Preparing Data¶

LLM Extractinator expects that each row (CSV) or item (JSON) contains the text you want to extract from.

The extractor needs to know which field/column to read — you’ll tell it that in the task JSON ("Input_Field": "text").

CSV example¶

id,text
1,"This is the first report..."
2,"This is the second report..."

text is the column we will extract from
you can have more columns; they will be carried through if the task supports it

JSON example¶

[
  {
    "id": 1,
    "text": "This is the first report..."
  },
  {
    "id": 2,
    "text": "This is the second report..."
  }
]

Again, text is the field we will extract from.

Data path¶

In your task JSON you will refer to the data:

{
  "Data_Path": "reports.csv",
  "Input_Field": "text"
}

Data_Path is relative to the data directory you pass to the CLI (--data_dir)
Input_Field must match the column/key name exactly

Few-shot examples¶

If you want to use few-shot learning (providing examples to guide the model), you can create an examples file. Set --num_examples > 0 and reference the file in your task JSON.

Example file format¶

Few-shot examples should be in JSON format with input and output fields:

[
  {
    "input": "Patient presented with severe headache and nausea.",
    "output": "{\"symptoms\": [\"headache\", \"nausea\"], \"severity\": \"severe\"}"
  },
  {
    "input": "No significant findings. Patient is in good health.",
    "output": "{\"symptoms\": [], \"severity\": null}"
  }
]

Important notes:

Each example must have an input field (the text to extract from)
Each example must have an output field (the expected JSON output as a string)
The output should match the structure defined by your parser
Store examples in the examples/ directory (or the path specified by --example_dir)

Adding examples to your task¶

In your task JSON, add the Example_Path field:

{
  "Description": "Extract medical symptoms",
  "Data_Path": "medical_reports.csv",
  "Input_Field": "text",
  "Parser_Format": "medical_parser.py",
  "Example_Path": "medical_examples.json"
}

Then run with --num_examples:

extractinate --task_id 1 --model_name "phi4" --num_examples 3

This will use semantic similarity to select the 3 most relevant examples for each input.

Configuring the embedding model¶

By default, nomic-embed-text is used for finding similar examples. You can change this:

extractinate --task_id 1 --num_examples 3 --embedding_model "mxbai-embed-large"

Parsers and data¶

Once your data is ready, you must also define a parser (the output structure). See the Parser page for that.