Skip to content

Instantly share code, notes, and snippets.

@pzread
Created February 17, 2023 07:14
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save pzread/5a151eb8f3939e34550f231a878101c7 to your computer and use it in GitHub Desktop.
Save pzread/5a151eb8f3939e34550f231a878101c7 to your computer and use it in GitHub Desktop.

Full Benchmark Summary

Regressed Compilation Times 🚩

Benchmark Name Compilation Time (ms)
PoseNet\_fp32 [fp32] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 22160 (vs. 18127, 22.25%↑)
PoseNet\_fp32 [fp32] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 24222 (vs. 19940, 21.47%↑)
PoseNet\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 14149 (vs. 12263, 15.38%↑)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-vmvx-generic-vmvx] default-flags,compile-stats 30358 (vs. 27323, 11.11%↑)
DeepLabV3\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 17409 (vs. 16168, 7.68%↑)
BertLargeTF [fp32,seqlen384,tensorflow] (exported\_tf\_v1) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 24295 (vs. 23106, 5.15%↑)

Improved Compilation Times 🎉

Benchmark Name Compilation Time (ms)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 37028 (vs. 50746, 27.03%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 11220 (vs. 14403, 22.10%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 51424 (vs. 64071, 19.74%↓)
DeepLabV3\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 16983 (vs. 21075, 19.42%↓)
DeepLabV3\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 15678 (vs. 18505, 15.28%↓)
PoseNet\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 13822 (vs. 15975, 13.48%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 25039 (vs. 28823, 13.13%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 24409 (vs. 28069, 13.04%↓)
EfficientNet\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 23972 (vs. 27545, 12.97%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 38727 (vs. 44400, 12.78%↓)
PoseNet\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 13673 (vs. 15563, 12.14%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 25069 (vs. 28510, 12.07%↓)
MobileSSD\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 33496 (vs. 38088, 12.06%↓)
MobileNetV1\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-riscv\_64-generic-linux-gnu] default-flags,compile-stats 26602 (vs. 30118, 11.67%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 42674 (vs. 48109, 11.30%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [cpu-riscv\_32-generic-linux-gnu] default-flags,compile-stats 52864 (vs. 59515, 11.18%↓)
DeepLabV3\_fp32 [fp32] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 29443 (vs. 33003, 10.79%↓)
MobileNetV1\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 35185 (vs. 39408, 10.72%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 24981 (vs. 27976, 10.71%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 41490 (vs. 46455, 10.69%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 96750 (vs. 108317, 10.68%↓)
PoseNet\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 13519 (vs. 15090, 10.41%↓)
DeepLabV3\_fp32 [fp32] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 30986 (vs. 34586, 10.41%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 23955 (vs. 26668, 10.17%↓)
MobileNetV1\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 30932 (vs. 34301, 9.82%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 14875 (vs. 16472, 9.70%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 95194 (vs. 105187, 9.50%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 110658 (vs. 121819, 9.16%↓)
EfficientNet\_int8 [int8] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 58743 (vs. 64659, 9.15%↓)
EfficientNet\_int8 [int8] (exported\_tflite) [cpu-riscv\_64-generic-linux-gnu] default-flags,compile-stats 36545 (vs. 40127, 8.93%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 27275 (vs. 29915, 8.83%↓)
