vi-c0de/gemmaiku-3-270m-it-experimental
vi-c0de/gemmaiku-3-270m-it-experimental is an experimental 0.3 billion parameter instruction-tuned Gemma-3-270m-it model developed by vi-c0de. Fine-tuned on Apple Silicon using LoRA, this model specializes in generating traditional 5-7-5 syllable haikus on various topics. It targets strict syllable adherence, making it suitable for creative text generation focused on specific poetic forms.
Loading preview...
Overview
Gemmaiku-3-270m-it is an experimental 0.3 billion parameter language model, fine-tuned from Google's Gemma-3-270m-it by vi-c0de. This model is specifically designed and optimized for generating traditional 5-7-5 syllable haikus across diverse themes. It was trained locally using LoRA on Apple Silicon with the Apple MLX framework.
Key Capabilities
- Specialized Haiku Generation: Excels at producing poetic text adhering to the strict 5-7-5 syllable structure.
- Compact Size: At 270M parameters, it offers a lightweight solution for creative text generation tasks.
- Instruction-Tuned: Capable of following instructions to generate haikus based on user prompts.
Usage Considerations
While the model is trained for strict syllable adherence, its compact size means it may occasionally deviate on complex or out-of-distribution prompts. For optimal results, users are advised to:
- Keep generation temperatures low (e.g.,
0.1-0.3). - Utilize the correct chat template for prompt formatting.
Good For
- Creative writing applications requiring structured poetic forms.
- Educational tools for demonstrating haiku composition.
- Experiments with small, specialized language models for niche text generation tasks.