Overview

This model, tegana/qwen2.5-arabic-finance-news-parser, is a specialized fine-tuned version of the Qwen/Qwen2.5-1.5B-Instruct base model. It leverages LoRA fine-tuning (rank 64, all targets) using LLaMA-Factory to achieve its specific capabilities. The model was trained on a dataset of approximately 2,792 Egyptian stock-market news articles, with a maximum sequence length of 3,500 tokens, over 3 epochs.

Key Capabilities

• Structured Information Extraction: Given an Arabic financial news article and a JSON output schema, the model extracts predefined structured data.
• Financial Data Parsing: Capable of identifying and extracting entities like company names, event types (e.g., earnings, acquisition, dividends), sentiment (positive, negative, neutral), impact levels, and short summaries.
• Arabic Language Specialization: Specifically trained on Egyptian stock-market news, making it highly proficient in this domain and language dialect.

Use Cases

• Automated Financial News Analysis: Ideal for systems requiring automated parsing of Arabic financial news to extract key data points.
• Market Intelligence: Can be used to quickly process large volumes of news for insights into company events, market sentiment, and financial figures.

Limitations

• Primarily optimized for Egyptian stock-market news; performance may vary with other Arabic financial dialects.
• Numerical extraction quality is dependent on the clarity of figures in the source text.