Hugging Face has introduced a new open large language model (LLM) leaderboard, with Chinese models ranking at the top. The leaderboard evaluates open-source LLMs based on new benchmarks, including the MMLU-pro benchmark, using 300 NVIDIA H100 GPUs for re-evaluation. Alibaba’s Qwen-72B model emerged as the top performer, showcasing the rapid progress of Chinese AI companies in the LLM space. The dominance of Chinese models on the leaderboard highlights the competitive and global nature of the open-source AI ecosystem. While previous benchmarks have become too easy for the latest models, there is a need for more challenging evaluations as LLMs become more sophisticated. The results also show that increasing model size does not always lead to better performance. The launch of the new leaderboard by Hugging Face signifies a step towards transparent and comprehensive evaluation in the field. With Chinese models leading, the leaderboard is expected to drive further innovation and investment in open-source AI technologies globally. Hugging Face, established in 2016, is a prominent platform for open-source machine learning, hosting a large community of developers using its models and datasets.

Source link

Source link: https://analyticsindiamag.com/hugging-face-launches-open-llm-leaderboard-chinese-models-dominate/