Refactor GroupedLinear quantization dispatch

[negvet]

Description

Move grouped quantization helpers into a dedicated module and replace separate split APIs with one. Keeps quantizer validation at construction only for CustomRecipe while skipping validation for the built-in recipes.
Followup for #2817

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[greptile-apps]

Greptile Summary

The PR extracts GroupedLinear split-quantization logic into a dedicated module and consolidates split, quantization, and bias-gradient handling behind one dispatch API.

• Adds quantizer-driven dispatch for native, hybrid, identity, debug, and Python fallback paths.
• Restricts grouped-quantizer compatibility validation to CustomRecipe generations.
• Updates GroupedLinear forward and backward call sites and expands quantization equivalence and dispatch tests.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains.

No blocking failure remains.

Important Files Changed

Filename	Overview
transformer_engine/pytorch/module/_split_quantization.py	Introduces the consolidated grouped split-quantization, compatibility-validation, hybrid-dispatch, and optional bias-gradient implementation.
transformer_engine/pytorch/module/grouped_linear.py	Delegates forward and backward split quantization to the new helper module and limits compatibility validation to CustomRecipe generations.
tests/pytorch/test_hybrid_quantization.py	Expands coverage for native bias-gradient dispatch, hybrid source policies, validation caching, and built-in recipe behavior.
tests/pytorch/test_custom_recipe.py	Strengthens GroupedLinear CustomRecipe coverage with exact forward and backward equivalence checks.
tests/pytorch/test_identity_quantizer.py	Updates identity and hybrid split tests for the unified API and verifies optional bias-gradient results.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  GL["GroupedLinear forward/backward"] --> SQ["_split_quantize"]
  SQ --> P{"Quantized output?"}
  P -->|No| Plain["Cast and torch.split"]
  P -->|Yes, native bgrad| BGrad["Per-split bgrad_quantize"]
  P -->|Yes, DebugQuantizer| Debug["Debug multi-tensor quantize"]
  P -->|Yes, HybridQuantizer| Hybrid["Rowwise/columnwise split quantization"]
  P -->|Yes, native quantizer| Native["Bulk split_quantize"]
  P -->|Yes, custom/identity| Fallback["Python per-split fallback"]
  Plain --> Out["Split outputs and optional dbiases"]
  BGrad --> Out
  Debug --> Out
  Hybrid --> Out
  Native --> Out
  Fallback --> Out

Loading

_{Reviews (2): Last reviewed commit: "[pre-commit.ci] auto fixes from pre-comm..." | Re-trigger Greptile}

[negvet]

/te-ci pytorch L1

[zhongbozhu]

Can we rename this file to something like split quantization? It's not really a grouped quantization

[zhongbozhu]

Same comment for renaming

[zhongbozhu]

same comment

[zhongbozhu]

overall LGTM, just have some questions

[zhongbozhu]

why do we need this cast(Sequence[Quantizer], quantizers)? feels like some extra CPU overhead?

[negvet]

This is only for the type checker, like assume there are no None elements. At runtime, it returns the same object unchanged, it does not iterate through.

[zhongbozhu]

/te-ci pytorch

[negvet]

/te-ci pytorch