Llama-SEA-LION-v3-70B-IT: Southeast Asian Language Model

Llama-SEA-LION-v3-70B-IT is a 70 billion parameter instruction-tuned model developed by AI Singapore, based on the Llama 3.1 architecture. It is part of the SEA-LION (Southeast Asian Languages In One Network) collection, specifically designed and optimized for the Southeast Asia region. The model supports a wide array of languages including Burmese, Chinese, English, Filipino, Indonesian, Javanese, Khmer, Lao, Malay, Sundanese, Tamil, Thai, and Vietnamese, utilizing the default Llama 3.1 70B Instruct tokenizer and featuring a 32,768 token context length.

Key Capabilities

• Multilingual Support: Pretrained and instruction-tuned across 13 Southeast Asian languages, in addition to English.
• Instruction Following: Evaluated using SEA-IFEval and SEA-MTBench, which are localized versions of IFEval and MT-Bench, to assess adherence to constraints and multi-turn conversation abilities.
• General Language Understanding: Performance assessed on the SEA-HELM benchmark, covering tasks like Question Answering, Sentiment Analysis, Toxicity Detection, Translation, Abstractive Summarization, Causal Reasoning, and Natural Language Inference.

Good For

• Applications requiring strong language understanding and generation in a variety of Southeast Asian languages.
• Instruction-following tasks where adherence to specific prompt constraints is critical.
• Multi-turn conversational agents tailored for the SEA region.

Limitations

Users should be aware that the model may exhibit hallucinations and occasional irrelevant content. It has not been aligned for safety, and developers are advised to implement their own safety fine-tuning and security measures.