Related Experiment Video
Updated: May 21, 2025
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
1.3K
Localized large language model TCNNet 9B for Taiwanese networking and cybersecurity
Jiun-Yi Yang1,2, Chia-Chun Wu3
1Department of AI & Data, Hi5 Technology Incorporation, Taipei, Taiwan.
Scientific Reports
|March 21, 2025
Summary
TCNNet-9B, a specialized Traditional Chinese language model, significantly improves networking industry tasks in Taiwan. This domain-specific model shows enhanced accuracy and relevance, outperforming baseline models.
Area of Science:
- Artificial Intelligence
- Natural Language Processing
- Computer Science
Background:
- Large language models (LLMs) often lack domain-specific knowledge and localization for specialized industries.
- The Taiwanese networking industry has unique requirements not fully met by general-purpose LLMs.
Purpose of the Study:
- To introduce TCNNet-9B, a specialized Traditional Chinese language model tailored for the Taiwanese networking sector.
- To evaluate the performance and practical efficacy of TCNNet-9B compared to baseline models.
Main Methods:
- Developed TCNNet-9B based on the Yi-1.5-9B architecture.
- Conducted extensive pretraining and instruction finetuning using a curated dataset of networking information and localized regulations.
- Evaluated performance using custom benchmarks across English, Traditional Chinese, and Simplified Chinese.
Main Results:
- TCNNet-9B achieved a 2.35-fold improvement in Q&A accuracy.
- Demonstrated a 37.6% increase in domain expertise comprehension.
- Showcased a 29.5% enhancement in product recommendation relevance.
- Successfully integrated into an intelligent sales advisor system.
Conclusions:
- Domain-specific adaptation and localization are crucial for enhancing LLM performance in non-English and specialized contexts.
- TCNNet-9B provides a valuable reference for developing vertical-specific language models.
