Following on from its successful initial HAT-style AI add-on board for the Raspberry Pi 5 the foundation has release a second generation product, namely the $130 Raspberry Pi AI HAT+ 2.
This new, more powerful board can run a range of Generative AI models directly on the device, ensuring data privacy and security, while eliminating the need for subscriptions to expensive cloud-based AI services.
A new Hailo-10H neural network accelerator delivers 40 TOPS of inferencing performance with low latency. And with 8GB of dedicated on-board RAM it can handle more sophisticated language (LLM) and vision (VLM) tasks.
Other benefits include tight integration with the Pi Camera software stack (libcamera, rpicam-apps and Picamera2). In fact, you can train a model with custom image datasets and the Hailo Dataflow Compiler.
The initial models availability list - DeepSeek-R1-Distill, Llama3.2, Qwen2.5-Coder, Qwen2.5-Instruct and Qwen2 - will be supplimented by others during 2026.
Visit Hailo’s GitHub for examples, demos, frameworks and and GenAI-based applications, such as VLMs, voice assistants, and speech recognition.
No comments:
Post a Comment