---
language:
- ru
license: apache-2.0
library_name: transformers
pipeline_tag: text2text-generation
tags:
- keyphrase-generation
- russian
- t5

base_model:
- sberbank-ai/ruT5-base
datasets:
- aglazkova/keyphrase_extraction_russian
model-index:
- name: ruT5 keyphrase generator
  results:
  - task:
      type: text2text-generation
      name: Keyphrase Generation
    dataset:
      name: aglazkova/keyphrase_extraction_russian
      type: aglazkova/keyphrase_extraction_russian
    metrics:
    - type: rougeL
      value: 0.XX
    - type: exact_match
      value: 0.XX
---

# ruT5 Keyphrase Generator (Russian)

**Base model:** `sberbank-ai/ruT5-base`  
**Dataset:** `aglazkova/keyphrase_extraction_russian`  

## Usage
```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tok = AutoTokenizer.from_pretrained("denis-gordeev/rut5-keyphrase-ru", use_fast=False)
mdl = AutoModelForSeq2SeqLM.from_pretrained("denis-gordeev/rut5-keyphrase-ru")

text = "keyphrase: Новая модель обнаруживает аномалии в банковских транзакциях..."
inp = tok(text, return_tensors="pt")
out = mdl.generate(**inp, max_new_tokens=64, num_beams=4, no_repeat_ngram_size=3)
print(tok.decode(out[0], skip_special_tokens=True))