TensorBoard Data Server

0.7.2 · active · verified Sun Mar 29

The `tensorboard-data-server` library, currently at version 0.7.2, provides fast data loading capabilities specifically designed for TensorBoard. It's a key component within the larger TensorBoard ecosystem, optimizing the ingestion and serving of experiment data for visualization. It is released independently but closely aligned with TensorBoard's development, which has a more frequent release cadence.

Warnings

gotcha The `tensorboard-data-server` version (e.g., 0.7.2) is significantly lower and has an independent release cycle compared to the main `tensorboard` package (e.g., 2.20.0). Do not assume their versions are directly correlated or that features align strictly by version number.
Fix: Always refer to the `tensorboard` package version for overall compatibility and feature sets. The data server's version reflects its internal development.
gotcha The PyPI classification for `tensorboard-data-server` lists its Development Status as '2 - Pre-Alpha'. While it's a dependency of the stable `tensorboard` package, direct programmatic interaction with `tensorboard-data-server` may expose unstable APIs or behaviors.
Fix: Avoid direct imports or usage of `tensorboard-data-server` unless specifically instructed by TensorBoard documentation or for development purposes. Rely on `tensorboard` to manage its interaction.
gotcha Users on systems with older GLIBC versions (e.g., CentOS 7 with GLIBC_2.17) might encounter 'GLIBC_2.18' not found errors when trying to use the native data server binary. This can cause TensorBoard to fall back to a slower, Python-based log ingester.
Fix: Ensure your system's GLIBC version meets the requirements of the `tensorboard-data-server` wheel. If upgrading GLIBC is not an option, be aware that performance might be degraded without the native server, as TensorBoard will use a slower Python fallback.
deprecated The TensorBoard.dev hosted service has been shut down. Commands like `tensorboard dev upload` will fail. This is a breaking change for the broader TensorBoard ecosystem that relies on the data server.
Fix: Use TensorBoard locally or with self-hosted solutions. The `tensorboard dev upload` command is no longer functional.

Install

pip install tensorboard-data-server Install stable version
pip install tensorboard # This will also install tensorboard-data-server as a dependency Install via TensorBoard

Imports

tensorboard_data_server
```
import tensorboard_data_server
```
While the package can be imported, its primary use is as a backend component for TensorBoard, rather than direct user-level programmatic interaction. TensorBoard leverages `tensorboard_data_server.server_binary()` internally to locate the server executable.

Quickstart

The `tensorboard-data-server` is typically leveraged automatically by `tensorboard`. This example shows how to generate logs and then start TensorBoard, explicitly enabling the fast data server. The `--load_fast=true` flag instructs TensorBoard to use the Rust-based data server for improved performance.

import tensorflow as tf
import datetime

# Create a log directory
log_dir = "logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
writer = tf.summary.create_file_writer(log_dir)

# Log a scalar value
with writer.as_default():
    for i in range(100):
        tf.summary.scalar("my_metric", i * 0.1, step=i)
    writer.flush()

print(f"TensorBoard logs saved to: {log_dir}")
print("To view, run in your terminal: ")
print(f"tensorboard --logdir {log_dir} --load_fast=true")

view raw JSON →