AIHacker Noonabout 7 hours ago
GLM-4.7-Flash Crawls on Mini PC
6 min read
GLM-4.7-Flash, a 30B sparse mixture-of-experts model, ran at only 7 tokens per second on a Minisforum UM790 Pro mini PC. The bottleneck was the runtime mishandling Multi-head Latent Attention, not the model weights. A CPU-only llama.cpp fork with explicit MLA mode restored performance.
Level
Hype check
Tap to vote and see what everyone thinks.
