nvidia-tao/tao-skill-bank/model

Skill Name

Standalone install? If this session was not initialized by the TAO skill bank plugin, run the tao-setup skill first (host preflight, credentials, cross-skill discovery).

Two-line summary of the model. What it is, what it produces.

External dependencies

Train Action Policy

This model is AutoML-enabled at the model layer. Before handling any train-stage request, read references/skill_info.yaml and resolve the run override from either an explicit automl_policy value or the user's workflow request. Use automl_policy: on by default and only expose on / off in new launch prompts. Treat phrases like "turn off AutoML", "disable AutoML", "no HPO", or "plain training" as automl_policy: off for this run only. When automl_policy: on, automl_enabled: true, and both schemas/train.schema.json and references/spec_template_train.yaml are packaged, route the train action through tao-skill-bank:tao-run-automl by default with this model's skill_dir. Preserve workflow/application overrides for datasets, specs, output directories, GPU/platform settings, parent checkpoints, and automl_policy. Use direct model training only when automl_policy: off or the packaged train schema/template is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.

Non-train actions such as evaluate, inference, export, and deploy flows stay in this model skill. The per-run automl_policy override does not change model metadata.

Quick start (Docker)

Train

set -a; source /path/to/.env; set +a   # omit if already exported
docker run --gpus all --rm \
  -e HF_TOKEN \
  -v /path/to/spec.yaml:/spec.yaml \
  -v /path/to/data:/data \
  -v /path/to/results:/results \
  nvcr.io/nvidia/tao/tao-toolkit:<tag> \
  <entrypoint-cmd> train -e /spec.yaml

Evaluate

set -a; source /path/to/.env; set +a   # omit if already exported
docker run --gpus all --rm \
  -e HF_TOKEN \
  -v /path/to/eval_spec.yaml:/spec.yaml \
  -v /path/to/checkpoint:/checkpoint \
  -v /path/to/results:/results \
  nvcr.io/nvidia/tao/tao-toolkit:<tag> \
  <entrypoint-cmd> evaluate -e /spec.yaml

Inference (dry-run)

docker run --gpus all --rm \
  -v /path/to/inference_spec.yaml:/spec.yaml \
  -v /path/to/test_dir:/test \
  -v /path/to/output:/output \
  nvcr.io/nvidia/tao/tao-toolkit:<tag> \
  <entrypoint-cmd> inference -e /spec.yaml

Container image and per-action command are in references/skill_info.yaml. See tao-skill-bank:tao-run-on-docker for docker run conventions.

CLI Reference

Output structure

<results_dir>/
├── train/
│   ├── checkpoints/
│   ├── logs/
│   └── metrics.json
├── evaluate/
│   └── results.json
└── inference/
    └── predictions/

Credentials

• `HF_TOKEN` (if gated weights) — describe what the token is for.
• `NGC_KEY` — for pulling nvcr.io images.

Pretrained weights

Data format

Describe expected input structure — directory layout, file conventions, annotation formats.

Critical overrides

If the model's schema has defaults that don't work out of the box, document them as a table.

Important parameters (reference)

Group by subsystem — training loop, model, optimization, vision, checkpointing, etc.

Hardware

• Minimum: <GPU count> × <VRAM>
• Recommended: <GPU count> × <GPU type>

Known pitfalls

Running on a platform

This skill describes the model, its container command, and the spec. To run with job tracking, S3 staging, or on a remote/multi-node cluster, hand the spec to a platform skill — tao-run-on-docker, -slurm, -kubernetes, -brev, -virtualenv, or any installed external platform skill — which executes it through the four-verb contract (submit/status/logs/ cancel) with no SDK. See tao-skill-bank:tao-launch-workflow.