You can deploy this model and start earning money today!

cvalenciaunivalle/
ner-prostata-beto

dccuchile/bert-base-spanish-wwm-cased

finetune

cc-by-nc-sa-4.0

Views

No views yet

NER próstata BETO (cancer de próstata)

Fine-tuning de BETO (dccuchile/bert-base-spanish-wwm-cased) sobre el corpus clínico de cáncer de próstata. Tarea: NER (Token Classification) para extraer entidades médicas tipo: EDAD, BIOMARCADOR, CIRUGIA, DIAGNOSTICO, IMAGEN, EXAMEN, ANTECEDENTE, LABORATORIO, TRATAMIENTO, etc.

Subfolders (batch sizes)

Batch	F1	Precision	Recall	Épocas
bs8 ⭐	0.9815	0.9785	0.9844	6
bs16	0.9770	0.9751	0.9789	6
bs32	0.9705	0.9663	0.9747	6

Entrenamiento con EarlyStoppingCallback(patience=2) sobre la métrica f1.

Uso

python

from transformers import AutoModelForTokenClassification, AutoTokenizer

model = AutoModelForTokenClassification.from_pretrained(
    "cvalenciaunivalle/ner-prostata-beto",
    subfolder="bs8"
)
tokenizer = AutoTokenizer.from_pretrained(
    "cvalenciaunivalle/ner-prostata-beto",
    subfolder="bs8"
)

# Inferencia
from transformers import pipeline
nlp = pipeline("ner", model=model, tokenizer=tokenizer, aggregation_strategy="simple")
nlp("Paciente masculino de 72 años con adenocarcinoma de próstata, PSA 9.9 ng/dL.")

Detalles

Modelo base: dccuchile/bert-base-spanish-wwm-cased (110M parámetros)
Hardware: Kaggle GPU P100 (16GB)
Tarea: Maestría en Ingeniería de Sistemas — Univalle, PLN
Compañeros: Laura Valero, Cristian Valencia