Deepreneur/blue-lizard

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Feb 5, 2024License:llama2Architecture:Transformer0.0K Open Weights Cold

Deepreneur/blue-lizard is a 7 billion parameter instruction-tuned causal language model developed by Deepreneur. Based on Meta's Llama-2-7b, it has undergone additional pre-training and fine-tuning using Japanese datasets, including Wikipedia and books. This model achieves scores exceeding ChatGPT-3.5 on the JGLUE Japanese benchmark, positioning it as a top-performing Japanese language model in its size class.

Loading preview...

Deepreneur/blue-lizard: A High-Performance Japanese LLM

Deepreneur/blue-lizard is a 7 billion parameter language model developed by Deepreneur, built upon Meta's Llama-2-7b architecture. This model has been extensively pre-trained and fine-tuned using a diverse range of Japanese datasets, including Wikipedia and various books, to optimize its performance for the Japanese language.

Key Capabilities

  • Exceptional Japanese Language Performance: Despite its lightweight 7B parameter count, Deepreneur/blue-lizard demonstrates superior performance on the JGLUE (Japanese General Language Understanding Evaluation) benchmark, surpassing scores achieved by ChatGPT-3.5. This makes it one of the highest-performing publicly available Japanese models.
  • Instruction-Tuned: The model has undergone fine-tuning with proprietary data, enhancing its ability to follow instructions effectively.
  • Efficient for Japanese Tasks: Optimized specifically for Japanese, it offers a powerful solution for applications requiring strong understanding and generation in the language.

Good For

  • Japanese Natural Language Processing (NLP): Ideal for tasks such as text generation, summarization, translation, and question answering in Japanese.
  • Applications Requiring High Accuracy in Japanese: Suitable for use cases where precise and contextually relevant Japanese output is critical.
  • Resource-Efficient Deployment: Its 7B parameter size makes it a more accessible option for deployment compared to larger models, while still delivering high performance for Japanese tasks.