typhoon-ai/llama3.2-typhoon2-3b-instruct
The typhoon-ai/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. It is primarily designed for Thai and English language tasks, excelling in instruction-following and function calling, particularly for Thai language applications. This model demonstrates strong performance in Thai code-switching and function call benchmarks compared to similar-sized models.
Loading preview...
Overview
Llama3.2-Typhoon2-3B-instruct is a 3 billion parameter instruction-tuned large language model developed by SCB 10X, built upon the Llama3.2 foundation model. It is specifically optimized for Thai (🇹🇭) and English (🇬🇧) language processing, demonstrating strong capabilities in instruction-following and function calling.
Key Capabilities & Performance
- Bilingual Proficiency: Primarily supports Thai and English, with notable performance in Thai-specific tasks.
- Instruction Following: Achieves 68.36% on IFEval - TH and 72.18% on IFEval - EN, outperforming Qwen2.5 3B Instruct in Thai instruction following.
- Function Calling: Excels in function call scenarios, scoring 71.36% for Thai and 75.90% for English, significantly higher than comparative models.
- Code-Switching: Demonstrates high proficiency in Thai Code-Switching with 99.2% (t=0.7) and 96% (t=1.0).
- Llama Architecture: A decoder-only model based on the Llama architecture, requiring
transformers4.45.0 or newer.
Intended Uses & Limitations
This model is an instructional model suitable for various tasks including analysis, question answering, math, coding, creative writing, and role-play. While it incorporates guardrails, it is still under development and may occasionally produce inaccurate, biased, or objectionable responses. Developers are advised to assess risks within their specific use cases.