Tag
User reports a CUDA error when using tensor split mode with the latest llama.cpp and Qwen-3.6-27b model on dual RTX 3090s with Ubuntu Server 24.04 and Docker.
Reddit user reports Qwen 3.6-27B shows unusually proactive agent behavior, autonomously building, testing and fixing code without prompting.