Use the following command for the default installation, or check out the comprehensive installation guide if your needs are more complex.
python -m pip install \ iree-compiler \ iree-runtime \ iree-tools-tf \ iree-tools-tflite \ iree-tools-xla
See end-to-end examples of how to execute a variety models on IREE. This covers the import, compilation, and execution of the provided model.
Importing from other frameworks is planned - stay tuned!
Check out the samples in IREE's samples/colab/ directory, as well as the iree-samples repository, which contains workflow comparisons across frameworks.
Importing models takes known file types and imports into a form that the core IREE compiler is able to ingest. This import process is specific to each frontend and typically involves a number of stages:
This fully legalized form can then be compiled without dependencies on the source model language.
During compilation we load an MLIR file and compile for the specified set of backends (CPU, GPU, etc). Each of these backends creates custom native code to execute on the target device. Once compiled, the resulting bytecode is exported to an IREE bytecode file that can be executed on the specified devices.
The final stage is executing the now compiled module. This involves selecting what compute devices should be used, loading the module, and executing the module with the intended inputs. For testing, IREE includes a Python API. However, on mobile and embedded devices you will want to use the C API.