[Kumo-Anomaly 1/n] Add DPM-Solver++ sampler by aw471 · Pull Request #1190 · NVIDIA/structured-data-models

aw471 · 2026-10-19T23:23:33Z

Summary

Adds DPMSolver, the sampler Kumo-Anomaly uses to reconstruct windows from noise. It is the second-order, single-step DPM-Solver++ on a uniform time grid, built on the existing NoiseScheduleVP.

Description of the change

DPMSolver(betas, num_steps=20) computes the time grid, the solver coefficients and the denoiser step indices once, and stores them as non-persistent buffers. forward(x, predict_noise) then runs the updates with a noise-predicting callable. Odd step counts end with a first-order update. There is no final denoising or clipping, and no host-device synchronization.

Each solver time t maps to the training step index round(t * total_N - 1), the integer step the denoiser is trained on. With the default 500-step schedule and 20 steps, the denoiser sees steps 499, 474, ..., 25.

The implementation comes from Aditi Gautam's sampling.py in #1167, which adds the rest of Kumo-Anomaly in one PR. This is the first of the PRs that split #1167 into reviewable pieces, following #1083.

Validation

With the same step mapping, outputs are bitwise identical to the reference DPM-Solver++ implementation (single-step, order 2, uniform time) for 1, 2, 3 and 20 steps in float32 and float64.

copy-pr-bot · 2026-10-19T23:23:37Z

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.

coderabbitai · 2026-10-21T17:35:16Z

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Last reviewed commit: 3c91ed4
Latest commit, not reviewed: 9bd692b

Use the following commands to manage reviews:

@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

▶️ Resume reviews
🔍 Trigger review

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/structured-data-models/.coderabbit.yaml
Review profile: QUIET
Plan: Enterprise
Run ID: 97d2475f-adb4-4a2c-bbb7-3a649eb426e9

📥 Commits

Reviewing files that changed from the base of the PR and between 69adeed and 3c91ed4.

📒 Files selected for processing: 2

sdm/models/kumo/timeseries/anomaly/sampling.py
test/models/kumo/timeseries/anomaly/test_sampling.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 8 remain after this review.

📝 Summary

Summary by CodeRabbit

New Features
- Added DPM-Solver++-based sampling for anomaly time-series generation.
- Supports configurable sampling steps across supported devices and numeric precisions.
- Improves noise integration accuracy with first- and second-order updates, including first-order handling for final evaluations.
Tests
- Added coverage for timestep selection, constant and linear noise signals, CPU/CUDA execution, and multiple numeric precisions.

Adds DPMSolver, which selects timesteps from a noise schedule and integrates predicted noise with first- and second-order updates. Tests cover timestep selection, numerical accuracy, device behavior, and dtype behavior.

Changes: DPM-Solver++ sampling

Layer / File(s)	Summary
Schedule and step selection `sdm/models/kumo/timeseries/anomaly/sampling.py`, `test/models/kumo/timeseries/anomaly/test_sampling.py`	`DPMSolver` defines the sampling API, constructs schedule intervals and buffers, and selects training timesteps. Tests verify the timestep sequence.
Prediction and sample updates `sdm/models/kumo/timeseries/anomaly/sampling.py`, `test/models/kumo/timeseries/anomaly/test_sampling.py`	The solver predicts noise, converts predictions to clean-data estimates, applies midpoint corrections for second-order intervals, and returns the integrated sample. Tests cover constant and linear signals across supported dtypes and devices.

Priority: ⬇️ Low

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant SamplingCaller
  participant DPMSolver
  participant predict_noise
  SamplingCaller->>DPMSolver: forward(x, predict_noise)
  DPMSolver->>predict_noise: predict noise at interval timestep
  predict_noise-->>DPMSolver: noise prediction
  DPMSolver->>predict_noise: predict noise at midpoint for second-order interval
  predict_noise-->>DPMSolver: midpoint noise prediction
  DPMSolver-->>SamplingCaller: integrated sample

Merge Risk: ⚪ Minimal · up to 3c91e

No merge-blocking issue was identified. Run the normal test checks before merging.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 14.29% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title clearly identifies the main change: adding a DPM-Solver++ sampler for Kumo-Anomaly.
Description check		The description accurately explains the new DPMSolver implementation, its behavior, step mapping, and validation.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches: 💡 1

📝 Generate docstrings: 💡

Commit to this branch
Create a new PR

🧪 Generate unit tests (beta)

Commit to this branch
Create a new PR

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

codecov · 2026-10-21T20:18:43Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.71%. Comparing base (12e9053) to head (9bd692b).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1190      +/-   ##
==========================================
+ Coverage   82.63%   82.71%   +0.08%     
==========================================
  Files         171      172       +1     
  Lines       10979    11033      +54     
==========================================
+ Hits         9072     9126      +54     
  Misses       1907     1907

