Run and submit¶
Start here after the included-player quickstart works. These procedures take a model from local configuration through experiment execution, result inspection, and submission.
Complete the quickstart first
The procedures assume that gptnt doctor can reach Redis and KTANE and that the
quickstart has produced a queryable result.
Configure the inputs¶
| Goal | Procedure | Result |
|---|---|---|
| Assemble a model-backed player | Add a model | A player profile with identity, capabilities, model settings, and image-token calibration |
| Connect the player to an endpoint | Configure a provider | Credentials and any provider override supplied outside the player profile |
| Prepare the manual selected by a suite | Prepare manuals | Validated, content-addressed manual artefacts |
| Select suites, players, and runtime capacity | Create a run manifest | A schema-v2 manifest and generated experiment specifications |
Execute and submit¶
- Run your model to execute the generated interactive specifications.
- Run static evaluations for the required no-game tasks.
- Inspect and analyse results to check completion and build DuckDB.
- Submit your results after the interactive and static outputs are complete.
Use roles, protocols, and capabilities when a suite's participation rules affect a player choice. Use the reference for exact commands, fields, formats, and supported Python interfaces.