Bump PyTorch pin to 2.13

[JacobSzwejbka]

Summary
Bumps the ExecuTorch PyTorch pin to 2.13.0 and updates the matching torchvision/source branch pins to 0.28/release/0.28.

Also syncs c10 headers from PyTorch release/2.13, bumps third-party/ao to upstream commit 4aa810113, and pins source installs to torchao==0.18.0.dev20260715 from the PyTorch nightly wheel index. The torchao changes include pytorch/ao#4462 (Remove named tensor Dimname overload reference), needed for PyTorch 2.13 trunk quantization flows.

Pins torchtune installs to 3c1872ace149f03ef4ffec765d3f0a5fd0399497 from meta-pytorch/torchtune#2960 (Fix import error after torchao NF4Tensor move), because the latest stable PyPI release and the available PyTorch wheel-index torchtune versions still import torchao.dtypes.nf4tensor.

For Docker image builds, uses GCC/G++ for the PyTorch source build when the image default C++ compiler is Clang older than 16. PyTorch 2.13 rejects Clang 12 during configure, while the clang12 image should keep Clang 12 as its default compiler for ExecuTorch jobs.

Updates ExecuTorch's exportable NVFP4 quantization wrapper for the newer torchao API: when per-tensor scaling is disabled, pass None to torchao's nvfp4_quantize and serialize a scalar 1.0 tensor for ExecuTorch dequantization.

For Metal MPS torchao kernels, keeps the nightly torchao wheel as the default dependency but installs third-party/ao from the pinned submodule when TORCHAO_BUILD_EXPERIMENTAL_MPS=1 is set, so the Metal jobs get libtorchao_ops_mps_aten.dylib.

Adds the PyTorch 2.13 AOTI runtime aoti_torch_get_numel shim to ExecuTorch's common and SlimTensor AOTI shims, including coverage for the CUDA-Windows export path that links AOTInductor-generated wrappers. Regenerates the checked-in CUDA-Windows aoti_cuda_shims.lib import library so export-time wrapper links can resolve the new symbol.

Test Plan
python3 -m py_compile extension/llm/export/nvfp4.py install_requirements.py torch_pin.py
bash -n .ci/docker/common/install_pytorch.sh examples/models/llama/install_requirements.sh
c++ -std=c++17 -DC10_USING_CUSTOM_GENERATED_MACROS -I. -I.. -Iruntime/core/portable_type/c10 -c backends/aoti/common_shims_slim.cpp -o /tmp/common_shims_slim.o
c++ -std=c++17 -DC10_USING_CUSTOM_GENERATED_MACROS -I. -I.. -Iruntime/core/portable_type/c10 -c backends/aoti/common_shims.cpp -o /tmp/common_shims.o
nm -g backends/cuda/runtime/aoti_cuda_shims.lib | grep aoti_torch_get_numel
git diff --check
Verified torchao 0.18.0.dev20260715 exists in PyTorch nightly cpu/cu126/cu130/cu132 indexes.
Verified torchtune 0.6.1, 0.7.0+cpu, and 0.7.0.dev20250929+cpu still import torchao.dtypes.nf4tensor; torchtune commit 3c1872ace149f03ef4ffec765d3f0a5fd0399497 imports NF4Tensor from torchao.quantization.
.ci/scripts/compare_dirs.sh runtime/core/portable_type/c10/c10 /tmp/executorch-pytorch-c10-sync.hRGjXi/c10
.ci/scripts/compare_dirs.sh runtime/core/portable_type/c10/torch/headeronly /tmp/executorch-pytorch-c10-sync.hRGjXi/torch/headeronly

Note: could not run the failing XNNPACK export, Phi-4-mini aarch64 model job, MLX NVFP4 job, Metal model artifact export, CUDA-Windows export, or full Docker image build locally because this environment does not have torch/docker configured for those CI paths.

This PR was authored with Claude.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20967

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 6 New Failures, 2 Unclassified Failures

As of commit 91e7fb6 with merge base af62dce ():

NEW FAILURES - The following jobs have failed:

• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / windows-job (gh)
  Process completed with exit code 1.
• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-3B-2507, non-quantized) / windows-job (gh)
  Process completed with exit code 1.
• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / windows-job (gh)
  Process completed with exit code 1.
• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / windows-job (gh)
  Process completed with exit code 1.
• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (nvidia, parakeet-tdt, non-quantized) / windows-job (gh)
  Process completed with exit code 1.
• Test CUDA Windows Export and E2E / test-model-cuda-windows-e2e (nvidia, parakeet-tdt, quantized-int4-weight-only) / windows-job (gh)
  Process completed with exit code 1.

UNCLASSIFIED FAILURES - DrCI could not classify the following jobs because the workflow did not run on the merge base. The failures may be pre-existing on trunk or introduced by this PR:

• Test CUDA Builds / export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• trunk / test-llama-torchao-lowbit / macos-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.