You can deploy this model and start earning money today!

DanielNRU/
S7-relevance-rubert-20260823

Name: S7-relevance-rubert-20260823
Availability: InStock

Views

No views yet

S7-relevance-rubert-20260823

Бинарная классификация релевантности русскоязычных текстов для домена S7.

Model details

Base model: ai-forever/ruRoBERTa-large
Architecture: RuBERTBinaryClassifier
Best validation F1: 0.978417
Test F1: 0.971429
Test precision: 0.951049
Test recall: 0.992701
Decision threshold: 0.5
Maximum sequence length: 512
Training class ratio (relevant:irrelevant): 3.0:1

Usage

python

from huggingface_hub import hf_hub_download
from transformers import AutoTokenizer
import torch


repo = "DanielNRU/S7-relevance-rubert-20260823"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = RuBERTBinaryClassifier("ai-forever/ruRoBERTa-large", hidden_size=512)
weights = hf_hub_download(repo_id=repo, filename="model.safetensors" if (output_dir / "model.safetensors").exists() else "pytorch_model.bin")
model.load_state_dict(torch.load(weights, map_location="cpu"))
model.eval()

Limitations

Модель предназначена для S7-домена. При изменении распределения данных рекомендуется повторно проверить threshold и метрики.