Skip to content
Simulo

simulo run

Submit the file’s one typed job to Simulo, or write a local package when no cloud target is selected.

usage: simulo run [--detach] [--viewstream] [--frozen] [--strict-assets] [--from POLICY:KIND]
                  [--policy POLICY:KIND] [--compare POLICY:KIND] [--require-success RATE]
                  [--fail-if-worse] [--skip-preflight] [--tag NAME]
                  [app_file]

Arguments

Section titled “Arguments”
Argument Description
APP_FILE (optional) Path to the Python file declaring the simulo.App.

Options

Section titled “Options”
Option Description
--detach Submit a cloud job without waiting for its logs to finish (prints attach hints instead).
--viewstream Enable the live 3D viewport stream for this cloud job (simulo view watches it once minted; no effect on a local-disk submit, which never reaches a worker). Training runs noticeably slower while streaming (roughly 4-5x, measured on a 512-environment benchmark) even with the platform’s built-in low-overhead render settings applied.
--frozen Fail the submit if any mounted registry asset is unpinned (has no :vN), instead of resolving it to the current latest. Opt CI and release pipelines into full version explicitness — a cloud submit only (a local-disk submit resolves no assets).
--strict-assets Treat asset warnings as errors: a mounted asset that is deprecated, or has never validated on the job’s runtime, fails the submit instead of printing a warning (cloud submit only).
--from POLICY:KIND Start a training run from an explicit policy checkpoint such as policy_x:best or policy_x:latest. The default is always a fresh run. This is available only for @app.job(type=“train”) and requires a cloud control plane (simulo login).
--policy POLICY:KIND Play or evaluate an explicit policy checkpoint such as policy_x:best or policy_x:latest. This is available only for @app.job(type=“play”) and @app.job(type=“eval”), which need it, and requires a cloud control plane (simulo login).
--compare POLICY:KIND Compare an eval policy with policy_x:best or policy_x:latest on the same episodes (eval only).
--require-success RATE Exit 1 when the share of successful episodes is below RATE (0 to 1, such as 0.9 for 90%%). Available only on an eval file, and needs simulo run to wait for the result, so it is refused with –detach. Exit 0 means it passed; exit 2 means the evaluation did not finish.
--fail-if-worse Exit 1 when –policy did worse than –compare. Available only with –compare on an eval file, and refused with –detach.
--skip-preflight Bypass the pre-submit platform check entirely (a cloud submit only; a local-disk submit never calls it) — no request is made, not just ‘ignore the result’. Use this if simulo prepare/simulo run’s preflight check is wrong about your app, or the platform is unreachable and you still want to try the submit.
--tag NAME Tag the submitted job; repeat for multiple tags.