Bumps torch to 2.12

[kellyguo11]

Description

Bumps pytorch to latest 2.12 stable release

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (existing functionality will not work without user modification)
• Documentation update

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[greptile-apps]

Greptile Summary

This PR bumps the required PyTorch version from 2.10 to 2.12 (stable) and torchvision from 0.25.x to 0.27.x across all setup.py files, the wheel-builder config, and all documentation install guides.

• source/isaaclab/setup.py, source/isaaclab_rl/setup.py, source/isaaclab_tasks/setup.py, and tools/wheel_builder/res/python_packages.toml all consistently raise the torch lower bound to >=2.12 and torchvision to >=0.27.0.
• All four documentation .rst files update the pinned install commands to torch==2.12.0 torchvision==0.27.0, but the Linux x86_64 and Windows x86_64 tabs still reference --index-url https://download.pytorch.org/whl/cu128 (CUDA 12.8), which PyTorch 2.12 dropped — users on those platforms will hit a package-not-found error and the install will fail.

Confidence Score: 3/5

The Python package version bumps are clean and consistent, but all four documentation files direct x86_64 users to a CUDA 12.8 wheel index that does not carry torch 2.12 packages, making the documented install commands non-functional on those platforms.

The setup.py and python_packages.toml changes are correct. However, every documentation install guide still points x86_64 users (Linux and Windows) to the cu128 index, and PyTorch 2.12 dropped CUDA 12.8 support entirely. Anyone following the quickstart or pip-installation guides on the two most common platforms will get a failed install. The aarch64 tab uses cu130 and is unaffected.

All four .rst documentation files need the cu128 index URL changed to cu126 for the Linux x86_64 and Windows x86_64 tabs.

Important Files Changed

Filename	Overview
docs/source/setup/quickstart.rst	Updated torch/torchvision versions to 2.12.0/0.27.0, but Linux x86_64 and Windows tabs still use cu128 index URL which has no torch 2.12 wheels.
docs/source/setup/installation/pip_installation.rst	Same cu128 index URL issue present for both x86_64 platform tabs; aarch64 (cu130) is correct.
docs/source/setup/installation/isaaclab_pip_installation.rst	Updated torch/torchvision versions correctly, but cu128 index URL will fail for Linux/Windows x86_64 under torch 2.12.
docs/source/experimental-features/newton-physics-integration/installation.rst	Newton integration guide updated to torch 2.12 / torchvision 0.27, but cu128 index URL is invalid for torch 2.12.
source/isaaclab/setup.py	Bumps torch lower-bound from >=2.10 to >=2.12; straightforward and correct.
source/isaaclab_rl/setup.py	Bumps torch to >=2.12 and torchvision to >=0.27.0 with updated comment; correct version pairing.
source/isaaclab_tasks/setup.py	Bumps torch to >=2.12 and torchvision to >=0.27.0; consistent with other setup.py changes.
tools/wheel_builder/res/python_packages.toml	All three dependency blocks updated to torch>=2.12 / torchvision>=0.27.0 consistently; no issues.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User follows install guide] --> B{Platform}
    B -->|Linux aarch64| C["cu130 index URL - torch==2.12.0 available"]
    B -->|Linux x86_64| D["cu128 index URL - torch==2.12.0 NOT available (deprecated)"]
    B -->|Windows x86_64| E["cu128 index URL - torch==2.12.0 NOT available (deprecated)"]
    D --> F[Fix: change cu128 to cu126]
    E --> F
    F --> G["cu126 index URL - torch==2.12.0 available"]

Loading

_{Reviews (1): Last reviewed commit: "Bumps torch to 2.12" | Re-trigger Greptile}

[greptile-apps]

PyTorch 2.12 dropped the CUDA 12.8 wheel (cu128). The official PyTorch 2.12 release compatibility matrix lists only CUDA 12.6 and CUDA 13.0 as stable wheel targets, and the 2.12 release blog explicitly announces "Deprecation of the CUDA 12.8 Wheel." Using --index-url https://download.pytorch.org/whl/cu128 with torch==2.12.0 will result in a package-not-found error for both the Linux x86_64 and Windows x86_64 tabs. The index URL should be changed to cu126 (or cu130) to match available wheels.

Suggested change

	.. code-block:: bash
	uv pip install -U torch==2.10.0 torchvision==0.25.0 --index-url https://download.pytorch.org/whl/cu128
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	uv pip install -U torch==2.10.0 torchvision==0.25.0 --index-url https://download.pytorch.org/whl/cu128
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	.. code-block:: bash
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126

[greptile-apps]

Same cu128 issue: PyTorch 2.12 does not publish CUDA 12.8 wheels, so --index-url https://download.pytorch.org/whl/cu128 with torch==2.12.0 will fail. The stable CUDA targets for 2.12 are cu126 (CUDA 12.6) and cu130 (CUDA 13.0). Both x86_64 tabs need cu128 changed to cu126.

Suggested change

	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	uv pip install -U torch==2.10.0 torchvision==0.25.0 --index-url https://download.pytorch.org/whl/cu128
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	.. tab-item:: :icon:`fa-brands fa-linux` Linux (aarch64)
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126
	.. tab-item:: :icon:`fa-brands fa-linux` Linux (aarch64)

[greptile-apps]

Same cu128 issue for this installation guide. The torch==2.12.0 packages are not published under the CUDA 12.8 index; cu126 is the correct stable target for x86_64 platforms.

Suggested change

	pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	pip install -U torch==2.10.0 torchvision==0.25.0 --index-url https://download.pytorch.org/whl/cu128
	pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126
	.. tab-item:: :icon:`fa-brands fa-windows` Windows (x86_64)
	:sync: windows-x86_64
	.. code-block:: bash
	pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126

[greptile-apps]

The Newton physics integration guide also installs torch via the cu128 index, which has no torch==2.12.0 wheel. This should be updated to cu126.

Suggested change

	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu128
	uv pip install -U torch==2.12.0 torchvision==0.27.0 --index-url https://download.pytorch.org/whl/cu126

[isaaclab-review-bot]

🔬 Code Review: PR #5629

Summary

This PR bumps PyTorch from 2.10 to 2.12 and torchvision from 0.25 to 0.27 across documentation and package dependencies. The changes are straightforward version bumps with no functional code changes.

✅ What Looks Good

• Version consistency: All documentation files and setup.py files are updated consistently
• CUDA index URLs preserved: The platform-specific CUDA wheel URLs (cu128 for x86_64, cu130 for aarch64) remain unchanged
• Package availability verified: Both torch==2.12.0 and torchvision==0.27.0 are available on PyPI
• Comment updates: The inline comments in setup.py files correctly reflect the new version compatibility

📋 Checklist Items to Address

Per the PR template, the following items are unchecked and may need attention:

1. Changelog — The "Check changelog fragments" CI check is failing. Consider adding a changelog fragment for this dependency update if required by project policy.

2. License check — The license-check CI job is failing. This may be unrelated to this PR's changes, but worth investigating.

🔍 Minor Observations

File	Line	Note
tools/wheel_builder/res/python_packages.toml	57, 75	Comments updated from "torch 2.10" to "torch 2.12" ✓
source/isaaclab_rl/setup.py	24	Comment correctly updated to reference torch 2.12.0 ✓
source/isaaclab_tasks/setup.py	24	Comment correctly updated to reference torch 2.12.0 ✓

💡 Recommendation

This is a clean dependency version bump. Once the CI checks pass (changelog fragment added if required, license-check resolved), this should be good to merge.

Verdict: Minor administrative fixes needed (CI checks), but the code changes themselves are correct and complete.