ELYZA-japanese-Llama-2-7b
| Property | Value |
|---|---|
| Parameter Count | 6.27B |
| Vocabulary Size | 32,000 |
| License | LLAMA 2 Community License |
| Languages | Japanese, English |
| Paper | arXiv:2307.09288 |
What is ELYZA-japanese-Llama-2-7b?
ELYZA-japanese-Llama-2-7b is an advanced language model that extends Meta's LLaMA 2 architecture with enhanced Japanese language capabilities. The model underwent additional pre-training specifically to improve its Japanese language understanding and generation abilities while maintaining its English language capabilities.
Implementation Details
The model architecture is based on the LLaMA 2 framework and comes in several variants, including a base model, an instruct-tuned version, and fast versions with expanded vocabulary. The implementation uses PyTorch and the Transformers library, making it easily accessible for developers.
- Base model with 6.27B parameters and 32,000 vocab size
- Fast version with 6.37B parameters and 45,043 vocab size
- Includes both regular and instruct-tuned variants
- Supports text generation and inference endpoints
Core Capabilities
- Bilingual text generation in Japanese and English
- Instruction-following capabilities in Japanese
- Enhanced Japanese language understanding
- Seamless integration with Hugging Face Transformers
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its specialized Japanese language capabilities while maintaining the robust architecture of LLaMA 2. It's particularly noteworthy for being one of the few large language models specifically optimized for Japanese language tasks.
Q: What are the recommended use cases?
The model is well-suited for Japanese text generation, bilingual applications, and general language understanding tasks. It's particularly effective when used with the provided system prompts for Japanese language interaction.
