perf(optimizers): wire amsgrad to the fused compiled training path (#653)

[ooples]

Summary

#653 core-saturation, training-perf lever: widen fused-optimizer coverage so AMSGrad no longer falls off the fused compiled training path into the eager tape fallback.

AMSGradOptimizer<T> now implements IFusedOptimizerSpec and maps to the Tensors OptimizerType.AMSGrad fused kernel via TryMapToFusedOptimizerConfig, so TrainWithTape keeps AMSGrad on the fused fast path (which eliminates the bulk of the tape-walk backward cost) instead of dropping to the eager fallback cliff. The map is guarded: it declines (returns false) when UseAdaptiveLearningRate is set or the LR schedule isn't fused-representable, so those paths still run eager — correctness preserved.

Tests

FusedOptimizerParityTests.AMSGrad_FusedMatchesEager_NoWorseThanAdam — runs the fused (compiled) vs eager training step for 40 steps and asserts:

• the fused path actually engages (fusedSteps > 0),
• training makes progress (trainDelta > 1e-6),
• fused-vs-eager parameter divergence stays within the Adam-parity bound.

Green locally.

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • AMSGrad optimizer now supports fused-training compilation mode for optimized performance.

• Tests

  • Added integration test to verify training parity between fused and eager execution modes for AMSGrad.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
aidotnet_website	Ready	Preview, Comment	Jun 21, 2026 4:41pm
aidotnet-playground-api	Ready	Preview, Comment	Jun 21, 2026 4:41pm

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 14479147-565f-4cbf-9e41-222098d426e0

📥 Commits

Reviewing files that changed from the base of the PR and between 7a46620 and 16ee571.

📒 Files selected for processing (2)

• src/Optimizers/AMSGradOptimizer.cs
• tests/AiDotNet.Tests/IntegrationTests/Optimizers/FusedOptimizerParityTests.cs

---

Walkthrough

AMSGradOptimizer adds Fused.IFusedOptimizerSpec to its class declaration and implements TryGetFusedOptimizerConfig, which returns a FusedOptimizerConfig for OptimizerType.AMSGrad unless adaptive LR mode is active or no fused scheduler can be resolved. A new integration test verifies fused/eager parity against an Adam control.

Changes

AMSGrad Fused Optimizer Integration

Layer / File(s)	Summary
IFusedOptimizerSpec implementation on AMSGradOptimizer src/Optimizers/AMSGradOptimizer.cs	Class declaration updated to implement Fused.IFusedOptimizerSpec; TryGetFusedOptimizerConfig added with guard clauses for adaptive LR mode and unresolvable fused LR schedules, returning a populated FusedOptimizerConfig with Beta1, Beta2, Epsilon, and derived schedule otherwise.
Fused/eager parity integration test tests/AiDotNet.Tests/IntegrationTests/Optimizers/FusedOptimizerParityTests.cs	New [Fact] test trains Adam for a control divergence value, then runs AMSGradOptimizer on both fused and eager code paths, asserting parity via AssertOptimizerParity and a not-worse-than-Adam bound.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🧮 AMSGrad steps in, fused and keen,
No adaptive mode to intervene.
Beta1, Beta2, Epsilon align,
A config returned when the schedule is fine.
Parity tested, Adam's the bar—
The gradient flows, near and far. ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'perf(optimizers): wire amsgrad to the fused compiled training path' clearly and specifically describes the main change—enabling AMSGradOptimizer to use the fused compiled training path for performance improvement.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch perf/653-fused-optimizer-coverage

---

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