Add torch.config.show(), reporting detailed version of all libraries. #18579

ezyang · 2019-03-28T19:26:46Z

Stack from ghstack:

Add torch.__config__.show(), reporting detailed version of all libraries. #18579 Add torch.config.show(), reporting detailed version of all libraries.

Fixes #111.

Example output:

PyTorch built with:
  - GCC 4.8
  - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20)
  - OpenMP 201107 (a.k.a. OpenMP 3.1)
  - NNPACK is enabled
  - CUDA Runtime 9.2
  - NVCC architecture flags: -gencode;arch=compute_52,code=sm_52
  - CudNN 7.10.2  (built against CUDA 9.1)
  - Magma 2.5.0
  - Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS=  -Wno-deprecated -fvisibility-inlines-hidden -fopenmp -O2 -fPIC -Wno-narrowing -Wall -Wextra -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -Wno-unused-but-set-variable -Wno-maybe-uninitialized, DISABLE_NUMA=1, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, USE_CUDA=True, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=True, USE_NNPACK=True, USE_OPENMP=ON,

Signed-off-by: Edward Z. Yang [email protected]

gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

Differential Revision: D14778507

…all libraries. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]>

soumith · 2019-03-28T20:05:00Z

this fixes #111

You might want to rename the methods slightly to match numpy, if you think it makes sense

ezyang · 2019-03-28T20:12:24Z

Makes sense to me!

dzhulgakov

Would be nice to unify it with Caffe2's build options here: caffe2/core/macros.h.in caffe2/core/common.cc

Specifically, if you follow new format, I'd add the following compilation flags:

BUILD_TYPE
CXX_FLAGS
BLAS (as we will likely replace the ugly THBlas soon)
USE_NNPACK
USE_FBGEMM (as we're adding quantization support now) - and include its version

dzhulgakov · 2019-04-01T21:28:21Z

torch/csrc/Module.cpp

 #endif
 }

+static PyObject *THPModule_detailedVersion(PyObject *module)


nit: shall we start using pybind universally?

For things like this one sure, but I don't trust pybind enough to feel ok with using it for our performance-critical eager functions.

ezyang · 2019-04-04T15:40:25Z

Would be nice to unify it with Caffe2's build options here: caffe2/core/macros.h.in caffe2/core/common.cc

Thanks, was not aware of this.

ezyang · 2019-04-04T15:46:50Z

USE_FBGEMM (as we're adding quantization support now) - and include its version

fbgemm does not appear to have any useful versions. Determined by looking at Fbgemm.h and searching for configure_file in the cmake

… all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

…ailed version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

…iled version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

…sion of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

ppwwyyxx · 2019-04-05T07:58:58Z

For cuda/cudnn, etc, could build-time version be different from run-time version in its minor version number or patch number?
If yes, it would be great to have access to both build-time version and run-time version available for debugging (perhaps with cudnnGetVersion & cudaRuntimeGetVersion)

ezyang · 2019-04-05T14:24:59Z

@ppwwyyxx Makes sense to me!

… all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

ezyang · 2019-04-05T16:16:47Z

Now also with nvcc arch information

… all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

…detailed version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

…etailed version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

dzhulgakov

So, do you want to also include the following:
BUILD_TYPE
CXX_FLAGS
BLAS (as we will likely replace the ugly THBlas soon)
USE_NNPACK
USE_FBGEMM (as we're adding quantization support now) - and include its version

Even if FBGEMM doesn't have exported version, just the fact whether we build with it or not, same for NNPACK.

Even for MKL and others - I'd suggest to explicitly adding a line "MKL: none" to the output

ezyang · 2019-04-08T17:07:56Z

@dzhulgakov I'm assuming part of your comment is an accidental copy paste from the earlier comment.

