Li Auto Launches In-House Cloud Inference Chip Project with Dataflow Architecture

Li Auto is exploring the development of its own cloud-based inference chips. The project is still in its early stages and will adopt the same dataflow architecture used in its autonomous driving chips. These cloud inference chips could take over certain inference tasks previously handled by GPUs, such as data processing, testing, and simulation for autonomous driving models, as well as request handling for large language models. However, parameter updates involved in model training will still rely on dedicated training chips.