When I run train.py, train_loss becomes nan. Is this problem normal?

[boxbox2]

class carpet
args1.json defaultdict(<class 'str'>, {'img_size': [256, 256], 'Batch_Size': 4, 'EPOCHS': 3000, 'T': 1000, 'base_channels': 128, 'beta_schedule': 'linear', 'loss_type': 'l2', 'diffusion_lr': 0.0001, 'seg_lr': 1e-05, 'random_slice': True, 'weight_decay': 0.0, 'save_imgs': True, 'save_vids': False, 'dropout': 0, 'attention_resolutions': '32,16,8', 'num_heads': 4, 'num_head_channels': -1, 'noise_fn': 'gauss', 'channels': 3, 'mvtec_root_path': 'datasets/mvtec', 'visa_root_path': 'datasets/VisA/visa', 'dagm_root_path': 'datasets/dagm', 'mpdd_root_path': 'datasets/mpdd', 'anomaly_source_path': 'datasets/dtd', 'noisier_t_range': 600, 'less_t_range': 300, 'condition_w': 1, 'eval_normal_t': 200, 'eval_noisier_t': 400, 'output_path': 'outputs', 'arg_num': '1'})
Epoch:0, Train loss: nan:

train_loss increases by two points from 2.03 to 4.23, sometimes becoming nan when it increases to 10, and sometimes becoming nan when it reaches 20

[boxbox2]

[HuiZhang0812]

Please see #29 for more details.

[HuiZhang0812]

When the batch size is small, it may lead to an entire batch consisting solely of abnormal samples, thereby affecting the calculation of the paper's loss formula (Formula 9). We have fixed this. We have fixed this bug. Please try again.