LLM Attributor: Inspect LLM training-data attribution

LLM Attributor

LLM Attributor is a Python package for interactively visualizing training-data attribution for large language model text generation. It lets users select generated phrases to inspect associated training data points, edit generated text, and compare attribution results side by side.

Independent OpenViST project profile, based on linked sources.

Understand the project

Practical overview

When it fits

Use it when you have model-generated text and want an interactive view of training-data attribution for selected phrases, including a side-by-side comparison after editing.

Who it is for

Researchers examining relationships between LLM outputs and training examples
Developers exploring attribution changes after editing generated text

Documented uses

Inspect training data points associated with a selected generated phrase
Compare phrase-level attribution before and after editing generated text
Initialize an attribution workflow with a training dataset and model configuration in a notebook

Before you choose

Requirements & limits

The repository documents installation through PyPI and identifies the software as MIT-licensed.

The documented setup uses computational notebooks rather than describing a hosted interface.
Initialization requires a training dataset and model-save directory.
The README demonstrates LLaMA 2 model and tokenizer paths and notes these are needed when the model is not fine-tuned.
The repository includes disaster and finance demo notebooks.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install and initialize LLM Attributor in a notebook

Prepare the package and configure its model and dataset inputs in a computational notebook.

Before you start

Python package installation using pip
A computational notebook such as Jupyter Notebook or JupyterLab
A training dataset and model-save directory
LLaMA 2 model and tokenizer paths; the README says these are necessary when the model is not fine-tuned

Steps

Install the package with the command below.
Documented command
```
pip install llm-attributor
```
Import LLMAttributor in the notebook.
Initialize it with the LLaMA 2 model path, tokenizer path, model-save directory, and training dataset.

Expected outcome

An initialized LLM Attributor instance configured with the supplied model and data paths.

Guide limitations

The documented steps cover installation and initialization; they do not establish a complete model-training or attribution-run procedure.

Documentation 1 ↗

Evidence and transparency

Sources & review

Official repository ↗

Source record details

Source availability: Active
Sources checked: 17 October 2026
Original website title: poloclub/LLM-Attributor · Official homepage title · retained in its source language

How we review sources Suggest a correction