Reproduce perplexity

[deciding]

In the readme the ppl is

Llama-2-7b | 1x16 | 5.92 | 2.4

In the paper it is:

Llama-2-7b AQLM 2.29 6.29 8.11

When I run locally using the same command as in the readme

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py $MODEL_PATH $DATASET_PATH --nsamples=1024 \
 --num_codebooks=1 --nbits_per_codebook=16 --in_group_size=8 \
 --relative_mse_tolerance=0.01 --finetune_relative_mse_tolerance=0.001 \
 --finetune_batch_size=32 --local_batch_size=1 --offload_activations \
 --wandb --save $SAVE_PATH

it gives me

Llama-2-7b AQLM 2.29 6.45 8.39

Can I know why there is such a mismatch? Thanks for any clarifications.