search

BUTSpeechFIT / VBx

Variational Bayes HMM over x-vectors diarization
252 stars 57 forks source link

Number of CPU cores as argument #41

Closed lucgeo closed 2 years ago

lucgeo commented 2 years ago

Hello,

I noticed that "run_example.sh" uses all the cores of the machine. I looked through the code, but it doesn't seem very obvious to me how I could give as a parameter the maximum number of cores that will be used for "VBx/predict.py" and "VBx/vbhmm.py". I'm interested in this to evaluate the processing time related to the available hardware resources. So, is there a way to set the number of CPUs that will be used?

Thanks!

fnlandini commented 2 years ago

Hello, I believe the code runs both predict.py and vbhmm.py on single cores when you run in CPU. But, to enforce the code to run on a specific core, you could try something like taskset --cpu-list 0 ./run_example.sh I hope this helps but let me know if I did not understand the question properly.

lucgeo commented 2 years ago

Hi,

Thanks for your answer! Yes, your solution helps, that's what I'm looking for. But I was hoping that I can do that directly from the source code. In "htop" it looks like all the cores are busy when I run on CPU only.

fnlandini commented 2 years ago

I actually tried looking at htop before answering and I could see that only one CPU was used. If I did not run it with taskset, then, if the job lasted long enough, it would be moved around (which is expected as the OS tries to utilize the CPUs evenly). Perhaps, you have other tasks running that consume a significant amount of CPU and that is why you see many cores been utilized? I tried it out without running other heavy tasks and observed a single core with ~100% utilization (the one running the example).

I understand that it would be convenient to have something in the code to limit the number of cores but it is not our priority so I am afraid I will not be able to implement that. However, if you want to work on that, feel free to create a pull request, it might be useful for other users as well :)

fnlandini commented 2 years ago

I am closing the issue but feel free to reopen if necessary.