RUPunct_big - самая большая модель из семейства RUPunct. Подходит для большинства задач.
Код инференса:
from transformers import pipeline
from transformers import AutoTokenizer
pt = "RUPunct/RUPunct_big"
tk = AutoTokenizer.from_pretrained(pt, strip_accents=False, add_prefix_space=True)
classifier = pipeline("ner", model=pt, tokenizer=tk, aggregation_strategy="first")
def process_token(token, label):
if label == "LOWER_O":
return token
if label == "LOWER_PERIOD":
return token + "."
if label == "LOWER_COMMA":
return token + ","
if label == "LOWER_QUESTION":
return token + "?"
if label == "LOWER_TIRE":
return token + "—"
if label == "LOWER_DVOETOCHIE":
return token + ":"
if label == "LOWER_VOSKL":
return token + "!"
if label == "LOWER_PERIODCOMMA":
return token + ";"
if label == "LOWER_DEFIS":
return token + "-"
if label == "LOWER_MNOGOTOCHIE":
return token + "..."
if label == "LOWER_QUESTIONVOSKL":
return token + "?!"
if label == "UPPER_O":
return token.capitalize()
if label == "UPPER_PERIOD":
return token.capitalize() + "."
if label == "UPPER_COMMA":
return token.capitalize() + ","
if label == "UPPER_QUESTION":
return token.capitalize() + "?"
if label == "UPPER_TIRE":
return token.capitalize() + " —"
if label == "UPPER_DVOETOCHIE":
return token.capitalize() + ":"
if label == "UPPER_VOSKL":
return token.capitalize() + "!"
if label == "UPPER_PERIODCOMMA":
return token.capitalize() + ";"
if label == "UPPER_DEFIS":
return token.capitalize() + "-"
if label == "UPPER_MNOGOTOCHIE":
return token.capitalize() + "..."
if label == "UPPER_QUESTIONVOSKL":
return token.capitalize() + "?!"
if label == "UPPER_TOTAL_O":
return token.upper()
if label == "UPPER_TOTAL_PERIOD":
return token.upper() + "."
if label == "UPPER_TOTAL_COMMA":
return token.upper() + ","
if label == "UPPER_TOTAL_QUESTION":
return token.upper() + "?"
if label == "UPPER_TOTAL_TIRE":
return token.upper() + " —"
if label == "UPPER_TOTAL_DVOETOCHIE":
return token.upper() + ":"
if label == "UPPER_TOTAL_VOSKL":
return token.upper() + "!"
if label == "UPPER_TOTAL_PERIODCOMMA":
return token.upper() + ";"
if label == "UPPER_TOTAL_DEFIS":
return token.upper() + "-"
if label == "UPPER_TOTAL_MNOGOTOCHIE":
return token.upper() + "..."
if label == "UPPER_TOTAL_QUESTIONVOSKL":
return token.upper() + "?!"
while 1:
input_text = input(":> ")
preds = classifier(input_text)
output = ""
for item in preds:
output += " " + process_token(item['word'].strip(), item['entity_group'])
print(">>>", output)