Модель для распознавания именованных сущностей (NER) в спортивных текстах на русском языке. Разработана для проекта Sculptor Sport.
1from transformers import pipeline
2
3ner_pipeline = pipeline(
4 "ner",
5 model="ZPM07/sculptor_NER",
6 aggregation_strategy="first",
7 device=0
8)
9
10text = "3 подхода по 10 приседаний с весом 50 кг"
11entities = ner_pipeline(text)
12
13def clean_results(entities):
14 cleaned = []
15 for entity in entities:
16 word = entity['word'].replace('Ġ', ' ').strip()
17 if word:
18 cleaned.append({
19 'word': word,
20 'entity': entity['entity_group'],
21 'score': round(entity['score'], 4),
22 'start': entity['start'],
23 'end': entity['end']
24 })
25 return cleaned
26
27cleaned_entities = clean_results(entities)
28print("Найденные сущности:")
29for entity in cleaned_entities:
30 print(f"- {entity['word']} -> {entity['entity']} (доверие: {entity['score']:.2f}, позиция: {entity['start']}-{entity['end']})")
