WaseemGh98/bankassist-qwen3
BankAssist Qwen3 is a 0.8 billion parameter instruction-tuned model developed by Waseem Ghanem, fine-tuned from Qwen/Qwen3-0.6B. This model specializes in banking intent classification, accurately identifying one of 77 banking intent labels from customer messages. It processes banking queries as a text generation task, providing a standalone solution for direct integration without requiring separate LoRA adapters.
Loading preview...
Model Overview
BankAssist Qwen3 is a specialized language model developed by Waseem Ghanem, fine-tuned from the Qwen/Qwen3-0.6B base model. It is designed for banking intent classification, taking customer banking messages and generating one of 77 predefined intent labels from the Banking77 dataset. This model treats intent classification as a text generation task, making it distinct from traditional classification head approaches.
Key Capabilities & Features
- Banking Intent Classification: Accurately identifies 77 distinct banking intents from customer text inputs.
- Standalone Model: Unlike LoRA adapters, this version has the fine-tuned weights merged into the base model, allowing direct loading with Hugging Face Transformers without
PeftModel. - Improved Accuracy: Achieved a significant increase in accuracy from 22.07% to 47.79% on the Banking77 test set compared to the base Qwen3-0.6B model.
- Efficient Fine-tuning: Trained using LoRA with PEFT and TRL's
SFTTraineron limited hardware (4 GB VRAM), demonstrating effective resource management through techniques like gradient accumulation and FP16 precision.
Use Cases
This model is ideal for developers needing a dedicated solution for automating the categorization of banking customer inquiries. It can be integrated into customer support systems to route queries, provide quick responses, or analyze customer needs based on their stated intent. While experimental, it offers a strong foundation for intent recognition in banking contexts.