202 lines
3.8 KiB
JSON
202 lines
3.8 KiB
JSON
{
|
|||
|
|
"device": "NVIDIA GeForce RTX 3090",
|
||
|
|
"torch": "2.10.0+cu130",
|
||
|
|
"act_scale": 1.0,
|
||
|
|
"cfg": {},
|
||
|
|
"B": 1,
|
||
|
|
"commit_n": 2,
|
||
|
|
"fp64_note": "3090 fp64 \u541e\u5410\u7ea6 fp32 \u7684 1/54\uff1b\u8ba4\u8bc1\u76ee\u524d\u8981\u6c42 fp64",
|
||
|
|
"inference": {
|
||
|
|
"sample_K1_steps1": {
|
||
|
|
"p50": 7.453,
|
||
|
|
"p95": 9.912,
|
||
|
|
"p99": 10.766,
|
||
|
|
"max": 11.109,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K8_steps1": {
|
||
|
|
"p50": 8.208,
|
||
|
|
"p95": 10.685,
|
||
|
|
"p99": 11.22,
|
||
|
|
"max": 15.224,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K1_steps2": {
|
||
|
|
"p50": 13.149,
|
||
|
|
"p95": 17.621,
|
||
|
|
"p99": 18.438,
|
||
|
|
"max": 19.296,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K8_steps2": {
|
||
|
|
"p50": 14.167,
|
||
|
|
"p95": 19.241,
|
||
|
|
"p99": 24.77,
|
||
|
|
"max": 25.703,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K1_steps4": {
|
||
|
|
"p50": 19.193,
|
||
|
|
"p95": 22.338,
|
||
|
|
"p99": 23.666,
|
||
|
|
"max": 25.787,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K8_steps4": {
|
||
|
|
"p50": 21.509,
|
||
|
|
"p95": 24.732,
|
||
|
|
"p99": 35.655,
|
||
|
|
"max": 41.514,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K1_steps8": {
|
||
|
|
"p50": 40.658,
|
||
|
|
"p95": 55.162,
|
||
|
|
"p99": 60.394,
|
||
|
|
"max": 60.398,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K8_steps8": {
|
||
|
|
"p50": 40.416,
|
||
|
|
"p95": 47.997,
|
||
|
|
"p99": 62.49,
|
||
|
|
"max": 71.976,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K1_steps16": {
|
||
|
|
"p50": 72.429,
|
||
|
|
"p95": 90.188,
|
||
|
|
"p99": 107.536,
|
||
|
|
"max": 112.674,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"sample_K8_steps16": {
|
||
|
|
"p50": 79.893,
|
||
|
|
"p95": 99.933,
|
||
|
|
"p99": 105.746,
|
||
|
|
"max": 106.078,
|
||
|
|
"reps": 120
|
||
|
|
}
|
||
|
|
},
|
||
|
|
"encode_once": {
|
||
|
|
"p50": 2.265,
|
||
|
|
"p95": 4.001,
|
||
|
|
"p99": 6.086,
|
||
|
|
"max": 6.136,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"tiny_op_ms": 0.028,
|
||
|
|
"sync_only_ms": 0.005,
|
||
|
|
"components": {
|
||
|
|
"features_f32": {
|
||
|
|
"p50": 12.443,
|
||
|
|
"p95": 14.227,
|
||
|
|
"p99": 19.678,
|
||
|
|
"max": 20.531,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"endpoint_score_K8_f32": {
|
||
|
|
"p50": 4.278,
|
||
|
|
"p95": 5.841,
|
||
|
|
"p99": 6.322,
|
||
|
|
"max": 8.883,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"dls_fb_1step_f64": {
|
||
|
|
"p50": 7.191,
|
||
|
|
"p95": 12.26,
|
||
|
|
"p99": 14.642,
|
||
|
|
"max": 15.384,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"dls_fb_1step_f32": {
|
||
|
|
"p50": 7.619,
|
||
|
|
"p95": 10.142,
|
||
|
|
"p99": 11.597,
|
||
|
|
"max": 12.022,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"discrete_cert_4pts_f32": {
|
||
|
|
"p50": 8.214,
|
||
|
|
"p95": 13.365,
|
||
|
|
"p99": 14.235,
|
||
|
|
"max": 14.991,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cert_n8_f64_2roots": {
|
||
|
|
"p50": 11.265,
|
||
|
|
"p95": 14.64,
|
||
|
|
"p99": 16.788,
|
||
|
|
"max": 17.761,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cert_n16_f64_2roots": {
|
||
|
|
"p50": 14.574,
|
||
|
|
"p95": 16.731,
|
||
|
|
"p99": 18.459,
|
||
|
|
"max": 22.346,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cert_n32_f64_2roots": {
|
||
|
|
"p50": 21.528,
|
||
|
|
"p95": 24.746,
|
||
|
|
"p99": 28.037,
|
||
|
|
"max": 28.066,
|
||
|
|
"reps": 120
|
||
|
|
}
|
||
|
|
},
|
||
|
|
"cycle_K8_steps16_discrete": {
|
||
|
|
"p50": 116.474,
|
||
|
|
"p95": 134.474,
|
||
|
|
"p99": 142.593,
|
||
|
|
"max": 152.768,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K8_steps4_discrete": {
|
||
|
|
"p50": 55.919,
|
||
|
|
"p95": 65.767,
|
||
|
|
"p99": 74.809,
|
||
|
|
"max": 78.059,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K1_steps4_discrete": {
|
||
|
|
"p50": 51.087,
|
||
|
|
"p95": 58.614,
|
||
|
|
"p99": 62.326,
|
||
|
|
"max": 63.919,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K1_steps1_discrete": {
|
||
|
|
"p50": 34.493,
|
||
|
|
"p95": 42.078,
|
||
|
|
"p99": 43.906,
|
||
|
|
"max": 50.268,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K8_steps16_cert_n8": {
|
||
|
|
"p50": 132.027,
|
||
|
|
"p95": 150.463,
|
||
|
|
"p99": 160.83,
|
||
|
|
"max": 192.586,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K8_steps4_cert_n8": {
|
||
|
|
"p50": 64.812,
|
||
|
|
"p95": 79.192,
|
||
|
|
"p99": 89.629,
|
||
|
|
"max": 95.837,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"cycle_K1_steps4_cert_n16": {
|
||
|
|
"p50": 61.967,
|
||
|
|
"p95": 73.219,
|
||
|
|
"p99": 79.053,
|
||
|
|
"max": 83.18,
|
||
|
|
"reps": 120
|
||
|
|
},
|
||
|
|
"f32_vs_f64": {
|
||
|
|
"points": 3,
|
||
|
|
"max_abs_err_mm": 0.0,
|
||
|
|
"sign_flips": 0
|
||
|
|
}
|
||
|
|
}
|