Overview

Sailor-14B-Chat is a 14.2 billion parameter instruction-tuned language model developed by sail, specifically designed for South-East Asian (SEA) languages. Built on the robust Qwen 1.5 architecture, this model focuses on enhancing performance in Indonesian, Thai, Vietnamese, Malay, and Lao, alongside strong capabilities in English and Chinese.

Key Capabilities

• Multilingual Proficiency: Optimized for five key SEA languages (Indonesian, Thai, Vietnamese, Malay, Lao), in addition to English and Chinese.
• Instruction-Tuned: Fine-tuned with publicly available datasets like Aya Collection, OpenOrca, UltraChat, and UltraFeedback to improve conversational and instruction-following abilities.
• Strong Benchmarking: Demonstrates proficiency in tasks such as question answering and commonsense reasoning across its target languages.
• Extensive Training: Continuously pre-trained on 200 billion tokens, leveraging a diverse corpus including SlimPajama, SkyPile, CC100, and MADLAD-400, with aggressive data deduplication and cleaning.

Use Cases

Sailor-14B-Chat is ideal for applications requiring robust language understanding and generation in South-East Asian contexts. Its instruction-tuned nature makes it suitable for chatbots, content generation, and information retrieval systems that serve multilingual user bases in the SEA region.