Local LLM / Local prompting and model adaptation
Local LLM: Extraction null semantics
Define when an absent field should be null, omitted or explicitly unknown.
Environment and inputs: Local chat models and optional PEFT/LoRA adaptation; templates and adapter compatibility depend on the base model.
When to use it
Version prompts, templates and decoding settings alongside the model artifact. Test a change on unseen examples before adopting it, and keep domain improvements separate from regressions in capabilities that the application still requires.
Practical example
Illustrative extraction schema. Null records missing evidence; it is not a fabricated invoice number.
{
"type": "object",
"properties": {
"invoice_number": {
"type": [
"string",
"null"
]
}
},
"required": [
"invoice_number"
],
"additionalProperties": false
}Check the result
Use documents with missing values and verify the model does not invent replacements.
Save and compare results
Save the test input, configuration and result. Fix the failing boundary, repeat the same test and compare it with the saved working configuration.
Scope and limits
These exercises have not been tested on your systems. Check feature support and configuration names against the installed release and linked documentation.