volker42maru
volker42maru
CuBERT seems to be running on all threads of all CPUs now. It was an issue with the KMP flag it seems. But actually it's slower in benchmark when utilizing...
I am running both TF-BERT and CuBERT in python at the moment, because my server is also implemented in python. I included TF-BERT into the python benchmark script by loading...
# Do you run cuBERT inside docker with limited CPU quota? I am not running inside a docker container. I use the same conda environment and the same CPU server...