The challenge
A shared experiment needs more than a call. Participants need to understand the model, the task they are running, and how individual work contributes to the experiment.
A possible workflow
Define the experiment
The host application defines the model, data boundaries, task format, and how results will be checked.
Join a shared session
Participants connect in the application. The workflow checks available execution capabilities before assigning work.
Run and coordinate tasks
Use the distributed-training tooling and a task-specific connector to coordinate supported work and surface progress.
Review results
Validate outputs, inspect failures, and decide how accepted results update the experiment.
Building blocks
Session layer: participant coordination and communication. Compute tooling: supported distributed-training or parallel-computing widgets. Your application: model definition, task protocol, validation, persistence, and experiment policy.
Plan data boundaries first
Define exactly which inputs and results move between peers. Do not assume a peer session automatically supplies federated-learning privacy guarantees, trusted execution, or server-grade GPU capacity.
Implementation
Start with a small, verifiable task and a controlled set of participants. Follow Get Started, then plan your site widgets and required widgets.