Help mimicking TVM Result

Hi ,

I have been working on TVM for a month now in order to mimic the result from this paper : https://arxiv.org/pdf/1905.00641.pdf ( page 8, table 5) .
Unfortunately, I haven’t been able to achieve even slightly similar results and I wanted to know if someone would be willing to discuss on a video call.
I have the mxnet model and the script.

Thanks