Revival #12

h-vetinari · 2021-01-27T21:32:17Z

Saw this feedstock didn't have an update in a while - it's going to become a dependency for allennlp 2.0.0, and I'm co-maintaining that feedstock

Closes #9
Closes #11

Possible follow-ups:

build with cuda
build with ffmpeg support
turn into multi-output recipe (e.g. with/without ffmpeg/image support)?
package c++ lib separately?
build for win/osx

…da-forge-pinning 2021.01.15.19.38.06

conda-forge-linter · 2021-01-27T21:32:24Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe:

Whenever possible python packages should use pip. See https://conda-forge.org/docs/maintainer/adding_pkgs.html#use-pip
License is not an SPDX identifier (or a custom LicenseRef) nor an SPDX license expression.

Documentation on acceptable licenses can be found here.

Also enable png/jpeg support, see https://github.com/pytorch/vision/blob/v0.8.2/setup.py

conda-forge-linter · 2021-01-27T21:35:10Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe:

Whenever possible python packages should use pip. See https://conda-forge.org/docs/maintainer/adding_pkgs.html#use-pip

conda-forge-linter · 2021-01-27T21:36:18Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

This reverts commit bb8ddb3.

…da-forge-pinning 2021.01.27.21.31.49

h-vetinari · 2021-01-28T08:14:29Z

@isuruf @hmaarrfk @nehaljwani
This builds, runs the test-suite, and passes except for one py37-only failure (that I don't understand). Not sure if we want to have ffmpeg & cuda support already in this PR or as follow-ups, but I think this is pretty close.

This reverts commit fdd7140.

h-vetinari · 2021-01-31T06:42:23Z

recipe/meta.yaml

+        # TODO: remove this once it is properly represented in pytorch build metadata
+        - typing_extensions  # [py<38]


Has been swiftly fixed by conda-forge/conda-forge-repodata-patches-feedstock#118

h-vetinari · 2021-01-31T19:25:21Z

So my idea of patching this is not possible in this feedstock, since failing check is in CUDAExtension, which is not part of torchvision but comes from pytorch directly:

from torch.utils.cpp_extension import BuildExtension, CppExtension, CUDAExtension, CUDA_HOME

hmaarrfk · 2021-01-31T21:55:05Z

do you want to monkey patch?

h-vetinari · 2021-01-31T22:02:35Z

do you want to monkey patch?

Not sure what other options we'd have. Maybe @isuruf has some ideas?

hmaarrfk · 2021-02-06T04:22:06Z

@jakirkham do you have any ideas on how to get around this? torch tries to determine gpu capabilities at build time.

dirkgr · 2021-02-08T21:40:58Z

You probably already know this, but torch itself recommends installing torchvision with conda install pytorch torchvision -c pytorch or conda install pytorch torchvision cudatoolkit=11.0 -c pytorch. Their version works with and without GPUs, so there must be a way to compile it one way that works for both.

h-vetinari · 2021-02-09T09:46:15Z

You probably already know this, but torch itself recommends installing torchvision with conda install pytorch torchvision -c pytorch or conda install pytorch torchvision cudatoolkit=11.0 -c pytorch. Their version works with and without GPUs, so there must be a way to compile it one way that works for both.

It's certainly possible to compile both in principle. What's not possible is using packages from outside conda-forge to build packages for conda-forge (with the sole exception of the anaconda default channels). Maybe we can get some inspiration from the pytorch recipes, provided they're licensed appropriately.

jakirkham · 2021-02-12T19:26:25Z

do you have any ideas on how to get around this? torch tries to determine gpu capabilities at build time.

No idea. Would ask the PyTorch developers for guidance

isuruf · 2021-03-09T20:31:12Z

@conda-forge-admin, rerender

…da-forge-pinning 2021.03.09.17.59.11

github-actions · 2021-03-09T21:10:55Z

Hi! This is the friendly automated conda-forge-webservice.
I tried to rerender for you, but it looks like there was nothing to do.

