very nice, but I wish it had more RAM
The AI HAT+ 2 was easy to add to my Raspberry Pi 5. Adding some AIs was a bit more complicated. Luckily I used my laptop to log on to Claude AI online and it assisted me in setting up the AI HAT+ 2. I downloaded 5 AIs from Hailo: deepseek_r1_distill_qwen:1.5b - reasoning llama3.2:3b - general chat qwen2.5-coder:1.5b - coding qwen2.5-instruct:1.5b - instruction following qwen2:1.5b - older baseline They are very small models, and not very smart, but they do work and are interesting. I was surprised at how fast they run. I was fully expecting them to be excruciatingly slow, but they output faster than my reading aloud speed. Don't expect the models to be much good. They're mainly for experimenting with and learning. I'm hoping to put a larger model on there in the near future. Hailo produce a tool that converts an AI model into one suitable for their chip. I just wish they made the board with more memory. 8GB is too small for LLMs (large language models). 32GB would have been nice. But of course it would have been much more expensive. Three things I have yet to experiment with, but will soon, are AI computer vision, AI computer speech (both input and output), and robotics. I think the small RAM size will be much less of a limitation there.
Miriam English
