Qualcomm has acquired AI startup Edge Impulse. The startup offers an AI platform for training LLMs for edge devices and support for Qualcomm Dragonwing processors.
Recently acquired by Qualcomm, Edge Impulse offers a cloud-based platform that allows end users to train LLMs running on (IoT or IIoT) devices at the edge of the (enterprise) network, for example, in industrial environments.
These specific LLMs are often trained on (measurement) data from network sensors. The Edge Impulse cloud platform helps transform this data into training sets for LLMs.
In addition, specific tooling helps create data pipelines for automating workflows, especially workflows that aggregate sensor data into a dataset for training the model. Consider filtering, deduplication and removal of erroneous records.
Other platform capabilities
The platform also provides tools for feature engineering. This involves condensing raw sensor data into a format LLMs can process more easily, such as merging a collection of temperature measurement data into a single value that is easier to analyze.
The Edge Impulse platform further helps identify the right LLM for a specific project after creating a training dataset.
Also, due to the limited processing power of edge devices, the platform helps identify the right hardware requirements for different AI architectures. This makes it easier for companies to select the right LLMs to run on their edge devices.
Under the hood, an LLM algorithm trained in Edge Impulse is packaged in a C++ library instead of Python. This is because the “old” programming language offers more hardware efficiency. Ultimately, Edge Impulse claims that this could reduce the memory usage of an LLM by 60 percent.
By now, more than 170,000 developers worldwide are said to be already using the platform.
Brick ML device
In addition to its cloud-based software platform, Edge Impulse offers a dedicated edge device: the Brick ML. This compact rectangular computing device is designed specifically for running LLMs. It can be attached to devices in edge environments, collect sensor data, and analyze it locally with the onboard LLMs.
Support for Qualcomm Dragonwing processors
Qualcomm’s Edge Impulse acquisition offers new opportunities, such as for its recently launched range of Dragonwing processors for edge devices. Within this range are GPUs for running LLMs and processors that can operate at extremely low temperatures.
Edge Impulse, meanwhile, supports the Qualcomm Dragonwing processors QCS6490 and QCS5430. Both processors are suitable for rugged mobile devices, and the latest Dragonwing processor is additionally suitable for robotic solutions and applications.
The AI startup plans to support more Qualcomm Dragonwing processors later and edge hardware from other partners, such as MCUs, CPUs, GPUs, and NPUs.
No financial details about the acquisition were disclosed.
Also read: Qualcomm stretches life of Snapdragon 8 Elite: eight years of support