thanhhoangnvbg/empathAI-llama3.1-8b
EmpathAI-llama3.1-8b by thanhhoangnvbg is a fine-tuned Llama 3.1 8B model specifically designed for Vietnamese e-commerce customer support. It excels at handling difficult situations like toxic customers, delayed deliveries, and product issues, aiming to reduce hallucinations and improve realism in customer service interactions. The model is optimized for integration with RAG and tool systems, focusing on policy-aware support and de-escalation.
Loading preview...
EmpathAI-llama3.1-8b: Specialized Vietnamese E-commerce Customer Support
EmpathAI-llama3.1-8b is a Llama 3.1 8B model, fine-tuned by thanhhoangnvbg, dedicated to Vietnamese e-commerce customer support. Its primary goal is to address challenging scenarios such as toxic or angry customers, delayed/missing orders, faulty products, and refund/payment issues, while adhering to real-world policies and workflows.
Key Capabilities
- Emotional Intelligence: Trained to de-escalate angry customers, maintain a natural tone, and provide clear next steps without unnecessary arguments.
- RAG & Tool-Friendly: Designed for seamless integration with RAG pipelines, order lookup systems, and internal customer support tools. It intelligently identifies when more information (like order codes) is needed and avoids hallucinating order statuses or unsupported promises.
- Safety & Grounding Focus: Emphasizes reducing hallucinated order statuses, unauthorized refund/voucher offers, and unnecessary PII requests, enhancing privacy and data safety.
- Specialized Training: Utilizes a two-stage SFT and DPO pipeline, with a dataset specifically curated for toxic customer complaints, delayed delivery, and Vietnamese empathy/de-escalation in e-commerce contexts.
Good For
- Automating and improving Vietnamese customer support for e-commerce platforms.
- Handling and de-escalating toxic customer interactions effectively.
- Integrating with existing RAG and tool systems for policy-aware responses.
- Reducing hallucinations in customer service, particularly regarding order status and compensation.
This model is not intended for general reasoning, coding assistance, or as a general-purpose chatbot, but rather for highly specialized e-commerce customer support workflows.