@Retired_mod 

The DBR runtime compatible  cuda version (11.3) is DBR 11.X. unfortunately pyspark.ml.torch.distributed works only with DBR 13.X

So going back to DBR 11.X is not solving the problem