InfectA-Chat
To prevent adversial effects of infectious diseases, clear and accessible communication, tracking infectious diseases regularly is crucial. InfectA-Chat is a generative model specifically designed to address this need. Built upon the powerful AceGPT-7B-Chat pre-trained model, InfectA-Chat is fine-tuned to track infectious diseases outbreaks in the infectious diseases domain. This makes it a valuable tool for facilitating communication in both Arabic and English, potentially bridging language barriers and fostering a deeper understanding of infectious diseases.
Model Details
In the fight against infectious diseases in the Middle East, clear and effective communication is paramount. We're excited to announce the release of InfectA-Chat, a generative text model fine-tuned on the AceGPT-7B-Chat model. Designed specifically for the Arabic and English languages, InfectA-Chat excels at following instructions related to infectious disease topics. Notably, our models outperform existing Arabic and state-of-the-art LLMs on Q&A task involving infectious disease instructions while competing with GPT-4. This advancement has the potential to significantly improve communication and disease tracking efforts in the specific region.
- Developed by: Korea Institute of Science and Technology
- Language(s) (NLP): Arabic, English
- License: Creative Commons Attribution 2.0
- Finetuned from model [optional]: AceGPT-7B-Chat
- Repository: KISTI-AI/InfectA-Chat
Training Details
Training Data
InfectA-Chat was instruction fine-tuned with 55,400 infectious diseases-related instruction-following data.
Training Procedure
This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure.
Training Hyperparameters
- Training regime: fp32
Evaluation
Evaluation Results on Infectious Diseases-related Instruction-Following Dataset
Experiments on infectious diseases-related instruction-following data and Arabic MMLU benchmark dataset. ‘STEM’, ‘Humanities’, ‘Social Sciences’, ‘Others’ belong to Arabic MMLU.
Evaluation Results on Arabic MMLU Benchmark Dataset
- Downloads last month
- 11