linzju/Bio-Medical-Llama-3-8B_EnchTable_Attention

EnchTable: Unified Safety Alignment Transfer (Bio-Medical + Attention)

This repository contains the Bio-Medical Llama-3-8B model aligned using the Attention-based intervention of the EnchTable framework.

This model is part of the research presented in the paper: "EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models", accepted at IEEE S&P 2026.

Model Details

• —Name: Bio-Medical Llama-3-8B (EnchTable-Attention)
• —Domain: Bio-Medical / Healthcare
• —Method: EnchTable (Attention Layers)
• —Base Model: Llama-3-8B fine-tuned on medical datasets

EnchTable is a framework designed to transfer safety alignment capabilities from a general-purpose safe model to specialized downstream models.

In this checkpoint, we apply safety vectors specifically to the Self-Attention layers. This ensures the model reduces harmful outputs (hallucinations, toxicity) while maintaining high performance in medical question-answering tasks.