Revert "[None][test] Add support for nemotron_3_ultra_550b_nvfp4 model in performance tests and configurations"

[tburt-nv]

Reverts #15166

Summary by CodeRabbit

• Tests

  • Expanded HuggingFace model path mappings with new model variants (Llama v3.1, Phi 4 Mini).
  • Updated model resolution and engine output path handling in performance test framework.
  • Added new performance test case for Llama v3.1 8B Instruct model.

• Chores

  • Removed Nemotron 3 Ultra 550B model configuration and related test cases.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 8db44068-eb1b-4959-8514-ebee9cddf1dd

📥 Commits

Reviewing files that changed from the base of the PR and between db7161b and 0c2d91b.

📒 Files selected for processing (5)

• tests/integration/defs/perf/_model_paths.py
• tests/integration/defs/perf/pytorch_model_config.py
• tests/integration/defs/perf/test_perf.py
• tests/integration/test_lists/qa/llm_perf_core.yml
• tests/integration/test_lists/test-db/l0_perf.yml

---

📝 Walkthrough

Walkthrough

The PR updates the TensorRT-LLM performance test infrastructure to refactor HuggingFace model path handling. It expands model mappings, reorders test conditions, removes legacy Nemotron-3-Ultra-550B support, and adjusts test matrices to cover new model variants and improve test organization across multiple GPU configurations.

Changes

Performance Test Infrastructure Refactoring

Layer / File(s)	Summary
HuggingFace model path mapping expansion tests/integration/defs/perf/_model_paths.py	The HF_MODEL_PATH dictionary is expanded with new HuggingFace model entries (llama_v3.1_8b_hf, llama_v3.1_8b_instruct_hf, phi_4_mini_instruct_hf) while removing the old nemotron_3_ultra_550b_nvfp4 entry.
Model resolution and engine path refactoring tests/integration/defs/perf/test_perf.py, tests/integration/defs/perf/pytorch_model_config.py	get_model_dir is rewritten to resolve HuggingFace models through MODEL_PATH_DICT and llm_models_root() with *_hf suffix handling. Build command wiring is updated to use HF_MODEL_PATH lookups. Engine output directory structure is changed to nest under engine_dir/<hf_model_name>/<tp_pp_str>. Legacy model identifiers are removed from gating sets, and the Nemotron-3-Ultra-550B pattern configuration is removed.
Test matrix reorganization and expansion tests/integration/test_lists/qa/llm_perf_core.yml, tests/integration/test_lists/test-db/l0_perf.yml	Condition blocks are reorganized with new test entries added for nemotron_nano_12b_v2, qwen3.5 variants, llama_v3.3_nemotron_super_49b, and llama_v3.1_nemotron_ultra_253b across multiple GPU configurations. Duplicate test blocks are consolidated, and a new l0_perf test case is added for llama_v3.1_8b_instruct on H100 GPUs.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#15166: Modifies the same performance-test infrastructure for nemotron_3_ultra_550b_nvfp4, expanding HF_MODEL_PATH and refactoring model resolution logic in test_perf.py.

Suggested reviewers

• StanleySun639
• ruodil

✨ 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 revert-15166-bug

---

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