Refactor(handler part 15): extract training preset switching into dedicated mixin

[1larity]

Summary

Extract switch_to_training_preset from AceStepHandler into a dedicated TrainingPresetMixin to continue handler decomposition while preserving behavior.

What Changed

• Added TrainingPresetMixin:
  • acestep/core/generation/handler/training_preset.py
• Wired mixin into handler composition:
  • acestep/core/generation/handler/__init__.py
  • acestep/handler.py
• Removed inline switch_to_training_preset implementation from AceStepHandler.
• Added focused unit tests:
  • acestep/core/generation/handler/training_preset_test.py
• Follow-up fix: preserve MLX behavior on preset switch by forwarding use_mlx_dit from cached init params.

Behavioral Notes

• Public interface is unchanged: AceStepHandler.switch_to_training_preset(...) remains available via mixin inheritance.
• switch_to_training_preset now explicitly forwards:
  • prefer_source (existing behavior)
  • use_mlx_dit (regression fix for MLX users)

Tests

• python acestep/core/generation/handler/training_preset_test.py (6 tests passing)
• Attempted broader run python acestep/core/generation/handler/init_service_test.py, but this environment lacks torchaudio (import failure unrelated to this PR logic).

Scope / Risk

• Single-responsibility decomposition only.
• No non-target runtime-path changes beyond preserving use_mlx_dit passthrough on preset switching.
• Low risk, with focused regression coverage.

Summary by CodeRabbit

• New Features

  • Added a training-safe preset mode that disables quantization for compatible inference setups and is now available on the handler interface.

• Tests

  • Added comprehensive unit tests covering preset switching, edge cases, error propagation, and preservation of cached init parameters.

[coderabbitai]

📝 Walkthrough

Walkthrough

Extracts training-preset logic from AceStepHandler into a new exported TrainingPresetMixin; integrates the mixin into AceStepHandler and adds unit tests validating reinitialization behavior and error cases. (49 words)

Changes

Cohort / File(s)	Summary
Mixin implementation acestep/core/generation/handler/training_preset.py, acestep/core/generation/handler/__init__.py	Adds TrainingPresetMixin with switch_to_training_preset() that reconstructs init params with quantization=None and calls initialize_service; exports mixin in package __all__.
Handler integration acestep/handler.py	Removes switch_to_training_preset from AceStepHandler and adds TrainingPresetMixin to its base classes.
Tests acestep/core/generation/handler/training_preset_test.py	Adds unit tests covering: disabled-quantization early return, missing last_init_params, successful reinit with quantization=None, initialize_service error propagation, handling of absent prefer_source, and immutability of cached params.

Sequence Diagram(s)

(omitted)

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• Refactor(handler part 2): Extract init utility mixin and add focused init-service tests #456 — Refactors AceStepHandler by extracting preset-switching behavior into a mixin and updating handler imports/bases.

Suggested reviewers

• ChuxiJ

Poem

🐰 I hop from code to line,
A mixin neat, a preset fine,
Quantization set to none,
Tests run bright beneath the sun,
Hooray — a cleaner handler, done!

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and specifically describes the main change: extracting training preset switching functionality into a dedicated mixin as part of handler refactoring.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

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

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

Tip

Issue Planner is now in beta. Read the docs and try it out! Share your feedback on Discord.

---

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

[ChuxiJ]

conflicts need to solve

[coderabbitai]

🧹 Nitpick comments (1)

acestep/core/generation/handler/training_preset_test.py (1)

10-35: Consider documenting why custom module loading is used.

The custom module loading approach isolates the test from heavyweight package imports (e.g., torch, torchaudio). A brief inline comment explaining this rationale would help future maintainers understand this pattern.

📝 Suggested documentation addition

 def _load_training_preset_module():
-    """Load training_preset module directly from file to avoid package side effects."""
+    """Load training_preset module directly from file to avoid package side effects.
+    
+    This bypasses the heavyweight acestep package imports (torch, torchaudio, etc.)
+    allowing fast, isolated unit tests without GPU or audio library dependencies.
+    """
     repo_root = Path(__file__).resolve().parents[4]

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@acestep/core/generation/handler/training_preset_test.py` around lines 10 -
35, Add a brief inline comment at the top of the _load_training_preset_module
function explaining that it manually constructs package entries and loads
training_preset.py to avoid importing the full aconda/package graph (which pulls
heavyweight dependencies like torch/torchaudio) so tests remain fast and
isolated; mention that this prevents package side-effects and ensures the test
only loads the target module.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@acestep/core/generation/handler/training_preset_test.py`:
- Around line 10-35: Add a brief inline comment at the top of the
_load_training_preset_module function explaining that it manually constructs
package entries and loads training_preset.py to avoid importing the full
aconda/package graph (which pulls heavyweight dependencies like
torch/torchaudio) so tests remain fast and isolated; mention that this prevents
package side-effects and ensures the test only loads the target module.

[1larity]

@ChuxiJ Conflict resolved.