DVC "experiment ... failed" (dvc exp) in CI

dvc exp run reproduces the pipeline inside an experiment workspace and it failed. The cause is usually the same as a repro failure, a stage command error, or a workspace that is not clean enough for DVC to create the experiment.

What this error means

dvc exp run stops with "ERROR: Failed to reproduce experiment" after a stage error or a message about an unexpected workspace state.

ERROR: Failed to reproduce experiment 'quick-lynx':
failed to reproduce 'train': failed to run: python train.py, exited with 1

Common causes

A pipeline stage failed inside the experiment

The experiment reruns the DAG; a stage command that errors fails the whole experiment.

Missing data or a dirty workspace

Inputs were not pulled, or uncommitted changes block DVC from creating a clean experiment commit.

How to fix it

Fix the failing stage, then re-run

1. Read the stage error inside the experiment output.
2. Pull inputs and install deps so the stage can run.
3. Re-run dvc exp run.

pip install -r requirements.txt
dvc pull
dvc exp run

Run experiments on a clean checkout

Ensure the CI checkout is clean so DVC can snapshot the experiment.

git status --porcelain
dvc exp run

How to prevent it

• Pull data and install deps before dvc exp run.
• Keep the CI workspace clean so experiments can be recorded.
• Reproduce the pipeline with dvc repro before running experiments.

Related guides