Closed mahmoodn closed 1 year ago
We use nvbit with docker images regularly. I am not sure what could be the issue. Can you please try running with VERBOSE=3 in front of LD_PRELOAD?
OK. Apparently, when Nvbit is used, more memory is occupied on the device due to the overhead. Therefore, the batch_size of 128 which is fine when running the application alone, is not workable with Nvbit. When I reduced the batch_size, the problem was fixed.
Is it possible to use Nvbit with MLPerf docker image? I tried to run
Has this been tested before?