search

forever208 / DDPM-IP

[ICML 2023] official implementation for "Input Perturbation Reduces Exposure Bias in Diffusion Models"
MIT License
101 stars 9 forks source link

TypeError: forward() missing l required positional argument: 'timesteps' #10

Closed KevinWang676 closed 1 year ago

KevinWang676 commented 1 year ago

Hi, when I tried to train a model myself, I got TypeError: forward() missing l required positional argument: 'timesteps'. I tried both multi-node training and single gpu training, but I got the same error in both cases. Could you help me resolve the issue? Thank you!

Here is my code: mpiexec -n 2 python scripts/image_train.py --input_pertub 0.15 \ --data_dir cifar_train \ --image_size 32 --use_fp16 True --num_channels 128 --num_head_channels 32 --num_res_blocks 3 \ --attention_resolutions 16,8 --resblock_updown True --use_new_attention_order True \ --learn_sigma True --dropout 0.3 --diffusion_steps 1000 --noise_schedule cosine --use_scale_shift_norm True \ --rescale_learned_sigmas True --schedule_sampler loss-second-moment --lr 1e-4 --batch_size 64

5c8f2142136b7be396bdef33246b495

KevinWang676 commented 1 year ago

I figured it out. I need to run pip install torch-summary first. Thanks.