Sailor 7B

Sailor 7B is the 7-billion-parameter flagship of the Sailor family from Sea AI Lab in Singapore, introduced in April 2024. The family spans 0.5B to 7B parameters and is continually pre-trained from Qwen1.5 on 200B to 400B tokens covering English, Chinese, Vietnamese, Thai, Indonesian, Malay, and Lao.

Training uses BPE dropout for robustness, aggressive data cleaning, and small proxy models to tune the data mixture. Benchmarks show strong commonsense reasoning, QA, reading comprehension, and examination performance across the region's languages. Release timing below is era-level; the Sailor technical report is canonical.