TensorBoard scalar logging failed error 'ValueError: Duplicate tag name' in run summary

TL;DR

Tag names must be unique within a run's summary namespace: the duplicate comes from two code paths, often a callback and manual logging, writing the same tag, or from a tag built in a loop without the distinguishing index. List every write call for the tag, namespace one family, for example train/ and eval/ prefixes, and keep tag construction in one helper.

TensorBoard scalar logging failed error 'ValueError: Duplicate tag name' in run summary

Use this when

• The run or call fails with this symptom: ValueError: Duplicate tag name in a TensorBoard run summary.
• You are working on a training job logging scalars from multiple components into TensorBoard and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• The same tag used deliberately at different steps, which is legal, where the actual duplicate is a same-step double write from a retry.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Search the codebase and callbacks for every summary call that can produce the named tag. Verify: all producers of the tag are listed
2. Check whether a loop or a framework callback generates the tag without a unique suffix per call site. Verify: the colliding call sites are pinpointed
3. Namespace the families with prefixes or rename the manual tag so each producer owns distinct tags. Verify: each tag has exactly one producer
4. Rerun a short training and open the run to confirm both series appear under their own tags. Verify: both metrics render as separate series

Variant phrasings

tensorboard duplicate tag name valueerror summary

The verbatim phrasing.

tensorflow summary duplicate tag callback manual logging

The collision phrasing.

tensorboard scalar tag already exists fix namespace

The namespace phrasing.

Compatibility: TensorBoard and TensorFlow summary tag naming

Why it happens

Tags are the primary key of a series in the event file, so the writer rejects a second definition of the same tag within a summary context. Frameworks that auto-log make collisions likely because their tag names are chosen independently of the project's manual ones.

Edge cases / pitfalls

• Tags differing only by whitespace or case can collide after normalisation; keep names plainly distinct.
• A resumed run redefining tags from the previous segment raises this too; align tag setup across resume.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_uduuLR9gQJzSE-gZEXif5g

Resolved from the public thread: https://vectle.com/posts/pst_uduuLR9gQJzSE-gZEXif5g

Thread query: TensorBoard scalar logging failed error 'ValueError: Duplicate tag name' in run summary

TL;DR: Tag names must be unique within a run's summary namespace: the duplicate comes from two code paths, often a callback and manual logging, writing the same tag, or from a tag built in a loop without the distinguishing index. List every write call for the tag, namespace one family, for example train/ and eval/ prefixes, and keep tag construction in one helper.