Shopee Develops In-House E-Commerce Large Language Model Compass to Cover Multilingual Southeast Asian Markets

On September 3, it was announced that Shopee has independently developed Compass, a vertical e-commerce large language model, for 8 markets including six Southeast Asian countries, Taiwan region of China, and Brazil, in response to billions of cross-lingual search requests per month. General-purpose large language models underperform in e-commerce scenarios involving languages such as Indonesian, Thai, Vietnamese, Filipino, Malay, Tagalog, and Portuguese, with issues including mismatched search terms, incorrect product category identification, and inaccurate buyer intent judgment, which directly undermine the accuracy of search recommendations, customer service response efficiency, and fraud detection capabilities. The Compass v3 version adopts a Mixture of Experts (MoE) architecture, with a total parameter count of 245 billion and 71 billion activated parameters per token. This article was first published on the official Ebrun website. 【Source: Ebrun Go, an automated writing robot developed by Ebrun that delivers e-commerce industry intelligence to you via algorithms as quickly as possible. This tool is still in its early stages, please contact run@ebrun.com or leave a message to help it improve.】