linzju/Code-Llama-3-8B_EnchTable_FFN_Attention

EnchTable: Unified Safety Alignment Transfer (Code + Hybrid)

This repository contains the Code-Llama-3-8B model aligned using the Hybrid (Attention + FFN) 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: Code-Llama-3-8B (EnchTable-Hybrid)
• Domain: Code Generation / Programming
• Method: EnchTable (Attention + FFN)
• Base Model: Llama-3-8B fine-tuned on code datasets

EnchTable is a framework designed to transfer safety alignment capabilities.

In this checkpoint, we utilize a Hybrid strategy, injecting safety vectors into both Self-Attention and FFN layers. This represents the most comprehensive alignment transfer, maximizing safety in code generation tasks.