typhoon-ai/llama-3-typhoon-v1.5-8b-instruct

Llama-3-Typhoon-1.5-8B-instruct: A Bilingual Thai-English LLM

Llama-3-Typhoon-1.5-8B-instruct is an 8 billion parameter instruction-tuned model developed by SCB10X, built upon the Llama 3 foundation. This model is specifically optimized for both Thai and English languages, aiming to provide high-quality responses in a bilingual context.

Key Capabilities & Performance

• Bilingual Proficiency: Primarily supports Thai and English, making it suitable for applications requiring understanding and generation in both languages.
• Instruction Following: Designed as an instruct model, it is capable of following user instructions effectively.
• Improved Thai Benchmarks: Demonstrates significant performance improvements over its predecessor (Typhoon-1.0) and other comparable Thai models across various Thai-specific benchmarks such as ONET, IC, TGAT, TPAT-1, A-Level, and M3Exam. For instance, it achieves an average score of 0.506 on ThaiExam, outperforming other 7B models like Sailor and SeaLLM 2.0.
• Llama 3 Architecture: Benefits from the advancements of the Llama 3 base model, ensuring a robust and capable foundation.

Intended Use Cases

• Thai Language Applications: Ideal for chatbots, content generation, and language understanding tasks in Thai.
• Bilingual Assistants: Can be utilized in scenarios requiring interaction in both Thai and English.
• Research and Development: Serves as a strong base for further fine-tuning or research into bilingual LLMs, particularly for Southeast Asian languages.