MNL model estimated in Pyro has a different result


I am trying to estimate a Multinomial Logit Model in Pyro. To benchmark the performance of the model, I estimated the same models using the linear model (logistic regression) in sklearn.

The coefficients are also statistically insignificant in Pyro. But, I checked for the same using other software and the estimates are all statistically significant.

Could you help me resolve this issue? If you find something wrong with the way I have approached the model and coded the same, please point that out.

Thanks in advance.