Add Python-3.10 (build and test)

[malfet]

No description provided.

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit ab5eae0 (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

build_docs (1/1)

Step: "Build docs" (full log | diagnosis details | 🔁 rerun)

Warning, treated as error:

loading intersphinx inventory from https://matplotlib.org/stable/objects.inv...
generating gallery...
Using Sphinx-Gallery to convert rst text blocks to markdown for .ipynb files.
generating gallery for auto_examples... [ 16%] plot_scripted_tensor_transforms.py
generating gallery for auto_examples... [ 33%] plot_optical_flow.py
generating gallery for auto_examples... [ 50%] plot_repurposing_annotations.py
generating gallery for auto_examples... [ 66%] plot_transforms.py
generating gallery for auto_examples... [ 83%] plot_video_api.py
generating gallery for auto_examples... [100%] plot_visualization_utils.py

Warning, treated as error:
/root/project/gallery/plot_visualization_utils.py failed to execute correctly: Traceback (most recent call last):
  File "/root/project/gallery/plot_visualization_utils.py", line 86, in <module>
    outputs = model(batch)
  File "/root/project/env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1111, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/project/torchvision/models/detection/generalized_rcnn.py", line 98, in forward
    proposals, proposal_losses = self.rpn(images, features, targets)
  File "/root/project/env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1111, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/project/torchvision/models/detection/rpn.py", line 353, in forward

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[jdsgomes]

I think these fails can be solved by replacing this

with

gamma_map = [round((255 + 1 - 1e-3) * gain * pow(ele / 255.0, gamma)) for ele in range(256)] * 3

Looking at the code I believe this is equivalent and solves the float/int issue.

[malfet]

gamma_map = [round((255 + 1 - 1e-3) * gain * pow(ele / 255.0, gamma)) for ele in range(256)] * 3

round or int (also, not sure which one is faster)?
Indeed it needs int otherwise it causes another failure: https://app.circleci.com/pipelines/github/pytorch/vision/14807/workflows/baa8f094-988d-4ab1-a643-b7f34788a702/jobs/1191741

[malfet]

Moved adjust_gamma change to #5427

[datumbox]

@jdsgomes We will probably need to cherrypick #5427 on the release branch. I assume 3.10 support will be added to this release right?

[pmeier]

I'm guessing this is not synchronized with #5320? cc @vfdev-5

[datumbox]

@pmeier that was just a check, we can close the other one.

[pmeier]

Note that #5320 was mostly a superset of what is proposed here. It also handled the README compatibility table, as well as docker and OSX wheels.

[malfet]

Note that #5320 was mostly a superset of what is proposed here. It also handled the README compatibility table, as well as docker and OSX wheels.

Wow, I wasn't aware of osx_wheel.sh, but at glance at this script it does not seem to be used (nor work) right now, so I propose we delete them in #5433

[datumbox]

Seems like the failing tests are due to dependency issues from core and numpy.

[datumbox]

@malfet @atalman I had a look at the failing tests and it seems the issues are related to dependencies issues from Core.

The binary_win_conda_py3.10_cpu fails due to multiple conflicts:

Package pytorch-mutex conflicts for:
pytorch==1.12.0.dev20220207 -> pytorch-mutex==1.0[build='cuda|cpu']
pytorch-mutex==1.0=cpu

Package python conflicts for:
pytorch==1.12.0.dev20220207 -> typing_extensions -> python[version='>=2.7,<2.8.0a0|>=3.5|>=3.8,<3.9.0a0|>=3.6,<3.7.0a0|>=3.7,<3.8.0a0']
python=3.10
pytorch==1.12.0.dev20220207 -> python[version='>=3.10,<3.11.0a0']

Package setuptools conflicts for:
setuptools
python=3.10 -> pip -> setuptools

The unittest_windows_cpu_py3.10 fails because PyTorch core has been compiled with a different version of numpy than the one installed:

Failed to initialize NumPy: module compiled against API version 0xf but this version of numpy is 0xe 

Nevertheless all other unittest_windows_cpu_py* jobs pass which makes me wonder if core gets compiled on the specific configuration with a different numpy version accidentally.

I understand that these issues can be fixed upstream, but if you think otherwise let me know to further investigate. Thanks!

[malfet]

The binary_win_conda_py3.10_cpu fails due to multiple conflicts:

Package pytorch-mutex conflicts for:
pytorch==1.12.0.dev20220207 -> pytorch-mutex==1.0[build='cuda|cpu']
pytorch-mutex==1.0=cpu

Package python conflicts for:
pytorch==1.12.0.dev20220207 -> typing_extensions -> python[version='>=2.7,<2.8.0a0|>=3.5|>=3.8,<3.9.0a0|>=3.6,<3.7.0a0|>=3.7,<3.8.0a0']
python=3.10
pytorch==1.12.0.dev20220207 -> python[version='>=3.10,<3.11.0a0']

Package setuptools conflicts for:
setuptools
python=3.10 -> pip -> setuptools

This was resolved by pytorch/builder@3e6551c

The unittest_windows_cpu_py3.10 fails because PyTorch core has been compiled with a different version of numpy than the one installed:

Failed to initialize NumPy: module compiled against API version 0xf but this version of numpy is 0xe 

Nevertheless all other unittest_windows_cpu_py* jobs pass which makes me wonder if core gets compiled on the specific configuration with a different numpy version accidentally.

Seems to be connected to the previous failure

[github-actions]

Hey @malfet!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py