Bump torch from 2.7.0+cpu to 2.13.0

[dependabot]

Bumps torch from 2.7.0+cpu to 2.13.0.

Release notes

Sourced from torch's releases.

PyTorch 2.13.0 Release Notes

• Highlights
• Backwards Incompatible Changes
• Deprecations
• New Features
• Improvements
• Bug fixes
• Performance
• Documentation
• Developers

Highlights

For more details about these highlighted features, you can look at the release blogpost. Below are the full release notes for this release.

Tracked Regressions

ROCm wheels break torch.compile on CPU in environments without a GPU

Running a torch==2.13.0+rocm7.2 wheel in an environment where no GPU is available (torch.cuda.is_available() is False) breaks torch.compile on the CPU path: the first compile raises RuntimeError: Can't detect vectorized ISA for CPU (#189194). This is a regression from torch==2.12.1+rocm7.2, which compiles CPU code fine (detecting e.g. VecAVX2) in the same setup. The 2.13 ROCm wheel appears to rely on something present in the ROCm builder image to detect the CPU vectorized ISA, so it works when run on a ROCm image but fails on a plain CPU-only image.

Workaround: run the +rocm wheel on a ROCm image, or install a standard CPU/CUDA build for GPU-less environments.

Backwards Incompatible Changes

• Stop building CPython 3.13t (free-threaded) binaries (#182951)

  Upstream pypa/manylinux removed CPython 3.13t (free-threaded) on 2026-05-07, because 3.13t was experimental and has been superseded by the now-non-experimental CPython 3.14t. As a result, PyTorch 2.13 no longer ships cp313t wheels (Linux, Triton, and related artifacts). Users on the free-threaded interpreter should move to Python 3.14t.

  PyTorch 2.12:

  # cp313t (free-threaded 3.13) wheels were available
  python3.13t -m pip install torch

  PyTorch 2.13:

... (truncated)

Commits

• See full diff in compare view

[Dongxu-H]

This PR replaces vendor-specific torch builds with generic PyPI version torch 2.13.0, which breaks all backend compatibility.

torch-npu requires torch 2.9.0 or 2.10.0 to work with Ascend CANN, not torch 2.13.0.
torch-mlu requires torch 2.7.1 to work with Cambricon MLU.
torch-gcu requires torch 2.10.0 to work with Enflame GCU.
Metax backend uses torch 2.8.0+metax3.7.2.0 which is vendor-specific and not interchangeable with standard torch.
Kunlunxin backend uses torch 2.9.0+cu129 which includes vendor patches.
Nvidia backends use torch 2.10.0+cu128 and 2.11.0+cu130 with specific CUDA versions.

Each backend has a vendor-specific torch version that is tightly coupled with their hardware drivers and libraries. Replacing them with standard torch 2.13.0 will cause runtime failures.

Dependabot does not understand vendor-specific torch version constraints. This PR should be closed.

[Dongxu-H]

This PR breaks all vendor-specific torch dependencies.

ascend-cann850 uses torch==2.9.0+cpu which is tightly coupled with torch-npu==2.9.0. Replacing with torch==2.13.0 will cause runtime failures.

ascend-cann900 uses torch==2.10.0+cpu which is tightly coupled with torch-npu==2.10.0.

cambricon uses torch==2.7.1+cpu which is tightly coupled with torch-mlu==1.29.2+torch2.7.1 and torch-mlu-ops==1.8.0+torch2.7.1.

enflame uses torch==2.10.0+cpu which is tightly coupled with torch-gcu and flash-attn vendor builds.

hygon uses torch==2.9.0+das.opt1.dtk2604 which is a vendor-specific build.

iluvatar uses torch==2.7.1+corex.4.4.0 which is a vendor-specific build.

kunlunxin uses torch==2.9.0+cu129 which is tightly coupled with torch_plugin.

metax uses torch==2.8.0+metax3.7.2.0 which is tightly coupled with flash_attn==2.6.3+metax3.7.2.0torch2.8.

mthreads-musa436 uses torch==2.9.0+musa.4.3.6 which is a vendor-specific build.

mthreads-musa520 uses torch==2.9.1+musa5.2.0 which is a vendor-specific build.

nvidia-cuda128 and nvidia-cuda133 have mismatched torch/torchaudio/torchvision versions.

spacemit uses torch==2.8.0+spacemit.0 which is a vendor-specific build.

