liveBook · Manning

Structured extraction

As explained in LLM Customization and Fine-Tuning

The process of identifying and extracting specified fields from unstructured or semi-structured sources.

When invoice layouts vary across vendors, a small example set may leave some formats uncovered. Many-shot prompting can help by showing at least one example from each layout family; otherwise, the model may force varied invoices into the dominant template. section 4.3.1

Invoices with varied layouts

For invoices from many vendors, represent each layout family in the examples so the model encounters the range of formats it must extract fields from. section 4.3.1

How LLM Customization and Fine-Tuning covers it

How deep it goes
Explains why layout coverage matters and stops short of specifying an extraction implementation.
Examples
Uses invoices from vendors with varied layouts to illustrate the need for representative examples.
What you can do after
Apply many-shot prompting by including at least one example for each input layout family.
How it connects
Connects structured extraction to many-shot learning and in-context learning.

Also explained in

Compare how 6 books explain structured extraction →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of LLM Customization and Fine-Tuning. The section links point to the parts of the book it draws on.