- One of the most powerful features of OwLite is its web GUI, where you can review the model structure and apply compression.
Compression panel
- The compression panel provides options for compressing a specific layer.
- For detailed information about each option, please take a look at the following link.
Compression methods
Recommended setting
- A Recommended popup will automatically appear when creating the first Experiment from a Baseline in OwLite.
- In other cases, you can access the settings by going to Tools > Recommended setting in the editor's top menu.
- The engineers at SqueezeBits have optimized compression configurations for each model, which can be applied with a single click in OwLite.
For Free plan users, usage is confined to the recommendation feature. However, SqueezeBits will periodically display benchmark results of the recommended algorithms for enhanced insight.
OpType setting
- This functionality allows you to select a specific type of Operation and uniformly apply the desired compression settings.
- In the configuration hierarchy, the OpType setting always precedes the Overall setting.
- Therefore, if you apply the Overall setting first and then the OpType setting in OwLite, the OpType setting will overwrite the Overall setting without any separate notification.
For Free plan users, the OpType setting is limited to only one.
Layer setting
- You can select a specific Layer and apply your desired Compression setting to that particular Layer.
- By using CMD (on Mac) or Ctrl (on Windows), you can multi-select layers of the same type and apply settings to them in bulk.
- However, even if the layers are of a single type, multi-selection is impossible if the layers include a mix of weight-input cases and input-input cases.
- By using CMD (on Mac) or Ctrl (on Windows), you can multi-select layers of the same type and apply settings to them in bulk.
- In the configuration hierarchy, the Layer setting always precedes the Overall and OpType settings.
- Therefore, if you apply the Overall setting or the OpType setting and then apply the Layer setting in OwLite, the Layer setting will overwrite the previous settings without any separate notification.
For Free plan users, the Layer setting is limited up to 10.
Saving compression configuration
- In OwLite, if any Overall, OpType, or Layer settings are applied to a specific Experiment, the save button next to the Experiment name in the editor becomes active.
- After saving the Compression configuration in OwLite, you can return to the PyTorch environment and execute the code. This will result in obtaining ONNX and Tensor RT models with the applied compression configuration.
- For more detailed information, please refer to the information provided below.