Masahiro Tanaka

LLM-jp, Akiko Aizawa, Eiji Aramaki et al.

This paper introduces LLM-jp, a cross-organizational project for the research and development of Japanese large language models (LLMs). LLM-jp aims to develop open-source and strong Japanese LLMs, and as of this writing, more than 1,500 participants from academia and industry are working together for this purpose. This paper presents the background of the establishment of LLM-jp, summaries of its activities, and technical reports on the LLMs developed by LLM-jp. For the latest activities, visit https://llm-jp.nii.ac.jp/en/.

Masaaki Yamauchi, Masahiro Tanaka, Yuichi Ohsita et al.

Internet-of-things (IoT) devices are vulnerable to malicious operations by attackers, which can cause physical and economic harm to users; therefore, we previously proposed a sequence-based method that modeled user behavior as sequences of in-home events and a base home state to detect anomalous operations. However, that method modeled users' home states based on the time of day; hence, attackers could exploit the system to maximize attack opportunities. Therefore, we then proposed an estimation-based detection method that estimated the home state using not only the time of day but also the observable values of home IoT sensors and devices. However, it ignored short-term operational behaviors. Consequently, in the present work, we propose a behavior-modeling method that combines home state estimation and event sequences of IoT devices within the home to enable a detailed understanding of long- and short-term user behavior. We compared the proposed model to our previous methods using data collected from real homes. Compared with the estimation-based method, the proposed method achieved a 15.4% higher detection ratio with fewer than 10% misdetections. Compared with the sequence-based method, the proposed method achieved a 46.0% higher detection ratio with fewer than 10% misdetections.

Masahiro Tanaka, Kenjiro Taura, Toshihiro Hanawa et al.

This work proposes RaNNC (Rapid Neural Network Connector) as middleware for automatic hybrid parallelism. In recent deep learning research, as exemplified by T5 and GPT-3, the size of neural network models continues to grow. Since such models do not fit into the memory of accelerator devices, they need to be partitioned by model parallelism techniques. Moreover, to accelerate training for huge training data, we need a combination of model and data parallelisms, i.e., hybrid parallelism. Given a model description for PyTorch without any specification for model parallelism, RaNNC automatically partitions the model into a set of subcomponents so that (1) each subcomponent fits a device memory and (2) a high training throughput for pipeline parallelism is achieved by balancing the computation times of the subcomponents. In our experiments, we compared RaNNC with two popular frameworks, Megatron-LM (hybrid parallelism) and GPipe (originally proposed for model parallelism, but a version allowing hybrid parallelism also exists), for training models with increasingly greater numbers of parameters. In the pre-training of enlarged BERT models, RaNNC successfully trained models five times larger than those Megatron-LM could, and RaNNC's training throughputs were comparable to Megatron-LM's when pre-training the same models. RaNNC also achieved better training throughputs than GPipe on both the enlarged BERT model pre-training (GPipe with hybrid parallelism) and the enlarged ResNet models (GPipe with model parallelism) in all of the settings we tried. These results are remarkable, since RaNNC automatically partitions models without any modification to their descriptions; Megatron-LM and GPipe require users to manually rewrite the models' descriptions.