GanDLF train bug

[ibrahimethemhamamci]

When we installed GanDLF with the instructions given in github, cpu version of pytorch was installed. When we tried to train the model by using cpu, training gave an expand dims error. After we installed gpu version pytorch, it gives "BrokenPipeError: [Errno 32] Broken pipe" error. Therefore, we wanted to ask how we can fix this problem.
@sarthakpati

[sarthakpati]

Can you post the headers of the CSV and the full YAML configuration?

[ibrahimethemhamamci]

We solved that problem but we faced another problem. The yaml file is:
model.zip

The header of the CSV is:

The other error is:

/pytorch/aten/src/ATen/native/cuda/Loss.cu:111: operator(): block: [11,0,0], thread: [63,0,0] Assertion input_val >= zero && input_val <= one failed.
Looping over training data: 0%|▏ | 7/8000 [00:00<04:13, 31.50it/s]
Traceback (most recent call last):
File "gandlf_run", line 79, in
args.data, args.config, args.output, args.train, args.device, args.reset_prv
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/cli/main_run.py", line 88, in main_run
reset_prev=reset_prev,
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/training_manager.py", line 242, in TrainingManager
testing_data=testingData,
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/training_loop.py", line 809, in training_loop
model, train_dataloader, optimizer, params
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/training_loop.py", line 176, in train_network
loss, calculated_metrics, _ = step(model, image, label, params)
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/training_loop.py", line 94, in step
loss, metric_output = get_loss_and_metrics(label, output, params)
File "/home/nephila/personal_projects/ibrahimethem/GandALF/GaNDLF/GANDLF/parameterParsing.py", line 477, in get_loss_and_metrics
metric_function(predicted, ground_truth, params).cpu().data.item()
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

Thank you in advance

[sarthakpati]

Difficult to know what's going on. Can you try this to give a meaningful stack trace: https://stackoverflow.com/a/51701698/1228757

Also, if you have 0 in your class_list, I would recommend adding ignore_label_validation as well [ref] to obtain accuracte validation scores.