stabilityai/japanese-stablelm-base-beta-7b
Japanese-StableLM-Base-Beta-7B is a 7-billion parameter decoder-only language model developed by Stability AI Japan, based on the Llama-2 architecture with a 4096-token context length. It is specifically fine-tuned on a diverse collection of Japanese data to maximize performance on Japanese language tasks. This model serves as a foundational base for application-specific fine-tuning, offering strong capabilities for Japanese text generation and understanding.
Loading preview...
Japanese-StableLM-Base-Beta-7B Overview
japanese-stablelm-base-beta-7b is a 7-billion parameter decoder-only language model developed by Stability AI Japan. Built upon the Llama-2 transformer architecture, this model has been extensively fine-tuned on a diverse and substantial collection of Japanese data, totaling approximately 100 billion tokens. Its primary design goal is to achieve high performance across various Japanese language tasks.
Key Capabilities
- Japanese Language Specialization: Optimized for generating and understanding Japanese text, making it suitable for applications requiring strong Japanese linguistic proficiency.
- Foundation Model: Intended as a base model for further fine-tuning to specific downstream applications, offering flexibility for developers.
- Llama-2 Architecture: Benefits from the robust and widely recognized Llama-2 transformer architecture.
- Extensive Training Data: Pre-trained on a mixture of Japanese and English corpora including Japanese/English Wikipedia, Japanese mc4, Japanese CC-100, Japanese OSCAR, and SlimPajama.
Good For
- Application-Specific Fine-tuning: Ideal for developers looking to build custom Japanese language applications by fine-tuning a strong base model.
- Japanese Text Generation: Generating coherent and contextually relevant text in Japanese.
- Research and Development: Serving as a robust foundation for research into Japanese natural language processing.
- Commercial Use: Available under the Llama2 Community License, allowing for commercial applications with reasonable caution regarding generated content.