Hugging Face scb10x/llama3.2-typhoon2-3b-instruct is a 3 billion parameter instruction-tuned large language model developed by SCB 10X, based on the Llama3.2 architecture. This model is specifically optimized for Thai language instruction-following and function calling, demonstrating strong performance in Thai code-switching and instruction-following benchmarks. It excels in applications requiring robust Thai language understanding and generation, including complex function call scenarios.
Overview
scb10x/llama3.2-typhoon2-3b-instruct is a 3 billion parameter instruction-tuned large language model developed by SCB 10X, built upon the Llama3.2-3B foundation. This model is primarily designed for Thai (🇹🇭) and English (🇬🇧) language processing, with a strong emphasis on instruction-following and function calling capabilities.
Key Capabilities
- Superior Thai Language Performance: Outperforms Qwen2.5 3B Instruct and the base Llama3.2 3B Instruct on Thai-specific benchmarks like IFEval-TH, MT-Bench TH, and Thai Code-Switching tasks.
- Enhanced Function Calling: Demonstrates significantly higher accuracy in both Thai and English function call scenarios compared to its counterparts.
- Instruction Following: Shows strong instruction-following abilities, particularly in Thai contexts.
- Llama Architecture: Leverages the Llama architecture for its decoder-only design.
Good For
- Thai-centric Applications: Ideal for chatbots, virtual assistants, and other AI systems requiring high proficiency in the Thai language.
- Function Calling: Suitable for use cases that involve integrating with external tools and APIs through function calls.
- Multilingual Instruction Following: Effective for tasks requiring adherence to instructions in both Thai and English.
Limitations
As an instructional model still under development, it may occasionally produce inaccurate, biased, or objectionable responses. Developers should assess these risks for their specific use cases.