Harry He
Harry He
## 🐛 Bug I met this bug when I try to load a fairseq translation model. I could see the same issue is still open in fairseq GitHub repo. I...
## ✨ Description This PR aims to enhance the compatibility and efficiency of the librilight preprocessing scripts with the new NS2 framework to be updated by @HeCheng0625. By updating the...
首先感谢作者的复现,但是文中的evaluation最后用的micro-averaged F1 score, 而非macro-averaged。 micro -averaged F1==accuracy,会看起来比macro-average的高很多,实际的F1远不到80%