Training MnasNet on Cloud TPU

This tutorial shows you how to train the Tensorflow MnasNet model using a Cloud TPU device or Cloud TPU Pod slice (multiple TPU devices). You can apply the same pattern to other TPU-optimised image classification models that use TensorFlow and the ImageNet dataset.

Model description

The model in this tutorial is based on MnasNet: Platform-Aware Neural Architecture Search for Mobile, which first introduces the AutoML mobile neural network (MnasNet) architecture. The tutorial uses the state-of-the-art variant, 'mnasnet-a1', and demonstrates training the model using TPUEstimator.

Warning: This tutorial uses a third-party dataset. Google provides no representation, warranty, or other guarantees about the validity, or any other aspects of this dataset.

Special considerations when training on a Pod slice (v2-32/v3-32 and above)

If you plan to train on a TPU Pod slice, please make sure you read this document that explains the special considerations when training on a Pod slice.

Objectives

• Create a Cloud Storage bucket to hold your dataset and model output.
• Run the training job.
• Verify the output results.

Costs

This tutorial uses billable components of Google Cloud, including:

• Compute Engine
• Cloud TPU
• Cloud Storage

Use the pricing calculator to generate a cost estimate based on your projected usage.

New Google Cloud users might be eligible for a free trial.

Before you begin

Before starting this tutorial, check that your Google Cloud project is correctly set up.

1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

2. In the Google Cloud Console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

3. Make sure that billing is enabled for your Cloud project. Learn how to confirm that billing is enabled for your project.

This walkthrough uses billable components of Google Cloud. Check the Cloud TPU pricing page to estimate your costs. Be sure to clean up resources you create when you've finished with them to avoid unnecessary charges.

Set up your resources

This section provides information on setting up Cloud Storage, VM, and Cloud TPU resources for tutorials.

Important: Set up all resources in the same region/zone to reduce network latency and network costs.

1. Open a Cloud Shell window.

   Open Cloud Shell

2. Create a variable for your project's ID.

   export PROJECT_ID=project-id
   

3. Configure gcloud command-line tool to use the project where you want to create Cloud TPU.

   gcloud config set project ${PROJECT_ID}
   

   The first time you run this command in a new Cloud Shell VM, an Authorize Cloud Shell page is displayed. Click Authorize at the bottom of the page to allow gcloud to make GCP API calls with your credentials.

4. Create a Service Account for the Cloud TPU project.

   gcloud beta services identity create --service tpu.googleapis.com --project $PROJECT_ID
   

   The command returns a Cloud TPU Service Account with following format:

   [e-mail]
   

5. Create a Cloud Storage bucket using the following command. Replace bucket-name with a name for your bucket.

   gsutil mb -p ${PROJECT_ID} -c standard -l europe-west4 -b on gs://bucket-name
   

   This Cloud Storage bucket stores the data you use to train your model and the training results. The gcloud compute tpus execution-groups tool used in this tutorial sets up default permissions for the Cloud TPU Service Account. If you want finer-grain permissions, review the access level permissions.

   The bucket location must be in the same region as your virtual machine (VM) and your TPU node. VMs and TPU nodes are located in specific zones, which are subdivisions within a region.

6. Launch the Compute Engine and Cloud TPU resources required for this tutorial using the gcloud compute tpus execution-groups command.

   Note: This tutorial uses TensorFlow 1.15.5 which is based on the Debian 10 image and defaults to Python 3.6. Earlier versions of TensorFlow 1.15.x were based on Debian 9 and default to Python 3.5.

   gcloud compute tpus execution-groups create \
    --vm-only \
    --name=mnasnet-tutorial \
    --zone=europe-west4-a \
    --disk-size=300 \
    --machine-type=n1-standard-8 \
    --tf-version=1.15.5
   

   Command flag descriptions

   vm-only

   Create the Compute Engine VM only, do not create a Cloud TPU.

   name

   The name of the Cloud TPU to create.

   zone

   The zone where you plan to create your Cloud TPU.

   disk-size

   The size of the hard disk in GB of the VM created by the gcloud command.

   machine-type

   The machine type of the Compute Engine VM to create.

   tf-version

   The version of Tensorflow gcloud installs on the VM.

   For more information on the gcloud command, see the gcloud Reference.

