Update the docs to reflect the variable_setup

documentation/source/Recipes_Training.md

@@ -102,7 +102,7 @@ defaults:
   - arch_params: resnet18_cifar_arch_params
   - checkpoint_params: default_checkpoint_params
   - _self_
-
+  - variable_setup
 ...
 ```
 
@@ -111,6 +111,8 @@ defaults:
 - **Defaults**: The `defaults` section is critical, and it leverages the OmegaConf syntax. It serves to reference other recipes, allowing you to create modular and reusable configurations.
 - **Referencing Parameters**: This allows you to point to specific parameters in the YAML file according to where they originate. For example, `training_hyperparams.initial_lr` refers to the `initial_lr` parameter from the `cifar10_resnet_train_params.yaml` file.
 - **Recipe Parameters - `_self_`**: The `_self_` keyword has a special role. It permits the current recipe to override the defaults. Its impact depends on its position in the `defaults` list.
+- **Variable setup - `variable_setup`**: This section is required to enable use of shortcuts for most commonly used overrides which is covered in the next section. Please note it `variable_setup` **must be the last item** in the defaults list.
+
 
 ### Understanding Override Order
 
@@ -139,6 +141,24 @@ Your recipe folder should have a specific structure to match this composition:
 
 You're not restricted to this structure, but following it ensures compatibility with SuperGradients' expectations.
 
+### Command-Line Override Shortcuts
+
+Although you can override any parameter from the command line, writing the full path of the parameter can be tedious.
+For example, to change the learning rate one would have to write `training_hyperparams.initial_lr=0.02`. 
+To change the batch size one would have to write
+`dataset_params.train_dataloader_params.batch_size=128 dataset_params.val_dataloader_params.batch_size=128`.
+
+To make it easier, we have defined a few shortcuts for the most common parameters that aims to reduce the amount of typing required:
+
+* Learning rate: `lr=0.02` (same as `training_hyperparams.initial_lr=0.02`)
+* Batch size: `bs=128` (same as `dataset_params.train_dataloader_params.batch_size=128 dataset_params.val_dataloader_params.batch_size=128`)
+* Number of train epochs: `epochs=100` (same as `training_hyperparams.max_epochs=100`)
+* Number of workers: `num_workers=4` (same as `dataset_params.train_dataloader_params.num_workers=4 dataset_params.val_dataloader_params.num_workers=4`)
+* Resume training for a specific experiment: `resume=True` (same as `training_hyperparams.resume=True`)
+* Enable or disable EMA: `ema=true` (same as `training_hyperparams.ema=true`)
+
+To use these shortcuts, a `variable_setup` section should be a part of hydra defaults in the recipe file.
+Please note it `variable_setup` **must be the last item** in the defaults list.
 
 ## Conclusion
 

documentation/source/configuration_files.md

@@ -116,11 +116,46 @@ in the first arg of the command line.
 
 In the experiment directory a `.hydra` subdirectory will be created. The configuration files related to this run will be saved by hydra to that subdirectory.  
 
-
 Two Hydra features worth mentioning are [Command-Line Overrides](https://hydra.cc/docs/advanced/override_grammar/basic/) 
 and [YAML Composition](https://hydra.cc/docs/0.11/tutorial/composition/).
 We strongly recommend you to have a look at both of these pages.
 
+#### YAML Composition
+If you browse the YAML files in the `recipes` directory you will see some file containing the saved-key `defaults:` at the beginning of the file.
+```yaml
+defaults:
+  - training_hyperparams: cifar10_resnet_train_params
+  - dataset_params: cifar10_dataset_params
+  - arch_params: resnet18_cifar_arch_params
+  - checkpoint_params: default_checkpoint_params
+  - _self_
+  - variable_setup
+
+```
+The YAML file containing this header will inherit the configuration of the above files. So when building a training recipe, one can structure
+the configurations into a few files (for training hyper-params, dataset params, architecture params ect.) and Hydra will conveniently aggregate them all 
+into a single dictionary. 
+
+The parameters will be referenced inside the YAML according to their origin. i.e. in the example above we can reference `training_hyperparams.initial_lr` 
+(initial_lr parameter from the cifar10_resnet_train_params.yaml file)
+
+The aggregated configuration file will be saved in the `.hydra` subdirectory.
+
+Please note that `variable_setup` **must be the last item** in the defaults list. 
+
+#### Command-Line Overrides
+When running with Hydra, you can override or even add configuration from the command line. These override will apply to the specific run only.
+```shell
+python -m super_gradients.train_from_recipe --config-name=cifar10_resnet training_hyperparams.initial_lr=0.02 experiment_name=test_lr_002
+```
+In the example above, the same script we launched earlier is used, but this time it will run with a different experiment name and a different 
+initial learning-rate. This feature is extremely usefully when experimenting with different hyper-parameters.
+Note that the arguments are referenced without the `--` prefix and that each parameter is referenced with its full path in the 
+configuration tree, concatenated with a `.`.
+
+
+More informaiton can be found in corresponding YAML file in the `recipes` directory:
+https://github.com/Deci-AI/super-gradients/blob/master/src/super_gradients/recipes/variable_setup.yaml
 
 ### Conclusion
 This brief introduction has given you a glimpse into the functionality and importance of recipes within SuperGradients: