TensorBoard Not Supported During Training in New Version 2.1.0

Description

In the latest version, TensorBoard is not displaying during the training process. Specifically:

1. In the Projector tab, the message "No checkpoint was found" is displayed.
2. On the Home page, the message "No dashboards are active for the current data set" appears.

Actual Behavior

1. In the Projector tab, "No checkpoint was found" is displayed.
2. On the Home page, "No dashboards are active for the current data set."

Expected Behavior

TensorBoard should display the training metrics, and checkpoints should be found in the Projector tab. Dashboards should be active and show real-time updates on the Home page.

Steps to Reproduce

1. Train according to the provided example code and generate a local folder './full_embeddings_vis'
2. Attempt to view training metrics in TensorBoard.

Environment

TensorFlow version: 2.9.0
Python version: 3.10.14
OS: Windows11

Additional Information

Even after adding tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir="./full_embeddings_vis", histogram_freq=1) to the callbacks parameter in model.fit(), TensorBoard still does not display properly. Is this feature no longer supported in the new version?

Run the referenced training example with TensorFlow 2.9.0 on Windows 11 and generate ./full_embeddings_vis. Check the TensorBoard Projector and Home pages during training, including with the stated TensorBoard callback. Done means checkpoints are found and training metrics appear in active dashboards with real-time updates.