vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy
The vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy is an 8 billion parameter Qwen3-based language model developed by the Vector Institute. Fine-tuned for news media bias detection and neutral rewriting, it identifies biased language, classifies bias types, provides neutral replacements, and rewrites articles into unbiased versions. This model excels at generating structured JSON responses directly, making it suitable for production inference in bias analysis applications.
Loading preview...
Model Overview
The vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy is an 8 billion parameter model, fine-tuned by the Vector Institute as part of the UnBias-Plus project. It is built upon the Qwen3-8B base model and specializes in analyzing news articles for bias.
Key Capabilities
- Bias Detection: Identifies biased language segments within news articles.
- Bias Classification: Classifies detected biases by type (e.g., loaded language, dehumanizing framing, sensationalism) and severity.
- Neutral Rewriting: Provides neutral alternative phrases for biased segments and generates a fully rewritten, unbiased version of the article.
- Structured JSON Output: Designed to produce clean, structured JSON responses directly, without chain-of-thought thinking blocks, optimizing it for faster and more reliable production inference.
- Dedicated Training: Fine-tuned using Supervised Fine-Tuning (SFT) with LoRA on the UnBias-Plus dataset, which contains expert-annotated news articles.
Use Cases and Differentiators
This model is specifically engineered for applications requiring automated news media bias analysis and content neutralization. Its direct JSON output and disabled 'thinking mode' make it highly efficient for deployment with inference engines like vLLM. While primarily trained on English news, it offers a robust solution for identifying and mitigating various forms of bias in textual content, with best performance on articles under 5000 characters. It is particularly suited for scenarios where high recall and handling of longer articles are priorities, distinguishing it from smaller variants.