Llama-3-70B-Tool-Use
A large language model with 70B parameters, optimized for tool use
- Advanced tool usage and function call task optimization
- API interaction, structured data manipulation, and complex tool usage
- Excellent performance on Berkeley Function Call Ranking (BFCL)
- Optimized Transformer Architecture
- Complete fine-tuning and direct preference optimization (DPO) training method
- Both input and output are text, enhancing the tool's usability
- Sensitive to temperature and top-p sampling configuration, requiring appropriate adjustments to optimize performance
Product Details
Llama3-70B-Tool-Use is a large language model with 70B parameters, designed specifically for advanced tool usage and function calling tasks. The overall accuracy of this model on the Berkeley Function Call Ranking (BFCL) reaches 90.76%, outperforming all open-source 70B language models. The model optimized the converter architecture and was trained on the Llama 3 70B base model through complete fine-tuning and direct preference optimization (DPO). Input as text and output as text enhance the ability to use tools and call functions. Although its main purpose is tool usage and function invocation, it may be more suitable for general knowledge or open tasks using a universal language model. This model may produce inaccurate or biased content in certain situations, and users should pay attention to implementing appropriate security measures that are suitable for their specific use cases. This model is highly sensitive to temperature and top-p sampling configuration.
