SeaLLMs-v3-7B: Large Language Model for Southeast Asian Languages
SeaLLMs-v3-7B is the latest 7.6 billion parameter base model in the SeaLLMs series, optimized for Southeast Asian (SEA) languages. It achieves state-of-the-art performance compared to other open-source models of similar size across diverse tasks including world knowledge, mathematical reasoning, translation, and instruction following. A key focus of SeaLLMs-v3 is enhanced trustworthiness, featuring significantly reduced hallucination and improved safety, particularly for queries related to Southeast Asian culture.
Key Capabilities & Features
- Multilingual Proficiency: Tailored for a wide range of SEA languages including English, Chinese, Indonesian, Vietnamese, Thai, Tagalog, Malay, Burmese, Khmer, Lao, Tamil, and Javanese.
- Strong Performance: Outperforms comparable models in human exam questions (M3Exam), cross-lingual MMLU, and multilingual math (MGSM) benchmarks.
- Enhanced Instruction Following: Demonstrates improved capability in understanding and executing instructions, especially in multi-turn conversational settings.
- Cultural Sensitivity & Safety: Designed to provide safe and contextually appropriate responses, with a focus on reducing hallucination and respecting local nuances.
Use Cases
This base model is ideal for fine-tuning on specific downstream tasks, particularly those involving SEA languages. For direct application in conversational or instruction-following scenarios, the chat-tuned version, SeaLLMs-v3-7B-Chat, is recommended.