Views
No views yet
sentence-transformers) y 3 cabezas clasificadoras:macro: "1" o "0"intent: "value" , "method", "other"context: "standalone" o "followup"encoder/heads.ptlabel2id.jsonid2label.jsontrain_config.jsonpip install torch sentence-transformers huggingface-hubpip install -r requirements.txt1from huggingface_hub import snapshot_download
2
3artifact_dir = snapshot_download(repo_id="TU_USUARIO/TU_REPO")
4print(artifact_dir)1from pathlib import Path
2
3from src.serialization.artifacts import load_artifacts
4from src.infer.predict import predict_all_tasks
5
6artifact_dir = Path("RUTA_DESCARGADA_DESDE_SNAPSHOT")
7encoder, multitask_model, _, id2label = load_artifacts(artifact_dir, device="cpu")
8
9text = "quiero pagar mi factura"
10output = predict_all_tasks(
11 text=text,
12 encoder=encoder,
13 multitask_model=multitask_model,
14 id2label=id2label,
15 device="cpu",
16)
17print(output)1{
2 "macro": {"label": "1", "score": 0.97},
3 "intent": {"label": "method", "score": 0.92},
4 "context": {"label": "standalone", "score": 0.88}
5}python -m src.main test --artifact-dir models/artifacts --text "quiero pagar mi factura" --device cpuscore corresponde a la probabilidad de la clase predicha por cada tarea.handler.py y requirements.txt orientados al entorno de despliegue.