We’re happy to release the QAST dataset that was used to support the experiments in our ICLR 2019 workshop paper: Simulating Execution Time of Tensor Programs using Graph Neural Networks.
This dataset contains ASTs and their associated execution time for the Resnet-18 conv2d operators on an Intel Xeon CPU E5-1620, all measured using TVM. We hope this new regression task on graphs will benefit the graph research community and raise interest in Optimizing Compiler research. For simplicity, the dataset is fully numpy based and doesn’t require TVM to be used.
We’d like to take this opportunity to thank the TVM team for their work and enabling new and exciting Optimizing Compiler research.
If you have any question about the paper or the dataset, feel free to ask.