[s2s] distill: --normalize_hidden --supervise_forward (#6834)

2020-09-04 14:05:56 -04:00
parent c5d43a872f
commit 6078b12098
3 changed files with 45 additions and 18 deletions
--- a/examples/seq2seq/test_seq2seq_examples.py
+++ b/examples/seq2seq/test_seq2seq_examples.py
@@ -31,6 +31,8 @@ logging.basicConfig(level=logging.DEBUG)
 logger = logging.getLogger()
 CUDA_AVAILABLE = torch.cuda.is_available()
 CHEAP_ARGS = {
+    "supervise_forward": True,
+    "normalize_hidden": True,
    "label_smoothing": 0.2,
    "eval_beams": 1,
    "val_metric": "loss",