19 lines (13 loc) · 587 Bytes

Code for NEZHA

Data

Data can be downloaded from https://github.com/HonghuiBao2000/LETTER/tree/master/data

Implementation

Key components of NEZHA are included in models/hack.py:

MTP_Head for NEZHA architectures;
mtp_generate for GR inference;
ctr_forward for GR training;

Pipelines: fintune.py for training and test.py for inference.
Data process: within collator.py and dataset.py
The running script is in scripts/ (including the training and test).

Contact

Yejing Wang (email to:wangyejing.wyj@alibaba-inc.com)