[CharlelieLrt]

Description

This PR adds DiffusionExtraTimes and DiffusionExtraOutputs to
physicsnemo.diffusion for models that need more than one diffusion time or
produce auxiliary outputs. The primary t and prediction remain required
tensors. Named extra fields are independently optional and default to None.
The reserved fields are the MeanFlow destination time r and adaptive loss
weighting log-scale logvar. Custom consumers define their own fields' meaning.

Pass (t, DiffusionExtraTimes(r=r, batch_size=[B])) when a model needs a
destination time. Return (prediction, DiffusionExtraOutputs(logvar=logvar, batch_size=[B])) when its training loss needs the auxiliary log-scale.

Affine preconditioners encode every supplied time tensor with the same mapping
and scale only the prediction. Conditioning wrappers pass supported tuples
unchanged. Existing DSM and flow-matching losses and sampling APIs use tensors.

Checklist

• This PR follows the contribution guidelines.
• The documentation describes the API and its role in the stack.
• This PR remains a draft.

Dependencies

• Depends on

  • This PR starts from main and has no prerequisite PR.

• Depends on this PR

  • PhysicsNeMo #2095 adds inverse SNR and TrigFlow components. It follows this PR directly in the stack.

• Later follow-ups

  • PhysicsNeMo #2096 adds private adapters and public sCM and MeanFlow losses.
  • PhysicsNeMo #2097 extends the loss API to DMD2 and f-distill.
  • PhysicsNeMo #2098 adds checkpoint bundles for restarting distillation.
  • PhysicsNeMo #2099 adds the optional training strategy and the grouped changelog entry.

• Parallel FastGen work

  • FastGen #43 allows the NumPy 2 dependency used by PhysicsNeMo. PhysicsNeMo Add sCM and MeanFlow distillation losses #2096 uses an immutable fork pin, so it does not need to wait for this proposal to merge.
  • FastGen #44 provides shared loss hooks for external training loops. The initial PhysicsNeMo integration does not require these hooks.

• Conditional follow-up

  • PhysicsNeMo #2100 replaces local compatibility code with the proposed FastGen loss hooks. It depends on PhysicsNeMo Add optional diffusion distillation strategy #2099 and FastGen [other] Add modulus.models.fcn_mip_plugin #44, not the other way around.

If FastGen maintainers decline FastGen #44, we would close PhysicsNeMo #2100.
The preceding PhysicsNeMo stack would remain available.

Review process

Reviewers can compare this PR with its immediate parent. After a parent merges,
we update the remaining branches and their PR bases to keep each diff focused.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

CODEOWNERS review map

Current for commit fe02e1035bea. An approval covers every file listed for that owner; one owner is sufficient for shared files.

⏳ @CharlelieLrt — 7 file(s)

• docs/api/diffusion/models.rst
• docs/api/diffusion/preconditioners.rst
• physicsnemo/diffusion/init.py
• physicsnemo/diffusion/base.py
• physicsnemo/diffusion/preconditioners/preconditioners.py
• test/diffusion/helpers.py
• test/diffusion/test_model_io.py

⏳ @megnvidia — 2 file(s)

• docs/api/diffusion/models.rst
• docs/api/diffusion/preconditioners.rst

No CODEOWNER

• CHANGELOG.md

Comment /codeowners-info to refresh.

[CharlelieLrt]

Continued in PhysicsNeMo #2199, from branch diffusion-extra-fields in CharlelieLrt/modulus. The replacement preserves this implementation and includes the requested documentation clarifications. Please continue the review there. This PR retains the earlier discussion.