Llama-SEA-LION-v3.5-70B-R: A Hybrid Model for Southeast Asian Languages

Developed by AI Singapore, Llama-SEA-LION-v3.5-70B-R is a 70 billion parameter decoder-only LLM based on the Llama 3.1 architecture. It is part of the SEA-LION (Southeast Asian Languages In One Network) collection, specifically designed and instruction-tuned for the Southeast Asian region. This model offers versatile functionality, adept at handling both complex reasoning tasks and general text generation, with mode selection managed via the tokenizer's chat template.

Key Capabilities & Features

• Multilingual Support: Supports 13 languages including Burmese, Chinese, English, Filipino, Indonesian, Javanese, Khmer, Lao, Malay, Sundanese, Tamil, Thai, and Vietnamese.
• Hybrid Functionality: Capable of both complex reasoning and standard text generation, with a unique thinking_mode toggle for enhanced reasoning.
• Extended Context Length: Features a substantial 128k token context window.
• Instruction-Tuned: Further instruction-tuned in English and various SEA languages (Filipino, Indonesian, Tamil, Thai, Vietnamese) for improved instruction-following.
• Benchmarked Performance: Evaluated using SEA-HELM for general language capabilities (QA, Sentiment, Translation, Summarisation, Reasoning) and SEA-IFEval/SEA-MTBench for instruction-following, with results available on the SEA-HELM leaderboard.

When to Use This Model

This model is ideal for applications requiring strong performance in Southeast Asian languages, particularly for tasks involving complex reasoning, instruction-following, and general text generation. Its hybrid nature and multilingual capabilities make it suitable for diverse use cases across the SEA region. Users should note that the model has not been aligned for safety, requiring developers to implement their own safety fine-tuning.