Hugging Face OpenBuddy/openbuddy-llama3-70b-v21.2-32k Overview
This model is a 70 billion parameter instruction-tuned large language model developed by the OpenBuddy team, based on Meta's Llama 3 architecture. It is specifically designed as a multilingual chatbot, with a primary focus on English and Chinese language capabilities. The model utilizes a unique prompt format, including special tokens like <|role|>, <|says|>, and <|end|>, which is defined in its tokenizer_config.json for consistent performance across platforms like transformers and vllm.
Key Capabilities
- Multilingual Chatbot: Optimized for conversational interactions in both English and Chinese.
- Instruction Following: Fine-tuned to follow instructions and provide helpful, respectful, and honest responses.
- Safety and Ethics: Designed to avoid harmful, political, religious, unethical, racist, sexist, toxic, dangerous, or illegal content, ensuring socially unbiased and positive outputs.
- Specialized Prompt Format: Employs a custom prompt structure for effective dialogue management and response generation.
Performance and Limitations
Evaluation results from the Open LLM Leaderboard show an average score of 35.18, with specific metrics including 70.10 for IFEval (0-Shot) and 49.97 for BBH (3-Shot). The model has a knowledge cutoff of April 2023. As with all LLMs, it has inherent limitations and may produce erroneous or undesirable outputs, and is not recommended for critical or high-stakes applications.
Good for
- Developing conversational AI agents in English and Chinese.
- Applications requiring instruction-following capabilities.
- Research and development in multilingual LLMs based on Llama 3.