Why a run produces several
Training here is step-based and the trainer writes weights at intervals along the schedule. That is insurance against a run that ends badly, and it is the only way to compare the model at different points in its own training. A checkpoint you delete cannot be regenerated without paying for the run again.
| Model | Default steps | How checkpoint spacing is controlled |
|---|---|---|
| GR00T N1.7 | 20,000 | saveSteps, the one extra field GR00T exposes |
| GR00T N1.5 | 2,000 | saveSteps |
| Pi0.5 | 30,000 | No saveSteps field. You take what the trainer writes |
| SmolVLA | 20,000 | No saveSteps field. The form offers seed and logFreq instead |
| ACT | 100,000 | No saveSteps field. chunkSize, nActionSteps, seed and logFreq instead |
Choosing one
- Keep two or three from the second half of the run. Earlier ones are usually undertrained and the final one has no special status.
- Evaluate each on the arm with the same protocol: same reset, same lighting, same number of attempts, results written down as you go.
- Compare success rates, not losses. A lower training loss and a higher success rate on hardware are two different questions, and the second one is the one you care about.
- Record the winning step number alongside the job id and the dataset. That triple is what makes a result reproducible three months later.
Training runs on a rented cloud GPU, started in the dashboard under Training or through the start_training tool on the MCP server, and the run page lists the checkpoints it produced. What the CLI does is run one. Without a checkpoint, the run command executes the base model, which is a useful smoke test before your own policy exists.
# Run one specific checkpoint against the task it was trained on
ay-robots run --model smolvla --checkpoint <checkpoint-id> \
--task "pick up the red cube and place it in the bowl"
# Stop it
ay-robots stop/learn/train-your-first-policy covers loss curves and what they do and do not tell you, and /train has the per-model walkthroughs that end in a checkpoint on real hardware.