# Attribution and modifications

## Base model and tokenizer

This repository builds on IBM's `ibm-granite/granite-4.2-3b` model and tokenizer.

Pinned upstream revision: `e459acceac81e5fe67c07d9cfc72329a332e7eb1`

Upstream repository:
https://huggingface.co/ibm-granite/granite-4.2-3b

The upstream model card declares Apache 2.0. Upstream license and notice
files found at the pinned revision are preserved under `upstream/`.
The standard Apache 2.0 text is included in the root `LICENSE`.

## Project contribution

B. Goldberg / Hugging Face user `bengoldberg0` created this independent
research adapter through QLoRA fine-tuning and performed the benchmark
comparisons documented in the model card.

The adapter changes model behavior when applied to the base checkpoint.
This repository does not distribute the full base-model weights.

The project is not endorsed by IBM or the dataset authors.

## Training sources

- PhiUSIIL:
  https://archive.ics.uci.edu/dataset/967/phiusiil+phishing+url+dataset
- PhreshPhish:
  https://huggingface.co/datasets/phreshphish/phreshphish

## External evaluation source

- Malicious URLs Dataset:
  https://www.kaggle.com/datasets/sid321axn/malicious-urls-dataset

Source-specific preparation, exclusions, revisions, and limitations are
documented in the model card.

Raw datasets are not redistributed. Apache 2.0 for the adapter and
original project code does not relicense source datasets or override
their terms.