Upload benchmarks/bench_09_multi_gpu.json with huggingface_hub
Browse files
benchmarks/bench_09_multi_gpu.json
ADDED
|
@@ -0,0 +1,234 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"benchmark": "multi_gpu",
|
| 3 |
+
"timestamp": "2026-03-19T12:21:06.000529+00:00",
|
| 4 |
+
"n_gpus_available": 4,
|
| 5 |
+
"gpu_names": [
|
| 6 |
+
"NVIDIA L4",
|
| 7 |
+
"NVIDIA L4",
|
| 8 |
+
"NVIDIA L4",
|
| 9 |
+
"NVIDIA L4"
|
| 10 |
+
],
|
| 11 |
+
"inference": {
|
| 12 |
+
"gpu_1": {
|
| 13 |
+
"n_gpus": 1,
|
| 14 |
+
"batch_results": {
|
| 15 |
+
"batch_1": {
|
| 16 |
+
"p50_ms": 128.19,
|
| 17 |
+
"p95_ms": 130.86,
|
| 18 |
+
"mean_ms": 128.42,
|
| 19 |
+
"fps": 7.8,
|
| 20 |
+
"per_sample_ms": 128.42
|
| 21 |
+
},
|
| 22 |
+
"batch_4": {
|
| 23 |
+
"p50_ms": 430.18,
|
| 24 |
+
"p95_ms": 440.71,
|
| 25 |
+
"mean_ms": 429.76,
|
| 26 |
+
"fps": 9.3,
|
| 27 |
+
"per_sample_ms": 107.44
|
| 28 |
+
},
|
| 29 |
+
"batch_8": {
|
| 30 |
+
"p50_ms": 853.99,
|
| 31 |
+
"p95_ms": 873.4,
|
| 32 |
+
"mean_ms": 857.72,
|
| 33 |
+
"fps": 9.3,
|
| 34 |
+
"per_sample_ms": 107.21
|
| 35 |
+
},
|
| 36 |
+
"batch_16": {
|
| 37 |
+
"p50_ms": 1734.01,
|
| 38 |
+
"p95_ms": 1759.55,
|
| 39 |
+
"mean_ms": 1727.97,
|
| 40 |
+
"fps": 9.3,
|
| 41 |
+
"per_sample_ms": 108.0
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"memory": {
|
| 45 |
+
"gpu_0_allocated_gb": 3.65,
|
| 46 |
+
"gpu_0_reserved_gb": 5.07
|
| 47 |
+
}
|
| 48 |
+
},
|
| 49 |
+
"gpu_2": {
|
| 50 |
+
"n_gpus": 2,
|
| 51 |
+
"batch_results": {
|
| 52 |
+
"batch_1": {
|
| 53 |
+
"p50_ms": 162.23,
|
| 54 |
+
"p95_ms": 168.32,
|
| 55 |
+
"mean_ms": 164.65,
|
| 56 |
+
"fps": 6.1,
|
| 57 |
+
"per_sample_ms": 164.65
|
| 58 |
+
},
|
| 59 |
+
"batch_4": {
|
| 60 |
+
"p50_ms": 611.69,
|
| 61 |
+
"p95_ms": 613.66,
|
| 62 |
+
"mean_ms": 611.27,
|
| 63 |
+
"fps": 6.5,
|
| 64 |
+
"per_sample_ms": 152.82
|
| 65 |
+
},
|
| 66 |
+
"batch_8": {
|
| 67 |
+
"p50_ms": 799.07,
|
| 68 |
+
"p95_ms": 802.8,
|
| 69 |
+
"mean_ms": 799.2,
|
| 70 |
+
"fps": 10.0,
|
| 71 |
+
"per_sample_ms": 99.9
|
| 72 |
+
},
|
| 73 |
+
"batch_16": {
|
| 74 |
+
"p50_ms": 1185.3,
|
| 75 |
+
"p95_ms": 1190.2,
|
| 76 |
+
"mean_ms": 1184.72,
