Wallart edited a comment on issue #16754: Is mirroring working with MXNet 1.5.1 Gluon ? URL: https://github.com/apache/incubator-mxnet/issues/16754#issuecomment-561682740 https://github.com/Wallart/transformer-ts-forecasting For execution : ./main.py train synth ~/fakePath 24 -e 400 -b 64 --num-layers 2 --dim-model 32 --dim-ff 128 By default it will start a training using Gluon API. To switch on Module API, you have to paste the following lines in main.py line 131 ``` #trainer = TransformerTrainer(args, net, ctx) trainer = ModuleTrainer(args, net, ctx) ``` in transformer.py line 38 ``` #if autograd.is_training(): # return logits, attn_w_1, attn_w_2 ``` EDIT : Just ran the code on another machine with the default pip installed mxnet and the loss is normal EDIT2 : It works on a V100 but not on GTX 1080Ti (same environment both using docker images)
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
