FinQwen is an open-source project focused on building high-quality financial large language model (LLM) question-answering systems. It aims to foster "AI + Finance" through community collaboration, targeting developers and researchers interested in financial AI applications. The project provides a financial LLM, a specialized dataset, and evaluation tools to facilitate the development of intelligent financial Q&A.

How It Works

The project centers around a financial LLM, specifically the Tongyi Qianwen financial model, which has been enhanced with an expanded financial vocabulary and trained on 200B tokens of financial data including reports, news, and forums. It supports a 16K context window, extendable to 64K. The system processes structured financial data (10 tables in SQLite) and unstructured text from prospectuses. Evaluation combines recall of key information and semantic similarity using the shibing624/text2vec-base-chinese model.

Quick Start & Requirements

• Model: Access to Tongyi Qianwen financial models via ModelScope or Hugging Face.
• Dataset: Financial dataset including 10 SQLite tables and 80 prospectuses.
• Evaluation: Code available in the ./eval directory.
• Resources: Fine-tuning scripts and technical details are referenced from the Qwen GitHub repository and technique report.

Highlighted Details

• Offers the Tongyi Qianwen financial LLM with 16K context, extendable to 64K.
• Provides a financial dataset comprising 10 SQLite tables and 80 prospectuses.
• Includes evaluation code for assessing recall and semantic similarity of Q&A responses.
• Showcases top-performing solutions from the "2023 Bojin LLM Challenge" and a subsequent learning competition.

Maintenance & Community

The project is community-driven, originating from the "2023 Bojin LLM Challenge." It encourages contributions and participation in ongoing learning competitions on the Tianchi platform. Links to related projects like FinGLM and FinGPT are provided for community engagement.

Licensing & Compatibility

The project resources are primarily for research and exchange, with a disclaimer against commercial use. Commercial use is subject to the licenses of the underlying models, such as the Tongyi Finance model.

Limitations & Caveats

The project's disclaimer advises against commercial use, suggesting users bear legal responsibility if they proceed. Commercial use of the models must adhere to their specific licenses.