---

name: "deployment-and-extension" description: "Routes NeuralForecast save/load, serialization, export, logging, and model-extension workflows." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

deployment-and-extension

Use this sub-skill when the user wants to save or reload a fitted forecast, export a model, log to MLflow, or understand how to extend NeuralForecast with a new model or docs update.

What this route covers

• NeuralForecast.save and NeuralForecast.load.
• Save/load portability and overwrite behavior.
• Optional export surfaces such as ONNX or MLflow when the user explicitly needs them.
• Maintainer-facing extension guidance for adding or updating models.

What this route does not cover

• The core forecast loop itself.
• Dataframe preparation.
• Model selection.
• Auto* tuning or distributed execution.

If the user only needs to run the saved model, route to ../core-forecasting/SKILL.md. If the user first needs the panel schema or future exogenous layout, route to ../data-and-exogenous/SKILL.md.

Read these bundled references

• ../../references/deployment-extension.md for serialization and optional export notes.
• ../../references/workflows.md for save/load and quickstart recipes.
• references/troubleshooting.md for save/load, optional-dependency, and docs failures.

Run this bundled script

• ../../scripts/check_serialization.py for a small save/load round-trip.
• ../../scripts/core_smoke.py when you want to confirm a model still fits and predicts before serializing it.

Common triggers

• "Can I save this fitted model?"
• "How do I reload a NeuralForecast bundle?"
• "Why did my checkpoint path fail?"
• "How do I add a new model to the package?"
• "How do I make the docs or export workflow run again?"

Minimal route

1. Read ../../references/deployment-extension.md for the portability or extension path.
2. Use ../../scripts/check_serialization.py for a safe round-trip smoke.
3. Read references/troubleshooting.md when the save/load or optional export workflow fails.

Decision cues

• If the issue is model family selection, route to model-selection.
• If the issue is the panel dataframe, route to data-and-exogenous.
• If the issue is a forecast run before serialization, route to core-forecasting.

Safe default

When in doubt, confirm save/load first with a temporary directory and a tiny model, then move on to optional export or maintainer work only if needed.