Use CPU/GPU extras with uv sources for torch installation

[RajdeepKushwaha5]

Describe your changes

Move torch from direct [project] dependencies to [project.optional-dependencies] with mutually exclusive cpu and gpu extras. Add [tool.uv.sources], [[tool.uv.index]], and [tool.uv] conflicts so that uv sync --extra cpu (or --extra gpu) installs torch from the correct PyTorch index in a single command.

Simplify the uv CI path from manual uv venv + uv pip install torch + uv pip install -e . to uv sync --extra ${{ matrix.device }} --group dev. The pip CI path retains the dry-run hack since pip cannot read [tool.uv.sources]. README install instructions updated accordingly.

Issue Link

closes #600

Type of change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📖 Documentation (Addition or improvements to documentation)

Checklist before requesting a review

• My branch is up-to-date with the target branch - if not update your fork with the changes from the target branch (use pull with --rebase option if possible).
• I have performed a self-review of my code
• For any new/modified functions/classes I have added docstrings that clearly describe its purpose, expected inputs and returned values
• I have placed in-line comments to clarify the intent of any hard-to-understand passages of my code
• I have updated the README to cover introduced code changes
• I have added tests that prove my fix is effective or that my feature works
• I have given the PR a name that clearly describes the change, written in imperative form (context).
• I have requested a reviewer and an assignee (assignee is responsible for merging). This applies only if you have write access to the repo, otherwise feel free to tag a maintainer to add a reviewer and assignee.

Checklist for reviewers

Each PR comes with its own improvements and flaws. The reviewer should check the following:

• the code is readable
• the code is well tested
• the code is documented (including return types and parameters)
• the code is easy to maintain

Author checklist after completed review

• I have added a line to the CHANGELOG describing this change, in a section
  reflecting type of change (add section where missing):
  • maintenance: added entry under Maintenance in CHANGELOG.md

[Copilot]

Pull request overview

This PR restructures how PyTorch (torch) is installed by introducing mutually exclusive cpu/gpu extras and using uv source routing to automatically select the correct PyTorch index, simplifying both CI and contributor installation flows.

Changes:

• Move torch from base dependencies to cpu/gpu optional extras and add uv index/source routing plus extra conflicts.
• Simplify the uv CI installation path to uv sync --extra ${{ matrix.device }} --group dev.
• Update README installation guidance and add a Maintenance entry to the changelog.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
pyproject.toml	Adds cpu/gpu extras for torch and configures uv indexes/sources with conflict rules.
README.md	Updates installation instructions to use uv sync --extra cpu/gpu and clarifies pip’s manual torch install.
CHANGELOG.md	Documents the maintenance change for torch installation approach.
.github/workflows/install-and-test.yml	Updates CI to use uv sync for uv installs and adjusts the pip dry-run logic.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sadamov]

This looks good overall. The extras + [tool.uv.sources] approach is the right solution for uv sync. A few things to fix:

This PR has a big! side-effect, changing the cuda version from 11.1 to 12.8 during install instructions. This should be made very clear to everyone and also discussed at a future dev-meeting.

[sadamov]

torch must stay in core dependencies (CRITICAL)

Removing torch from [project.dependencies] means pip install neural-lam (or uv sync without an extra) produces a runtime ImportError. The extras should control the variant, not make torch optional.

Suggested change

	"plotly>=5.15.0",
	"torch>=2.3.0",
	"torch-geometric==2.3.1",
	"plotly>=5.15.0",
	"torch>=2.3.0",
	"torch-geometric==2.3.1",

[sadamov]

for some reason i cant get this inline suggestion to display properly, sorry

[sadamov]

Worth noting here that --torch-backend is not available for uv sync, so readers understand why the extras approach is used rather than a simpler flag.

Suggested change

	[tool.uv]
	conflicts = [[{ extra = "cpu" }, { extra = "gpu" }]]
	[tool.uv]
	# uv-specific configuration for PyTorch CPU/GPU variant selection.
	# NOTE: --torch-backend is only supported in `uv pip`, not `uv sync`.
	# Tracking issue for project-interface support: https://github.com/astral-sh/uv/issues/12994
	# Until then, we use extras + explicit index routing for `uv sync`.
	conflicts = [[{ extra = "cpu" }, { extra = "gpu" }]]

[sadamov]

Nit: name : has a spurious space (line 46 of the updated workflow file). Since this line is between diff hunks it cannot be an inline suggestion, but it should be fixed:

      - name: Install uv (if applicable)

[RajdeepKushwaha5]

Pushed fb8f4eb addressing all review comments. Summary of changes:

pyproject.toml

• Restored torch>=2.3.0 to [project.dependencies] so pip install neural-lam (and uv sync without an extra) keeps torch as a hard requirement. The cpu/gpu extras now only select the variant.
• Added the --torch-backend / uv issue #12994 note above [tool.uv].

.github/workflows/install-and-test.yml

• Rewrote the header comment per your suggestion (torch framed as required, extras select variant, --torch-backend note, pip routing explanation, hosts block preserved).
• Reverted the pip dry-run argument from ".[${{ matrix.device }}]" back to "." — pip ignores [tool.uv.sources] and torch is in core deps now, so the extra is unnecessary.
• Fixed the name : whitespace on the Install uv step.

README.md

• Replaced the **regarding torch installation** paragraph with your shorter wording that makes the hard requirement explicit.
• Expanded the uv install steps with the post-install uv pip install torch --torch-backend auto recipe for users on a CUDA version other than 12.8.

Commit message

• Updated to reflect that torch stays in core deps (the previous wording said "Move torch from direct dependencies to optional-dependencies" which is no longer accurate).

[sadamov]

Thanks for the round of revisions in fb8f4eb, all my previous comments are addressed. Three things left, and I'd like all of them in this PR rather than as follow-ups so we land the lockfile + cache-key story coherently with #606:

1. CHANGELOG: link the PR, split into two entries to surface the CUDA bump.
2. Commit uv.lock and switch CI to --locked. This is the big one.
3. README: lead the CUDA-swap recipe with --extra cpu (smaller download) and clarify the swap is reverted by the next uv sync.

Details inline.