alphaedge-ai/gemma-3-270m-it-kat-16384
alphaedge-ai/gemma-3-270m-it-kat-16384 is a 0.3 billion parameter instruction-tuned causal language model, derived from Google's Gemma-3-270m-it. It has been optimized for the Georgian language through a vocabulary reduction technique, resulting in a 58.67% smaller model size and a vocabulary of 16,384 tokens. This specialization makes it highly efficient for Georgian language tasks, though its performance may be limited for other languages due to the trimmed vocabulary.
Loading preview...
Overview
alphaedge-ai/gemma-3-270m-it-kat-16384 is a specialized version of Google's Gemma-3-270m-it model, significantly optimized for the Georgian language. This model achieves a 58.67% reduction in size compared to its base model by employing a vocabulary trimming method, reducing its vocabulary from 262,144 to just 16,384 tokens. This optimization aims to maintain similar performance for Georgian while drastically decreasing memory footprint.
Key Capabilities
- Georgian Language Specialization: Optimized specifically for processing and generating text in Georgian.
- Reduced Model Size: At 110,811,776 parameters, it is substantially smaller than the original 268,098,176 parameter model, leading to lower memory consumption.
- Efficient Inference: The smaller vocabulary and model size contribute to more efficient inference for its target language.
Good for
- Applications requiring a compact and efficient language model for Georgian text processing.
- Deployments with strict memory constraints where Georgian language support is primary.
- Research into vocabulary trimming techniques and their impact on language model performance for specific languages.