Jun 26, 02:24 PM

Open LLM Leaderboard 2 Released with New Benchmarks; Qwen 72B Leads on Huggingface Hub

The Open LLM Leaderboard 2 has been released, introducing new benchmarks and features to evaluate large language models (LLMs) more effectively. This update aims to address the plateauing performance of LLMs by incorporating new evaluation metrics such as IFEval, BBH, MATH Lvl 5, GPQA, MUSR, and MMLU-PRO. The leaderboard now includes high-quality datasets, chat templates, and a community voting system to prioritize model evaluations. Qwen 72B Instruct currently leads the leaderboard. The update also emphasizes fairer, transparent, and reproducible comparisons of LLMs, with new visualizations and technical details to enhance user experience. The leaderboard is available on the huggingface Hub, and 300 H100 GPUs were used to re-run new evaluations.

#Open LLM Leaderboard #Qwen 72B Instruct #Hub

Written with ChatGPT (GPT-4o).