Hugging Face has released its second LLM leaderboard to rank the best language models across various tasks. Alibaba’s Qwen models dominate the rankings, with three spots in the top ten. The leaderboard tests models on knowledge, reasoning, math, and instruction following using six benchmarks. Qwen, Alibaba’s LLM, leads the pack, followed by other models like Llama3-70B and Meta’s LLM. The tests are run on Hugging Face’s computers, powered by 300 Nvidia H100 GPUs. The leaderboard is open for submissions, with a new voting system to prioritize popular new entries for testing.
Hugging Face’s first leaderboard became popular among developers aiming for high ranks, but as models improved, the results became less meaningful, leading to the creation of a second leaderboard. Some models underperformed in the new leaderboard due to over-training on the first one’s benchmarks. This trend reflects a decline in AI performance over time, highlighting the importance of training data in LLM performance. True artificial intelligence remains a distant goal, as evidenced by the limitations of current language models. Hugging Face’s collaborative approach and commitment to transparency make it a trusted source in the LLM space.
Source link
Source link: https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-llms-storm-hugging-faces-chatbot-benchmark-leaderboard-alibaba-runs-the-board-as-major-us-competitors-have-worsened