Unable to reproduce RTN results in paper

[Golden-Wang]

Hi, thanks for the great work. I'm a big fan of how the configs are laid out.
I've been trying to reproduce the results mentioned in the paper using this repo. It was easy to do so for the methods that has a .yaml file in configs. However, no config is provided for RTN.
The best W4A8KV4 RTN result for Llama-2-7b I can get using this repo is 7.99 PPL on wikitext2. The paper gives 6.51 PPL for W4A8KV4 RTN and 5.99 PPL for W4A8KV4g128 RTN.
How were these results obtained? Thanks in advance.

[synxlin]

Hi,
For RTN results, please turn off all optimizations in QOQ configurations including smooth, rotation and weight clipping. We keep the same weight data type (INT4 with zero points) for RTN and QOQ experiments.