Spaces:
Sleeping
Sleeping
| #Milo WBA Model (Phi-3 Fine-Tuned) | |
| Version: v1.0 | |
| Purpose: WhatsApp Business Assistant | |
| Base Model: Microsoft Phi-3 Medium 4K | |
| #Features | |
| - Customer service responses (short, polite, WhatsApp tone) | |
| - Delivery information and updates | |
| - Complaint handling with empathy | |
| - Business templates (messages, confirmations) | |
| - Quotes & invoices (PDF-style text outputs) | |
| - Product descriptions (concise and clear) | |
| #Training Details | |
| - Fine-tuning method:LoRA (Low-Rank Adaptation) | |
| - Datasets used: | |
| - Customer service Q&A dataset | |
| - Delivery updates dataset | |
| - Quotes & invoices dataset | |
| - Split: 90% training / 10% validation | |
| - Steps: 2000 training steps with warmup | |
| - Batch size: 2 (gradient accumulation = 8) | |
| - Learning rate: 2e-4 | |
| #Intended Use | |
| Milo AI is designed to: | |
| - Handle WhatsApp business conversations quickly and politely | |
| - Generate structured responses (quotes, invoices, delivery updates) | |
| - Support small businesses with customer engagement | |
| #Limitations | |
| - Not suitable for medical, legal, or financial advice | |
| - Focused on short WhatsApp-style messages, not long-form documents | |
| - May need retraining for niche industries or highly technical queries | |
| #Future Improvements | |
| - Add multilingual support (e.g., isiZulu, Afrikaans, Xhosa) | |
| - Expand datasets for more industries (retail, health, education) | |
| - Integrate richer templates (contracts, receipts, promotions) | |