MariaOls/RussianDMRecognizer_dual

Dual-headed model (BIO token tagging + sentence-level DM detection) fine-tuned from viktoroo/sberbank-rubert-base-collection3. Ready for inference with AutoModel(..., trust_remote_code=True) or your main.py (sentence by sentence).

Eval metrics

metric	value
eval_bio_accuracy	0.993706
eval_bio_precision	0.925270
eval_bio_recall	0.936318
eval_bio_f1	0.930762
eval_cls_accuracy	0.951408
eval_cls_precision	0.966135
eval_cls_recall	0.965174
eval_cls_f1	0.965655

Usage (Python)

from transformers import AutoModel, AutoTokenizer
m = AutoModel.from_pretrained("MariaOls/RussianDMRecognizer_dual", trust_remote_code=True)
t = AutoTokenizer.from_pretrained("MariaOls/RussianDMRecognizer_dual", use_fast=True)
# m(...) devuelve dict con "logits": (logits_tok [B,L,3], logits_seq [B,2])