search

maikherbig / AIDeveloper

GUI-based software for training, evaluating and applying deep neural nets for image classification
BSD 2-Clause "Simplified" License
111 stars 20 forks source link

Being able to use multiple GPUs in AIDeveloper #8

Open alfrem opened 4 years ago

alfrem commented 4 years ago

Would be good to have the option to train networks using multiple GPUs as a possibility:

https://www.tensorflow.org/api_docs/python/tf/keras/utils/multi_gpu_model

or newest: https://www.tensorflow.org/api_docs/python/tf/distribute/MirroredStrategy

With the different distribution strategies.

Thank you for your great job.

maikherbig commented 4 years ago

Hi, I think this should already work since multi_gpu_model has to be used when defining the model. Models are defined in a separate script called model_zoo.py. You can modify this the model_zoo and include your own model which uses multi_gpu_model. I recorded a tutorial video showing how to add models to the model zoo (Tutorial 4): https://figshare.com/articles/Krater_et_al_2020_Data_zip/9902636 I hope this helps!

alfrem commented 4 years ago

I will try as soon as I have access to new multiple GPU server. Should be in 2 weeks :). Thank you!

alfrem commented 4 years ago

By the way I didn't realise there was a 4th tutorial. Thanks!