Skip to content

🧪 pxl-pipeline test

The test command runs a pipeline locally in a virtual environment, using the same logic that will later run on Picsellia. It uses the unified entrypoint pipeline.py with --mode local.

Usage

pxl-pipeline test PIPELINE_NAME [OPTIONS]

Example:

pxl-pipeline test yolov8 --run-config-file runs/run_config.toml

This runs: 

python pipeline.py --mode local --config-file runs/run_config.toml

Options

Option Description Default
--run-config-file Path to a TOML file containing pipeline inputs, outputs, and parameters. If omitted, you’ll be prompted for required values, and a run_config.toml will be generated automatically. None
--reuse-dir Reuse the last run directory instead of creating a new one. This avoids re-downloading datasets and models. False
--organization Picsellia organization name (will be inferred from environment if missing). Prompted
--env Picsellia environment: PROD, STAGING, LOCAL. PROD

Run directories and configs

At initialization

When you create a pipeline with pxl-pipeline init, a template run_config.toml is generated under runs/. This file acts as a starting point you can edit to configure inputs, outputs, and parameters.

At each test run

Each time you run pxl-pipeline test:

  • A new run directory is created under runs/runX/ (unless --reuse-dir is used).
  • A fresh run_config.toml is copied or generated in that folder.
  • Required values are auto-filled or prompted if missing.
  • Datasets and model versions are downloaded inside the run folder.
  • Outputs generated by the pipeline are written there as well.
  • The run_config.toml is updated at launch and completion with extra metadata (IDs, URLs to datasets, outputs, etc.).

Example structure:

my_pipeline/
├── runs/
│   ├── run1/
│   │   ├── run_config.toml
│   │   ├── dataset/        # downloaded dataset files
│   │   ├── model/          # downloaded model weights
│   │   └── outputs/        # generated results
│   ├── run2/
│   │   └── run_config.toml
│   └── run3/
│       └── run_config.toml

💡 This makes each run self-contained and reproducible.