> For the complete documentation index, see [llms.txt](https://docs.zingg.ai/zingg/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zingg.ai/zingg/recipes-and-integration/pre-trained-models.md).

# Pre-trained Models

Use a Zingg pre-trained model to start matching without going through the findTrainingData and label phases.

Zingg ships with a pre-trained model built on the `FEBRL` dataset; for example, a synthetic Australian population dataset with name, address, date of birth, and SSN fields. The `FEBRL` pre-trained model is bundled in the Zingg Docker image and available from the Zingg GitHub\
repository.

Use a pre-trained model when you want to run a quick proof of concept on data that matches the `FEBRL` schema, or as a starting point that you fine-tune with your own labels.

### Using the FEBRL pre-trained model

#### Step 1: Download the pre-trained model

The FEBRL pre-trained model is bundled in the Zingg Docker image. It is available in the models/ directory inside the container.

```bash
docker pull zingg/zingg:0.5.0
docker run -it zingg/zingg:0.5.0 bash
```

Inside the container, models are at:

```
/zingg/models/
```

{% hint style="success" icon="right-long" %}
Pre-trained models are also available from the Zingg GitHub releases page: `github.com/zinggAI/zingg/releases`
{% endhint %}

#### Step 2: Use the pre-trained model in your config

```json
{
  "fieldDefinition": [
    {
      "fieldName": "fname",
      "matchType": "fuzzy",
      "fields": "fname",
      "dataType": "string"
    },
    {
      "fieldName": "lname",
      "matchType": "fuzzy",
      "fields": "lname",
      "dataType": "string"
    },
    {
      "fieldName": "dob",
      "matchType": "exact",
      "fields": "dob",
      "dataType": "string"
    },
    {
      "fieldName": "ssn",
      "matchType": "exact",
      "fields": "ssn",
      "dataType": "string"
    }
  ],
  "data": [
    {
      "name": "yourInput",
      "format": "csv",
      "props": {
        "location": "path/to/your/data.csv",
        "delimiter": ",",
        "header": "true"
      }
    }
  ],
  "output": [
    {
      "name": "yourOutput",
      "format": "csv",
      "props": {
        "location": "/tmp/zinggOutput"
      }
    }
  ],
  "modelId": "100",
  "zinggDir": "models",
  "labelDataSampleSize": 0.5,
  "numPartitions": 4
}
```

#### Step 3: Run match directly (no training needed)

**Python**

```python
options = ClientOptions([ClientOptions.PHASE, "match"])
zingg = Zingg(args, options)
zingg.initAndExecute()
```

**CLI**

Skip `findTrainingData` and `label`. Run `match` directly using the pre-trained model.

```bash
./scripts/zingg.sh --phase match --conf config.json
```

{% hint style="success" icon="right-long" %}
The pre-trained `FEBRL` model works best on data with the same field schema: `fname`, `lname`, `stNo`,\
`add1`, `add2`, `city`, `areacode`, `state`, `dob`, `ssn`.

For significantly different schemas or data distributions, fine-tuning with your own labels will improve accuracy.
{% endhint %}

### Fine-tuning a pre-trained model

You can use a pre-trained model as a starting point and improve it with your own labeled data. Run `findTrainingData` and `label` on your own dataset, then run train. Zingg will continue training using the pre-trained model weights rather than starting from scratch.

This is faster than building a model from scratch when your data is structurally similar to the pre-trained model's training data.

{% hint style="success" icon="right-long" %}
**Read more**:

* Create training data - [Create training data](/zingg/running-zingg/create-training-data.md)
* Label training pairs - [Label training pairs](/zingg/running-zingg/create-training-data/label-training-pairs.md)
* Build and save the model - [Build and save the model](/zingg/running-zingg/build-and-save-the-model.md)
  {% endhint %}

### Other available pre-trained models

Zingg ships with pre-trained models for several datasets. All models are bundled under the `models/` directory in the Zingg installation. Check `models/modelIds.txt` for the complete list of available models and their IDs.

<table><thead><tr><th valign="top">Model</th><th valign="top">Dataset</th><th valign="top">Fields</th></tr></thead><tbody><tr><td valign="top">FEBRL</td><td valign="top">Synthetic Australian population</td><td valign="top"><code>fname</code>, <code>lname</code>, <code>stNo</code>, <code>add1</code>, <code>add2</code>, <code>city</code>, <code>areacode</code>, <code>state</code>, <code>dob</code>, <code>ssn</code></td></tr><tr><td valign="top">North Carolina Voters</td><td valign="top">US voter registration records</td><td valign="top">Name, address, date of birth</td></tr><tr><td valign="top">Amazon-Google Products</td><td valign="top">Product catalog matching</td><td valign="top">Product title, description</td></tr></tbody></table>

For schemas and model IDs, see `models/modelIds.txt` inside your Zingg installation.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.zingg.ai/zingg/recipes-and-integration/pre-trained-models.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.