Decide which training experiment to run next for a project and launch it. Use when the user floats a training idea ("is a smaller model as good as a larger one on this benchmark?", "should we try more SFT data?", "what should we run next?") or asks to design/launch/continue experiments. Reads the project goal + past experiments via the evsys-sdk SDK, reasons over the evidence, then crafts and runs the next experiment through the project's own train/benchmark skills.
Scaffold or migrate a repo to the evsys-sdk research-project layout (data/, src/, experiments/, .evsys/). Use when starting a new project that will use the SDK, or when the user has an existing ad-hoc evsys-sdk project (loose `training/` scripts, scattered data files) they want to bring into the standard shape.
Scaffold or migrate a repo to the evsys-sdk research-project layout (data/, src/, experiments/, .evsys/). Use when starting a new project that will use the SDK, or when the user has an existing ad-hoc evsys-sdk project (loose `training/` scripts, scattered data files) they want to bring into the standard shape.
How to push training runs (SFT/RL/distillation) to the EvolvingSystems dashboard. Use when writing experiment scripts that should appear in the dashboard, or when wiring metrics/predictions/conclusions into an existing training loop.
How to pull history of past experiments, generations, eval results, predictions, and rendered training data from the EvolvingSystems dashboard backend — so you can build on prior runs instead of starting cold. Use before proposing a new hypothesis or recipe.
How to use the evsys-sdk SDK to read project goals + experiment history and to create/launch experiments: EvsysStore (backend-routed data access) and Workspace (local dataset cache). Use when writing code that reads or writes experiments/runs/datasets/benchmarks/metrics, or materializes data for training.