shikunpunk/Qwen2.5-3B-GuCheng
shikunpunk/Qwen2.5-3B-GuCheng is a 3.1 billion parameter language model fine-tuned from Qwen2.5-3B-Instruct by shikunpunk. This model specializes in generating modern poetry in the distinctive style of the poet Gu Cheng, capturing his unique imagery, rhythm, and emotional tone. It is optimized for creative text generation, specifically for short, medium, and long-form modern Chinese poetry.
Loading preview...
Model Overview
shikunpunk/Qwen2.5-3B-GuCheng is a 3.1 billion parameter language model, fine-tuned from the Qwen2.5-3B-Instruct base model. Its primary purpose is to generate modern Chinese poetry in the unique style of the renowned poet Gu Cheng.
Key Capabilities
- Gu Cheng Style Poetry Generation: The model has learned Gu Cheng's characteristic imagery (e.g., natural elements like stars, wind, rain, ocean), language rhythm, and emotional depth.
- Variable Length Output: Capable of generating short poems (up to 6 lines), medium-length poems (10-15 lines), and longer poems (20+ lines) based on user prompts.
- Distinctive Poetic Features: Generates text with abundant natural imagery, concise phrasing, use of white space, a blend of imagination and reality, and a tone of tender observation mixed with subtle melancholy.
Training Details
The model was trained using QLoRA (4-bit NF4 quantization) on a dataset of 183 de-duplicated Gu Cheng poems. This method allowed for training on hardware with limited VRAM (e.g., 8GB RTX 4060 Laptop). A high LoRA rank (32) and multiple epochs (8) were used to thoroughly capture the poet's individual linguistic style.
Good For
- Developers and researchers interested in stylized text generation, particularly for poetry.
- Applications requiring creative content generation with a specific artistic voice.
- Exploring fine-tuning techniques for capturing nuanced linguistic styles from limited datasets.