Core Viewpoint - The Singapore National Artificial Intelligence Program (AISG) is undergoing a significant strategic adjustment by shifting from the Meta Llama model system to Alibaba's Qwen open-source architecture for its Southeast Asian language model project [1]. Group 1: Project Overview - AISG has launched the "Qwen-SEA-LION-v4" model, which has achieved top-tier performance among open-source instruction models with under 200 billion parameters, surpassing previous versions based on Llama in several metrics, including Indonesian, Malay, and Thai [1]. - The latest version utilizes Alibaba Cloud's Qwen3-32B as the base model, combined with AISG's extensive regional corpus for training, emphasizing understanding of Southeast Asia's multilingual and multicultural contexts [1]. Group 2: Strategic Goals - AISG aims to enhance Singapore's digital economy capabilities through artificial intelligence, addressing the shortcomings of "Western-centric" models in regional languages from the outset of the SEA-LION project [1]. - The collaboration with Qwen is intended to create a tailored open-source system specifically designed for Southeast Asian languages and scenarios [1].
新加坡AI项目引入中国开源模型