picsellia-pipelines-cli 0.2.0

No project description provided

🧪 Picsellia Pipelines CLI

A command-line tool to create, test, deploy, and manage training and processing pipelines for Picsellia.

Built with Typer for intuitive CLI usage.

---

Installation

Install the CLI directly from GitHub (no need to clone the repo):

Using Poetry:

poetry add picsellia-pipelines-cli

Using uv (faster, works with requirements.txt or pyproject.toml):

uv pip install picsellia-pipelines-cli

You can now use:

pxl-pipeline --help

Available Commands

🔹 All CLI commands are structured like this:

pxl-pipeline [init|test|deploy|smoke-test|sync] <pipeline_name>

The pipeline type is resolved automatically from its config file (config.toml), except during init.

🔧 Initialize a Pipeline

pxl-pipeline init <pipeline_name> --type [training|processing] --template <template>

Examples:

pxl-pipeline init yolov8 --type training --template ultralytics
pxl-pipeline init resize-images --type processing --template dataset_version_creation

This generates:

• a config.toml with pipeline metadata

• a Dockerfile, .dockerignore, and dependency file (pyproject.toml or requirements.txt)

• scripts for both picsellia_pipeline.py and local_pipeline.py

• pre-filled steps.py and utility files

You can customize templates in your own extensions later.

After Initialization

Once the pipeline is initialized:

• A local virtual environment is automatically created inside the pipeline directory (<pipeline_name>/.venv)
• You can directly activate it and run any pipeline-related commands:

cd <pipeline_name>
source .venv/bin/activate  # or .venv\Scripts\activate.bat on Windows

You can still use the CLI from inside the venv:

pxl-pipeline test <pipeline_name>

🧪 Test Locally

pxl-pipeline test <pipeline_name>

Runs the pipeline in a local virtualenv (.venv/) and prompts for required parameters (e.g., dataset version ID, experiment ID, etc.).

🔥 Smoke Test in Docker

pxl-pipeline smoke-test <pipeline_name>

Builds the Docker image for the pipeline and runs it locally to validate that everything (code + dependencies + env) works inside the container.

🚀 Deploy to Picsellia

pxl-pipeline deploy <pipeline_name>

Builds and pushes the Docker image to your configured registry and registers the pipeline in Picsellia (either as a training pipeline or a dataset processing job).

🔁 Sync Parameters (Processing Only)

pxl-pipeline sync <pipeline_name>

For processing pipelines, this updates the default parameters stored in Picsellia based on your Parameters class and config.toml.

Training sync is not yet implemented.

📁 Project Structure Example

resize-images/
├── config.toml
├── Dockerfile
├── picsellia_pipeline.py
├── local_pipeline.py
├── steps.py
├── utils/
│   ├── parameters.py
│   └── processing.py
└── pyproject.toml

💡 Tips

• You can override the output directory on init with --output-dir
• Virtual environments are created in <pipeline_name>/.venv by default
• You can always edit config.toml to change pipeline metadata or execution scripts

---

Made with ❤️ by the Picsellia team.