Alibaba updated its open-source Tongyi Qianwen family of large language models with Qwen2, outperforming Meta's Llama 3 in multiple benchmarks, including maths, coding, natural and social sciences, engineering, and humanities.
Alibaba's Qwen2-72B model is the strongest among the five variations, with up to 72.7 billion parameters trained on 27 languages, surpassing Meta's Llama 3-70B in benchmark tests across different domains.
Collection
[
|
...
]