[tools/onert_train] Add onert_train usage example to readme (Samsung#…

…12714)

This PR adds a example to use onert_train in the readme.

ONE-DCO-1.0-Signed-off-by: SeungHui Youn <[email protected]>

tests/tools/onert_train/README.md

@@ -34,3 +34,61 @@ onert_train \
 ```
 
 `onert_train --help` would help you to set each parameter.
+
+## Example
+
+To deliver a quick insight to use `onert_train`, let's train a simple mnist model. You could get a mnist tensroflow model code from [here](https://www.kaggle.com/code/amyjang/tensorflow-mnist-cnn-tutorial). 
+
+Before using `onert_train`, training data files and a model file have to be ready.
+
+### Prepare training data files
+
+`onert_train` expects that a preprocessed dataset is given as binary files. <br/>
+For convenience, we provide a tool([tf dataset convert](https://github.com/Samsung/ONE/tree/master/tools/generate_datafile/tf_dataset_converter)) that preprocesses tensorflow dataset and save it as binary files.
+
+You could use the tool like this. For detailed usage, please refer [here](https://github.com/Samsung/ONE/tree/master/tools/generate_datafile/tf_dataset_converter#readme).
+```bash
+# Move to tf_dataset_convert directory 
+$ cd ONE/tools/generate_datafile/tf_dataset_converter
+
+# install prerequisites
+$ pip3 install -r requirements.txt
+
+# generate binary data files
+$ python3 main.py \ 
+--dataset-name mnist \ 
+--prefix-name mnist \ 
+--model mnist 
+
+# check data files are generated
+# There are 'mnist.train.input.1000.bin' and 'mnist.train.output.1000.bin'
+$ tree out
+```
+
+### Prepare a circle model file
+
+`onert_train` use a `*.circle` file or a nnpackage as input. <br/>
+
+<!-- This readme is for the ONE developers, so they might know the onecc usage.--> 
+You could convert tf/tflite/onnx model file into circle file using [`onecc`](https://github.com/Samsung/ONE/tree/master/compiler/one-cmds). <br/> 
+If you start with tensorflow code, you could first save it as saved format and then convert it to a circle file by using `onecc`. 
+
+<!--TODO : Add how to inject training parameter in the circle model -->
+
+### Run onert_train
+Now you're ready to run `onert_train`. <br/>
+Please pass your model file to `--modelfile` and data files to `--load_input:raw` and `--load_expected:raw`. <br/>
+Also, you could set training parameter using options like `--batch_size`, `--epoch`.. etc.
+
+```bash 
+$ onert_train \
+--modelfile mnist.circle \
+--load_input:raw mnist.train.input.1000.bin \
+--load_expected:raw mnist.train.output.1000.bin \
+--batch_size 32 \ 
+--epoch 5 \
+--optimizer 2 \          # adam
+--learning_rate 0.001 \
+--loss 2 \               # cateogrical crossentropy
+--loss_reduction_type 1  # sum over batch size
+```