TensorBoard.dev TensorBoard TensorFlow MY EXPERIMENTS


TensorBoard.dev is shutting down. Please export your experiments by December 31, 2023.
Thank you for being part of the TensorBoard.dev journey.


Do I have any affected experiments?

Affected users will be receiving an email with instructions. If you created any TensorBoard.dev experiments, you can also view them by signing in at tensorboard.dev/experiments/.

What is the timeline for shutting down?

Effective immediately, TensorBoard.dev is no longer accepting new users.
  • Nov 1, 2023: Existing users can no longer upload new experiments.
  • Dec 1, 2023: Experiments can no longer be accessed via the TensorBoard.dev website. Links to existing experiments will redirect to this page.
  • Dec 31, 2023: Experiments can no longer be exported, and will be permanently deleted.

What can I use instead of TensorBoard.dev?

You can continue to use TensorBoard as a local tool via the open source project, which is unaffected by this shutdown, with the exception of the removal of the `tensorboard dev` subcommand in our command line tool. For a refresher, please see our documentation.

For sharing TensorBoard results, we recommend the TensorBoard integration with Google Colab. TensorBoard log files that are generated or downloaded by code in the Colab notebook can be directly visualized by running TensorBoard within the notebook.

How do I get a copy of my data?

If you do not have a copy of your original data that was uploaded to TensorBoard.dev, you must export your experiments by December 31, 2023 to avoid permanent deletion.

To export, run the command `tensorboard dev export --outdir OUTPUT_PATH`. This command will export all your TensorBoard.dev experiments to a local directory of your choice.

What format is the data export produced by `tensorboard dev export`?

The export command produces one subfolder per experiment. Each subfolder contains several newline-delimited JSON files:

  • metadata.json: experiment-level metadata such as title and description
  • scalars.json: full data points for scalars (shown on the Scalars dashboard)
  • tensors.json: data point metadata for tensor-type data (shown on the Histograms, Distributions, Text, and Hparams dashboard)
  • blob_sequences.json: data point metadata for blob-type data (shown on the Graphs dashboard)

The data point values for tensors and blob sequences are written separately as individual files. Tensor values are written to a "tensors" subfolder as Numpy zip files (.npz), and blob values are written to a "blobs" subfolder as binary files (.bin). For the Graphs plugin, the binary file format will be a serialized GraphDef protocol buffer.

How do I delete an experiment prior to the shutdown?

Run the command `tensorboard dev delete --experiment_id EXPERIMENT_ID`.

The experiment ID is the trailing part of the experiment URL, without any slashes, for example "QFRIzZJpTZCNRzi8N7zomA".

What if I encounter an error running the `tensorboard dev` command?

Stale credentials can lead to an "UNAVAILABLE" error when running the `tensorboard dev` command. For example, the error message might include "invalid_grant: Bad Request" or similar details.

This can typically be addressed by refreshing your credentials. To do this, run the command `tensorboard dev auth revoke` and then retry the original command, which will prompt you to sign in again. Please ensure you sign in using the same account that you previously used with TensorBoard.dev.

What if I have a problem not addressed here?

For problems that cannot be addressed by this FAQ, please reach out to tensorboard.dev-support@google.com.