BigDL-2.x
BigDL-2.x copied to clipboard
intel tensorflow 1.15.0up3 testing
- [ ] bf16 verification
- [x] make sure oneDNN ops can be turned off
- [ ] inception v1 example
- [x] other examples, ut
- [ ] wnd memory leak problem
@sgwhat
@sgwhat
Roger that
In the tf2/resnet training data part, we'd better change the URL of instruction. The original tensorflow/model repository's slim bazel BUILD file is broken and not working, this repository updated BUILD and it works . @yangw1234
In the tf2/resnet training data part, we'd better change the URL of instruction. The original tensorflow/model repository's slim bazel BUILD file is broken and not working, this repository updated BUILD and it works . @yangw1234
Let's use dummpy data first
- [ ] bf16 verification
- [x] make sure oneDNN ops can be turned off
- [x] inception v1 example
- [x] other examples, ut
- [x] wnd memory leak problem
Continue adding more results of other examples
- v1.15.0up3 Tensorflow
Task Name | Loss | Top1Accuracy | Running Time(s) |
---|---|---|---|
Inception v1(Part) | 5.744978 | 0.0025575447 | 1258.7487390041351 |
Basic_text_classification | 4.087298393249512 | 0.8654000163078308 | 151.16134071350098 |
Transfer_learning | 2.854679584503174 | 0.9340000152587891 | 355.7099766731262 |
Image_segmentation | 0.07223675400018692 | N/A | 269.35604453086853 |
Keras_dataset | 0.114413782954216 | 0.9639999866485596 | 32.79020833969116 |
Keras_ndarray | 0.09849555790424347 | 0.9689000248908997 | 55.92670226097107 |
- v1.15.0up3 Disable MKL-DNN Tensorflow
Task Name | Loss | Top1Accuracy | Running Time(s) |
---|---|---|---|
Inception v1(Part) | 4.9309444 | 0.0051150895 | 1634.7301709651947 |
Basic_text_classification | 4.24573278427124 | 0.8626999855041504 | 148.5364646911621 |
Transfer_learning | 2.6535277366638184 | 0.9449999928474426 | 446.9547288417816 |
Image_segmentation | 0.07158134132623672 | N/A | 480.3870759010315 |
Keras_dataset | 0.0991317480802536 | 0.9692000150680542 | 34.12212657928467 |
Keras_ndarray | 0.09951204806566238 | 0.9718000292778015 | 58.7927188873291 |
- Original Version
Task Name | Loss | Top1Accuracy | Running Time(s) |
---|---|---|---|
Inception V1(Part) | 5.7137446 | 0.0076726344 | 1312.0718114376068 |
Basic_Text_Classification | 4.267351150512695 | 0.8640000224113464 | 171.9112410545349 |
Transfer_Learning | 2.574239730834961 | 0.9629999995231628 | 475.75219345092773 |
Image_segmentation | 0.07637828588485718 | N/A | 305.888587474823 |
Keras_dataset | 0.1009390577673912 | 0.9696999788284302 | 34.34745478630066 |
Keras_ndarray | 0.09711404144763947 | 0.9707000255584717 | 60.50141477584839 |
- Performance(Speed) Comparison
Task Name | Original Version Running Time(s) | v1.15.0up3 Running Time(s) | v1.15.0up3 Disable MKL-DNN Running Time(s) | Speed Up (v1.15.0up3&Original) |
---|---|---|---|---|
Inception V1(Part) | 1312.0718114376068 | 1258.7487390041351 | 1634.7301709651947 | 4.06% |
Basic_Text_Classification | 171.9112410545349 | 151.16134071350098 | 148.5364646911621 | 12.07% |
Transfer_Learning | 475.75219345092773 | 355.7099766731262 | 446.9547288417816 | 25.23% |
Image_segmentation | 305.888587474823 | 269.35604453086853 | 480.3870759010315 | 11.94% |
Keras_dataset | 34.34745478630066 | 32.79020833969116 | 34.12212657928467 | 4.53% |
Keras_ndarray | 60.50141477584839 | 55.92670226097107 | 58.7927188873291 | 7.56% |
- Performance(Accuracy) Comparison
Task Name | Original Version Running Time(s) | v1.15.0up3 Running Time(s) | v1.15.0up3 Disable MKL-DNN Running Time(s) | Optimization (v1.15.0up3&Original) |
---|---|---|---|---|
Inception V1(Part) | 0.0076726344 | 0.0025575447 | 0.0051150895 | |
Basic_Text_Classification | 0.8640000224113464 | 0.8654000163078308 | 0.8626999855041504 | |
Transfer_Learning | 0.9629999995231628 | 0.9340000152587891 | 0.9449999928474426 | |
Image_segmentation | N/A | N/A | N/A | N/A |
Keras_dataset | 0.9696999788284302 | 0.9639999866485596 | 0.9692000150680542 | |
Keras_ndarray | 0.09711404144763947 | 0.9689000248908997 | 0.9718000292778015 |