fix(recipes): sync INT4 blockwise recipe defaults

[realAsma]

What does this PR do?

Fixes the public general/ptq/int4_blockwise_weight_only recipe added in PR #1172 by composing it from the shared recipe units instead of carrying a manually duplicated quantizer exclusion list.

The duplicated list had drifted from configs/ptq/units/default_disabled_quantizers, so the public INT4 recipe missed newer standard exclusions such as linear-attention projections, VLM vision/projector modules, and nn.Embedding. Recipe mode can apply directly to the broader model path, so those missing exclusions can quantize modules that the shared preset/qformat path avoids.

NVBug: 6311597

Testing

• python_pwd tools/precommit/check_modelopt_recipes.py modelopt_recipes/general/ptq/int4_blockwise_weight_only.yaml
• pytest_pwd tests/unit/recipe/test_loader.py::test_load_recipe_all_builtins tests/unit/recipe/test_loader.py::test_general_ptq_yaml_matches_config_dicts -q
• pytest_pwd tests/unit/recipe/test_presets.py -q
• Direct assertion that general/ptq/int4_blockwise_weight_only includes current standard exclusions for *linear_attn.in_proj_a*, *linear_attn.in_proj_b*, VLM modules, and nn.Embedding.

Notes

The NVBug page redirects to Microsoft sign-in from my shell, so this fix is based on the Slack-linked PR diff anchor resolving to modelopt_recipes/general/ptq/int4_blockwise_weight_only.yaml and local recipe/preset drift analysis.

Summary by CodeRabbit

• New Features

  • Added a new INT4 blockwise weight-only PTQ recipe for easier low-bit quantization setup.
  • Streamlined recipe configuration so shared defaults are reused automatically.

• Bug Fixes

  • Improved recipe loading and validation for cases where some configuration entries are not present.
  • Updated consistency checks to handle these configurations more reliably.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4201f0af-ee20-414d-9249-4be696da4efc

📥 Commits

Reviewing files that changed from the base of the PR and between 33bfa8b and 8a5cf80.

📒 Files selected for processing (2)

• modelopt_recipes/general/ptq/int4_blockwise_weight_only.yaml
• tests/unit/recipe/test_loader.py

---

📝 Walkthrough

Walkthrough

The INT4 blockwise weight-only PTQ recipe now assembles its quantization config from imported fragments instead of an inline disable list. Loader tests add the recipe, allow mappings without a kv config, and conditionally build the expected quant config.

Changes

INT4 blockwise PTQ recipe

Layer / File(s)	Summary
Recipe imports and quantizer overrides modelopt_recipes/general/ptq/int4_blockwise_weight_only.yaml	imports now assembles quantize.quant_cfg from base_disable_all, int4_per_block for *weight_quantizer, *input_quantizer disabled, and default_disabled_quantizers, replacing the inline disable list.
Loader test updates for the new recipe tests/unit/recipe/test_loader.py	Adds the recipe to built-in PTQ coverage, maps its YAML with kv_cfg_name=None, and makes the YAML/config comparison skip kv_cfg concatenation when no kv config exists.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• NVIDIA/Model-Optimizer#1826: Updates default_disabled_quantizers.yaml, which is imported by the new INT4 blockwise weight-only recipe.

Suggested reviewers

• shengliangxu
• juhi10071998
• kevalmorabia97
• cjluo-nv

🚥 Pre-merge checks | ✅ 6

✅ Passed checks (6 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: syncing the INT4 blockwise recipe with shared default exclusions.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.
Security Anti-Patterns	✅ Passed	Only a YAML recipe and test changed; no banned patterns (weights_only=False, allow_pickle=True, trust_remote_code=True, eval/exec, # nosec, or new deps) were found.

✨ 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 beebot/nvbug-6311597

---

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.29%. Comparing base (33bfa8b) to head (8a5cf80).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1836   +/-   ##
=======================================
  Coverage   77.29%   77.29%           
=======================================
  Files         513      513           
  Lines       56920    56920           
=======================================
  Hits        43997    43997           
  Misses      12923    12923           

Flag	Coverage Δ
unit	54.61% <ø> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[realAsma]

Moved this diff into draft PR #1838 per owner request; closing this separate draft PR.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-06-26 22:29 UTC