How long does it take to train Reuters on your machine?

Roche / BalancedLossNLP

Other

128 stars 25 forks source link

How long does it take to train Reuters on your machine? #15

Open jinweijoyce opened 3 months ago

jinweijoyce commented 3 months ago

hi, I think I followed the instructions to run train.py with DBloss, but the progress bar is not moving on both local (CPU) and colab T4 GPU, so I wonder what I can do to investigate and how long it takes to run 1 epoch if everything it correct? Thank you! Epoch: 0%| | 0/40 [00:00<?, ?it/s]

blessu commented 2 months ago

Hi @jinweijoyce, as mentioned in the paper - "For Reuters-21578 dataset we use one-GPU (V100) experiments which takes 5 minutes for one epoch". For your question, training Reuters data with DBloss on the Colab T4 GPU would take ~11 minutes per epoch. Epoch: 48%|████▊ | 19/40 [3:26:46<3:49:02, 654.41s/it]