MiniLML12H384Uncased [int32,seqlen128] (exported\_tf\_v2) [gpu-cuda-sm\_80-linux-gnu] default-flags,compile-stats 10281 (vs. 11261, 8.70%↓)
EfficientNetV2STF [fp32,cnn,tensorflow] (exported\_tf\_v2) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 51066 (vs. 55613, 8.18%↓)
MobileBertSquad\_fp16 [fp16] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 102947 (vs. 111978, 8.06%↓)
EfficientNetV2STF [fp32,cnn,tensorflow] (exported\_tf\_v2) [gpu-cuda-sm\_80-linux-gnu] default-flags,compile-stats 33314 (vs. 36236, 8.06%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 28272 (vs. 30696, 7.90%↓)
EfficientNet\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 24124 (vs. 26115, 7.62%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 40240 (vs. 43489, 7.47%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [cpu-riscv\_64-generic-linux-gnu] default-flags,compile-stats 17997 (vs. 19439, 7.42%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 33811 (vs. 36518, 7.41%↓)
MobileSSD\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 32589 (vs. 35198, 7.41%↓)
PersonDetect\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 11979 (vs. 12934, 7.38%↓)
EfficientNet\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 29732 (vs. 32101, 7.38%↓)
MobileNetV2\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 27121 (vs. 29267, 7.33%↓)
MobileSSD\_fp32 [fp32] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 63510 (vs. 68368, 7.11%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 95324 (vs. 102526, 7.02%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 101556 (vs. 109036, 6.86%↓)
BertForMaskedLMTF [fp32,seqlen512,tensorflow] (exported\_tf\_v2) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 15024 (vs. 16094, 6.65%↓)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 26317 (vs. 28119, 6.41%↓)
MobileBertSquad\_int8 [int8] (exported\_tflite) [cpu-x86\_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 228256 (vs. 243742, 6.35%↓)
MobileBertSquad\_fp16 [fp16] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 115844 (vs. 123611, 6.28%↓)
MiniLML12H384Uncased [int32,seqlen128] (exported\_tf\_v2) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 14258 (vs. 15185, 6.10%↓)
PoseNet\_fp32 [fp32] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 24614 (vs. 26207, 6.08%↓)
BertForMaskedLMTF [fp32,seqlen512,tensorflow] (exported\_tf\_v2) [cpu-x86\_64-cascadelake-linux-gnu] default-flags,compile-stats 15658 (vs. 16669, 6.07%↓)
MobileBertSquad\_fp32 [fp32] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 87968 (vs. 93633, 6.05%↓)
MobileBertSquad\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 118003 (vs. 125199, 5.75%↓)
PoseNet\_fp32 [fp32] (exported\_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 13473 (vs. 14293, 5.74%↓)
MobileBertSquad\_fp16 [fp16] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 120793 (vs. 128004, 5.63%↓)
MobileBertSquad\_int8 [int8] (exported\_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 127350 (vs. 134730, 5.48%↓)
MobileBertSquad\_int8 [int8] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 171205 (vs. 180772, 5.29%↓)
MobileBertSquad\_int8 [int8] (exported\_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,dotprod,compile-stats 228290 (vs. 240425, 5.05%↓)

All Compilation Metrics

Benchmark Name Compilation Time (ms) Total Dispatch Size (bytes) Total Artifact Size (bytes)
DeepLabV3_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 29443 (vs. 33003, 10.79%↓) 140408 (vs. 140408, 0.00%) 2921587 (vs. 2921587, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 61613 (vs. 61428, 0.30%↑) 562184 (vs. 562184, 0.00%) 5412598 (vs. 5412598, 0.00%)
