jason

Hello, I met the same problem too when I ran "bash scripts/release/dsnet/train_dsnet_slurm_dist_ii.sh". File "/mnt/cache/gongjunchao/workdir/DS-Net/network/modules/pytorch_meanshift.py", line 92, in calc_shifted_matrix_flat_kernel_bandwidth_weight new_X = torch.sum(torch.stack(new_X_list), dim=0) / torch.sum(weights, dim=1).view(-1) (function _print_stack) RuntimeError: Function 'DivBackward0'...

Hello, I found this bug is caused by numerical error between operator "**" and "torch.mm" in my case. To fix it, I changed function "pariwise_distance" in network/loss/instance_losses.py.  I hope...

### base config ### full_field: &FULL_FIELD loss: 'l2' lr: 1E-3 scheduler: 'ReduceLROnPlateau' num_data_workers: 4 dt: 1 # how many timesteps ahead the model will predict n_history: 0 #how many previous...

I have the same problem.