Rename SelectiveMixedPrecision heuristic algorithms

[Ti-Tai Wang (titaiwangms)]

Describe your changes

Rename the misleading k_quant_* SelectiveMixedPrecision heuristic names to describe which layers are promoted to high precision:

• k_quant_last -> high_precision_lm_head
• k_quant_down -> high_precision_mlp_down
• k_quant_mixed -> high_precision_mlp_down_qkv

Legacy configuration strings remain available and emit FutureWarning. Deprecated Python enum names remain available as silent aliases, while get_k_quant_config remains available and emits FutureWarning. The independent ORT GenAI ModelBuilder int4_algo_config values remain unchanged. Tests and quantization documentation are updated accordingly.

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.

Release note: SelectiveMixedPrecision now exposes descriptive high_precision_* names for its layer-based heuristics. Existing k_quant_* configuration strings continue to work with deprecation warnings.

(Optional) Issue link

Related to #2638.

[Copilot]

Pull request overview

This PR renames SelectiveMixedPrecision’s heuristic algorithm identifiers from the misleading k_quant_* names to descriptive high_precision_* names, while keeping legacy strings and helper APIs working via deprecation warnings. This helps reduce confusion with the separate KQuant pass and makes mixed-precision intent clearer to users.

Changes:

• Renamed heuristic algorithms to high_precision_lm_head, high_precision_mlp, and high_precision_mlp_qkv, with legacy k_quant_* strings mapped via FutureWarning.
• Added/updated unit tests to validate warning behavior, schema/search exposure, and algorithm validation.
• Updated documentation and clarified that ORT GenAI ModelBuilder int4_algo_config values are unrelated and remain unchanged.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
olive/passes/pytorch/selective_mixed_precision.py	Introduces new canonical heuristic names, legacy-string normalization with warnings, and renames the helper to get_high_precision_config while keeping get_k_quant_config deprecated.
test/passes/pytorch/test_selective_mixed_precision.py	Updates end-to-end heuristic tests to new names and adds deprecation/normalization/schema coverage tests.
olive/passes/onnx/model_builder.py	Adds a clarifying comment that int4_algo_config values are forwarded to ORT GenAI and unrelated to SelectiveMixedPrecision’s algorithm enum.
docs/source/blogs/quant-slms.md	Updates the blog to reference the new high_precision_* mixed-precision configurations.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

Suppressed comments (1)

olive/passes/pytorch/selective_mixed_precision.py:663

• Direct access to these deprecated Python names does not emit the promised FutureWarning: standard Enum alias lookup returns the canonical member directly and never invokes _missing_, which only handles value construction such as Algorithm("k_quant_down"). The new test likewise accesses Algorithm.K_QUANT_DOWN without expecting a warning. Please either add and test a warning-aware mechanism for deprecated member access or narrow the PR's compatibility/deprecation contract.

        # Deprecated Python API names. Their canonical values make these Enum aliases, so normal
        # iteration (and consequently schema/search generation) only exposes the canonical names.
        K_QUANT_LAST = "high_precision_lm_head"
        K_QUANT_DOWN = "high_precision_mlp"
        K_QUANT_MIXED = "high_precision_mlp_qkv"