OpenNMT

Can you give me some advice?

I tried English to Chinese training several times.

The total corpus size of M6 model is 19855197
v1

  1. English is not case converted
    2.train:19656646 val:198551
  2. Align files without adding words

step 20w
r:0.494755170243611
p:0.4779859562907683
f:0.48265058031158203
Model size: 986M

v2

  1. English is not case converted
    2.train:19656646 val:198551
  2. Add word alignment file

step 20w
r:0.4976210266977913
p:0.4767123516687056
f:0.48350180947632854
Model size: 986m

v3

  1. English to lowercase
    2.train:19656646 val:198551
  2. Add word alignment file

step 20w
r:0.4917068697418362
p:0.4751876029938719
f:0.4798988116173338
Model size: 986m

I use transformer model.
Do I need to increase the corpus size and model size?
Is there any way to further improve the model score. Please help me.