Skip to content

commandline.md

Latest commit

 

History

History
102 lines (88 loc) · 5.53 KB

commandline.md

File metadata and controls

102 lines (88 loc) · 5.53 KB
Raw

Neuracore CLI

The neuracore command helps you authenticate, choose an organization, monitor training runs (cloud and local), and launch a local policy server for quick validation. It does not yet create datasets or start training jobs—use the SDK for those and come back here to monitor.

Configure once

  • Login and save an API key:
    neuracore login
    Non-interactive: neuracore login --email you@example.com --password '<PASSWORD>'
  • Select your organization:
    neuracore select-org --org-name "<ORG_NAME>"
  • Config is stored at ~/.neuracore/config.json. Override the backend with NEURACORE_API_URL=https://staging.api.neuracore.com/api if needed.

Training commands require neuracore[ml] (pip install "neuracore[ml]").

10-minute journey: sign in → monitor → inspect → serve

  1. Sign in and pick your org
    (Run this once per machine; re-run if you switch orgs.)
neuracore login
neuracore select-org --org-name "<ORG_NAME>"
  1. List recent training runs (cloud and local by default)
neuracore training list --limit 5
# Expected: a table plus plain-text rows like:
# run-2024-08-12 | Yes | diffusion_policy | <DATASET_ID>

# Cloud only:
neuracore training list --cloud --limit 5
# Local only:
neuracore training list --local --limit 5
  1. Inspect a run
# Cloud (default)
neuracore training inspect --training-name <RUN_NAME_OR_ID>
# Add --config to print algorithm config, or --json for raw JSON.

# Local
neuracore training inspect --local --training-name <RUN_NAME>
  1. Launch a local policy server to sanity-check inference
neuracore launch-server \
  --input_embodiment_description '{"RGB_IMAGES": {"0": "front_cam"}}' \
  --output_embodiment_description '{"JOINT_TARGET_POSITIONS": {"0": "arm"}}' \
  --job_id <RUN_ID> \
  --org_id <ORG_ID> \
  --port 8080
  1. Clean up
neuracore training delete --training-name <RUN_NAME_OR_ID> --yes
# Delete a local run directory instead:
# neuracore training delete --local --root <RUNS_DIR> --training-name <RUN_NAME> --yes

Core commands reference

  • neuracore login [--email <EMAIL>] [--password <PASSWORD>] — saves your API key to ~/.neuracore/config.json.
  • neuracore select-org [--org-name <NAME> | --org-id <ID>] — sets the active organization.
  • neuracore training list [--cloud | --local | --all] [--status <STATUS>] [--limit <N>] [--root <DIR>]
    • Default is both cloud and local (~/.neuracore/training/runs).
    • Status examples: PENDING, RUNNING, COMPLETED.
  • neuracore training inspect --training-name <RUN> [--cloud | --local] [--root <DIR>] [--config] [--json]
    • Cloud is the default; add --local to read training_run.json from a local run directory.
  • neuracore training monitor [--training-name <RUN>] [--root <DIR>] [--port <PORT>] [--host <HOST>] [--bind-all]
    • Launches TensorBoard for local runs. With --training-name, opens that run’s tensorboard/ logs; without it, opens all runs under --root for comparison. Requires the tensorboard executable on PATH (neuracore[ml]).
  • neuracore training delete --training-name <RUN> [--cloud | --local] [--root <DIR>] [--yes]
    • Cloud is the default; add --local to remove a run directory under --root. Prompts unless --yes is used.
  • neuracore training start — currently a placeholder; use the SDK to launch training and return here to monitor.
  • neuracore launch-server --input_embodiment_description '<JSON>' --output_embodiment_description '<JSON>' [--job_id <RUN_ID>] [--org_id <ORG_ID>] [--host <HOST>] [--port <PORT>]
    • JSON must map DataType strings (e.g., "RGB_IMAGES") to indexed name mappings, with JSON string keys for the indexes (for example, {"RGB_IMAGES": {"0": "front_cam"}}).

Troubleshooting

  • Authentication failed. Please run 'neuracore login' first.
    • Re-run neuracore login; confirm ~/.neuracore/config.json exists and contains an API key.
  • Training commands require optional ML dependencies / import errors on neuracore training.
    • Install with pip install "neuracore[ml]".
  • No runs show up but you know they exist.
    • Ensure you selected the right org (neuracore select-org) and used the correct --status filter.
  • Invalid status filter when listing runs.
    • Use one of PENDING, RUNNING, COMPLETED, FAILED, CANCELLED.
  • Local inspect fails with missing training_run.json.
    • Point --root to the directory where your Hydra runs live (default ~/.neuracore/training/runs).
  • TensorBoard executable not found when running training monitor.
    • Install neuracore[ml]. If you see No module named 'pkg_resources', run pip install "setuptools<82".
  • TensorBoard log directory not found when running training monitor.
    • Confirm the run finished at least one logging step, or use the correct --training-name / --root.
  • launch-server JSON errors.
    • Ensure the JSON strings use double quotes and DataType keys match your model ("RGB_IMAGES", "JOINT_TARGET_POSITIONS", etc.).

FAQ

  • Where is CLI config stored? ~/.neuracore/config.json (API key and current org).
  • How do I switch API environments? Set NEURACORE_API_URL before running commands.
  • Where are local runs written? ~/.neuracore/training/runs (override with --root).
  • How do I view local training metrics? neuracore training monitor --training-name <RUN_NAME>.
  • Can the CLI upload datasets or start training? Not yet. Dataset creation/upload and training launch are available through the Python API; neuracore training start is a placeholder today.