Issue: TensorBoard wheels on PyPI fail with setuptools >81 due to deprecated pkg_resources removal

Summary

The latest released TensorBoard packages on PyPI currently fail to run with newer versions of setuptools (>81), because they still depend on pkg_resources, which has been removed/deprecated in recent setuptools releases.

This issue appears to already be fixed in PR #7057, but no new wheel/package containing the fix has been released to PyPI yet.

As a result, fresh environments using modern setuptools versions cannot run TensorBoard from the latest published release.

---

Affected Packages

• tensorboard==2.20.0

  • Released: Jul 18, 2025

• tb-nightly==2.21.0a20251023

  • Released: Oct 23, 2025

---

Problem

Running TensorBoard in an environment with newer setuptools versions results in import/runtime failures because pkg_resources is no longer available.

Example environment:

pip install -U setuptools
pip install tensorboard
tensorboard

Observed failure:

ModuleNotFoundError: No module named 'pkg_resources'

---

Root Cause

TensorBoard still imports/depends on pkg_resources, which has been deprecated for a long time and removed in newer setuptools releases (>81).

The fix already exists in:

• PR #7057

However, no released wheel currently includes this fix.

---

Request

Please publish a new release (or at minimum a new wheel) containing the changes from PR #7057 so TensorBoard works correctly with modern setuptools versions.

This currently breaks clean installs in up-to-date Python environments and affects downstream tooling relying on TensorBoard.

---

Additional Notes

A temporary workaround is pinning setuptools to an older version:

pip install "setuptools<81"

But this is only a workaround and not ideal for reproducible modern environments.

Start by reviewing PR #7057 and compare its changes with the published tensorboard 2.20.0 and tb-nightly 2.21.0a20251023 packages. Verify the setuptools compatibility issue using the provided pip installation steps; done means a new wheel or release containing the fix is available on PyPI.