Add python 3.14 support for PyTorch

[ScottTodd]

Python 3.14 was released on October 7, 2025 but we do not yet publish packages for it and it is already the default Python version on Fedora and other operating systems.

References:

• https://www.python.org/downloads/release/python-3140/
• https://docs.python.org/3/whatsnew/3.14.html

Related issues:

• Python 3.14 support for PyTorch pytorch/pytorch#156856
  • We generally track what PyTorch publishes
• Add python 3.10 support for Pytorch #2449
  • Add Python 3.10 Support for PyTorch Builds #2422
  • Revert "Add Python 3.10 Support for PyTorch Builds" #2580
  • [Pytorch] Python 3.10 venv setup failing: Package 'networkx' requires a different Python: 3.10.11 not in '>=3.11' #2463

Code:

• TheRock/RELEASES.md

  Line 41 in 41b9e77

  We currently support Python 3.11, 3.12, and 3.13.

• TheRock/.github/workflows/release_portable_linux_pytorch_wheels.yml

  Line 90 in 41b9e77

  python_version: ["3.11", "3.12", "3.13"]

• TheRock/.github/workflows/release_windows_pytorch_wheels.yml

  Line 90 in 41b9e77

  python_version: ["3.11", "3.12", "3.13"]

[ScottTodd]

The error messages from Python when a package is not supported on a given Python version can be confusing:

D:\scratch\therock
λ py -V:3.14 -m venv 3.14.venv && .\3.14.venv\Scripts\activate.bat

D:\scratch\therock
(3.14.venv) λ pip install --index-url=https://rocm.nightlies.amd.com/v2-staging/gfx110X-all/ torch
Looking in indexes: https://rocm.nightlies.amd.com/v2-staging/gfx110X-all/
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

Users are confused too: https://old.reddit.com/r/ROCm/comments/1pkxihc/amd_radeon_rx_9070_xt_not_a_supported_wheel_on/. There they were following https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html, which uses an explicit wheel version (see cp312 in the wheel names and the "To install the following wheels, Python 3.12 must be installed." text at the top of the page):

(3.14.venv) λ pip install --no-cache-dir ^     https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torch-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^     https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchaudio-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^     https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchvision-0.24.0+rocmsdk20251116-cp312-cp312-win_amd64.whl
ERROR: torch-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl is not a supported wheel on this platform

[tlee933]

nice! i spent days. working with rdna2 (gfx1031 native NO_HSA) fedora43
2.62x faster than RTX 3070 in Conv2D operations

[tlee933]

Yes, I will look into it tonight. I do have it working on my workstation. I issue is missing composable kernel for gfx103X and rocm lagging behind python 3.14. it prefers python 3.12. I will try push up built pytorch packages. I'm guessing you are on Fedora43 as well. :)

…

On Mon, Jan 5, 2026, 3:53 AM Megara ***@***.***> wrote: *Megara18* left a comment (ROCm/TheRock#2640) <#2640 (comment)> There is Any ETA for this? Im studying with Fedora Server and i would need PyTorch for 3.14 — Reply to this email directly, view it on GitHub <#2640 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B2ADJ5FI4JBFCKDSRWFBHYL4FI7A7AVCNFSM6AAAAACPPJB5ESVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMBZHEZDONBVHE> . You are receiving this because you commented.Message ID: ***@***.***>

[tlee933]

That's awesome. Happy to help! Are you running the 6700 gfx1031?

…

On Mon, Jan 5, 2026, 5:50 AM Megara ***@***.***> wrote: *Megara18* left a comment (ROCm/TheRock#2640) <#2640 (comment)> I am! and thank you so much, is my firts year at the University, yes is Fedora 43 server — Reply to this email directly, view it on GitHub <#2640 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B2ADJ5HPZ4HCBZR6UAKUDX34FJMZLAVCNFSM6AAAAACPPJB5ESVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMJQGI4TEMRUGA> . You are receiving this because you commented.Message ID: ***@***.***>

[tlee933]

Very nice. I can configure for that device. AMD has been spotty for support with various devices. Fortunately for HSA_OVERRIDE we can force support for some. Will look into it that model. Often family of supported device. (Same Silicon)

…

On Mon, Jan 5, 2026, 8:42 AM Megara ***@***.***> wrote: *Megara18* left a comment (ROCm/TheRock#2640) <#2640 (comment)> Oh nono, i did an "auto-gift" to myself and im running the new Ryzen Ai Max+395, so is gfx1151 haha — Reply to this email directly, view it on GitHub <#2640 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B2ADJ5GJZ3LQKPENIDC4WJ34FKA5VAVCNFSM6AAAAACPPJB5ESVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMJQHE2TEMZZHE> . You are receiving this because you commented.Message ID: ***@***.***>

[tlee933]

Oh! Also https://GitHub.com/rikimeow/pytorch-gfx1151 :)

…

On Mon, Jan 5, 2026, 9:57 AM Tony Lee ***@***.***> wrote: Very nice. I can configure for that device. AMD has been spotty for support with various devices. Fortunately for HSA_OVERRIDE we can force support for some. Will look into it that model. Often family of supported device. (Same Silicon) On Mon, Jan 5, 2026, 8:42 AM Megara ***@***.***> wrote: > *Megara18* left a comment (ROCm/TheRock#2640) > <#2640 (comment)> > > Oh nono, i did an "auto-gift" to myself and im running the new Ryzen Ai > Max+395, so is gfx1151 haha > > — > Reply to this email directly, view it on GitHub > <#2640 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/B2ADJ5GJZ3LQKPENIDC4WJ34FKA5VAVCNFSM6AAAAACPPJB5ESVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMJQHE2TEMZZHE> > . > You are receiving this because you commented.Message ID: > ***@***.***> >

[marbre]

@Megara18 and @tlee933 please either open a discussion at https://github.com/ROCm/TheRock/discussions or consider switching to Discord eventually. It makes it hard to follow on the actual issue. That said we will start work to enable 3.14 this week, hoping that things go smooth.