chancatNVIDIA发布新模型 性能直逼 DeepSeek-R1 中发帖

今天英伟达开源了一款名为 Llama-3.1-Nemotron-Ultra-253B-v1 的新模型,性能逼近拥有 6710 亿参数的 DeepSeek R1,但只用了不到一半的参数量。