AutonomoV3

SlotExtractor_MDM_Autonomo

Modelo BERT en español fine-tuned para clasificación de intenciones.

🧠 Arquitectura

• Base: dccuchile/bert-base-spanish-wwm-cased
• Idioma: Español

📊 Métricas

• Accuracy: 0.967
• Macro F1: 0.967

🔤 Intenciones (27)

• 0 → add_reminder
• 1 → analyze_image
• 2 → archive_trello_cards
• 3 → budget_control
• 4 → contact_with_live_agent
• 5 → create_image
• 6 → create_lead
• 7 → create_music
• 8 → delete_reminder
• 9 → get_drive_file
• 10 → get_information
• 11 → get_map
• 12 → get_reminder
• 13 → get_security_code
• 14 → get_trello_cards
• 15 → list_filenames
• 16 → make_call
• 17 → routine_tracking
• 18 → search_filtered_exercises
• 19 → search_information
• 20 → search_products
• 21 → send_email_attachment
• 22 → send_gmail
• 23 → simulate_operations
• 24 → start_conversation
• 25 → upload_drive_file
• 26 → upload_image

⚙️ Uso rápido

from transformers import AutoTokenizer, AutoModelForSequenceClassification
MODEL_ID = 'ArmandoRockYourCloud/SlotExtractor_MDM_Autonomo'

tok = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
id2label = model.config.id2label

text = 'Agrega un recordatorio para mañana a las 9'
enc = tok(text, return_tensors='pt', truncation=True, max_length=128)
with torch.no_grad():
    logits = model(**enc).logits
pred = logits.argmax(-1).item()
print(id2label[str(pred)])

🧾 Notas

• config.json incluye los mapeos id2label y label2id.
• labels.json se incluye como referencia adicional.