jburnford/dyslexic-writer-qwen3-4b

Overview

The jburnford/dyslexic-writer-qwen3-4b is a 4 billion parameter model built upon the Qwen3 architecture, specifically fine-tuned by jburnford for advanced spelling and grammar correction. Its primary focus is to provide robust assistance to dyslexic writers by accurately identifying and correcting textual errors.

Key Capabilities

• Specialized Correction: Optimized for spelling and grammar correction, with a focus on common errors made by dyslexic writers.
• High Accuracy: Achieves an 85.6% Exact Match Accuracy and an 80.4% Error Fix Rate, while preserving correct text with 99.3% No-Error Preservation.
• Extensive Training: Trained on approximately 495,000 examples, including diverse datasets of word pairs, sentence corrections, and paragraph-level error injection from synthetic stories.
• Flexible Deployment: Available for use with both Ollama (GGUF format) and the Hugging Face Transformers library.

Good For

• Automated Proofreading: Ideal for applications requiring automated spelling and grammar checks.
• Dyslexia Support Tools: Excellent for integrating into tools designed to assist dyslexic individuals with writing.
• High-Quality Correction: This 4B parameter variant offers the best quality correction among the dyslexic-writer-qwen3 series, making it suitable for scenarios where accuracy is paramount.