About training detail on graspanything++ dataset

[ReTass]

python -m torch.distributed.launch --nproc_per_node=2 --use_env -m train_network_diffusion --dataset grasp-anywhere --dataset-path data/grasp-anything/ --add-file-path data/grasp-anything++/ --description training_grasp_anything++_lgd --use-depth 0 --seen 1 --network lgdm --epochs 1000

Hello, thank you for your excellent work! When I run this command, two 4090 GPUs and no one else is using it, one epoch training time is 5 minutes, but in the verification, it takes nearly 1.5 hours to finish, and the info during training is as follows, is this normal? Is there something wrong?

[andvg3]

Hi @ReTass, it is completely normal as the inference time of diffusion model is extremely slow.

[ReTass]

I got it. Thank you very much for your reply.

你好@ReTass，这是完全正常的，因为扩散模型的推理时间非常慢。 I got it. Thanks so much for your reply!