MachinaCheck在AMD MI300X上运行多代理CNC分析系统
machinacheck利用amd mi300x硬件上的本地qwen 2.5 7b模型分析cad文件的可制造性，无需将数据发送至云端。
developers built machinacheck to solve a bottleneck in cnc manufacturing where managers spend hours manually reviewing drawings for feasibility. the application uses a local llm deployment on amd mi300x, utilizing 192gb of vram to run qwen 2.5 7b via vllm. this ensures sensitive cad geometry never leaves the customer's infrastructure. the pipeline combines mathematical cad parsing with llm reasoning to classify operations and match tools. by keeping the inference on-premise, the system adheres to strict nda requirements common in aerospace and medical device manufacturing.