Open crazilizer opened 1 year ago
micro torch.Size([1, 1, 512, 512]) grad0 None micro torch.Size([1, 1, 512, 512]) grad0 None my command is: python scripts/image_train.py --data_dir /home/dell/guimu/data/dataset/norm --image_size 512 --num_channels 128 --num_res_blocks 2 --num_heads 1 --learn_sigma True --use_scale_shift_norm False --attention_resolutions 16 --diffusion_steps 1000 --noise_schedule linear --rescale_learned_sigmas False --rescale_timesteps False --lr 1e-4 --batch_size 10
Same problem as you, is it normal?
| grad_norm | 0.348 | | loss | 0.384 | | loss_q0 | 0.893 | | loss_q1 | 0.483 | | loss_q2 | 0.0522 | | loss_q3 | 0.00336 | | mse | 0.378 | | mse_q0 | 0.873 | | mse_q1 | 0.479 | | mse_q2 | 0.0517 | | mse_q3 | 0.00332 | | param_norm | 205 | | samples | 3.3e+05 | | step | 3.3e+03 | | vb | 0.00691 | | vb_q0 | 0.0201 | | vb_q1 | 0.00353 | | vb_q2 | 0.000483 | | vb_q3 | 4.19e-05 |
micro torch.Size([1, 1, 512, 512]) grad0 None micro torch.Size([1, 1, 512, 512]) grad0 None my command is: python scripts/image_train.py --data_dir /home/dell/guimu/data/dataset/norm --image_size 512 --num_channels 128 --num_res_blocks 2 --num_heads 1 --learn_sigma True --use_scale_shift_norm False --attention_resolutions 16 --diffusion_steps 1000 --noise_schedule linear --rescale_learned_sigmas False --rescale_timesteps False --lr 1e-4 --batch_size 10