|
| 77 |
+
"fps": 13.5,
|
| 78 |
+
"per_sample_ms": 74.04
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"memory": {
|
| 82 |
+
"gpu_0_allocated_gb": 3.66,
|
| 83 |
+
"gpu_0_reserved_gb": 4.47,
|
| 84 |
+
"gpu_1_allocated_gb": 0.01,
|
| 85 |
+
"gpu_1_reserved_gb": 4.47
|
| 86 |
+
}
|
| 87 |
+
},
|
| 88 |
+
"gpu_4": {
|
| 89 |
+
"n_gpus": 4,
|
| 90 |
+
"batch_results": {
|
| 91 |
+
"batch_1": {
|
| 92 |
+
"p50_ms": 164.7,
|
| 93 |
+
"p95_ms": 171.49,
|
| 94 |
+
"mean_ms": 167.37,
|
| 95 |
+
"fps": 6.0,
|
| 96 |
+
"per_sample_ms": 167.37
|
| 97 |
+
},
|
| 98 |
+
"batch_4": {
|
| 99 |
+
"p50_ms": 913.4,
|
| 100 |
+
"p95_ms": 915.7,
|
| 101 |
+
"mean_ms": 912.74,
|
| 102 |
+
"fps": 4.4,
|
| 103 |
+
"per_sample_ms": 228.19
|
| 104 |
+
},
|
| 105 |
+
"batch_8": {
|
| 106 |
+
"p50_ms": 1003.53,
|
| 107 |
+
"p95_ms": 1007.34,
|
| 108 |
+
"mean_ms": 1002.41,
|
| 109 |
+
"fps": 8.0,
|
| 110 |
+
"per_sample_ms": 125.3
|
| 111 |
+
},
|
| 112 |
+
"batch_16": {
|
| 113 |
+
"p50_ms": 1178.17,
|
| 114 |
+
"p95_ms": 1182.78,
|
| 115 |
+
"mean_ms": 1178.2,
|
| 116 |
+
"fps": 13.6,
|
| 117 |
+
"per_sample_ms": 73.64
|
| 118 |
+
}
|
| 119 |
+
},
|
| 120 |
+
"memory": {
|
| 121 |
+
"gpu_0_allocated_gb": 3.66,
|
| 122 |
+
"gpu_0_reserved_gb": 4.49,
|
| 123 |
+
"gpu_1_allocated_gb": 0.01,
|
| 124 |
+
"gpu_1_reserved_gb": 4.16,
|
| 125 |
+
"gpu_2_allocated_gb": 0.01,
|
| 126 |
+
"gpu_2_reserved_gb": 4.16,
|
| 127 |
+
"gpu_3_allocated_gb": 0.01,
|
| 128 |
+
"gpu_3_reserved_gb": 4.16
|
| 129 |
+
}
|
| 130 |
+
}
|
| 131 |
+
},
|
| 132 |
+
"training": {
|
| 133 |
+
"gpu_1": {
|
| 134 |
+
"n_gpus": 1,
|
| 135 |
+
"batch_size": 2,
|
| 136 |
+
"n_steps": 30,
|
| 137 |
+
"step_time_mean_ms": 432.4,
|
| 138 |
+
"steps_per_sec": 2.31,
|
| 139 |
+
"samples_per_sec": 4.63,
|
| 140 |
+
"loss_start": 4.0196,
|
| 141 |
+
"loss_end": 1.7553,
|
| 142 |
+
"loss_reduction_pct": 56.3,
|
| 143 |
+
"memory": {
|
| 144 |
+
"gpu_0_peak_gb": 9.0
|
| 145 |
+
}
|
| 146 |
+
},
|
| 147 |
+
"gpu_2": {
|
| 148 |
+
"n_gpus": 2,
|
| 149 |
+
"batch_size": 4,
|
| 150 |
+
"n_steps": 30,
|
| 151 |
+
"step_time_mean_ms": 1264.8,
|
| 152 |
+
"steps_per_sec": 0.79,
|
| 153 |
+