…da-forge-pinning 2021.03.09.17.59.11

isuruf · 2021-03-09T22:51:55Z

We can restart cuda 11 builds when pytorch cuda 11 builds are uploaded.

h-vetinari · 2021-03-09T23:24:04Z

Thanks for finishing this off, @isuruf!

hmaarrfk · 2021-03-09T23:40:50Z

awesome!

h-vetinari · 2021-03-10T08:14:21Z

recipe/build-torch.sh

+  export TORCH_CUDA_ARCH_LIST="3.5;5.0+PTX"
+  if [[ ${cuda_compiler_version} == 9.0* ]]; then
+      export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;7.0"
+  elif [[ ${cuda_compiler_version} == 9.2* ]]; then
+      export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0"
+  elif [[ ${cuda_compiler_version} == 10.* ]]; then
+      export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0;7.5"
+  elif [[ ${cuda_compiler_version} == 11.0* ]]; then
+      export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0;7.5;8.0"
+  elif [[ ${cuda_compiler_version} == 11.1* ]]; then
+      export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0;7.5;8.0;8.6"
+  else
+      echo "unsupported cuda version. edit build.sh"
+      exit 1
+  fi


@isuruf
Shouldn't that +PTX be on the last available architecture?

h-vetinari · 2021-03-16T22:04:24Z

recipe/meta.yaml

-  skip: True  # [not linux64]
-  script: python setup.py install --single-version-externally-managed --record=record.txt
+  number: {{ number }}
+  skip: true  # [not linux]


Setting an anchor for easy referencing here...

regro-cf-autotick-bot added 2 commits January 17, 2021 23:32

updated v0.8.2

eaa25e9

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.8.6, and con…

b40926b

…da-forge-pinning 2021.01.15.19.38.06

h-vetinari added 3 commits January 27, 2021 22:34

update requirements for 8.2

534aedf

Also enable png/jpeg support, see https://github.com/pytorch/vision/blob/v0.8.2/setup.py

run upstream test suite

afddfc0

fix SPDX

7e45c0b

h-vetinari force-pushed the revival branch from 82428d5 to 7cb0d48 Compare January 27, 2021 21:35

use pip

4faf85d

h-vetinari force-pushed the revival branch from 7cb0d48 to 4faf85d Compare January 27, 2021 21:36

h-vetinari added 6 commits January 27, 2021 22:44

move most dependencies to host

58c17fd

add c++ compiler

64c6fa0

re-add pillow as a runtime dep

5692d89

add scipy as test dependency

4cc12bc

disable image support for now

bb8ddb3

Revert "disable image support for now"

502df3f

This reverts commit bb8ddb3.

h-vetinari force-pushed the revival branch from 3e72880 to 3f07804 Compare January 28, 2021 00:08

set TORCHVISION_INCLUDE to find correct headers

2083f36

h-vetinari force-pushed the revival branch from fd59e07 to 2083f36 Compare January 28, 2021 00:29

h-vetinari mentioned this pull request Jan 28, 2021

Add hmaarrfk as maintainer #9

Closed

5 tasks

h-vetinari added 3 commits January 28, 2021 01:49

add hmaarrfk & h-vetinari to maintainers

ec51400

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.8.6, and con…

b8247e8

…da-forge-pinning 2021.01.27.21.31.49

use jpeg instead of libjpeg-turbo

a549abc

h-vetinari closed this Jan 28, 2021

h-vetinari reopened this Jan 28, 2021

h-vetinari closed this Jan 29, 2021

h-vetinari reopened this Jan 29, 2021

h-vetinari mentioned this pull request Jan 29, 2021

Cannot install torchvision 0.8.2 with conda-forge without defaults channel pytorch/vision#3264

Closed

Revert "add typing_extensions that aren't being pulled in for pytorch"

fd9a471

This reverts commit fdd7140.

h-vetinari commented Jan 31, 2021

View reviewed changes

dpryan79 mentioned this pull request Feb 4, 2021

add new recipe simba_pbg bioconda/bioconda-recipes#26392

Merged

conda-forge-linter and others added 3 commits March 9, 2021 20:32

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.9.0, and con…

2b5aa6a

…da-forge-pinning 2021.03.09.17.59.11

set TORCH_CUDA_ARCH_LIST

da965c7

Use cuda110.yaml for compiler version

6943ed6

isuruf and others added 6 commits March 9, 2021 15:11

Update cuda110.yaml

49c4b01

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.9.0, and con…

1bd1795

…da-forge-pinning 2021.03.09.17.59.11

Update 0001-skip-flaky-test.patch

d84400b

python 3.9

edaf7f9

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.9.0, and con…

5b162f3

…da-forge-pinning 2021.03.09.17.59.11

MNT: Re-rendered with conda-build 3.21.4, conda-smithy 3.9.0, and con…

17cd898

…da-forge-pinning 2021.03.09.17.59.11

isuruf merged commit b8ce1ae into conda-forge:master Mar 9, 2021

h-vetinari deleted the revival branch March 9, 2021 23:23

h-vetinari mentioned this pull request Mar 9, 2021

MNT: rerender #14

Merged

1 task

h-vetinari commented Mar 10, 2021

View reviewed changes

h-vetinari mentioned this pull request Mar 16, 2021

Recipe improvements #22

Open

10 tasks

h-vetinari commented Mar 16, 2021

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Revival #12

Revival #12

h-vetinari commented Jan 27, 2021 •

edited

Loading

conda-forge-linter commented Jan 27, 2021

conda-forge-linter commented Jan 27, 2021

conda-forge-linter commented Jan 27, 2021

h-vetinari commented Jan 28, 2021

h-vetinari Jan 31, 2021

h-vetinari commented Jan 31, 2021 •

edited

Loading

hmaarrfk commented Jan 31, 2021

h-vetinari commented Jan 31, 2021

hmaarrfk commented Feb 6, 2021

dirkgr commented Feb 8, 2021

h-vetinari commented Feb 9, 2021

jakirkham commented Feb 12, 2021

isuruf commented Mar 9, 2021 •

edited

Loading

github-actions bot commented Mar 9, 2021

isuruf commented Mar 9, 2021

h-vetinari commented Mar 9, 2021

hmaarrfk commented Mar 9, 2021

h-vetinari Mar 10, 2021

h-vetinari Mar 16, 2021

		# TODO: remove this once it is properly represented in pytorch build metadata
		- typing_extensions # [py<38]

Revival #12

Revival #12

Conversation

h-vetinari commented Jan 27, 2021 • edited Loading

conda-forge-linter commented Jan 27, 2021

conda-forge-linter commented Jan 27, 2021

conda-forge-linter commented Jan 27, 2021

h-vetinari commented Jan 28, 2021

h-vetinari Jan 31, 2021

Choose a reason for hiding this comment

h-vetinari commented Jan 31, 2021 • edited Loading

hmaarrfk commented Jan 31, 2021

h-vetinari commented Jan 31, 2021

hmaarrfk commented Feb 6, 2021

dirkgr commented Feb 8, 2021

h-vetinari commented Feb 9, 2021

jakirkham commented Feb 12, 2021

isuruf commented Mar 9, 2021 • edited Loading

github-actions bot commented Mar 9, 2021

isuruf commented Mar 9, 2021

h-vetinari commented Mar 9, 2021

hmaarrfk commented Mar 9, 2021

h-vetinari Mar 10, 2021

Choose a reason for hiding this comment

h-vetinari Mar 16, 2021

Choose a reason for hiding this comment

h-vetinari commented Jan 27, 2021 •

edited

Loading

h-vetinari commented Jan 31, 2021 •

edited

Loading

isuruf commented Mar 9, 2021 •

edited

Loading