Meituan Unveils the LongCat-Flash-Lite Model
21 hour ago / Read about 0 minute
Author:小编   

On February 6th, Meituan made an official announcement regarding the launch of its latest innovation—the LongCat-Flash-Lite model. This cutting-edge model is equipped with a staggering 68.5 billion parameters. However, during each inference process, it intelligently activates only between 2.9 and 4.5 billion parameters, thereby establishing itself as a lightweight Mixture of Experts (MoE) model. By adeptly utilizing over 30 billion parameters within the embedding layer, the LongCat-Flash-Lite model not only outperforms other MoE baseline models that possess a similar number of parameters but also showcases remarkable competitiveness when pitted against models of an equivalent scale. It particularly shines in the realms of agent applications and code-related tasks. Moreover, harnessing the power of YARN technology, this model is capable of supporting context lengths of up to 256K. This enables it to efficiently tackle complex tasks such as processing lengthy documents and conducting large-scale code analyses with ease and precision.