empty line hallucinations fix

app.py

@@ -40,7 +40,9 @@ else:
     lang2id['isv_Cyrl'] = tokenizer(['isv_Cyrl'], add_special_tokens=False)['input_ids'][0][0]
             
     def translate(text, from_, to_):
-        inputs = tokenizer(text.split('\n'), return_tensors="pt", padding=True)
+        # empty line hallucinations fix
+        lines = [f'{line} ' for line text.split('\n')]
+        inputs = tokenizer(lines, return_tensors="pt", padding=True)
         inputs['input_ids'][:, 0] = lang2id[LANGS[from_]]
         translated_tokens = model.generate(**inputs, max_length=400, forced_bos_token_id=lang2id[LANGS[to_]])
         result = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)