Can not reproduce results of Table 2.

[shuhao02]

Hi, thanks for your outstanding work.

I am trying to reproduce the result of Table 2, especially the Wanda w. DSnoT of LLaMA-V1-7B under the 70% sparsity.

Here is my command:

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py \ --model decapoda-research/llama-7b-hf \ --prune_method DSnoT \ --initial_method wanda \ --sparsity_ratio 0.7 \ --sparsity_type unstructured \ --max_cycle_time 50 \ --update_threshold 0.1 \ --pow_of_var_regrowing 1

The perplexity is 72.9418335, while the reported perplexity is 62.05.

Can you release the command to reproduce the result?