Tsunami-1.0-14B-Instruct: A Specialized Thai LLM

Tsunami-1.0-14B-Instruct is a 14 billion parameter instruction-tuned large language model developed by Pollakrit Lorprasertkul. It is fine-tuned from the Qwen2.5-14B architecture, with a primary focus on enhancing performance for the Thai language.

Key Capabilities and Performance

This model excels in Thai language understanding and generation, as evidenced by its benchmark results. It outperforms several comparable models, including the base Qwen2.5-14B-Instruct, Meta-Llama-3.1-70B-Instruct, llama-3-typhoon-v1.5x-70b-instruct, and openthaigpt1.5-14b-instruct, on specific Thai-centric evaluations.

• Superior Thai Language Performance: Achieves an average score of 62.05, with 61.06 on Thai Exam and 63.05 on M3Exam, surpassing other models in its class.
• Instruction Following: Fine-tuned for instruction-based tasks, utilizing the ChatML prompt template for structured conversations.

Ideal Use Cases

Tsunami-1.0-14B-Instruct is particularly well-suited for applications requiring high-quality Thai language processing.

• Thai-specific Chatbots and Assistants: Its strong performance in Thai makes it ideal for conversational AI in Thai.
• Content Generation in Thai: Generating articles, summaries, or creative text in the Thai language.
• Educational Tools: Assisting with Thai language learning or content creation for Thai speakers.