TAIPEI, Aug. 14, 2024 /PRNewswire/ — As AI technology continues its rapid advancement, the demand for traditional Chinese models in the Taiwanese market is increasing. In response to this need, iKala, a leading AI transformation solutions provider, has unveiled its new traditional Chinese validation set, “TMMLU+.” This innovative tool has been officially featured in the inaugural issue of the prestigious Conference on Language Modeling (COLM).
The conference was founded by renowned experts from DeepMind, Meta, Microsoft, and Ivy League institutions, is a highly respected authority in the field of AI. The TMMLU+ validation set achieved a top 10% score during the review process of numerous papers, highlighting iKala’s terrific expertise in AI technology.
iKala Launches TMMLU+ Validation Set to Enhance Accuracy and Localization of Traditional Chinese LLMs
The launch of TMMLU+ validation set is poised to significantly boost the accuracy and localization of large language models for traditional Chinese. With the rise of generative AI, nations worldwide are focusing on developing language models that cater to their specific linguistic needs. In Taiwan, several companies are working on creating traditional Chinese AI models. Given the linguistic and cultural nuances unique to Taiwan, such as terms like “8+9,” “Beicheng,” and local Taiwanese dialects, specialized validation sets are essential for fine tuning these models to achieve precise and relevant AI applications.
iKala’s AI team has leveraged the TMMLU framework from MediaTek and expanded it using social media data to encompass a broader array of Taiwanese local usage and cultural references. This enhancement allows the TMMLU+ validation set to better reflect the linguistic context and expressions specific to traditional Chinese. The open source TMMLU+ validation set is designed to support local enterprises in validating their traditional Chinese language models, ensuring they will be fine tuned to meet the needs.
iKala Traditional Chinese Validation Set TMMLU+ Stands Out Globally,
On October 23, the AI team began to develop TMMLU+ validation set which underwent intensive work over several months. It had been submitted to COLM for review in the spring of 2024. During the process of review, iKala’s AI team provided a comprehensive explanation of the unique characteristics in traditional Chinese and the value of TMMLU+. This presentation significantly impressed the reviewers and effectively enhanced the TMMLU+ validation set’s paper rating, allowing it to stand out from global competitors.
The TMMLU+ validation set has already been adopted by several Taiwanese companies for evaluating large language models, establishing a reliable and standardized benchmark for the tech industry in Taiwan. This recognition not only affirms iKala’s years of dedication to advancing AI technology but also enhances its influence and authority in the industry.
iKala Co-founder and CEO Sega Cheng commented, “As a multinational AI company based in Taiwan, iKala is dedicated to solving complex and meaningful AI challenges, with a focus on delivering business value and practical applications. We are extremely grateful for the efforts of our AI team for their hard work, which has allowed us to gain international recognition. We also extend our thanks to MediaTek for leading the way in Taiwan’s AI open source technology.”
【About iKala】
iKala is a leading AI transformation solutions provider, with a mission to “enable AI competencies” of enterprises by providing AI-driven cloud management and MarTech solutions, to optimize their operational efficiency and increase customer engagement. iKala’s solutions and SaaS products are available in 7 countries, enabling over 1,000 enterprises and top-tier brands, including Fortune 500 companies, to transform their business.
iKala Official Website: https://ikala.ai/zh-tw/
View original content to download multimedia:https://www.prnewswire.com/apac/news-releases/ikalas-traditional-chinese-validation-set-tmmlu-gains-top-ai-conference-endorsement-showcasing-ikalas-expertise-302222047.html
SOURCE iKala