"samples_per_sec": 3.16,
|
| 154 |
+
"loss_start": 1.3165,
|
| 155 |
+
"loss_end": 1.4857,
|
| 156 |
+
"loss_reduction_pct": -12.9,
|
| 157 |
+
"memory": {
|
| 158 |
+
"gpu_0_peak_gb": 14.59,
|
| 159 |
+
"gpu_1_peak_gb": 4.07
|
| 160 |
+
}
|
| 161 |
+
},
|
| 162 |
+
"gpu_4": {
|
| 163 |
+
"n_gpus": 4,
|
| 164 |
+
"batch_size": 8,
|
| 165 |
+
"n_steps": 30,
|
| 166 |
+
"step_time_mean_ms": 2005.1,
|
| 167 |
+
"steps_per_sec": 0.5,
|
| 168 |
+
"samples_per_sec": 3.99,
|
| 169 |
+
"loss_start": 6.7918,
|
| 170 |
+
"loss_end": 1.182,
|
| 171 |
+
"loss_reduction_pct": 82.6,
|
| 172 |
+
"memory": {
|
| 173 |
+
"gpu_0_peak_gb": 14.6,
|
| 174 |
+
"gpu_1_peak_gb": 4.07,
|
| 175 |
+
"gpu_2_peak_gb": 4.07,
|
| 176 |
+
"gpu_3_peak_gb": 4.07
|
| 177 |
+
}
|
| 178 |
+
}
|
| 179 |
+
},
|
| 180 |
+
"fp16": {
|
| 181 |
+
"fp16_gpu_1": {
|
| 182 |
+
"n_gpus": 1,
|
| 183 |
+
"precision": "fp16",
|
| 184 |
+
"batch_results": {
|
| 185 |
+
"batch_4": {
|
| 186 |
+
"p50_ms": 122.14,
|
| 187 |
+
"fps": 32.7,
|
| 188 |
+
"per_sample_ms": 30.56
|
| 189 |
+
},
|
| 190 |
+
"batch_8": {
|
| 191 |
+
"p50_ms": 234.11,
|
| 192 |
+
"fps": 34.2,
|
| 193 |
+
"per_sample_ms": 29.26
|
| 194 |
+
},
|
| 195 |
+
"batch_16": {
|
| 196 |
+
"p50_ms": 486.22,
|
| 197 |
+
"fps": 32.9,
|
| 198 |
+
"per_sample_ms": 30.43
|
| 199 |
+
},
|
| 200 |
+
"batch_32": {
|
| 201 |
+
"p50_ms": 950.24,
|
| 202 |
+
"fps": 33.6,
|
| 203 |
+
"per_sample_ms": 29.73
|
| 204 |
+
}
|
| 205 |
+
}
|
| 206 |
+
},
|
| 207 |
+
"fp16_gpu_4": {
|
| 208 |
+
"n_gpus": 4,
|
| 209 |
+
"precision": "fp16",
|
| 210 |
+
"batch_results": {
|
| 211 |
+
"batch_4": {
|
| 212 |
+
"p50_ms": 901.49,
|
| 213 |
+
"fps": 4.4,
|
| 214 |
+
"per_sample_ms": 225.52
|
| 215 |
+
},
|
| 216 |
+
"batch_8": {
|
| 217 |
+
"p50_ms": 903.67,
|
| 218 |
+
"fps": 8.8,
|
| 219 |
+
"per_sample_ms": 113.02
|
| 220 |
+
},
|
| 221 |
+
"batch_16": {
|
| 222 |
+
"p50_ms": 911.97,
|
| 223 |
+
"fps": 17.5,
|
| 224 |
+
"per_sample_ms": 57.07
|
| 225 |
+
},
|
| 226 |
+
"batch_32": {
|
| 227 |
+
"p50_ms": 1013.11,
|
| 228 |
+
"fps": 31.6,
|
| 229 |
+
"per_sample_ms": 31.67
|
| 230 |
+
}
|
| 231 |
+
}
|
| 232 |
+
}
|
| 233 |
+
}
|
| 234 |
+
}
|