Tutorial 1 training still get nan loss

[xiaogangzhu]

Hi,
for short tutorial 1 I still get training loss == nan

[xiaogangzhu]

And also Tutorial 3: Semi-parametric extensions to TARNet get nan loss.

Why get nan loss? Is the tensorflow version or GPU problem?

[kochbj]

Thanks for letting me know. I'll look into it as soon as I get the chance!

Best,
Bernie

[kochbj]

Hi Xiaogang,

I just tried the short tutorial 1 again and get no issues, using either CPU or the T4 GPU on collab. In general, using GPUs is probably slower than using CPU for these small networks anyways. Are you not running it on collab?

Best,
Bernie

[xiaogangzhu]

I tried different version of tensorflow and find that tensorflow>2.10 will get the error. I use tensorflow=2.10.0 and it is fine now!