feat(guidance): add --alignment-reverse-diffusion arg

[k-chrispens]

Pass through the alignment reverse diffusion arg.

Summary by CodeRabbit

• New Features
  • Added a new command-line flag to enable user control over reverse-diffusion alignment behavior in model inference operations.
  • Configuration is optional with intelligent fallback to model-specific defaults, providing flexibility for advanced use cases requiring alignment adjustments.

[coderabbitai]

Warning

Review limit reached

@k-chrispens, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 59 minutes and 28 seconds. Learn how PR review limits work.

Your organization has used up its prepaid credits, and credit purchases are no longer available. Enable the review add-on in the billing tab to keep reviews running — you're only billed for reviews past your plan's rate limits ($0.25/file).

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based credits.

🚦 How do rate limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan refill rate.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, the refill rate gradually slows as usage increases. The highest same-day bursts are limited more strictly.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 41a11e53-8d61-4c19-8ed2-732eb836059d

📥 Commits

Reviewing files that changed from the base of the PR and between 9c1b278 and 578c8d2.

📒 Files selected for processing (4)

• src/sampleworks/utils/guidance_script_arguments.py
• src/sampleworks/utils/guidance_script_utils.py
• tests/cli/test_guidance_cli.py
• tests/utils/test_guidance_script_utils.py

📝 Walkthrough

Walkthrough

Adds a new optional boolean field alignment_reverse_diffusion to GuidanceConfig, wires it to a new --alignment-reverse-diffusion CLI flag in add_generic_args, populates it through from_cli, and updates _run_guidance to use the override when set, falling back to is_boltz when not provided.

Changes

Alignment Reverse Diffusion Toggle

Layer / File(s)	Summary
GuidanceConfig field, CLI flag, and _run_guidance override src/sampleworks/utils/guidance_script_arguments.py, src/sampleworks/utils/guidance_script_utils.py	GuidanceConfig gains alignment_reverse_diffusion: bool | None; add_generic_args registers --alignment-reverse-diffusion as store_const=True with default=None; from_cli populates the field; _run_guidance assigns use_alignment_for_reverse_diffusion from the override when non-None, otherwise from is_boltz.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~5 minutes

Possibly related PRs

• diff-use/sampleworks#207: Introduced the same GuidanceConfig.from_cli() and _run_guidance wiring pattern that this PR extends with the new --alignment-reverse-diffusion option.

Suggested reviewers

• saada
• marcuscollins

Poem

🐇 A flag for the diffusion, reversed if you please,
--alignment-reverse-diffusion set with ease!
If None, fall back to is_boltz we say,
else use what the user has passed along the way.
The config grows richer, one field at a time — hooray! 🌿

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: adding a new CLI argument --alignment-reverse-diffusion to the guidance feature, which matches the primary objective and code modifications.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch kmc/alignment-reverse-diffusion

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

This PR adds a new CLI/config knob to control whether the EDM sampler performs “reverse diffusion” alignment (align noisy state to denoised prediction), defaulting to enabled for Boltz wrappers while allowing non-Boltz models to opt in.

Changes:

• Add alignment_reverse_diffusion: bool | None to GuidanceConfig and plumb it through GuidanceConfig.from_cli().
• Add --alignment-reverse-diffusion CLI flag and use it in _run_guidance() to override the per-model default.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/sampleworks/utils/guidance_script_utils.py	Computes the effective alignment_reverse_diffusion setting (model default vs. user override) and passes it into EDMSamplerConfig.
src/sampleworks/utils/guidance_script_arguments.py	Adds the new config field and CLI argument, and forwards it into the constructed GuidanceConfig.

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/sampleworks/utils/guidance_script_arguments.py`:
- Around line 457-467: The --alignment-reverse-diffusion argument uses
action="store_const" with const=True, which only allows users to set the value
to True (when the flag is provided) or None (default), preventing explicit
disabling of the feature. Replace the action="store_const" with const=True
approach with a mechanism that supports both enabling and disabling the feature,
such as using action="store_true" and "store_false" for separate flags
(--alignment-reverse-diffusion to enable and --no-alignment-reverse-diffusion to
disable), or use argparse.BooleanOptionalAction if available, ensuring users can
explicitly override the default behavior in both directions.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: bae053c9-ede6-4852-a105-98f9a7f1c618

📥 Commits

Reviewing files that changed from the base of the PR and between 3e6aba6 and 9c1b278.

📒 Files selected for processing (2)

• src/sampleworks/utils/guidance_script_arguments.py
• src/sampleworks/utils/guidance_script_utils.py

[marcuscollins]

I'd make a couple changes, and it probably does make sense to add some tests for these arguments to make sure they propagate correctly.

[marcuscollins]

No reason to use getattr here. The code explicitly constructs args.alignment_reverse_direction, its default is None anyway, and it should raise an error if that attribute doesn't exist. I would change the default to False and then the next like can just be use_alignment_for_reverse_diffusion = override | is_boltz

[marcuscollins]

Thanks for the changes.