TensorBoard

Sync TensorBoard logs to W&B for cloud-hosted visualization, sharing, and centralized analysis alongside system metrics.

W&B supports embedded TensorBoard for W&B Multi-tenant Cloud.

This page shows how to sync TensorBoard logs to W&B so you can upload your TensorBoard logs to the cloud, share your results among colleagues and classmates, and keep your analysis in one centralized location. This integration is for users who already log to TensorBoard and want cloud-hosted visualization, sharing, and side-by-side comparison with W&B system metrics.

Get started

To enable TensorBoard syncing, set sync_tensorboard=True when you initialize a W&B run. W&B automatically uploads any TensorBoard events your training code emits.

import wandb

# Start a wandb run with `sync_tensorboard=True`
wandb.init(project="my-project", sync_tensorboard=True) as run:
  # Your training code using TensorBoard
  ...

Review an example TensorBoard integration run.

After your run finishes, you can access your TensorBoard event files in W&B and visualize your metrics in native W&B charts. W&B also captures additional information such as system CPU or GPU utilization, the git state, and the terminal command the run used.

W&B supports TensorBoard with all versions of TensorFlow. W&B also supports TensorBoard 1.14 and later with PyTorch as well as TensorBoardX.

Frequently asked questions

The following sections answer common questions about customizing the TensorBoard integration, including logging extra metrics, configuring the patch, syncing historical runs, and using notebook environments.

How can I log metrics to W&B that aren’t logged to TensorBoard?

If you need to log additional custom metrics that aren’t logged to TensorBoard, you can call wandb.Run.log() in your code: run.log({"custom": 0.8}).

Setting the step argument in run.log() is turned off when syncing TensorBoard. If you’d like to set a different step count, you can log the metrics with a step metric as:

run.log({"custom": 0.8, "global_step": global_step})

How do I configure TensorBoard when I’m using it with wandb?

If you want more control over how W&B patches TensorBoard, call wandb.tensorboard.patch() instead of passing sync_tensorboard=True to wandb.init().

import wandb

wandb.tensorboard.patch(root_logdir="<logging_directory>")
run = wandb.init()

# Finish the wandb run to upload the tensorboard logs to W&B (if running in Notebook)
run.finish()

To patch vanilla TensorBoard, pass tensorboard_x=False to this method. If you’re using TensorBoard later than 1.14 with PyTorch, pass pytorch=True to patch it. Both of these options have sensible defaults depending on what versions of these libraries you’ve imported.

By default, W&B also syncs the tfevents files and any .pbtxt files. This lets W&B launch a TensorBoard instance on your behalf. You see a TensorBoard tab on the run page. To turn off this behavior, pass save=False to wandb.tensorboard.patch.

import wandb

run = wandb.init()
wandb.tensorboard.patch(save=False, tensorboard_x=True)

# If running in a notebook, finish the wandb run to upload the tensorboard logs to W&B
run.finish()

You must call either wandb.init() or wandb.tensorboard.patch() before calling tf.summary.create_file_writer() or constructing a SummaryWriter via torch.utils.tensorboard.

How do I sync historical TensorBoard runs?

If you have existing tfevents files stored locally and you would like to import them into W&B, you can run wandb sync log_dir, where log_dir is a local directory containing the tfevents files.

How do I use Google Colab or Jupyter with TensorBoard?

If you run your code in a Jupyter or Colab notebook, make sure to call wandb.Run.finish() at the end of your training. This finishes the wandb run and uploads the TensorBoard logs to W&B so they can be visualized. This isn’t necessary when you run a .py script, because wandb finishes automatically when a script finishes.

To run shell commands in a notebook environment, you must prepend a !, as in !wandb sync directoryname.

How do I use PyTorch with TensorBoard?

If you use PyTorch’s TensorBoard integration, you may need to manually upload the PyTorch Profiler JSON file.

with wandb.init(project="my-project", sync_tensorboard=True) as run:
    run.save(glob.glob(f"runs/*.pt.trace.json")[0], base_path=f"runs")

Can I sync tfevents files stored in the cloud?

wandb 0.20.0 and later supports syncing tfevents files stored in S3, GCS, or Azure. wandb uses the default credentials for each cloud provider. The following table lists the command to configure credentials and the expected logging directory format for each provider:

Cloud provider	Credentials	Logging directory format
S3	aws configure	s3://bucket/path/to/logs
GCS	gcloud auth application-default login	gs://bucket/path/to/logs
Azure	az login1	az://account/container/path/to/logs

Footnotes

1. You must also set the AZURE_STORAGE_ACCOUNT and AZURE_STORAGE_KEY environment variables. ↩