NEC Innovation Challenge 2023

Deliver the solutions of tomorrow – NEC is looking for startups with LLM, Healthcare, Smart City, and Urban Agriculture solutions to collaborate together!

Large Language Model

 

   This year, NEC has developed its own generative AI Large Language Model (LLM) for the Japanese language. Although there are other LLMs all over the world today, most existing LLMs are trained mainly in English, and there are almost no LLMs that can be customized for use in a variety of industries while possessing high Japanese language ability.

   The development of NEC's new LLM by NEC, is expected to further accelerate the company’s business use and help improve corporate productivity. We believe that there is no limitation of expanding this LLM to other fields, such as healthcare, finance, marketing, robotics, etc.

 

 

Inspirations: Additional Context

Here are some additional background information on the nature of the current situation and challenge statement to help you come up with relevant solutions:   
 

This LLM, developed by NEC, is a general-purpose model, also known as a foundation model. It was developed using multilingual data which NEC collected and processed independently.  The LLM achieves high performance while reducing the number of parameters to 13 billion through unique innovations, which helps reduce power consumption, but also enables operation in cloud and on-premises environments due to its light weight and high speed.  

In terms of performance, the LLM has achieved world-class Japanese language proficiency based on a  Japanese knowledge and reading comprehension benchmark, which corresponds to the model's high reasoning ability.

NEC has already begun using this LLM for internal operations, applying it  for common operations and for improving the efficiency of various tasks, such as document creation and coding support. 
 
In recent years, generative AI such as "ChatGPT" has been attracting worldwide attention, and its use is rapidly increasing across a wide range of industries. However, most existing LLMs are trained mainly in English, and there are almost no LLMs that can be customized for use in a variety of industries while possessing high Japanese language ability. We believe that the development of this new LLM is expected to further accelerate business utilization and contribute to increased corporate productivity.
 

NEC's LLM features
1. High Japanese language proficiency 
In order to use LLM in actual business, high performance is required in terms of the amount of Japanese language knowledge and the document reading comprehension.  

NEC’s LLM was evaluated using JGLUE, a standard Japanese Language comprehension benchmark in the field of natural language processing. (*1) NEC’s LLM achieved the world-top level performance at 81.1% regarding the amount of knowledge and 84.3% regarding the document reading comprehension. It is expected to perform well in a variety of industries.

 




2. Light weight
This LLM has high performance, yet NEC's  technology has reduced the size of the model to a compact size with 13 billion parameters. While conventional LLMs with high performance require many GPUs, this LLM can be run on a standard server with a single GPU. As a result, business applications that incorporate the LLM can respond well, which helps reduce power consumption and the cost of servers during business operations. In addition, the LLM can be easily built in a short period of time and can be run in a customer's on-premises environment, making it safe even for highly confidential operations. 

The size of the parameter is often used to feature one of the characteristics of its LLM. . However, increasing the parameter size leads to a decrease in inference speed, as well as an increase in the number of GPUs and the power consumption required to operate the model. Consequently, if the same performance can be provided, it is desirable to have fewer parameters. In this study, NEC focused on the fact that LLM performance depends not only on parameter size, but also on the amount of high-quality data used for training and training time. As a result, NEC achieved high performance by limiting the parameter size to the range that can be operated with a single GPU, and by using a large amount of both data and calculation time.

NEC independently developed the largest supercomputer for AI research among Japanese companies, and it has been fully operational since March 2023. By utilizing this supercomputer, NEC was able to construct a high-performance LLM in a short period of time as within approximately one month.

Going forward, NEC aims to leverage the foundation model in order to actively promote the development of LLM for each company using their closed customer data. NEC also plans to improve the performance of the foundation model itself, and to apply these technologies to practical uses as soon as possible through the NEC Generative AI Hub. 

(*1) A JCommonSenseQA dataset was performed for these questionnaires. Although Fine-tune from the training data was not performed, NEC evaluated the results of predictions from 3 examples for each LLM at the In-Context Learning setting which provides few examples during reasoning. For the evaluation of the document reading  comprehension, a JSQuAD dataset was performed, and 2 examples were used at the In Context Learning setting. An Exact Match score was used for evaluation parameters. “The Top LLM '' (Japanese) was selected as the best performing model among the opened Japanese LLMs, whose evaluation scores are either obtained from experiments conducted by NEC, and the closed LLMs which were evaluated with experiment reports conducted at the same setting as NEC’s.

 

 

Sponsor of This Theme