OpenThaiGPT 1.0.0-alpha-7b-chat-ckpt-hf Overview
This model is the first Thai implementation of a 7-billion parameter LLaMA v2 Chat model, fine-tuned by OpenThaiGPT to understand and respond to Thai translated instructions. It leverages the robust LLaMA v2 architecture, which was pretrained on over 2 trillion tokens, providing a strong foundation for its language capabilities.
Key Capabilities & Features
- Thai Instruction Following: Specifically optimized for processing and generating responses based on instructions provided in Thai.
- Enhanced Context Length: Features an upgraded context window of 4096 tokens, an increase from the previous 2048 tokens, allowing for more extensive conversations and complex queries.
- Commercial Use: Licensed for both research and commercial applications, making it suitable for a wide range of deployments.
- Base Model: Built upon the
meta-llama/Llama-2-7b-chat model.
Performance Benchmarks
Evaluated on the Open LLM Leaderboard, the model achieved an average score of 42.05. Notable scores include 50.85 on ARC (25-shot) and 74.89 on HellaSwag (10-shot), indicating its general reasoning and common sense capabilities. MMLU (5-shot) scored 40.02, and TruthfulQA (0-shot) scored 47.23.
Good for
- Applications requiring a large language model with strong Thai language understanding and generation.
- Developers and researchers building Thai-centric AI solutions.
- Commercial projects that need a performant and commercially usable Thai-instruction-following model.