See: Description
| Interface | Description |
|---|---|
| GradientCollector |
An interface that provides a mechanism to collect gradients during training.
|
| ParameterServer |
An interface for a key-value store to store parameters, and their corresponding gradients.
|
| TrainingConfig |
An interface that is responsible for holding the configuration required by
Trainer. |
| Class | Description |
|---|---|
| DataManager |
DataManager is an class that is used primarily in the Trainer to manipulate data
from the dataset to be appropriate for the current model. |
| DefaultTrainingConfig |
DefaultTrainingConfig is an implementation of the TrainingConfig interface. |
| EasyTrain |
Helper for easy training of a whole model, a trainining batch, or a validation batch.
|
| LocalParameterServer |
LocalParameterServer is an implementation of the ParameterServer interface. |
| ParallelTrain |
Helper for easy training of a whole model on multiple GPUs in parallel.
|
| ParameterStore |
The
ParameterStore contains a map from a parameter to the mirrors of it on other devices. |
| Trainer |
The
Trainer interface provides a session for model training. |
| TrainingResult |
A class that is responsible for holding the training result produced by
Trainer. |
Trainer