replace-with-skill-name

name	REPLACE-WITH-SKILL-NAME
description	One-to-three-sentence description of what the model does and when to use it. Use when the user asks to "fine-tune REPLACE-WITH-NETWORK", "train REPLACE-WITH-NETWORK on REPLACE-WITH-DATA-TYPE", or mentions REPLACE-WITH-DOMAIN-TERMS. Include literal trigger phrases.
license	Apache-2.0
compatibility	Requires docker + nvidia-container-toolkit + NGC API key.
metadata	{"author":"REPLACE-WITH-AUTHOR-NAME","version":"0.1"}
allowed-tools	Read Bash

Skill Name

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

External dependencies

Dependency	Purpose	Install
docker	Run the training container	https://docs.docker.com/engine/install/
nvidia-container-toolkit	GPU access in containers	https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
NGC API key	Pull nvcr.io images	https://ngc.nvidia.com/

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

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

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

Argument	Required	Default	Description
-e, --experiment_spec	Yes	—	Path to YAML spec file (mounted into container)
-r, --results_dir	No	from spec	Output directory
-k, --key	No	—	TLT encryption key (legacy models)

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

Model	Path in container	Source
Base	<container_path>	NGC / HuggingFace / URL

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.

Parameter	Schema default	Required value	Why
...	...	...	...

Important parameters (reference)

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

Hardware

• Minimum: ×
• Recommended: ×

Known pitfalls

Symptom	Cause	Fix
FileNotFoundError on pretrained model	Missing mount	Add -v for the checkpoint dir
OOM during training	Batch size too large	Reduce train.batch_size or add GPUs

Optional: running via the TAO SDK

If nvidia-tao-sdk is installed and you want job tracking + S3 I/O wrapping:

from tao_sdk.platforms.brev import BrevSDK   # or: from tao_sdk.platforms.slurm import SlurmSDK
sdk = BrevSDK()
job = sdk.create_job(
    image='nvcr.io/nvidia/tao/tao-toolkit:<tag>',
    command='<entrypoint-cmd> train -e /spec.yaml',
    gpu_count=1,
    env_vars={'HF_TOKEN': os.environ['HF_TOKEN']},
    inputs={'/spec.yaml': 's3://bucket/specs/train.yaml',
            '/data/':     's3://bucket/datasets/...'},
    outputs=['/results/'],
)

See tao-skill-bank:tao-run-platform for full SDK semantics.