Fix some bugs in onnxmlirtorch package #3319
Conversation
Signed-off-by: Tung D. Le <[email protected]>
| add_custom_target(OMCreateONNXMLIRTorchPackage | ||
| COMMAND cp -r ${CMAKE_CURRENT_SOURCE_DIR}/onnxmlirtorch ${CMAKE_CURRENT_BINARY_DIR} | ||
| COMMAND cp ${ONNX_MLIR_BIN_ROOT}/${CMAKE_BUILD_TYPE}/lib/PyRuntimeC.*.so ${CMAKE_CURRENT_BINARY_DIR}/onnxmlirtorch/src/onnxmlirtorch/ | ||
| COMMAND cp ${ONNX_MLIR_BIN_ROOT}/${CMAKE_BUILD_TYPE}/lib/PyRuntimeC.*.so ${CMAKE_CURRENT_BINARY_DIR}/onnxmlirtorch/src/onnxmlirtorch/libs |
There was a problem hiding this comment.
Missing libs folder so that OMExecutionSession was not found.
| def __init__(self, model_path, **kwargs): | ||
| self.debug = False | ||
| self.session = None | ||
| self.output_dir = tempfile.TemporaryDirectory() |
There was a problem hiding this comment.
We want to keep this temp folder, so move it to the class member. Otherwise, the temp folder is cleaned up automatically.
|
Jenkins s390x sometimes failed because of a bug that is fixed in #3302. |
Signed-off-by: Tung D. Le <[email protected]>
Signed-off-by: Tung D. Le <[email protected]>
| cache_size = 3 | ||
|
|
||
|
|
||
| glocalSessionCache = SessionCache(config.cache_size) |
There was a problem hiding this comment.
Make it global. Otherwise, the cache is local to each pytorch call to the backend.
|
Jenkins Linux s390x Build #17586 [push] Fix some bugs in onnxmli... started at 01:01 |
|
Jenkins Linux amd64 Build #17579 [push] Fix some bugs in onnxmli... started at 00:01 |
|
Jenkins Linux s390x Build #17586 [push] Fix some bugs in onnxmli... passed after 1 hr 33 min |
|
Jenkins Linux amd64 Build #17579 [push] Fix some bugs in onnxmli... passed after 1 hr 42 min |
3 participants
I have tried to run an example
torch_compile_add.pythat usestorch.compile(mod, backend=onnxmlirtorch.onnxmlir_backend. Compilation is done in the local environment, not a docker image. I found some errors such as:and
This patch is to fix such errors and now
torch_compile_add.pyis runnable in my environment.