Modernize docker builds for 2.0 release

[ktangsali]

PhysicsNeMo Pull Request

Description

This PR cleans up the dockerfile and makes following changes:

1. Uses uv to install packages (system wide and not in a virtual env)
2. Tries to leverage the dependency groups specified in the toml file as much as possible to ensure consistency between the two approaches
3. Simplifies the stages. Builder contains all the if-else logic, deploy is builder - mlflow and wandb (these two are removed as these packages typically have CVEs). We can also remove any other packages from builder here depending on the license / security guidance as need be. CI is builder + dev group, and a few other dependencies
4. Locks the version of cupy. cupy-cuda13x==14.0.0 (released on Feb 16, 2026) introduces a bug (see below). Locking the version fixes it.

name = 'cupy_backends.cuda.libs.torch', import_ = <function _gcd_import at 0x7c5fb53000e0>

>   ???
E   torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
E   ModuleNotFoundError: No module named 'cupy_backends.cuda.libs.torch'
E   
E   Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

<frozen importlib._bootstrap>:1324: BackendCompilerFailed

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• The CHANGELOG.md is up to date with these changes.
• An issue is linked to this pull request.
• If I am implementing a new model or modifying any existing model, I have followed the Models Implementation Coding Standards.

Dependencies

Review Process

All PRs are reviewed by the PhysicsNeMo team before merging.

Depending on which files are changed, GitHub may automatically assign a maintainer for review.

We are also testing AI-based code review tools (e.g., Greptile), which may add automated comments with a confidence score.
This score reflects the AI’s assessment of merge readiness and is not a qualitative judgment of your work, nor is
it an indication that the PR will be accepted / rejected.

AI-generated feedback should be reviewed critically for usefulness.
You are not required to respond to every AI comment, but they are intended to help both authors and reviewers.
Please react to Greptile comments with 👍 or 👎 to provide feedback on their accuracy.

[greptile-apps]

Greptile Summary

This PR modernizes the Docker build by replacing pip with uv for package installation, bumping the base image from pytorch:25.09-py3 to pytorch:26.01-py3, simplifying the stage hierarchy (builder -> ci/deploy -> docs), moving torch_cluster installation from CI-only to the builder stage, and consolidating many explicit dependency installs into uv pip install ".[cu13,utils-extras,...]" leveraging pyproject.toml extras. The Makefile is updated to rename MODULUS_GIT_HASH to PHYSICSNEMO_GIT_HASH.

• Build-breaking bug: uv pip install --group dev on line 217 is invalid -- --group is not a recognized flag for uv pip install (it's only available on uv sync/uv run). The CI image build will fail at this step.
• Editable install + source deletion: The CI stage installs physicsnemo in editable mode then deletes /physicsnemo/, which breaks import physicsnemo unless the CI runner remounts the source. This should either be documented or changed to a non-editable install.
• UV_CONSTRAINT may point to nonexistent file: The env var is set unconditionally, but the constraint file may not exist in all base images, which would cause all subsequent uv pip install calls to fail.
• Unpinned uv:latest: Using ghcr.io/astral-sh/uv:latest risks non-reproducible builds; consider pinning a specific version.

Important Files Changed

Filename	Overview
Dockerfile	Major Dockerfile modernization using uv, but contains a build-breaking bug (uv pip install --group dev is invalid syntax), a logic issue with editable install + source deletion, and a potential issue with UV_CONSTRAINT pointing to a nonexistent file.
Makefile	Simple rename of MODULUS_GIT_HASH to PHYSICSNEMO_GIT_HASH build arg to match the Dockerfile's ARG name. Correct and straightforward.

_{Last reviewed commit: fb02afd}

[greptile-apps]

_{2 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[NickGeneva]

LGTM to me thanks for addressing feedback

[ktangsali]

/blossom-ci