fairseq_cli/train.py · 0a70c5c505452b7fa2bd7feb1b1e6f6d7f8808c4 · xuchen / Fairseq-S2T · GitLab

Blame History Permalink

I valid the results of embedding norm and no scale embedding for speech-to-text encoder. · d946bc3b
```
Yeah, it is better.
```
xuchen committed May 27, 2022
d946bc3b

train.py 17.6 KB