AMD Supports Local Operation of Alibaba's Qwen 27B Model

AMD has announced that it will support the local operation of Alibaba's Qwen 27B model on its AI PCs and workstations from the day of launch. This initiative aims to provide an environment for running large-scale open-source AI models without cloud access. The Qwen 27B model is a dense model with 27 billion parameters, optimized for code generation, practical tasks, research, and long-context AI applications. The operating environment is based on the open-source inference framework llama.cpp, and it supports AI PCs and workstations equipped with AMD processors, as well as a single AMD 32GB graphics card. Initial tests showed local inference performance of up to 24.5 tokens per second on the Ryzen AI Max Plus 395 processor and up to 51.8 tokens per second on a single Radeon AI Pro R9700 GPU. 'Day 0' support means that a support system is prepared to allow immediate execution in specific hardware and software environments upon the model's release. Alibaba is making the Qwen series models available as open-source for developers, research institutions, and businesses. This support is part of the competition in AI semiconductors and local AI infrastructure, as AMD expands its ecosystem that allows for shared use of cloud and local environments.