Get and verify checkpoints

Every training job makes one policy, and the policy holds that job’s checkpoints. You find checkpoints by policy, then download the one you want.

1. List your policies

simulo policy list

The table has one row per policy, most recent first:

Column	What it shows
POLICY	The policy ID, such as policy_brisk-heron-4f8k2m.
STATUS	The status of the training job that made it.
CHECKPOINTS	The checkpoints it holds: best, latest, or both.
BEST REWARD	The reward recorded for the best checkpoint.
TAGS	The training job’s tags. Shown when any listed policy has one.
UPDATED	When the training job ended.

Filter and widen the list with these options:

• --org lists your organization’s policies, not only the ones you trained.
• --tag NAME keeps policies whose training job has that tag. Repeat it to require several tags.
• --limit N shows up to N policies (default 20, at most 250).
• --json prints the full records, including each checkpoint’s iteration, reward and SHA-256 digest.

A policy’s checkpoints appear when its training job ends. To see one policy on its own, with the task and robot it was trained on and every checkpoint’s iteration and reward, inspect it by ID:

simulo policy inspect policy_brisk-heron-4f8k2m

See Policies and checkpoints for the full output and what --json adds. You can also browse policies on the Policies page of Simulo Console; open one to see its checkpoints, the jobs that used it, and cards to play or export it.

2. Download a checkpoint

Name the policy and the checkpoint kind:

simulo policy get policy_brisk-heron-4f8k2m:best

Saved /home/you/cartpole_demo/policy_brisk-heron-4f8k2m-best.pt (sha256 verified)

Without -o, the file is saved in the current folder as <policy>-<kind>.pt, and the command refuses to overwrite a file already there. -o PATH writes to a path you choose and replaces an existing file:

simulo policy get policy_brisk-heron-4f8k2m:best -o ./best.pt

The kind is required. A job ID gets a hint, for example job_brisk-heron-4f8k2m is a job; did you mean policy_brisk-heron-4f8k2m:best?.

How the digest check works

The platform records a SHA-256 digest for every checkpoint. simulo policy get downloads the bytes, compares their digest with the recorded one, and writes the file only when they match. If they differ, the command fails and writes nothing, so you never keep a damaged or altered checkpoint.

To check a file yourself, compare its digest with the one the platform recorded:

sha256sum best.pt
simulo policy list --json

The second command shows the recorded value in each checkpoint’s digest_sha256 field. The two hex strings match, ignoring any sha256: prefix.

What a checkpoint is for

A checkpoint is training state for Simulo. Use it to continue training with --from, or play it back with --policy to watch it or record it. To run the policy outside Simulo, export it to an ONNX bundle:

simulo export policy_brisk-heron-4f8k2m:best