During the process of configuring the IREE environment, I encountered a multitude of issues. There are two main reasons for these problems. First, the official documentation for IREE is quite limited. The documentation on the official website and in the repository is incomplete and not up to date. Second, IREE itself is still in the development phase, and some bugs are continually being fixed. Therefore, some bugs have not been resolved yet. Additionally, apart from convolution operators, which are faster compared to PyTorch's native inference speed, most other operators can be considered as performance drawbacks.
During the process of configuring the IREE environment, I encountered a multitude of issues. There are two main reasons for these problems. First, the official documentation for IREE is quite limited. The documentation on the official website and in the repository is incomplete and not up to date. Second, IREE itself is still in the development phase, and some bugs are continually being fixed. Therefore, some bugs have not been resolved yet. Additionally, apart from convolution operators, which are faster compared to PyTorch's native inference speed, most other operators can be considered as performance drawbacks.