suicao / coleridge-gpt

Part of 1st place solution for Coleridge Initiative - Show US the Data.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Loss is too high, and no result output by infer.py

PubMedKG opened this issue · comments

commented

image
image

Hi suicao,

I want cite your code for dataset NER, but I don't know why loss is too high, and no result output by infer.py.

I have tried the code for 5 times at least. Could you please check the code?

Thank you very much.

Sorry for the late reply, have you fixed the issue?
It seems to me the p_mask parameter in the forward function of the model messed with the cross entropy loss, you can send p_mask = None from the training loop and the loss would converge.

commented

Sorry for the late reply, have you fixed the issue? It seems to me the p_mask parameter in the forward function of the model messed with the cross entropy loss, you can send p_mask = None from the training loop and the loss would converge.

Oh, thank you, it works for me. You are really a cool man!