Quickstart¶
Submit a training run in under a minute.
1. Create a Client¶
2. Submit a Training Run¶
run = client.submit_training_run(
name="my-first-run",
task_name="Template-Go2-Standing-Direct-v0",
code_url="https://github.com/canard-cloud/go2-standing-env",
num_envs=4096,
max_iterations=5000,
gpu_count=2,
)
The SDK prints an estimated cost and a link to the dashboard:
Submitting training run: my-first-run
Task: Template-Go2-Standing-Direct-v0
Envs: 4096 | Iterations: 5000 | GPUs: 2
Est. ~7.2 hr · $1.00/hr · ~$7.23 total
Dashboard: https://app.canard.cloud/runs/run_a1b2c3d4
3. Monitor Progress¶
From the dashboard¶
Open the dashboard URL printed above. You'll see live reward curves, GPU utilization, and training videos.
From Python¶
# Block until complete (with progress bar)
run.wait_for_completion(show_progress=True)
# Or poll manually
run.refresh()
print(f"Status: {run.status}, Progress: {run.progress:.0%}")
4. Download Results¶
# Download the best checkpoint
path = run.download_best_checkpoint("./checkpoints")
print(f"Checkpoint saved to {path}")
# Or download everything
run.download_results("./results")
5. List Artifacts¶
# See what's available
for artifact in run.list_artifacts():
print(f"{artifact['filename']} ({artifact['artifact_type']})")
# Filter by type
checkpoints = run.list_checkpoints()
videos = run.list_videos()
What's Next?¶
- Parameter Sweeps — Search across learning rates, entropy coefficients, etc.
- Custom Environments — Bring your own Isaac Lab environment
- Cost Estimation — Understand pricing before you submit