---

name: export-and-prompt description: "Prompt trained experiments locally, tune generation parameters, and publish or export to Hugging Face Hub." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

export-and-prompt

Use this sub-skill when a trained experiment already exists and the task is to interactively prompt it, adjust generation settings during the prompt session, preflight the saved artifacts, or prepare a Hugging Face Hub export.

Route here when

• The user wants to chat with a saved experiment locally.
• The user wants to change generation settings at prompt time.
• The user wants to verify that the experiment directory is ready for prompt or export.
• The user wants to publish a trained checkpoint to Hugging Face Hub.
• The user wants to hand a downloaded or published model to h2oGPT.

Route elsewhere when

• The user still needs dataset preparation or experiment creation.
• The user needs to train, resume, or compare experiments.
• The user wants model internals, loss logic, or metric behavior.

Bundled references

• references/prompting-and-generation.md
• references/hugging-face-export.md
• references/troubleshooting.md

Bundled scripts

• scripts/check_experiment_artifacts.py
• scripts/check_publish_inputs.py

Fast checks

• python scripts/check_experiment_artifacts.py --help
• python scripts/check_publish_inputs.py --help
• python llm_studio/prompt.py -h
• python llm_studio/publish_to_hugging_face.py -h

Working rule

Use the preflight scripts first, then the runtime command that matches the user's goal. Prompt-session parameter changes are local to that session and do not rewrite the saved experiment.