
"threads": 16
|
| 242 |
+
}
|