MobileBertSquad_fp16 [fp16] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 97743 (vs. 101688, 3.88%↓) 73840 (vs. 73840, 0.00%) 99929081 (vs. 99929081, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 95324 (vs. 102526, 7.02%↓) 80064 (vs. 80064, 0.00%) 98482937 (vs. 98482937, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 228555 (vs. 234513, 2.54%↓) 6046848 (vs. 6046848, 0.00%) 31249593 (vs. 31249593, 0.00%)
MobileNetV1_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 30932 (vs. 34301, 9.82%↓) 83832 (vs. 83832, 0.00%) 17000317 (vs. 17000317, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 32242 (vs. 33250, 3.03%↓) 146984 (vs. 146984, 0.00%) 14130555 (vs. 14130555, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 42674 (vs. 48109, 11.30%↓) 203432 (vs. 203432, 0.00%) 10424954 (vs. 10424954, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 49712 (vs. 50252, 1.07%↓) 258568 (vs. 258568, 0.00%) 18198771 (vs. 18198771, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 27563 (vs. 27878, 1.13%↓) 169272 (vs. 169272, 0.00%) 416630 (vs. 416630, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 22160 (vs. 18127, 22.25%↑) 81032 (vs. 81032, 0.00%) 5134193 (vs. 5134193, 0.00%)
BertForMaskedLMTF [fp32,seqlen512,tensorflow] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 15658 (vs. 16669, 6.07%↓) 62880 (vs. 62880, 0.00%) 438231621 (vs. 438231621, 0.00%)
BertLargeTF [fp32,seqlen384,tensorflow] (exported_tf_v1) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 23646 (vs. 23439, 0.88%↑) 58256 (vs. 58256, 0.00%) 1336064773 (vs. 1336064773, 0.00%)
EfficientNetV2STF [fp32,cnn,tensorflow] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 46975 (vs. 49266, 4.65%↓) 230216 (vs. 230216, 0.00%) 86832709 (vs. 86832709, 0.00%)
MiniLML12H384Uncased [int32,seqlen128] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 14258 (vs. 15185, 6.10%↓) 67104 (vs. 67104, 0.00%) 133566597 (vs. 133566597, 0.00%)
Resnet50TF [fp32] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] default-flags,compile-stats 23386 (vs. 24470, 4.43%↓) 138616 (vs. 138616, 0.00%) 102622149 (vs. 102622149, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 30986 (vs. 34586, 10.41%↓) 149208 (vs. 149208, 0.00%) 2925491 (vs. 2925491, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 58743 (vs. 64659, 9.15%↓) 568760 (vs. 568760, 0.00%) 5419062 (vs. 5419062, 0.00%)
MobileBertSquad_fp16 [fp16] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 102947 (vs. 111978, 8.06%↓) 73840 (vs. 73840, 0.00%) 99929081 (vs. 99929081, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 96750 (vs. 108317, 10.68%↓) 80064 (vs. 80064, 0.00%) 98482937 (vs. 98482937, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 228256 (vs. 243742, 6.35%↓) 6046848 (vs. 6046848, 0.00%) 31249593 (vs. 31249593, 0.00%)
MobileNetV1_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 35185 (vs. 39408, 10.72%↓) 103608 (vs. 103608, 0.00%) 17017277 (vs. 17017277, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 33811 (vs. 36518, 7.41%↓) 170216 (vs. 170216, 0.00%) 14149115 (vs. 14149115, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 38727 (vs. 44400, 12.78%↓) 234776 (vs. 234776, 0.00%) 10454714 (vs. 10454714, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 48260 (vs. 50612, 4.65%↓) 282360 (vs. 282360, 0.00%) 18217203 (vs. 18217203, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 24409 (vs. 28069, 13.04%↓) 169272 (vs. 169272, 0.00%) 416630 (vs. 416630, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 22200 (vs. 22170, 0.14%↑) 86104 (vs. 86104, 0.00%) 5134897 (vs. 5134897, 0.00%)
BertForMaskedLMTF [fp32,seqlen512,tensorflow] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 15024 (vs. 16094, 6.65%↓) 62880 (vs. 62880, 0.00%) 438231621 (vs. 438231621, 0.00%)
BertLargeTF [fp32,seqlen384,tensorflow] (exported_tf_v1) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 24295 (vs. 23106, 5.15%↑) 58256 (vs. 58256, 0.00%) 1336064773 (vs. 1336064773, 0.00%)
