What is EXAONE 3.5?
EXAONE 3.5 is a series of advanced AI models developed by LG AI Research. These models are known for their superior performance and cost-effectiveness. They excel in areas like model training efficiency, text processing, long text understanding, and instruction following. The development process adheres to LG’s AI ethics principles, ensuring responsible use.
Who Can Use EXAONE 3.5?
EXAONE 3.5 is ideal for researchers, developers, and enterprise users. Researchers can utilize these models for cutting-edge AI research. Developers can integrate them into applications to enhance user experience. Enterprises can leverage these models to boost productivity and innovate services.
Where Is EXAONE 3.5 Being Used?
Korean companies Polaris Office and Hancom are exploring ways to use EXAONE 3.5-based AI services to improve their software offerings.
A Proof of Concept (PoC) project integrates EXAONE 3.5 into Hancom Office to enhance work efficiency in public institutions.
EXAONE 3.5 shows strong performance in math and coding tasks, achieving competitive results across nine general benchmarks.
What Makes EXAONE 3.5 Special?
Long Text Understanding: EXAONE 3.5 handles up to 32K tokens of context, outperforming similar global models.
Instruction Following: EXAONE 3.5 ranks first in seven benchmark tests for instruction following, surpassing comparable models.
Training Efficiency: Improvements in dataset deduplication and data anonymization enhance training performance.
Pollution Removal: Detailed pollution removal processes increase confidence in performance assessments.
Multilingual Support: Top-tier performance in English and Korean text understanding.
Business Partnerships: LG AI Research collaborates with multiple Korean and international companies to achieve tangible business outcomes.
Responsible AI: LG AI Research focuses on ethical practices during EXAONE 3.5 development, including AI ethics impact assessments.
How Do You Use EXAONE 3.5?
1. Access EXAONE 3.5 from LG AI Research's website or Hugging Face platform.
2. Choose the appropriate version (2.4B, 7.8B, or 32B) based on your needs.
3. Read and understand the technical report to grasp the model's capabilities and application scenarios.
4. Customize the model through fine-tuning to fit specific tasks or domains.
5. Leverage the model’s strengths in long text understanding and instruction following to develop or refine applications.
6. Follow LG AI Ethics Principles during development to ensure responsible AI usage.
7. Gather feedback and continuously optimize the model to meet evolving business requirements.