fix NER output; fix BERT Tokenizer

5 years ago · 3eabe5054a
parent 9bf2e6f51b
commit 3eabe5054a
1 changed files with 1 additions and 1 deletions
--- a/qurator/sbb_ner/webapp/app.py
+++ b/qurator/sbb_ner/webapp/app.py
@ -271,7 +271,7 @@ def ner(model_id):
            if token == '[UNK]':
                orig_pos = len("".join([pred['word'] for pred in output_sentence]))

-                output_sentence.append({'word': original_text[orig_pos], 'prediction': 'O'})
+                output_sentence.append({'word': original_text[orig_pos], 'prediction': last_prediction})
                continue

            token = token[2:] if token.startswith('##') else token