ado

What ado is

ado is a free, open-source Python platform for planning computational experiment campaigns and running them at scale. Its workflow guides researchers through selecting experiments, defining parameter spaces, choosing a strategy, executing runs, and analysing results. Teams of researchers and engineers can collaborate across locations, run experiments, and share data. Python functions can extend experiments and operators through ado’s plugin model. The sample store keeps measurement results and can reuse existing results through memoization.

The documentation describes validated schemas, dry runs, structured results, and provenance records that track resource relationships and plugin versions. It also covers remote execution on Ray clusters, deployment of backends with Kubernetes or OpenShift, and a distributed MySQL setup for project data. Example applications include benchmarking fine-tuning and inference performance, and creating predictive performance models. Bundled agent skills can help AI coding assistants build research campaigns, execute experiments, query data, and analyse findings. The user guide is intended for researchers and benchmarkers exploring parameter spaces and results. The project welcomes contributions and directs feature proposals and bug reports to its GitHub issue tracker.

Who it suits

ado suits researchers and benchmarkers who need to explore parameter spaces, run computational experiments, and analyse results. Distributed research or engineering teams can use it to collaborate and share experiment data.

Getting started

1. Open the ado documentation at https://ibm.github.io/ado/.
2. Use the documented workflow to choose experiments and define a parameter space.
3. Configure a strategy, then execute experiments and analyse results.
4. Extend experiments or operators with Python functions if needed.

Verdict

Choose ado for research and benchmarking campaigns that need a documented workflow, reusable results, and extensibility through Python. Its listed platforms are API and self-hosted, so it suits teams prepared to work with those deployment options.