EfficientNetV2STF [fp32,cnn,tensorflow] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 51066 (vs. 55613, 8.18%↓) 541944 (vs. 541944, 0.00%) 87136325 (vs. 87136325, 0.00%)
MiniLML12H384Uncased [int32,seqlen128] (exported_tf_v2) [cpu-x86_64-cascadelake-linux-gnu] experimental-flags,fuse-padding,compile-stats 14331 (vs. 14811, 3.24%↓) 67104 (vs. 67104, 0.00%) 133566597 (vs. 133566597, 0.00%)
BertForMaskedLMTF [fp32,seqlen512,tensorflow] (exported_tf_v2) [gpu-cuda-sm_80-linux-gnu] default-flags,compile-stats 11848 (vs. 12393, 4.40%↓) 170712 (vs. 170712, 0.00%) 438345937 (vs. 438345937, 0.00%)
BertLargeTF [fp32,seqlen384,tensorflow] (exported_tf_v1) [gpu-cuda-sm_80-linux-gnu] default-flags,compile-stats 21774 (vs. 22075, 1.36%↓) 120028 (vs. 120028, 0.00%) 1336132459 (vs. 1336132459, 0.00%)
EfficientNetV2STF [fp32,cnn,tensorflow] (exported_tf_v2) [gpu-cuda-sm_80-linux-gnu] default-flags,compile-stats 33314 (vs. 36236, 8.06%↓) 801848 (vs. 801848, 0.00%) 87443313 (vs. 87443313, 0.00%)
MiniLML12H384Uncased [int32,seqlen128] (exported_tf_v2) [gpu-cuda-sm_80-linux-gnu] default-flags,compile-stats 10281 (vs. 11261, 8.70%↓) 126076 (vs. 126076, 0.00%) 133633979 (vs. 133633979, 0.00%)
Resnet50TF [fp32] (exported_tf_v2) [gpu-cuda-sm_80-linux-gnu] default-flags,compile-stats 15325 (vs. 15481, 1.01%↓) 345816 (vs. 345816, 0.00%) 102845707 (vs. 102845707, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 19913 (vs. 20206, 1.45%↓) 47552 (vs. 47552, 0.00%) 2828789 (vs. 2828789, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 90172 (vs. 93340, 3.39%↓) 22432 (vs. 22432, 0.00%) 98425275 (vs. 98425275, 0.00%)
MobileNetV1_fp32 [fp32,imagenet] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 26602 (vs. 30118, 11.67%↓) 37272 (vs. 37272, 0.00%) 16953727 (vs. 16953727, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 157069 (vs. 162319, 3.23%↓) 1992952 (vs. 1992952, 0.00%) 27195707 (vs. 27195707, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 17997 (vs. 19439, 7.42%↓) 64336 (vs. 64336, 0.00%) 311672 (vs. 311672, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [cpu-riscv_64-generic-linux-gnu] default-flags,compile-stats 36545 (vs. 40127, 8.93%↓) 144768 (vs. 144768, 0.00%) 4995256 (vs. 4995256, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [cpu-riscv_32-generic-linux-gnu] default-flags,compile-stats 121892 (vs. 127757, 4.59%↓) 2541728 (vs. 2541728, 0.00%) 7392184 (vs. 7392184, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-riscv_32-generic-linux-gnu] default-flags,compile-stats 924230 (vs. 940304, 1.71%↓) 49611248 (vs. 49611248, 0.00%) 74814011 (vs. 74814011, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [cpu-riscv_32-generic-linux-gnu] default-flags,compile-stats 52864 (vs. 59515, 11.18%↓) 915184 (vs. 915184, 0.00%) 1162552 (vs. 1162552, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 33966 (vs. 35664, 4.76%↓) 98056 (vs. 98056, 0.00%) 2879219 (vs. 2879219, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 74263 (vs. 77626, 4.33%↓) 290584 (vs. 290584, 0.00%) 18230771 (vs. 18230771, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 24614 (vs. 26207, 6.08%↓) 34136 (vs. 34136, 0.00%) 5087345 (vs. 5087345, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 101556 (vs. 109036, 6.86%↓) 109176 (vs. 109176, 0.00%) 98512057 (vs. 98512057, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 40240 (vs. 43489, 7.47%↓) 102824 (vs. 102824, 0.00%) 14086459 (vs. 14086459, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 37028 (vs. 50746, 27.03%↓) 91208 (vs. 91208, 0.00%) 10312698 (vs. 10312698, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] default-flags,compile-stats 171205 (vs. 180772, 5.29%↓) 1883544 (vs. 1883544, 0.00%) 27086265 (vs. 27086265, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 37575 (vs. 38930, 3.48%↓) 90992 (vs. 90992, 0.00%) 2889779 (vs. 2889779, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 63510 (vs. 68368, 7.11%↓) 177280 (vs. 177280, 0.00%) 