THUDM / GLM

GLM (General Language Model)

The attention_mask dimension not right？

zhangyipin opened this issue a year ago · comments

yiy commented a year ago

GLM/tasks/seq2seq/dataset.py

Line 856 in b8ea51c

attention_mask[self.max_src_length:, self.max_src_length:] = np.tril(

should output attention mask dimension 3 but 2.
just like

Zhengxiao Du commented a year ago

Thank you for pointing that out. Already fixed.