Cheng Zeyi

Transliterated name · original name retained below

Cheng Zeyi graduated in computer science from Wuhan University and focuses on generative AI inference optimization and system performance engineering. As a technical expert at Alibaba, led design and deployment of inference acceleration frameworks widely used in image recognition and AIGC. At a Silicon Valley AI infrastructure company, developed image-generation inference engines with industry-leading performance and key support for commercialization. Authors multiple open-source optimization projects under GitHub username @chengzeyi, with thousands of stars. His Comfy-WaveSpeed image and video inference plugin is widely used in the ComfyUI community, delivering 2-fold acceleration across hardware platforms. ParaAttention supports multi-GPU parallel DiT inference and is used by fal.ai and Replicate. His stable-fast project is a reference for Hugging Face Diffusers inference optimization, with more than 1000 stars and adoption by multiple community projects. As an engineer-founder, he uses open source to improve and broaden access to the AIGC inference ecosystem.