[Parsing] Allow workload-specific scenario overrides

[podkidyshev]

Summary

Scenario entries can override workload-specific test definition fields without adding each field to TestRunModel. Explicitly supplied workload fields pass through to the existing merge and are validated by the selected workload. Scenario run controls remain separate, and referenced tests still cannot override test_template_name.

The NeMo Launcher FP8 two-node scenario now references the BF16 test TOML and overrides cmd_args.training.model.fp8 and extra_cmd_args. Both checked-in scenarios that used the duplicate FP8 test file were updated, and the duplicate was removed. The tutorial documents the shared override rule.

Test Plan

• Automated CI (parsing affected only)

Additional Notes

N/A

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/cloudai/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Enterprise

Run ID: 09b194cd-af61-412f-ae48-dbef96395b1e

📥 Commits

Reviewing files that changed from the base of the PR and between 01582d9 and 3519c88.

📒 Files selected for processing (1)

• conf/experimental/test_scenario/nemo_launcher_nemotron_15b_fp8.toml

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 10 remain after this review.

---

📝 Walkthrough

Walkthrough

Scenario serialization now includes explicitly set workload-specific fields. Nemotron FP8 scenarios reference BF16 test configurations and specify FP8 arguments. Documentation describes scenario override limits and removes prior guidance against workload-specific scenario arguments.

Changes

Scenario workload overrides

Layer / File(s)	Summary
Serialize workload-specific scenario overrides src/cloudai/models/scenario.py, tests/test_test_scenario.py, doc/Tutorial.rst, doc/workloads/sglang.rst, doc/workloads/vllm.rst	TestRunModel allows extra fields, and tdef_model_dump serializes explicitly set fields outside the scenario-level field set. Tests cover SGLang semantic evaluation argument overrides through path and test_name references, and check that an unknown field raises TestConfigParsingError. The documentation describes scenario override limits and removes guidance that workload-specific arguments require a separate test definition.
Update Nemotron FP8 scenario references conf/experimental/test_scenario/nemo_launcher_nemotron_15b_fp8.toml, conf/experimental/test_scenario/nemo_launcher_nemotron_15b_fp8_2_node.toml, conf/experimental/test/nemo_launcher_nemotron_15b_fp8_2_node.toml	Both scenarios reference BF16 test configurations and add names, descriptions, and FP8 arguments. The two-node scenario retains its node count. The separate two-node FP8 test configuration is removed.

Priority: ⬇️ Low

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Feature

Merge Risk: ⚪ Minimal · up to 3519c

The scenarios preserve the two-node FP8 configuration while reusing the BF16 definition. No actionable merge risk was identified.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: enabling workload-specific scenario overrides.
Description check	✅ Passed	The description accurately covers the parsing changes, NeMo configuration updates, documentation changes, and test plan.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

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

[podkidyshev]

@coderabbitai full review

[coderabbitai]

✅ Action performed

Full review finished.