Skip to content

Instantly share code, notes, and snippets.

@rwightman
Last active July 13, 2022 03:55
Show Gist options
  • Save rwightman/8b2c72d3e02147f4be75bb8165a588a7 to your computer and use it in GitHub Desktop.
Save rwightman/8b2c72d3e02147f4be75bb8165a588a7 to your computer and use it in GitHub Desktop.

comparing ln

Comparing some LayerNorm for 2D rank-4 NCHW tensors via ConvNeXt models on 3090 and V100.

All runs done with native torch AMP, PyTorch 1.12 cu113.

Some col descriptions

Models with _c use conv_mlp=True arg to use 1x1 convs instead of nn.Linear for MLP blocks. While for ConvNeXt they are usually worse across the board, this mode does better represent a mix of layers found in some other models where this LN is useful... (where it could be a kxk conv, 2D pool layers, stride > 1, etc mixed in the sequence).

csvs w/ all combine all cogegen, with eager are showing just eager (with the ts of layer only included for rw_hack). Sorted by inference and train throughputs...

model fmt cg layer infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count infer_gmacs infer_macts
convnext_tiny nhwc torchscript permute 3364.46 76.075 256 224 877.71 144.926 128 224 28.59
convnext_tiny nchw torchscript permute 3330.51 76.85 256 224 861.39 147.693 128 224 28.59
convnext_tiny nhwc torchscript rw_hack 3299.9 77.564 256 224 765.84 166.191 128 224 28.59
convnext_tiny nchw torchscript rw_hack 3277.37 78.096 256 224 775.44 164.174 128 224 28.59
convnext_tiny_c nhwc torchscript permute 2837.58 90.203 256 224 811.07 156.88 128 224 28.59
convnext_tiny_c nchw torchscript permute 2805.44 91.234 256 224 798.45 159.418 128 224 28.59
convnext_tiny nhwc aot ng_hack 2743.59 93.293 256 224 827.46 153.853 128 224 28.59
convnext_tiny nhwc aot rw_hack 2729.9 93.761 256 224 734.61 173.385 128 224 28.59
convnext_tiny_c nhwc torchscript rw_hack 2699.6 94.813 256 224 601.07 211.979 128 224 28.59
convnext_tiny_c nchw torchscript rw_hack 2694.33 94.999 256 224 605.98 210.22 128 224 28.59
convnext_tiny nhwc aot permute 2668.55 95.917 256 224 819.95 155.287 128 224 28.59
convnext_tiny nchw aot rw_hack 2651.38 96.539 256 224 782.49 162.762 128 224 28.59
convnext_tiny nchw aot permute 2634.11 97.171 256 224 788.72 161.334 128 224 28.59
convnext_tiny_c nhwc aot ng_hack 2607.56 98.162 256 224 799.89 319.219 256 224 28.59
convnext_tiny_c nhwc aot rw_hack 2597.44 98.543 256 224 606.81 420.99 256 224 28.59
convnext_tiny_c nchw aot rw_hack 2552.02 100.299 256 224 647.69 394.423 256 224 28.59
convnext_tiny nchw aot ng_hack 2509.07 102.013 256 224 763.84 166.723 128 224 28.59
convnext_tiny_c nhwc eager rw_hack_ts 2487.57 102.897 256 224 646.34 395.096 256 224 28.59
convnext_tiny_c nchw aot permute 2382.17 107.45 256 224 770.43 331.449 256 224 28.59
convnext_tiny_c nhwc aot permute 2309.24 110.833 256 224 747.49 341.568 256 224 28.59
convnext_tiny nhwc eager rw_hack_ts 2298.36 111.369 256 224 724.19 352.506 256 224 28.59
convnext_tiny_c nhwc eager permute 2288.94 111.828 256 224 794.23 321.435 256 224 28.59 4.47 13.44
convnext_tiny_c nchw eager permute 2285.72 111.986 256 224 790.3 323.035 256 224 28.59 4.47 13.44
convnext_tiny nhwc eager permute 2250.61 113.733 256 224 792.56 322.103 256 224 28.59 4.47 13.44
convnext_tiny nchw eager permute 2232.09 114.674 256 224 791.37 322.585 256 224 28.59 4.47 13.44
convnext_tiny nhwc eager ng_hack 2217.22 115.445 256 224 787.31 324.191 256 224 28.59
convnext_tiny nchw eager rw_hack_ts 2195.18 116.602 256 224 766.21 333.081 256 224 28.59
convnext_tiny nchw eager rw_hack 2165.01 118.229 256 224 759.56 336.101 256 224 28.59
convnext_tiny nhwc eager rw_hack 2163.16 118.331 256 224 751.81 339.603 256 224 28.59
convnext_tiny_c nhwc eager ng_hack 2162.07 118.39 256 224 763.11 334.57 256 224 28.59
convnext_tiny nchw eager ng_hack 2100.81 121.843 256 224 733.75 347.957 256 224 28.59
convnext_tiny_c nchw eager rw_hack 2090.56 122.44 256 224 685.44 372.539 256 224 28.59
convnext_tiny_c nhwc eager rw_hack 2080.55 123.031 256 224 680.73 375.149 256 224 28.59
convnext_tiny_c nchw eager rw_hack_ts 1999.74 128.002 256 224 671.62 380.171 256 224 28.59
convnext_small nhwc torchscript permute 1927.41 132.805 256 224 504.01 252.232 128 224 50.22
convnext_small nchw torchscript permute 1916.93 133.531 256 224 508.55 249.996 128 224 50.22
convnext_small nhwc torchscript rw_hack 1906.58 134.258 256 224 473.71 268.535 128 224 50.22
convnext_small nchw torchscript rw_hack 1898.71 134.811 256 224 477.39 266.497 128 224 50.22
convnext_small_c nchw torchscript permute 1652.98 154.855 256 224 473.78 268.408 128 224 50.22
convnext_small nhwc aot ng_hack 1614.36 158.559 256 224 498.73 255.192 128 224 50.22
convnext_small nhwc aot rw_hack 1613.63 158.631 256 224 465.18 273.701 128 224 50.22
convnext_small_c nhwc torchscript rw_hack 1607.13 159.274 256 224 346.62 182.939 64 224 50.22
convnext_small_c nchw torchscript rw_hack 1604.59 159.525 256 224 349.37 181.504 64 224 50.22
convnext_small_c nhwc torchscript permute 1595.57 160.421 256 224 458.92 277.03 128 224 50.22
convnext_small nhwc aot permute 1590.94 160.892 256 224 493.49 257.925 128 224 50.22
convnext_small nchw aot rw_hack 1587.74 161.22 256 224 481.35 264.473 128 224 50.22
convnext_small nchw aot permute 1564.47 163.612 256 224 490.25 259.414 128 224 50.22
convnext_small_c nhwc aot rw_hack 1548.09 165.347 256 224 371.33 343.245 128 224 50.22
convnext_small nchw aot ng_hack 1532.08 167.074 256 224 474.15 268.507 128 224 50.22
convnext_small_c nchw aot rw_hack 1530.51 167.249 256 224 380.85 334.595 128 224 50.22
convnext_small_c nhwc aot ng_hack 1521.95 168.187 256 224 457.44 278.213 128 224 50.22
convnext_small_c nhwc eager rw_hack_ts 1479.75 172.985 256 224 371.02 343.251 128 224 50.22
convnext_small_c nhwc aot permute 1451.85 176.308 256 224 457.4 278.366 128 224 50.22
convnext_small_c nchw aot permute 1442.18 177.493 256 224 453.51 280.621 128 224 50.22
convnext_tiny_c nchw aot ng_hack 1416.88 180.661 256 224 438.08 583.537 256 224 28.59
convnext_small_c nchw eager permute 1389.39 184.239 256 224 462.57 275.083 128 224 50.22 8.71 21.56
convnext_small_c nhwc eager permute 1387.23 184.523 256 224 464.43 274.049 128 224 50.22 8.71 21.56
convnext_tiny_c nchw eager ng_hack 1375.98 186.032 256 224 441.37 579.066 256 224 28.59
convnext_small nhwc eager rw_hack_ts 1367.93 187.128 256 224 438.44 290.306 128 224 50.22
convnext_small nchw eager permute 1351.2 189.446 256 224 461.49 275.767 128 224 50.22 8.71 21.56
convnext_small nhwc eager ng_hack 1339.83 191.053 256 224 461.05 276.022 128 224 50.22
convnext_small nchw eager rw_hack_ts 1338.01 191.312 256 224 453.84 280.375 128 224 50.22
convnext_small nhwc eager permute 1333.5 191.956 256 224 461.32 275.83 128 224 50.22 8.71 21.56
convnext_small nchw eager rw_hack 1328.35 192.703 256 224 450.11 282.768 128 224 50.22
convnext_small nhwc eager rw_hack 1324.29 193.296 256 224 447.81 284.127 128 224 50.22
convnext_small_c nhwc eager ng_hack 1321.24 193.739 256 224 448.36 283.776 128 224 50.22
convnext_small nchw eager ng_hack 1293.56 197.885 256 224 441.5 288.235 128 224 50.22
convnext_base nchw torchscript permute 1287.32 198.845 256 224 322.5 196.805 64 224 88.59
convnext_base nhwc torchscript rw_hack 1283.11 199.497 256 224 297.33 213.607 64 224 88.59
convnext_base nchw torchscript rw_hack 1278.18 200.266 256 224 306.72 206.975 64 224 88.59
convnext_small_c nchw eager rw_hack 1269.52 201.632 256 224 398.95 319.187 128 224 50.22
convnext_small_c nhwc eager rw_hack 1268.07 201.866 256 224 398.03 319.951 128 224 50.22
convnext_base nhwc torchscript permute 1250.67 204.67 256 224 311.85 203.505 64 224 88.59
convnext_small_c nchw eager rw_hack_ts 1210.47 211.473 256 224 389.84 326.674 128 224 50.22
convnext_base_c nchw torchscript permute 1120.65 228.42 256 224 304.39 208.473 64 224 88.59
convnext_base nhwc aot ng_hack 1119.98 228.558 256 224 324.59 195.7 64 224 88.59
convnext_base nhwc aot rw_hack 1116.84 229.198 256 224 297.31 213.705 64 224 88.59
convnext_base_c nhwc torchscript permute 1111.51 230.293 256 224 301.68 210.307 64 224 88.59
convnext_base nhwc aot permute 1110.2 230.569 256 224 314.73 201.687 64 224 88.59
convnext_base nchw aot rw_hack 1105.11 231.634 256 224 311.86 203.775 64 224 88.59
convnext_base_c nhwc torchscript rw_hack 1102.76 232.127 256 224 229.42 277.173 64 224 88.59
convnext_base nchw aot permute 1100.94 232.51 256 224 320.76 197.961 64 224 88.59
convnext_base_c nchw torchscript rw_hack 1100.6 232.582 256 224 234.78 270.831 64 224 88.59
convnext_base_c nhwc aot rw_hack 1068.56 239.558 256 224 250.93 508.62 128 224 88.59
convnext_base nchw aot ng_hack 1064.84 240.394 256 224 310.86 204.438 64 224 88.59
convnext_base_c nchw aot rw_hack 1057.34 242.099 256 224 261.65 487.764 128 224 88.59
convnext_base_c nhwc aot ng_hack 1043.95 245.203 256 224 312.98 407.53 128 224 88.59
convnext_base_c nhwc eager rw_hack_ts 1020.57 250.823 256 224 246.93 516.607 128 224 88.59
convnext_base_c nchw aot permute 1008.67 253.78 256 224 316.45 402.965 128 224 88.59
convnext_base_c nhwc aot permute 975.86 262.308 256 224 301.41 423.09 128 224 88.59
convnext_base_c nhwc eager permute 975.5 262.412 256 224 323.27 394.372 128 224 88.59 15.38 28.75
convnext_base_c nchw eager permute 972.0 263.354 256 224 322.5 395.237 128 224 88.59 15.38 28.75
convnext_base nhwc eager rw_hack_ts 958.99 266.928 256 224 299.15 426.176 128 224 88.59
convnext_base nhwc eager permute 953.71 268.409 256 224 327.23 389.549 128 224 88.59 15.38 28.75
convnext_base nchw eager permute 949.2 269.683 256 224 325.43 391.656 128 224 88.59 15.38 28.75
convnext_base nhwc eager ng_hack 944.94 270.898 256 224 325.29 391.65 128 224 88.59
convnext_base nchw eager rw_hack_ts 938.8 272.669 256 224 318.84 399.83 128 224 88.59
convnext_base nchw eager rw_hack 933.4 274.245 256 224 317.11 402.017 128 224 88.59
convnext_base nhwc eager rw_hack 929.31 275.457 256 224 315.24 404.457 128 224 88.59
convnext_base_c nhwc eager ng_hack 928.06 275.822 256 224 313.06 407.254 128 224 88.59
convnext_base nchw eager ng_hack 911.86 280.725 256 224 311.3 409.399 128 224 88.59
convnext_base_c nchw eager rw_hack 888.26 288.186 256 224 278.23 458.409 128 224 88.59
convnext_base_c nhwc eager rw_hack 887.55 288.418 256 224 276.74 460.897 128 224 88.59
convnext_base_c nchw eager rw_hack_ts 848.68 301.625 256 224 271.44 469.878 128 224 88.59
convnext_small_c nchw aot ng_hack 846.52 302.392 256 224 261.31 488.258 128 224 50.22
convnext_small_c nchw eager ng_hack 824.79 310.36 256 224 261.48 487.93 128 224 50.22
convnext_large nhwc torchscript rw_hack 679.33 376.821 256 224 157.16 202.005 32 224 197.77
convnext_large nchw torchscript permute 677.92 377.601 256 224 181.33 351.33 64 224 197.77
convnext_large nchw torchscript rw_hack 676.72 378.274 256 224 158.41 200.414 32 224 197.77
convnext_large nhwc torchscript permute 665.97 384.376 256 224 182.53 348.966 64 224 197.77
convnext_large_c nchw torchscript permute 611.21 418.819 256 224 170.39 373.947 64 224 197.77
convnext_large nhwc aot ng_hack 611.01 418.955 256 224 180.07 353.903 64 224 197.77
convnext_large nhwc aot rw_hack 610.55 419.273 256 224 174.22 365.887 64 224 197.77
convnext_large_c nhwc torchscript permute 609.61 419.916 256 224 170.95 372.567 64 224 197.77
convnext_large_c nhwc torchscript rw_hack 607.6 421.308 256 224 131.28 242.001 32 224 197.77
convnext_large nhwc aot permute 607.19 421.59 256 224 181.27 351.53 64 224 197.77
convnext_large nchw aot rw_hack 605.93 422.472 256 224 177.53 359.099 64 224 197.77
convnext_large_c nchw torchscript rw_hack 605.02 423.103 256 224 132.1 240.504 32 224 197.77
convnext_large nchw aot permute 604.61 423.393 256 224 181.06 351.973 64 224 197.77
convnext_base_c nchw aot ng_hack 601.03 425.917 256 224 185.56 688.336 128 224 88.59
convnext_large_c nhwc aot rw_hack 591.73 432.609 256 224 141.75 450.036 64 224 197.77
convnext_large_c nchw aot rw_hack 587.87 435.449 256 224 143.85 443.445 64 224 197.77
convnext_base_c nchw eager ng_hack 585.95 436.878 256 224 184.97 690.427 128 224 88.59
convnext_large nchw aot ng_hack 579.61 441.65 256 224 174.98 364.366 64 224 197.77
convnext_large_c nhwc aot ng_hack 577.23 443.475 256 224 166.17 383.655 64 224 197.77
convnext_large_c nhwc eager rw_hack_ts 570.02 449.085 256 224 143.18 445.257 64 224 197.77
convnext_large_c nchw aot permute 566.95 451.518 256 224 169.71 375.64 64 224 197.77
convnext_large_c nhwc aot permute 562.7 454.923 256 224 169.11 376.888 64 224 197.77
convnext_large_c nhwc eager permute 548.42 466.77 256 224 172.28 369.859 64 224 197.77 34.4 43.13
convnext_large_c nchw eager permute 547.38 467.658 256 224 172.21 370.055 64 224 197.77 34.4 43.13
convnext_large nhwc eager rw_hack_ts 536.41 477.226 256 224 167.99 379.206 64 224 197.77
convnext_large nchw eager permute 533.71 479.629 256 224 173.78 366.771 64 224 197.77 34.4 43.13
convnext_large nhwc eager ng_hack 531.16 481.933 256 224 173.38 367.456 64 224 197.77
convnext_large nhwc eager permute 527.4 485.373 256 224 171.93 370.585 64 224 197.77 34.4 43.13
convnext_large nchw eager rw_hack_ts 526.37 486.323 256 224 169.6 375.688 64 224 197.77
convnext_large_c nhwc eager ng_hack 525.77 486.883 256 224 167.53 380.432 64 224 197.77
convnext_large nchw eager rw_hack 524.46 488.089 256 224 169.68 375.488 64 224 197.77
convnext_large nhwc eager rw_hack 523.26 489.214 256 224 168.42 378.378 64 224 197.77
convnext_large nchw eager ng_hack 513.97 498.061 256 224 166.97 381.48 64 224 197.77
convnext_large_c nchw eager rw_hack 506.51 505.39 256 224 150.96 422.328 64 224 197.77
convnext_large_c nhwc eager rw_hack 505.83 506.074 256 224 150.14 424.639 64 224 197.77
convnext_large_c nchw eager rw_hack_ts 484.87 527.957 256 224 147.55 432.081 64 224 197.77
convnext_large_c nchw aot ng_hack 361.68 707.792 256 224 106.97 596.757 64 224 197.77
convnext_large_c nchw eager ng_hack 353.0 725.187 256 224 106.83 597.551 64 224 197.77
model fmt cg layer train_samples_per_sec train_step_time train_batch_size
convnext_tiny nhwc torchscript permute 877.71 144.926 128
convnext_tiny nchw torchscript permute 861.39 147.693 128
convnext_tiny nhwc aot ng_hack 827.46 153.853 128
convnext_tiny nhwc aot permute 819.95 155.287 128
convnext_tiny_c nhwc torchscript permute 811.07 156.88 128
convnext_tiny_c nhwc aot ng_hack 799.89 319.219 256
convnext_tiny_c nchw torchscript permute 798.45 159.418 128
convnext_tiny_c nhwc eager permute 794.23 321.435 256
convnext_tiny nhwc eager permute 792.56 322.103 256
convnext_tiny nchw eager permute 791.37 322.585 256
convnext_tiny_c nchw eager permute 790.3 323.035 256
convnext_tiny nchw aot permute 788.72 161.334 128
convnext_tiny nhwc eager ng_hack 787.31 324.191 256
convnext_tiny nchw aot rw_hack 782.49 162.762 128
convnext_tiny nchw torchscript rw_hack 775.44 164.174 128
convnext_tiny_c nchw aot permute 770.43 331.449 256
convnext_tiny nchw eager rw_hack_ts 766.21 333.081 256
convnext_tiny nhwc torchscript rw_hack 765.84 166.191 128
convnext_tiny nchw aot ng_hack 763.84 166.723 128
convnext_tiny_c nhwc eager ng_hack 763.11 334.57 256
convnext_tiny nchw eager rw_hack 759.56 336.101 256
convnext_tiny nhwc eager rw_hack 751.81 339.603 256
convnext_tiny_c nhwc aot permute 747.49 341.568 256
