alphaedge-ai/gemma-3-270m-it-kor-32768
alphaedge-ai/gemma-3-270m-it-kor-32768 is a 0.3 billion parameter instruction-tuned Gemma 3 model, derived from google/gemma-3-270m-it. It features a reduced vocabulary of 32,768 tokens and a 32768 context length, achieved through a trimming method that optimizes it specifically for the Korean language. This optimization results in a 54.76% smaller model size while aiming for similar performance to the original for Korean language tasks.
Loading preview...
Overview
This model, alphaedge-ai/gemma-3-270m-it-kor-32768, is a specialized version of the google/gemma-3-270m-it model, optimized for the Korean language. It has been significantly reduced in size and vocabulary through a "trimming" method, making it more efficient for Korean-specific applications.
Key Characteristics
- Parameter Count: 0.3 billion parameters, a 54.76% reduction from the original
gemma-3-270m-it. - Vocabulary Size: Reduced from 262,144 tokens to 32,768 tokens, an 87.50% reduction.
- Context Length: Supports a context of 32,768 tokens.
- Optimization: Specifically optimized for the Korean language by removing tokens not commonly used in Korean, leading to a smaller memory footprint.
- Performance: Expected to perform similarly to the original model for Korean language tasks, though performance for other languages may be degraded due to vocabulary reduction.
Use Cases
- Korean Language Processing: Ideal for applications requiring efficient and performant instruction-tuned language understanding and generation in Korean.
- Resource-Constrained Environments: Suitable for deployment where memory and computational resources are limited, thanks to its significantly smaller model size.
Limitations
- May not perform well for languages other than Korean, as its vocabulary has been tailored to Korean linguistic patterns.