There is no checkpoint output,

[zhangbaijin]

Thanks for your contribution, and after train, i get nothing, no checkpoint, why?

`----------------------------
| grad_norm | 0.00858 |
| lg_loss_scale | 20.9 |
| loss | 0.00124 |
| loss_q0 | 0.00374 |
| loss_q1 | 0.000934 |
| loss_q2 | 0.00063 |
| loss_q3 | 0.000106 |
| mse | 0.00124 |
| mse_q0 | 0.00374 |
| mse_q1 | 0.000934 |
| mse_q2 | 0.00063 |
| mse_q3 | 0.000106 |
| param_norm | 240 |
| samples | 4e+05 |
| step | 5e+04 |

saving model 0...
saving model 0.9999...`

[dillfrescott]

Take a look at how I'm training mine, I'm getting a model output with this configuration...

!LOGDIR="OUTPUT/sinddpm-yourimage-day-commitseq" python image_train.py --data_dir /content/upscaled2.png --lr 5e-4 --diffusion_steps 1000 --image_size 256 \
                                   --noise_schedule linear --num_channels 64 --num_head_channels 16 --channel_mult "1,2,4" \
                                   --attention_resolutions "2" --num_res_blocks 1 --resblock_updown False --use_fp16 True \
                                   --use_scale_shift_norm True --use_checkpoint True --batch_size 16

Hope this helps!

[zhangbaijin]

It seems that bash: !LOGDIR="OUTPUT/sinddpm: event not found. Is that true?

[zhangbaijin]

It has been sovled, thanks for a lot

[dillfrescott]

No problem

[cwy08090014]

@zhangbaijin May I ask how did this problem solve? I meet with the same problem. Thanks.

[KevinWang676]

Hi @dillfrescott, I used the same code as yours and the size of my input image is 256x256. But I got an error called RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 54 but got size 53 for tensor number 1 in the list. Could you help me resolve this issue? Thank you in advance!

[KevinWang676]

Hi @dillfrescott, I resolved the problem just now by adding \ after each line. But I wonder if you know how many steps the training process takes in total. Thank you.

[dillfrescott]

@KevinWang676 It's been a while since ive used this project. I'm glad you were able to fix it though but I do not remember how many steps the training process takes in total.

[Shaohua987]

It has been sovled, thanks for a lot

Hello, may I ask how many steps did you train to complete the training？