convnext_tiny nhwc aot rw_hack 734.61 173.385 128
convnext_tiny nchw eager ng_hack 733.75 347.957 256
convnext_tiny nhwc eager rw_hack_ts 724.19 352.506 256
convnext_tiny_c nchw eager rw_hack 685.44 372.539 256
convnext_tiny_c nhwc eager rw_hack 680.73 375.149 256
convnext_tiny_c nchw eager rw_hack_ts 671.62 380.171 256
convnext_tiny_c nchw aot rw_hack 647.69 394.423 256
convnext_tiny_c nhwc eager rw_hack_ts 646.34 395.096 256
convnext_tiny_c nhwc aot rw_hack 606.81 420.99 256
convnext_tiny_c nchw torchscript rw_hack 605.98 210.22 128
convnext_tiny_c nhwc torchscript rw_hack 601.07 211.979 128
convnext_small nchw torchscript permute 508.55 249.996 128
convnext_small nhwc torchscript permute 504.01 252.232 128
convnext_small nhwc aot ng_hack 498.73 255.192 128
convnext_small nhwc aot permute 493.49 257.925 128
convnext_small nchw aot permute 490.25 259.414 128
convnext_small nchw aot rw_hack 481.35 264.473 128
convnext_small nchw torchscript rw_hack 477.39 266.497 128
convnext_small nchw aot ng_hack 474.15 268.507 128
convnext_small_c nchw torchscript permute 473.78 268.408 128
convnext_small nhwc torchscript rw_hack 473.71 268.535 128
convnext_small nhwc aot rw_hack 465.18 273.701 128
convnext_small_c nhwc eager permute 464.43 274.049 128
convnext_small_c nchw eager permute 462.57 275.083 128
convnext_small nchw eager permute 461.49 275.767 128
convnext_small nhwc eager permute 461.32 275.83 128
convnext_small nhwc eager ng_hack 461.05 276.022 128
convnext_small_c nhwc torchscript permute 458.92 277.03 128
convnext_small_c nhwc aot ng_hack 457.44 278.213 128
convnext_small_c nhwc aot permute 457.4 278.366 128
convnext_small nchw eager rw_hack_ts 453.84 280.375 128
convnext_small_c nchw aot permute 453.51 280.621 128
convnext_small nchw eager rw_hack 450.11 282.768 128
convnext_small_c nhwc eager ng_hack 448.36 283.776 128
convnext_small nhwc eager rw_hack 447.81 284.127 128
convnext_small nchw eager ng_hack 441.5 288.235 128
convnext_tiny_c nchw eager ng_hack 441.37 579.066 256
convnext_small nhwc eager rw_hack_ts 438.44 290.306 128
convnext_tiny_c nchw aot ng_hack 438.08 583.537 256
convnext_small_c nchw eager rw_hack 398.95 319.187 128
convnext_small_c nhwc eager rw_hack 398.03 319.951 128
convnext_small_c nchw eager rw_hack_ts 389.84 326.674 128
convnext_small_c nchw aot rw_hack 380.85 334.595 128
convnext_small_c nhwc aot rw_hack 371.33 343.245 128
convnext_small_c nhwc eager rw_hack_ts 371.02 343.251 128
convnext_small_c nchw torchscript rw_hack 349.37 181.504 64
convnext_small_c nhwc torchscript rw_hack 346.62 182.939 64
convnext_base nhwc eager permute 327.23 389.549 128
convnext_base nchw eager permute 325.43 391.656 128
convnext_base nhwc eager ng_hack 325.29 391.65 128
convnext_base nhwc aot ng_hack 324.59 195.7 64
convnext_base_c nhwc eager permute 323.27 394.372 128
convnext_base nchw torchscript permute 322.5 196.805 64
convnext_base_c nchw eager permute 322.5 395.237 128
convnext_base nchw aot permute 320.76 197.961 64
convnext_base nchw eager rw_hack_ts 318.84 399.83 128
convnext_base nchw eager rw_hack 317.11 402.017 128
convnext_base_c nchw aot permute 316.45 402.965 128
convnext_base nhwc eager rw_hack 315.24 404.457 128
convnext_base nhwc aot permute 314.73 201.687 64
convnext_base_c nhwc eager ng_hack 313.06 407.254 128
convnext_base_c nhwc aot ng_hack 312.98 407.53 128
convnext_base nchw aot rw_hack 311.86 203.775 64
convnext_base nhwc torchscript permute 311.85 203.505 64
convnext_base nchw eager ng_hack 311.3 409.399 128
convnext_base nchw aot ng_hack 310.86 204.438 64
convnext_base nchw torchscript rw_hack 306.72 206.975 64
convnext_base_c nchw torchscript permute 304.39 208.473 64
convnext_base_c nhwc torchscript permute 301.68 210.307 64
convnext_base_c nhwc aot permute 301.41 423.09 128
convnext_base nhwc eager rw_hack_ts 299.15 426.176 128
convnext_base nhwc torchscript rw_hack 297.33 213.607 64
convnext_base nhwc aot rw_hack 297.31 213.705 64
convnext_base_c nchw eager rw_hack 278.23 458.409 128
convnext_base_c nhwc eager rw_hack 276.74 460.897 128
convnext_base_c nchw eager rw_hack_ts 271.44 469.878 128
convnext_base_c nchw aot rw_hack 261.65 487.764 128
convnext_small_c nchw eager ng_hack 261.48 487.93 128
convnext_small_c nchw aot ng_hack 261.31 488.258 128
convnext_base_c nhwc aot rw_hack 250.93 508.62 128
convnext_base_c nhwc eager rw_hack_ts 246.93 516.607 128
convnext_base_c nchw torchscript rw_hack 234.78 270.831 64
convnext_base_c nhwc torchscript rw_hack 229.42 277.173 64
convnext_base_c nchw aot ng_hack 185.56 688.336 128
convnext_base_c nchw eager ng_hack 184.97 690.427 128
convnext_large nhwc torchscript permute 182.53 348.966 64
convnext_large nchw torchscript permute 181.33 351.33 64
convnext_large nhwc aot permute 181.27 351.53 64
convnext_large nchw aot permute 181.06 351.973 64
convnext_large nhwc aot ng_hack 180.07 353.903 64
convnext_large nchw aot rw_hack 177.53 359.099 64
convnext_large nchw aot ng_hack 174.98 364.366 64
convnext_large nhwc aot rw_hack 174.22 365.887 64
convnext_large nchw eager permute 173.78 366.771 64
convnext_large nhwc eager ng_hack 173.38 367.456 64
convnext_large_c nhwc eager permute 172.28 369.859 64
convnext_large_c nchw eager permute 172.21 370.055 64
convnext_large nhwc eager permute 171.93 370.585 64
convnext_large_c nhwc torchscript permute 170.95 372.567 64
convnext_large_c nchw torchscript permute 170.39 373.947 64
convnext_large_c nchw aot permute 169.71 375.64 64
convnext_large nchw eager rw_hack 169.68 375.488 64
convnext_large nchw eager rw_hack_ts 169.6 375.688 64
convnext_large_c nhwc aot permute 169.11 376.888 64
convnext_large nhwc eager rw_hack 168.42 378.378 64
convnext_large nhwc eager rw_hack_ts 167.99 379.206 64
convnext_large_c nhwc eager ng_hack 167.53 380.432 64
convnext_large nchw eager ng_hack 166.97 381.48 64
convnext_large_c nhwc aot ng_hack 166.17 383.655 64
convnext_large nchw torchscript rw_hack 158.41 200.414 32
convnext_large nhwc torchscript rw_hack 157.16 202.005 32
convnext_large_c nchw eager rw_hack 150.96 422.328 64
convnext_large_c nhwc eager rw_hack 150.14 424.639 64
convnext_large_c nchw eager rw_hack_ts 147.55 432.081 64
convnext_large_c nchw aot rw_hack 143.85 443.445 64
convnext_large_c nhwc eager rw_hack_ts 143.18 445.257 64
convnext_large_c nhwc aot rw_hack 141.75 450.036 64
convnext_large_c nchw torchscript rw_hack 132.1 240.504 32
convnext_large_c nhwc torchscript rw_hack 131.28 242.001 32
convnext_large_c nchw aot ng_hack 106.97 596.757 64
convnext_large_c nchw eager ng_hack 106.83 597.551 64
model fmt cg layer infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count infer_gmacs infer_macts
convnext_tiny_c nhwc eager rw_hack_ts 2487.57 102.897 256 224 646.34 395.096 256 224 28.59
convnext_tiny nhwc eager rw_hack_ts 2298.36 111.369 256 224 724.19 352.506 256 224 28.59
convnext_tiny_c nhwc eager permute 2288.94 111.828 256 224 794.23 321.435 256 224 28.59 4.47 13.44
convnext_tiny_c nchw eager permute 2285.72 111.986 256 224 790.3 323.035 256 224 28.59 4.47 13.44
convnext_tiny nhwc eager permute 2250.61 113.733 256 224 792.56 322.103 256 224 28.59 4.47 13.44
convnext_tiny nchw eager permute 2232.09 114.674 256 224 791.37 322.585 256 224 28.59 4.47 13.44
convnext_tiny nhwc eager ng_hack 2217.22 115.445 256 224 787.31 324.191 256 224 28.59
convnext_tiny nchw eager rw_hack_ts 2195.18 116.602 256 224 766.21 333.081 256 224 28.59
convnext_tiny nchw eager rw_hack 2165.01 118.229 256 224 759.56 336.101 256 224 28.59
convnext_tiny nhwc eager rw_hack 2163.16 118.331 256 224 751.81 339.603 256 224 28.59
convnext_tiny_c nhwc eager ng_hack 2162.07 118.39 256 224 763.11 334.57 256 224 28.59
