Loss goes to Zero in first epoch!

vision
1

Hi,

I am getting a weird case in which my loss just vanishes within an epoch:


Train Epoch: 1 [0/33005 (0%)] Loss: 84108.492188

Train Epoch: 1 [32000/33005 (97%)] Loss: 0.000000

Epoch: 2 LR: [0.001]

Starting Training for Epoch 2

Train Epoch: 2 [0/33005 (0%)] Loss: 0.000000

Train Epoch: 2 [32000/33005 (97%)] Loss: 0.000000

Epoch: 3 LR: [0.001]

Starting Training for Epoch 3

Train Epoch: 3 [0/33005 (0%)] Loss: 0.000000

I am at loss for what is the issue, I tried different random initializations as well, same thing happens.

Any common tips on debugging the model?

2

Can you print your output to check if there are some unusual things or paste it to discuss？

3

Can you post model architecture. And also the training loop