Did you split the model into small parts and then your custom webgpu kernel loads them if im not mistaken?
· Sign up or log in to comment