convnext_tiny nchw eager ng_hack 2100.81 121.843 256 224 733.75 347.957 256 224 28.59
convnext_tiny_c nchw eager rw_hack 2090.56 122.44 256 224 685.44 372.539 256 224 28.59
convnext_tiny_c nhwc eager rw_hack 2080.55 123.031 256 224 680.73 375.149 256 224 28.59
convnext_tiny_c nchw eager rw_hack_ts 1999.74 128.002 256 224 671.62 380.171 256 224 28.59
convnext_small_c nhwc eager rw_hack_ts 1479.75 172.985 256 224 371.02 343.251 128 224 50.22
convnext_small_c nchw eager permute 1389.39 184.239 256 224 462.57 275.083 128 224 50.22 8.71 21.56
convnext_small_c nhwc eager permute 1387.23 184.523 256 224 464.43 274.049 128 224 50.22 8.71 21.56
convnext_tiny_c nchw eager ng_hack 1375.98 186.032 256 224 441.37 579.066 256 224 28.59
convnext_small nhwc eager rw_hack_ts 1367.93 187.128 256 224 438.44 290.306 128 224 50.22
convnext_small nchw eager permute 1351.2 189.446 256 224 461.49 275.767 128 224 50.22 8.71 21.56
convnext_small nhwc eager ng_hack 1339.83 191.053 256 224 461.05 276.022 128 224 50.22
convnext_small nchw eager rw_hack_ts 1338.01 191.312 256 224 453.84 280.375 128 224 50.22
convnext_small nhwc eager permute 1333.5 191.956 256 224 461.32 275.83 128 224 50.22 8.71 21.56
convnext_small nchw eager rw_hack 1328.35 192.703 256 224 450.11 282.768 128 224 50.22
convnext_small nhwc eager rw_hack 1324.29 193.296 256 224 447.81 284.127 128 224 50.22
convnext_small_c nhwc eager ng_hack 1321.24 193.739 256 224 448.36 283.776 128 224 50.22
convnext_small nchw eager ng_hack 1293.56 197.885 256 224 441.5 288.235 128 224 50.22
convnext_small_c nchw eager rw_hack 1269.52 201.632 256 224 398.95 319.187 128 224 50.22
convnext_small_c nhwc eager rw_hack 1268.07 201.866 256 224 398.03 319.951 128 224 50.22
convnext_small_c nchw eager rw_hack_ts 1210.47 211.473 256 224 389.84 326.674 128 224 50.22
convnext_base_c nhwc eager rw_hack_ts 1020.57 250.823 256 224 246.93 516.607 128 224 88.59
convnext_base_c nhwc eager permute 975.5 262.412 256 224 323.27 394.372 128 224 88.59 15.38 28.75
convnext_base_c nchw eager permute 972.0 263.354 256 224 322.5 395.237 128 224 88.59 15.38 28.75
convnext_base nhwc eager rw_hack_ts 958.99 266.928 256 224 299.15 426.176 128 224 88.59
convnext_base nhwc eager permute 953.71 268.409 256 224 327.23 389.549 128 224 88.59 15.38 28.75
convnext_base nchw eager permute 949.2 269.683 256 224 325.43 391.656 128 224 88.59 15.38 28.75
convnext_base nhwc eager ng_hack 944.94 270.898 256 224 325.29 391.65 128 224 88.59
convnext_base nchw eager rw_hack_ts 938.8 272.669 256 224 318.84 399.83 128 224 88.59
convnext_base nchw eager rw_hack 933.4 274.245 256 224 317.11 402.017 128 224 88.59
convnext_base nhwc eager rw_hack 929.31 275.457 256 224 315.24 404.457 128 224 88.59
convnext_base_c nhwc eager ng_hack 928.06 275.822 256 224 313.06 407.254 128 224 88.59
convnext_base nchw eager ng_hack 911.86 280.725 256 224 311.3 409.399 128 224 88.59
convnext_base_c nchw eager rw_hack 888.26 288.186 256 224 278.23 458.409 128 224 88.59
convnext_base_c nhwc eager rw_hack 887.55 288.418 256 224 276.74 460.897 128 224 88.59
convnext_base_c nchw eager rw_hack_ts 848.68 301.625 256 224 271.44 469.878 128 224 88.59
convnext_small_c nchw eager ng_hack 824.79 310.36 256 224 261.48 487.93 128 224 50.22
convnext_base_c nchw eager ng_hack 585.95 436.878 256 224 184.97 690.427 128 224 88.59
convnext_large_c nhwc eager rw_hack_ts 570.02 449.085 256 224 143.18 445.257 64 224 197.77
convnext_large_c nhwc eager permute 548.42 466.77 256 224 172.28 369.859 64 224 197.77 34.4 43.13
convnext_large_c nchw eager permute 547.38 467.658 256 224 172.21 370.055 64 224 197.77 34.4 43.13
convnext_large nhwc eager rw_hack_ts 536.41 477.226 256 224 167.99 379.206 64 224 197.77
convnext_large nchw eager permute 533.71 479.629 256 224 173.78 366.771 64 224 197.77 34.4 43.13
convnext_large nhwc eager ng_hack 531.16 481.933 256 224 173.38 367.456 64 224 197.77
convnext_large nhwc eager permute 527.4 485.373 256 224 171.93 370.585 64 224 197.77 34.4 43.13
convnext_large nchw eager rw_hack_ts 526.37 486.323 256 224 169.6 375.688 64 224 197.77
convnext_large_c nhwc eager ng_hack 525.77 486.883 256 224 167.53 380.432 64 224 197.77
convnext_large nchw eager rw_hack 524.46 488.089 256 224 169.68 375.488 64 224 197.77
convnext_large nhwc eager rw_hack 523.26 489.214 256 224 168.42 378.378 64 224 197.77
convnext_large nchw eager ng_hack 513.97 498.061 256 224 166.97 381.48 64 224 197.77
convnext_large_c nchw eager rw_hack 506.51 505.39 256 224 150.96 422.328 64 224 197.77
convnext_large_c nhwc eager rw_hack 505.83 506.074 256 224 150.14 424.639 64 224 197.77
convnext_large_c nchw eager rw_hack_ts 484.87 527.957 256 224 147.55 432.081 64 224 197.77
convnext_large_c nchw eager ng_hack 353.0 725.187 256 224 106.83 597.551 64 224 197.77
model fmt cg layer train_samples_per_sec train_step_time train_batch_size
convnext_tiny_c nhwc eager permute 794.23 321.435 256
convnext_tiny nhwc eager permute 792.56 322.103 256
convnext_tiny nchw eager permute 791.37 322.585 256
convnext_tiny_c nchw eager permute 790.3 323.035 256
convnext_tiny nhwc eager ng_hack 787.31 324.191 256
convnext_tiny nchw eager rw_hack_ts 766.21 333.081 256
convnext_tiny_c nhwc eager ng_hack 763.11 334.57 256
convnext_tiny nchw eager rw_hack 759.56 336.101 256
convnext_tiny nhwc eager rw_hack 751.81 339.603 256
convnext_tiny nchw eager ng_hack 733.75 347.957 256
convnext_tiny nhwc eager rw_hack_ts 724.19 352.506 256
convnext_tiny_c nchw eager rw_hack 685.44 372.539 256
convnext_tiny_c nhwc eager rw_hack 680.73 375.149 256
convnext_tiny_c nchw eager rw_hack_ts 671.62 380.171 256
convnext_tiny_c nhwc eager rw_hack_ts 646.34 395.096 256
convnext_small_c nhwc eager permute 464.43 274.049 128
convnext_small_c nchw eager permute 462.57 275.083 128
convnext_small nchw eager permute 461.49 275.767 128
convnext_small nhwc eager permute 461.32 275.83 128
convnext_small nhwc eager ng_hack 461.05 276.022 128
convnext_small nchw eager rw_hack_ts 453.84 280.375 128
convnext_small nchw eager rw_hack 450.11 282.768 128
convnext_small_c nhwc eager ng_hack 448.36 283.776 128
convnext_small nhwc eager rw_hack 447.81 284.127 128
convnext_small nchw eager ng_hack 441.5 288.235 128
convnext_tiny_c nchw eager ng_hack 441.37 579.066 256
convnext_small nhwc eager rw_hack_ts 438.44 290.306 128
convnext_small_c nchw eager rw_hack 398.95 319.187 128
convnext_small_c nhwc eager rw_hack 398.03 319.951 128
convnext_small_c nchw eager rw_hack_ts 389.84 326.674 128
convnext_small_c nhwc eager rw_hack_ts 371.02 343.251 128
convnext_base nhwc eager permute 327.23 389.549 128
convnext_base nchw eager permute 325.43 391.656 128
convnext_base nhwc eager ng_hack 325.29 391.65 128
convnext_base_c nhwc eager permute 323.27 394.372 128
convnext_base_c nchw eager permute 322.5 395.237 128
convnext_base nchw eager rw_hack_ts 318.84 399.83 128
convnext_base nchw eager rw_hack 317.11 402.017 128
convnext_base nhwc eager rw_hack 315.24 404.457 128
convnext_base_c nhwc eager ng_hack 313.06 407.254 128
convnext_base nchw eager ng_hack 311.3 409.399 128
convnext_base nhwc eager rw_hack_ts 299.15 426.176 128
convnext_base_c nchw eager rw_hack 278.23 458.409 128
convnext_base_c nhwc eager rw_hack 276.74 460.897 128
convnext_base_c nchw eager rw_hack_ts 271.44 469.878 128
convnext_small_c nchw eager ng_hack 261.48 487.93 128
convnext_base_c nhwc eager rw_hack_ts 246.93 516.607 128
convnext_base_c nchw eager ng_hack 184.97 690.427 128
convnext_large nchw eager permute 173.78 366.771 64
convnext_large nhwc eager ng_hack 173.38 367.456 64
convnext_large_c nhwc eager permute 172.28 369.859 64
