[pull] master from pytorch:master #167
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
![pull[bot]](https://avatars.githubusercontent.com/in/12910?s=60&v=4){kind=small}
Co-authored-by: Jane (Yuan) Xu <[email protected]>
Signed-off-by: Eli Uriegas <[email protected]>
We shouldn't be relying on conda-forge for our testing purposes since it contains packages that can conflict with the current way we build pytorch. Signed-off-by: Eli Uriegas <[email protected]>
Signed-off-by: Eli Uriegas <[email protected]>
See pytorch/pytorch#57190 for details Co-authored-by: driazati <[email protected]>
Otherwise, mkl-2021.2 gets installed which contains both .so and .so.1 binaries
Installing mkl-2021.2 renders numpy-11.1 unimportable on Windows: From da1928b ``` Traceback (most recent call last): File "<string>", line 1, in <module> File "C:\w\b\windows\conda\envs\py36\lib\site-packages\numpy\__init__.py", line 142, in <module> from . import add_newdocs File "C:\w\b\windows\conda\envs\py36\lib\site-packages\numpy\add_newdocs.py", line 13, in <module> from numpy.lib import add_newdoc File "C:\w\b\windows\conda\envs\py36\lib\site-packages\numpy\lib\__init__.py", line 8, in <module> from .type_check import * File "C:\w\b\windows\conda\envs\py36\lib\site-packages\numpy\lib\type_check.py", line 11, in <module> import numpy.core.numeric as _nx File "C:\w\b\windows\conda\envs\py36\lib\site-packages\numpy\core\__init__.py", line 14, in <module> from . import multiarray ImportError: DLL load failed: The specified module could not be found. -- Could NOT find NumPy (missing: NUMPY_INCLUDE_DIR NUMPY_VERSION) ```
* Add github_analyze So far, very simple script only capable of counting month commits, authors and reverts * Add option to analyze reverts * Add `Authorization` option to `fetch_json`
PyTorch conda package no longer have a runtime numpy dependency, as binding is dynamic, but caffe2 still needs it
* update install_miopen.sh for rocm 4.2 * update manywheel/build_all_docker.sh for rocm 4.2 * add rocm4.2 to update_s3_htmls.sh * fix copy/paste error for rocm in manywheel/build_all_docker.sh * add rocm 4.2 to build-manywheel-images.yml action
Numpy is an optional dependency for PyTorch, but not for caffe2
Prints duplicated symbols between various libraries For example: ``` build/lib/libtorch_python.dylib symbols size 15.81 Mb libshm.dylib overlap is 2 Kb libtorch_cpu.dylib overlap is 1.12 Mb libc10.dylib overlap is 32 Kb build/lib/libshm.dylib symbols size 31 Kb libtorch.dylib overlap is 0 bytes libtorch_cpu.dylib overlap is 2 Kb libc10.dylib overlap is 2 Kb build/lib/libtorch.dylib symbols size 0 bytes build/lib/libtorch_cpu.dylib symbols size 103.98 Mb libc10.dylib overlap is 134 Kb ```
* Install ninja in manywheel images * pip isn't there, so use yum * push 11.3 images * remove WITH_PUSH as we don't have permissions anyway
* enable credentials for push * move authentication step, add paths guarding * document WITH_PUSH == true condition
Loooking at this failing CI job: https://app.circleci.com/pipelines/github/pytorch/pytorch/368932/workflows/1f95e9a9-04d4-4d30-866c-95c60bd62cc7/jobs/15575031 I see that even though `DESIRED_PYTHON` is 3.7m, the actual python on path is 3.9 from conda. This happens with all libtorch nightly builds. The build scripts add `/opt/python/$DESIRED_PYTHON` to the path, but the directory doesn't exist so conda's python is used instead.
Follow up to the previous PR which didn't work. The previous PR was missing the `COPY` directive, so the command wasn't actually being run. Once I fixed this, the docker build failed because the manylinux scripts use yum instead of apt. So, I've factored the cpython build script into `common/install_cpython.sh` and added the `apt-get` calls into the Dockerfile.
The preprocess selector for this platform is actually osx
Add trailing '/' to subdir prefix to avoid adding packages from `rocm4.0.1` to rocm4.0 filelist
Example: ``` ./github_analyze.py --missing-in-branch --branch release/1.9 --milestone-id 23 URL;Title;Status https://github.com/pytorch/pytorch/pull/64036;[torch/elastic] Pretty print the failure message captured by @record;closed pytorch/pytorch#63953 torch.distributed.run OMP_NUM_THREADS message to log.warning;closed pytorch/pytorch#63910 Add torch.distributed.is_torchelastic_launched() util method + make init_method=tcp:// compatible with torchelastic;closed https://github.com/pytorch/pytorch/issues/62586;[v.1.9.1] Release Tracker ;open pytorch/pytorch#62378: Fix elastic launch doc;closed pytorch/pytorch#62139 option to skip GH validation for torch.hub;closed pytorch/pytorch#62072 infinite loop in `_validate_not_a_forked_repo()`;closed pytorch/pytorch#61394 mm not correctly report TORCH_CHECK failure issue;closed https://github.com/pytorch/pytorch/pull/61294;[torch] Various improvements to `torch.distributed.launch` and `torch.distributed.run` (#60925);closed pytorch/pytorch#60754 docs need an urgent serious update;closed pytorch/pytorch#60752 `torch.distributed.launch` issues in 1.9.0;closed https://github.com/pytorch/pytorch/issues/60717;torch.distributed.run future warning in 1.9.0;closed https://github.com/pytorch/pytorch/issues/60716;torch.distributed runs under INFO log level in 1.9.0;closed pytorch/pytorch#60318 Caffe2 thread-pool leak warning;closed pytorch/pytorch#60059 warning on .names() access in max_pool2d and max_pool2d_backward;closed pytorch/pytorch#58849 git stamp from release documentation version;open pytorch/pytorch#57555 function for skipping module parameter / buffer initialization;closed ```
* move 2 binaries to S3 * minor change * only update download URL
* improve handling windows update * Update windows/internal/cuda_install.bat Co-authored-by: Nikita Shulga <[email protected]>
* analytics: Add ability to pass remote Adds the ability to pass remote ass an arg to the cli, also adds a log message to note which remote is being used Signed-off-by: Eli Uriegas <[email protected]> * Update analytics/github_analyze.py Co-authored-by: Nikita Shulga <[email protected]> Co-authored-by: Nikita Shulga <[email protected]>
Ping libcxx to 10.0.0 when cross-compiling for M1
This reverts commit 580b3b9.
Installing llvm-9 while cross-compiling alters include order, which results in wrong libuv being used
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot]
Can you help keep this open source service alive? 💖 Please sponsor : )