Llama-3-70B-Tool-Use

Groq has released a model specifically optimized for advanced tool use and function call tasks. Training Method: The model was comprehensively fine-tuned and directly preference optimized (DPO) on the basis of the Llama 3 70B base model. Berkeley Function Call List (BFCL) Score: Overall accuracy of 90.76%. This score represents the best performance among all open-source 70B LLMs on BFCL.

Groq has made a splash with the release of a distinctive model that has been meticulously optimized for advanced tool use and function call tasks. In terms of training, Groq adopted a strategy of comprehensive fine-tuning and direct preference optimization (DPO) on the Llama 3 70B base model. After such training, the model achieved remarkable results on the Berkeley Function Call List (BFCL), with an overall accuracy rate of 90.76%. This score is of great significance, as it represents the best performance among all open-source 70B large language models on BFCL, fully demonstrating the exceptional capabilities and strong advantages of Groq's released model in advanced tool use and function call tasks.