Skip to content

CLI reference

This page is generated from the current Typer application, so it is the source of truth for command names, arguments, and options. Task guides intentionally do not copy every flag.

The CLI covers scenario resolution/build/training, dataset export, model export, smoke tests, tuning, registry inspection, backend inspection, and local-library management. Evaluation itself currently runs inside the training/test lifecycle rather than through a separate top-level evaluate command.

nexuml

NexuML pipeline framework CLI

Usage:

nexuml [OPTIONS] COMMAND [ARGS]...

Options:

  --install-completion  Install completion for the current shell.
  --show-completion     Show completion for the current shell, to copy it or
                        customize the installation.
  --help                Show this message and exit.

backend

Inspect available backend implementations

Usage:

nexuml backend [OPTIONS] COMMAND [ARGS]...

Options:

  --help  Show this message and exit.
list

List available backend implementations

Usage:

nexuml backend list [OPTIONS] [CATEGORY]

Options:

  [CATEGORY]  Optional category filter, e.g. data-export, data-loader,
              training
  --help      Show this message and exit.

build

Compile and validate a pipeline from config

Usage:

nexuml build [OPTIONS] CONFIG_PATH

Options:

  CONFIG_PATH  Path to resolved YAML config  [required]
  --help       Show this message and exit.

export

Export a trained pipeline

Usage:

nexuml export [OPTIONS] SCENARIO_NAME

Options:

  SCENARIO_NAME      Scenario name to train and export  [required]
  -o, --output PATH  Export directory  [default: exported_model]
  --checkpoint PATH  Optional checkpoint to export
  --help             Show this message and exit.

export-dataset

Export a dataset view from a scenario or config

Usage:

nexuml export-dataset [OPTIONS] [SCENARIO_NAME]

Options:

  [SCENARIO_NAME]                 Scenario name
  -c, --config PATH               Config YAML path
  -o, --output TEXT               Path or s3:// URI  [default:
                                  exported_dataset]
  --backend TEXT                  Dataset export backend  [default: numpy]
  --split TEXT                    Split to export. Repeat the option to export
                                  multiple splits.
  --preprocess / --no-preprocess  Run the compiled pipeline until the
                                  requested preprocessing keys exist.
                                  [default: no-preprocess]
  --preprocess-until-key TEXT     TensorDict x key marking the preprocessing
                                  boundary. Repeatable.
  --x-key TEXT                    x TensorDict keys to persist. Repeatable.
                                  Default: all.
  --y-key TEXT                    label TensorDict keys to persist.
                                  Repeatable. Default: all.
  --labels / --no-labels          Include labels from the batch y TensorDict.
                                  [default: labels]
  --dtype TEXT                    Optional storage dtype passed to the export
                                  backend, e.g. float16.
  --samples-per-shard INTEGER     Samples per WebDataset tar shard  [default:
                                  256]
  --s3-endpoint-url TEXT          Optional S3-compatible endpoint
  --s3-region TEXT                Optional S3 region
  --s3-profile TEXT               Optional AWS credential profile name
  --help                          Show this message and exit.

library

Manage local library roots

Usage:

nexuml library [OPTIONS] COMMAND [ARGS]...

Options:

  --help  Show this message and exit.
add

Add a local library root path

Usage:

nexuml library add [OPTIONS] PATH

Options:

  PATH    Path to local library root  [required]
  --help  Show this message and exit.
delete

Remove a local library root path

Usage:

nexuml library delete [OPTIONS] PATH

Options:

  PATH    Path to local library root  [required]
  --help  Show this message and exit.
list

List available library sources

Usage:

nexuml library list [OPTIONS]

Options:

  --help  Show this message and exit.

registry

Inspect discovered registry contents

Usage:

nexuml registry [OPTIONS] COMMAND [ARGS]...

Options:

  --help  Show this message and exit.
list

List registered items by kind: layers|data|scenarios|eval

Usage:

nexuml registry list [OPTIONS] [KIND]

Options:

  [KIND]         Kind to list: layers, data, scenarios, eval  [default:
                 layers]
  -v, --verbose  Show full tracebacks for discovery errors
  --help         Show this message and exit.

resolve

Resolve a scenario to YAML config

Usage:

nexuml resolve [OPTIONS] NAME

Options:

  NAME               Scenario name  [required]
  -o, --output PATH  Output YAML path
  --help             Show this message and exit.

smoke

Run full smoke test: resolve → build → train → export → reload → infer

Usage:

nexuml smoke [OPTIONS] [SCENARIO_NAME]

Options:

  [SCENARIO_NAME]       Scenario name  [default: synthetic-linear-ae-
                        reconstruction]
  --max-epochs INTEGER  Training epochs  [default: 3]
  --download            Download datasets before training
  --help                Show this message and exit.

train

Train a pipeline from a scenario or config

Usage:

nexuml train [OPTIONS] [SCENARIO_NAME]

Options:

  [SCENARIO_NAME]            Scenario name
  -c, --config PATH          Config YAML path
  --scenario-file PATH       Trusted Python file exposing scenario() ->
                             ScenarioSpec
  --artifact-dir PATH        Optional directory for scenario-file provenance
                             snapshots
  --max-epochs INTEGER       Override max epochs
  --trainer-checkpoint PATH  Optional Lightning Trainer checkpoint to resume
                             from.
  -O, --override TEXT        Override a scenario field: key.path=value.
                             Repeatable.
  --help                     Show this message and exit.

tune

Run Optuna hyperparameter search on a scenario

Usage:

nexuml tune [OPTIONS] [SCENARIO_NAME]

Options:

  [SCENARIO_NAME]       Scenario name
  --scenario-file PATH  Trusted Python file exposing scenario() ->
                        ScenarioSpec
  --artifact-dir PATH   Optional directory for scenario-file provenance
                        snapshots
  --n-trials INTEGER    Number of Optuna trials
  --metric TEXT         Metric to optimise (default: tuning_spec.metric_key or
                        val/loss)
  --direction TEXT      Optuna direction: minimize or maximize (default: from
                        tuning_spec)
  --storage TEXT        Optuna storage path
  --prune / --no-prune  Enable Optuna pruning
  -O, --override TEXT   Override a scenario field: key.path=value. Repeatable.
  --help                Show this message and exit.