7. When prompted, press y to create your Cloud TPU resources.

   Note: The first time you run gcloud compute tpus execution-groups on a project it takes about 5 minutes to perform startup tasks such as SSH key propagation and API turnup.

   When the gcloud compute tpus execution-groups command has finished executing, verify that your shell prompt has changed from username@projectname to username@vm-name. This change shows that you are now logged into your Compute Engine VM. If you are not connected to the Compute Engine instance, you can do so by running the following command:

   gcloud compute ssh mnasnet-tutorial --zone=europe-west4-a
   

   From this point on, a prefix of (vm)$ means you should run the command on the Compute Engine VM instance.

8. Create an environment variable for the storage bucket. Replace bucket-name with the name of your Cloud Storage bucket.

   (vm)$ export STORAGE_BUCKET=gs://bucket-name
   

9. Create an environment variable for the model directory.

   (vm)$ export MODEL_DIR=${STORAGE_BUCKET}/mnasnet
   

10. Add the top-level /models folder to the Python path with the command

    (vm)$ export PYTHONPATH=${PYTHONPATH}:/usr/share/tpu/models:/usr/share/tpu/models/official/efficientnet
    

Locate the data

ImageNet is an image database. The images in the database are organized into a hierarchy, each node of the hierarchy contains hundreds and thousands of images.

This tutorial uses a demonstration version of the full ImageNet dataset, referred to as fake_imagenet. This demonstration version allows you to test the tutorial, while reducing the storage and time requirements typically associated with running a model against the full ImageNet database.

The fake_imagenet dataset is at this location on Cloud Storage:

gs://cloud-tpu-test-datasets/fake_imagenet

1. Create an environment variable for the data directory.

   (vm)$ export DATA_DIR=gs://cloud-tpu-test-datasets/fake_imagenet
   

The fake_imagenet dataset is only useful for understanding how to use a Cloud TPU and validating end-to-end performance. The accuracy numbers and saved model will not be meaningful.

For information on how to download and process the full ImageNet dataset, see Downloading, preprocessing, and uploading the ImageNet dataset.

Caution: For this tutorial, make sure you don't set the STORAGE_BUCKET environment variable to the path of the fake_imagenet dataset. You can read from gs://cloud-tpu-test-datasets but you can't write to it. As a result, you can't use it to write out training logs. Make sure the STORAGE_BUCKET environment variable is set to your own Cloud Storage bucket, as shown above.

Train the MnasNet model with fake_imagenet

The Mnasnet TPU model is pre-installed on your Compute Engine VM in the following directory:

/usr/share/tpu/models/official/mnasnet/

Note: If you want to monitor the model's output and performance, follow the guide to setting up TensorBoard.

1. Run the following command to create your Cloud TPU.

   (vm)$ gcloud compute tpus execution-groups create \
    --tpu-only \
    --accelerator-type=v3-8 \
    --name=mnasnet-tutorial \
    --zone=europe-west4-a \
    --tf-version=1.15.5
   

   Command flag descriptions

   tpu-only

   Create the Cloud TPU only, does not create a Compute Engine.

   accelerator-type

   The type of the Cloud TPU to create.

   name

   The name of the Cloud TPU to create.

   zone

   The zone where you plan to create your Cloud TPU.

   tf-version

   The version of Tensorflow gcloud installs on the VM.

2. Create an environment variable for the Cloud TPU name.

   (vm)$ export TPU_NAME=mnasnet-tutorial
   

3. Navigate to the model directory.

   (vm)$ cd /usr/share/tpu/models/official/mnasnet/
   

4. Run the training script.

   (vm)$ python3 mnasnet_main.py \
     --tpu=${TPU_NAME} \
     --data_dir=${DATA_DIR} \
     --model_dir=${MODEL_DIR} \
     --model_name="mnasnet-a1" \
     --skip_host_call=true \
     --train_steps=109474 \
     --train_batch_size=4096

   Command flag descriptions

   tpu

   The name of the Cloud TPU to run training or evaluation.

   data_dir

   Specifies the Cloud Storage path for training input. It is set to the fake_imagenet dataset in this example.

   model_dir

   The Cloud Storage path where checkpoints and summaries are stored during model training. You can reuse an existing folder to load previously generated checkpoints and to store additional checkpoints as long as the previous checkpoints were created using a Cloud TPU of the same size and TensorFlow version.

   model_name

   The name of the model to train. For example, mnasnet-a1.

   skip_host_call

   Set to true to instruct the script skip the host_call which is executed every training step. This is generally used for generating training summaries (training loss, learning rate, and so on). When skip_host_call=false, there could be a performance drop if the host_call function is slow and cannot keep up with the TPU-side computation.

   train_steps

   The number of steps to use for training. The default value is 218,949 which is approximately 350 epochs at a batch size of 2048. This flag should be adjusted according to the train_batch_size value.

   train_batch_size

   The training batch size.

