suna999/inquiry_merged_model
The suna999/inquiry_merged_model is a 2.5 billion parameter language model, fine-tuned from Google's Gemma 2B-IT using LoRA. It specializes in generating responses for civil complaint Q&A, trained on 2,000 samples of private civil complaint data. This model is designed for standalone inference, having fully merged its LoRA adapters with the base model, and is optimized for customer service applications.
Loading preview...
Model Overview
The suna999/inquiry_merged_model is a 2.5 billion parameter language model developed by suna999, based on Google's Gemma 2B-IT architecture. It has been specifically fine-tuned using LoRA (Low-Rank Adaptation) to excel at generating responses for civil complaint Q&A scenarios. The model is fully merged, allowing for standalone inference without requiring the base Gemma model.
Key Capabilities
- Specialized Q&A Generation: Optimized for handling and responding to civil complaint inquiries.
- Financial/Customer Service Focus: Trained on private civil complaint data from financial institutions (Hana Card, Activenture, LG Uplus).
- Efficient Inference: The LoRA adapters are fully merged with the base model, resulting in a 4.7GB model capable of independent operation.
Training Details
The model was fine-tuned on a dataset of 2,000 Q&A samples sourced from the AI Hub's private civil complaint LLM pre-training data. This dataset consists of masked Q&A dialogues between counselors and customers. The fine-tuning process involved 2 epochs, a batch size of 1, a learning rate of 2e-4, and LoRA rank/alpha settings of 32/64 respectively.
Good For
- Automated Customer Support: Ideal for building systems that automatically answer common civil complaint questions.
- Financial Services: Particularly well-suited for applications within banking, credit card services, and telecommunications due to its training data origin.
- Rapid Deployment: Its merged nature simplifies deployment, as it functions as a complete model.