Pytorch for Python 3.12 not available #110436
Labels
high priority
module: build
Build system issues
module: python frontend
For issues relating to PyTorch's Python frontend
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Comments
colesbury
added
module: build
|
It takes some time to make torch compatible with new versions. Be patient! |
|
Hi, |
b3n4kh
added a commit
to ait-cs-IaaS/Taranis-NG
that referenced
this issue
Oct 6, 2023
With worker beeing incompatible with python 3.12: pytorch/pytorch#110436 With core having an incompatible dependency with flask 3 python-restx/flask-restx#567
1 task
dstein64
added a commit
to dstein64/pastiche
that referenced
this issue
Oct 6, 2023
pytorch is not available yet. pytorch/pytorch#110436
b3n4kh
added a commit
to taranis-ai/taranis-ai
that referenced
this issue
Oct 17, 2023
With worker beeing incompatible with python 3.12: pytorch/pytorch#110436 With core having an incompatible dependency with flask 3 python-restx/flask-restx#567
|
Is it possible to have core-functionality released / RC version? Considering Pytorch build from source on Windows is that friendly. |
5 tasks
Closed
1 task
|
I tried to build from source, but it failed at the last step: full logs... |
|
Also need Windows py3.12 version for CPU |
3 tasks
|
The version of Python shipped with Homebrew on macOS upgraded to Python 3.12 recently, which is causing problems for my https://llm.datasette.io tool as some of its plugins depend on PyTorch: |
|
Bumping priority due to user activity. Expected steps are:
The whole thing is going to take some time but hopefully we can get mostly working binaries to unblock sooner. |
|
Hey! Closing this as done and you can follow #120233 for torch.compile support specifically. |
4 tasks
aavashsubedi
added a commit
to aavashsubedi/graph_weather
that referenced
this issue
Mar 11, 2024
Removed check for torch 2.0.0 and 2.1.0 for Python version 3.12.0. pytorch/pytorch#110436
aavashsubedi
added a commit
to aavashsubedi/graph_weather
that referenced
this issue
Mar 11, 2024
Excluding torch version 2.2.0 with Python 3.12 until fixed. pytorch/pytorch#110436 Also updated py-coverage versions and directory.
2 tasks
jacobbieker
pushed a commit
to openclimatefix/graph_weather
that referenced
this issue
Mar 11, 2024
Excluding torch version 2.2.0 with Python 3.12 until fixed. pytorch/pytorch#110436 Also updated py-coverage versions and directory.
1 task
32 tasks
|
What about pytorch for Python 3.12, as opposed to torch? When I try to install pytorch, it fails: $ python --version Edition Windows 10 Pro |
|
@LyleZiegelmiller it might be useful to see that package's PyPI page:
EDIT: |
|
Thanks for your response. I'll take your word for it that I should install torch instead of pytorch, but it's odd that it even tries to install pytorch. Consider this: Notice it doesn't progress anywhere near as far with the installation as it did when I tried to install pytorch. Anyway, thanks for your help. |
|
@LyleZiegelmiller the Pytorch package was initially created by @soumith in 2017 for the early version (0.1.2) of pytorch, after that IMHO it'd be reasonable to make pytorch Pypi package to point to torch package so users are not confused. |
|
Sounds like a good idea. |
|
Can we have |
|
Ho very good point @kevalmorabia97 |
|
Python versions are moving so fast. 3.13 is almost out. Given the PyTorch development capabilities, have you guys considered rather than trying to make it compatible for each version. maybe every 5 versions? Like 3.10, 3.15, 3.20 etc.? |
Object your opinion，since python is moving fast, pytorch also needs to catch up . |
|
IMO the correct long term strategy would be to just use Py_LIMITED_API(https://docs.python.org/3/c-api/stable.html#limited-c-api) macro and make the wheels python version agnostic (after a min version), so the native binaries are decoupled from python versions and dont need to recompiled for each version separately. |
|
Yes moving faster is what we're going for. See https://github.com/pytorch/rfcs/blob/master/RFC-0038-cpython-support.md for the plan to make sure we can get 3.13 working in a timely manner ! @ankurvdev I'm afraid the limited API doesn't provide nearly enough stuff for PyTorch to work. |
jvansanten
added a commit
to AmpelAstro/Ampel-HU-astro
that referenced
this issue
Dec 3, 2024
