
Longsys runs a 397B-parameter AI model derived from Alibaba's Qwen 3.5 A17B on a Ryzen AI Halo with 16-core Ryzen AI Max+ 395 and 128GB RAM. The system uses custom SPU and iSA to compress data in real time, offloading unused experts to a fast storage buffer.
Tap to vote and see what everyone thinks.
Summary by ByteBrief
4 tiny Docker containers for reliable homelabs