commit | 4f5d55beec79cfa7a27c95c823d4255f716b492e | [log] [tgz] |
---|---|---|
author | Scott Todd <scotttodd@google.com> | Tue Apr 25 09:26:30 2023 -0700 |
committer | GitHub <noreply@github.com> | Tue Apr 25 09:26:30 2023 -0700 |
tree | f4bace898551ac3d6dd374a07527743549faf480 | |
parent | eca606e48ef900d3723a1e211893538dc9864ec7 [diff] |
Handle supported ImportOptions in tf.py and fix Windows compatibility. (#13287) This adds back support for `import_only` and `save_temp_iree_input` to our TensorFlow `compile_saved_model` API. I also removed unsupported options (`import_extra_args`, `save_temp_tf_input`, `save_temp_mid_level_input`, and `use_tosa`). Those flags were dropped in https://github.com/openxla/iree/pull/12758 / https://github.com/openxla/iree/pull/13025 , but they are still useful in Colab notebooks and when debugging tests. --- Progress on https://github.com/openxla/iree/issues/13148, though some further updates will be needed to our Colab notebooks, such as ```python # before: compiler_module = tfc.compile_module( EdgeDetectionModule(), import_only=True, import_extra_args=["--output-format=mlir-ir"]) print("Edge Detection MLIR: ", compiler_module.decode('utf-8')) # after: compiler_module = tfc.compile_module( EdgeDetectionModule(), import_only=True) print("Edge Detection MLIR: ", compiler_module) ``` --- I developed this change on Windows (Yes! Finally, I can use Python that touches TF on Windows without needing to build TF from source!), where I found that this pattern is broken: ```python with tempfile.NamedTemporaryFile(mode="w") as temp_file: __main__.import_saved_model(output_path=temp_file.name, ``` See https://stackoverflow.com/a/23212515 - `NamedTemporaryFile` _creates and opens_ the file, and the file _cannot be opened again_... on Windows (it can be opened again on Unix). I used the trick from https://stackoverflow.com/a/45803022 to work around this: ```python with tempfile.TemporaryDirectory() as tmpdir: # ... # Not saving the file, so generate a loose temp file without tfs. tf_iree_input = os.path.join(tmpdir, 'tf-iree-input.mlir') ```
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.