convnext_large_c nchw eager permute 172.21 370.055 64
convnext_large nhwc eager permute 171.93 370.585 64
convnext_large nchw eager rw_hack 169.68 375.488 64
convnext_large nchw eager rw_hack_ts 169.6 375.688 64
convnext_large nhwc eager rw_hack 168.42 378.378 64
convnext_large nhwc eager rw_hack_ts 167.99 379.206 64
convnext_large_c nhwc eager ng_hack 167.53 380.432 64
convnext_large nchw eager ng_hack 166.97 381.48 64
convnext_large_c nchw eager rw_hack 150.96 422.328 64
convnext_large_c nhwc eager rw_hack 150.14 424.639 64
convnext_large_c nchw eager rw_hack_ts 147.55 432.081 64
convnext_large_c nhwc eager rw_hack_ts 143.18 445.257 64
convnext_large_c nchw eager ng_hack 106.83 597.551 64
model fmt cg layer infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count
convnext_tiny nhwc torchscript permute 3023.91 84.643 256 224 805.78 316.24 256 224 28.59
convnext_tiny nchw torchscript permute 2989.54 85.615 256 224 786.85 323.78 256 224 28.59
convnext_tiny nhwc aot ng_hack 2968.06 86.233 256 224 827.51 307.932 256 224 28.59
convnext_tiny nhwc aot rw_hack 2941.24 87.02 256 224 651.75 391.449 256 224 28.59
convnext_tiny nchw aot rw_hack 2931.87 87.298 256 224 766.97 332.386 256 224 28.59
convnext_tiny nchw torchscript rw_hack 2927.56 87.427 256 224 661.11 385.512 256 224 28.59
convnext_tiny nhwc torchscript rw_hack 2907.18 88.04 256 224 653.1 390.386 256 224 28.59
convnext_tiny nhwc aot permute 2899.61 88.27 256 224 815.04 312.58 256 224 28.59
convnext_tiny nchw aot permute 2868.5 89.227 256 224 796.65 319.983 256 224 28.59
convnext_tiny nchw aot ng_hack 2696.44 94.923 256 224 692.13 368.511 256 224 28.59
convnext_tiny_c nhwc torchscript permute 2368.89 108.05 256 224 673.06 378.66 256 224 28.59
convnext_tiny_c nchw aot rw_hack 2345.25 109.137 256 224 498.35 512.304 256 224 28.59
convnext_tiny_c nhwc aot ng_hack 2342.13 109.283 256 224 672.51 379.173 256 224 28.59
convnext_tiny_c nhwc aot permute 2340.02 109.383 256 224 660.62 386.11 256 224 28.59
convnext_tiny_c nhwc aot rw_hack 2327.09 109.99 256 224 448.9 568.856 256 224 28.59
convnext_tiny_c nchw aot permute 2315.17 110.556 256 224 646.96 394.141 256 224 28.59
convnext_tiny_c nchw torchscript permute 2312.78 110.672 256 224 666.05 382.712 256 224 28.59
convnext_tiny_c nchw torchscript rw_hack 2091.27 122.395 256 224 469.96 542.956 256 224 28.59
convnext_tiny_c nhwc torchscript rw_hack 2007.69 127.493 256 224 454.32 561.759 256 224 28.59
convnext_tiny nhwc eager rw_hack_ts 1996.25 128.223 256 224 573.17 444.905 256 224 28.59
convnext_tiny nhwc eager permute 1975.29 129.584 256 224 690.18 369.352 256 224 28.59
convnext_tiny nchw eager rw_hack_ts 1962.79 130.409 256 224 600.74 424.488 256 224 28.59
convnext_tiny nchw eager permute 1956.4 130.836 256 224 667.5 382.023 256 224 28.59
convnext_tiny nhwc eager ng_hack 1945.41 131.573 256 224 684.85 372.206 256 224 28.59
convnext_tiny nchw eager rw_hack 1903.62 134.463 256 224 630.53 404.528 256 224 28.59
convnext_tiny nhwc eager rw_hack 1889.58 135.461 256 224 587.89 433.765 256 224 28.59
convnext_tiny_c nhwc eager permute 1884.06 135.858 256 224 623.6 409.02 256 224 28.59
convnext_tiny_c nhwc eager rw_hack_ts 1811.37 141.312 256 224 448.71 568.931 256 224 28.59
convnext_tiny_c nchw eager permute 1774.78 144.223 256 224 603.65 422.557 256 224 28.59
convnext_tiny nchw eager ng_hack 1756.08 145.761 256 224 562.33 453.651 256 224 28.59
convnext_tiny_c nchw eager rw_hack_ts 1752.3 146.077 256 224 454.94 561.108 256 224 28.59
convnext_tiny_c nhwc eager ng_hack 1725.17 148.371 256 224 599.13 425.666 256 224 28.59
convnext_tiny_c nchw eager rw_hack 1583.8 161.618 256 224 507.43 502.86 256 224 28.59
convnext_small nhwc aot ng_hack 1547.3 165.431 256 224 458.19 276.691 128 224 50.22
convnext_small nhwc torchscript permute 1546.35 165.534 256 224 430.88 294.284 128 224 50.22
convnext_small nchw torchscript permute 1538.02 166.431 256 224 426.79 297.103 128 224 50.22
convnext_small nchw torchscript rw_hack 1520.31 168.37 256 224 397.6 319.031 128 224 50.22
convnext_tiny_c nhwc eager rw_hack 1503.91 170.206 256 224 487.46 523.637 256 224 28.59
convnext_small nchw aot rw_hack 1502.19 170.4 256 224 433.31 292.765 128 224 50.22
convnext_small nhwc aot rw_hack 1493.94 171.34 256 224 413.16 307.11 128 224 50.22
convnext_small nhwc torchscript rw_hack 1491.47 171.626 256 224 391.12 324.275 128 224 50.22
convnext_small nhwc aot permute 1476.52 173.362 256 224 450.4 281.669 128 224 50.22
convnext_small nchw aot permute 1457.35 175.642 256 224 434.45 292.187 128 224 50.22
convnext_small nchw aot ng_hack 1334.99 191.743 256 224 406.24 312.668 128 224 50.22
convnext_small_c nhwc torchscript permute 1221.23 209.608 256 224 382.35 331.792 128 224 50.22
convnext_small_c nchw torchscript permute 1199.71 213.368 256 224 376.79 336.783 128 224 50.22
convnext_tiny_c nchw aot ng_hack 1191.91 214.762 256 224 312.04 818.947 256 224 28.59
convnext_small_c nhwc aot rw_hack 1169.28 218.92 256 224 280.95 908.672 256 224 50.22
convnext_small_c nhwc aot ng_hack 1167.52 219.249 256 224 365.09 698.546 256 224 50.22
convnext_small_c nchw torchscript rw_hack 1157.71 221.109 256 224 286.75 443.268 128 224 50.22
convnext_small_c nhwc torchscript rw_hack 1151.47 222.308 256 224 282.58 449.95 128 224 50.22
convnext_small_c nchw aot rw_hack 1148.24 222.932 256 224 296.15 861.845 256 224 50.22
convnext_small_c nchw aot permute 1116.25 229.318 256 224 354.97 718.562 256 224 50.22
convnext_small_c nhwc aot permute 1080.56 236.895 256 224 355.73 716.936 256 224 50.22
convnext_tiny_c nchw eager ng_hack 1028.52 248.885 256 224 309.24 826.374 256 224 28.59
convnext_small nhwc eager permute 1004.53 254.829 256 224 362.99 702.462 256 224 50.22
convnext_base nhwc torchscript permute 1000.73 255.797 256 224 298.07 426.518 128 224 88.59
convnext_base nchw torchscript permute 988.99 258.831 256 224 293.38 433.542 128 224 88.59
convnext_base nhwc torchscript rw_hack 986.99 259.355 256 224 256.34 496.532 128 224 88.59
convnext_base nchw torchscript rw_hack 984.76 259.944 256 224 262.16 485.562 128 224 88.59
convnext_small nhwc eager ng_hack 966.82 264.769 256 224 356.52 715.044 256 224 50.22
convnext_small nchw eager rw_hack 966.11 264.948 256 224 344.18 741.05 256 224 50.22
convnext_small nchw eager permute 964.77 265.33 256 224 350.69 727.147 256 224 50.22
convnext_base nhwc aot ng_hack 955.63 267.868 256 224 315.01 403.86 128 224 88.59
convnext_base nchw aot rw_hack 947.13 270.272 256 224 298.13 426.714 128 224 88.59
convnext_small nchw eager rw_hack_ts 946.84 270.355 256 224 326.99 779.921 256 224 50.22
convnext_base nhwc aot rw_hack 943.88 271.203 256 224 273.27 465.805 128 224 88.59
convnext_base nhwc aot permute 941.35 271.93 256 224 306.97 414.202 128 224 88.59
convnext_small nhwc eager rw_hack_ts 934.17 274.022 256 224 322.12 791.862 256 224 50.22
convnext_base nchw aot permute 928.64 275.652 256 224 305.97 415.761 128 224 88.59
convnext_small_c nhwc eager rw_hack_ts 928.51 275.695 256 224 255.61 497.697 128 224 50.22
convnext_small_c nchw eager rw_hack_ts 925.56 276.571 256 224 267.96 952.418 256 224 50.22
convnext_small_c nhwc eager permute 919.97 278.251 256 224 341.2 747.55 256 224 50.22
convnext_small nhwc eager rw_hack 900.04 284.414 256 224 331.45 769.39 256 224 50.22
convnext_small_c nchw eager permute 889.44 287.798 256 224 334.33 763.03 256 224 50.22
convnext_small_c nhwc eager ng_hack 879.17 291.148 256 224 328.46 776.664 256 224 50.22
convnext_base nchw aot ng_hack 859.05 297.984 256 224 285.95 445.07 128 224 88.59
convnext_small nchw eager ng_hack 841.59 304.169 256 224 322.14 791.84 256 224 50.22