Right now, I print all of the existing Caffe2 settings as part of the Build settings key, filtering out keys which are set to an empty string. Do you want each of these to be given their own line in the output? Also, are you asking to stop filtering out empty string keys? (I feel they add unnecessary noise to the output)

… detailed version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

dzhulgakov

Apologies, I was looking at the old code version (GH UI is confusing)

…ch.__config__.show(), reporting detailed version of all libraries." Add torch.__config__.show(), reporting detailed version of all libraries. Fixes #111. Example output: ``` PyTorch built with: - GCC 4.8 - Intel(R) Math Kernel Library Version 2019.0.3 Product Build 20190125 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v0.18.1 (Git Hash 7de7e5d02bf687f971e7668963649728356e0c20) - OpenMP 201107 (a.k.a. OpenMP 3.1) - CUDA Runtime 9.2.0 - CuDNN 7.1.2 ``` Signed-off-by: Edward Z. Yang <[email protected]> gh-metadata: pytorch pytorch 18579 gh/ezyang/48/head

vadimkantorov · 2019-04-09T18:21:42Z

would this be useful for collect_env script to report build time config (if torch can be loaded and not completely broken)?

…ies. (#18579) Summary: Pull Request resolved: pytorch/pytorch#18579 ghimport-source-id: 65124c95e49423de4ad1008c65e75057fea09b94 Differential Revision: D14778507 Pulled By: ezyang fbshipit-source-id: 1e4bb79f4800a116ce8fb7af2fefbd34da8d102c

facebook-github-bot · 2019-04-09T20:34:59Z

@ezyang merged this pull request in 29ea086.

ppwwyyxx · 2019-04-10T00:22:54Z

@ezyang

CudNN 7.10.2 (built against CUDA 9.1)

There is no cudnn 7.10.2.

Cudnn version numbers work a bit different from cuda. 7102 means 7.1.2.

ezyang · 2019-04-10T02:04:13Z

@ppwwyyxx Oops! Thanks for the catch.

…ies. (pytorch#18579) Summary: Pull Request resolved: pytorch#18579 ghimport-source-id: 65124c9 Differential Revision: D14778507 Pulled By: ezyang fbshipit-source-id: 1e4bb79f4800a116ce8fb7af2fefbd34da8d102c

ezyang requested a review from soumith March 28, 2019 19:28

dzhulgakov approved these changes Apr 1, 2019

View reviewed changes

soumith mentioned this pull request Apr 2, 2019

Clean up detection of library versions #18760

Closed

ezyang changed the title ~~Add torch.backends.detailed_version(), reporting detailed version of all libraries.~~ Add torch.__config__.show(), reporting detailed version of all libraries. Apr 4, 2019

ezyang added 4 commits April 4, 2019 12:06

ezyang added 3 commits April 5, 2019 14:53

dzhulgakov reviewed Apr 8, 2019

View reviewed changes

dzhulgakov approved these changes Apr 8, 2019

View reviewed changes

facebook-github-bot closed this in 29ea086 Apr 9, 2019

facebook-github-bot added the merged label Apr 9, 2019

dzhulgakov mentioned this pull request Apr 26, 2019

Allow a non-OpenMP based build #19749

Closed

ezyang deleted the gh/ezyang/48/head branch May 30, 2019 15:11

great-way mentioned this pull request Mar 24, 2020

how can i make sure that when my pytorch code run in cpu can use avx2 #34574

Closed

Add torch.__config__.show(), reporting detailed version of all libraries. #18579

Add torch.__config__.show(), reporting detailed version of all libraries. #18579

Uh oh!

Conversation

ezyang commented Mar 28, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

soumith commented Mar 28, 2019

Uh oh!

ezyang commented Mar 28, 2019

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

dzhulgakov Apr 1, 2019

Choose a reason for hiding this comment

Uh oh!

apaszke Apr 4, 2019

Choose a reason for hiding this comment

Uh oh!

ezyang commented Apr 4, 2019

Uh oh!

ezyang commented Apr 4, 2019

Uh oh!

ppwwyyxx commented Apr 5, 2019

Uh oh!

ezyang commented Apr 5, 2019

Uh oh!

ezyang commented Apr 5, 2019

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

ezyang commented Apr 8, 2019

Uh oh!

dzhulgakov left a comment

Choose a reason for hiding this comment

Uh oh!

vadimkantorov commented Apr 9, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Apr 9, 2019

Uh oh!

ppwwyyxx commented Apr 10, 2019

Uh oh!

ezyang commented Apr 10, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

Add torch.config.show(), reporting detailed version of all libraries. #18579

Add torch.config.show(), reporting detailed version of all libraries. #18579

ezyang commented Mar 28, 2019 •

edited

Loading

vadimkantorov commented Apr 9, 2019 •

edited

Loading