Hugging Face Overview
OpenThaiGPT 1.5 14b Instruct: A Thai-Centric LLM
OpenThaiGPT 1.5 14b Instruct is a 14-billion-parameter language model developed by the OpenThaiGPT team, built upon the Qwen v2.5 architecture. Released on October 13, 2024, this model is specifically fine-tuned on over 2,000,000 Thai instruction pairs, making it highly proficient in understanding and generating Thai language.
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.
- Extensive Context Handling: Processes up to 131,072 input tokens and generates up to 8,192 tokens, supporting complex and detailed interactions. It utilizes YaRN for enhanced length extrapolation beyond 32,768 tokens.
- Advanced Interaction: Supports multi-turn conversations for extended dialogues, Retrieval Augmented Generation (RAG) for enhanced response quality, and robust tool calling capabilities for integrating external functions and APIs.
- Qwen-based Architecture: Leverages the Qwen v2.5 foundation, allowing for both research and commercial uses under its license terms.
Ideal Use Cases
- Thai Language Applications: Best suited for applications requiring deep understanding and generation of Thai text, including chatbots, content creation, and customer support in Thai.
- Educational & Research Tools: Excellent for academic purposes, particularly in evaluating and developing Thai language models.
- Integration with External Systems: Its tool calling feature makes it suitable for scenarios requiring interaction with APIs for real-time data retrieval (e.g., weather, stock market) or custom functions.