I don’t have access to high-end hardware to run models but I have multiple personal computers that can add up together. since tvm can compile model for multiple hardware and support for popular libraries, it would be good solution to solve this by splitting the original model into small pieces and run it across different machines connected through network. walking through tvm source code and documentation, I found a discussion titled [RFC] Compute graph pipeline with new subgraph executor which I think is pretty close to what I want but corresponding RFC was outdated if I’m not mistaken, contributed code was moved to [RFC] Compute graph pipeline with new subgraph executor) but I can’t make it work either