Skip to content

Latest commit

 

History

History
96 lines (68 loc) · 3.66 KB

README.md

File metadata and controls

96 lines (68 loc) · 3.66 KB

PyPI version

Machine Learning Exchange catalog connector

This catalog connector enables Elyra to load Kubeflow Pipelines components from Machine Learning Exchange (MLX) deployments.

Install the connector

You can install the MLX catalog connector from PyPI or source code. Note that a rebuild of JupyterLab is not required.

Prerequisites

Install from PyPI

To install the connector from PyPI:

$ pip install mlx-component-catalog-connector

Install from source code

To install the connector from source:

$ git clone https://github.com/elyra-ai/examples.git
$ cd examples/component-catalog-connectors/mlx-connector/
$ make source-install

Run connector unit tests

To run the unit tests:

$ git clone https://github.com/elyra-ai/examples.git
$ cd examples/component-catalog-connectors/mlx-connector/
$ make test

Note: The tests don't require access to a running MLX instance.

Use the connector

  1. Launch JupyterLab.
  2. Open the 'Manage Components' panel.
  3. Add a new MLX component catalog ('+' > 'New Machine Learning Exchange Component Catalog').
  4. Specify a catalog name, e.g. 'MLX dev catalog'.
  5. (Optional) Specify a category under which the catalog's components will be organized in the palette.
  6. Configure the MLX API URL, e.g. 'http://my-mlx-server.mydomain:8080/'.

    Note: The Machine Learning Exchange API URL is different from the GUI URL!

  7. Apply an optional filter expression to the component names. The * (zero or more characters) and ? (zero or one character) wildcards are supported.
  8. Save the catalog entry.
  9. Open the Kubeflow Pipelines Visual Pipeline Editor and expand the palette. The catalog components are displayed.

Uninstall the connector

  1. Remove all MLX catalog entries from the 'Manage Components' panel.
  2. Stop JupyterLab.
  3. Uninstall the mlx-component-catalog-connector package. 
    $ pip uninstall -y mlx-component-catalog-connector

Troubleshooting

Problem: The palette does not display any components from the configured catalog.

Solution: If the the Elyra GUI does not display any error message indicating that a problem was encountered, inspect the JupyterLab log file.

Example error message (The specified MLX URL is invalid):

Error fetching component list from MLX catalog http://localhost:8080: ... Failed to establish a new connection: [Errno 61] Connection refused'