aisingapore/Apertus-SEA-LION-v4-8B-IT
Apertus-SEA-LION-v4-8B-IT is an 8-billion parameter decoder-based large language model developed by AI Singapore, built upon the Apertus-8B-Instruct architecture. It is specifically post-trained on 6.4 million instruction-text pairs for domain adaptation to the Southeast Asian (SEA) region. This model excels in multilingual and multicultural fluency across key SEA languages, including Indonesian, Vietnamese, Thai, Filipino, Tamil, Burmese, and Malay, and incorporates tool-calling capabilities.
Loading preview...
Apertus-SEA-LION-v4-8B-IT: Southeast Asian Language Model
Apertus-SEA-LION-v4-8B-IT is an 8-billion parameter instruction-tuned model from AI Singapore, designed for the Southeast Asian (SEA) region. It is based on the Apertus-8B-Instruct architecture and has undergone extensive post-training on approximately 6.4 million instruction-text pairs to achieve strong domain adaptation.
Key Capabilities
- Multilingual and Multicultural Fluency: Proficient in Indonesian, Vietnamese, Thai, Filipino, Tamil, Burmese, and Malay, making it suitable for diverse regional applications.
- Tool-Calling: Includes capabilities for function calling, demonstrated with examples like searching HDB listings.
- Open Resources: AI Singapore has released the post-training datasets and evaluation codes/datasets (SEA-HELM) to promote transparency and further development.
- Context Length: Features a context length of 65,000 tokens.
Evaluation and Limitations
The model's performance is evaluated using the SEA-HELM benchmark for general language capabilities (QA, Sentiment, Translation, etc.) and SEA-IFEval/SEA-MTBench for instruction-following and multi-turn chat. While designed for regional fluency, users should be aware of common LLM limitations such as potential for hallucination and the need for safety fine-tuning, as the model has not been aligned for safety.