XGBoost preds uniform probability

I trained a 101 class model on a GPU instance (with xgb 0.90) and saved the model using joblib. I was able to run it just fine to get classifications. Upon moving to a different CPU-only instance with a fresh install of xgboost (0.90) + other libraries, I get a weird prediction of :

    [[0.00990099 0.00990099 0.00990099 0.00990099 0.00990099 0.00990099
     ....
      0.00990099 0.00990099 0.00990099 0.00990099 0.00990099]]    

This is my code and I can see the kf_descs_list populates just fine

clf = joblib.load("./models/clf_101_classes.job")
dquery = xgb.DMatrix(np.array(kf_descs_list)) 
preds_list = clf.predict(dquery)
print(preds_list)

I have checked that my numpy and other dependencies match and other (non-xgboost) parts of the code work fine.

Could this because I am running on a CPU-only instance? something else I am missing?

This looks like a bug. Can you put up a script that I can run to reproduce the bug?