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

Co-authored-by: Aditi Gautam <adgautam@nvidia.com>

aw471 · 2026-10-21T21:10:08Z

/ok to test 9bd692b

# Summary > Implements `KumoAnomaly.forward` and the `_KumoAnomaly` core for the contract from #1220, so the model scores each time step of a multivariate series by its reconstruction error. Weights are randomly initialized here; loading the checkpoint follows later in this stack. # Description of the change > `model(x, *, recipe=None, num_samples=30, num_steps=20, callbacks=None, generator=None)` fits the recipe from #1221 on `x` (shape `[..., R, D]`, with rows as ordered time steps) and returns scores with shape `[..., R, 1]`. Callbacks follow `ICLModel`: `on_query_preprocessing_end` runs after pre-processing and `on_model_forward_end` after the forward pass, and gradients are enabled when training or when a callback requires them. Unsupported feature stypes trigger the same warning as in `ICLModel`, and more than 40 features after pre-processing raise. > > Rows are split into windows of 100 time steps, and the final partial window is padded by repeating its last row. Each window is normalized by its scalar mean and standard deviation, and four alternating blocks are reconstructed under two complementary masks with the DPM-Solver++ sampler from #1190. The lower median over `num_samples` draws per mask gives the reconstruction, and the score is the mean absolute reconstruction error across features. > > The implementation comes from Aditi Gautam's `model.py` in #1167, adapted to the `Model` contract from #1220. This PR is stacked on #1190. # Validation > n/a --------- Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de> Co-authored-by: Aditi Gautam <adgautam@nvidia.com>

aw471 mentioned this pull request Oct 20, 2026

[Kumo-Anomaly 2/n] Add MinMaxScale processor #1208

Closed

aw471 force-pushed the aw471/kumo-anomaly-2-sampler branch from 6d2b046 to 4965266 Compare October 20, 2026 17:33

aw471 changed the title ~~[Kumo-Anomaly 1/8] Add DPM-Solver++ sampler~~ [Kumo-Anomaly 1/n] Add DPM-Solver++ sampler Oct 20, 2026

aw471 force-pushed the aw471/kumo-anomaly-2-sampler branch 2 times, most recently from 4645e56 to dc8dc89 Compare October 21, 2026 17:07

aw471 mentioned this pull request Oct 21, 2026

[Kumo-Anomaly 2/n] Add KumoAnomaly reconstruction and scoring #1211

Merged

aw471 marked this pull request as ready for review October 21, 2026 17:34

aw471 requested review from RBendias, akihironitta and rusty1s as code owners October 21, 2026 17:34

aw471 force-pushed the aw471/kumo-anomaly-2-sampler branch 3 times, most recently from 57c2782 to 69adeed Compare October 21, 2026 20:16

aw471 force-pushed the aw471/kumo-anomaly-2-sampler branch from 69adeed to 3c91ed4 Compare October 21, 2026 20:34

Add Kumo-Anomaly DPM-Solver++ sampler

9bd692b

Co-authored-by: Aditi Gautam <adgautam@nvidia.com>

aw471 force-pushed the aw471/kumo-anomaly-2-sampler branch from 3c91ed4 to 9bd692b Compare October 21, 2026 21:08

aw471 merged commit fadaa01 into main Oct 21, 2026
10 checks passed

aw471 deleted the aw471/kumo-anomaly-2-sampler branch October 21, 2026 21:17

Navigation Menu

Repository navigation

[Kumo-Anomaly 1/n] Add DPM-Solver++ sampler - #1190

[Kumo-Anomaly 1/n] Add DPM-Solver++ sampler#1190
aw471 merged 1 commit into
mainfrom
aw471/kumo-anomaly-2-sampler

aw471 commented Oct 19, 2026 •

edited

Loading

Uh oh!

copy-pr-bot Bot commented Oct 19, 2026

Uh oh!

coderabbitai Bot commented Oct 21, 2026 •

edited

Loading

Reviews paused

Summary by CodeRabbit

Sequence Diagram(s)

Uh oh!

codecov Bot commented Oct 21, 2026 •

edited

Loading

Uh oh!

aw471 commented Oct 21, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

aw471 commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Description of the change

Validation

Uh oh!

copy-pr-bot Bot commented Oct 19, 2026

Uh oh!

coderabbitai Bot commented Oct 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Summary by CodeRabbit

Sequence Diagram(s)

Uh oh!

codecov Bot commented Oct 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

aw471 commented Oct 21, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

aw471 commented Oct 19, 2026 •

edited

Loading

coderabbitai Bot commented Oct 21, 2026 •

edited

Loading

codecov Bot commented Oct 21, 2026 •

edited

Loading