Dependabot does not understand vendor-specific torch version constraints. Each backend has a torch variant tightly coupled with their hardware drivers. Replacing them with standard torch==2.13.0 will cause runtime failures across all backends. This PR should be closed.

[Dongxu-H]

torch is a critical package and must use vendor-specific versions for different backends. Each backend (ascend, cambricon, enflame, hygon, iluvatar, kunlunxin, metax, mthreads, nvidia, spacemit, sunrise, thead, tsingmicro) requires a specific torch build with platform-specific optimizations and compatibility. Replacing all vendor-specific torch versions with a unified 2.13.0 will break compatibility across all backends. This change is not acceptable.

[Dongxu-H]

torch 2.13.0 移除了所有厂商特定的版本后缀（如 +das.opt1.dtk2604、+metax3.7.2.0、+musa.4.3.6、+corex.4.4.0）。这些后缀标识了针对特定硬件平台定制的 PyTorch 版本。移除后缀将导致 ascend、hygon、iluvatar、metax、mthreads、thead 等后端无法正常工作。不同厂商需要使用特定版本的 PyTorch 才能正确调用底层硬件。建议回退此更改，保留厂商特定的 torch 版本。

[Dongxu-H]

CI failed: code-style check failed. This major version upgrade from torch 2.7.0 to 2.13.0 requires careful review for compatibility with existing Triton kernels and CUDA operations. The jump from 2.7 to 2.13 spans multiple major versions and likely includes breaking changes. Need to verify all tests pass before merging.

[Dongxu-H]

CI code-style check failed. Updating torch versions across all backends to 2.13.0 will break vendor-specific SDK compatibility. Each backend requires specific torch versions to match their vendor SDKs (e.g., ascend-cann850 needs torch==2.9.0+cpu with torch-npu==2.9.0, metax needs torch==2.8.0+metax3.7.2.0, cambricon needs torch==2.7.1+cpu with torch-mlu==1.29.2+torch2.7.1). This update would break all backend-specific builds.

[Dongxu-H]

CI code-style check failed.
Updating torch from 2.7.0+cpu to 2.13.0 is a major version upgrade that changes torch versions across all vendor-specific configurations from their specialized builds to a generic 2.13.0 version. This may break compatibility with vendor-specific backends like ascend, cambricon, iluvatar, metax, mthreads, kunlunxin, etc. Each vendor typically requires specific torch builds that are compatible with their hardware and runtime. Verify that this upgrade is intentional and that all vendor backends will work with torch 2.13.0 before merging.

[Dongxu-H]

CI code-style check failed.

torch version mismatch with backend-specific packages:
cambricon-neuware443: torch-mlu==1.29.2+torch2.7.1 requires torch 2.7.1, not 2.13.0
cambricon-neuware472: torch-mlu==1.33.1+torch2.11.0 requires torch 2.11.0
enflame-tops1910: torch-gcu==2.10.0 requires torch 2.10.0
enflame-tops1106: torch-gcu==2.11.0 requires torch 2.11.0
metax-maca3810: flash_linear_attention shows torch2.10 dependency
iluvatar: torch==2.7.1+corex.4.4.0 version indicates specific torch version requirement

Backend-specific torch packages (torch-mlu, torch-npu, torch-gcu, etc.) have strict torch version requirements. Bumping torch to 2.13.0 may cause runtime incompatibility or import failures.

[dependabot]

OK, I won't notify you again about this release, but will get in touch when a new version is available. If you'd rather skip all updates until the next major or minor version, let me know by commenting @dependabot ignore this major version or @dependabot ignore this minor version. You can also ignore all major, minor, or patch releases for a dependency by adding an ignore condition with the desired update_types to your config file.

If you change your mind, just re-open this PR and I'll resolve any conflicts on it.