OpenThaiGPT 1.5 Instruct Overview

OpenThaiGPT 1.5 Instruct is a 72.7 billion parameter Thai language chat model, built upon the Qwen v2.5 architecture. Developed by the OpenThaiGPT team, this model is specifically fine-tuned on over 2,000,000 Thai instruction pairs, making it highly proficient in handling Thai-specific domain questions and conversations. It demonstrates leading performance among open-source Thai LLMs, achieving the highest average scores across various Thai language examinations.

Key Capabilities

• Advanced Thai Language Proficiency: Achieves the highest average scores on Thai language exams compared to other open-source Thai LLMs, as evidenced by benchmarks on OpenThaiGPT Eval and scb10x/thai_exam.
• Extensive Context Handling: Processes up to 131,072 input tokens and generates up to 8,192 output tokens, facilitating detailed and complex interactions. It utilizes YaRN for enhanced length extrapolation, supporting contexts beyond 32,768 tokens.
• Multi-turn Conversation: Supports extended dialogues, maintaining coherence and context across multiple exchanges.
• Retrieval Augmented Generation (RAG) Compatibility: Designed to integrate with RAG systems for enhanced response generation.
• Tool Calling Support: Enables efficient function calls, including external API interactions for real-time data retrieval (e.g., weather information) or predictions, by intelligently responding to user queries.

Good for

• Applications requiring highly accurate and nuanced Thai language understanding and generation.
• Building chatbots or conversational AI systems for the Thai market.
• Use cases benefiting from long context windows and multi-turn dialogue capabilities.
• Integrating external tools and APIs through intelligent function calling.