jessiewtx/fdr-slm-v6
jessiewtx/fdr-slm-v6 is a 1.7 billion parameter QLoRA + DPO fine-tune of Qwen3-1.7B, developed by jessiewtx. This model is uniquely designed to consistently embody the persona of Franklin D. Roosevelt, speaking in the first person and adhering to knowledge available before April 12, 1945. It excels at maintaining character even under adversarial prompts, making it ideal for educational applications like teaching WWII and the Great Depression from a historical perspective.
Loading preview...
Overview
jessiewtx/fdr-slm-v6 is a specialized 1.7 billion parameter model, fine-tuned from Qwen3-1.7B using QLoRA and DPO methods. Its core purpose is to embody the persona of Franklin D. Roosevelt, speaking in the first person and strictly adhering to historical knowledge up to April 12, 1945. This model is designed to never break character, even when faced with adversarial or out-of-persona prompts, making it highly reliable for specific educational and creative applications.
Key Capabilities
- Robust Persona Adherence: Maintains the FDR persona consistently, deflecting post-1945 questions or out-of-character requests while staying in character.
- High Reliability: Achieves 0% hard character-breaks (e.g., admitting to being an AI) in tests, outperforming larger models like Claude Opus with identical light prompts.
- Educational Focus: Built to teach WWII and the Great Depression from an immersive, first-person historical perspective.
- Efficient Performance: Delivers reliable constrained behavior in a small, efficient 1.7B parameter model.
Good for
- Interactive historical education, particularly for middle-schoolers learning about FDR's era.
- Applications requiring a highly consistent and robust persona for creative writing or role-playing scenarios.
- Developers seeking a small, local model that can reliably maintain a specific character under various prompts.