Hugging Face Overview
OpenThaiGPT 1.5-7b-Instruct: Thai-Centric LLM
OpenThaiGPT 1.5-7b-Instruct is a 7.6 billion parameter instruction-tuned model built upon the Qwen v2.5 architecture, specifically designed for the Thai language. Released on September 30, 2024, it has been extensively fine-tuned using over 2 million Thai instruction pairs, making it highly proficient in understanding and generating Thai-specific content.
Key Capabilities & Features
- Superior Thai Language Performance: Achieves the highest average scores across various Thai language exams compared to other open-source Thai LLMs, as demonstrated on the OpenThaiGPT Eval and scb10x/thai_exam benchmarks.
- Extended Context Handling: Capable of processing up to 131,072 tokens of input and generating up to 8,192 tokens, facilitating detailed and complex interactions. It utilizes YaRN for enhanced length extrapolation beyond 32,768 tokens.
- Multi-turn Conversation: Supports seamless and extended dialogues.
- Retrieval Augmented Generation (RAG) Compatibility: Designed to work with RAG systems for improved response generation.
- Tool Calling Support: Enables efficient function calls, including external API integrations for real-time data retrieval (e.g., weather, stock market information).
Ideal Use Cases
- Thai-specific Chatbots: Developing conversational AI agents that require deep understanding and generation of Thai language.
- Educational Applications: Assisting with Thai language learning or answering questions related to Thai academic subjects.
- Information Retrieval: Building systems that need to extract and synthesize information from Thai documents or databases.
- Automated Customer Support: Providing intelligent responses for customer service in Thai.
- Applications requiring Tool Use: Integrating with external systems or APIs to perform actions or retrieve dynamic data based on user queries in Thai.