Model Card for Beyond Reality

Model Details

Basic Information

• Model Type: Language Model
• Base Model: Qwen/Qwen3-8B
• Training Type: Fine-tuned
• Version: 1.0
• Language(s): English

Model Architecture

• Architecture: Qwen/Qwen3-8B
• Parameters: 8 billion
• Training Procedure: Fine-tuned on a custom dataset of interactive fiction scenarios

Intended Use

• Primary intended uses: Interactive storytelling, text-based adventure games, narrative exploration
• Primary intended users: Game developers, writers, AI researchers, interactive fiction enthusiasts

Limitations and Bias

• Limited to 5-6 coherent actions in sequence before potential degradation
• May exhibit biases present in the original LLaMA model and the fine-tuning dataset
• Not suitable for factual information retrieval or real-world decision making

Training Data

• Multi-choice action systems (options A-D)
• Custom user-defined actions (E+)
• Various narrative genres and settings

Performance and Evaluation

• Maintains coherence for 5-6 sequential actions on average
• Evaluated primarily through user testing and narrative consistency
• Already starting as a small model, quantization noticeably reduces the ability of the model to follow the multiple-choice narrative

Ethical Considerations

• Model outputs are fictional and should not be used as a source of factual information
• Users should be aware of potential biases in generated content