Training in progress, step 3000

.ipynb_checkpoints/eval-checkpoint.py

@@ -56,7 +56,7 @@ def normalize_text(text: str) -> str:
     kakasi = kakasi()
     kakasi.setMode('J', 'H') #Convert from kanji to hiragana
     conv = kakasi.getConverter()
-    chars_to_ignore_regex = '[\,\?\!\-\;\:\"\“\%\‘\”\�\—\’\…\–\（\，\[\]\)\(\！]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+    chars_to_ignore_regex = '[\,\?\!\-\;\:\"\“\%\‘\”\�\—\’\…\–\（\，\[\]\)\(\！\/\「\」\『\』]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
 
     text = text.lower()
     # normalize non-standard (stylized) unicode characters

eval.py

@@ -56,7 +56,7 @@ def normalize_text(text: str) -> str:
     kakasi = kakasi()
     kakasi.setMode('J', 'H') #Convert from kanji to hiragana
     conv = kakasi.getConverter()
-    chars_to_ignore_regex = '[\,\?\!\-\;\:\"\“\%\‘\”\�\—\’\…\–\（\，\[\]\)\(\！]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+    chars_to_ignore_regex = '[\,\?\!\-\;\:\"\“\%\‘\”\�\—\’\…\–\（\，\[\]\)\(\！\/\「\」\『\』]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
 
     text = text.lower()
     # normalize non-standard (stylized) unicode characters

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1f4cb898aaf03bd9a4ecfe06f7a7819fb4d6857ead19e101dd0c31279732765
+oid sha256:43c659f9abb94c30da63bde3ee1435e3e4783c4b9d9dc7d2d57150191a20eaaa
 size 1262665777