This repository is the official implementation of our paper "Exploring Language-Specific Regions in Large Language Models".

We provide our found language-specific neurons in LLaMA-2 (7B), LLaMA-2 (13B), LLaMA-2 (70B), BLOOM (7B), OPT (6.7B), Mistral (7B), and Phi-2 (2.7B).

You should use torch.load to load xxx.neuron.pth, each of which is a List[List[LongTensor]], neuron[i][j] represents the neuron indice of the i-th language in the j-th layer in the model. The language 0-6 indice stand for en, zh, fr, es, vi, id, ja. For example, LLaMA-2-7B[1][4]=tensor([6147, 9114, 9292]), which means that the Chinese neurons inside the 4-th layer of LLaMA-2-7B are of the indice 6147, 9114, and 9292.

Record the activation state:

python activation.py -m meta-llama/Llama-2-7b-hf -l xx

Identifying language-specific neurons:

python identify.py

python ppl.py -m meta-llama/Llama-2-7b-hf -a activation_mask/xxx

python generation.py -m meta-llama/Llama-2-7b-hf -a activation_mask/xxx