Fix typo for auto_detect_ep function #40
Conversation
Signed-off-by: Wang, Mengni <[email protected]>
|
it seems the CI failure is due to latest torch can't export > 2GB model to ONNX model: |
|
@hshen14 it's been 4 days since the provided changes were approved, please merge |
@Crafter-san，since there is no available solution to fix export issue on latest Torch currently, we are trying to find out the proper Torch version, which will be limited in CI test. The PR will be merged soon after this walk around applied in CI |
Signed-off-by: Wang, Mengni <[email protected]>
|
Unit Test ONNXRT baseline runs the unittest of main branch, its failure makes sense. |
|
@thuang6 here is the possible offender for the size limit, according to another user: If you could share this with the rest of the team, you guys could possibly go through and fix the issue and make a pull request. I don't have enough Python experience to fix the issue myself. |
|
additionally, the issue is within it requiring a filename to properly write it to a file. I don't know how, but perhaps including an arg for filename, and a default filename (for people using things like Optimum CLI that don't have access to the direct args) |
|
Pytorch issue is in the process of being fixed: pytorch/pytorch#140962 |
|
Reversion complete |
Hi, thank you for your reminder, we will remove the version control after torch latest version is released which contains this commit. |
|
@mengniwang95 when will it be fixed? There's no new versions/releases |
|
Hi @Crafter-san , this fix is in pytorch source code and now what we need to do is waiting for the new release of pytorch to fix this issue. If you failed to export onnx model due to this issue, you can downgrade your pytorch version or build pytorch from its main branch source code. |
Type of Change
bug fix