typhoon-ai/llama3.2-typhoon2-1b-instruct
Llama3.2-Typhoon2-1B-instruct by typhoon-ai is a 1 billion parameter instruction-tuned large language model based on the Llama3.2 architecture, primarily supporting Thai and English. It demonstrates strong performance in Thai instruction-following, code-switching, and function calling, outperforming similar-sized models like Qwen2.5 1.5B instruct in several key Thai benchmarks. This model is optimized for applications requiring robust bilingual (Thai/English) conversational AI and tool use capabilities.
Loading preview...
Llama3.2-Typhoon2-1B-instruct: A Thai-Centric LLM
Llama3.2-Typhoon2-1B-instruct is a 1 billion parameter instruction-tuned large language model developed by typhoon-ai, built upon the Llama3.2 architecture. It is designed for strong performance in both Thai and English, with a particular focus on Thai language capabilities.
Key Capabilities & Performance
This model excels in several areas, especially when compared to other models in its size class:
- Bilingual Instruction Following: Achieves 52.46% on IFEval-TH and 53.35% on IFEval-EN, demonstrating strong instruction adherence in both languages.
- Thai Code-Switching: Shows high proficiency with 96.4% accuracy at temperature 0.7 and 88% at temperature 1.0, indicating robust handling of mixed-language inputs.
- Function Calling: Records 34.96% on FunctionCall-TH and 45.60% on FunctionCall-EN, making it suitable for tool-use applications.
- MT-Bench Scores: Scores 3.9725 on MT-Bench TH and 5.2125 on MT-Bench EN.
Differentiators & Use Cases
Llama3.2-Typhoon2-1B-instruct stands out for its optimized performance in Thai language tasks, particularly in instruction following and function calling, where it surpasses models like Qwen2.5 1.5B instruct. Its strong code-switching ability makes it highly effective for real-world bilingual interactions. Developers can leverage this model for building conversational agents, chatbots, and applications requiring tool integration with a focus on the Thai language market. While it includes guardrails, developers should assess its suitability for specific use cases, acknowledging it is still under development.