- The document describes the process of compiling an ONNX model to XCVM using the Chainer compiler. It involves parsing the ONNX model, applying optimization passes like fusion and constant propagation, lowering it to an XCVM program, and then executing the program on XCVM to run the model.
- The compiler represents the model as a Model object containing a Graph. Nodes in the graph are represented by Node objects. Values are represented by Value objects. Tensor data is stored in Tensor objects.
- The XCVM execution engine takes the compiled XCVM program and runs it, interpreting each XCVM operation by calling methods on classes like ConvOp that implement the operations.