OpenThaiGPT-ThaiLLM-8b-instruct-v0.7.2-research-preview: Thai Knowledge Specialist
This is an experimental 8-billion parameter Thai language model, developed by the OpenThaiGPT team in collaboration with ThaiLLM. It is fine-tuned from Qwen3-8B and ThaiLLM-8B, with a strong focus on Thai knowledge, history, culture, and government information.
Key Capabilities & Features
- Deep Thai Knowledge: Accurately answers questions related to Thai history, culture, government agencies, and local context.
- Chain-of-Thought Reasoning: Supports step-by-step reasoning using the
<think>...</think> format, enhancing complex problem-solving. - Paraphrase Robustness: Understands diverse question phrasings due to Paraphrase Augmentation during training.
- Bilingual Support: Functions in both Thai and English.
- High Instruction Following: Achieves 87.6% on IFEval and 75.5% on IFEval-TH, indicating strong adherence to instructions.
- Training Details: Utilizes Full Supervised Fine-tuning (SFT) on a Qwen3-v5IFEval-SLERP Merged base, with a maximum context length of 8,192 tokens.
Intended Use Cases
This model is specifically optimized for:
- Thai FAQ Systems: Building intelligent Q&A systems for Thai-specific information.
- Thai Context Question Answering: Providing accurate answers to questions deeply rooted in Thai culture, history, and public administration.
Limitations
It's important to note that this model is not designed for advanced mathematics or competitive coding, where its performance is lower compared to specialized models. Its knowledge is limited to its training data, and like all LLMs, it may occasionally produce inaccurate information (hallucinations).