Nace AI Research Lab has published a technical report detailing their Drex decision models, marking a significant contribution to the community's understanding of how to train and evaluate such models. The report covers various architectures, training techniques including Reinforcement Learning from Adjusted Feedback (RLAF), and evaluation metrics, with models ranging from 9B to 35B parameters. The Drex models have demonstrated strong performance on long-context benchmarks and offer low latency, with weights and code now available for local use. AI
IMPACT Provides a foundational playbook for training state-of-the-art decision models, potentially accelerating development in agent routing and tool selection.
RANK_REASON The cluster describes a technical report detailing the architecture, datasets, training, and evaluation of decision models, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
