waihongteh/llama-finetuned30
The waihongteh/llama-finetuned30 is a 7 billion parameter Llama 2-based language model, fine-tuned from Meta's Llama-2-7b-chat-hf. It was specifically retrained on the TOFU benchmark, a dataset of fictitious authors, for machine unlearning research. This model is intended for reproducibility in studies concerning machine unlearning and is not designed for general deployment.
Loading preview...
Overview
The waihongteh/llama-finetuned30 is a 7 billion parameter language model derived from Meta's Llama-2-7b-chat-hf. This specific checkpoint was fine-tuned as part of a machine unlearning study, focusing on the TOFU (Test Of Fictitious Unlearning) benchmark. Its primary purpose is to serve as a reproducible artifact for research into machine unlearning techniques.
Key Characteristics
- Base Model: Fine-tuned from
meta-llama/Llama-2-7b-chat-hf. - Training Data: Retrained on the TOFU benchmark, which consists of fictitious author data.
- Research Focus: Developed for reproducibility in machine unlearning studies, specifically as a 'retained' model (TOFU retain90, meaning it never encountered the 'forget' set).
- Context Length: Supports a context length of 4096 tokens.
Intended Use
This model is not intended for general deployment or production applications. Its sole purpose is for research and reproducibility within the context of machine unlearning experiments. Users should refer to the original Llama 2 Community License and Acceptable Use Policy for licensing details, as this model is subject to those terms.