convnext_small_c nchw eager rw_hack 823.76 310.741 256 224 287.61 887.187 256 224 50.22
convnext_base_c nhwc torchscript permute 792.08 323.182 256 224 259.19 490.747 128 224 88.59
convnext_small_c nhwc eager rw_hack 782.61 327.09 256 224 275.22 927.236 256 224 50.22
convnext_base_c nchw torchscript permute 777.31 329.321 256 224 262.12 485.186 128 224 88.59
convnext_base_c nhwc aot rw_hack 739.9 345.976 256 224 194.19 656.666 128 224 88.59
convnext_base_c nchw aot rw_hack 739.32 346.246 256 224 208.12 612.406 128 224 88.59
convnext_base_c nhwc aot ng_hack 739.1 346.348 256 224 255.15 499.107 128 224 88.59
convnext_base_c nhwc torchscript rw_hack 737.37 347.163 256 224 182.79 697.122 128 224 88.59
convnext_base_c nchw torchscript rw_hack 730.5 350.428 256 224 188.38 676.383 128 224 88.59
convnext_base_c nhwc aot permute 725.43 352.876 256 224 256.82 495.839 128 224 88.59
convnext_base_c nchw aot permute 720.62 355.231 256 224 256.27 496.865 128 224 88.59
convnext_small_c nchw aot ng_hack 655.73 390.387 256 224 179.35 1424.921 256 224 50.22
convnext_base_c nhwc eager rw_hack_ts 602.51 424.87 256 224 172.61 738.43 128 224 88.59
convnext_base nhwc eager rw_hack_ts 601.57 425.538 256 224 219.86 579.023 128 224 88.59
convnext_base nhwc eager permute 601.43 425.633 256 224 243.09 523.686 128 224 88.59
convnext_base_c nchw eager rw_hack_ts 601.42 425.643 256 224 180.77 705.006 128 224 88.59
convnext_base nhwc eager ng_hack 595.92 429.571 256 224 238.55 533.696 128 224 88.59
convnext_base_c nhwc eager permute 595.48 429.865 256 224 237.65 535.861 128 224 88.59
convnext_base nchw eager rw_hack 594.5 430.599 256 224 234.73 542.375 128 224 88.59
convnext_base nhwc eager rw_hack 593.82 431.088 256 224 232.14 548.499 128 224 88.59
convnext_base nchw eager permute 587.78 435.517 256 224 237.08 537.14 128 224 88.59
convnext_base_c nchw eager permute 584.6 437.886 256 224 230.78 551.927 128 224 88.59
convnext_base nchw eager rw_hack_ts 581.66 440.098 256 224 222.42 572.35 128 224 88.59
convnext_base_c nhwc eager ng_hack 564.96 453.092 256 224 228.34 557.696 128 224 88.59
convnext_small_c nchw eager ng_hack 552.16 463.619 256 224 176.94 1443.868 256 224 50.22
convnext_base nchw eager ng_hack 547.32 467.718 256 224 225.58 564.519 128 224 88.59
convnext_base_c nchw eager rw_hack 533.24 480.063 256 224 202.53 629.0 128 224 88.59
convnext_base_c nhwc eager rw_hack 528.7 484.185 256 224 196.89 647.016 128 224 88.59
convnext_large nhwc torchscript permute 461.51 554.676 256 224 150.45 422.39 64 224 197.77
convnext_large nchw torchscript permute 453.47 564.513 256 224 148.73 427.612 64 224 197.77
convnext_large nhwc aot rw_hack 449.77 569.157 256 224 152.25 417.628 64 224 197.77
convnext_large nhwc aot ng_hack 448.58 570.675 256 224 155.14 409.485 64 224 197.77
convnext_large nchw torchscript rw_hack 447.27 572.347 256 224 143.4 443.423 64 224 197.77
convnext_large nhwc torchscript rw_hack 443.82 576.797 256 224 141.95 448.099 64 224 197.77
convnext_large nchw aot rw_hack 442.48 578.541 256 224 151.95 418.682 64 224 197.77
convnext_large nhwc aot permute 438.31 584.042 256 224 154.95 410.54 64 224 197.77
convnext_large nchw aot permute 435.9 587.275 256 224 156.16 406.63 64 224 197.77
convnext_base_c nchw aot ng_hack 422.71 605.601 256 224 130.79 976.12 128 224 88.59
convnext_large nchw aot ng_hack 415.24 616.496 256 224 147.55 430.966 64 224 197.77
convnext_large_c nhwc torchscript permute 386.58 662.198 256 224 135.9 468.059 64 224 197.77
convnext_large_c nhwc aot rw_hack 382.73 668.865 256 224 120.56 528.212 64 224 197.77
convnext_base_c nchw eager ng_hack 380.08 673.53 256 224 126.84 1006.37 128 224 88.59
convnext_large_c nchw torchscript permute 367.51 696.564 256 224 134.5 472.887 64 224 197.77
convnext_large_c nhwc aot ng_hack 362.53 706.119 256 224 128.52 993.407 128 224 197.77
convnext_large_c nchw torchscript rw_hack 361.94 707.286 256 224 109.46 581.567 64 224 197.77
convnext_large_c nhwc torchscript rw_hack 361.37 708.399 256 224 108.92 584.642 64 224 197.77
convnext_large_c nhwc aot permute 350.83 729.668 256 224 128.61 992.703 128 224 197.77
convnext_large_c nchw aot permute 345.4 741.14 256 224 127.25 1003.393 128 224 197.77
convnext_large_c nhwc eager rw_hack_ts 317.88 805.324 256 224 98.96 643.76 64 224 197.77
convnext_large nchw eager rw_hack 314.61 813.683 256 224 124.68 1023.618 128 224 197.77
convnext_large nhwc eager rw_hack 311.63 821.461 256 224 121.06 1054.197 128 224 197.77
convnext_large nhwc eager rw_hack_ts 311.33 822.262 256 224 121.03 1054.776 128 224 197.77
convnext_large_c nchw eager rw_hack_ts 311.04 823.035 256 224 102.57 620.844 64 224 197.77
convnext_large nhwc eager ng_hack 310.17 825.322 256 224 125.29 1018.734 128 224 197.77
convnext_large nhwc eager permute 309.49 827.134 256 224 125.57 1016.477 128 224 197.77
convnext_large_c nhwc eager permute 307.64 832.131 256 224 119.12 1071.582 128 224 197.77
convnext_large nchw eager rw_hack_ts 305.21 838.745 256 224 119.88 1064.756 128 224 197.77
convnext_large nchw eager permute 305.2 838.754 256 224 125.98 1013.284 128 224 197.77
convnext_large_c nchw eager permute 304.56 840.529 256 224 118.15 1080.296 128 224 197.77
convnext_large_c nhwc eager ng_hack 296.82 862.444 256 224 116.27 1098.151 128 224 197.77
convnext_large nchw eager ng_hack 292.69 874.627 256 224 121.93 1046.427 128 224 197.77
convnext_large_c nhwc eager rw_hack 281.81 908.387 256 224 102.4 622.012 64 224 197.77
convnext_large_c nchw eager rw_hack 281.1 910.679 256 224 104.38 1223.167 128 224 197.77
convnext_large_c nchw aot ng_hack 236.18 1083.877 256 224 73.51 1738.218 128 224 197.77
convnext_large_c nchw eager ng_hack 214.31 1194.533 256 224 72.34 1766.633 128 224 197.77
model fmt cg layer train_samples_per_sec train_step_time train_batch_size
convnext_tiny nhwc aot ng_hack 827.51 307.932 256
convnext_tiny nhwc aot permute 815.04 312.58 256
convnext_tiny nhwc torchscript permute 805.78 316.24 256
convnext_tiny nchw aot permute 796.65 319.983 256
convnext_tiny nchw torchscript permute 786.85 323.78 256
convnext_tiny nchw aot rw_hack 766.97 332.386 256
convnext_tiny nchw aot ng_hack 692.13 368.511 256
convnext_tiny nhwc eager permute 690.18 369.352 256
convnext_tiny nhwc eager ng_hack 684.85 372.206 256
convnext_tiny_c nhwc torchscript permute 673.06 378.66 256
convnext_tiny_c nhwc aot ng_hack 672.51 379.173 256
convnext_tiny nchw eager permute 667.5 382.023 256
convnext_tiny_c nchw torchscript permute 666.05 382.712 256
convnext_tiny nchw torchscript rw_hack 661.11 385.512 256
convnext_tiny_c nhwc aot permute 660.62 386.11 256
convnext_tiny nhwc torchscript rw_hack 653.1 390.386 256
convnext_tiny nhwc aot rw_hack 651.75 391.449 256
convnext_tiny_c nchw aot permute 646.96 394.141 256
convnext_tiny nchw eager rw_hack 630.53 404.528 256
convnext_tiny_c nhwc eager permute 623.6 409.02 256
convnext_tiny_c nchw eager permute 603.65 422.557 256
convnext_tiny nchw eager rw_hack_ts 600.74 424.488 256
convnext_tiny_c nhwc eager ng_hack 599.13 425.666 256
convnext_tiny nhwc eager rw_hack 587.89 433.765 256
convnext_tiny nhwc eager rw_hack_ts 573.17 444.905 256
convnext_tiny nchw eager ng_hack 562.33 453.651 256
convnext_tiny_c nchw eager rw_hack 507.43 502.86 256
convnext_tiny_c nchw aot rw_hack 498.35 512.304 256
convnext_tiny_c nhwc eager rw_hack 487.46 523.637 256
convnext_tiny_c nchw torchscript rw_hack 469.96 542.956 256
convnext_small nhwc aot ng_hack 458.19 276.691 128
convnext_tiny_c nchw eager rw_hack_ts 454.94 561.108 256
convnext_tiny_c nhwc torchscript rw_hack 454.32 561.759 256
convnext_small nhwc aot permute 450.4 281.669 128
