apple coreml - YingkunZhou/EdgeTransformerBench GitHub Wiki
- M1 (8-core GPU)
- M1 max (32-core GPU)
- M2 (8-core GPU)
Converting Deep Learning Models
- torch==2.2.0
- coremltools==7.1
- compute: Mixed (FLoat16, Float32, Int32)
- Storage: Float16
- OS: macOS 14.1
注意：笔记本电脑测试的时候要接电源！
| Model | M1 Mapping(C/G/N) M2 Mapping(C/G/N) |
ALL mini/MAX |
GPU mini/MAX |
CPU | #params | GMACs |
|---|---|---|---|---|---|---|
| efficientformerv2_s0 | 4/0/213 0/0/217 |
0.73/0.68 0.51 |
4.44/4.69 4.08 |
50.15 48.27 |
3.5M | 0.40G |
| efficientformerv2_s1 | 4/0/255 0/0/259 |
0.86/0.81 0.62 |
4.43/4.36 5.04 |
73.93 71.37 |
6.1M | 0.65G |
| efficientformerv2_s2 | 4/0/389 0/0/393 |
1.26/1.25 0.94 |
5.92/4.60 6.40 |
112.40 108.27 |
12.6M | 1.25G |
| SwiftFormer_XS | 4/0/246 0/0/250 |
0.81/0.76 0.63 |
4.22/4.52 5.21 |
50.95 48.93 |
3.5M | 0.4G |
| SwiftFormer_S | 4/0/271 0/0/275 |
0.95/0.88 0.73 |
5.25/4.18 5.00 |
62.24 59.70 |
6.1M | 1.0G |
| SwiftFormer_L1 | 4/0/276 0/0/280 |
1.21/1.18 0.98 |
6.20/4.42 5.60 |
81.57 78.08 |
12.1M | 1.6G |
| EMO_1M | 4/0/346 0/0/350 |
2.47/3.93 2.70 |
4.12/4.43 6.43 |
20.38 19.92 |
1.3M | 0.26G |
| EMO_2M | 4/0/394 0/0/398 |
3.70/5.69 4.29 |
4.79/4.27 7.20 |
28.61 27.65 |
2.3M | 0.44G |
| EMO_5M | 4/0/394 0/0/398 |
6.51/6.93 7.80 |
6.07/4.46 7.46 |
44.52 42.90 |
5.1M | 0.90G |
| EMO_6M | 4/0/394 0/0/398 |
6.81/7.59 8.02 |
6.25/4.49 7.43 |
47.13 46.13 |
6.1M | 0.96G |
| edgenext_xx_small | 2/163/110 0/167/108 |
4.28/5.94 4.69 |
3.66/4.27 5.49 |
44.31 43.09 |
1.3M | 0.26G |
| edgenext_x_small | 2/199/128 0/203/126 |
5.30/8.24 6.81 |
4.47/4.64 6.45 |
72.47 69.26 |
2.3M | 0.54G |
| edgenext_small/usi | 2/267/60 0/199/130 |
6.91/9.56 9.11 |
4.63/4.45 7.41 |
118.40 112.80 |
5.6M | 1.26G |
| mobilevitv2_050 | 4/0/396 0/0/400 |
0.84/0.80 0.64 |
4.18/4.23 4.87 |
19.07 18.13 |
1.4M | 0.5G |
| mobilevitv2_075 | 4/0/396 0/0/400 |
1.15/1.13 0.93 |
4.88/4.26 5.75 |
28.20 26.97 |
2.9M | 1.0G |
| mobilevitv2_100 | 4/0/396 0/0/400 |
1.64/1.64 1.32 |
5.99/4.29 6.38 |
37.84 36.05 |
4.9M | 1.8G |
| mobilevitv2_125 | 4/0/396 0/0/400 |
1.98/2.04 1.62 |
8.73/4.34 7.66 |
48.27 45.56 |
7.5M | 2.8G |
| mobilevitv2_150 | 4/0/396 0/0/400 |
2.43/2.50 2.06 |
10.33/4.93 9.13 |
58.72 56.20 |
10.6M | 4.0G |
| mobilevitv2_175 | 4/0/396 0/0/400 |
2.88/2.96 2.46 |
13.52/5.71 11.88 |
70.38 65.23 |
14.3M | 5.5G |
| mobilevitv2_200 | 4/0/396 0/0/400 |
3.34/3.46 2.88 |
14.84/6.22 13.00 |
81.28 76.22 |
18.4M | 7.2G |
| mobilevit_xx_small | 4/0/381 0/0/385 |
1.44/1.51 1.30 |
4.23/4.35 5.74 |
15.54 15.07 |
1.3M | 0.36G |
| mobilevit_x_small | 4/0/380 0/0/384 |
2.04/2.34 1.79 |
6.04/4.28 7.19 |
28.73 28.20 |
2.3M | 0.89G |
| mobilevit_small | 4/0/380 0/0/384 |
2.74/3.05 2.48 |
7.62/4.79 7.36 |
35.77 34.71 |
5.6M | 2.0 G |
| LeViT_128S | 2/266/30 0/273/25 |
3.83/5.63 5.32 |
3.37/4.08 3.68 |
6.15 5.86 |
7.8M | 0.30G |
| LeViT_128 | 2/338/30 0/345/25 |
4.65/6.76 6.46 |
4.16/4.13 3.64 |
7.44 7.25 |
9.2M | 0.41G |
| LeViT_192 | 2/339/29 0/337/33 |
5.48/7.58 7.39 |
4.16/4.19 4.61 |
8.90 8.82 |
11 M | 0.66G |
| LeViT_256 | 2/236/132 0/235/135 |
4.85/8.39 6.72 |
4.13/4.30 6.70 |
13.62 13.25 |
19 M | 1.12G |
| resnet50 | 4/0/122 0/0/126 |
1.67/1.67 1.33 |
8.32/3.92 6.83 |
12.83 12.11 |
25.6M | 4.1G |
| mobilenetv3_large_100 | 4/0/182 0/0/186 |
0.59/0.53 0.41 |
4.49/2.01 5.06 |
5.90 4.49 |
5.5M | 0.29G |
| tf_efficientnetv2_b0 | 4/0/217 0/0/221 |
0.69/0.65 0.51 |
4.62/4.32 6.35 |
11.40 10.64 |
7.1M | 0.72G |
| tf_efficientnetv2_b1 | 4/0/264 0/0/268 |
0.83/0.79 0.63 |
5.10/4.38 7.15 |
14.23 14.42 |
8.1M | 1.2G |
| tf_efficientnetv2_b2 | 4/0/276 0/0/280 |
1.34/1.25 0.88 |
7.27/4.56 7.25 |
20.08 19.81 |
10.1M | 1.7G |
| tf_efficientnetv2_b3 | 4/0/324 0/0/328 |
1.82/1.72 1.22 |
11.33/6.08 10.13 |
32.04 31.65 |
14.4M | 3.0G |