SeaLLM 2
About
The SeaLLM family is a series of large language models designed to cater to Southeast Asian languages, showcasing superior multilingual capabilities. The SeaLLM-7B-v2 marks a significant advancement, effectively outperforming larger models like its predecessor and GPT-3.5 in various multilingual tasks such as world knowledge, math reasoning, and instruction following. It notably excels in the zero-shot GSM8K task, achieving state-of-the-art performance. Building on this, the SeaLLM-7B-v2.5 further enhances its proficiency in reasoning benchmarks and world knowledge. Despite these strengths, users are advised to exercise caution as the models, even with safety fine-tuning, might still produce misleading or harmful content. The recent release of SeaLLMs-v3 improves on trustworthiness and reduces hallucinations, furthering the advancement in handling non-Latin Southeast Asian languages 145.