convnext_tiny_c nhwc aot rw_hack 448.9 568.856 256
convnext_tiny_c nhwc eager rw_hack_ts 448.71 568.931 256
convnext_small nchw aot permute 434.45 292.187 128
convnext_small nchw aot rw_hack 433.31 292.765 128
convnext_small nhwc torchscript permute 430.88 294.284 128
convnext_small nchw torchscript permute 426.79 297.103 128
convnext_small nhwc aot rw_hack 413.16 307.11 128
convnext_small nchw aot ng_hack 406.24 312.668 128
convnext_small nchw torchscript rw_hack 397.6 319.031 128
convnext_small nhwc torchscript rw_hack 391.12 324.275 128
convnext_small_c nhwc torchscript permute 382.35 331.792 128
convnext_small_c nchw torchscript permute 376.79 336.783 128
convnext_small_c nhwc aot ng_hack 365.09 698.546 256
convnext_small nhwc eager permute 362.99 702.462 256
convnext_small nhwc eager ng_hack 356.52 715.044 256
convnext_small_c nhwc aot permute 355.73 716.936 256
convnext_small_c nchw aot permute 354.97 718.562 256
convnext_small nchw eager permute 350.69 727.147 256
convnext_small nchw eager rw_hack 344.18 741.05 256
convnext_small_c nhwc eager permute 341.2 747.55 256
convnext_small_c nchw eager permute 334.33 763.03 256
convnext_small nhwc eager rw_hack 331.45 769.39 256
convnext_small_c nhwc eager ng_hack 328.46 776.664 256
convnext_small nchw eager rw_hack_ts 326.99 779.921 256
convnext_small nchw eager ng_hack 322.14 791.84 256
convnext_small nhwc eager rw_hack_ts 322.12 791.862 256
convnext_base nhwc aot ng_hack 315.01 403.86 128
convnext_tiny_c nchw aot ng_hack 312.04 818.947 256
convnext_tiny_c nchw eager ng_hack 309.24 826.374 256
convnext_base nhwc aot permute 306.97 414.202 128
convnext_base nchw aot permute 305.97 415.761 128
convnext_base nchw aot rw_hack 298.13 426.714 128
convnext_base nhwc torchscript permute 298.07 426.518 128
convnext_small_c nchw aot rw_hack 296.15 861.845 256
convnext_base nchw torchscript permute 293.38 433.542 128
convnext_small_c nchw eager rw_hack 287.61 887.187 256
convnext_small_c nchw torchscript rw_hack 286.75 443.268 128
convnext_base nchw aot ng_hack 285.95 445.07 128
convnext_small_c nhwc torchscript rw_hack 282.58 449.95 128
convnext_small_c nhwc aot rw_hack 280.95 908.672 256
convnext_small_c nhwc eager rw_hack 275.22 927.236 256
convnext_base nhwc aot rw_hack 273.27 465.805 128
convnext_small_c nchw eager rw_hack_ts 267.96 952.418 256
convnext_base nchw torchscript rw_hack 262.16 485.562 128
convnext_base_c nchw torchscript permute 262.12 485.186 128
convnext_base_c nhwc torchscript permute 259.19 490.747 128
convnext_base_c nhwc aot permute 256.82 495.839 128
convnext_base nhwc torchscript rw_hack 256.34 496.532 128
convnext_base_c nchw aot permute 256.27 496.865 128
convnext_small_c nhwc eager rw_hack_ts 255.61 497.697 128
convnext_base_c nhwc aot ng_hack 255.15 499.107 128
convnext_base nhwc eager permute 243.09 523.686 128
convnext_base nhwc eager ng_hack 238.55 533.696 128
convnext_base_c nhwc eager permute 237.65 535.861 128
convnext_base nchw eager permute 237.08 537.14 128
convnext_base nchw eager rw_hack 234.73 542.375 128
convnext_base nhwc eager rw_hack 232.14 548.499 128
convnext_base_c nchw eager permute 230.78 551.927 128
convnext_base_c nhwc eager ng_hack 228.34 557.696 128
convnext_base nchw eager ng_hack 225.58 564.519 128
convnext_base nchw eager rw_hack_ts 222.42 572.35 128
convnext_base nhwc eager rw_hack_ts 219.86 579.023 128
convnext_base_c nchw aot rw_hack 208.12 612.406 128
convnext_base_c nchw eager rw_hack 202.53 629.0 128
convnext_base_c nhwc eager rw_hack 196.89 647.016 128
convnext_base_c nhwc aot rw_hack 194.19 656.666 128
convnext_base_c nchw torchscript rw_hack 188.38 676.383 128
convnext_base_c nhwc torchscript rw_hack 182.79 697.122 128
convnext_base_c nchw eager rw_hack_ts 180.77 705.006 128
convnext_small_c nchw aot ng_hack 179.35 1424.921 256
convnext_small_c nchw eager ng_hack 176.94 1443.868 256
convnext_base_c nhwc eager rw_hack_ts 172.61 738.43 128
convnext_large nchw aot permute 156.16 406.63 64
convnext_large nhwc aot ng_hack 155.14 409.485 64
convnext_large nhwc aot permute 154.95 410.54 64
convnext_large nhwc aot rw_hack 152.25 417.628 64
convnext_large nchw aot rw_hack 151.95 418.682 64
convnext_large nhwc torchscript permute 150.45 422.39 64
convnext_large nchw torchscript permute 148.73 427.612 64
convnext_large nchw aot ng_hack 147.55 430.966 64
convnext_large nchw torchscript rw_hack 143.4 443.423 64
convnext_large nhwc torchscript rw_hack 141.95 448.099 64
convnext_large_c nhwc torchscript permute 135.9 468.059 64
convnext_large_c nchw torchscript permute 134.5 472.887 64
convnext_base_c nchw aot ng_hack 130.79 976.12 128
convnext_large_c nhwc aot permute 128.61 992.703 128
convnext_large_c nhwc aot ng_hack 128.52 993.407 128
convnext_large_c nchw aot permute 127.25 1003.393 128
convnext_base_c nchw eager ng_hack 126.84 1006.37 128
convnext_large nchw eager permute 125.98 1013.284 128
convnext_large nhwc eager permute 125.57 1016.477 128
convnext_large nhwc eager ng_hack 125.29 1018.734 128
convnext_large nchw eager rw_hack 124.68 1023.618 128
convnext_large nchw eager ng_hack 121.93 1046.427 128
convnext_large nhwc eager rw_hack 121.06 1054.197 128
convnext_large nhwc eager rw_hack_ts 121.03 1054.776 128
convnext_large_c nhwc aot rw_hack 120.56 528.212 64
convnext_large nchw eager rw_hack_ts 119.88 1064.756 128
convnext_large_c nhwc eager permute 119.12 1071.582 128
convnext_large_c nchw eager permute 118.15 1080.296 128
convnext_large_c nhwc eager ng_hack 116.27 1098.151 128
convnext_large_c nchw torchscript rw_hack 109.46 581.567 64
convnext_large_c nhwc torchscript rw_hack 108.92 584.642 64
convnext_large_c nchw eager rw_hack 104.38 1223.167 128
convnext_large_c nchw eager rw_hack_ts 102.57 620.844 64
convnext_large_c nhwc eager rw_hack 102.4 622.012 64
convnext_large_c nhwc eager rw_hack_ts 98.96 643.76 64
convnext_large_c nchw aot ng_hack 73.51 1738.218 128
convnext_large_c nchw eager ng_hack 72.34 1766.633 128
model fmt cg layer infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count
convnext_tiny nhwc eager rw_hack_ts 1996.25 128.223 256 224 573.17 444.905 256 224 28.59
convnext_tiny nhwc eager permute 1975.29 129.584 256 224 690.18 369.352 256 224 28.59
convnext_tiny nchw eager rw_hack_ts 1962.79 130.409 256 224 600.74 424.488 256 224 28.59
convnext_tiny nchw eager permute 1956.4 130.836 256 224 667.5 382.023 256 224 28.59
convnext_tiny nhwc eager ng_hack 1945.41 131.573 256 224 684.85 372.206 256 224 28.59
convnext_tiny nchw eager rw_hack 1903.62 134.463 256 224 630.53 404.528 256 224 28.59
convnext_tiny nhwc eager rw_hack 1889.58 135.461 256 224 587.89 433.765 256 224 28.59
convnext_tiny_c nhwc eager permute 1884.06 135.858 256 224 623.6 409.02 256 224 28.59
convnext_tiny_c nhwc eager rw_hack_ts 1811.37 141.312 256 224 448.71 568.931 256 224 28.59
convnext_tiny_c nchw eager permute 1774.78 144.223 256 224 603.65 422.557 256 224 28.59
convnext_tiny nchw eager ng_hack 1756.08 145.761 256 224 562.33 453.651 256 224 28.59
convnext_tiny_c nchw eager rw_hack_ts 1752.3 146.077 256 224 454.94 561.108 256 224 28.59
convnext_tiny_c nhwc eager ng_hack 1725.17 148.371 256 224 599.13 425.666 256 224 28.59
convnext_tiny_c nchw eager rw_hack 1583.8 161.618 256 224 507.43 502.86 256 224 28.59
convnext_tiny_c nhwc eager rw_hack 1503.91 170.206 256 224 487.46 523.637 256 224 28.59
convnext_tiny_c nchw eager ng_hack 1028.52 248.885 256 224 309.24 826.374 256 224 28.59
convnext_small nhwc eager permute 1004.53 254.829 256 224 362.99 702.462 256 224 50.22
convnext_small nhwc eager ng_hack 966.82 264.769 256 224 356.52 715.044 256 224 50.22
