Data Field Container - Build Fine Tuning File

Can be one of the following types:

Specifies which Data Fill Method will be used to generate the fine-tuning data. The selected fill method determines which extraction logic and schema are used when building the training file.

Usage Guidance

• Use the drop-down list to select from available fill methods configured on the current Data Field Container.
• Typically, you will select an AI Extract fill method, as this is the method that benefits from fine-tuning.

> Tip: If you have multiple fill methods (such as AI Extract and Fill Descendants), choose the one that matches your intended extraction scenario.

Selects the Batches whose documents will be used to generate the fine-tuning data. The documents in these batches should have already been extracted and then reviewed/corrected by human operators to represent the ideal state after extraction.

Usage Guidance

• Use the batch selector to choose one or more Batches containing representative and high-quality training samples.
• Only documents matching the declaring Content Type of the selected Data Field Container will be included.

Example Scenario

• After running extraction and review on a batch named Training Set, select that batch here to use its documents for fine-tuning.

> Tip: The more diverse and accurate your training samples, the better your fine-tuned model will perform.

Determines whether the schema for the selected Data Field Container is included in the fine-tuning data. Including the schema can help guide the LLM to produce output that matches the expected structure, improving extraction accuracy.

Usage Guidance

• Set to true to include the schema in each training example. This is recommended for most fine-tuning scenarios.
• Set to false only if you have a specific reason to exclude the schema (such as advanced prompt engineering).

> Note: Including the schema is generally recommended unless you are fine-tuning for a highly specialized use case.