提交 9b86f200 编写于 作者: L lifuchen

fix a bug of transformertts when use data parallel.

上级 6428ce54
......@@ -94,6 +94,12 @@ def main(args):
if args.stop_token:
writer.add_scalar('stop_loss', stop_loss.numpy(), global_step)
if args.use_data_parallel:
writer.add_scalars('alphas', {
'encoder_alpha':model._layers.encoder.alpha.numpy(),
'decoder_alpha':model._layers.decoder.alpha.numpy(),
}, global_step)
else:
writer.add_scalars('alphas', {
'encoder_alpha':model.encoder.alpha.numpy(),
'decoder_alpha':model.decoder.alpha.numpy(),
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册