convnext_small nchw eager rw_hack 966.11 264.948 256 224 344.18 741.05 256 224 50.22
convnext_small nchw eager permute 964.77 265.33 256 224 350.69 727.147 256 224 50.22
convnext_small nchw eager rw_hack_ts 946.84 270.355 256 224 326.99 779.921 256 224 50.22
convnext_small nhwc eager rw_hack_ts 934.17 274.022 256 224 322.12 791.862 256 224 50.22
convnext_small_c nhwc eager rw_hack_ts 928.51 275.695 256 224 255.61 497.697 128 224 50.22
convnext_small_c nchw eager rw_hack_ts 925.56 276.571 256 224 267.96 952.418 256 224 50.22
convnext_small_c nhwc eager permute 919.97 278.251 256 224 341.2 747.55 256 224 50.22
convnext_small nhwc eager rw_hack 900.04 284.414 256 224 331.45 769.39 256 224 50.22
convnext_small_c nchw eager permute 889.44 287.798 256 224 334.33 763.03 256 224 50.22
convnext_small_c nhwc eager ng_hack 879.17 291.148 256 224 328.46 776.664 256 224 50.22
convnext_small nchw eager ng_hack 841.59 304.169 256 224 322.14 791.84 256 224 50.22
convnext_small_c nchw eager rw_hack 823.76 310.741 256 224 287.61 887.187 256 224 50.22
convnext_small_c nhwc eager rw_hack 782.61 327.09 256 224 275.22 927.236 256 224 50.22
convnext_base_c nhwc eager rw_hack_ts 602.51 424.87 256 224 172.61 738.43 128 224 88.59
convnext_base nhwc eager rw_hack_ts 601.57 425.538 256 224 219.86 579.023 128 224 88.59
convnext_base nhwc eager permute 601.43 425.633 256 224 243.09 523.686 128 224 88.59
convnext_base_c nchw eager rw_hack_ts 601.42 425.643 256 224 180.77 705.006 128 224 88.59
convnext_base nhwc eager ng_hack 595.92 429.571 256 224 238.55 533.696 128 224 88.59
convnext_base_c nhwc eager permute 595.48 429.865 256 224 237.65 535.861 128 224 88.59
convnext_base nchw eager rw_hack 594.5 430.599 256 224 234.73 542.375 128 224 88.59
convnext_base nhwc eager rw_hack 593.82 431.088 256 224 232.14 548.499 128 224 88.59
convnext_base nchw eager permute 587.78 435.517 256 224 237.08 537.14 128 224 88.59
convnext_base_c nchw eager permute 584.6 437.886 256 224 230.78 551.927 128 224 88.59
convnext_base nchw eager rw_hack_ts 581.66 440.098 256 224 222.42 572.35 128 224 88.59
convnext_base_c nhwc eager ng_hack 564.96 453.092 256 224 228.34 557.696 128 224 88.59
convnext_small_c nchw eager ng_hack 552.16 463.619 256 224 176.94 1443.868 256 224 50.22
convnext_base nchw eager ng_hack 547.32 467.718 256 224 225.58 564.519 128 224 88.59
convnext_base_c nchw eager rw_hack 533.24 480.063 256 224 202.53 629.0 128 224 88.59
convnext_base_c nhwc eager rw_hack 528.7 484.185 256 224 196.89 647.016 128 224 88.59
convnext_base_c nchw eager ng_hack 380.08 673.53 256 224 126.84 1006.37 128 224 88.59
convnext_large_c nhwc eager rw_hack_ts 317.88 805.324 256 224 98.96 643.76 64 224 197.77
convnext_large nchw eager rw_hack 314.61 813.683 256 224 124.68 1023.618 128 224 197.77
convnext_large nhwc eager rw_hack 311.63 821.461 256 224 121.06 1054.197 128 224 197.77
convnext_large nhwc eager rw_hack_ts 311.33 822.262 256 224 121.03 1054.776 128 224 197.77
convnext_large_c nchw eager rw_hack_ts 311.04 823.035 256 224 102.57 620.844 64 224 197.77
convnext_large nhwc eager ng_hack 310.17 825.322 256 224 125.29 1018.734 128 224 197.77
convnext_large nhwc eager permute 309.49 827.134 256 224 125.57 1016.477 128 224 197.77
convnext_large_c nhwc eager permute 307.64 832.131 256 224 119.12 1071.582 128 224 197.77
convnext_large nchw eager rw_hack_ts 305.21 838.745 256 224 119.88 1064.756 128 224 197.77
convnext_large nchw eager permute 305.2 838.754 256 224 125.98 1013.284 128 224 197.77
convnext_large_c nchw eager permute 304.56 840.529 256 224 118.15 1080.296 128 224 197.77
convnext_large_c nhwc eager ng_hack 296.82 862.444 256 224 116.27 1098.151 128 224 197.77
convnext_large nchw eager ng_hack 292.69 874.627 256 224 121.93 1046.427 128 224 197.77
convnext_large_c nhwc eager rw_hack 281.81 908.387 256 224 102.4 622.012 64 224 197.77
convnext_large_c nchw eager rw_hack 281.1 910.679 256 224 104.38 1223.167 128 224 197.77
convnext_large_c nchw eager ng_hack 214.31 1194.533 256 224 72.34 1766.633 128 224 197.77
model fmt cg layer train_samples_per_sec train_step_time train_batch_size
convnext_tiny nhwc eager permute 690.18 369.352 256
convnext_tiny nhwc eager ng_hack 684.85 372.206 256
convnext_tiny nchw eager permute 667.5 382.023 256
convnext_tiny nchw eager rw_hack 630.53 404.528 256
convnext_tiny_c nhwc eager permute 623.6 409.02 256
convnext_tiny_c nchw eager permute 603.65 422.557 256
convnext_tiny nchw eager rw_hack_ts 600.74 424.488 256
convnext_tiny_c nhwc eager ng_hack 599.13 425.666 256
convnext_tiny nhwc eager rw_hack 587.89 433.765 256
convnext_tiny nhwc eager rw_hack_ts 573.17 444.905 256
convnext_tiny nchw eager ng_hack 562.33 453.651 256
convnext_tiny_c nchw eager rw_hack 507.43 502.86 256
convnext_tiny_c nhwc eager rw_hack 487.46 523.637 256
convnext_tiny_c nchw eager rw_hack_ts 454.94 561.108 256
convnext_tiny_c nhwc eager rw_hack_ts 448.71 568.931 256
convnext_small nhwc eager permute 362.99 702.462 256
convnext_small nhwc eager ng_hack 356.52 715.044 256
convnext_small nchw eager permute 350.69 727.147 256
convnext_small nchw eager rw_hack 344.18 741.05 256
convnext_small_c nhwc eager permute 341.2 747.55 256
convnext_small_c nchw eager permute 334.33 763.03 256
convnext_small nhwc eager rw_hack 331.45 769.39 256
convnext_small_c nhwc eager ng_hack 328.46 776.664 256
convnext_small nchw eager rw_hack_ts 326.99 779.921 256
convnext_small nchw eager ng_hack 322.14 791.84 256
convnext_small nhwc eager rw_hack_ts 322.12 791.862 256
convnext_tiny_c nchw eager ng_hack 309.24 826.374 256
convnext_small_c nchw eager rw_hack 287.61 887.187 256
convnext_small_c nhwc eager rw_hack 275.22 927.236 256
convnext_small_c nchw eager rw_hack_ts 267.96 952.418 256
convnext_small_c nhwc eager rw_hack_ts 255.61 497.697 128
convnext_base nhwc eager permute 243.09 523.686 128
convnext_base nhwc eager ng_hack 238.55 533.696 128
convnext_base_c nhwc eager permute 237.65 535.861 128
convnext_base nchw eager permute 237.08 537.14 128
convnext_base nchw eager rw_hack 234.73 542.375 128
convnext_base nhwc eager rw_hack 232.14 548.499 128
convnext_base_c nchw eager permute 230.78 551.927 128
convnext_base_c nhwc eager ng_hack 228.34 557.696 128
convnext_base nchw eager ng_hack 225.58 564.519 128
convnext_base nchw eager rw_hack_ts 222.42 572.35 128
convnext_base nhwc eager rw_hack_ts 219.86 579.023 128
convnext_base_c nchw eager rw_hack 202.53 629.0 128
convnext_base_c nhwc eager rw_hack 196.89 647.016 128
convnext_base_c nchw eager rw_hack_ts 180.77 705.006 128
convnext_small_c nchw eager ng_hack 176.94 1443.868 256
convnext_base_c nhwc eager rw_hack_ts 172.61 738.43 128
convnext_base_c nchw eager ng_hack 126.84 1006.37 128
convnext_large nchw eager permute 125.98 1013.284 128
convnext_large nhwc eager permute 125.57 1016.477 128
convnext_large nhwc eager ng_hack 125.29 1018.734 128
convnext_large nchw eager rw_hack 124.68 1023.618 128
convnext_large nchw eager ng_hack 121.93 1046.427 128
convnext_large nhwc eager rw_hack 121.06 1054.197 128
convnext_large nhwc eager rw_hack_ts 121.03 1054.776 128
convnext_large nchw eager rw_hack_ts 119.88 1064.756 128
convnext_large_c nhwc eager permute 119.12 1071.582 128
convnext_large_c nchw eager permute 118.15 1080.296 128
convnext_large_c nhwc eager ng_hack 116.27 1098.151 128
convnext_large_c nchw eager rw_hack 104.38 1223.167 128
convnext_large_c nchw eager rw_hack_ts 102.57 620.844 64
convnext_large_c nhwc eager rw_hack 102.4 622.012 64
convnext_large_c nhwc eager rw_hack_ts 98.96 643.76 64
convnext_large_c nchw eager ng_hack 72.34 1766.633 128
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment