Skip to content

Key repositories

composable-logs/mnist-digits-demo-pipeline

  • Repo for the mnist demo training pipeline.
  • A public task and experiment tracker for this pipeline is hosted on Github Pages:

screenshot-task-list.png

https://composable-logs.github.io/mnist-digits-demo-pipeline/

composable-logs/composable-logs (Python)

  • Repo for the main library (Python, Ray, OpenTelemetry). This is used when running the demo pipeline.
  • Dependency for the above demo pipeline.
  • Install via PyPI:
    • pip install composable-logs (latest release, PyPI link)
    • pip install composable-logs-snapshot (dev-release, latest commit to main branch, PyPI link)

composable-logs/mlflow (Javascript)

  • A slightly modified version of MLFlow, that can build static MLFlow-like sites where all metadata is included in the front end.
  • Dependency for the above demo pipeline.
  • The static assets are made available as a Python package via PyPI:
    • Install as pip install composable-logs-webui --target <install dir> (latest release, PyPI link). After installation the assets will be in the assets directory.

Development environments are dockerized, and makefile:s should be available for most common tasks.

Inspecting the Github actions in the composable-logs and mnist-digits-demo-pipeline repos can also be useful since they show how various tests (unit, static code analysis) and pipelines are run in docker.

Instructions assume linux-based setup (eg. Ubuntu), but should also work on macs with some changes (or at least on non-M1 ones).