Hugging Face Model Overview
hiig-ai-lab/simba_best_092024 is an 8 billion parameter German language model developed by the Public Interest AI research group at HIIG Berlin. It is fine-tuned from meta-llama/Meta-Llama-3-8B-Instruct with approximately 800 German newspaper articles that were simplified by the Austrian Press Agency. The primary goal of this model is to simplify German-language text, making it more understandable for a broader audience.
Key Capabilities
- German Text Simplification: Specializes in simplifying German newspaper articles to an A2 language level.
- Text Generation: Functions as a text generation model, producing simplified versions of input texts.
- Instruction Following: Utilizes a specific prompt structure for inference, guiding the model to summarize and simplify.
Use Cases
- Direct Simplification: Best suited for simplifying German-language newspaper articles (news items, not commentaries or editorials).
- Accessibility: Can be used to make complex German texts more accessible.
Limitations and Recommendations
As with many text generation models, simba_best_092024 may occasionally produce incorrect information. Users are advised to manually verify the output text against the input to ensure factual consistency. While primarily trained on newspaper articles, its capabilities might be extended by fine-tuning on more diverse parallel text datasets (standard German and simplified German).