18137907 (vs. 18137907, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 24222 (vs. 19940, 21.47%↑) 45744 (vs. 45744, 0.00%) 5104881 (vs. 5104881, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 87968 (vs. 93633, 6.05%↓) 37952 (vs. 37952, 0.00%) 98622777 (vs. 98622777, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 41490 (vs. 46455, 10.69%↓) 112768 (vs. 112768, 0.00%) 14111355 (vs. 14111355, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,compile-stats 51424 (vs. 64071, 19.74%↓) 138176 (vs. 138176, 0.00%) 10379706 (vs. 10379706, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [cpu-armv8.2-a-generic-linux-android29] experimental-flags,mmt4d,dotprod,compile-stats 228290 (vs. 240425, 5.05%↓) 4090288 (vs. 4090288, 0.00%) 29496121 (vs. 29496121, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 16983 (vs. 21075, 19.42%↓) 389644 (vs. 389644, 0.00%) 3189359 (vs. 3189359, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 33496 (vs. 38088, 12.06%↓) 542338 (vs. 542338, 0.00%) 18516251 (vs. 18516251, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 14149 (vs. 12263, 15.38%↑) 137108 (vs. 137108, 0.00%) 5201067 (vs. 5201067, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 94290 (vs. 97348, 3.14%↓) 226556 (vs. 226556, 0.00%) 98640839 (vs. 98640839, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 25658 (vs. 26115, 1.75%↓) 329596 (vs. 329596, 0.00%) 14331695 (vs. 14331695, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] default-flags,compile-stats 28272 (vs. 30696, 7.90%↓) 394660 (vs. 394660, 0.00%) 10647977 (vs. 10647977, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 16220 (vs. 16206, 0.09%↑) 412560 (vs. 412560, 0.00%) 3204232 (vs. 3204232, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 32589 (vs. 35198, 7.41%↓) 604820 (vs. 604820, 0.00%) 18568322 (vs. 18568322, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 13473 (vs. 14293, 5.74%↓) 159596 (vs. 159596, 0.00%) 5216797 (vs. 5216797, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 95566 (vs. 100266, 4.69%↓) 226556 (vs. 226556, 0.00%) 98640839 (vs. 98640839, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 23145 (vs. 24227, 4.47%↓) 374542 (vs. 374542, 0.00%) 14367681 (vs. 14367681, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,compile-stats 23955 (vs. 26668, 10.17%↓) 411694 (vs. 411694, 0.00%) 10660782 (vs. 10660782, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 34845 (vs. 34023, 2.42%↑) 604820 (vs. 604820, 0.00%) 18651522 (vs. 18651522, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 13822 (vs. 15975, 13.48%↓) 159596 (vs. 159596, 0.00%) 5245405 (vs. 5245405, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 24981 (vs. 27976, 10.71%↓) 374542 (vs. 374542, 0.00%) 14414721 (vs. 14414721, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-adreno-generic-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 25069 (vs. 28510, 12.07%↓) 411694 (vs. 411694, 0.00%) 10739758 (vs. 10739758, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 17409 (vs. 16168, 7.68%↑) 178304 (vs. 178304, 0.00%) 2978159 (vs. 2978159, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 35095 (vs. 34313, 2.28%↑) 353770 (vs. 353770, 0.00%) 18327707 (vs. 18327707, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 13519 (vs. 15090, 10.41%↓) 95576 (vs. 95576, 0.00%) 5159595 (vs. 5159595, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 91290 (vs. 95998, 4.90%↓) 129892 (vs. 129892, 0.00%) 98544135 (vs. 98544135, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 27121 (vs. 29267, 7.33%↓) 194320 (vs. 194320, 0.00%) 14196527 (vs. 14196527, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] default-flags,compile-stats 26317 (vs. 28119, 6.41%↓) 263936 (vs. 263936, 0.00%) 10517353 (vs. 10517353, 0.00%)
