[Question] Support for Custom Attention Mask

[Peng-YM]

❓ General Questions

Many thanks for your effort to develop such a great library. I want to add support for the ChatGLM (not the 3rd generation) model in mlc-llm, however, it seems to me that currently attention mask is not supported as documented in the following file:

# 3rdparty/tvm/python/tvm/relax/frontend/nn/modules.py:924

assert attention_mask is None, "Attention mask not yet supported."

How can I resolve this issue, sincerely wish to hear from you.