profile
viewpoint
Soumith Chintala soumith Facebook AI Research New York, USA http://soumith.ch /\︿╱\ ___________________________ \0_ 0 /╱\╱ _____________________ \▁︹_/

pyro-ppl/pyro 6013

Deep universal probabilistic programming with Python and PyTorch

OpenNMT/OpenNMT 2250

Open Source Neural Machine Translation in Torch (deprecated)

dmlc/dlpack 285

RFC for common in-memory tensor structure and operator interface for deep learning system

clementfarabet/torch-ios 201

Torch7 for iOS.

prigoyal/pytorch_memonger 146

Experimental ground for optimizing memory of pytorch models

clementfarabet/graphicsmagick 127

A simple Lua wrapper to graphicsmagick.

clementfarabet/lua---nnx 96

An extension to Torch7's nn package.

clementfarabet/lua---parallel 93

A (simple) parallel computing framework for Lua

koraykv/unsup 86

Some unsupervised learning modules using Torch

issue closedpytorch/vision

Downloading pretrained models with torchvision gives HTTP Error 403

I'm getting a 403 error when I try to download pre-trained weights with torchvision 0.5.

torchvision.models.mobilenet_v2(pretrained=True) results in HTTPError: HTTP Error 403: Forbidden. I've tried multiple models on multiple machines on multiple networks, always get a 403 error.

I can download the model directly from the URL with wget. But for some reason downloading it through torchvision doesn't work. I can progress since I can download the weights, but it's unfortunate I have to load them in manually.

closed time in 4 days

mcleonard

issue commentpytorch/vision

Downloading pretrained models with torchvision gives HTTP Error 403

The SSL Certificate issue is fixed now. The Cert was expired for ~4 hours, and we root-caused the auto-renewal problem, it wont happen again.

mcleonard

comment created time in 4 days

issue closedpytorch/pytorch

SSL: CERTIFICATE_VERIFY_FAILED while trying to download pretrained model

🐛 Bug

Trying to use pretrained model from torchvision and receiving urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed (_ssl.c:852)> when downloading. In similar issues #33233 and #33234 it was said that problem is solved but I'm facing the issue right now.

To Reproduce

vgg16 = torchvision.models.vgg16(pretrained=True)

Expected behavior

Pre-trained models should be downloaded without certificate failures

Environment

PyTorch version: 1.4.0 Is debug build: No CUDA used to build PyTorch: 10.1

OS: Ubuntu 18.04.3 LTS GCC version: (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0 CMake version: version 3.12.0

Python version: 3.6 Is CUDA available: Yes CUDA runtime version: 10.0.130 GPU models and configuration: GPU 0: Tesla P100-PCIE-16GB Nvidia driver version: 418.67 cuDNN version: /usr/lib/x86_64-linux-gnu/libcudnn.so.7.6.5

Versions of relevant libraries: [pip3] numpy==1.17.5 [pip3] torch==1.4.0 [pip3] torchfile==0.1.0 [pip3] torchsummary==1.5.1 [pip3] torchtext==0.3.1 [pip3] torchvision==0.5.0 [conda] Could not collect



cc @ezyang @gchanan @zou3519

closed time in 4 days

muxgt

issue commentpytorch/pytorch

SSL: CERTIFICATE_VERIFY_FAILED while trying to download pretrained model

The SSL Certificate issue is fixed now. The Cert was expired for ~4 hours, and we root-caused the auto-renewal problem, it wont happen again.

muxgt

comment created time in 4 days

issue closedpytorch/pytorch

Installation on Windows fails due to expired certificate of download.pytorch.org

🐛 Bug

Installation of torch and torchvision fails on windows.

To Reproduce

Steps to reproduce the behavior:

  1. Install on windows
  2. RUN pip install torch===1.4.0 torchvision===0.5.0 -f https://download.pytorch.org/whl/torch_stable.html

Error log:

WARNING: Retrying (Retry(total=0, connect=None, read=None, redirect=None, status=None)) after connection broken by 'SSLError(SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: certificate has expired (_ssl.c:1076)'))': /whl/torch_stable.html
Could not fetch URL https://download.pytorch.org/whl/torch_stable.html: There was a problem confirming the ssl certificate: HTTPSConnectionPool(host='download.pytorch.org', port=443): Max retries exceeded with url: /whl/torch_stable.html (Caused by SSLError(SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: certificate has expired (_ssl.c:1076)'))) - skipping
ERROR: Could not find a version that satisfies the requirement torch===1.4.0 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)
ERROR: No matching distribution found for torch===1.4.0

cc @ezyang @gchanan @zou3519

closed time in 4 days

ZoranPandovski

issue commentpytorch/pytorch

Installation on Windows fails due to expired certificate of download.pytorch.org

The SSL Certificate issue is fixed now. The Cert was expired for ~4 hours, and we root-caused the auto-renewal problem, it wont happen again.

ZoranPandovski

comment created time in 4 days

issue closedpytorch/examples

cannot open ImageNet dataset with .sh file

Hello i can't open dataset in this code I've open it in GitBash (my os is windows10) Script is launching, and folders creating but later mv ILSVRC2012_val_00000001.JPEG n01751748 when commands like this launch - my console give me: mv: cannot stat no such file or directory

closed time in 9 days

germanjke

issue commentpytorch/examples

cannot open ImageNet dataset with .sh file

clarified via the commit https://github.com/pytorch/examples/commit/e9e76722dad4f4569651a8d67ca1d10607db58f9

germanjke

comment created time in 9 days

issue commentpytorch/examples

cannot open ImageNet dataset with .sh file

I've clarified language in README that the imagenet dataset has to be downloaded, and then the valprep.sh only reorganizes the validation set into labeled folders.

germanjke

comment created time in 9 days

push eventpytorch/examples

Soumith Chintala

commit sha e9e76722dad4f4569651a8d67ca1d10607db58f9

clarify imagenet download

view details

push time in 9 days

pull request commentpytorch/pytorch.github.io

fixed typo in quick start module

thanks @shahrouzz !

shahrouzz

comment created time in 10 days

push eventpytorch/pytorch.github.io

shahrouzz

commit sha 0dd0ed64968d2dfb9356006dbba66bda99536712

fixed typo in quick start module (#344) Signed-off-by: Shahrouz Zolfaghari <shahrouz@senri.se>

view details

push time in 10 days

PR merged pytorch/pytorch.github.io

Reviewers
fixed typo in quick start module

Fixed typo that affected one of the windows download links

+1 -1

0 comment

1 changed file

shahrouzz

pr closed time in 10 days

startedfacebookresearch/pytorch3d

started time in 11 days

issue commentpytorch/vision

Pre-built wheels unexpectedly changed their hashes in the last days

to be clear, the hashes only changed for the wheels in https://download.pytorch.org/whl/ and not for the wheels on https://pypi.org/project/torch/

daniel-j-h

comment created time in 14 days

issue closedpytorch/vision

Pre-built wheels unexpectedly changed their hashes in the last days

When using pip to install the pre-build torchvision wheels you can specify a --hash option to let pip verify the wheel's integrity. The hash can be computed e.g. via pip hash *.whl.

I noticed the cpu and gpu wheel hashes for torchvision 0.5.0 on Python 3.6 changing in the last days.

https://download.pytorch.org/whl/cpu/torchvision-0.5.0%2Bcpu-cp36-cp36m-linux_x86_64.whl

-          --hash=sha256:12a5d92dca112cfbfb6c2a4225fa2c281f77b0ae22e7b2c6cb80d7b2daa04b1e           \
+          --hash=sha256:ae3686f57a56c71ee7cc001bb067b3ebb246f1d6abc8b9023f7d7d043a4f8b83           \

https://download.pytorch.org/whl/cu101/torchvision-0.5.0-cp36-cp36m-linux_x86_64.whl

-          --hash=sha256:fea3d431bf639c0719afff5972eb568ebe143eba447c1c8bb491c7dfb0025ed6       \
+          --hash=sha256:df994daf1b04a183779538022fb4345b66364fccf66b33e1095ae8d4955a5406       \

This is unexpected and I would like to understand what the cause of the hash change is. Did you folks re-publish with slightly different wheels? Or what happened here?

closed time in 14 days

daniel-j-h

issue commentpytorch/vision

Pre-built wheels unexpectedly changed their hashes in the last days

@daniel-j-h this is expected but accidental. We were building Python 3.8 binaries a few days after 1.4.0 release, and we inadvertently triggered all builds. Our build system is non-deterministic, and hence the hashes changed. The functionality in the wheel didn't actually change.

@seemethere and I will make sure this doesn't happen again, by making sure if we have any post-release builds, they happen with a bumped version.

daniel-j-h

comment created time in 14 days

pull request commentpytorch/pytorch

Test type hints using tutorials

as discussed offline with Edward, I think submoduling tutorials is too heavy-weight, and we need to move to something like hash-recording (a text file in .circleci folder that has hashes, and the hashes get updated by a bot that checks for stable).

csvoss

comment created time in 14 days

push eventpytorch/contrib

Katherine Crowson

commit sha 24fa0f14d0c9892c9c0b97eada6b96affe280a02

Don't try to use CuDNN without confirming CUDA is available

view details

Soumith Chintala

commit sha c545fedf4f73c8e95f91fd81f2d5bf7fa9c62a61

Merge pull request #34 from crowsonkb/cudnn_used Don't try to use CuDNN attributes without confirming CUDA is available

view details

push time in 14 days

PR merged pytorch/contrib

Don't try to use CuDNN attributes without confirming CUDA is available

torch.backends.cudnn.disable_global_flags() (this line of code here) gives an error when it is run on a machine without CUDA:

AttributeError: module 'torch.backends.cudnn' has no attribute 'disable_global_flags'

and all test cases fail. I put it behind an if torch.cuda.is_available() test. I am now able to run PyTorch Contrib's test cases on my GPU-less laptop. I think your Travis CI checks have also been failing due to this error, as well.

+2 -1

0 comment

1 changed file

crowsonkb

pr closed time in 14 days

pull request commentpytorch/pytorch

Add nn.ResidualBlock and nn.ResidualBlockWithShortcut

Seems like a good idea, @fmassa @colesbury is the API sufficiently general to cover all popular resnet stuff?

crowsonkb

comment created time in 14 days

pull request commentpytorch/pytorch

Enable MKL on MacOS if installed

the OSX binaries already link with MKL, so hopefully it should be okay. Our smoke tests should tell.

CaoZhongZ

comment created time in 14 days

issue closedpytorch/pytorch

No PyPI packages for Python 3.8 which will be released soon (expected October)

~ ❯❯❯ LD_LIBRARY_PATH=$HOME/pysandbox/lib ~/pysandbox/bin/pip3 install torch
/Users/btv/pysandbox/lib/python3.8/site-packages/pip/_vendor/html5lib/_trie/_base.py:3: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  from collections import Mapping
Collecting torch
/Users/btv/pysandbox/lib/python3.8/site-packages/pip/_vendor/msgpack/fallback.py:220: PendingDeprecationWarning: encoding is deprecated, Use raw=False instead.
  warnings.warn(
  Using cached https://files.pythonhosted.org/packages/5f/e9/bac4204fe9cb1a002ec6140b47f51affda1655379fe302a1caef421f9846/torch-0.1.2.post1.tar.gz
    Complete output from command python setup.py egg_info:
    /Users/btv/pysandbox/lib/python3.8/site-packages/setuptools/depends.py:2: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
      import imp
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/private/var/folders/xm/k62yj65n0qd9n40myt10y8sjpk8rbt/T/pip-install-6cdjtra9/torch/setup.py", line 11, in <module>
        raise RuntimeError(README)
    RuntimeError: PyTorch does not currently provide packages for PyPI (see status at https://github.com/pytorch/pytorch/issues/566).
    
    Please follow the instructions at http://pytorch.org/ to install with miniconda instead.
    
    
    ----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /private/var/folders/xm/k62yj65n0qd9n40myt10y8sjpk8rbt/T/pip-install-6cdjtra9/torch/
You are using pip version 19.0.3, however version 19.1.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.

closed time in 14 days

umanwizard

issue commentpytorch/pytorch

No PyPI packages for Python 3.8 which will be released soon (expected October)

python 3.8 packages on PyPI are live for both OSX and Linux.

https://pypi.org/project/torch/#files

umanwizard

comment created time in 14 days

pull request commentpytorch/builder

Add CUDA 10.2 support

or else, the binary size bug affects us

seemethere

comment created time in 17 days

pull request commentpytorch/builder

Add CUDA 10.2 support

all looks reasonable, but one thing we forgot to add back was that magma binaries have to be built with devtoolset3 in https://github.com/pytorch/builder/tree/master/conda#building-magma-cuda91

Basically, revert the lines 44 to 48 of README.md in https://github.com/pytorch/builder/commit/dd7edfdf40e7eb75f0d561512e31b31c8319d9c9

seemethere

comment created time in 17 days

issue openedgooglecolab/colabtools

Upgrading PyTorch to v1.4.0 and torchvision to v0.5.0

Hi folks,

We released a new PyTorch version v1.4.0 and torchvision v0.5.0.

The wheel links are:

PyTorch

CUDA 10.0

https://download.pytorch.org/whl/cu100/torch-1.4.0%2Bcu100-cp27-cp27m-linux_x86_64.whl https://download.pytorch.org/whl/cu100/torch-1.4.0%2Bcu100-cp27-cp27mu-linux_x86_64.whl https://download.pytorch.org/whl/cu100/torch-1.4.0%2Bcu100-cp36-cp36m-linux_x86_64.whl

CUDA 10.1

https://download.pytorch.org/whl/cu101/torch-1.4.0-cp27-cp27m-linux_x86_64.whl https://download.pytorch.org/whl/cu101/torch-1.4.0-cp27-cp27mu-linux_x86_64.whl https://download.pytorch.org/whl/cu101/torch-1.4.0-cp36-cp36m-linux_x86_64.whl

TorchVision

CUDA 10.0

https://download.pytorch.org/whl/cu100/torchvision-0.5.0%2Bcu100-cp27-cp27m-linux_x86_64.whl https://download.pytorch.org/whl/cu100/torchvision-0.5.0%2Bcu100-cp27-cp27mu-linux_x86_64.whl https://download.pytorch.org/whl/cu100/torchvision-0.5.0%2Bcu100-cp36-cp36m-linux_x86_64.whl

CUDA 10.1

https://download.pytorch.org/whl/cu101/torchvision-0.5.0-cp27-cp27m-linux_x86_64.whl https://download.pytorch.org/whl/cu101/torchvision-0.5.0-cp27-cp27mu-linux_x86_64.whl https://download.pytorch.org/whl/cu101/torchvision-0.5.0-cp36-cp36m-linux_x86_64.whl

If you require wheels for Python 3.7 / 3.8, you can find the links here: https://download.pytorch.org/whl/torch_stable.html

We're looking to having it updated in Colab.

Thanks very much.

cc'ing @colaboratory-team

created time in 20 days

issue commentpytorch/pytorch

[v1.4.0] Release Tracker

release is closed

mingbowan

comment created time in 20 days

issue closedpytorch/pytorch

[v1.4.0] Release Tracker

The branch v1.4.0 has been cut.

If you need any particular patches onto this branch, please comment below and send a PR to v1.4.0 branch (instead of master).


Current PRs open against the v1.4.0 branch

closed time in 20 days

mingbowan

issue closedpytorch/pytorch

[Feature request] Provide binaries for Python 3.8

🐛 Bug

Trying to install torch results in a python error.

To Reproduce

Steps to reproduce the behavior:

  1. Install Python (python-3.8.0-amd64.exe)
  2. run pip install numpy
  3. run pip install torch torchvision -f https://download.pytorch.org/whl/torch_stable.html

Expected behavior

Pytorch gets installed

Environment

C:\Windows\system32>python C:\temp\collect_env.py
Collecting environment information...
PyTorch version: N/A
Is debug build: N/A
CUDA used to build PyTorch: N/A
OS: Microsoft Windows 10 Home
GCC version: (GCC) 4.6.1
CMake version: Could not collect
Python version: 3.8
Is CUDA available: N/A
CUDA runtime version: Could not collect
GPU models and configuration: GPU 0: GeForce GTX 980 Ti
Nvidia driver version: 430.86
cuDNN version: Could not collect
Versions of relevant libraries:
[pip3] numpy==1.17.3
[conda] Could not collect

Additional context

Full console output:

C:\Windows\system32>pip install torch torchvision -f https://download.pytorch.org/whl/torch_stable.html
Looking in links: https://download.pytorch.org/whl/torch_stable.html
Collecting torch
  Using cached https://files.pythonhosted.org/packages/f8/02/880b468bd382dc79896eaecbeb8ce95e9c4b99a24902874a2cef0b562cea/torch-0.1.2.post2.tar.gz
Collecting torchvision
  Using cached https://files.pythonhosted.org/packages/fb/01/03fd7e503c16b3dc262483e5555ad40974ab5da8b9879e164b56c1f4ef6f/torchvision-0.2.2.post3-py2.py3-none-any.whl
Collecting pyyaml
  Downloading https://files.pythonhosted.org/packages/e3/e8/b3212641ee2718d556df0f23f78de8303f068fe29cdaa7a91018849582fe/PyYAML-5.1.2.tar.gz (265kB)
     |████████████████████████████████| 266kB 1.3MB/s
Collecting six
  Using cached https://files.pythonhosted.org/packages/73/fb/00a976f728d0d1fecfe898238ce23f502a721c0ac0ecfedb80e0d88c64e9/six-1.12.0-py2.py3-none-any.whl
Requirement already satisfied: numpy in c:\program files\python38\lib\site-packages (from torchvision) (1.17.3)
Collecting pillow>=4.1.1
  Downloading https://files.pythonhosted.org/packages/b6/5a/ee223b0503ebba3d6adcf24516cbae85d581058f2bac635fc82e4bd6c2bf/Pillow-6.2.1-cp38-cp38-win_amd64.whl (2.0MB)
     |████████████████████████████████| 2.0MB 6.4MB/s
Installing collected packages: pyyaml, torch, six, pillow, torchvision
    Running setup.py install for pyyaml ... done
    Running setup.py install for torch ... error
    ERROR: Command errored out with exit status 1:
     command: 'c:\program files\python38\python.exe' -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\\Users\\Tyler\\AppData\\Local\\Temp\\pip-install-qo2q0sjc\\torch\\setup.py'"'"'; __file__='"'"'C:\\Users\\Tyler\\AppData\\Local\\Temp\\pip-install-qo2q0sjc\\torch\\setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record 'C:\Users\Tyler\AppData\Local\Temp\pip-record-72uhhcea\install-record.txt' --single-version-externally-managed --compile
         cwd: C:\Users\Tyler\AppData\Local\Temp\pip-install-qo2q0sjc\torch\
    Complete output (23 lines):
    running install
    running build_deps
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "C:\Users\Tyler\AppData\Local\Temp\pip-install-qo2q0sjc\torch\setup.py", line 225, in <module>
        setup(name="torch", version="0.1.2.post2",
      File "c:\program files\python38\lib\site-packages\setuptools\__init__.py", line 145, in setup
        return distutils.core.setup(**attrs)
      File "c:\program files\python38\lib\distutils\core.py", line 148, in setup
        dist.run_commands()
      File "c:\program files\python38\lib\distutils\dist.py", line 966, in run_commands
        self.run_command(cmd)
      File "c:\program files\python38\lib\distutils\dist.py", line 985, in run_command
        cmd_obj.run()
      File "C:\Users\Tyler\AppData\Local\Temp\pip-install-qo2q0sjc\torch\setup.py", line 99, in run
        self.run_command('build_deps')
      File "c:\program files\python38\lib\distutils\cmd.py", line 313, in run_command
        self.distribution.run_command(command)
      File "c:\program files\python38\lib\distutils\dist.py", line 985, in run_command
        cmd_obj.run()
      File "C:\Users\Tyler\AppData\Local\Temp\pip-install-qo2q0sjc\torch\setup.py", line 51, in run
        from tools.nnwrap import generate_wrappers as generate_nn_wrappers
    ModuleNotFoundError: No module named 'tools.nnwrap'
    ----------------------------------------
ERROR: Command errored out with exit status 1: 'c:\program files\python38\python.exe' -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\\Users\\Tyler\\AppData\\Local\\Temp\\pip-install-qo2q0sjc\\torch\\setup.py'"'"'; __file__='"'"'C:\\Users\\Tyler\\AppData\\Local\\Temp\\pip-install-qo2q0sjc\\torch\\setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record 'C:\Users\Tyler\AppData\Local\Temp\pip-record-72uhhcea\install-record.txt' --single-version-externally-managed --compile Check the logs for full command output.

cc @ezyang @gchanan @zou3519

closed time in 20 days

tjf801

issue commentpytorch/pytorch

[Feature request] Provide binaries for Python 3.8

  • website wizard updated.
  • torch, torchvision, torchaudio stable binaries for python 3.8 are live.
tjf801

comment created time in 20 days

issue commentpytorch/pytorch

Python 3.8 serialization error

not just windows. reproduced on OSX / Linux

peterjc123

comment created time in 20 days

PR closed pytorch/vision

Reviewers
Preparing builds for 0.5.0 release
+138 -806

2 comments

7 changed files

fmassa

pr closed time in 20 days

pull request commentpytorch/vision

Preparing builds for 0.5.0 release

python 3.8 binaries are live as well.

fmassa

comment created time in 20 days

PR closed pytorch/audio

Build/v0.4.0

building py 3.8 binaries

+41 -372

0 comment

4 changed files

soumith

pr closed time in 20 days

push eventpytorch/pytorch.github.io

Soumith Chintala

commit sha a35da59f5017ca52e996467610c1ddd026d91b16

remove other python selectors except 2.7 and 3

view details

Soumith Chintala

commit sha b7d6bfe14abb0b0f7eebf7824bb5db71e00d8563

for quick start commands, remove python versions, just keep Python as a single option

view details

push time in 20 days

push eventpytorch/vision

Soumith Chintala

commit sha 14d65951ce87c8c2eb1547a39ce42fbafc599eab

make typing py2k only

view details

push time in 20 days

push eventpytorch/audio

Vincent QB

commit sha 82e02e853364120ed66c748df74ad1f35909e6ac

Update version number to v0.4.0 (#410) * update version number. * add upload jobs too Co-authored-by: Soumith Chintala <soumith@gmail.com>

view details

Soumith Chintala

commit sha fa137d5f4db50c39f26a70111d1d28b361a615e2

make typing py2 only runtime dependency in conda

view details

Soumith Chintala

commit sha acaa3ef310e1e59b73bda7754f11f1ec3dcc98ca

build python 3.8 binaries

view details

Soumith Chintala

commit sha 5e2f1d21bc9fe92ba1c66ec13db6911bf58ea1ba

bump

view details

Soumith Chintala

commit sha ba6d67ba383b215e8be4672b554142a92d01920f

fix

view details

push time in 20 days

push eventpytorch/audio

Soumith Chintala

commit sha fa137d5f4db50c39f26a70111d1d28b361a615e2

make typing py2 only runtime dependency in conda

view details

push time in 20 days

push eventpytorch/audio

Soumith Chintala

commit sha 18eb10e00a5de6ab08bd3998a7d5b5cdc785c7d6

fix

view details

push time in 20 days

push eventpytorch/vision

Soumith Chintala

commit sha ea8a30e2b5da924a8a60dfc113f08511f6583854

fix

view details

push time in 20 days

push eventpytorch/audio

Soumith Chintala

commit sha ea67fdd207eeff32649292156678e3f1ca69b218

bump

view details

push time in 20 days

push eventpytorch/vision

Soumith Chintala

commit sha 526de3ee2ed8d34ff0ab374a9b67ca04bc025e76

bump

view details

push time in 20 days

push eventpytorch/builder

Soumith Chintala

commit sha 67a4b2ccadda2e662e32b7e3548354875b00c4fc

add python 3.8

view details

Soumith Chintala

commit sha a70b202cc557bee6899956e253c345e7b99adfb8

add 3.8 to windows upload script

view details

push time in 20 days

PR opened pytorch/audio

Build/v0.4.0

building py 3.8 binaries

+54 -296

0 comment

2 changed files

pr created time in 20 days

push eventpytorch/audio

Soumith Chintala

commit sha 39bc4d5f6676271c33a9294b7e3f0c5c1691a554

build python 3.8 binaries

view details

push time in 20 days

PR closed pytorch/pytorch

V1.4.0 RC builds open source
+74 -3507

2 comments

12 changed files

mingbowan

pr closed time in 20 days

pull request commentpytorch/pytorch

V1.4.0 RC builds

all 1.4.0 binaries are built and live, including python 3.8 ones. Closing the PR

mingbowan

comment created time in 20 days

push eventpytorch/vision

Soumith Chintala

commit sha d3a6a59406d4b54834c33e42564d33f0bc28b2e0

rebuild python 3.8 only

view details

push time in 20 days

issue commentpytorch/pytorch

[Feature request] Provide binaries for Python 3.8

after they finish, for Linux, macOS, need to build torchvision binaries

tjf801

comment created time in 21 days

issue commentpytorch/pytorch

[Feature request] Provide binaries for Python 3.8

waiting on macOS v1.4.0 builds to finish in https://github.com/pytorch/pytorch/pull/31542

tjf801

comment created time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha 5a21ea12d715cbc0a3c976007b1958c711d8270f

fix

view details

push time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha 54502e697e915fde69702239c99e7fc5efca3880

fix

view details

push time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha 682fca2116ace0125686f848128e91aa09672c6e

fix

view details

push time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha 8bd03ab1f16946a54a8bc19103a21406a12a3d2e

fix

view details

push time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha d322311b7e30cb6f462bf9e819d9d3ee1fa73f13

fix

view details

push time in 21 days

push eventmingbowan/pytorch

Soumith Chintala

commit sha 81c7435d594c99d6c7681f8017e89116d45efa9c

macos 3.8 builds re-kick

view details

push time in 21 days

issue commentpytorch/pytorch

[Proposal] Data reading framework for PyTorch (Hive, MySQL, S3 etc.)

@gourav-sg pretty much all of the bullet points that you gave sound more like standard high-level client requirements, I think you can make them whether you are picking Pandas or Spark -- i.e. regardless of the nature of the project. And, if you are building custom production pipelines, you likely have to do non-trivial engineering work regardless of the product you pick.

Do you think that Petastorm will be supported by the Pytorch community in general, and later on integrated? The ability to transform data from objects stores, or SQL (JDBC) into tensors will be a mandatory requirement today.

Looks like Petastorm already supports clean PyTorch integration. I don't see what else one has to do to integrate it further -- we wont be swallowing Petastorm into the PyTorch codebase, it doesn't make any sense for modularity. About support itself, unlike Databricks or Cloudera, we don't have a Commercial Support arm, so yes we will endorse good projects in the PyTorch community by carefully vetting them, but we won't have a group of engineers doing client support in exchange for money. Our work on vetting pytorch-based projects carefully for quality is consolidated in this page: https://pytorch.org/ecosystem/

I think the disconnect between your thinking and ours is that, we see PyTorch as a library that integrates with other libraries as equal partners, and people use these freely to build their solutions. Whereas you see PyTorch as a complete AI solution, with various official / well-supported integrations and a good commercial support arm. It is not the latter, unless someone starts a company around commercializing PyTorch. From the limited knowledge I have, TensorFlow isn't that project either, though they put a lot more emphasis on having a TensorFlow API for everything (with less emphasis on reusing other projects' APIs).

So, with that disconnect in place, do choose what suits best for your clients. If they really want a commercial-grade endorsement and guarantees (sort of how one used to say "you never get fired for buying IBM"), I am not sure PyTorch at this stage of it's lifetime is the right product for your clients. If they perceive TF as being that, then that's fine. Maybe PyTorch will get their some day, organically, but it's not in our strategic priorities to exactly be that.

When we say PyTorch is production-ready, it really is -- used by a different set of consumers. It is used by large and small software companies {FB, Uber, Tesla, Matroid, etc.} who build their own end-to-end solutions for serving / inference.

pritamdamania87

comment created time in 21 days

IssuesEvent

push eventpytorch/tutorials

Vipul Vaibhaw

commit sha 967d22b93b17e18f0371b15ffd5850dace3dd7f5

torch.uint8 is now deprecated, moving to torch.bool (#812) Co-authored-by: Soumith Chintala <soumith@gmail.com>

view details

push time in 22 days

PR merged pytorch/tutorials

reinforcement_q_learning: resolving torch.uint8 is now deprecated warning
  • torch.uint8 is now deprecated, moving to torch.bool
+1 -1

1 comment

1 changed file

vaibhawvipul

pr closed time in 22 days

push eventvaibhawvipul/tutorials

prajjwal1

commit sha 60443862d54945db4a23f513bed241cbccdd0e6f

added sample code for fasterrcnn_resnet50_fpn (optional)

view details

prajjwal1

commit sha 676c17d8ef2b21f022dbd0a61054bd05710c6891

added sample code for fasterrcnn_resnet50_fpn (optional)

view details

prajjwal1

commit sha 88b6d4531e6c49cd0bbf26b65662768a3e0f75b8

removed comment

view details

Shen Li

commit sha b4e9f214a89e8cc6aef8e37692f6d4ddfcea33b3

Adding an RPC Tutorial

view details

Shen Li

commit sha 494f02f1c937df1afae029262dac99c2c1b9c913

Address comments

view details

Shen Li

commit sha 12925975281f60dc8a036747a9898f71e9ab89dd

address comments

view details

Shen Li

commit sha 7c0b10ec19726c6e995938b796c080404395078f

address comments

view details

prajjwal1

commit sha d5df20a5bc0806f4c27c254236736ce75e61f9fc

changed heading and removed random init data in favor of sample data from dataset

view details

Edward Z. Yang

commit sha bc0e315f1755358da9a272e74190c46830b723cc

Update sample makefiles to C++14, as that's what PyTorch requires now Signed-off-by: Edward Z. Yang <ezyang@fb.com>

view details

Will Feng

commit sha 0720795fd24d63d16024faede3b32dfbbcacd740

Remove context: org-member from config.yml to fix CI build issues

view details

Shen Li

commit sha 6b1d31049ca7b97ecdefeea7a5ee606bbbb5e3de

address comments

view details

Shen Li

commit sha 2ac179473ca0831a9f49dff38691280e686d6fbd

Add sample training outputs for RL example

view details

Brian Johnson

commit sha 3472ee2de659c4a7c19473cd2e88ad116d94dfcf

Merge pull request #814 from pytorch/circleci_remove_context Remove context: org-member from config.yml to try to fix CI build issues

view details

Will Feng

commit sha eca84ca77caaa15d02e132906a8e9555763e08de

Create separate build jobs for PR CI and master CI

view details

Will Feng

commit sha 16e8be26eae16740a7cd5095300e11220d479d32

Revert "Create separate build jobs for PR CI and master CI" This reverts commit eca84ca77caaa15d02e132906a8e9555763e08de.

view details

Will Feng

commit sha 0434f834b55dc417cc96af3bb427b575718a820d

Create separate build jobs for PR CI and master CI

view details

Prajjwal Bhargava

commit sha 30ca017a54b546c6040be1abf4faafd7f8410813

Merge branch 'master' into master

view details

Will Feng

commit sha 11876d8b39a54d3a4ccb9d02ce1288fd102cbe23

try to fix

view details

Will Feng

commit sha 94753f52faae3b6c784629863a97a86259de7871

try to fix

view details

Will Feng

commit sha 68031d7a2645d5f2a101bfc5b1f96840ae956c62

try to fix

view details

push time in 22 days

issue commentpytorch/pytorch

[Proposal] Data reading framework for PyTorch (Hive, MySQL, S3 etc.)

I took a look at the petastorm API, and it looks pretty reasonable:

from petastorm import make_reader
from petastorm.pytorch import DataLoader

with DataLoader(make_reader('hdfs://myhadoop/some_dataset', num_epochs=10,
                            transform_spec=transform), batch_size=64) as train_loader:
    with data, label in train_loader:
        # training loop

It has the following expected features from a high-performant production-ready data / object store reader:

  • Selective column readout
  • Multiple parallelism strategies: thread, process, single-threaded (for debug)
  • N-grams readout support
  • Row filtering (row predicates)
  • Shuffling
  • Partitioning for multi-GPU training
  • Local caching
pritamdamania87

comment created time in 22 days

issue commentpytorch/pytorch

[Proposal] Data reading framework for PyTorch (Hive, MySQL, S3 etc.)

in some fast-forwarded time, if and when @pritamdamania87 and team decide to open-source their Data pipeline framework (and there are similar efforts, as @jaliyae pointed out such as petastorm), one has to think about whether to yield NumPy ndarrays or PyTorch Tensors, depending on where the open-source user-base would find this generally useful.

pritamdamania87

comment created time in 22 days

issue commentpytorch/pytorch

[Proposal] Data reading framework for PyTorch (Hive, MySQL, S3 etc.)

does not make any sense at all. We should not be creating massive technical overheads just for reading data from object stores (S3, HDFS, etc) and SQL databases. For example, if I am running Kafka SQL to get me the latest images recorded off an IoT device, and train and respond in real time, I want this functionality to be there in Pytorch API's directly.

Yet you can see that Tensorflow does have those libraries for S3: https://apimirror.com/tensorflow~guide/deploy/s3 which is another object store.

@gourav-sg the whole notion that PyTorch is some special snowflake in the Python ecosystem, and it requires new APIs or tooling to read a pickle (of Tensors or ndarrays) from S3 or Parquet seems pretty wrong, and questionably misguided.

There are mature Python APIs around reading from HDFS or Parquet or S3. Ultimately, one wants to write a generator that yields a Tensor.

Yes, TensorFlow made a whole new API from scratch for it, and with TFRecords and stuff, but it is literally to read a TensorFlow format into a TensorFlow Tensor.

But if there is mature Python API that reads data from Datastores and Cloudstores and returns it as a NumPy ndarray or a Pandas dataframe, we absolutely be leveraging that, not really sit and redesign something from scratch, because it's somehow not "PyTorch API". Converting a NumPy array to a PyTorch Tensor is zero mem-copy and zero overhead.

pritamdamania87

comment created time in 22 days

issue closedpytorch/pytorch

Feature Request: Add API to anneal LR and WD of optimizer

In PyTorch ImageNet example, the approach to anneal LR is to change internal parameters, which is defined in here. An API to handle hyper parameter state might be better.

Possible solutions might be

  1. A member function to directly change internal members
  2. A annealing scheme function (epoch) -> {lr, wd, ...} as an argument when initializing the opimizers

cc @vincentqb

closed time in 22 days

IanChen83

issue commentpytorch/pytorch

Feature Request: Add API to anneal LR and WD of optimizer

LRSchedulers reasonably handle this aspect now, and they are in the PyTorch API.

IanChen83

comment created time in 22 days

PR closed pytorch/pytorch

Reviewers
Add zoneout awaiting response (this tag is deprecated) open source

This PR adds a new module for regularizing LSTMs and GRUs implementing zoneout, a technique recently pushing state-of-the-art in language modelling. It was requested in #561.

A naive code snippet for using it with RNNCellBase subclasses is provided in the docstring. Ideally, more convenient and faster utilities for using zoneout could be implemented, including a wrapper around RNNCellBase (what they did in seq2seq.pytorch looks like a good design) and argument-level support in nn.LSTM and nn.GRU. This is not straightforward since current RNN API is poorly extensible (as pointed out in #711) and zoneout could only work with autograd RNN backend (as opposed to cudnn). Considering that, I decided integrating zoneout with RNN API deserves its own PR and should be discussed in the meantime. Suggestions are most welcome.

+243 -3

13 comments

6 changed files

tomekkorbak

pr closed time in 22 days

pull request commentpytorch/pytorch

Add zoneout

the method is a bit more niche, and we were waiting for it to pick up in popularity. Considering it didn't, I'm going to close this Feature Request as wontfix, we dont want to increase the maintenance surface of core,

tomekkorbak

comment created time in 22 days

issue closedpytorch/pytorch

Feature request: Zoneout LSTM

Zoneout is one method to perform recurrent dropout on the hidden state of an RNN and has been shown to work quite well on a variety of cases.

Implementing a zoneout LSTM, such as in Chainer's implementation, requires reasonably trivial modifications to implement and would be a good early project for someone new to contributing to PyTorch. If no-one has started by the point I have free time I'll look at doing it myself :)

Note: other forms of recurrent dropout are also worth looking at but I feel zoneout is likely the most broadly applicable version.

closed time in 22 days

Smerity

issue commentpytorch/pytorch

Feature request: Zoneout LSTM

the method is a bit more niche, and we were waiting for it to pick up in popularity. Considering it didn't, I'm going to close this Feature Request as wontfix, we dont want to increase the maintenance surface of core,

Smerity

comment created time in 22 days

push eventpytorch/builder

peterjc123

commit sha dc7d9e53bf5b951fff6a172396e72d995cf0c713

Fix Windows LibTorch build (#396)

view details

push time in 22 days

PR merged pytorch/builder

Fix Windows LibTorch build

cc @soumith

+1 -0

0 comment

1 changed file

peterjc123

pr closed time in 22 days

startedprabhuomkar/pytorch-cpp

started time in 23 days

push eventpytorch/builder

peterjc123

commit sha cbbe5edee6f0e1ef80b8237011b2a609045f44f0

Add Python 3.8 binaries for nightlies (#395)

view details

push time in 23 days

PR merged pytorch/builder

Add Python 3.8 binaries for nightlies

@soumith

+13 -1

0 comment

2 changed files

peterjc123

pr closed time in 23 days

push eventpytorch/pytorch

peterjc123

commit sha f067088a8f0facc6cde7187e278540add7d5f47b

Unify changes and skip the test (#32339) Skip three failed tests in test_torch.py skip test_unmatched_type_annotation for Windows and Python 3.8 don't check existence for dirs in PATH Revert "Fix pip install" This reverts commit 614b63b57f63f21c12b1812b49359be2a7d8d109. Fix pip install Use absolute path for ctypes.cdll.LoadLibrary Fixes for conda and avoid importing torch in caffe2 fix lint again One more for nvtoolsext Skip check if it's available in one more path Correct relative import Use relative import More lint fixes Fix lint Apply changes to caffe2 Skip the check for CPU builds Fix dll load logic for Python 3.8 on Windows # Conflicts: # test/test_jit.py # torch/__init__.py

view details

push time in 25 days

PR merged pytorch/pytorch

Reviewers
Fix dll load logic for Python 3.8 on Windows (v1.4.0) open source

https://github.com/pytorch/pytorch/pull/32215 for v1.4.0.

+67 -31

1 comment

6 changed files

peterjc123

pr closed time in 25 days

push eventpytorch/cpuinfo

Ashkan Aliabadi

commit sha 0e6bde92b343c5fbcfe34ecd41abf9515d54b4a7

Upstream the cpuinfo updates in XNNPACK/third_party/cpuinfo.patch. Author: The XNNPACK Team.

view details

Soumith Chintala

commit sha e39a5790059b6b8274ed91f7b5b5b13641dff267

Merge pull request #29 from AshkanAliabadi/xnnpack Upstream the cpuinfo updates in XNNPACK/third_party/cpuinfo.patch

view details

push time in a month

PR merged pytorch/cpuinfo

Reviewers
Upstream the cpuinfo updates in XNNPACK/third_party/cpuinfo.patch

Author: The XNNPACK Team.

+1419 -145

0 comment

22 changed files

AshkanAliabadi

pr closed time in a month

startedfacebookresearch/pythia

started time in a month

issue commentpytorch/pytorch

pytorch 1.4 bug in conv2D kernel size 1x1 RuntimeError: cuDNN error: CUDNN_STATUS_NOT_SUPPORTED. This error may appear if you passed in a non-contiguous input

1.4.0 is frozen, if we plan for a 1.4.1, we should mark this as 1.4.1 backport-target. Right now, we have no plans for 1.4.1

One-sixth

comment created time in a month

startedfacebookresearch/SlowFast

started time in a month

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

the third way is not possible, as it's not clear that we will have a user-writeable folder available after the wheel is installed

asears

comment created time in a month

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

torch-deps which only includes our DLLs is not okay. We dont have license to redistribute DLLs separately, we can only redistributed them with our package's binary bits.

asears

comment created time in a month

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

yes, CUDA and NVIDIA refuse to provide one, I asked them several times.

asears

comment created time in a month

issue closedpytorch/pytorch

showing symbolic representation

current scenario - when I do

import torch
torch.nn.Sigmoid()

then it gives output

Sigmoid()

expected scenario - I do something like

import torch
torch.nn.Sigmoid(symbolic=True)

gives Screenshot (230) or Screenshot (232)

closed time in a month

vainaijr

issue commentpytorch/pytorch

showing symbolic representation

no, i don't think this is useful. It sounds like documentation? We should put that / already have the tex formulas in docstrings

vainaijr

comment created time in a month

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

the size limit right now is 750MB, and we already asked for way too much, compared to everyone else. If you see pypa/packaging-problems, you will find issues that I filed. The windows binaries are larger than 750MB

asears

comment created time in a month

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

Okay, I am wrong in the sense that, I didn't notice that we pushed 0.4.1 Windows binaries to torchvision pip -- we shouldn't have. If we shouldn't have, but we did, might as well do 0.5.0.

I've uploaded torchvision 0.5.0 to PyPI now, and updated our release runbook to do so in the future.

the torch package which torchvision depends on will not be uploaded for Windows, until if and when we figure out a size issue and we have more people working on Windows.

asears

comment created time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha 552b0e5861928d0151ece5e714b3cd419cb74793

http -> https

view details

push time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha 70fe29b4028438fd4ac9a8bf782736ce756b5f53

http -> https

view details

push time in a month

issue closedpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

The integration with PIL 7.0 VERSION API change is breaking some dependencies. I can see these are fixed in Linux / Mac releases.

Doesn't look like Windows version has been published to pypi? https://pypi.org/project/torchvision/0.5.0/#files

Last release appears to be https://pypi.org/project/torchvision/0.4.1/#files

References #1759 #1712 #1718 #1726

closed time in a month

asears

issue commentpytorch/vision

Pytorchvision 0.5.0 not published to pypi or Conda for Windows

The windows version of torchvision (and pytorch) were never uploaded to PyPI. The Windows PyTorch version exceeds the binary size limits, and hence we cant upload it. Because torchvision depends on the pytorch package, it wont make sense to upload just the torchvision Windows package to PyPI, as a pip install will surely fail.

I suggest that in your requirements.txt you add something like this:

-f https://download.pytorch.org/whl/torch_stable.html
torch===1.4.0
torchvision===0.5.0 
asears

comment created time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha f71ea6b608565500e5c6230e56069383438edd7b

add fair reference

view details

push time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha fef759f55c55ffb7bc354be69dbd1da1b5c85656

disable sitemap

view details

push time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha c9b15812b4c4b5b159ab2821d2744b447c4ad4ec

disable projects

view details

push time in a month

push eventsoumith/soumith.github.io

Soumith Chintala

commit sha 013bd68a70c5f19438c933e2d5419d803c31aa02

http until dns is updated

view details

push time in a month

more