MobileBertSquad_fp16 [fp16] (exported_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 115844 (vs. 123611, 6.28%↓) 2973114 (vs. 2973114, 0.00%) 52951950 (vs. 52951950, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 118003 (vs. 125199, 5.75%↓) 2554516 (vs. 2554516, 0.00%) 27996660 (vs. 27996660, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 23972 (vs. 27545, 12.97%↓) 306826 (vs. 306826, 0.00%) 5195036 (vs. 5195036, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] default-flags,demote-f32-to-f16,compile-stats 11220 (vs. 14403, 22.10%↓) 199310 (vs. 199310, 0.00%) 470866 (vs. 470866, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 14495 (vs. 14236, 1.82%↑) 202756 (vs. 202756, 0.00%) 2994568 (vs. 2994568, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 32937 (vs. 34082, 3.36%↓) 378628 (vs. 378628, 0.00%) 18342146 (vs. 18342146, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 13239 (vs. 13584, 2.54%↓) 118720 (vs. 118720, 0.00%) 5176029 (vs. 5176029, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 95194 (vs. 105187, 9.50%↓) 129892 (vs. 129892, 0.00%) 98544135 (vs. 98544135, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 21262 (vs. 20941, 1.53%↑) 213966 (vs. 213966, 0.00%) 14207169 (vs. 14207169, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,compile-stats 25039 (vs. 28823, 13.13%↓) 288894 (vs. 288894, 0.00%) 10538094 (vs. 10538094, 0.00%)
MobileBertSquad_fp16 [fp16] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 120793 (vs. 128004, 5.63%↓) 2973114 (vs. 2973114, 0.00%) 52951950 (vs. 52951950, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 127350 (vs. 134730, 5.48%↓) 2554516 (vs. 2554516, 0.00%) 27996660 (vs. 27996660, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 24124 (vs. 26115, 7.62%↓) 308184 (vs. 308184, 0.00%) 5195822 (vs. 5195822, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,demote-f32-to-f16,compile-stats 11979 (vs. 12934, 7.38%↓) 199310 (vs. 199310, 0.00%) 470866 (vs. 470866, 0.00%)
DeepLabV3_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 15678 (vs. 18505, 15.28%↓) 202756 (vs. 202756, 0.00%) 3098760 (vs. 3098760, 0.00%)
MobileSSD_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 35085 (vs. 36508, 3.90%↓) 378628 (vs. 378628, 0.00%) 18513986 (vs. 18513986, 0.00%)
PoseNet_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 13673 (vs. 15563, 12.14%↓) 118720 (vs. 118720, 0.00%) 5235101 (vs. 5235101, 0.00%)
MobileBertSquad_fp32 [fp32] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 110658 (vs. 121819, 9.16%↓) 129892 (vs. 129892, 0.00%) 99849607 (vs. 99849607, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 27275 (vs. 29915, 8.83%↓) 213966 (vs. 213966, 0.00%) 14304385 (vs. 14304385, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,compile-stats 28318 (vs. 29248, 3.18%↓) 288894 (vs. 288894, 0.00%) 10701294 (vs. 10701294, 0.00%)
MobileBertSquad_fp16 [fp16] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 157910 (vs. 166002, 4.87%↓) 2973114 (vs. 2973114, 0.00%) 54257486 (vs. 54257486, 0.00%)
MobileBertSquad_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 231969 (vs. 236775, 2.03%↓) 2554516 (vs. 2554516, 0.00%) 30552052 (vs. 30552052, 0.00%)
EfficientNet_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 29732 (vs. 32101, 7.38%↓) 308184 (vs. 308184, 0.00%) 5405870 (vs. 5405870, 0.00%)
PersonDetect_int8 [int8] (exported_tflite) [gpu-valhall-mali-android31] experimental-flags,fuse-padding,repeated-kernel,demote-f32-to-f16,compile-stats 14875 (vs. 16472, 9.70%↓) 199310 (vs. 199310, 0.00%) 601106 (vs. 601106, 0.00%)
MobileNetV2_fp32 [fp32,imagenet] (exported_tflite) [cpu-vmvx-generic-vmvx] default-flags,compile-stats 26897 (vs. 27000, 0.38%↓) 96949 (vs. 96949, 0.00%) 14080564 (vs. 14080564, 0.00%)
MobileNetV3Small_fp32 [fp32,imagenet] (exported_tflite) [cpu-vmvx-generic-vmvx] default-flags,compile-stats 30358 (vs. 27323, 11.11%↑) 165685 (vs. 165685, 0.00%) 10387187 (vs. 10387187, 0.00%)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment