-
Notifications
You must be signed in to change notification settings - Fork 722
Arm backend: Add pypi package install option for VGF backend #15551
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
For Linux/Windows x86 machines, PyPi packages of MLSDK repository for VGF backend may be used. This will eventually be the default. Reason it is not yet default is because of a limitation of model-converter to handle large models. Signed-off-by: Måns Nilsson <mans.nilsson@arm.com> Co-authored-by: Per Held <per.held@arm.com>, Ryan O'Shea <ryan.oshea3@arm.com> Change-Id: I292bf74a3db9c129ac9bc7550575cfae6f85e9a0
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15551
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Cancelled Job, 5 Unrelated FailuresAs of commit 3a7584f with merge base dcc9a91 ( CANCELLED JOB - The following job was cancelled. Please retry:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
Hi @SS-JIA this is a PR you might want to test internally if you tests VGF backend. |
|
@SS-JIA Some background: This is needed to be able to add test for VGF backend, currently installing and building VGF tools from source take too long on GitHub to be useful (e.g. jobs times out), but with this we can install a pip packages for most of it and we hope it will be fast enough (like minutes or even seconds) to be able to start adding unit/trunk tests and maybe even the new backend suit tests for VGF. :) |
|
Failed tests are unrelated |
…#15681) Reverts #15551 This is causing internal test failures: * Error outside of ExecuTorch: ModuleNotFoundError: No module named 'executorch.backends.arm.vgf.model_converter' Errors inside of ExecuTorch: * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:cos * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:custom_partition * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:debug_hook * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:decompose_avg_pool2d_pass * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:decompose_cosine_similarity_pass * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:decompose_div_pass * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:decompose_layernorm_pass * buck test '@fbcode//mode/opt' fbcode//executorch/backends/arm/test:decompose_linalg_vector_norm_pass
|
Go ahead we have not merged the followup PRs that start using this yet so it will probably not be that messy, e.g. only need to rever this so get stuff back to a working state and we reinvestigte, sorry it broke stuff. |
|
Thanks a lot @metascroy for adding the internal fails to the revert PR, with a bit of luck it gives some help resloving this.
|
New setup option is added: --install-mlsdk-deps-with-pip
For Linux/Windows x86 machines, PyPi packages of MLSDK repository for VGF backend may be used.
This will eventually be the default. Reason it is not yet default is because of a limitation of model-converter to handle large models. The new option will decrease setup time, which can enable VGF backend testing in github.
Co-authored-by: Per Held <per.held@arm.com>, Ryan O'Shea <ryan.oshea3@arm.com>
cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai