Run

ia run starts a workload on insaali. Each workload type is its own subcommand. The first is simeval.

simeval

Runs one episode of a policy in a simulator.

ia run simeval --policy hf://<owner>/<repo> [--sim HalfCheetah-v5] [--max-steps 100]

flag	required	default	description
--policy	yes	—	HuggingFace reference hf://<owner>/<repo>
--sim	no	HalfCheetah-v5	Gymnasium env id
--max-steps	no	100	Maximum environment steps per episode

The command prints the new run id and its initial status, then returns. The run continues on the compute backend.

$ ia run simeval --policy hf://acme/policy
3f0b7c1e-4a8e-4d6e-9f2b-1b3c0e7f1d2a
status: running

Following a run

ia status <run-id> and ia logs <run-id> fetch the live state from the compute backend on each call — nothing is mirrored.

$ ia status 3f0b7c1e-4a8e-4d6e-9f2b-1b3c0e7f1d2a
id:        3f0b7c1e-4a8e-4d6e-9f2b-1b3c0e7f1d2a
simulator: HalfCheetah-v5
policy:    hf://acme/policy
status:    succeeded

$ ia logs 3f0b7c1e-4a8e-4d6e-9f2b-1b3c0e7f1d2a
[insaali] run_id=3f0b7c1e simulator=HalfCheetah-v5 policy_ref=hf://acme/policy
[insaali] step=1/100
[insaali] step=2/100
...
[insaali] done run_id=3f0b7c1e

Lifecycle

status	meaning
running	The job is provisioning or executing on the compute backend.
succeeded	The run finished without error.
failed	The run errored. ia status shows the message; ia logs has more.
unknown	The backend could not be reached for a status call. Retry.

Status is whatever the compute backend reports at the moment of the call, lowercased. Once a run is succeeded or failed, it stays that way; backends retain logs for a limited window.