My model is taking forever (about 45-60 min) to complete an epoch (which I think is due to the long sequences) for a daily simple (baseline) network. I haven't even added "bells and whistles". So I want to find out what an ideal epoch execution time is. I am afraid I cannot even train larger and better networks with this execution time
With 5 epochs, I start at 30.0% accuracy and end at 29.9% (basically, no learning).
Is my architecture wrong? Am I calculating accuracy wrongly? I will appreciate any help.
Thanks.
What model architecture you are using? Try using
1 embed -> 1 LSTM -> 2 Linear (relu and softmax)
Try TPU in colab. In GPU, the running time is very
Thanks for your assistance. I will try that.
However, your responses seem to be incomplete. Your initial reply said "try using..." this one says the running time is very..." both incomplete
I think you should check that. Thanks
Sorry I don't know why the last word got cutoff. I meant that running time in GPU is very high. Try using colab tpu.