Junxian He

icon indicating a website link jxhe.github.io

icon company The Hong Kong University of Science and Technology icon location Assistant Professor@HKUST. PhD@CMU LTI. Working on NLP/ML.

Results 2 issues of Junxian He

overfitting when template.txt has duplicate examples

1 comment

overflow of int16

1 comment

## 🐛 Bug The `int16` dtype (when `--dstore-fp16` is activated) will cause overflow issue since int16 cannot handle all integer ids in a large vocab like Wikitext-103. This will make...

bug