tsilva/qwen2.5-3b-trump-style-merged-v1
The tsilva/qwen2.5-3b-trump-style-merged-v1 is a 3 billion parameter Qwen2.5-based causal language model, fine-tuned by tsilva to generate text in a brash, punchy, high-confidence public-speaking style associated with Donald Trump. This merged 16-bit model is specifically designed for style-following chat experiments and evaluation, aiming to reproduce a distinct rhetorical pattern. It is optimized for direct Transformer loading and is part of a versioned style-tuning release.
Loading preview...
Model Overview
This model, tsilva/qwen2.5-3b-trump-style-merged-v1, is a 3 billion parameter Qwen2.5-based language model fine-tuned by tsilva. It is specifically designed to generate text in a brash, punchy, high-confidence public-speaking style reminiscent of Donald Trump. The model was created using an llmstyler Runbook training job, resulting in a merged 16-bit model artifact.
Key Capabilities & Differentiators
- Distinct Style Emulation: The primary feature is its ability to adopt a specific public-speaking style, characterized by short, emphatic sentences, simple strong wording, conversational asides, repetition for emphasis, and frequent superlatives.
- Base Model: Built upon
unsloth/Qwen2.5-3B-Instruct-bnb-4bit, providing a solid foundation for instruction-following before style application. - Training Details: Trained for 1 epoch with a max sequence length of 2048, using a LoRA rank of 32 and alpha of 32, on the
tsilva/stylemix_trump-v1dataset. - Artifact Versatility: Available as a merged 16-bit model for direct Transformer loading, with companion QLoRA adapter, GGUF, and ONNX exports for various deployment scenarios.
Intended Use Cases
- Style-Following Chat Experiments: Ideal for research and development in controlled style generation.
- Evaluation: Useful for assessing the strength and consistency of style transfer in language models.
- Creative Content Generation: Can be used to generate text with a very specific, recognizable rhetorical flair.
Limitations
Users should be aware that the model may over-apply the target style, potentially leading to a loss of factual nuance. It may also reproduce limitations inherited from its base model and training data. Thorough evaluation of task accuracy, safety behavior, refusal behavior, and style strength is recommended before deployment.