Add Script Mode example #83
Conversation
| "source": [ | ||
| "# Using the Script Mode to train any TensorFlow script from GitHub in SageMaker\n", | ||
| "\n", | ||
| "In this tutorial, we will show how is simple to train a TensorFlow script in SageMaker using the new Script Mode Tensorflow Container.\n", |
There was a problem hiding this comment.
"we show"
present tense, when possible, is easier to read and translate
| "\n", | ||
| "In this tutorial, we will show how is simple to train a TensorFlow script in SageMaker using the new Script Mode Tensorflow Container.\n", | ||
| "\n", | ||
| "The example the we choosed is [Multi-layer Recurrent Neural Networks (LSTM, RNN) for character-level language models in Python using Tensorflow](https://github.com/sherjilozair/char-rnn-tensorflow) but this same technique can be use to other scripts/repos including [TensorFlow Model Zoo](https://github.com/tensorflow/models) and [TensorFlow benchmark scripts](https://github.com/tensorflow/benchmarks/tree/master/scripts/tf_cnn_benchmarks).\n" |
| "source": [ | ||
| "## How does it work\n", | ||
| "\n", | ||
| "The cell above downloaded a Python 3 CPU container locally and used it to simulate SageMaker training. When training starts, the script mode will invoke the following command inside the container:\n", |
There was a problem hiding this comment.
When training starts, the container will invoke the 'train.py' script as shown bellow:
There was a problem hiding this comment.
I have more edits to suggest, but it would be easier for me to do a full edit in Jupyter, and submit that as a PR.
| "source": [ | ||
| "# Using the Script Mode to train any TensorFlow script from GitHub in SageMaker\n", | ||
| "\n", | ||
| "In this tutorial, we will show how is simple to train a TensorFlow script in SageMaker using the new Script Mode Tensorflow Container.\n", |
There was a problem hiding this comment.
"we show"
present tense, when possible, is easier to read and translate
| "source": [ | ||
| "## How does it work\n", | ||
| "\n", | ||
| "The cell above downloaded a Python 3 CPU container locally and used it to simulate SageMaker training. When training starts, the script mode will invoke the following command inside the container:\n", |
There was a problem hiding this comment.
I have more edits to suggest, but it would be easier for me to do a full edit in Jupyter, and submit that as a PR.
| "from sagemaker.estimator import Framework\n", | ||
| "from sagemaker.tensorflow import TensorFlow\n", | ||
| "\n", | ||
| "class ScriptModeTensorFlow(Framework):\n", |
There was a problem hiding this comment.
Current design is to add a flag to the existing TensorFlow estimator, script_mode=true. I talked to yuan briefly last week, we are going to deprioritize the sdk change until we get some feedbacks from the beta customers.
Beta users can use the current TensorFlow estimator with the script-mode containers. They just have to specify the entire image uri and ignore the extra arguments we passed in as hyperparamters, eg, training_steps and eval_steps. I think we can explain all of that here.
There was a problem hiding this comment.
The script in this example breaks with the TF estimator because the additional hyperparameters, which is why I had to create the estimator.
There was a problem hiding this comment.
I see. I think Nadia's idea of use scriptmode with py3 might work. This is fine as is and we can modify this notebook once we made the changes to the sdk.
| "metadata": {}, | ||
| "outputs": [], | ||
| "source": [ | ||
| "inputs = sagemaker_session.upload_data(path='sherlock', bucket=bucket, key_prefix='datasets/sherlock')" |
There was a problem hiding this comment.
For testing with localmode, we can point it out that the users can use the training data on disk.
* Add Script Mode example
* Add Script Mode example
Description of changes:
Example on how to use the Script Mode container.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.