   For a single Cloud TPU device, the procedure trains the MnasNet model ('mnasnet-a1' variant) for 350 epochs and evaluates every fixed number of steps. Using the specified flags, the model should train in about 23 hours. With real imagenet data, the settings will reproduce the state-of-the-art research result, while users should be able to tune up the training speed.

Scaling your model with Cloud TPU Pods

You can get results faster by scaling your model with Cloud TPU Pods. The fully supported Mnasnet model can work with the following Pod slices:

• v2-32
• v2-128
• v2-256
• v2-512
• v3-32
• v3-128
• v3-256
• v3-512
• v3-1024
• v3-2048

When working with Cloud TPU Pods, you first train the model using a Pod, then use a single Cloud TPU device to evaluate the model.

Training with Cloud TPU Pods

Note: If you have already deleted your Compute Engine instance, create a new one following the steps in Set up your resources.

1. Delete the Cloud TPU resource you created for training the model on a single device.

   (vm)$ gcloud compute tpus execution-groups delete mnasnet-tutorial \
   --tpu-only \
   --zone=europe-west4-a

2. Create a new Cloud TPU pod.

   (vm)$ gcloud compute tpus execution-groups create --tpu-only \
     --accelerator-type=v2-32 \
     --zone=europe-west4-a \
     --name=mnasnet-tutorial \
     --tf-version=1.15.5
   

   Command flag descriptions

   tpu-only

   Create a Cloud TPU only. By default the gcloud command creates a VM and a Cloud TPU.

   accelerator-type

   The type of the Cloud TPU to create.

   zone

   The zone where you plan to create your Cloud TPU.

   name

   The name of the Cloud TPU to create.

   tf-version

   The version of Tensorflow gcloud installs on the VM.

3. Update the TPU_NAME environment variable.

   (vm)$ export TPU_NAME=Cloud TPU
   

4. Update the MODEL_DIR directory to store the training data.

   (vm)$ export MODEL_DIR=${STORAGE_BUCKET}/mnasnet-tutorial
   

5. Train the model by running the following script.

   The script trains the model on the fake_imagnet dataset to 35 epochs. This takes approximately 90 minutes to run on a v3-128 Cloud TPU.

   (vm)$ python3 mnasnet_main.py \
     --tpu=${TPU_NAME} \
     --data_dir=${DATA_DIR} \
     --model_dir=${MODEL_DIR} \
     --model_name="mnasnet-a1" \
     --skip_host_call=true \
     --train_steps=109474 \
     --train_batch_size=4096
   

   Command flag descriptions

   tpu

   The name of the Cloud TPU.

   data_dir

   Specifies the Cloud Storage path for training input. It is set to the fake_imagenet dataset in this example.

   model_dir

   The Cloud Storage path where checkpoints and summaries are stored during model training. You can reuse an existing folder to load previously generated checkpoints and to store additional checkpoints as long as the previous checkpoints were created using a Cloud TPU of the same size and TensorFlow version.

   model_name

   The name of the model to train. For example, efficientnet.

   skip_host_call

   Set to true to instruct the script skip the host_call which is executed every training step. This is generally used for generating training summaries (training loss, learning rate, and so on). When skip_host_call=false, there could be a performance drop if the host_call function is slow and cannot keep up with the TPU-side computation.

   train_steps

   The number of steps to use for training. The default value is 218,949 steps which is approximately 350 epochs at a batch size of 2048. This flag should be adjusted according to the train_batch_size flag value.

   train_batch_size

   The training batch size.

   The procedure trains the MnasNet model ('mnasnet-a1' variant) on the fake_imagent data set to 350 epochs. It should finish in around 5 hours.

Evaluating the model

In this set of steps, you use Cloud TPU to evaluate the above trained model against the fake_imagenet validation data.

