eval entrypoint:
.toml files for configuration:
uv run eval @ config.toml --dry-run. To run the evaluation, use uv run eval @ config.toml.
Use dotted arguments to set values using the CLI, e.g. --sampling.temperature 0.5. CLI arguments overwrite toml arguments when both are present.
The output from evaluations are written into outputs/<taskset>--<model>--<harness>/<uuid>/ by default (use output_dir to overwrite the folder). The folder contains the used config.toml, all the traces in results.jsonl, as well as logs of the run and workers in eval.log.
Common config values
model— the model id to evaluate, e.g.nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55Bsampling— generation params passed to the model, e.g.sampling.temperaturetaskset.id/harness.id— pick the taskset and harnessnum_tasks— how many tasks to evaluate. Not setting a value means all tasksnum_rollouts— rollouts per taskmax_concurrent— caps how many rollouts are in flight at onceverbose— log at debug instead of infoshuffle— randomizes the order of tasks (fixed seed)
Resuming evaluations
--resume <output-dir> re-runs only the rollouts a previous run left missing or errored, appending to that run’s own results.jsonl. It reloads the run’s saved config.toml verbatim, so it takes no other arguments. Good rollouts are kept, while errored ones are dropped and redone.
Disabling tools
Almost every harness comes with adisabled_tools list, which can be used to disable one or multiple tools: