Meituan launches domestic large model Lomao-2.0, achieving fully autonomous training of a trillion-parameter model across all stacks

Meituan Breaks Free from NVIDIA's Grip

According to the South China Morning Post, China's leading food delivery giant Meituan has unveiled "LongCat-2.0," a large language model with 1.6 trillion parameters and a context window of 1 million token units, fully trained on domestic hardware.

Previously, AI models built on domestic chips in China were limited to inference tasks, while pre-training remained highly dependent on NVIDIA hardware.

Main Features:

???? Trained on a domestic computing cluster comprising 50,000 chips

???? The first trillion-parameter model fully trained and deployed on domestic hardware

???? Built using AI-specific ASIC intelligent computing super nodes and Huawei's Collective Communication Library (HCCL)

???? Outperforms Google Gemini 3.1 Pro in certain benchmark tests

This release marks rapid progress in China's development of a domestically driven AI hardware ecosystem.

Original source: toutiao.com/article/1869461445577740/

Disclaimer: The views expressed in this article are solely those of the author.