1. Delete the Cloud TPU resource you created to train the model on a Pod.

   (vm)$ gcloud compute tpus execution-groups delete mnasnet-tutorial \
     --tpu-only \
     --zone=europe-west4-a
   

2. Start a v2-8 Cloud TPU. Use the same name that you used for the Compute Engine VM, which should still be running.

   (vm)$ gcloud compute tpus execution-groups create --tpu-only \
     --accelerator-type=v3-8 \
     --zone=europe-west4-a \
     --name=mnasnet-tutorial \
     --tf-version=1.15.5
   

   Command flag descriptions

   tpu-only

   Create a Cloud TPU only. By default the gcloud command creates a VM and a Cloud TPU.

   accelerator-type

   The type of the Cloud TPU to create.

   zone

   The zone where you plan to create your Cloud TPU.

   name

   The name of the Cloud TPU to create.

   tf-version

   The version of Tensorflow gcloud installs on the VM.

3. Create an environment variable for your accelerator type.

   (vm)$ export ACCELERATOR_TYPE=v3-8
   

4. Run the model evaluation. This time, add the mode flag and set it to eval.

   (vm)$ python3 mnasnet_main.py \
     --tpu=${TPU_NAME} \
     --data_dir=${DATA_DIR} \
     --model_dir=${MODEL_DIR} \
     --mode=eval \
     --config_file=configs/cloud/${ACCELERATOR_TYPE}.yaml
   

   Command flag descriptions

   tpu

   The name of the Cloud TPU.

   data_dir

   Specifies the Cloud Storage path for training input. It is set to the fake_imagenet dataset in this example.

   model_dir

   The Cloud Storage path where checkpoints and summaries are stored during model training. You can reuse an existing folder to load previously generated checkpoints and to store additional checkpoints as long as the previous checkpoints were created using a Cloud TPU of the same size and TensorFlow version.

   mode

   One of train_and_eval, train or eval. train_and_eval trains and evaluates the model. train trains the model. eval evaluates the model.

   config_file

   The configuration file used by the training/evaluation script.

   This generates output similar to the following:

   Eval results: {
   'loss': 7.532023,
   'top_1_accuracy': 0.0010172526,
   'global_step': 100,
   'top_5_accuracy': 0.005065918
   }
   Elapsed seconds: 88

Cleaning up

To avoid incurring charges to your GCP account for the resources used in this topic:

1. Disconnect from the Compute Engine VM:

   (vm)$ exit
   

   Your prompt should now be user@projectname, showing you are in the Cloud Shell.

2. Delete your Cloud TPU and Compute Engine resources.

   $ gcloud compute tpus execution-groups delete mnasnet-tutorial \
     --zone=europe-west4-a
   

3. Verify the resources have been deleted by running gcloud compute tpus execution-groups list. The deletion might take several minutes. A response like the one below indicates your instances have been successfully deleted.

   $ gcloud compute tpus execution-groups list \
     --zone=europe-west4-a
   

      NAME             STATUS
   

4. Delete your Cloud Storage bucket using gsutil as shown below. Replace bucket-name with the name of your Cloud Storage bucket.

   $ gsutil rm -r gs://bucket-name
   

Note: For free storage limits and other pricing information, see the Cloud Storage pricing guide.

What's next

In this tutorial you have trained the MNASNET model using a sample dataset. The results of this training are (in most cases) not usable for inference. To use a model for inference you can train the data on a publicly available dataset or your own data set. Models trained on Cloud TPUs require datasets to be in TFRecord format.

You can use the dataset conversion tool sample to convert an image classification dataset into TFRecord format. If you are not using an image classification model, you will have to convert your dataset to TFRecord format yourself. For more information, see TFRecord and tf.Example

Hyperparameter tuning

To improve the model's performance with your dataset, you can tune the model's hyperparameters. You can find information about hyperparameters common to all TPU supported models on GitHub. Information about model-specific hyperparameters can be found in the source code for each model. For more information on hyperparameter tuning, see Overview of hyperparameter tuning, Using the Hyperparameter tuning service and Tune hyperparameters.

Inference

Once you have trained your model you can use it for inference (also called prediction). AI Platform is a cloud-based solution for developing, training, and deploying machine learning models. Once a model is deployed, you can use the AI Platform Prediction service.

• Explore the TPU tools in TensorBoard.