maywell/koOpenChat-sft
koOpenChat-sft is a 7 billion parameter instruction-tuned causal language model developed by maywell, based on the OpenChat3.5 architecture. This model is specifically fine-tuned for Korean language tasks, leveraging the ChatML and Alpaca (No-Input) instruction formats. It is designed for general-purpose conversational AI in Korean, offering a balance of performance and efficiency for various applications.
Loading preview...
koOpenChat-sft: A Korean-Optimized OpenChat Model
koOpenChat-sft is a 7 billion parameter language model developed by maywell, built upon the OpenChat3.5 base architecture. This model has undergone supervised fine-tuning (SFT) to enhance its performance, particularly for Korean language understanding and generation. It was trained on a single A100 80GB GPU, indicating a focused and efficient development process.
Key Capabilities & Features
- Korean Language Optimization: Specifically fine-tuned to excel in Korean conversational and instructional tasks.
- Instruction Following: Supports both ChatML and Alpaca (No-Input) instruction formats, making it versatile for various prompt structures.
- Open-source Base: Leverages the robust OpenChat3.5 architecture, known for its strong performance in open-source benchmarks.
- Benchmark Performance: Achieves an average score of 51.36 on the Open LLM Leaderboard, with notable scores in HellaSwag (78.73) and MMLU (61.32), demonstrating its general reasoning and knowledge capabilities.
Good For
- Korean Chatbots and Conversational AI: Its instruction-following capabilities and Korean optimization make it suitable for building interactive agents.
- General Korean Text Generation: Can be used for various tasks requiring coherent and contextually relevant Korean text output.
- Research and Development: Provides a solid base for further fine-tuning or experimentation in Korean NLP applications.