kmseong/llama2_7b-chat_gsm8k_full_ft_lr5e-5
The kmseong/llama2_7b-chat_gsm8k_full_ft_lr5e-5 is a 7 billion parameter Llama 2-based chat model, specifically fine-tuned using the SN-Tune (Safety Neuron Tuning) method. This approach enhances safety alignment by selectively fine-tuning only critical safety neurons on the Circuit Breakers dataset, while freezing other parameters. It is designed to provide improved safety alignment with minimal impact on general capabilities, making it suitable for applications requiring robust safety features.
Loading preview...
Model Overview
This model, kmseong/llama2_7b-chat_gsm8k_full_ft_lr5e-5, is a 7 billion parameter Llama 2-based chat model that has undergone Safety Neuron Tuning (SN-Tune). It is built upon the meta-llama/Llama-3.2-3B-Instruct base model and was uploaded on 2026-04-22.
Key Capabilities & Features
- Enhanced Safety Alignment: Utilizes the SN-Tune method to specifically improve the model's safety responses.
- Parameter-Efficient Fine-tuning: Achieves safety improvements by only fine-tuning a small subset of "safety neurons" while keeping most parameters frozen.
- Minimal Impact on General Capabilities: The selective fine-tuning approach aims to preserve the base model's general performance.
- Targeted Training: Fine-tuned on the Circuit Breakers dataset, which is designed for safety alignment.
What is SN-Tune?
SN-Tune is a novel fine-tuning approach that involves:
- Identifying specific "safety neurons" within the model.
- Freezing all other non-safety related parameters.
- Training only these identified safety neurons on dedicated safety datasets.
When to Use This Model
This model is particularly well-suited for use cases where:
- Safety is a primary concern: Applications requiring a higher degree of safety alignment in conversational AI.
- Efficiency is important: The parameter-efficient fine-tuning means it can be adapted for safety without extensive retraining of the entire model.
- Building on Llama 2: Developers already using Llama 2-based models and seeking an enhanced safety version.