Groq, in partnership with Glaive, has recently introduced two state-of-the-art AI models for tool use: Llama-3-Groq-70B-Tool-Use and Llama-3-Groq-8B-Tool-Use. By outperforming all previous models, these innovations have achieved over 90% accuracy on the Berkeley Function Calling Leaderboard (BFCL) and are now open-sourced and available on GroqCloud Developer Hub and Hugging Face. The models leveraged ethically generated data instead of user data, adhering to strict ethical AI development standards.
Both models were crafted via a careful and overall fine-tuning training approach combined with Direct Preference Optimization (DPO) to enhance tool use and function-calling capabilities. Ensuring minimal overfitting, contamination analysis was employed using the LMSYS method. The results were a mere 5.6% contamination rate for the SFT data and 1.3% for the DPO data.
A highlight of the Llama-3 Groq Tool Use models is their compatibility with the hybrid approach, wherein they can be integrated with general-purpose language models. Correspondingly, a routing system analyzes incoming requests, utilizing the Llama-3 Groq Tool Use model for API interactions, function calls, and structured data manipulation. In contrast, the unmodified Llama-3 70B serves general knowledge, open-ended conversations, or tasks unrelated to tool use. This guarantees that every request is handled by the most suitable model, ensuring optimal performance.
Both innovative AI models can be previewed through the Groq API, specifically with model IDs – llama3-groq-70b-8192-tool-use-preview and llama3-groq-8b-8192-tool-use-preview. Groq encourages its community to leverage the models in creating and experimenting via the GroqCloud Developer Hub to spur AI tool use innovations.
To conclude, the introduction of the Llama-3 Groq Tool Use models marks a considerable milestone in AI research and development. Groq’s commitment to ethical AI development, coupled with its open collaboration ethos, indicate its leadership in the industry, setting the stage for future advancements in AI tool use. TL;DR: Groq, in collaboration with Glaive, has launched two new AI models that have achieved over 90% accuracy on the BFCL. Now open for public use, these models adhere to ethical AI standards and can be paired with general-purpose models to maximize AI performance.