* Bump version to 0.10a0 * Follow v0.8 -> v0.9 API churn AbsPlainT3Unit -> AbsT4Unit AbsReviewT3Unit -> AbsT3Unit * Follow AlertConsumer API churn * RapidBase: slack -> slack_sdk * ruff: ignore more files * ci: run on py311 PyTorch stable does not support 3.12 yet See: pytorch/pytorch#110436 * T2RiseDeclineBase: inherit from LogicalUnit AmpelBaseModel subclasses can't set extra attributes by default. * test: init parameterized model via dict * conf: follow T3ReviewUnit churn * chore: activate warnings * ligo not our problem * ignore numpy header size warning * fixup! chore: activate warnings * T2RiseDeclineStat: squash numpy scalar conversion warning * SlackSummaryPublisher: fillna in bulk for pandas 3.0 compat * chore(deps): refresh lockfile * Bump version to 0.10.0a0 * HealpixCorrPlotter: conform to matplotlib 3.9 API * Remove vestigial ResourceDependentConsumer * Bump version to 0.10.0a1 * T2RiseDeclineBase: demote to plain class Since this will always be mixed into the inheritance chain of AmpelUnit or BaseModel, both of which harvest annotations from their bases, there's not actually a reason for this to be a BaseModel itself. * PlotTransientLightCurves: require no slack token by default * Convert stellar_outburst example to v0.10 syntax * chore(deps): advance ampel-core to 0.10.1 * chore: explciitly annotate typeddict arg * chore: cast t2 payloads to dict Ideally there would be a declared result type, such that downstream consumers can be type checked * ruff: remove empty comments * ruff: simplify * ruff: % to f-strings * ruff: simplify min calls * T2RunTDE: unnecessary staticmethod * ci: update ruff * T2ElasticcReport: bail if either z is None * ruff: format * ruff: fixup line endings * chore: hide type checking behind facade * Remove remaining casts * Ensure payload is Mapping * fix(deps): advance ampel-interface to v0.10.3a0 * fix(deps): advance ampel-interface to v0.10.3a1 * ruff: cleanup * thou shalt not print() * renovate: baseBranches * fix: use typing_extensions.TypedDict in models for python<3.12 compatibility. typing.TypedDict lacks necessary metadata before python 3.12; see https://docs.pydantic.dev/2.3/usage/types/dicts_mapping/#typeddict * CostCounter: return TypedDict * chore(deps): flip snpy back to mainline now compatible with both PEP517 _and_ scipy 1.14 * DCachePublisher: reimplement in terms of requests * TNSClient: replace aiohttp with requests good riddance, i say * chore(deps): remove ruff * chore: move t4 units to t4 directory * chore: sort unit config * chore: remove more commented-out print() * chore: Update README for 73f2591
jvansanten
added a commit
to AmpelAstro/Ampel-PanSTARRS
that referenced
this issue
Dec 5, 2024
* Bump version to 0.10a0 * Follow v0.8 -> v0.9 API churn AbsPlainT3Unit -> AbsT4Unit AbsReviewT3Unit -> AbsT3Unit * Follow AlertConsumer API churn * RapidBase: slack -> slack_sdk * ruff: ignore more files * ci: run on py311 PyTorch stable does not support 3.12 yet See: pytorch/pytorch#110436 * T2RiseDeclineBase: inherit from LogicalUnit AmpelBaseModel subclasses can't set extra attributes by default. * test: init parameterized model via dict * conf: follow T3ReviewUnit churn * chore: activate warnings * ligo not our problem * ignore numpy header size warning * fixup! chore: activate warnings * T2RiseDeclineStat: squash numpy scalar conversion warning * SlackSummaryPublisher: fillna in bulk for pandas 3.0 compat * chore(deps): refresh lockfile * Bump version to 0.10.0a0 * HealpixCorrPlotter: conform to matplotlib 3.9 API * Remove vestigial ResourceDependentConsumer * Bump version to 0.10.0a1 * T2RiseDeclineBase: demote to plain class Since this will always be mixed into the inheritance chain of AmpelUnit or BaseModel, both of which harvest annotations from their bases, there's not actually a reason for this to be a BaseModel itself. * PlotTransientLightCurves: require no slack token by default * Convert stellar_outburst example to v0.10 syntax * chore(deps): advance ampel-core to 0.10.1 * chore: explciitly annotate typeddict arg * chore: cast t2 payloads to dict Ideally there would be a declared result type, such that downstream consumers can be type checked * ruff: remove empty comments * ruff: simplify * ruff: % to f-strings * ruff: simplify min calls * T2RunTDE: unnecessary staticmethod * ci: update ruff * T2ElasticcReport: bail if either z is None * ruff: format * ruff: fixup line endings * chore: hide type checking behind facade * Remove remaining casts * Ensure payload is Mapping * fix(deps): advance ampel-interface to v0.10.3a0 * fix(deps): advance ampel-interface to v0.10.3a1 * ruff: cleanup * thou shalt not print() * renovate: baseBranches * fix: use typing_extensions.TypedDict in models for python<3.12 compatibility. typing.TypedDict lacks necessary metadata before python 3.12; see https://docs.pydantic.dev/2.3/usage/types/dicts_mapping/#typeddict * CostCounter: return TypedDict * chore(deps): flip snpy back to mainline now compatible with both PEP517 _and_ scipy 1.14 * DCachePublisher: reimplement in terms of requests * TNSClient: replace aiohttp with requests good riddance, i say * chore(deps): remove ruff * chore: move t4 units to t4 directory * chore: sort unit config * chore: remove more commented-out print() * chore: Update README for 73f2591d4e6d3de8579accbe2bd0b114ebeca8a5
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
and others
🚀 The feature, motivation and pitch
On Windows, Pytorch installation is not available:
Alternatives
No response
Additional context
No response
cc @ezyang @gchanan @zou3519 @kadeng @malfet @seemethere @albanD
The text was updated successfully, but these errors were encountered: