The Laya model, a 421‑million‑parameter network, was converted to OpenVINO format and quantized to INT8 for CPU inference. Running on an Intel Core i7‑12th generation processor, the model successfully controlled Flappy Bird in real time. This demonstrates efficient low‑precision deployment of large language models on consumer‑grade CPUs.